Journal of Communication and Computer（Issue 8，2013）

Volume 10, Number 8, August 2013 (Serial Number 105)

Journal of

Communication and Computer

David Publishing Company

www.davidpublishing.com

PublishingDavid

Publication Information: Journal of Communication and Computer is published monthly in hard copy (ISSN 1548-7709) and online (ISSN 1930-1553) by David Publishing Company located at 3592 Rosemead Blvd #220, Rosemead, CA 91770, USA. Aims and Scope: Journal of Communication and Computer, a monthly professional academic journal, covers all sorts of researches on Theoretical Computer Science, Network and Information Technology, Communication and Information Processing, Electronic Engineering as well as other issues. Contributing Editors: YANG Chun-lai, male, Ph.D. of Boston College (1998), Senior System Analyst of Technology Division, Chicago Mercantile Exchange. DUAN Xiao-xia, female, Master of Information and Communications of Tokyo Metropolitan University, Chairman of Phonamic Technology Ltd. (Chengdu, China). Editors: Cecily Z., Lily L., Ken S., Gavin D., Jim Q., Jimmy W., Hiller H., Martina M., Susan H., Jane C., Betty Z., Gloria G., Stella H., Clio Y., Grace P., Caroline L., Alina Y.. Manuscripts and correspondence are invited for publication. You can submit your papers via Web Submission, or E-mail to [email protected]. Submission guidelines and Web Submission system are available at http://www.davidpublishing.org, www.davidpublishing.com. Editorial Office: 3592 Rosemead Blvd #220, Rosemead, CA 91770, USA Tel：1-323-984-7526, Fax: 1-323-984-7374 E-mail: [email protected]; [email protected] Copyright©2013 by David Publishing Company and individual contributors. All rights reserved. David Publishing Company holds the exclusive copyright of all the contents of this journal. In accordance with the international convention, no part of this journal may be reproduced or transmitted by any media or publishing organs (including various websites) without the written permission of the copyright holder. Otherwise, any conduct would be considered as the violation of the copyright. The contents of this journal are available for any citation. However, all the citations should be clearly indicated with the title of this journal, serial number and the name of the author. Abstracted / Indexed in: Database of EBSCO, Massachusetts, USA Chinese Database of CEPS, Airiti Inc. & OCLC Chinese Scientific Journals Database, VIP Corporation, Chongqing, P.R.China CSA Technology Research Database Ulrich’s Periodicals Directory Summon Serials Solutions Subscription Information: Price (per year): Print $520; Online $360; Print and Online $680 David Publishing Company 3592 Rosemead Blvd #220, Rosemead, CA 91770, USA Tel：1-323-984-7526, Fax: 1-323-984-7374 E-mail: [email protected]

David Publishing Company

www.davidpublishing.com

DAVID PUBLISHING

D

Journal of Communication and Computer

Volume 10, Number 8, August 2013 (Serial Number 105)

Contents Computer Theory and Computational Science

1019 Research in the Development of Finite Element Software for Creep Damage Analysis

Dezheng Liu, Qiang Xu, Zhongyu Lu and Donglai Xu

1031 The Global Crisis and Academic Communication: The Challenge of Social Networks in Research

Sandra Martorell and Fernando Canet

1042 Can You Explain This? Personality and Willingness to Commit Various Acts of Academic Misconduct

Yovav Eshet, Yehuda Peled, Keren Grinautski and Casimir Barczyk

Network and Information Technology

1047 Data Security Model for Cloud Computing

Eman M. Mohamed, Hatem S. Abdelkader and Sherif El-Etriby

1063 The Retraining Churn Data Mining Model in DMAIC Phases

Andrej Trnka

1070 Codebook Subsampling and Rearrangement Method for Large Scale MIMO Systems

Xin Su, Tianxiao Zhang, Jie Zeng, Limin Xiao, Xibin Xu and Jingyu Li

1076 A High-Precision Time Handling Library

Irina Fedotova, Eduard Siemens and Hao Hu

1087 New Hybrid Access Method for Femtocell through Adjusting QoS

Mansour Zuair, Abdul Malik Bacheer Rahhal and Mohamad Mahmoud Alrahhal

Communications and Electronic Engineering

1092 Design of an Information Connection Model Using Rule-Based Connection Platform

Heeseok Choi and Jaesoo Kim

1099 Communication Methods: Instructors’ Positions at Istanbul Aydin University Distance Education Institute

Kubilay Kaptan and Onur Yilmaz

1105 Coordination in Competitive Environments

Salvador Ibarra-Martinez, Jose A. Castan-Rocha and Julio Laria-Menchaca

1114 Logistics Customer Segmentation Modeling on Attribute Reduction and K-Means Clustering

Youquan He and Qianqian Zhen

1120 UHF Propagation Parameters to Support Wireless Sensor Networks for Onboard Trains

B. Nkakanou, G.Y. Delisle, N. Hakem and Y. Coulibaly

1131 A Novel Matlab-Based Underwater Acoustic Channel Simulator

Zarnescu George

1139 Normalized Efficient Routing Protocol for WSN

Rushdi Hamamreh and Mahmoud I Arda

Journal of Communication and Computer 10 (2013) 1019-1030

Research in the Development of Finite Element Software

for Creep Damage Analysis

Dezheng Liu1, Qiang Xu

2, Zhongyu Lu

2 and Donglai Xu

1

1. School of Science and Engineering, Teesside University, Middlesbrough TS1 3BA, UK

2. School of Computing and Engineering, University of Huddersfield, Huddersfield HD1 3DA, UK

Received: July 29, 2013 / Accepted: August 20, 2013 / Published: August 31, 2013.

Abstract: This paper reports the development of finite element software for creep damage analysis. Creep damage deformation and

failure of high temperature structure is a serious problem for power generation and it is even more technically demanding under the

current increasing demand of power and economic and sustainability pressure. This paper primarily consists of three parts: (1) the need

and the justification of the development of in-house software; (2) the techniques in developing such software for creep damage analysis;

(3) the validation of the finite element software conducted under plane stress, plane strain, axisymmetric, and 3 dimensional cases. This

paper contributes to the computational creep damage mechanics in general.

Key words: Finite element software, creep damage, CDM, axisymmetric, validation.

1. Introduction

Creep damage mechanics has been developed and

applied to analyze creep deformation and the

simulation for the creep damage evolution and rupture

of high temperature components [1].

The computational capability can only be obtained

by either the development and the application of

special material subroutine in junction with standard

commercial software (such as ABAQUS or ANSYS)

or the development and application of dedicated

in-house software.

The needs of such computational capability and the

justification for developing in-house software were

reported in the early stage of this research [2, 3].

Essentially, the creep damage problem is of time

dependent, non-linear material behavior, and

multi-material zones. Becker et al. [4] and Hayhurst

and Krzeczkowski [5] have reported the development

and the use of their in-house software for creep damage

Corresponding author: De Zheng Liu, Ph.D. student,

research fields: mechanical engineering, finite element method. E-mail: [email protected].

analysis; furthermore, Ling et al. [6] have presented a

detailed discussion and the use of Runge-Kutta type

integration algorithm. On the other hand, it was noted

that Xu [7] revealed the deficiency of KRH

(Kachanov-Rabatnov-Hayhurst) formulation and

proposed a new formulation for the multi-axial

generalization in the development of creep damage

constitutive equations. The new creep damage

constitutive equations for low Cr-Mo steel and for high

Cr-Mo steel are under development in this research

group [8, 9].

The purpose of this paper is to present the finite

element method based on CDM (continuum damage

mechanics) to develop FE software for creep damage

mechanics. More specifically, it summarizes the

current state of how to obtain such computational

capability then it concludes with a preference of

in-house software; secondly, it reports the development

of such software including the development of finite

element algorithms based on CDM for creep damage

analysis, and a flow diagram of the structure of new

finite element software has been completed to be

Research in the Development of Finite Element Software for Creep Damage Analysis

1020

guided in developing in-house FE software, and the use

of some standard subroutines in programming; thirdly,

the development and the validation of the finite

element software conducted so far include plane stress,

plane strain, axisymmetric case, and 3D case.

2. Current Finite Element Software for

Creep Damage Analysis

2.1 The Industrial Standard FE Package

The current industrial standard FE packages are

listed and commented in Table 1. The standard FE

package is not able to provide the creep damage

analysis capability and it can be expanded with the

development and use of special subroutine.

2.2 The In-house Finite Element Software

The in-house finite element softwares developed and

used for creep damage analysis are listed and

commented in Table 2.

2.3 Why the In-house Computational Software.

FE standard packages can only obtain the capability

for creep damage analysis by developing material user

subroutine for investigating creep damage problems,

which is very complex and not accurate [21].

Computational capability such as CDM for creep

damage analysis is not readily available in the

industrial standard FE packages. FE standard packages

such as ABAQUS does not currently permit the failure

of and the removal of the failed element from the

boundary-value problems during the solution process

[11]. Thus, there still have advantages in developing

and using in-house finite element software.

3. The Development of the New Finite

Element Software

3.1 The General Structure of the Finite Element

Software

The structure of developing in-house finite element

Table 1 The main industrial standard FE package.

Industrial standard FE

package Samples of application Observation and comment

ABAQUS

Numerical investigation on the creep damage induced by void growth in

heat affected zone of weldments [10] Benchmarks for finite element analysis of creep continuum damage mechanics [4]

The developer must develop a user-subroutine in junction with ABAQUS commercial FE code such as ABAQUS-UMAT damage code to analysis the creep CDM numerical problem [4]. It can access to a wide range of element types, material models and other facilities such as efficient equation solvers, not normally available in in-house FE codes. It does not currently permit the removal of failed elements from the

boundary-value problem during the solution process [11].

ANSYS

Development of a creep-damage model for non-isothermal long-term strength analysis of high-temperature components operating in a wide stress range [12]

Numerical benchmarks for creep-damage modelling [13]

ANSYS uses full Newton-Raphson scheme for global solution to achieve better convergence rate [13]. The material matrix must be consistent with the material constitutive integration scheme for the better convergence rate of the overall Newton-Raphson scheme. To ensure the overall numerical stability, the user should ensure that the integration scheme implemented in subroutine is stable.

Developing the user-subroutine for analyzing creep damage problems is very complex and inefficient [14].

MSC.Marc software Numerical modelling of GFRP laminates with MSC.Marc system and experimental validation [15]

Marc is a powerful, general-purpose, nonlinear finite element analysis solution to accurately simulate the response of your products under static, dynamic and multi-physics loading scenarios. Developing the user-subroutine for analyzing creep damage problems is very complex and inefficient [14].

RFPA2D-Creep Research on the closure and creep mechanism of circular tunnels [16]

RFPA2D-Creep introduces the degeneration equation on the mechanical characteristics of micro-element based on the meso-damage model in order to reveal the relationship between the damage accumulation, deformation localization and integral accelerated creep. The failed element can not be removed and the accuracy should be improved [11].


1021

Table 2 The main in-house finite element software.

FE software & author Characterization Observation and comment

FE-DAMAGE T.H. Hyde et al.

FE-DAMAGE is written in FORTRAN and developed at University of Nottingham [4]. Facilities for creep continuum damage analysis are included in which a single damage parameter constitutive equation is adopted. The failed elements from the boundary-value problem can be removed during the solution process.

The OOP (object oriented programming)

approach is not used in programming this software [4]. The OOP (object oriented programming) approach could be used in future.

DAMAGE XX D.R. Hayhurst et al.

DAMAGE XX is 2-D CDM-based FE solver, which has been developed over three decades by a number of researchers [17].

The failed elements from the boundary-value problem can be removed during the solution process. The running speed of the computer code has been increased by vectorization and parallel processing on the Cray X-MP/416.

The inability to solve problems with large numbers of elements and degrees of freedom [18]. An inefficiency numerical equation solver occupies a large proportion of the computer resource. Fourth order integration scheme was used in program, but the details have published might be incorrect according to Ling et al.

[6].

DAMAGE XXX R.J. Hayhurst M.T. Wang

DAMAGE XXX is developed to model high-temperature creep damage initiation, evolution and crack growth in 3-D engineering components [19]. The failed elements from the boundary-value problem can be removed during the solution process. It is running on parallel computer parallel architectures. The

tetrahedral elements are used in the DAMAGE XXX [17].

An inefficiency numerical equation solver occupies a large proportion of the computer resource. Fourth order integration scheme was used in program, but the details have published might be incorrect according to Ling et al

[6].

DNA (damage non-linear analysis) G.Z. Voyiadjis et al.

DNA stands for damage nonlinear analysis. It was at Louisiana State University in Baton Rouge. It includes the elastic, plastic and creep damage analysis of materials incorporating damage effects [20]. Both linear and nonlinear analysis options are available in DNA. The failed elements from the boundary-value problem can not be removed during the solution process.

Number of nodes in a problem must not exceed 3,000, the number of elements in a problem must not exceed 400 [20]. It is a 32-bit DOS executable file which can only run undue the Windows 96/98/NT operating system.

software for creep damage analysis is listed in Fig. 1.

The steps for the development of finite element

software can be summarized in:

(1) Input the definition of a specific FE model

including nodes, element, material property, boundary

condition, as well as the computational control

parameters;

(2) Calculate the initial elastic stress and strain;

(3) Integrate the constitutive equation and update the

field variables such as creep strain, damage, stress; the

time step is controlled;

(4) Remove the failed element [17] and update the

stiffness matrix;

(5) Stop execution and output results.

3.2 The Equilibrium Equations

Assume that the total strain ε can be partitioned into

the elastic and creep strains, thus the total strain

increment can be expressed as:

Δε = Δεe + Δε

c (1)

where Δε, Δεe and Δε

c are increments in total, elastic

and creep strain components, respectively [22].

The stress increment is related to the elastic and

creep strain increments by:

Δσ = D(Δε – Δεc) (2)

where D is the stress-strain matrix and it contains the

elastic constants.

The stress increments are related to the incremental

displacement vector Δu by:

Δσ = D(BΔu – Δεc) (3)

where B is strain matrix. The equilibrium equation to

be satisfied any time can be expressed by:

∫vBTΔσ dv = ΔR (4)

where ΔR is the vector of the equivalent nodal

mechanical load and v is the element volume.

Combining Eqs. (3) and (4):

∫vBTD(BΔu – Δε

c)dv = ΔR (5)

3.3 Sample Creep Damage Constitutive Equations

The creep damage constitutive equations are


1022

Fig. 1 The structure of developing new FE software.

proposed to depict the behaviors of material during

creep damage (deformation and rupture) process,

especially for predicting the lifetime of material. One

example is KRH constitutive equations which is

popular and is introduced here [23].

Uni-axial form

𝜀 = 𝐴 𝑠𝑖𝑛ℎ(

𝐵𝜎 1−𝐻

1−𝜑 1−𝜔 ) (6.1)

𝐻 =ℎ

𝜎 1 −

𝐻

𝐻∗ 𝜀 (6.2)

𝜑 =𝐾𝐶

3 1 − 𝜑 4 (6.3)

𝜔 = 𝐶𝜀 ∗ 6.4

(6)


1023

where A, B, C, h, H* and Kc are material parameters. H

(0 < H < H*) indicates strain hardening during primary

creep, φ (0< φ < 1) describes the evolution of spacing

of the carbide precipitates [23].

Multi-axial form

𝜀𝑖𝑗 =

3𝑆𝑖𝑗

2𝜎𝑒𝐴𝑠𝑖𝑛ℎ(

𝐵𝜎𝑒(1−𝐻)

(1−𝜑)(1−𝜔)) (7.1)

𝐻 =ℎ

𝜎𝑒 1 −

𝐻

𝐻∗ 𝜀 (7.2)

𝜑 =𝐾𝐶

3 1 −𝜑 4 (7.3)

𝜔 = 𝐶𝜀𝑒 𝜎1

𝜎𝑒 𝜈 7.4

(7)

where 𝜎𝑒 is the Von Mises stress, 𝜎1 is the maximum

principal stress and 𝜈 is the stress state index defining

the multi-axial stress rupture criterion [23].

The intergranular cavitation damage varies from

zero, for the material in the virgin state, to 1/3, when all

of the grain boundaries normal to the applied stress

have completely cavitated, at which time the material is

considered to have failed [24]. Thus, the critical value

of creep damage is set to 0.3333333 in the current

program. Once the creep damage reaches the critical

value, the program will stop execution and the results

will be output automatically. Other type of creep

damage constitutive equations will be incorporated in

the FE software in future.

3.4 The Integration Scheme

The FEA solution critically depends on the selection

of the size of time steps associated with an appropriate

integration method. Some integration methods have

been reviewed in previous work [3]. In the current

version, Euler forward integration subroutine,

developed by colleagues [25], was adopted here for

simplicity.

𝜀𝑛+1 = 𝜀𝑛 + 𝜀 ∗ ∆𝑡 (8.1)

𝐻𝑛+1 = 𝐻𝑛 + 𝐻 ∗ ∆𝑡 (8.2)𝜑𝑛+1 = 𝜑𝑛 + 𝜑 ∗ ∆𝑡 (8.3)

𝜔𝑛+1 = 𝜔𝑛 + 𝜔 ∗ ∆𝑡 8.4

𝑡𝑛+1 = 𝑡𝑛 + ∆𝑡 8.5

(8)

It is noted that D02BHF (NAG) [26] integrates a

system of first-order ordinary differential equations

solution using Runge-Kutta-Merson method. This

subroutine can be adopted in the FEA software of creep

damage analysis development, and a detailed

instruction on how to use it was published by the

company [26]. A more sophisticated Runge-Kutta type

integration scheme will be adopted and explored in

future.

3.5 The Finite Element Algorithm for Updating Stress

The Absolute Method [27] has been given for the

solution of the structural creep damage problems. The

principle of virtual work applied to the boundary value

problem is given:

Pload = [Kv] × TOTD – Pc (9)

where Pload is applied force vector, and [Kv] is the

global stiffness matrix, which is assembled by the

element stiffness matrices [Km]; TOTD is the global

vector of the nodal displacements and Pc is the global

creep force vector.

[Km] = ∫∫[B]T[D][B]dxdy (10)

The [B] and [D] represent the strain-displacement

and stress-strain matrices, respectively.

TOTD = [Kv]–1

× (Pload + Pc) (11)

The initial Pc is zero and the Choleski Method [27] is

used for the inverse of the global stiffness matrix [Kv].

By giving the Pload, the elastic strain εek and the elastic

stress σek for each element can be obtained:

εek = [B] × ELD (12)

σek = [D] × ε (13)

The element node displacement ELD can be found

from the global displacement vector and the creep

strain rate εckrate for each element can be obtained by

substituting the element elastic stress into the creep

damage constitutive equations. The creep strain can be

calculated as:

εck(t + △t) = εcK(t) + εcKrate × △t (14)

The node creep force vectors for each element are

given by:

Pck = [B]T[D] × εcK (15)

The node creep force vector Pck can be assembled

into the global creep force vector Pc and the Pc is used


1024

to up-date Eq. (9). Thus, the elastic strain can be

updated:

εtotk= [B] × ELD = εek + εck (16)

εek= [B] × ELD – εck (17)

where the εtotk and εck represent the total strain and

creep strain for each element, respectively; and the

elastic strain εek is used to up-date Eq. (13).

3.6 The Standard FE Library and Subroutines

In the development of this software, the existing FE

library and subroutines such as Smith’s [27] were used

in programming. The subroutines can perform the tasks

of finite element meshing, assemble element matrices

into system matrices and carry out appropriate

equilibrium, eigenvalue or propagation calculations.

Some subroutines used in programming are reviewed

in Table 3.

4. Preliminary Validation and Discussion

4.1 The Validation of Plane Stress Problem

The validation of new software for plane stress was

performed and it was conducted via a two-dimensional

tension model in Fig. 2. The length of a side is set to 1

m. The Young’s modulus E and Poisson’s ratio υ are

set to 1,000 MPa and 0.3, respectively. A uniformly

distributed linear load 40 kN/m was applied on the top

line of this uni-axial tension model.

The theoretical stress in Y direction is 40 kN/m2.

The stress in X direction should be zero. These stress

values should remain the same throughout the creep

test up to failure.

Samples of the stress obtained from FE software

with the stress updating invoked due to creep

deformation are shown in Fig. 3 and Fig. 4.

Using the theoretical stress value into the uni-axial

version of creep constitutive equations and a 0.1 h time

step with Euler forward integration method, the

theoretical rupture time, creep strain rate, creep strain

and damage can be obtained by a excel program [28]

and some of them are shown in Table 4.

Using the uni-axial version of creep constitutive

equations and a 0.1 h time step with Euler forward

integration method, the rupture time, creep strain rate,

creep strain and damage obtained from FE software at

failure were obtained and are shown in Table 5.

Table 3 The existing FE library and subroutines.

The standard subroutine Function

Subroutine geometry_3tx To form the coordinates and node vector for a rectangular mesh of uniform three-node triangles

Subroutine formkb and Subroutine formkv To assemble the individual element matrices to form the global matrices

Subroutine sparin and Subroutine spabac To solve the sets of linear algebraic equations based on the Cholesky direct solution method

Fig. 2 2D plane stress tension mode.

Fig. 3 The stress distribution in Y direction at rupture

time.


1025

Fig. 4 The stress distribution in X direction at rupture

time.

Table 4 The results obtained from excel program.

Rupture time Creep strain Creep damage

104,062 0.179934333 0.33333335

The percentage errors of FE results against

theoretical results are shown in Table 6.

A comparison of the results shown in Table 4 and

Table 5 and an examination of the percentage errors

shown in Table 6 clearly show that the results obtained

from the FE software do agree with the expected

theoretical values and the percentage errors are

negligible.

In the current version, Euler forward integration

subroutine, developed by colleagues [25], was adopted

here. Rupture time, strain rate, creep strain and damage

obtained from FE software have revealed that the FE

results have a good agreement with the theoretical

values.

4.2 The Validation of Plane Strain Problem

The validation of this software for plane stress was

performed and it was conducted via a uni-axial tension

model in Fig. 5. The width of this model is set to 5 m.

The Young’s modulus E and Poisson’s ratio υ are set to

1,000 MPa and 0.3, respectively. A uniformly linear

distributed load 10 kN/m was applied on the top line of

this model.

The theoretical stress in Y direction can be shown as:

𝜎𝑦 =𝑃

𝐴=

50

5.0= 10 kN/m2

The theoretical stress in Z direction can be shown as:

𝜎𝑧 = 𝐸 × 𝜖𝑧 = 𝐸 × 𝜐 × 𝜖𝑦 = 𝐸 × 𝜐 ×𝜎𝑦𝐸

= 3kN/m2

The stress and displacement obtained from FE

software with the stress updating invoked due to creep

deformation are shown in Fig. 6 and Fig. 7. The

displacements in x and y direction is shown in Fig. 8

and Fig. 9, respectively.

Using the theoretical stress value into the multi-axial

version of creep constitutive equations, the theoretical

rupture time and damage can be obtained without stress

update by a testified subroutine [25] and the results

obtained without stress update are shown in Table 7.

Table 5 The results obtained from FE software for plane stress problem.

Element number Rupture time Strain rate Creep strain Creep damage

Element No. 1 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 2 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 3 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 4 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 5 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 6 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 7 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Element No. 8 0.1040E+06 0.6540E-04 0.1798E+00 0.3334E+00

Table 6 The percentage errors.

Rupture time percentage error = 104,000 − 104,062

104,062 × 100 = 0.0596%

Creep strain percentage error = 0.1798 − 0.179934333

0.179934333 × 100 = 0.0747%

Damage percentage error = 0.3334 − 0.33333335

0.33333335 × 100 = 0.02%


1026

Fig. 5 Plane strain tension model.

Fig. 6 Stress distribution in Y direction at rupture time.

Fig. 7 Stress distribution in Z direction at rupture time.

Fig. 8 Displacement distribution in Y axis at rupture time.

Fig. 9 Displacement distribution in X axis at rupture time.

Table 7 The theoretical rupture time and creep damage

for plane strain case.

Rupture time Creep damage

180,460 0.3333334

Fig. 6 and Fig. 7 show that the results obtained from

the FE software do agree with the expected theoretical

values.

The displacement is distributed reasonable in Fig. 8

and Fig. 9. Table 7 and Fig. 10 have revealed that

rupture time and damage obtained from FE software

have a good agreement with the theoretical values

obtained from the subroutine [29].


1027

4.3 The Validation of Axisymmetric Problem

The validation of new software for the axisymmetric

problem was performed and it was conducted via a

uni-axial tension model in Fig. 11. A uniformly

distributed tensile force 0.375 kN/m2 was applied on

the top line of this uni-axial tension model.

Using the theoretical stress value into the multi-axial

version of creep constitutive equations and a 0.1 h time

step with Euler forward integration method, the

theoretical rupture time and damage can be obtained by

a subroutine [29] and the results are shown in Table 8.

The stress and displacement obtained from FE

software with the stress updating invoked due to creep

deformation are shown in Fig. 12 and Fig. 13.

The stress has been uniformly distributed in Fig. 14

and do agree with the theoretical values.

Other results are shown in Table 8 and Fig. 15.

Rupture time and damage obtained from FE software

have been revealed that have a good agreement with

the theoretical values obtained from the subroutine [29].

4.4 The Validation of 3D Problem

A preliminary validation of such software was

performed and it was conducted via a three-

dimensional uni-axial tension model in Fig. 16. The

length of a side is set to 1 m and a uniformly distributed

load 5 kN was applied on the top surface of this

uni-axial tension model.

The theoretical stress in Z direction is 5 kN/m2. The

stress in X and Y direction should be zero and these

stress values should remain the same throughout the

creep test up to failure. The stress obtained from FE

software with the stress update program is shown in

Table 9 at a 0.1 h time step with Euler forward

integration method.

Table 9 shows that the results obtained from the FE

software do agree with the expected theoretical values.

The stress involving creep deformation and stress

updating confirmed the uniform distribution of stresses,

and the values of stress in Z direction obtained from FE

software are correct, and the stress in X and Y direction

Fig. 10 The damage distribution on 180,230 h.

Fig. 11 The axisymmetric FE model.

Table 8 The theoretical rupture time and creep damage

for axisymmetric case.

Rupture time Creep damage

146,080 0.3333334

Fig. 12 Displacement distribution in Z axis.

Fig. 13 The displacement distribution in r axis.


1028

Fig. 14 Stress distribution in Z direction.

Fig. 15 Damage distribution on 143,060 h.

Fig. 16 The three-dimensional uni-axial tension model.

Table 9 The stress obtained from FE software with the

stress update program.

Integration point 𝜎𝑥 𝜎𝑦 𝜎𝑧

No. 1 0.1545E-05 0.1377E-06 0.5000E+01

No. 2 0.4970E-06 0.7690E-06 0.5000E+01

No. 3 0.1068E-05 0.2017E-07 0.5000E+01

No. 4 0.5675E-06 0.1478E-06 0.5000E+01

No. 5 0.1760E-05 0.3392E-06 0.5000E+01

No. 6 0.1212E-05 0.8395E-06 0.5000E+01

No. 7 0.6717E-07 0.8630E-06 0.5000E+01

No. 8 0.6380E-06 0.1648E-06 0.5000E+01

Table 10 The theoretical values and FE results.

The results Theoretical value FE results

Rupture time 98,046 93,540

Strain rate 0.000065438 0.000067522

Creep strain 0.179934333 0.182658312

Damage 0.33333337 0.33333334

is negligible.

The lifetime and creep strain at failure, and other

field variable can be obtained for the simple tensile

case illustrated above. They have been obtained by

direct integration of the uni-axial version of

constitutive equation for a given stress [30]. They have

also been produced by the FE software. Table 10 is a

summary and comparison of them.

Table 10 reveals that real values have a good

agreement with the theoretical values obtained from

the subroutine [30]. Work in this area is ongoing and

will be reported in future.

5. Conclusion

This paper is to present the finite element method

based on CDM to design FE software for creep damage

mechanics. More specifically, it summarizes the

current state of how to obtain such computational

capability then it concludes with a preference of

in-house software; secondly, it reports the development

of such software including the development of finite

element algorithms based on CDM for creep damage

analysis, and a flow diagram of the structure of new

finite element software has been completed to be

guided in developing in-house FE software, and the use

of some standard subroutines in programming; thirdly,

the development and the validation of the finite

element software conducted so far include plane stress,

plane strain, axisymmetric case, and 3D case were

reported.

Work in this area is ongoing and future development

work includes: (1) development and incorporation of

the new constitutive equation subroutines; (2)

intelligent and practical control of the time steps; (3)

removal of failed element and update stiffness matrix;

and (4) further validation. Further real case study to be

conducted and reported.

References

[1] J. Lemaitre, J.L. Chaboche, Mechanics of Solid Materials,

1st ed., Cambridge University Press, Cambridge, 1994.

[2] D. Liu, Q. Xu, Z. Lu, D. Xu, The review of computational


1029

FE software for creep damage mechanics, Advanced

Materials Research 510 (2012) 495-499.

[3] D. Liu, Q. Xu, D. Xu, Z. Lu, F. Tan, The validation of

computational FE software for creep damage mechanics,

Advanced Materials Research 744 (2013) 205-210.

[4] A.A. Becker, T.H. Hyde, W. Sun, P. Andersson,

Benchmarks for finite element analysis of creep

continuum damage mechanics, Computational Materials

Science 25 (2002) 34-41.

[5] D.R. Hayhurst, A.J. Krzeczkowski, Numerical solution of

creep problems, Comput. Methods App. Mech. Eng 20

(1979) 151-171.

[6] X. Ling, S.T. Tu, J.M. Gong, Application of

Runge-Kutta-Merson algorithm for creep damage analysis,

International Journal of Pressure Vessels and Piping 77

(2000) 243-248.

[7] Q. Xu, Creep damage constitutive equations for

multi-axial states of stress for 0.5Cr0.5Mo0.25V ferritic

steel at 590 °C, Theoretical and Applied Fracture

Mechanics 36 (2001) 99-107.

[8] L. An, Q. Xu, D. Xu, Z. Lu, Review on the current state of

developing of advanced creep damage constitutive

equations for high chromium alloy, Advanced Materials

Research 510 (2012) 776-780.

[9] Q.H. Xu, Q. Xu, Y.X. Pan, M. Short, Current state of

developing creep damage constitutive equation for

0.5Cr0.5Mo0.25V ferritic steel, Advanced Materials

Research 510 (2012) 812-816.

[10] Y. Tao, Y. Masataka, H.J. Shi, Numerical investigation on

the creep damage induced by void growth in heat affected

zone of weldments, International Journal of Pressure

Vessels and Piping 86(2009) 578-584.

[11] R. Mustata, R.J. Hayhurst, D.R. Hayhurst, F.

Vakili-Tahami, CDM predictions of creep damage

initiation and growth in ferritic steel weldments in a

medium-bore branched pipe under constant pressure at

590 °C using a four-material weld model, Arch Appl

Mech 75 (2006) 475-495.

[12] M.S. Herrn, G. Yevgen, Development of a creep-damage

model for non-isothermal long-term strength analysis of

high-temperature components operating in a wide stress

range, 1st ed., ULB Sachsen-Anhalt, Kharkiv, 2008.

[13] H. Altenbach, N. Konstantin, G. Yevgen, Numerical

benchmarks for creep-damage modelling, in: Proceeding

in Applied Mathematics and Mechanics, Bremen, May 22,

2008.

[14] H. Altenbach, G. Kolarow, O.K. Morachkovsky, K.

Naumenko, On the accuracy of creep-damage predictions

in thinwalled structures using the finite element method,

Computational Mechanics 25(2000) 87-98.

[15] M. Klasztorny, A. Bondyra, P. Szurgott, D. Nycz,

Numerical modelling of GFRP laminates with MSC. Marc

system and experimental validation, Computational

Materials Science 64 (2012) 151-156.

[16] H.C. Li, C.A. Tang, T.H. Ma, X.D. Zhao, Research on the

closure and creep mechanism of circular tunnels, Journal

of Coal Science and Engineering 14 (2008) 195-199.

[17] M.T. Wong, Three-dimensional finite element analysis of

creep continuum damage growth and failure in weldments,

Ph.D. Thesis, University of Manchester, UK, 1999.

[18] D.R. Hayhurst, R.J. Hayhurst, F. Vakili-Tahami,

Continuum damage mechanics predictions of creep

damage initiation and growth in ferritic steel weldments in

a medium-bore branched pipe under constant pressure at

590 °C using a five-material weld model, Proc. R. Soc. A

461 (2005) 2303-2326.

[19] R.J. Hayhurst, F.Vakili-Tahami, D.R. Hayhurst,

Verification of 3-D parallel CDM software for the analysis

of creep failure in the HAZ region of Cr-Mo-V crosswelds,

International Journal of Pressure Vessels and Piping 86

(2009) 475-485.

[20] K. Naumenko, H. Altenbach, Modeling of Creep for

Structural Analysis, Springer Berlin Heidelberg, New

York, 2004.

[21] F. Dunne, J. Lin, D.R. Hayhurst, J. Makin, Modelling

creep, ratchetting and failure in structural components

subjected to thermo-mechanical loading, Journal of

Theoretical and Applied Mechanics 44 (2006) 87-98.

[22] F.A. Leckie, D.R. Hayhurst, Constitutive equations for

creep rupture, Acta Metallurgical 25 (1977) 1059-1070.

[23] D.R. Hayhurst, P.R. Dimmer, G.J. Morrison,

Development of continuum damage in the creep rupture of

notched bars, Trans R Soc London A 311 (1984) 103-129.

[24] I.J. Perrin, D.R. Hayhurst, Creep constitutive equations for

a 0.5Cr-0.5Mo-0.25V ferritic steel in the temperature

range 600-675 °C, J. Strain Anal. Eng 31 (1996) 299-314.

[25] F. Tan, Q. Xu, Z. Lu, D. Xu, The preliminary development

of computational software system for creep damage

analysis in weldment, in: Proceedings of the 18th

International Conference on Automation & Computing,

Loughborough, Sep. 7-8, 2012.

[26] M. Dewar, Embedding a general-purpose numerical

library in an interactive environment, Journal of

Numerical Analysis 3 (2008) 17-26.

[27] I.M. Smith, D.V. Griffiths, Programming the Finite

Element Method, 4th ed., John Wiley & Sons Ltd., Sussex,

2004.

[28] Q. Xu, M. Wright, Q.H. Xu, The development and

validation of multi-axial creep damage constitutive

equations for P91, in: ICAC'11: The 17th International

Conference on Automation and Computing, Huddersfield,

Sep. 10, 2011.

[29] F. Tan, Q. Xu., Z. Lu, D. Xu, D. Liu, Practical guidance on

the application of R-K-integration method in finite


1030

element analysis of creep damage problem, in: The 2013

World Congress in Computer Science, Computer

Engineering, and Applied Computing, Nevada, July.

22-25, 2013.

[30] D. Liu, Q. Xu, Z. Lu, D. Xu, Q.H. Xu, The techniques in

developing finite element software for creep damage

analysis, Advanced Materials Research 744 (2013)

199-204.


The Global Crisis and Academic Communication: The

Challenge of Social Networks in Research

Sandra Martorell and Fernando Canet

Department of Media Communication, Information System and Art History, Polytechnic University of Valencia, Valencia 46022,

Spain

Received: June 21, 2013 / Accepted: July 30, 2013 / Published: August 31, 2013.

Abstract: The global economic crisis is seriously affecting academic research. The situation is provoking some big changes and an urgent need to seek alternatives to traditional models. It is as if the academic community was reinventing itself; and this reinvention is happening online. Faced with a lack of funding, researchers have determined to help each other develop their projects and they are doing so on social knowledge networks that they have created for this mission. The purpose of this paper is to analyze different social networks designed for academic online research. To this end, we have made a selection of these networks and established the parameters for their study in order to determine what they consist of, what tools they make use of, what advantages they offer and the degree to which they are bringing about a revolution in how research is carried out. This analysis is conducted from both a qualitative and a quantitative perspective, allowing us to identify the percentage of these networks that approach what would be the ideal social knowledge network. As we will be able to confirm, the closer they are to this ideal, the more effective they will be and the better future they will have, which will also depend on the commitment of users to participation and the quality of their contributions. Key words: Academic social networks, Web 2.0, research, participatory knowledge.

1. Introduction

“It is a change of epoch, a change of era. Many

things are changing, both in public life and in private

life. The mentalities of the people are changing too. I

believe that it is a change similar to what Europe went

through in the shift from the Middle Ages to the

Renaissance, except that then it took a century and

now we are going through it in just two or three

decades. We are experiencing a change of coordinates,

of mentality and of sensibility.” These are the words

of Professor Emeritus in Sociology Amando de

Miguel Rodriguez [1] in reference to the economic

crisis that we have been experiencing since the

collapse of Lehman Brothers Holdings in 2008.

Many countries, especially in Europe, are facing a

Corresponding author: Sandra Martorell, Ph.D. candidate,

research field: media communication, E-mail: [email protected].

period of huge changes, brought about largely by the

economic cutbacks that they have been subjected to.

One sector affected by the devastation arising from

the current crisis is the scientific and academic

community. This has been made clear by scientists

themselves in texts such as the open letter signed by

42 Nobel Prize and Fields Medal winners to the heads

of state and government of the European Union,

expressing the idea that science is fundamental for

progress [2]. In the face of the crisis, while continuing

to call for greater investment, many scientists have

diligently gone on pursuing their work by all means

available, one such means being the Internet, where

they have begun working in groups through social

networks. These are not general social networks like

Facebook or Twitter, but social networks created by

and for researchers where they can exchange

knowledge. This gives them, in addition to the usual

The Global Crisis and Academic Communication: The Challenge of Social Networks in Research

1032

resources, tools that serve to facilitate their everyday

research activities, which can be summed up in three

basic tasks: communicate, collaborate and share

(hereinafter referred to as “CCS”).

These three functions together allow researchers to

use these networks to work in groups, help each other,

and engage in group discussion. In this way, through

shared research, other researchers or academics can

take over a research project so that it can progress

exponentially, or so that new avenues of study can be

opened up. This has resulted in a constant increase in

articles and other publications, a worldwide scientific

revolution that has been possible in part thanks to this

kind of network in which researchers commit to

thinking collectively, as Levy suggests in a clear

reference to Descartes, from the perspective of

cogitamus (“we think”) rather than cogito (“I think”).

From this we can see a clear relationship between

the changes in researcher practice and technology,

specifically ICTs (information and communication

technologies). The concept of ICT refers to the set of

technological tools that allow us to access information

and share it with others [3]. Thanks to these tools,

relationships with knowledge sources have increased

and individuals are now able to communicate with

each other in a different way, which in turn has

changed traditional conceptions of communication of

and access to knowledge [4]. But it is not simply that

these new technologies have facilitated advances in

this sense, but that the change is being brought about

by the volition of thousands of users. In other words,

technology alone can not force people to participate

against their will; however, for those who are willing,

it can provide the environment necessary to facilitate

collaboration and communication [5].

Evidence of this can be found in the concept of the

collaboratory, a term coined by former UNESCO

Director-General 1 Koichiro Matsuura, which

combines the words “collaboration” and “laboratory”.

The concept defines the combination of technology,

1From 1999 to 2009.

instruments and infrastructure that allows scientists to

work with remote facilities and other colleagues as if

they were located in the same place and with effective

interface communication [6]. As Jane Russell points

out in Ref. [7], these “centres” without walls’ are

associated with a new paradigm in scientific practice

that gives researchers in any field easy access to

people, data, instruments and results; a kind of virtual

research lab which, judging by the figures provided by

the National Science Board, represents a significant

challenge to traditional research methods that has been

growing and gaining force gradually for a few decades:

from 1981 to 1995, the number of articles with more

than one author increased by 80% and the number of

articles based on international collaboration increased

by 200%, while there was a total increase in the

production of articles of 20% [7]. These data make it

clear that the first collaborative applications in the

field of research focused on speeding up and enriching

the process of writing scientific articles, as a direct

consequence of the adaptation of scientific production

methods to the new digital environment [8].

Today this is even more evident and relations

between researchers working in the same field in

different parts of the world have intensified thanks to

Web 2.0. Also known as the social web, this network

is based to a large extent on interactive relations open

to Internet surfers who want to participate in

communicative processes of production,

dissemination, reception and exchange of all kinds of

files [9], an activity that finds its finest expression in

social networks.

Social knowledge networks are also collaboratories,

serving as a meeting and discussion point where users

can work collectively. Moreover, online social

networks in general, as Flores-Vivar suggests in Ref.

[10], are the flagship of Web 2.0. The combination

these two aspects—their importance within the web

universe and their capacity to put members of the

academic community in contact with each

other—make them a powerful tool driving a new


1033

revolution in knowledge that is bringing about an

epistemological paradigm shift. To highlight this

change we have decided to conduct a study based on

the analysis of different social knowledge networks

that connect researchers from all over the world. The

results of this project are outlined in this article, which

we have organized as follows:

First of all, we will discuss the state of the question

in order to contextualize the study. To do this, we will

offer an overview of social knowledge networks and

the different types thereof in the context of Web 2.0.

We will then establish the methodology and the

different parameters for analysis that led to the series

of results presented under the heading Analysis and

results.

Following this, the final section will set forth the

general conclusions of this study, which aim to cover

the following objectives:

to establish an experience-based definition of the

academic social networks created on the Internet;

to list the main characteristics of these types of

networks;

to examine the basic principles underpinning

such networks;

to highlight their potential;

to identify their deficiencies or weak points and

the importance of correcting them in the interests of

ensuring their successful future development.

2. State of the Question

Social knowledge networks arise out of the

academic community’s need to reinvent itself and to

find new ways of ensuring its survival and evolution

even in the hardest times.

They form part of what is known as Science 2.0, a

term that covers the whole range of applications and

platforms designed to help scientists in their daily

activities, offering them different tools to manage

their work flows, facilitate the search for pertinent

information or provide them with new ways of

communicating their findings [8]. The concept

therefore includes networks of scientific blogs, 2.0

journals and reference managers, as well as the

academic social networks that are our object of study.

There are many different names for these networks,

which, apart from bringing together researchers from

all over the world, are focal points of constant creation

and shared development of knowledge. What we refer

to here as knowledge networks2 other authors call

research networks or academic social networks. Their

essential priority is to communicate and disseminate

scientific information, seeking to reach a large number

of readers, and to this end they make use of the web,

so that through a message or a link or a file

attachment, information can be shared with all their

members [11].

In Ref. [12] Garcia-Aretio attributes to these

networks the objectives of sharing, co-creating and

building knowledge through their relations and

communication exchanges, while for Salinas et al. [13]

the basic principles are information exchange and an

adequate flow of information which, according to

these authors, depend on accessibility, the culture of

participation, collaboration, diversity and sharing that

condition the quality of life of the community, the

communication skills of their members and the

relevant content. For Sanudo [14], central to their

activities are knowledge production, resource

management and achieving results geared towards

innovation, among others.

Some networks of this type outline their own

definition, such as ResearchGate, which does so using

the graphic explanation shown in Fig. 1.

These are different ways of referring to the same

functions or objectives, the aforementioned CCS, key

elements underpinning these kinds of networks for

which, based on our analysis, we have established our

own definition:

“Academic social knowledge networks are a

meeting point for researchers from all over the world, 2 A concept coined decades ago but that has now been consolidated with the arrival of Web 2.0 and online social networks.


1034

Fig. 1 Diagram of the three pillars that define ResearchGate.

who join forces in an effort to advance their studies on

the basis of three basic principles: communication,

collaboration and sharing their knowledge in a

democratic virtual environment that is optimal for

dissemination provided there is a commitment to

participation and a faithfulness to academic rigour.”

These networks have two different types of

idiosyncrasies: the first relates to the topic they

address, and the second to their operating policy. With

regard to the first, two basic types can be identified:

general networks and specialist networks. General

networks cover a more diverse range of disciplines,

allowing for interdisciplinary exchange on a single

platform, thereby fostering transversality of

knowledge.

Specialist networks, as their name suggests, focus

on specific fields, although the degree of specificity

may vary (ranging from fields as broad as the social

sciences to others limited to the study of history or

even further to the history of a particular discipline,

movement or period).

In terms of operating policy, we are particularly

interested in addressing the question of whether the

networks are free or require payment of a subscription

fee to gain access.

In this regard we have aimed to take samples of

both categories, although we have considered

dedicating special attention to free or open access

networks, which are based on a philosophy that is

becoming increasingly predominant, fostered to a

great extent by those voices calling for the publication

of raw data compiled in publicly funded research [8].

Open access is a movement that advocates free

access to scientific or academic online resources,

which should not be restricted by any impositions

other than technological limitations or the Internet

connection of the user [15]. The resources may

therefore be downloaded, read, distributed and

otherwise used in accordance with the licence, which

includes what is normally referred to as Creative

Commons, one of the more common systems for open

access publication, encompassing diverse categories

depending on the restrictions applicable, such as

author acknowledgement, non-commercial use or a

prohibition on modifications to the work.

Open access is a philosophy whose basic principles,

according to Tapscott [16], are collaboration,

transparency, sharing and empowerment. It has now

become a viable option endorsed in international

declarations that seek to define the concept, such as

the Budapest Open Access Initiative signed in 2002,

the Bethesda Statement on Open Access Publishing in

June 2003, or the Berlin Declaration on Open Access

to Knowledge in the Sciences and Humanities in

October 2003.

These declarations and others that have followed

them uphold the need to promote the principle of open

access, based on the idea that if we can make the best


1035

use of information technologies we will be able to

expand distribution capacity while reducing costs in

order to provide wider and easier access to research

results, thanks to the advantages offered [17], which

are:

The cost is low and the results can have a big

impact in a short period of time, facilitated to a large

extent by the viral nature of the Internet, as well as the

reduction of time needed for the evaluation and

publication process compared to the time needed to

produce a print publication;

The results obtained can be compared with other

previously published results, or the data can be reused

for further research without the need for a new

investment, which constitutes a vital advantage for

small research groups with limited resources.

Added to the above is the fact that all scholars in a

discipline will have equal access to the information

provided they have internet access without censorship

or government restrictions, thereby liberating research

from the constraints of intellectual inbreeding to open

it up to the world in the interests of development

fostered by the “collective intelligence”, meaning

simply “a form of universally distributed intelligence,

constantly enhanced, coordinated in real time, and

resulting in the effective mobilization of skills” whose

basis and objective is the “mutual recognition and

enrichment of individuals rather than the cult of

fetishized communities in hypostasis” [18].

In this regard, we could also cite Bailon-Moreno et

al. (quoted in Ref. [8]) in relation to the Ortega

hypothesis, according to which scientific progress is

based on the minimal contributions of a multitude of

scientists. Because, as will be shown below, these

types of networks can only function positively with

the commitment of users, who collectively form what

Surowiecki analysed in The Wisdom of Crowds [19]

or Rheingold in Smart Mobs [20] and to which Cobo

Romaní and Pardo Kuklinski refer in Ref. [21] as a

form of knowledge that is more valuable when

multiplied because, according to the authors, shared or

distributed knowledge is on average much more

effective and accurate than the knowledge that may be

produced by the most acclaimed or accomplished

expert.

3. Materials and Methods

We apply a methodological system based, on the

one hand, on the theories proposed by the authors

mentioned above, and on the other, on a qualitative

study for which a series of analysis criteria have been

established through the comparison of different

platforms of the same kind.

To conduct this study, we have first made a

selection of the knowledge networks to be analysed.

The basic premise has been that they need to be

networks whose mission is to bring the academic

community together, and that have a marked social

character3, i.e., they allow dialogue by connecting

users to each other. In addition to this, we have had to

distinguish between two types of networks of this kind:

general networks on one hand and, on the other,

networks focused on a specific field.

For general networks, the selection has been made

taking into account the number of users registered and

the quantity of documents stored, and considering

Metcalfe’s Law, according to which the value of a

network increases in proportion with the square of the

number of system users (n2), which Foglia [22] shows

using the graph in Fig. 2.

Fig. 2 Metcalfe’s law.

3Taking advantage of the resources offered by Web 2.0.


1036

We therefore chose three basic networks:

ResearchGate (2.2 million users and 35 million

documents), Academia.edu (2,201,270 users and

1,661,926 documents as of February 6, 2013) and

Mendeley (2,153,818 users and 351,357,178

documents as of February 8, 2013). The supremacy of

these networks is also reflected by their media

exposure and the interest that investors have taken in

them, as well as awards received. Evidence of this is

the space dedicated to Mendeley on the blogs of the

Wall Street Journal, Tech Europe, and The Guardian,

which rated it at number 6 among the “Top 100 Tech

Media Companies” [23], and awards such as

“European Start-up of the Year 2009” [24] and “Best

Social Innovation Which Benefits Society 2009” [25].

In terms of the interest that these kinds of networks

arouse outside the academic community, it is worth

noting that ResearchGate benefits from powerful

investors such as Founders Fund, and from

collaborations with Benchmark Capital, Accel

Partners and others such as Michael Birch and David

O. Sacks, who trust in the network’s potential, as

clearly expressed by Luke Nosek, Founders Fund

coordinator and partner [26]: “We have a genuine

appreciation for the considerable success that the team

at ResearchGate has demonstrated since the company

was founded. We truly believe that the network has

the potential to disrupt a much-outdated system”.

For specialist networks, the selection criteria have

been different. There are networks of this kind

associated with a wide range of disciplines, with some

of the most prolific fields being those related to the

natural sciences. These include the networks Biomed

Experts, Epernicus, Scilife and Nature Work, and

many other networks with large numbers of users that

have been the subject of numerous studies. There are

others, however, which to date have not had so much

visibility, such as those associated with the social

sciences, which are the very networks we have

determined to focus our attention on given their

increasing proliferation and the lack of articles

studying and analysing them, despite the fact they

constitute a substantial change in terms of the

knowledge models used in their different research

areas.

Of these we have selected five for their affinity with

our field of study, which is essentially the field of

communication. We have therefore focused on the

following networks: Social Science Research Network

(hereinafter SSRN), H-net, ECREA, NECS and Portal

de la Communication.

We have thus made a selection of eight (three

general and five specialist) networks for study using a

qualitative analysis, for which we have established a

series of variables (a total of 70) grouped into five

categories, which in turn are broken down into more

specific subcategories, allowing us to extract the

characteristics not only of the networks but also of the

users who participate and their content, and to

determine their nature, what they offer and how they

contribute to communication and exchange, among

other aspects. These five categories are outlined

below:

(1) General parameters: This section offers a

general idea of the network, both with regard to its

size and to the basic characteristics that define it, such

as the type of users it targets, the geographical regions

it covers and its objectives (plus eleven other

parameters).

(2) User data: This section is made up of

twenty-two items consisting of the fields to be filled

in every time a new registration is completed. This

allows us to determine the type of information that

this kind of network considers relevant for the

creation of user profiles.

(3) Services and resources: This is a list of 28

actions and resources that determine the possibilities

that network users have, ranging from conducting

searches to the option of contributing files or creating

work groups. Many of these features originate from

conventional social networks, such as the use of a wall

or chat function, but there are also others that are


1037

highly useful to academics, such as repositories for

storing users’ documents and consulting the

documents of other users, bookmarking, and the

facility to create quotes or links to scientific or

academic databases. This section also determines the

involvement of the network and its tools and resources

in the achievement of CCS, which are the

fundamental pillars for this kind of network.

(4) Content: This section allows us to analyse the

kind of files stored on the network and the nature of

their organization or access (whether you need to be a

registered user to view them, whether they can be

downloaded or whether all or only a part of the

information stored is accessible).

(5) Miscellaneous: Here we include other types of

data that did not fit into previous sections but that are

of relevance.

Upon completion of the qualitative analysis based

on the parameters encompassed by each category, we

have sought to extract a numeric representation of the

data through the use of percentages. Our aim is to

confirm, on the basis of a figure, the extent to which

each network conforms to our concept of knowledge

networks, irrespective of whether they are general or

specialist networks.

We have not been able to determine this from the

initial parameters, as among the seventy that we have

established there are many that have no special

relevance or are descriptive in nature and therefore not

applicable for this purpose. Thus, based on our ideal

conception of knowledge platforms, we have made a

selection of the 25 most important aspects that define

them, as shown in Table 1, giving each one a value of

four points4, i.e., 4% of the total.

4. Analysis and Results

Based on the 25 parameters established and after

conducting the quantitative analysis, we obtained the

results summarised in Table 2, regarding the degree to

which the networks studied conform to the ideal for

425 parameters with a value of 4% each = 100% of the total.

participatory knowledge networks developed on the

Internet by collectives of researchers and

academics:

The figures show that the general networks

conform more closely to the idea that we have of a

knowledge network than the specialist networks, with

ResearchGate (which is also the most popular)

standing out above the rest. This may be due to the

fact that because it has the largest number of users and

the highest user participation, it is able to monitor

actual user needs more dynamically and adapt the

network accordingly. Another determining factor is a

network’s international character; we therefore

especially take into account the languages in which it

is established, which as a general rule is English. The

one exception is Portal de la Communication, which

has opted for Spanish and Portuguese, which thus,

despite not operating in English like the others, also

expands its potential by reaching beyond national

borders. As can be seen, this platform is located at the

halfway point towards the ideal and is designed more

as a portal than a network as such, although we have

decided to include it because of its uniqueness, the

work it performs, and its marked social character,

which bring it closer to our idea of a knowledge

platform.

In terms of user fees, as noted above we have

sought a mixture of options. The three general

networks studied offer free access, unlike some of the

specialist networks such as ECREA and NECS, both

of which finished in last place, below those without

user fees. This makes it clear that the option of open

access is viable, and that there is no reason that the

quality of the platform will be lower if payment is not

required, but rather that free networks can be just as

sustainable. Moreover, the platforms analysed (both

general and specialist) that do not charge user fees

have more users (while NECS has around 1,100 users

and ECREA has 3,500, Social Science Research

Network reports more than 1.3 million and H-net

more than 100,000).


1038

Table 1 Important aspects for defining a knowledge platform.

Participation on social networks

Communication with users

Communication between users

Global character Follow/be followed

Free to users Search engine Subscription to topics of interest

Upload files Download files

Invite contacts Citation Creation of work groups Share links Wall

Chat Forum User recommendation Sending updates Repository

Calendar of events Job offers Statistics News Bookmarking

Table 2 Percentage of conformity to ideal for online knowledge networks.

General networks

ResearchGate 84%

Academia.edu 75%

Mendeley 75%

Specialist networks

SSRN 61%

H-net 52%

Portal de la Communication 49%

ECREA 39%

NECS 33%

In this respect, several aspects should be considered:

On the one hand, the wider the network’s field of

study, the more users will join, which in itself places

NECS and ECREA at a disadvantage due to their very

narrow focus (the first is the European Network for

Cinema and Media Studies and the second is the

European Communication Research and Education

Association), something that may be favorable for

certain researchers not seeking transversality between

disciplines but instead wishing to focus on a specific

field. On this basis, it is clear that they have fewer

users, while others like SSRN with many more users

cover the wide range of all the social sciences.

On the other hand, it is true that many of the users

registered on these networks are not willing to pay,

either because initially they will only be exploring and

getting to know the platform and refuse to pay for

something that they are not certain they will benefit

from, or because they are in favour of the philosophy

of open access, or perhaps even because they are

reluctant to pay for certain services online. In this

sense, we find that often the number of users is not

representative of the use of the network, since many

users registered on a network do not engage in any

activity on it. This tends to occur more often on the

networks with no user fees, where many register to try

it out but soon stop using it. On networks with user

fees, however, people may think it over more

carefully but if they ultimately decide to register it is

because they are truly convinced or at least have the

intention to use the network. As a result we find that

although they may have fewer users, the users they

have may participate more than users on free access

networks.

Indeed, low participation is one of the issues that

most severely afflict these types of networks in

general, constituting one of their most common weak

points. Thousands of registered users do not

participate, or if they do, they often abandon the

network to a certain degree once they have covered

their information needs and make no new

contributions. We can affirm that only a portion of

registered users participate actively and with a certain

degree of regularity in the achievement of CCS.

However, for the network to function properly

participation is essential, because to truly build

knowledge in virtual environments, according to No

Sanchez [27], the conditions of active commitment,

participation, frequent interaction and connection with

the real world need to be met, a point also underlined

by Arriaga Mendez et al. [11], who argue that the

meaning and objectives of a network will only be

made a reality through the work of the participants.

We therefore need to ask what the low participation

of certain groups of users could be due to. There may

be various reasons for the reluctance of researchers to

participate in these networks [8]. One factor may be


1039

the highly competitive nature of scientific work,

which fosters a certain degree of discretion in the

dissemination of results until those results are

published by conventional means. Another factor may

be the age of the researchers, i.e., the fact that the

more established researchers do not tend to be so

familiar with the Internet and the new possibilities it

offers, and prefer traditional methods, a situation that

nevertheless is changing thanks to the up-and-coming

generations of academics who have grown up with

ICTs and who apply them in practically all spheres of

action, both personal and professional.

Another aspect is the fact that there are knowledge

networks where there is total freedom to post content,

without the need for that content to undergo any type

of review process, the most common type being peer

review. While it is true that there are networks that do

include a review requirement, such as H-net and

SSRN, on others there is no filter whatsoever; this,

rather than favouring collective progress, is actually

harmful to it, given the hazard to scientific rigour

constituted by the possible inclusion of erroneous

information. Also this in a way keeps researchers

from publishing freely [28], as any contribution not

submitted to the scrutiny of their peers is always

under suspicion. Moreover, any unreviewed

publication would most probably not be taken into

account in the evaluation processes to which

researchers are submitted.

Of course, the review process does not guarantee

total accuracy of information, as we have seen in

cases such as that of Woo Suk Hwang, who published

a fraudulent scientific finding in the journal Science in

2005, and which the publication subsequently

withdrew, or Alan Sokal and Jean Bricmont’s book

Fashionable Nonsense [29], in which, to expose the

cultural relativism and confusing and pretentious use

of scientific terms by some intellectuals, the authors

revealed that they succeeded in publishing a farcical

article in the journal Social Text [30]. This

demonstrates the fact that reviews, and thus the filters

established to ensure maximum reliability, sometimes

fail, but at present they are the forms of legitimation

that are most widespread and commonly considered to

be the most reliable, and we therefore can not sidestep

them, either for journals or for the knowledge

networks that concern us here, which they endow with

scientific rigour, trustworthiness and prestige.

5. Conclusions

A Spanish newspaper has asserted that “things are

as bad now as in the worst moments of Spanish

history” [31]. Nevertheless, crisis and change always

go hand in hand. The current crisis is no exception,

and while it affects many sectors of the population,

those sectors will try to survive it however they can.

This is true of the academic community, which is

gradually embracing the idea that together we can

move forward.

To this end, academics are making use of the

resources available, including new tools that enable

them to publish and share their knowledge with a

great advantage over the conventional tools used in

the past [32].

Most of these tools are available on the Internet,

such as the social knowledge networks designed for

the academic community. These networks have been

developing for years but now more than ever have the

potential to become a fundamental resource for

research, not only at the national level but globally,

given that the current crisis is not only affecting Spain

but the whole world.

These networks did not appear with the crisis, but

they can help to make the crisis more bearable as they

offer a multitude of possibilities for communication

and exchange of knowledge.

To this end, they offer a series of resources and

services that have been developed through the

application of the advantages of Web 2.0 to the field

of research, such as work and collaboration online, the

creation of interest groups, communication via chats

or other types of messaging, and the possibility of


1040

document sharing.

In this way, these knowledge platforms or networks

have the virtue of offering two basic benefits,

especially those that are open access:

They benefit participants individually, as we

must not forget that sharing research data publicly can

have a positive effect on citation [33], thereby

contributing to an increase in productivity and in

impact;

They benefit society in general, given that,

according to the theories of Avalos [34] and Aguilera

[35], research and education constitute the

cornerstones of the economic policy of developed

nations. Toffler suggests something similar in arguing

that knowledge is the central element of our society

today. In this context the search for knowledge guides

our actions, is the source for the production of goods

and services, and the means that allows us to pursue

greater development [36].

We see the potential of these networks as lying in

the fact that they allow academics to develop

professionally while also pursuing the good of the

public in general, both inside and outside the

academic world.

To this end, the agents who participate in these

networks are at once apprentices and masters,

contributing their own experience and benefiting from

the experience of others, so that traditional

hierarchical structures give way to collaborative work,

shared leadership, participation and coordination [37].

It should be noted, however, that all these synergies

are based on an ideal conception of these networks.

We conceive of a dynamic and constant exchange

between all members of information that is checked,

analysed in depth, in a reliable and thorough manner,

which is not always the case.

In view of the above, we can conclude that this new

research model is currently in an incipient phase and

still needs to develop and mature, especially in terms

of the quality and indexing of content, as well as the

raising of awareness of the importance of advancing

together, because only in this way, united in practice,

can we ensure the dynamic and stable development of

research, without barriers and as a collective.

Acknowledgments

The research for this article was enabled with the

support of the Research Project “Study and analysis

for development of Research Network on Film

Studies through Web 2.0 platforms”, financed by the

National R + D + i Plan of the Spanish Ministry of

Economy and Competitivity (code HAR2010-18648).

References

[1] D. Zamora [Online], http://www.lne.es/sociedad-cultura/2012/09/11/vivimos-crisis-cambio-apenas-decadas/129365.html (accessed Apr. 21, 2013).

[2] D. Ahlstrom [Online], http://www.irishtimes.com/newspaper/ireland/2012/1023/1224325578288.html (accessed Apr.05, 2013).

[3] A. Solano [Online], http://www.slideshare.net/ppitufo/tic-y-web-20 (accessed Apr. 08, 2013).

[4] J.E. Reyes Iriarte, Conocimiento y tecnología; una relación esencial en la reinvención humana (Knowledge and technology, an essential link in the human reinvention), Akademeia 5 (2010) 32-37.

[5] National Science Board, Science and Engineering Indicators, US Government Printing Office, 1998.

[6] P. Glasner, From community to “collaborator”? The human genome project and the changing culture of science, Science and Public Policy 23 (1996) 109-116.

[7] J.M. Russell, Scientific communication at the beginning of the 21st century, International Social Science Journal 168 (2001) 271-282.

[8] A. Cabezas-Clavijo, D. Torres-Salinas, E. Delgado-Lopez-Cozar, Ciencia 2.0: Catálogo de herramientas e implicaciones para la actividad investigadora (Science 2.0: Catalog of tools and implications for research activity), El profesional de la información (The Information Professional) 18 (2009) 72-79.

[9] M. Cebrián Herreros, La Web 2.0 como red social de comunicación e información (Web 2.0 as a social network of communication and information), Estudios sobre el Mensaje Periodístico (Post Journalism Studies) 14 (2008) 345-361.

[10] J.M. Flores-Vivar, Nuevos modelos de comunicación, perfiles y tendencias en las redes socials (New


1041

communication patterns, profiles and trends in social networks), Comunicar 33 (2009) 73-81.

[11] J.A. Méndez, M.G.M. Jiménez, M.L.P. Cervantes, Retos y desafíos de las redes de investigación (Challenges of research network), Revista Iberoamericana Sobre Calidad, Eficacia y Cambio en Educación (Iberoamerican Journal on Quality, Efficiency and Change in Education) 10 (2012) 178-183.

[12] L. García-Aretio, Redes y comunidades (Networks and communities), Comunicación y Pedagogía (Communication and Education) 223 (2007) 28-33

[13] J. Salinas, A. Pérez, B. de Benito, Metodologías Centradas en el Alumno para el aprendizaje en Red (Learner-centered Methodologies for Networked Learning), Síntesis, 2008.

[14] L. Sañudo, El papel de las redes profesionales de investigación e un mundo globalizado (The role of research professional networking in a globalized world), Revista Iberoamericana sobre Calidad, Eficacia y Cambio en Educación (Revista Iberoamericana on Quality , Efficiency and Education)10 (2012) 136-143

[15] P. Suber, An Introduction to Open Access [Online], http://bibliotecavirtual.clacso.org.ar/ar/libros/secret/babini/Peter%20Suber.pdf (accessed May 27, 2013).

[16] D. Tapscott, Four principles for the open world [Online], http://www.ted.com/talks/don_tapscott_four_principles_for_the_open_world_1.html (accessed Feb. 1, 2013).

[17] J. Alonso, I. Subirats, M.L. Martínez Conde, Informe APEI sobre acceso abierto (AOSIS on Open Access), Asociación profesional de especialistas en información (Professional Association of Information Specialists), 2008.

[18] P. Levy, Collective Intelligence: Mankind’s Emerging World in Cyberspace, Basic Books, 1999.

[19] J. Surowiecki, The Wisdom of Crowds, Anchor Books, 2004.

[20] H. Rheingold, Smart Mobs, Editorial Gedisa, S.A, 2004 [21] C.C. Romaní, H.P. Kuklinski, Planeta Web 2.0,

Inteligencia colectiva o medios fast food (Planet Web 2.0. Collective Intelligence or Means Fast Food), Grup de Recerca d'Interaccions Digitals Universitat de Vic (Digital Interactions Research Group UVic), 2007.

[22] G. Foglia, [Online], http://www.iprofesional.com/notas/87619-Como-se-comportan-las-redes-sociales-segun-las-leyes-de-Moore-y-Metcalfe.html (accessed May 11, 2013).

[23] The top 100 tech media companies, [Online], http://www.guardian.co.uk/tech-media-invest-100/top-100 (accessed Feb. 14, 2013).

[24] R. Wauters, [Online], http://plugg.eu/media/blog/p/detail/winners-for-plugg-star

t-ups-rally-2009-announced (accessed May 22, 2013). [25] M. Butcher, The europas: The winners and finalists,

[Online], techcrunch.com/2009/07/09/the-europas-the-winners-and-finalists/ (accessed Jan. 22, 2013).

[26] ResearchGate: Founders Fund invests in science-network, [Online], http://www.eu-startups.com/2012/02/researchgate-founders-fund-invests-in-science-network/ (accessed Jan. 22, 2013).

[27] J.N. Sánchez, Comunicación y construcción del conocimiento en el nuevo espacio tecnológico (Communication and knowledge construction in the new technology space), Revista de universidad y sociedad del conocimiento (Journal of College and the knowledge society) 5 (2008) 2-3.

[28] D. Torres-Salinas, El paradigma 2.0 en las grandes revistas científicas (The paradigm 2.0 in major scientific journals), in: 3rd International LIS-EPI Meeting, Valencia, 2008.

[29] A. Sokal, J. Bricmont, Fashionable Nonsense: Postmodern Intellectuals’ Abuse of Science, Picador, New York, 1998.

[30] T. Delclòs, Scientific journals and reliability [Online], http://elpais.com/elpais/2013/03/02/opinion/1362219068_768198.html (accessed Mar. 03, 2013).

[31] E. Capurro, The current economic crisis reached the worst levels of the recession of the 90 [Online], http://www.abc.es/20120608/economia/abci-fedea-crisis-comparacion-201206071755.html (accessed Nov. 30, 2012).

[32] S. Biro, Miradas Desde Afuera: Investigacion Sobre Divulgacion, Universidad Nacional Autónoma de México, 2007.

[33] H.A. Piwowar, R.S. Day, D.B. Fridsma, Sharing detailed research data is associated with increased citation rate, PLoS One 2 (2007) 308.

[34] I. Avalos, La investigación universitaria en tiempos de la sociedad del conocimiento, Revista Venezolana de Economía y Ciencias Sociales 11 (2005) 89-105.

[35] M. Aguilera, La Investigación Universitaria: indispensable en los nuevos tiempos, Universitas 29 (2000) 11-12.

[36] J.G. Dominguez, R.D. Monaco, L.G. Garcia, et al., La investigación universitaria como eje de la transferencia social del conocimiento (University research as the center of social transfer of knowledge), Publicaciones en Ciencia y Tecnología (Publications in Science and Technology) 6 (2012) 41-51.

[37] J. Gairín, Las comunidades virtuales de aprendizaje, Educar 37 (2006) 41-51.


Can You Explain This? Personality and Willingness to

Commit Various Acts of Academic Misconduct

Yovav Eshet1, Yehuda Peled2, Keren Grinautski1 and Casimir Barczyk3

1. School of Management, University of Haifa, Haifa 3190501, Israel

2. Department of Education, Western Galilee College, Akko 24121, Israel

3. School of Management, Purdue University Calumet, Hammond IN 46323, United States


Abstract: The rapid development of IT has created a problematic situation in higher education by providing individuals with a greater opportunity to engage in academic dishonesty especially in online courses, in contrast to traditional classroom courses. There are various factors that were used in research to explain the phenomenon of academic dishonesty. Among them are personality traits that were found to be effective in explaining unethical behaviors. Therefore, this study explores students’ personality traits as predictors of academic dishonesty in the context of traditional and distance-learning courses in higher education. Data from 1,365 students enrolled in academic institutes in the U.S.A and Israel were surveyed to assess their personality and their willingness to commit various acts of academic misconduct. The findings indicate that in both countries dishonest behaviors are greater in face-to-face than in online courses. In addition, both American and Israeli students identified with the personality trait of agreeableness showed a negative correlation with academic dishonesty. Furthermore, Israeli students identified with the personality traits of conscientiousness and emotional stability demonstrated a negative correlation with academic dishonesty. In contrast, the personality trait of extraversion among American students was positively correlated with academic misconduct. Implications for further research are discussed.

Key words: Academic dishonesty, personality traits, OCEAN, online courses.

1. Introduction

The concept of academic dishonesty is frequently

addressed in research along with various factors that

serve to explain the phenomenon. For instance, the

rapid development of IT has created a problematic

situation because the lessening of personal contact

between students and faculty provides individuals

with a greater opportunity to engage in academic

dishonesty [1, 2]. Therefore, online courses, in

contrast to traditional classroom courses, may

contribute to a higher incidence of academic

dishonesty among students by making them feel

“distant” from others [3-6]. In addition, scholars

suggest that students’ perception of what constitutes

Corresponding Author: Yovav Eshet, Ph.D., research fields: academic dishonesty and on-line education. E-mail: [email protected].

cheating has changed. For instance, working together

on a “take home” exam is considered “postmodern

learning”, and text-messaging answers is not

considered cheating by some students [7]. Another

important factor that might influence students’

tendency to engage in academic misconduct is related

to various personality traits.

This paper investigates the ubiquitous and

somewhat universal concept of academic dishonesty.

While the precise definition of the term has not yet

crystallized, we choose to define it as “forms of

cheating and plagiarism that involve students giving

or receiving unauthorized assistance in an academic

exercise or receiving credit for work that is not their

own” [8]. This definition is sufficiently broad to

include behaviors such as numerous forms of cheating,

intentional and non-intentional plagiarism,

Can You Explain This? Personality and Willingness to Commit Various Acts of Academic Misconduct

1043

falsification, bribery and collusion.

The paper is organized as follows: Section 2

discusses the research literature review regarding

Students’ personality as a predictor of academic

dishonesty. Section 3 introduces the research method.

Section 4 presents the results. Section 5 gives

conclusions.

2. Students’ Personality as a Predictor of Academic Dishonesty

There are few research studies linking unethical

behavior and personality traits, but each study uses a

different measure of dishonesty. Hence, the results are

often contradictory [9-11]. Although measures based

on the “Big Five” personality traits have been shown

to be effective in explaining unethical behaviors [12],

they are not frequently used in the context of

academic dishonesty. Most researchers who have used

the “Big Five”, which consists of openness to

experience, conscientiousness, extraversion,

agreeableness, and neuroticism (OCEAN), addressed

only a few traits instead of the whole model [11, 13].

Descriptions of the personality traits associated with

the “Big Five” model in the context of academic

dishonesty are shown below.

The conscientious student may be described as

dependable, achievement-oriented, persistent,

responsible and honest [14]. He operates as an

effective regulator of his own actions, who is able to

restrain and regulate behavior through “effortful

control”, thus, he can resist cheating [15] and hold

more negative attitudes toward cheating [16]. By

contrast, the student with lower conscientiousness is

expected to be irresponsible, disorganized and

impulsive. As a consequence, these characteristics

might lead to poorer study skills, which in turn might

increase the tendency to cheat. Another personality

trait—emotional stability (which is the reverse of

neuroticism)—reflects a student’s enhanced feeling of

competence and a sense of security [14]. This trait

allows him/her to be more relaxed, unworried and less

likely to become strained under stressful conditions,

such as test-taking or meeting deadlines. Thus,

students with this trait are considered to be less

inclined toward cheating behaviors [16].

Agreeableness involves cooperating with others and

maintaining harmony. Thus, an individual who is low

on this trait is expected to show lower levels of

cooperativeness. The personality trait of extraversion

is characterized as the tendency to be sociable,

talkative, energetic and sensation-seeking. Studies that

addressed this trait’s effect on dishonesty are scarce

and their results are contradictory [13]. Finally, high

openness to experience includes tendencies toward

intellectualism, imagination and broad-mindedness

[14]. Research shows that this personality trait is

related to academic success and to learning orientation,

reflecting a desire to understand concepts and master

material. Furthermore, learning orientation predicted

lower inclination to cheat [16].

Empirical research confirmed the relationship

between personality traits and an individual’s

tendency to cheat for emotional stability such that

students who are high on neuroticism (low on

emotional stability) have higher tendency to engage in

unethical academic behaviors and in cheating [11]. In

addition, low conscientiousness and low

agreeableness were found to predict cheating

behaviors as well [13]. More recently, the effects of

conscientiousness, emotional stability and openness to

experience on students’ attitudes towards cheating,

combined with two context variables—classroom

culture and pedagogy—were examined [16]. The

findings showed that while conscientiousness was the

sole personality measure that directly predicted

negative attitudes towards cheating, emotional

stability and openness to experience also led to

negative attitudes towards academic misconduct, but

only when combined with classroom context variables.

Based on these studies, we hypothesize that there will

be differences in students’ propensity to engage in

academic dishonesty based on various personality


1044

traits and whether they are face-to-face or

e-learners.

3. Method

3.1 Participants

The sample consisted of 1,574 participants with 803

from two American universities and 771 from four

Israeli academic institutes. 65% of the participants

were women and 35% were men. Their age ranged

from 17 to 59 (the mean was 26.4 years). 26% of the

participants were freshmen, 32%—sophomores,

20%—juniors, 19%—seniors and 3%—graduate

students. 46% were Christians, 38% were Jews, and

16% were Muslims. 13% of the participants were

excluded from the analysis because their surveys were

incomplete or carelessly completed. Therefore, the

final data set consisted of 1,365 participants.

3.2 Survey Instrument

A three part survey instrument was used in the

current study. Part 1 included the TIPI scale, which

consisted 10 items assessing the participants’

personality traits [17]. The reliability of this

questionnaire, measured by Cronbach’s alpha, was

0.72. Part 2 consisted of questions that examined

academic integrity using the Academic Integrity

Inventory [18]. These questions investigated the

students’ likelihood to engage in various forms of

academic misconduct. The instrument was validated

and reliability of this questionnaire, measured by

Cronbach’s alpha, was 0.75 [18]. Part 3 presented a

series of socio-demographic questions.

3.3 Procedure

In order to encourage the participants to think in the

frame of a specific type of course, we administered a

printed version of the survey instrument in the

traditional face-to-face courses and an on-line version

of the survey instrument in the e-learning courses. The

survey instruments were coded and grouped according

to the location of the participants’ university or

college (USA or Israel). The questionnaires were

distributed at the end of the courses.

4. Results

Table 1 summarizes the results of independent

sample t-test analyses, which indicate that there were

statistically significant differences in students’

likelihood to engage in academic dishonesty based on

the type of course in which they were enrolled.

Specifically, it was found that students in face-to-face

courses were more likely to engage in acts of

academic dishonesty than their counterparts in

e-learning courses.

Based on a MANOVA (multiple analysis of

variance) we found significant 2-way interaction

effect between country (Israel vs. United States) and

course type (on-line vs. face-to-face) [F (1, 1361) = 57.16,

p < 0.001].

Table 2 shows that there is a significant negative

correlation between the personality trait of

agreeableness and academic dishonesty indicating that

the more the students are cooperative with others, the

less are they to be academically dishonest in both

countries—Israel and USA. In addition, among Israeli

students that are identified with higher

conscientiousness and emotional stability

demonstrated a significant negative correlation with

academic dishonesty. General Linear Model revealed

that there is a significant 2-way interaction effect

among Israeli students between course type (on-line

vs. face-to-face) and the personality trait of

conscientiousness [F = 2.058, p < 0.05] and between

course type and the personality trait of emotional

stability [F = 2.047, p < 0.05]. Interestingly, the

personality trait of extraversion among American

students was found to be positively correlated with

academic dishonesty, indicating that the tendency to

be sociable is correlated with a higher inclination to

cheat.


1045

Table 1 Differences in academic dishonesty by course type and country.

Country Course type N Mean S.D. T-Test F

USA E-learning 287 1.61 0.52

12.70***

57.16*** Face-to-Face 470 2.16 0.66

Israel E-learning 293 1.78 0.60

5.33*** Face-to-Face 315 2.52 0.65

***P < 0.001, **P < 0.01, *P < 0.05.

Table 2 Correlation between personality and academic dishonesty by course type and country.

Course type 1 2 3 4 5

Israel E-learning -.038 -.149* -.125* -.246** -.068

Face-to-Face -.090 -.131 -.237** -.151** -.063

USA E-learning -.100 -.090 -.057 -.121* -.038

Face-to-Face -.016 -.040 -.031 -.114* .105*

***P < 0.001, **P < 0.01, *P < 0.05 Israel (N = 608) USA (N = 757). Note: 1 = Openness to Experiences, 2 = Emotional Stability, 3 = Consciousness, 4 = Agreeableness, 5 = Extraversion.

5. Conclusions

Our research found that there is less overall

cheating in the virtual than in the traditional classroom

setting. These findings are similar to those found by

other researchers [19, 20], who explained this

phenomenon by the notion that students in virtual or

online courses may have a higher motivation to learn

or are able to learn without being dependent on the

typical structure of traditional classroom settings.

Our research also indicates that the personality

traits of emotional stability, agreeableness and

conscientiousness are negatively related to academic

dishonesty. These results are similar to the findings

reported by other researchers [11, 13, 16]. One of the

explanations for the notion that personality trait of

conscientiousness predicts academic dishonesty is that

conscientious students have less need to cheat since

they tend to be better prepared academically [21].

Conscientious students are able to resist cheating

since they are achievement-oriented, but at the same

time responsible and honest [14] and are able to

regulate their behavior [15]. The personality trait of

emotional stability also can help students to avoid

unethical academic behaviors, since students that are

high on this trait have sense of security [14], which

allows them to be less influenced by stressful

conditions [16].

The findings of this study revealed that the effects

of conscientiousness and emotional stability on

academic dishonesty that appeared among Israeli

students were not observed among their American

counterparts. This might be explained by the cultural

differences, as several studies that compared US

students with students in Lebanon [22], China [23]

and non-Western countries [24] indicated that

Americans tend to show less acceptance for cheating

and to possess higher standards with regard to

honesty.

Classroom contextual effects, such as those

presented in another study [16], may be worth

investigating in further research, since they seem to

contribute to the literature linking the effects of

personality traits to attitudes toward cheating. The

main practical implication of this research is its

contribution to our knowledge on the process of

student profiling, since we found that students who

use cheating practices are less emotionally stable, less

conscientious and less agreeable. Further research

should focus on how to amplify cooperative tasks in

online courses in order to reduce academic dishonesty.

References

[1] C. Robinson-Zanartu, E.D. Pena, V. Cook-Morales, A.M. Pena, R. Afshani, L. Nguyen, Academic crime and punishment: Faculty members’ perceptions of and


1046

responses to plagiarism, School Psychology Quarterly 20 (2005) 318-337.

[2] J. Walker, Measuring plagiarism: Researching what students do, not what they say they do, Studies in Higher Education 35 (2010) 41-59.

[3] K. Kelley, K. Bonner, Distance education and academic dishonesty: Faculty and administrator perception and responses, Journal of Asynchronous Learning Network 9 (2005) 43-52.

[4] J. Burgoon, M. Stoner, J. Bonita, N. Dunbar, Trust and Deception in Mediated Communication, in: 36th Hawaii International Conference on Systems Sciences, Big Island, Jan. 7-9, 2003.

[5] N. Rowe, Cheating in online student assessment: Beyond plagiarism, Online Journal of Distance Learning 7 (2004).

[6] M. Heberling, Maintaining academic integrity in online education, Online Journal of Distance Learning Administration 5 (2004).

[7] I. Anitsal, M. Anitsal, R. Elmore, Academic dishonesty and intention to cheat: A model on active versus passive academic dishonesty as perceived by business students, Academy of Educational Leadership Journal 13 (2009) 17-26.

[8] W. Kibler, Academic dishonesty: A student development dilemma, National Association of Student Personnel Administrators. NASPA Journal 30 (1993) 252-267.

[9] D.E. Allmon, D. Page, R. Roberts, Determinants of perceptions of cheating: Ethical orientation, personality and demographics, Journal of Business Ethics 23 (2000) 411-422.

[10] S. Etter, J. Cramer, S. Finn, Origins of academic dishonesty: Ethical orientations and personality factors associated with attitudes about cheating with information technology, Journal of Research on Technology in Education 39 (2006) 133-155.

[11] C.J. Jackson, S.Z. Levine, A. Furnham, N. Burr, Predictors of cheating behavior at a university: A lesson from the psychology of work, Journal of Applied Social Psychology 32 (2002) 1031-1046.

[12] P.R. Sackett, J. E Wanek, New developments in the use of measures of honesty, integrity, conscientiousness, dependability, trustworthiness, and reliability for personnel selection. Personnel Psychology 42 (1996) 787- 829.

[13] K.M. Williams, C. Nathanson, D.L. Paulhus, Identifying

and profiling scholastic cheaters: Their personality, cognitive ability and motivation, Journal of Experimental Psychology 16 (2010) 293-307.

[14] M.R. Barrick, M.K. Mount, The big five personality dimensions and job performance: A meta-analysis. Personnel Psychology 44 (1991) 1-26.

[15] L.A., Jensen-Campbell, W.G. Graziano, The two faces of temptation: Differing motives for self-control, Merrill-Palmer Quarterly 51 (2005) 287–314.

[16] N.E. Day, D. Hudson, P.R. Dobies, R. Waris, Student or situation? Personality and classroom context as predictors of attitudes about business school cheating, Social Psychology of Education 14 (2001) 261-282.

[17] S.D. Gosling, P.J. Rentfrow, W.B. Swann Jr., A very brief measure of the big five personality domains, Journal of Research in Personality 37 (2003) 504-528.

[18] J.L. Kisamore, T.H. Stone, I.M. Jawahar, Academic integrity: The relationship between individual and situational factors on misconduct contemplations, Journal of Business Ethics, 75 (2007) 381-394.

[19] D. Stuber-McEwen, P. Wiseley, S. Hoggatt, Point, click, and cheat: Frequency and type of academic dishonesty in the virtual classroom, Online Journal of Distance Learning Administration 7 (2009).

[20] Y. Peled, C. Barczyk, Y. Eshet, K. Grinautski, Learning motivation and student academic dishonesty—A comparison between face-to-face and online courses, In P. Resta (Ed.), in: Proceedings of Society for Information Technology & Teacher Education International Conference, Chesapeake, Mar. 5-9, 2012, 752-759, VA: AACE.

[21] J. Hogan, R. Hogan, How to measure employee reliability, Journal of Applied Psychology 74 (1989) 273-279.

[22] D.L. McCabe, T. Feghali, H. Abdallah, Academic dishonesty in the Middle East: Individual and contextual factors, Research in Higher Education 49 (2008) 451-467.

[23] M. Rawwas, J. Al-Khatib, S. Vitell, Academic dishonesty: A cross-cultural comparison of U.S. and Chinese marketing students, Journal of Marketing Education 26 (2004) 89-100.

[24] P.W. Grimes, Dishonesty in academics and business: A cross-cultural evaluation of student attitudes, Journal of Business Ethics 49 (2004) 273-290.


Data Security Model for Cloud Computing

Eman M. Mohamed, Hatem S. Abdelkader and Sherif El-Etriby

Department of Computer Science, Faculty of Computers and Information, Menofia University, Menofia 32511, Egypt

Received: May 16, 2013 / Accepted: June 09, 2013 / Published: August 31, 2013.

Abstract: From the perspective of data security, which has always been an important aspect of quality of service, cloud computing focuses a new challenging security threats. Therefore, a data security model must solve the most challenges of cloud computing security. The proposed data security model provides a single default gateway as a platform. It used to secure sensitive user data across multiple public and private cloud applications, including Salesforce, Chatter, Gmail, and Amazon Web Services, without influencing functionality or performance. Default gateway platform encrypts sensitive data automatically in a real time before sending to the cloud storage without breaking cloud application. It did not effect on user functionality and visibility. If an unauthorized person gets data from cloud storage, he only sees encrypted data. If authorized person accesses successfully in his cloud, the data is decrypted in real time for your use. The default gateway platform must contain strong and fast encryption algorithm, file integrity, malware detection, firewall, tokenization and more. This paper interested about authentication, stronger and faster encryption algorithm, and file integrity. Key words: Cloud computing, data security model in cloud computing, randomness testing, cryptography for cloud computing, OTP (one time password).

1. Introduction

In the traditional model of computing, both data and

software are fully contained on the user’s computer; in

cloud computing, the user’s computer may contain

almost no software or data (perhaps a minimal

operating system and web browser, display terminal

for processes occurring on a network).

Cloud computing is based on five attributes:

multi-tenancy (shared resources), massive scalability,

elasticity, pay as you go, and self-provisioning of

resources, it makes new improvements in processors,

Virtualization technology, disk storage, broadband

Internet connection, and combined fast, inexpensive

servers to make the cloud to be a more compelling

solution.

The main attributes of cloud computing are

illustrated as follows [1]:

Multi-tenancy (shared resources): Cloud computing

Corresponding author: Eman Meslhy Mohamed Elsdody, M.Sc., Lecturer, research fields: cloud computing and information security. E-mail: [email protected].

is based on a business model in which resources are

shared (i.e., multiple users use the same resource) at

the network level, host level and application level;

Massive scalability: Cloud computing provides the

ability to scale to tens of thousands of systems, as well

as the ability to massively scale bandwidth and

storage space;

Elasticity: Users can rapidly increase and decrease

their computing resources as needed;

Pay as you used: Users to pay for only the resources

they actually use and for only the time they require

them;

Self-provisioning of resources: Users’

self-provision resources, such as additional systems

(processing capability, software, storage) and network

resources.

Cloud computing can be confused with distributed

system, grid computing, utility computing, service

oriented architecture, web application, web 2.0,

broadband network, browser as a platform,

Virtualization and free/open software [2].


1048

Cloud computing is a natural evolution of the

widespread adoption of virtualization, service-oriented

architecture, autonomic and utility computing [3].

Details are abstracted from end-users, who no longer

have a need for expertise in, or control over, the

technology infrastructure “in the cloud” that supports

them as shown in Fig. 1.

Cloud services exhibit five essential characteristics

that demonstrate their relation to, and differences from,

traditional computing approaches such as on-demand

self-service, broad network access, resource pooling,

rapid elasticity and measured service [4].

Cloud computing often leverages massive scale,

homogeneity, virtualization, resilient computing (no

stop computing), low cost/free software, geographic

distribution, service orientation software and advanced

security technologies [4].

The main objective of this paper is to enhance data

security model for cloud computing. The proposed

data security model solves cloud user security

problems, help cloud provider to select the most

suitable encryption algorithm to its cloud. We also

help user cloud to select the highest security

encryption algorithm.

The proposed data security model is composed of

three-phase defense system structure, in which each

floor performs its own duty to ensure the data security

of cloud. The first phase is responsible for strong

authentication. It applies the OTP (one time password)

as a two-factor authentication system. OTP provides

high security because it used one password in a

session and can not be cracked. The second phase

selects the stronger and a faster encryption algorithm

by proposing algorithm called “Evaluation algorithm”.

This algorithm used for selected eight modern

encryption techniques namely: RC4, RC6, MARS,

AES, DES, 3DES, Two-Fish, and Blowfish. The

evaluation has performed for those encryption

algorithms according to randomness testing by using

NIST statistical testing. This evaluation uses PRNG

(pseudo random number generator) to determine the

most suitable. This evaluation algorithm performed at

Amazon EC2 Micro Instance cloud computing

environment. In addition, this phase checks the

integrity of user data. It encourages cloud users to

encrypt his data by using “TrueCrypt” software or

proposed software called “CloudCrypt V.10”. The

third phase, ensure fast recovery of user data.

The paper is organized as follows: in Section 2

cloud computing architecture is defined; Cloud

computing security is discussed in Section 3; in

Section 4 methodology is described and finally in

Section 5 interruptions of the results are described.

2. Cloud Computing Architecture

2.1 Cloud Computing Service Models

Cloud SaaS (software as a service): Application and

information clouds, use provider’s applications over a

network, cloud provider examples Zoho,

Salesforce.com, and Google Apps.

Cloud PaaS (platform as a service): Development

clouds, deploy customer-created applications to a

cloud, cloud provider examples Windows Azure,

Google App Engine and Aptana Cloud.

Cloud IaaS (infrastructure as a service):

Infrastructure clouds, Rent processing, storage,

network capacity, and other fundamental computing

resources, Dropbox, Amazon Web Services, Mozy

and Akamai.

Fig. 1 Evolution of cloud computing.


1049

2.2 Cloud Computing Deployment Models

Private cloud : Enterprise owned or leased;

Community cloud: Shared infrastructure for

specific community;

Public cloud: Sold to the public, mega-scale

infrastructure;

Hybrid cloud: Composition of two or more clouds.

2.3 Cloud Computing Sub-services Models

IaaS: DBaaS (database-as-a-service): DBaaS allows

the access and use of a database management system

as a service.

PaaS: STaaS (storage-as-a-service): STaaS involves

the delivery of data storage as a service, including

database-like services, often billed on a utility

computing basis, e.g., per gigabyte per month.

SaaS: CaaS (communications-as-a-service): CaaS is

the delivery of an enterprise communications solution,

such as Voice over IP, instant messaging, and video

conferencing applications as a service.

SaaS: SECaaS (security-as-a-service): SECaaS is

the security of business networks and mobile networks

through the Internet for events, database, application,

transaction, and system incidents.

SaaS: MaaS (monitoring-as-a-service): MaaS refers

to the delivery of second-tier infrastructure

components, such as log management and asset

tracking, as a service.

PaaS: DTaaS (desktop-as-a-service): DTaaS is the

decoupling of a user’s physical machine from the

desktop and software he or she uses to work.

IaaS: CCaaS (compute capacity-as-a-service):

CCaaS is the provision of “raw” computing resource,

typically used in the execution of mathematically

complex models from either a single “supercomputer”

resource or a large number of distributed computing

resources where the task performs well [5].

2.4 Cloud Computing Benefits

Lower computer costs, improved performance,

reduced software costs, instant software updates,

improved document format compatibility, unlimited

storage capacity, device independence, and increased

data reliability

2.5 Cloud Computing Drawbacks

Requires a constant Internet connection, does not

work well with low-speed connections, can be slow,

features might be limited, stored data might not be

secure, and stored data can be lost.

2.6 Cloud Computing Providers

AWS (amazon web services)—include Amazon S3,

Amazon EC2, Amazon Simple-DB, Amazon SQS,

Amazon FPS, and others. Salesforce.com—Delivers

businesses over the internet using the software as a

service model. Google Apps—Software-as-a-service

for business email, information sharing and security.

And others providers such as Microsoft Azure

Services Platform, Proof-point, Sun Open Cloud

Platform, Workday, etc..

3. Cloud Computing Security

With cloud computing, all your data is stored on the

cloud. So cloud users ask some questions like: How

secure is the cloud? Can unauthorized users gain

access to your confidential data?

Cloud computing companies say that data is secure,

but it is too early to be completely sure of that. Only

time will tell if your data is secure in the cloud. Cloud

security concerns arising which both customer data

and program are residing in provider premises.

Security is always a major concern in Open System

Architectures as shown in Fig. 2.

While cost and ease of use are two great benefits of

cloud computing, there are significant security

concerns that need to be addressed when considering

moving critical applications and sensitive data to

public and shared cloud environments. To address

these concerns, the cloud provider must develop

sufficient controls to provide the same or a greater


1050

level of security than the organization would have if

the cloud were not used.

There are three types of data in cloud computing.

The first type is a data in transit (transmission data),

the second data at rest (storage data), and finally data

in processing (processing data).

Clouds are massively complex systems can be

reduced to simple primitives that are replicated

thousands of times and common functional units.

These complexities create many issues related to

security as well as all aspects of Cloud computing. So

users always worry about its data and ask where the

data is? And who has access? Every cloud provider

encrypts the data in three types according to Table 1.

4. Methodology

Security of data and trust problem has always been

a primary and challenging issue in cloud computing.

This section describes a proposed data security model

in cloud computing. In addition, focuses on enhancing

security by using an OTP authentication system,

check data integrity by using hashing algorithms,

encrypt data automatically with the highest strong/

fast encryption algorithm and finally ensure the fast

recovery of data.

4.1 Existing Data Security Model

Most cloud computing providers [6-8] in first,

authenticates (e.g., Transfer usernames and password)

via secure connections and secondly, transfer (e.g., via

HTTPS) data securely to/from their servers (so-called

“data in transit”), but, as far as I can tell, none finally,

encrypts stored data (so-called “data at rest”)

automatically.

In cloud computing, to ensure correctness of user

data, in first, user must be make authentication.

Authentication is the process of validating or

confirming that access credentials provided by a user

(for instance, a user ID and password) are valid. A

user in this case could be a person, another application,

or a service; all should be required to authenticate.

Many enterprise applications require that users

authenticate before allowing access. The authorization,

the process of granting access to requested resources,

is pointless without suitable authentication. When

organizations begin to utilize applications in the cloud,

authenticating users in a trustworthy and manageable

manner becomes an additional challenge.

Organizations must address authentication-related

challenges such as credential management, strong

Fig. 2 Security is a major concern to cloud computing [8].

Table 1 Data security (encryption) in cloud computing.

Storage Processing Transmission

Symmetric encryption Homomophric encryption Secret socket layer SSL encryption

AES-DES-3DES-Blowfish-MARS Unpadded RSA-ElGamal SSL 1.0-SSL 3.0-SSL 3.1-SSL 3.2


1051

authentication, delegated authentication, and trust

across all types of cloud delivery models (SPI).

Two-factor authentication became solution for any

cloud application.

After authentication, cloud user can access data that

stored on servers at a remote location, and can access

it from any device/ anywhere.

So if you want your data to be secure in the cloud,

then consider encrypting the stored data. In addition,

do not store your encryption keys on the same server.

It is unclear whether a cloud computing provider

could be compelled by law enforcement agencies to

decrypt data that (1) it has encrypted or that (2) users

have encrypted, but if the provider has the keys,

decryption is at least possible.

4.2 Problems in Existing Data Security Model

The cloud users ask your cloud providers about

some problems:

Are files stored on cloud servers encrypted

automatically?

Can other users boot my machine?

Have unauthorized users of my machine

detected?

Is my data accessed solely by my virtual

machine?

Is the system should own my key?

Is my data retrieving fast?

Therefore, a data security model must solve the

previous problems. DSM (data security model) must:

ensure that data must be encrypted automatically;

use a strong encryption algorithm;

use the strong encryption algorithm that must be

fast to retrieve data faster;

use strong authentication;

ensure file integrity;

be kept not in cloud system; most of these

problems appear in previous data security model [9].

Clouds typically have a single security architecture.

This section provides a detailed overview of the

existing data security model, as well as explains the

data security model for Amazon web services as

shown in Table 2.

Amazon web services encourage user’s to encrypt

sensitive data by using TrueCrypt software, this free

encryption product provides the following:

Creates a virtual encrypted disk within a file and

mounts it as a real disk;

Encrypts an entire partition or storage device

such as USB flash drive or hard drive;

Encrypts a partition or drive where Windows is

installed;

Encryption is automatic, real-time (on-the-fly)

and transparent;

Provides two levels of plausible deniability, in

case an adversary forces you to reveal the password:

Hidden volume (steganography). No TrueCrypt

volume can be identified (volumes cannot be

distinguished from random data);

Encryption algorithms: AES, Serpent, and

Twofish.

Finally mode of operation is XTS.

TrueCrypt is an outstanding encryption solution for

anyone familiar with managing volumes and a slight

knowledge of encryption technology. For the rest, it

can be a bit daunting. Any organization planning to

deploy TrueCrypt as a cloud-data protection solution

must consider the cost and logistics of training and

supporting users, managing versions, and recovering

damages.

TrueCrypt is a computer software program whose

primary purposes are to: Secure data by encrypting it

before it is written to a disk. Decrypt encrypted data

after it is read from the disk. However, TrueCrypt uses

only three methods (AES, Serpent and Twofish) to

encrypt data as shown in Fig. 3.

4.3 Proposed Data at Rest Security Model

The proposed data security model used three-level

defense system structure, in which each floor performs

its own duty to ensure that the data security of cloud

as shown in Fig. 4.


1052

Table 2 Data security model in Amazon EC2.

Homoromphic encryption algorithm, e.g., RSA. Data for processing Secure Sockets Layer, e.g., HTTPS Data in transit Depend on the OS used In windows EFS are used. It is not full encryption. It encrypts individual files. If you need a full encrypted volume, consider using the open-source TrueCrypt product. Linux uses AES, it encrypts individual files. OpenSolaris can take advantage of ZFS Encryption

Data at rest

The key kept in the cloud as a list but it makes key rotation Key management Two factor authentication (e-mail and password) (account of AWS AMI) X.509 certificates for authentication

Authentication

AWS encourages users to encrypt their sensitive data before it upload into Amazon S3. User protection

The first phase: strong authentication is achieved by

using OTP.

The second phase: data are encrypted automatically

by using strong/fast encryption algorithm. In addition

to encrypt data, users can encrypt his sensitive data by

using TrueCrypt software or proposed software

CloudCrypt V.10. CloudCrypt software uses eight

modern/strong encryption algorithms. Finally, data

integrity is achieved by using hashing algorithms.

The third phase: fast recovery of user data is

achieved in this phase.

The three phases are implemented in default

gateway, as shown in Fig. 5. The proposed data

security model provides a single default gateway as a

platform to secure sensitive customer data across

multiple public and private cloud applications,

including Salesforce, Gmail, and Amazon Web

Services, without affecting functionality or

performance.

Default gateway platform tasks:

Encrypt sensitive data automatically on a real

time before sending to the cloud without breaking

cloud application;

The default gateway platform did not effect on

user functionality and visibility;

If an unauthorized person gets data from cloud

storage, he can see the encrypted data;

If authorized person access success in his cloud,

the data is decrypted in real time for your use;

Fig. 3 TrueCrypt encryption options.

Fig. 4 Proposed data security model in cloud computing.

Fig. 5 How data stored in the cloud by using the proposed data security model.

Phase 1 Phase 2

Phase 3

Automatically Data Encryption

Automatically Data Integrity

Data Fast Recovery

OTP Authentication

Private User Protection


1053

The default gateway platform must contain

Strong/Fast Encryption Algorithm by using one or

more of encryption algorithms namely RC4, RC6,

MARS, AES, DES, 3DES, Two-Fish, and Blowfish

[10-16];

The default gateway platform must contain File

integrity;

The default gateway platform must contain

Malware detection, Firewall, Tokenization and more.

Proposed data security model implemented and

applied to cloudsim 3.0 by using HDFS architecture

and Amazon web services (S3 and EC2).

In this paper, automatically encryption, integrity,

fast recovery and private user encryption all are

achieved in the proposed data security model.

4.4 Implementation Details

4.4.1 First Phase, Authentication

The cloud user select company, then create an

account;

Cloud provider upload user information in DB in

cloud storage;

Cloud Provider confirms user with his username

and password;

Cloud user request login page;

The cloud provider displays login screen;

Cloud user login with username and password;

A cloud provider check is valid username and

password by searching in DB in cloud storage. If user

information not valid display error message else

display reserve a PC page;

Cloud user reserves your PC.

4.4.2 OTP Authentication Steps

Cloud user enters passphrase, challenge and

sequence number for OTP authentication;

Cloud user generates an OTP;

The cloud provider generates the OTP temporary

DB based on user information;

Cloud user login with OTP;

A cloud provider check is valid OTP by searching

in temporary DB for OTP in cloud storage. If OTP not

valid display error message else display user PC page.

4.4.3 In Second Phase, Private User Protection

Before adding data, cloud user can encrypt data by

using TrueCrypt or CloudCrypt software’s;

In second phase, Automatic data encryption;

Cloud user adds data;

Cloud server encrypt data automatically by using

fast/strong encryption algorithm that selected based on

an evaluation algorithm for the cloud company.

4.4.4 Second Phase, Automatic Check Data

Integrity

The cloud server generates file hash value;

Cloud server store data with its hash value;

When a cloud user requests his data, cloud server

decrypt data automatically, check integrity by check

the hash value.

4.4.5 Third Phase, Fast Recovery of Data

Finally, cloud server retrieves data with message of

file integrity.

4.5 Proposed Evaluation Algorithm

We use NIST statistical tests to get the highest

security encryption algorithm from eight algorithms

namely RC4, RC6, MARS, AES, DES, 3DES,

Two-Fish, and Blowfish as shown in Fig. 6. NIST

Developed to test the randomness of binary sequences

produced by either hardware or software based

cryptographic random or pseudorandom number

generators.

NIST statistical tests has 16 tests, namely: The

Frequency (Mon-obit) Test, Frequency Test within a

Fig. 6 Steps to select the highest encryption algorithm.


1054

Block, The Runs Test, Tests for the

Longest-Run-of-Ones in a Block, The Binary Matrix

Rank Test, The Discrete Fourier Transform (Spectral)

Test, The Non-overlapping Template Matching Test,

The Overlapping Template Matching Test, Maurer’s

“Universal Statistical” Test, The Linear Complexity

Test, The Serial Test, The Approximate Entropy Test,

The Cumulative Sums (Cusums) Test, The Random

Excursions Test, and The Random Excursions Variant

Test.

We also compare between eight encryption

algorithms based on speed of encryption to achieve

faster recovery.

We use Amazon EC2 as a case study of our

software. Amazon EC2 Load your image onto S3 and

register it. Boot your image from the Web Service.

Open up the required ports for your image. Connect to

your image through SSH. And finally execute your

application.

For our experiment in a cloud computing

environment, we use Micro Instances of this Amazon

EC2 family, provide a small amount of consistent

CPU resources, they are well suited for lower

throughput applications, 613 MB memory, up to 2

EC2 Compute Units (for short periodic bursts), EBS

(elastic block store) storage only from 1 GB to 1 TB,

64-bit platform, low I/O Performance, t1.micro API

name, We use Ubuntu Linux to run NIST Statistical

test package [17-19].

4.6 Selection the Highest Encryption Algorithm Steps

Sign up for Amazon web service to create an

account. Lunch Micro instance Windows (64 bit)

Amazon EC2. Connect to Amazon EC2 [20-22]

Windows Micro Instance. Generate 128 plain stream

sequences as PRNG, each sequence is 7,929,856 bits

in length (991,232 bytes in length) and key stream

(length of key 128 bits). Apply cryptography

algorithms to get ciphers text. Lunch Micro instance

Amazon EC2 Ubuntu Linux Connect to Amazon EC2

Ubuntu Linux Micro instance Run NIST statistical

tests for each sequence to eight encryption algorithms

to get P-value Compare P-value to 0.01, if P-value

less than 0.01 then reject the sequence.

We compare between eight encryption methods

based on P-value, Rejection rate and finally based on

time consuming for each method.

We have 128 sequences (128-cipher text) for each

eight-encryption algorithm.

Each sequence has 7,929,856 bits in length

(991,232 bytes in length). Additionally, the P-values

reported in the tables can find in the results.txt files

for each of the individual test—not in the

finalAnalysisReport.txt file in NIST package.

The P-value represents the probability of observing

the value of the test statistic which is more extreme in

the direction of non-randomness. P-value measures

the support for the randomness hypothesis on the basis

of a particular test Rejection Rate number of rejected

sequences (P-value less than significance level α may

be equal 0.01 or 0.1 or 0.05). The higher P-value the

better and vice versa with rejection rate, the lower the

better [17].

For each statistical test, a set of P-values

(corresponding to the set of sequences) is produced.

For a fixed significance level α, a certain percentage

of P-values are expected to indicate failure. For

example, if the significance level is chosen to be

0.01 (i.e., α ≥ 0.01), then about 1% of the sequences

are expected to fail. A sequence passes a statistical

test whenever the P-value ≥ α and fails

otherwise.

We produce P-value, which small P-value(less than

0.01) support non-randomness. For example, if the

sample consists of 128 sequences, the rejection rate

should not exceed 4.657, or simply expressed 4

sequences with α = 0.01. The maximum number of

rejections was computed using the formula [17]:

)1(

3= # rateRejection

ss

(1)

where s is the sample size and α is the significance

level is chosen to be 0.01.


1055

5. Simulation Results

In this section, we show and describe the simulation

results of the proposed data security model.

5.1 OTP Authentication

OTP [23] System steps as shown in Fig. 7.

The users connect to the cloud provider. Then the

user gets the username (e-mail), password and finally

account password.

Users login to the cloud provider website by getting

username (e-mail), password and account password.

Cloud node controller verifies user info. If user info

is true, controller-node send that login authentication

success and require OTP.

OTP generation software used to generate OTP as

shown in Fig. 8.

Users generate OTP by using MD5 hash function

and sequence number based on user name, password

and account password.

Then users login to cloud website with OTP as

shown in Fig. 9.

The cloud controller node generates 1000 OTP

based on user info by using the MD5 hash function.

Then the cloud controller saves 1000 OTP in the

temporary OTP database.

The cloud controller verifies user OTP from the

temporary OTP database.

If OTP is true, send OTP login success.

Password Space is a function of the size of the

alphabet and the number of characters from that

alphabet that are used to create passwords. To

determine the minimum size of the password space

needed to satisfy the security requirements for an

operational environment. In the text-based password, a

password space is computed as [23] :

S = A·M (2)

Password length and alphabet size are factors in

computing the maximum password space

requirements. Eq. 2 expresses the relationship between

S, A, and M where:

S = password space;

A = number of alphabet symbols;

M = password length.

To illustrate: If passwords consisting of 4 digits

using an alphabet of 10 digits (e.g., 0-9) are to be

generated: S = 104

That is, 10,000 unique 4-digit passwords could be

generated. Likewise, to generate random 6-character

passwords from an alphabet of 26 characters (e.g.,

A-Z): S = 266

Password entropy is usually used to measure the

security of a generated password which mean how

hard to blindly guess out the password.

Entropy = M × log2(A). (3)

Fig. 7 OTP authentication in PDSM.


1056

Fig. 8 Proposed software for OTP generation.

Fig. 9 Proposed OTP login screen.

In other word, entropy tries to measure the

probabilities that the attacker obtains the correct

password based on random guessing. For example,

text message with 6 characters length with capital and

small alphabets then the number of entropy bits = 6 ×

log2(52) = 34.32 [23].

We have compared password space with different

password schemas we can identify the most secure

approaches with respect to brute force attack as shown

in Table 3. This table demonstrates the comparison of

the password space and password length for popular

user authentication schemas for cloud computing. It

shows that the approach presented by us is both more

secured and the easiest to remember. At the same time,

it is relatively fast to produce during an authentication

procedure as shown in Figs. 10 and 11.

Table 3 Password space comparison.

Authentication system Alphabet Password length Password space size Entropy bits

Static password 82 12 92.4 × 1021 22.96

PIN number 10 12 1 × 1012 12

OTP 40 30 1.15 × 1048 48.06

Fig. 10 Security strength comparison based on entropy bits.

Fig. 11 Security strength comparison based on password space size.


1057

We must remember that, an OTP (one-time

password) is a password that is valid for only one

login session or transaction. OTPs avoid a number of

shortcomings that are associated with traditional

(static) passwords. the most important shortcoming

which is addressed by OTPs is that, in contrast to

static passwords, they are not vulnerable to replay

attacks. This means that, if a potential intruder

manages to record an OTP that was already used to

log into a service or to conduct a transaction, he or she

will not be able to abuse it since it will be no longer

valid. On the downside, OTPs are difficult for human

beings to memorize. Therefore they require additional

technology in order to work.

Benefits of OTP in cloud computing

OTP offers strong two-factor authentication;

The OTP is unique to this session and can not be

used again;

OTP offers strong security because they cannot

be guessed or hacked;

Provides protection from unauthorized access;

Easier to use for the employee than complex

frequently changing passwords;

Easy to deploy for the administrator;

Good first step to strong authentication in an

organization;

Low cost way to deploy strong authentication.

5.2 Evaluation Algorithm Results

In this paper, we select the strongest and the fastest

encryption algorithm by proposing algorithm called

“Evaluation algorithm”. This algorithm used for

selecting eight modern encryption techniques namely:

RC4, RC6, MARS, AES, DES, 3DES, Two-Fish and

Blowfish. The evaluation has performed for those

encryption algorithms according to randomness

testing by using NIST statistical testing. This

evaluation uses PRNG (pseudo random number

generator) to determine the most suitable. This

evaluation algorithm performed at Amazon EC2

Micro Instance cloud computing environment.

Simulation results are given in Table 4 for the

selected eight-encryption algorithms rejection rate in

Amazon EC2. This table shows the rejection rate

results at Amazon EC2.

We notice that there are no strong indications of

statistical weaknesses about Amazon EC2 rejection

rate in results and the rejection rate for selected

modern encryption techniques does not exceed the

maximum number of rejection rate (expected 4)

Table 4 Amazon EC2 rejection rate for modern encryption algorithms.

AES 3DES Blowfish Two-Fish MARS DES RC4 RC6 Nist tests AcceptRejectAccept Reject Accept RejectAcceptRejectAcceptRejectAcceptReject Accept Reject Accept Reject

128 0 128 0 127 1 127 1 127 1 127 1 127 1 127 1 1

126 2 127 1 127 1 127 1 124 4 125 3 126 2 127 1 2

128 0 126 2 127 1 127 1 127 1 128 0 126 2 127 1 3

128 0 127 1 126 2 128 0 127 1 127 1 127 1 125 3 4

127 1 125 3 127 1 125 3 127 1 127 1 128 0 126 2 5

128 0 127 1 125 3 128 0 127 1 128 0 127 1 127 1 6

125 3 126 2 127 1 127 1 126 2 127 1 128 0 128 0 7

127 1 128 0 127 1 126 2 126 2 126 2 127 1 128 0 8

127 1 127 1 127 1 128 0 127 1 127 1 126 2 127 1 9

126 2 127 1 126 2 125 3 127 1 126 2 127 1 127 1 10

80 48 74 54 77 51 78 50 66 62 76 52 77 51 77 51 11

81 47 72 56 76 52 80 48 69 59 77 51 78 50 75 53 12

127 1 126 2 128 0 125 3 128 0 128 0 126 2 126 2 13

126 2 128 0 127 1 125 3 127 1 127 1 126 2 127 1 14

127 1 126 2 127 1 127 1 127 1 128 0 126 2 128 0 15

127 1 126 2 126 2 126 2 125 3 127 1 128 0 124 4 16


1058

expect two tests. The first is the test number 11:

Random Excursions and the other is the test number

12: Random Excursions Variant.

The random excursion test and the random

excursion variant test only apply whenever the

number of cycles exceeds 500. If a sample has

sequences with cycles fewer than 500, then they will

not be evaluated by the random excursion tests, and

thus the proportion of applicable sequences will be

reduced (by as much as 42%). In this event, a small

sample size may incorrectly suggest deviation from

randomness. It is important to keep in mind that only

one sample was constructed for each algorithm and

data category.

The random excursion test and the random

excursion variant test depend on Eq. (4):

500,005.0max nJ (4)

where J is denoting the total number of cycles in the

string and the randomness hypothesis will be rejected

when the P-value is too small. The result appears in

file stats.txt in NIST package for test 11 and test 12 as

shown in Figs. 12 and 13.

Experimental results for this comparison point are

shown in Table 5 to indicate the most suitable

encryption algorithm to Amazon EC2. The results

show the superiority of the AES algorithm over other

algorithms in terms of highest security and faster

encryption algorithm. Another point can be noticed

here that Blowfish and DES require less time and

higher security than all algorithms except AES.

Finally, it is found that Twofish has low performance

when compared with other algorithms. All this results

are from Amazon EC2 Ubunto 11.10 Micro Instance.

This evaluation must apply to all cloud companies

with different operating systems.


shown in Fig. 14 to indicate the highest security for

modern encryption techniques based on rejection rate.

The results show the superiority of the AES algorithm

over other algorithms in terms of the rejection rate

(small number of rejection).


shown in Fig. 15 to indicate the highest security for

modern encryption techniques based on P-value. The

results show the superiority of the AES algorithm over

other algorithms in terms of the P-value. Another

point can be noticed here that RC6 requires more

P-value than all algorithms except AES. A third point

can be noticed here that 3DES has an advantage over

other DES, RC4, MARS, 3DES and Twofish in terms

of P-value. Finally, it is found that Twofish has low

security when compared with other algorithms.

Fig. 12 Screenshot of stats .txt file in NIST for random excursion test at Amazon EC2.

Fig. 13 Screenshot of stats .txt file in NIST for random excursion variant test at Amazon EC2.

Table 5 Modern encryption algorithms evaluation in Amazon EC2.

Evaluation parameters 1 2 3 4 5 6

Rejection Rate AES DES Blowfish RC4 Twofish RC6

P-value AES RC6 3DES MARS DES Blowfish

Enc/Dec speed Blowfish AES RC4 DES RC6 MARS


1059

Fig. 14 Amazon EC2 average rejection rate for eight modern encryption algorithms.

Fig. 15 Amazon EC2 Average P-value for eight modern encryption algorithms based on 16 NIST test.


shown in Fig. 16 to indicate the speed of

encryption/decryption. The results show the

superiority of the Blowfish algorithm over other

algorithms in terms of the processing time. Another

point can be noticed here that AES requires less time

than all algorithms except Blowfish. A third point can

be noticed here that RC4 has an advantage over other

DES, RC6, MARS, 3DES and Twofish in terms of

time consumption. A fourth point can be noticed here

that 3DES has low performance in terms of power

consumption when compared with DES. It always

requires more time than DES because of its triple

phase encryption characteristics. Finally, it is found

that Twofish has low performance when compared

with other algorithms.

5.3 Security Desktop vs. Cloud

We notice from results that, the Desktop rejection

Fig. 16 Encryption/decryption comparison with different size in Amazon EC2.

rate more than Amazon EC2 rejection rates. In

addition, Desktop P-value is less than Amazon EC2

P-value. That is because cloud computing has

advanced security technologies than Desktop. The

number of P-value in cloud computing (that is less

than 0.1 thresholds) is less than number of P-value in

desktop (that is less than 0.1 thresholds). This

conclusion ensures that the security in cloud

computing is higher than security in traditional

desktop as shown in Fig. 17.

5.4 Private User Protection

Amazon web services encourage users to encrypt

sensitive data by using TrueCrypt software. A new

computer software program is implemented to encrypt

data before storing in cloud storage devices. This

software enables users to choose from eight

encryption techniques namely: AES, DES, 3DES,

RC4, RC6, Twofish, Blowfish, and MARS as shown

in Fig. 18.

5.5 Ensuring Integrity

This is an extra concern for customers that now

they have to worry about how to keep data hidden

from auditors. The actual problem of “trust” remains

the same. In order to avoid third party auditors in this

chain, this paper propose that the integrity check of

data stored in the cloud can be checked on customer’s

side. This integrity check can be done by using

cryptographic hash functions.


1060

Fig. 17 Desktop and Amazon EC2 P-value comparison.

Fig. 18 Proposed encryption software CloudCrypt at runtime in Amazon EC2.

For integrity check, we have to think about a simple

solution that is feasible and easy to implement for a

common user. The trust problem between Cloud

storage and customer can be solved, if users can check

the integrity of data themselves instead of renting an

auditing service to do the same. This can be achieved

by hashing the data on user’s side and storing the hash

values in the cloud with the original data. As shown in

Fig. 19. This figure presents the overview of the

scheme.

(1) The program takes file path that as shown in Fig.

20;

(2) The program computes a four-hash values in

this file based on the four hash functions (MD4, MD5,

SHA-1 and SHA-2) as shown in Fig. 21;

(3) When users store data in cloud storage devices,

server store filled with four hash values;

(4) When a user retrieve data file, server generate

four hash values;

Fig. 19 Overview of integrity check with hash functions.

Fig. 20 Screen shot of check integrity program.

Retrieve file and check integrity

Store file with hash value

Cloud server Or Owner of data

File Hash value


1061

Fig. 21 Check integrity program calculating hash values.

Table 6 Summarized results of the proposed data security model in cloud computing.

Features Description

Authentication OTP Authentication System (mathematical generation).

Provider encryption Software implemented to select the highest security and faster encryption algorithm based on NIST statistical tests. This software select AES algorithm to Micro Instance ubunto Amazon EC2 with Amazon S3.

Private user encryption TrueCrypt system or proposed software CloudCrypt v.10.

Data integrity Hashing-MD5-MD4-SHA-1-SHA-2.

Data fast recovery Based on decryption algorithm speed.

Key management User keys not stored in provider control domain.

(5) Server check integrity by comparing new four

hash values with stored four hash values.

The following are the advantages of using the

utility:

Not much implementation effort required.

Cost effective and more secured.

Do not require much time to compute the hash

values.

Flexible enough to change the security level as

required.

Not much space required to store the hash values.

6. Conclusions

According to the simulation results, in the

authentication phase in the proposed data security

model, OTP is used as two-factor authentication

software. OTP archived more password strength

security than other authentication systems (BIN and

static password). This appears by comparing between

OTP, BIN, and static password authentication systems

based on the space time size and entropy bits.

From the simulation results of the second phase in

the proposed data security model, test the proposed

system in Ubunto Amazon Micro Instance EC2, and

from randomness and performance evaluation to eight

modern encryption algorithms AES is the best

encryption algorithm in Ubunto Amazon Micro

Instance EC2. In addition to the randomness and

performance evaluation, data integrity must be

ensured. Moreover, the proposed data security model

encourages users to use true-crypt to encrypt his/her

sensitive data.

From the comparison and performance evaluation,

fast recovery of data achieved to the user. These

appear in the proposed data security model third

phase.

From the comparison and performance evaluation,

cloud computing depend on some condition, however

it has advanced security technologies rather than

traditional desktop. The summarized results of

proposed data security model are shown in

Table 6.

References

[1] Information Security Briefing Cloud Computing, Vol.1,

Center of The Protection of National Infrastructure CPNI

by Deloitte, 2010, p. 71.

[2] L. Foster, Y. Zhao, I. Raicu, S.Y. Lu, Cloud computing

and grid computing 360-degree compared, in: Grid

Computing Environments Workshop, GCE’08, Austin,

TX, Nov. 12-16, 2008, pp. 1-10.

[3] The NIST Definition of Cloud Computing, National

Institute of Science and Technology, p.7. Retrieved July

24 2011.


1062

[4] J.W. Rittinghouse, J.F. Ransome, Cloud Computing Implementation, Management, and Security, CRC Press, Boca Raton, 2009.

[5] Security Guidance for Critical Areas of Focus in Cloud Computing V1.0 [Online], Apr. 2009, Cloud Security Alliance Guidance, www.cloudsecurityalliance.org/guidance/csaguide.v1.0.pdf.

[6] M.A. Vouk, Cloud computing-issues, research and

implementations, Journal of Computing and Information

Technology 16 (2008) 235-246

[7] Security Guidance for Critical Areas of Focus in Cloud Computing V2.1 [Online], Dec. 2009, Cloud Security Alliance Guidance.

[8] Security Guidance for Critical Areas of Focus in Cloud Computing V3.0 [Online], Cloud Security Alliance Guidance.

[9] Y.F. Dai, B. Wu, Y.Q. Gu, Q. Zhang, C.J Tang, Data security model for cloud computing, in: Proceedings of The 2009 International Workshop on Information Security and Application IWISA 2009, Qingdao, China, Nov. 21-22, 2009.

[10] C. Burwick, D. Coppersmith, The MARS Encryption Algorithm, Aug. 27, 1999

[11] E. Dawson, H. Gustafson, M. Henricksen, B. Millan, Evaluation of RC4 Stream Cipher, Information Security Research Centre Queensland University of Technology, Jul. 31, 2002

[12] W. Stallings, Cryptography and Network Security, 4th ed., Prentice Hall, N.J, 2005, pp. 58-309.

[13] D. Coppersmith, The Data Encryption Standard (DES) and Its Strength Against Attacks, IBM Journal of Research and Development 38 (1994) 243-250.

[14] J. Daemen, V. Rijmen, Rijndael: The advanced encryption standard, D r. Dobb's Journal 26 (2001)

137-139. [15] B. Schneier, The Blowfish Encryption Algorithm

[Online], http://www.schneier.com/blowfish.html (accessed: Oct. 25, 2008).

[16] B. Schneier, J. Kelseyy, D. Whitingz, D. Wagnerx, C. Hall, N. Ferguson, Twofish: A 128-bit block cipher, Jun. 15, 1998.

[17] J. Soto, Randomness Testing of the Advanced Encryption Standard Candidate Algorithms, U.S. Dept. of Commerce, Technology Administration, National Institute of Standards and Technology, 1999.

[18] R. Buyya, C.S. Yeo, S. Venugopal, Market-oriented

cloud computing: Vision, hype, and reality for delivering

IT services as computing utilities, in: 9th IEEE/ACM

International Symposium on Cluster Computing and the

Grid, CCGRID’09, Shanghai, May 18-21, 2009.

[19] A. Rukhin, J. Soto, J. Nechvatal, M. Smid, A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications, NIST Special Publication, Apr. 2010.

[20] Amazon EC2 API, Amazon Elastic Compute Cloud Developer Guide [Online], Oct. 2006, http://docs.amazonwebservices.com/AWSEC2/2006-10-01/DeveloperGuide/.

[21] Amazon Web Services, Amazon Simple Storage Service Developer Guide [Online], Mar. 2006.

[22] Amazon Web Services, Overview of Security Processes [Online], Sep. 2009 http://aws.typepad.com/aws/2009/08/introducing-amazon-virtual-private-cloud-vpc.html.

[23] V. Prakash, A. Infant, J. Shobana, Eliminating vulnerable attacks using One-Time Password and PassText—Analytical study of blended schema, Universal Journal of Computer Science and Engineering Technology 1 (2010) 133-140.


The Retraining Churn Data Mining Model in DMAIC

Phases

Andrej Trnka

Department of Mass Media Communication, Faculty of Mass Media Communication, University of SS. Cyril and Methodius, Trnava

917 01, Slovak Republic

Received: May 30, 2013 / Accepted: July 01, 2013 / Published: August 31, 2013.

Abstract: Six Sigma is a rigorous, focused, and highly effective implementation of proven quality principles and techniques. A company’s performance is measured by the sigma level of their business processes. Traditionally companies accepted three or four sigma performance levels as the norm. The Six Sigma standard of 3.4 problems-per-million opportunities is a response to the increasing expectations of customers. DMAIC is an acronym for five phases of Six Sigma methodology: Define, Measure, Analyze, Improve, Control. This paper describes possibility of using Bayesian Network for retraining data mining model. Concrete application of this proposal is in the field of the churn. Churn is a derivation from change and turn. It can be defined as a discontinuation of a contract. Data mining methods and algorithms can predict behavior of customers. We can get better results using Six Sigma methodology. The goal of this paper is proposal of implementation churn (with Bayesian network) to the phases of Six Sigma methodology.

Key words: Bayesian network, data mining, DMAIC, Churn, Six Sigma.

1. Introduction

Six Sigma methodology and its phases (Fig. 1) have

been widely adopted by industries and non-profit

organizations throughout the world. Six Sigma

methodology was first espoused by Motorola in the

mid-1980s. The successful implementation of the Six

Sigma program in Motorola led to huge benefits.

Motorola recorded a reduction in defects and

manufacturing time, and also began to reap financial

rewards. The Six Sigma has become the most

prominent trend in quality management not only for

manufacturing and service industries, but also for

non-profit organizations and government institutes

[1-5].

The main target of Six Sigma is to minimize

variation because it is somehow impossible to

Corresponding author: Andrej Trnka, Ph.D., research

fields: data mining, Six Sigma, statistical process control. E-mail: [email protected].

eliminate it totally. Sigma (σ) in the statistical field is

a metric used to represent the distance in standard

deviation units from the mean to a specific limit. Six

Sigma is a representation of six standard deviations

from the distribution mean. If a process is described as

within Six Sigma, the term quantitatively means that

the process produces fewer than 3.4 DPMO (defects

per million opportunities). Table 1 shows how

exponential the sigma scale is between levels 1 and 6

[6].

We can talk about Lean Six Sigma, too. Lean Six

Sigma for services is a business improvement

methodology that maximizes shareholder value by

achieving the fastest rate of improvement in customer

satisfaction, cost, quality, process speed, and invested

capital [7]. But for our research, we can ignore the fast,

so using the Six Methodology is proper.

In our previous research we implemented selected

data mining methods and algorithms to the DMAIC

phases of Six Sigma Methodology. The main area of

The Retraining Churn Data Mining Model in DMAIC Phases

1064

Fig. 1 Representation of Six Sigma methodology by BPMN.

Table 1 Six Sigma scale.

Sigma DPMO Efficiency (%)

1 691,462 30.9

2 308,538 69.1

3 66,807 93.3

4 6,210 99.4

5 233 99.98

6 3.4 99.9999966

Source: Ref. [6].

the implementation was manufacturing processes. But

Six Sigma methodology can be used in customer

services, too.

Some authors have used data mining algorithms in

manufacturing processes, but without Six Sigma

methodology [8-11].

Data mining is the process of discovering

interesting patterns and knowledge from large

amounts of data. The data sources can include

databases, data warehouses, the Web, other

information repositories, or data that are streamed into

the system dynamically [12]. One of the data mining

task is to predict the customer’s churn.

2. Churn

Mobile phone providers fight churn by detecting

patterns of behavior that could benefit from new

services, and then advertise such services to retain

their customer base. Incentives provided specifically

to retain existing customers can be expensive, and

successful data mining allows them to be precisely

targeted to those customers who are likely to yield

maximum benefit [13]. Churn is defined as a

discontinuation of a contract. Reducing churn is

important because acquiring new customers is more

expensive than retaining existing customers. In order

to manage customer churn to increase profitability,

companies need to predict churn behavior,

however, this problem not yet well understood

[14, 15].

Churning customers can be divided into two main

groups, voluntary and non-voluntary churners.

Non-voluntary churners are the easiest to identify, as

these are the customers who have had their service

withdrawn by the company. There are several reasons

why a company could revoke a customer’s service,

including abuse of service and non-payment of service.

Voluntary churn is more difficult to determine,

because this type of churn occurs when a customer

makes a conscious decision to terminate his/her

service with the provider. Voluntary churn can be


1065

sub-divided into two main categories, incidental churn

and deliberate churn.

Incidental churn happens when changes in

circumstances prevent the customer from further

requiring the provided service. Examples of incidental

churn include changes in the customer’s financial

circumstances, so that the customer can no longer

afford the service, or a move to a different

geographical location where the company’s service is

unavailable. Incidental churn usually only explains a

small percentage of a company’s voluntary churn.

Deliberate churn is the problems that most churn

management solutions try to battle. This type of churn

occurs when a customer decides to move his/her

custom to a competing company. Reasons that could

lead to a customer’s deliberate churn include

technology-based reasons, when a customer discovers

that a competitor is offering the latest products, while

their existing supplier can not provide them.

Economic reasons include finding the product at a

better price from a competing company. Examples of

other reasons for deliberate churn include quality

factors such as poor coverage, or possibly bad

experiences with call centers [16-17].

3. Data Mining Model

For our research we used IBM SPSS Modeler 14.

Telecommunications provider is concerned about the

number of customers it is losing to competitors.

Historic customer data can be used to predict which

customers are more likely to churn in the future.

These customers can be targeted with offers to

discourage them from transferring to another service

provider.

This model focuses on using an existing churn data

to predict which customers may be likely to churn in

the future and then adding the following data to refine

and retrain the model [18].

Fig. 2 shows the built model in IBM SPSS Modeler,

which contains the historical data.

In analysis we used two data sets. These data sets

had identical structure of variables. First data set

contained 412 rows (records) and the second data set

contained 451 rows.

Fig. 2 Summary model for churn.


1066

The first analysis with Feature Selection showed

that several variables were unimportant when

predicting churn. These variables were filtered from

data set to increase the speed of processing when the

model is built.

The step in analysis is using Bayesian networks to

predict the churn. A Bayesian network provides a

succinct way of describing the joint probability

distribution for a given set of random variables. In our

analysis we used Tree Augmented Naive Bayes. This

algorithm is used mainly for classification. It

efficiently creates a simple Bayesian network model.

The model is an improvement over the naive Bayes

model as it allows for each predictor to depend on

another predictor in addition to the target variable. Its

main advantages are its classification accuracy and

favorable performance compared with general

Bayesian network models. Its disadvantage is also due

to its simplicity; it imposes much restriction on the

dependency structure uncovered among its nodes [19].

After learning the model from first data set we

attached the second data set and we trained the

existing model.

4. Results

To compare and evaluate the generated models we

had to combine the two data sets. The generated

Bayesian Network model shows two columns. The

first column contains a network graph of nodes that

displays the relationship between the target and its

most important predictors. The second column

indicates the relative importance of each predictor in

estimating the model, or the conditional probability

value for each node value and each combination of

values in its parent nodes.

Fig. 3 shows relationship between the target

variable. Due to confidentiality of provider data, we

changed the names of variables and we used generic

names of variables.

Fig. 4 shows predictors (variables) importance.

To display the conditional probabilities for any

node, it is necessary to click on the concrete node and

the conditional probability is generated. Fig. 5 shows

conditional probability for most important

variable—variable 2.

To check how well each model predicts churn, we

used an analysis node. This node shows the accuracy

in terms of percentage for both correct and incorrect

predictions. The analysis shows that both models have

a similar degree of accuracy when predicting churn.

Tables 2-5 show results for output variable churn.

For the other view to data analysis we used

Fig. 3 Created Bayesian network.


1067

Fig. 4 Predictors importance.

Fig. 5 Conditional probability of variable 2.

Table 2 Comparing churn_1 with churn.

Total 863

Correct 654 75.78%

Wrong 209 24.22%

Table 3 Comparing churn_2 with churn.

Total 863

Correct 655 75.9%

Wrong 208 24.1%

Table 4 Agreement between churn_1 and churn_2.

Total 863

Correct 682 79.03%

Wrong 181 20.97%

Table 5 Comparing agreement with churn.

Total 682

Correct 565 82.84%

Wrong 117 17.16%

evaluation graph to compare the model’s predicted

accuracy by building a gains chart. Fig. 6 shows

evaluating model accuracy.

The graph shows that each model type produces

similar results. However, the retrained model

(churn_2) using both data sets is slightly better

because it has a higher level of confidence in its

predictions. Therefore, we used another algorithm of

Bayesian network—Markov Blanket.

The Markov Blanket [19] for the target variable

node in a Bayesian network is the set of nodes

containing target’s parents, its children, and its

children’s parents. Markov blanket identifies all the

Fig. 6 Evaluation graph of analysis (TAN Bayes Network).


1068

Fig. 7 Evaluation graph of analysis (Markov Blanket Bayes Network).

Fig. 8 Churn in proposed control phase.

variables in the network that are needed to predict the

target variable. This can produce more complex

networks, but also takes longer to produce. Using

feature selection preprocessing can significantly

improve performance of this algorithm.

Fig. 7 shows the same analysis, but with using

Markov Blanket algorithm. The evaluation graph

shows that churn_2 has higher level of confidence

than churn_2 with TAN Bayes Network.

5. Conclusions

The churn can be implemented to the DMAIC

phases of Six Sigma methodology. We suggest

implementing churn in to the control phase with

message event to the step process control. Fig. 8

shows proposed place of churn in control phase.

The red tasks and gateways represent our origin

proposal. The green task churn is the new proposed

task in Control phase of DMAIC.

Acknowledgments

This paper supports the project VEGA 1/0214/11.

Grateful acknowledgment for translating the English

edition goes to Juraj Mistina. The results of this article

were published in World Congress on Engineering

and Computer Science 2012, WCECS 2012, San

Francisco, USA, October 24-26, 2012.

References

[1] C.C. Yang, Six Sigma and Total Quality Management, in: Quality Management and Six Sigma, ed. A Coskun, Croatia, Sciyo, 2010.

[2] J. Antony, R. Banuelas, Key ingredients for the effective implementation of Six Sigma Program, Measuring Business Excellence 6 (2002) 20-27.

[3] H. Wiklund, P.S. Wiklund, Widening the Six Sigma concept: An approach to improve organizational learning, Total Quality Management 13 (2002) 233-239.

[4] L. Sandholm, L. Sorqvist, 12 requirements for Six Sigma success, Six Sigma Forum Magazine 2 (2002) 17-22.

[5] Ch.CH. Yang, An integrated model of TQM and GE Six Sigma, International Journal of Six Sigma and Competitive Advantage 1 (2004) 97-111.

[6] B. El-Haik, A. Shaout, Software Design for Six Sigma, John Wiley & Son, Hoboken, New Jersey, 2010.


1069

[7] W. Bentley, P.T. Davis, Lean Six Sigma Secrets for the CIO, CRC Press, Boca Raton, Florida, 2010.

[8] M. Kebisek, P. Schreiber, I. Halenar, Knowledge discovery in databases and its application in manufacturing, in: Proceedings of the International Workshop “Innovation Information Technologies: Theory and Practice”, Dresden, Sep. 6-20, 2010, pp. 204-207.

[9] R. Halenar, Matlab possibilities for real time ETL method, Acta Technica Corviniensis: Bulletin of Engineering 5 (2012) 51-53.

[10] P. Vazan, P. Tanuska, M. Kebisek, The data mining usage in production system management, World Academy of Science, Engineering and Technology 7 (2011) 1304-1308.

[11] M. Kudla, M. Stremy, Alternatívne metódy ukladania pološtruktúrovaných dát (Alternative methods for storing semi-structured data,), in: Applied Natural Sciences 2007, International Conference on Applied Natural Sciences, Trnava, Slovak Republic, Nov. 7-9, 2007, pp. 404-409.

[12] J. Han, M. Kamber, J. Pei, Data Mining Concepts and Techniques, Elsevier, Waltham, Massachusets, 2012.

[13] I. Witten, E. Frank, M. Hall, Data Mining Practical Machine Learning Tools and Techniques, Elsevier, Burlington, Massachusets, 2012.

[14] J. Ahn, S. Han, Y. Lee, Customer churn analysis: Churn determinants and mediation effects of partial defection in the Korean mobile telecommunications service industry, Telecommunications Policy 30 (2006) 552-568.

[15] K.Ch. Lee, N.Y. Jo, Bayesian network approach to predict mobile churn motivations: Emphasis on general Bayesian network, Markov blanket, and what-if simulation, in: Second International Conference, FGIT 2010, Jeju Island, Korea, Dec. 13-15, 2010.

[16] H. Kim, C. Yoon, Determinants of subscriber churn and customer loyalty in the Korean mobile telephony market, Telecommunications Policy 28 (2004) 751-765.

[17] J. Hadden, A. Tiwari, R. Rajkumar, D. Ruta, Churn prediction: Does technology matter?, International Journal of Electrical and Computer Engineering 1 (2006) 397-403.

[18] IBM SPSS Modeler 14.2 Applications Guide, IBM, 2011.

[19] IBM SPSS Modeler 14.2 Algorithms Guide, IBM, 2011.


Codebook Subsampling and Rearrangement Method for

Large Scale MIMO Systems

Xin Su1, Tianxiao Zhang2, Jie Zeng1, Limin Xiao1, Xibin Xu1 and Jingyu Li1

1. Tsinghua National Laboratory for Information Science and Technology, Research Institute of Information Technology, Tsinghua

University, Beijing 100000, China

2. School of Advanced Engineering, Beihang University, Beijing 100000, China

Received: April 28, 2013 / Accepted: June 01, 2013 / Published: August 31, 2013.

Abstract: In large scale MIMO (multiple-input multiple-output) systems, the size of codebook increases greatly when transmitters and receivers are equipped with more antennas. Thus, there are demands to select subsets of the codebook for usage to reduce the huge feedback overhead. In this paper, we propose a novel codebook subsampling method using chordal distance of different codewords and deleting them to affordable payload of PUCCH (physical uplink control channel). Besides, we design a related codebook rearrangement algorithm to mitigate the system performance loss when there are bit errors in the feedback channel.

Key words: Large scale MIMO, codebook subsampling, codebook rearrangement, PMI feedback.

1. Introduction

The explosive growth of wireless data service calls

for new spectrum resource. Meanwhile, the available

spectrum for further wireless communication systems

is very limited and expensive. Since the capacity of a

MIMO (multiple-input multiple-output) system

greatly increases with the minimum number of

antennas at the transmitter and receiver sides under

rich scattering environments [1], the large scale

MIMO [2] shown in Fig. 1 is one of the most

important techniques to address the issue of

exponential increasing in wireless data service by

using spatial multiplexing and interference mitigating.

For the consideration of practical application, the

number of antennas on the terminal side is restricted,

and thus the number of multiplexing layers is limited

though the number of antennas on the BS (base station)

could be very large. As a result, we should explore the

Corresponding author: Jie Zeng, M.Sc., engineer, research

fields: broadband wireless access, software defined radio, 4G/B4G technology and standardization. E-mail: [email protected].

large scale MIMO system potentials by utilizing

beamforming technologies. The performance of

beamforming relies on the accuracy of precoding.

However, the size of codebook can be very large

when antennas are increased, considering that the

payload capacity of PUCCH (physical uplink control

channel) is limited to 11 bits [3]. To decrease the

overhead in CSI (channel state information) feedback,

the choice of codebook subsampling for transmission

is necessary [4, 5].

Fig. 1 The close-loop MIMO system model.

Codebook Subsampling and Rearrangement Method for Large Scale MIMO Systems

1071

Several subsampling methods have been proposed

in Rel-10 in 2, 4, 8 Tx scenario. The subset selection

in this case naturally corresponds to the reduction of

granularity in direction and/or phase offset [6], such

as uniform subsampling or staggered subsampling

which keeps better granularity. However, the

codebook design for large scale MIMO system may

not be based on direction for each polarization and

phase offset between polarizations; hence the

application above for large scale MIMO is restricted.

In this case, we propose a novel codebook

subsampling method which applies to all kinds of

codebook design, in which we select the subset of

codebook using chordal distance of different

codewords and delete them to affordable payload of

PUCCH. In addition, to further optimize the

performance of PMI (precoding matrix indication)

when errors occur, we propose the codebook

rearrangement method to decrease the impact of

mismatch between PMI and the channel.

This paper is organized as follows: Section 2

introduces the model of the precoding system. Section

3 presents the codebook subsampling method and the

codebook rearrangement method. Section 4 shows the

simulation results. And finally, Section 5 concludes

the paper.

2. System Model

2.1 System Model

In this paper, we discuss about a close-loop MIMO

system withtN transmit antennas and

rN receive

antennas depicted in Fig. 1. For massive MIMO

system tN

could be very large.

Firstly, the data vector S in the system is

de-multiplexed into L streams. L is limited

by 1 min( , )t rL N N . When L = 1 we call the

transmission beamforming, while L > 1 we call it

multiplexing. After the data vector s is preprocessed

by an Nt × L precoding matrix Wi, we get a Nt × 1

signal vector x for the tN

antennas to transmit:

x Ws (1)

Thus, after x passing the channel and being added

the noise, we will get the received signal y, which can

be expressed as:

iy HW s v (2)

where, H ( r tN NH C ) denotes the fading channel matrix

with its entry ijH

denoting the channel response

from the thj transmit antenna to the thi receive

antenna, and v denotes the white complex Gaussian

noise vector with covariance matrix 0 rNN I .

The precoding matrix is selected from the

predesigned codebook which is known to the

transmitter and the receiver. Taking the downlink as

an example, when UEs have received the pilots from

the BS, the receiver can choose the optimal codeword

after the channel estimation. Then the receiver reports

the PMI with limited bits to BS [7] through the uplink

channel. If the feedback is limited to B bits, the size of

codebook satisfies BN 2 . Thus the transmitter can

retrieve the precoding matrix and perform the

precoding.

2.2 Kerdock Codebook

The basic idea of the Kerdock codebook design is

utilizing the feature of MUB (mutually unbiased bases)

to construct precoding matrices. The main

characteristic of the Kerdock codebook is that all the

elements of the matrix are ±1 or ±j. Hence, the

Kerdock codebook has some advantages, such as low

requirement for storage, low computational

complexity for codeword search, and the simple

systematic construction.

The MUB property is described as follows:

1{ ,..., }tNS s s ,

1{ ,..., }tNU u u are two orthonormal

bases with sizet tN N . If the column vectors drawn

from S and U satisfy , 1 /i j ts u N , we can say

that they have the mutually unbiased property [8].

An MUB is the set 1{ ,..., }

tNS s s satisfying the

mutually unbiased property. The Kerdock codebook

has several construction methods such as


1072

Sylvester–Hadamard construction and power

construction. In this paper, we use the

Sylvester–Hadamard construction:

First, we construct the generating matrices Dn

(t tN N

diagonal matrices with ±1, ±j elements) for n

= 0, 1, 2…Nt–1 according to Ref. [9].

Then we construct the corresponding orthonormal

matrix: 1 ˆ , 0,1,..., 1

tn n N t

t

W D H n NN

(3)

where, ˆtNH is the

t tN N Sylvester–Hadamard

matrix:

2 2ˆ ˆ ˆ ....

tNH H H

(4)

where 2

1 1ˆ1 1

H

For the beamforming, we can construct the

codebook by selecting each column of all the bases as

the precoding vector: { }{1} {2}

1 0 2 0 1{ , ,..., }t

t

N

N NC f W f W f W (5)

And for an L-layer spatial multiplexing codebook,

the largest codebook is derived by taking all L-column

combinations from each nW .

2.3 Codeword Search

We can choose the optimal main codeword from

1 2{ , ,..., }NK K K through the estimate of the channel.

The codebook is shared by the transmitter and

receiver.

Codeword selection criteria: for 1-layer

beamforming, the beamformer that minimizes the

probability of symbol error for maximum ratio

combining receiver is expressed as [10]: 2

2ˆ[ ] arg max [ ]

f Cf i H i f

(6)

where, f denotes a 1tN matrix. For spatial

multiplexing with a zero forcing receiver, the

minimum singular value selection criterion is

expressed as:

minˆ[ ] arg max { [ ] }

K CF i H i K

(7)

where, min

denotes the minimum singular value of

the argument. This selection criterion approximately

maximizes the minimum sub-stream SNR

(signal-to-noise ratio).

3. The Novel Codebook Subsampling and Rearrangement Method

3.1 Codebook Subsampling

To pursue the maximum SNR, we select the

codeword with the smallest chordal distance from the

transmission channel. The basic idea of the codebook

subsampling method is to delete one codeword of the

codeword pairs which have the smallest chordal

distance. The chordal distance between two precoding

vector is represented by 2

( , ) 1,

chord i j

i j

i j

d f ff f

f f

(8)

with ||.|| being the norm of the vector. If the chord

distance between two codewords is the smallest

among the codewords pool, we may reserve only one

of them and delete another. Therefore, we could

decrease the overhead as well as remain the

performance of precoding at the utmost.

The process of subsampling is shown as follows:

(1) Suppose the codebook includes K codewords.

Divide the codewords into g groups.

(1-1) Compute the chordal distance between any

two codewords ( , )i jd f f . Choose if and

jf as

reference codewords if their distance is the largest.

(1-2) Compute the chordal distance between the rest

( (0, ], ! & & ! )Lf L K L i L j and reference

codeword. If ( , ) ( , )i L L jd f f d f f , put Lf in if ’s

group. Otherwise put Lf in jf ’s group.

(1-3) Repeat the procedure until the number of

groups is g.


1073

(2) Delete codewords and related PMI.

(2-1) Compute the chordal distance between any two codewords in ),,2,1( gll th group. Find the

min ( , )l i jd f f

(2-2) Choose the min{min ( , )}( 1,2, , )l i jd f f l g

as the codeword pair to deal with (suppose in the thm

group).

(2-3) Compute the chordal distance between the rest ( ! & & ! )Lf L i L j in thm group and reference

codeword. If ( , ) ( , )i L L jd f f d f f , delete if ,

otherwise delete jf .

(2-4) Select the new min ( , )m i jd f f . Back to (a)

until the number of codewords satisfies the

requirement of PMI feedback.

The summary of the algorithm is given in Table 1.

3.2 Codebook Rearrangement

The error in PMI feedback could lead to severe

mismatch of precoding vector and user’s channel,

thus greatly decreasing transmission gain and

increasing unreliability. By decreasing the mismatch

caused by PMI transmission error, we could

compensate for the performance loss, even the

precoding vector is not optimal. Therefore, we

rearrange the PMI, reduce the Hamming distance of

binary indexes of codewords with high correlation.

Consider one bit error in PMI feedback. When the

two indexes with one bit of Hamming distance are

arranged to codewords with high correlation, even if

the error occurs and the base station uses the wrong

codeword, the wrong codeword could still perform

well due to the high correlation with the correct one,

thus ensuring the compatibility with the channel and

decreasing the gain loss.

The process of subsampling is presented in Table 2

and the description is given as follows:

Table 1 Simulated subsampling algorithm.

//K: the total number of codewords //B: cycling times //N=2B : groups of codewords //n: current number of groups //dis[i][j]: matrix of chordal distance between codewords n=1; loop for B cycles loop for n cycles calculate any of two codewords fi, fj with dis[i][j] (fi, fj the nk group) get the two codewords fi, fj with the largest codewords distance in the nk group if dis(i,t)<dis(j,t) [tsize(nk)，t≠i,j] then allocate the codeword to the group 2*(nk-1) else allocate to the group 2*nk

end if end loop n=n*2 end loop loop for N cycle calculate the nk group any of two codewords fi, fj with dis{nk}[i][j] end loop while(K>payload) dis(nk)=min(dis{nk}(:)) m=argmin(dis(nk)) In the m group if ∑dis(i,t)<∑dis1(j,t) then delete the codeword fi in the m group else delete the codeword fj in the m group end if renew the m group with a minimum distance end while

Table 2 Simulated rearrangement algorithm.

//G=log2(payload) //PMI_B : groups of PMI based on code weight loop for payload cycles restore each PMI in PMI_B{i} with code weight i; end loop D0 =f0 for i=1:G for j=1:size(PMI_B{i}) Pij

=PMI_B{i}( j)

Uji-1={ fji-1

, d(Pi-1,Pij)=1}

ji’=argmin∑d(fm,Uji-1

) (fm the nk group)

Dji=f ji’

if the codeword in the nk group all have been allocated new PMI then go to next group else continue in this group end if end end


1074

(1) Divide PMI into B PMI groups based on binary

code weight. if denotes the original codeword

associated with PMI iw , and

iU denotes the new one.

Let 0 0U f .

(2) Select one ibw ,

in ),,2,1( Bbbth PMI

group. Find all thejbw ,1in thb )1( PMI group that

1),( ,1, jbibbinary wwd . (3) In ),,2,1( gll th codeword group, compute

1,( , ), thk b j k

j

d f f f l , if ' 1, 1,( , ) min ( , )b j k b jk

j j

d f f d f f

( thkf l codeword group), then ',b i k

U f .

(4) If all codewords in thl codeword group are

rearranged with new PMI, turn to thl )1( codeword

group.

(5) If all PMI in thb PMI group are rearranged

with new codeword, turn to thb )1( PMI group.

4. Simulation Results

This section we present the simulation results under

the configuration given in Table 3. The simulation

procedure follows the system model in Fig. 1.

Fig. 2 shows the performance of downlink

transmission BER vs. feedback error probability. The

feedback error probability means the probability of

each bit-error occurs in the PMI feedback, and the

downlink transmission BER means the bit error rate of

the downlink transmission. From the results, we can

see the codebook after subsampling has significant

BER performance gain compared with the original

Kerdock codebook, because of the low probability of

Table 3 Simulation parameters.

Parameter Value

Frequency 2.1 GHz

System bandwidth 10 MHz

Channel modelling i.i.d, CN(0, 1)

Number of BS antennas 32

Number of UE antennas 1

Channel estimation Ideal

UE receiver MMSE

SNR 10 dB

10-3

10-2

10-5

10-4

10-3

10-2

10-1

SNR = 10 dB

Feedback error probability

Tra

ns.

BE

R

Kerdock

Sampled

Proposed

Fig. 2 The BER performance of different codebooks.

error occurrence due to the fewer bits for feedback.

And the proposed codebook after rearrangement has

the further BER performance gain since to configure

high correlation codewords with reduced code

distance, we decrease the performance loss of system

when the mismatch of precoding vector and the

channel occurs.

5. Conclusions

In this paper, we proposed a novel codebook

subsampling method based on chordal distance as

well as the related codebook rearrangement algorithm

for codebook designs in large scale MIMO system.

The codebook subsampling method can reduce the

feedback overhead without impacting the system

performance, and the rearrangement algorithm can

significantly mitigate the system performance loss

when errors in the feedback channel occur. Simulation

results show that the Kerdock codebook after

subsampling and rearrangement has significant

performance gain under the non-ideal uplink feedback

channel in large scale MIMO system.

Acknowledgment

This work was supported by Beijing Natural

Science Foundation Funded Project (No.4110001),

National S&T Major Project (No. 2011ZX03003-002),

Tsinghua Independent Research (No. 2010TH203-02)

and Samsung Company.


1075

References

[1] E. Telatar, Capacity of multi-antenna gaussian channels, European Transactions on Telecommunications 10 (1999) 585-595.

[2] F. Rusek, D. Persson, B.K. Lau, E.G. Larsson, T.L.

Marzetta, O. Edfors, et al., Scaling up MIMO:

opportunities and challenges with very large arrays, IEEE

Signal Processing Magazine 30 (2012) 40-60.

[3] 3GPP TS 36.213: Evolved universal terrestrial radio

access (E-UTRA), Physical layer procedures,

pp. 56-64.

[4] R1-104164 Way forward on 8Tx codebook for Rel.10 DL

MIMO, CATT, Ericsson, LG Electronics, Mitsubishi

Electric, Nokia, Nokia Siemens Networks, NTT DoCoMo,

Panasonic, Sharp, ST-Ericsson, Texas Instruments,

RAN1 #61bis, June 2010, Dresden, Germany.

[5] R1-104259 Way Forward on CSI Feedback for Rel.10 DL MIMO Alcatel-Lucent, Alcatel-Lucent Shanghai Bell,

CATT, Ericsson, Huawei, Marvell, Mitsubishi Electric, NEC, Nokia, Nokia Siemens Networks, NTT DoCoMo, Panasonic, Samsung, Sharp, ST-Ericsson, Texas Instruments, ZTE, RAN1 #61bis, June 2010, Dresden, Germany.

[6] R1-104901 8Tx Codebook Subsampling, Panasonic, August 2010, Madrid, Spain.

[7] 3GPP TS 36.211: Evolved Universal Terrestrial Radio Access (E-UTRA), Physical channels and modulation, pp. 17-20.

[8] A. Klappenecker, M. Roetteler, Constructions of mutually unbi-ased bases, Finite Fields Appl. 2948 (2004) 137-144.

[9] R.W. Heath Jr., T. Strohmer, A.J. Paulraj, On quasi-orthogonal signatures for CDMA systems, IEEE Trans. Inf. Theory 52 (2006) 1217-1226.

[10] D.J. Love, R.W. Heath Jr., Limited feedback unitary precoding for spatial multiplexing systems, IEEE Trans. Inf. Theory 51 (2005) 2967-2976.


A High-Precision Time Handling Library

Irina Fedotova1, Eduard Siemens2 and Hao Hu2

1 Faculty of Information Science and Computer Engineering, Siberian State University of Telecommunication and Information

Sciences, Novosibirsk 630102, Russia

2 Faculty of Electrical, Mechanical and Industrial Engineering, Anhalt University of Applied Sciences, Koethen 06366, Germany

Received: April 28, 2013 / Accepted: June 01, 2013 / Published: August 31, 2013.

Abstract: An appropriate assessment of end-to-end network performance presumes highly efficient time tracking and measurement with precise time control of the stopping and resuming of program operation. In this paper, a novel approach to solving the problems of highly efficient and precise time measurements on PC-platforms and on ARM-architectures is proposed. A new unified High Performance Timer and a corresponding software library offer a unified interface to the known time counters and automatically identify the fastest and most reliable time source, available in the user space of a computing system. The research is focused on developing an approach of unified time acquisition from the PC hardware and accordingly substituting the common way of getting the time value through Linux system calls. The presented approach provides a much faster means of obtaining the time values with a nanosecond precision than by using conventional means. Moreover, it is capable of handling the sequential time value, precise sleep functions and process resuming. This ability means the reduction of wasting computer resources during the execution of a sleeping process from 100% (busy-wait) to 1-1.5%, whereas the benefits of very accurate process resuming times on long waits are maintained.

Key words: High-performance computing, network measurement, timestamp precision, time-keeping, wall clock.

1. Introduction

Estimation of the achieved quality of the network

performance requires high-resolution, low CPU-cost

time interval measurements along with an efficient

handling of process delays and sleeps [1, 2]. The

importance on controlling these parameters can be

shown on the example of a transport layer protocol. Its

implementation may need up to 10 time fetches and

time operations per transmitted and received data

packet. However, performing accurate time interval

measurements, even on high-end computing systems,

faces significant challenges.

Even though Linux (and in general UNIX timing

subsystems) uses auto-identification of the available

hardware time source and provides nanosecond

resolution, these interfaces are always accessed from

Corresponding author: Fedotova Irina, M.Eng., research

fields: hard real-time system, for OS Linux. E-mail: [email protected].

user space applications through system calls. Thus it

costs extra time in the range of up to a few

microseconds—even on contemporary high-end PCs

[3]. Therefore, direct interaction with the timing

hardware from the user space can help to reduce time

fetching overhead from the user space and to increase

timing precision. The Linux kernel can use different

hardware sources, whereby time acquisition

capabilities depend on the actual hardware

environment and kernel boot parameterization. While

the time acquisition of some time sources costs up to 2

microseconds, others need about 20 nanoseconds. In

the course of this work, a new High Performance

Timer and a corresponding library HighPerTimer have

been developed. They provide a unified user-space

interface to time counters available in the system and

automatically identify the fastest and the most reliable

time source (e.g., TSC (time stamp counter) [4, 5] or

HPET (high-performance event counter) [6, 7]). In the


1077

context of this paper, the expression time source

means one of the available time hardware or

alternatively the native timer of the operating system,

usually provided by the standard C library.

Linux (as well as other UNIX operating systems)

faces a significant problem of inaccurate sleep time,

which is known for many years, especially in older

kernel versions, when Linux has provided a process

sleep time resolution of 10 m. This leads to a

minimum sleep time of about 20 ms [8]. Even

nowadays, when Linux kernels usually reduce this

resolution down to 1 m, waking up from sleeps can

take up to 1-2 m. With kernel 2.6 the timer handling

under Linux has been changed significantly. This

change has reduced the wakeup misses of sleep calls

to 51 µs on average and to 200-300 µs in peaks.

However, for many soft-real-time and

high-performance applications, this reduction is not

sufficient. Presented High Performance Timer not

only significantly improves the time fetching accuracy,

but also addresses the problem of those imprecise

wakeups from sleep calls under Linux.

These precision issues lead to the fact that, for

high-precision timing within state machines and

communication protocols, busy-waiting loops are

currently commonly used for waits, preventing other

threads from using the given CPU. The approach of

the High Performance Timer library aims at reducing

the CPU load down to an average of 1-1.5% within

the sleep calls of the library and at raising the wakeup

precision to 70-160 ns. Reaching these values enables

users of this library to implement many protocols and

state machines with soft real-time requirements in

user space.

The remainder of the paper is organized as follows.

In Section 2, related work is described. Section 3

shows the specific details of each time source within

the suggested single unified High-Performance Timer

class interface. In Section 4, we briefly describe the

implemented library interface. Some experimental

results of identifying appropriate timer source along

with their performance characteristics are shown in

Section 5. In Section 6, precise process sleeping

aspects are shown. Finally, Section 7 describes next

steps and future work in our effort to develop a tool

for highly efficient high-performance network

measurements.

2. Related Work

Since the problem of inefficient time keeping in

Linux operating system implementation has become

apparent, several research projects have suggested to

access the timing hardware directly from user space [1,

9-10]. However, most of this research considers

handling of a single time hardware source only,

predominantly the Time Stamp Counter [1, 9, 11].

Other solutions provide just wrappers around

timer-related system calls and so inherit their

disadvantages such as the high time overhead [12, 13].

In other proposals, the entire time capturing process is

integrated into dedicated hardware devices [14, 15].

Most of this research focuses only on a subset of the

problems, addressed in this work. Our work with the

HighPerTimer library improves timing support by

eliminating the system call overhead and also by

application of more precise process sleep techniques.

3. Unified Time Source of the HighPerTimer Library

While most of the current software solutions on

Linux and Unix use the timing interface by issuing

clock_gettime() or gettimeofday() system calls,

HighPerTimer tries to invoke the most reliable time

source directly from the user space. Towards the user,

the library provides a unified timing interface for time

period computation methods along with sleep and

wakeup interfaces, independently from the used

underlying time hardware. So, the user sees a “unified

time source” that accesses the best possible on the

underlying hardware, and that generally avoids system

call overheads. The HighPerTimer interface supports

access the mostly used time counters: TSC, HPET and,


1078

as the last alternative, the native timer of the operating

system, through one of the said Unix system calls.

The latter time source we call the OS Timer.

Using the Time Stamp Counter is the fastest way of

getting CPU time. It has the lowest CPU overhead and

provides the highest possible time resolution,

available for the particular processor. Therefore, in the

context of our library, the TSC is the most preferable

time source. In newer PC systems, the TSC may

support an enhancement, referred to as an Invariant

TSC feature. Invariant TSC is not tightly bound to a

particular processor core and has, in contrary to many

older processor families, a guaranteed constant

frequency [16]. The presence of the Invariant TSC

feature in the system can be tested by the Invariant

TSC flag, indicated by the cpuid processor instruction.

For most cases, the presence of this Invariant TSC

flag is essential in order to accept it as a

HighPerTimer source.

Formerly referred by Intel as a Multimedia Timer

[7], the High Precision Event Timer is another

hardware timer used in personal computers. The

HPET circuit is integrated into the south bridge chip

and consists of a 64-bit or 32-bit main counter register

counting at a given constant frequency between 10

MHz and 25 MHz. Difficulties are faced when the

HPET main counter register is running in 32-bit mode

because overflows of the main counter arise at least

every 7.16 min. With a frequency of 25 MHz, register

overflows would occur even within less than 3 min.

So, time periods longer than 3 min can not reliably

measured in 32 bit mode. So, in the HighPerTimer

library, we decided to generally avoid using the HPET

time source in case of a 32-bit main counter.

For systems, on which neither TSC nor HPET are

accessible or TSC is unreliable, an alternative option

of using the OS Timer is envisaged. This alternative is

a wrapper issuing the system call clock_gettime().

This source is safely accessible on any platform.

However, it has the lowest priority because it issues

system calls, with their time costs of up to 2

microseconds in worst case [17, 18]. Depending on

the particular computer architecture and used OS,

these costs can be less due to the support of the

so-called virtual system calls. These calls provide

faster access to time hardware and avoid expensive

context switches between user and kernel modes [19].

Nevertheless, invocation of clock_gettime() through a

virtual system call is still slower than the acquisition

time value from current time hardware directly. The

difference between getting the time value using virtual

system calls and getting the time values directly from

the hardware is about 3 to 17 ns, as measurement

results, discussed in Section 5, show.

4. The HighPerTimer Interface

The common guidelines on designing any interfaces

cover efficiency, encapsulation, maintainability and

extensibility. Accordingly, the implementation of the

HighPerTimer library pays particular attention to

these aspects. Using the new C++11 programming

language standard [20], the library achieves high

efficiency and easy code maintainability. Furthermore,

regarding the platform-specific aspects, HighPerTimer

runs on different 64-bit and 32-bit processors of Intel,

AMD, VIA and ARM, and considers their general

features along with specialties of time keeping.

However, some attention must be paid to obtaining

a clean encapsulation of hardware access when using

C++. For this encapsulation, the HighPerTimer library

comprises two header files and two implementation

files called HighPerTimer and TimeHardware. Each

of them contains three classes. HighPerTimer files

contain HighPerTimer, HPTimerInitAndClear and

AccessTimeHardware classes, as described below. In

TimeHardware files, the classes TSCTimer,

HPETTimer and OSTimer corresponding to the

respective time sources TSC, HPET and the OS

source have been implemented. Through an assembly

code within the C++ methods, they provide direct

access to the timer hardware, initialize the respective

timer source, retrieve their time value and are at only


1079

HighPerTimer class’s disposal. Dependencies

between the classes are presented in Fig. 1.

TSCTimer, HPETTimer and OSTimer classes have

a “friend” relationship with the HighPerTimer class,

which means that HighPerTimer places their private

and protected methods and members at friend classes’

disposal. For safety and security reasons, we protect

the hardware access from use by application users

directly and permit access only from special classes.

An AccessTimeHardware class provides a limited

read-only access to some information on CPU and

specific time hardware features, obtained in a

protected interface. For example, some advanced

users can find out failure reasons of the initialization

routine of the HPET device and get a corresponding

error message:

std::cout <<

AccessTimeHardware::HpetFailReason();

However, all the routines of time handling along

with access to the actual timer attributes such as clock

frequency are accessed by the library users via the

HighPerTimer class. For interfacing with other time

formats, HighPerTimer class provides a set of

constructors that sets its object to the given time

provided in seconds, nanoseconds or in the native

clock ticks of the used time source. Via specific

constructor, a time value in a Unix-specific time

format [21] can also be assigned to a HighPerTimer

object. The current time value is retrieved using the

following piece of code:

// declare HighPerTimer objects

HighPerTimer timer1, timer2;

HighPerTimer::Now (timer1);

// measured operation

HighPerTimer::Now (timer2);

Comparison operators allow effective comparison

to be performed using the main counter values. Some

of these methods are declared as follows:

bool operator >= (const HighPerTimer & timer)

const;

bool operator<= (const HighPerTimer& timer)

const;

bool operator!= (const HighPerTimer& timer)

const;

Fig. 1 Simplified class diagram of HighPerTimer library.


1080

The user can also set the value of a timer object

explicitly to zero and add or subtract the time values

in terms of timer objects, tics, nanoseconds or seconds.

Since the main “time” capability of a timer object is

kept in the main counter only, the comparison

operations between timer objects, as well as

arithmetical operations on them, are nearly as fast as

comparisons and elementary arithmetical operations

on two int 64 variables. Recalculations between tics,

seconds, microseconds and nanoseconds are only

done in the “late initialization” fashion when string

representations of the timer object or seconds,

microseconds or nanoseconds of the object are

explicitly requested via the class interface:

// subtract from timer object

HighPerTimer & SecSub (const uint64_t Seconds);

HighPerTimer & USecSub (const uint64_t

USeconds);

HighPerTimer & NSecSub (const uint64_t

NSeconds);

HighPerTimer & TicSub (const uint64_t Tics);

// add to timer object

HighPerTimer & SecAdd (const uint64_t Seconds);

HighPerTimer & USecAdd (const uint64_t

USeconds);

HighPerTimer & NSecAdd (const uint64_t

NSeconds);

HighPerTimer & TicAdd (const uint64_t Tics);

Assignment operators allow a HighPerTimer object

to be set from the Unix-format of time values -

timeval or timespec structs [21]. Both of these

structures represent time, elapsed since 00:00:00 UTC

on 01.01.1970. They consist of two elements: the

number of seconds and the rest of the elapsed time

represented either in microseconds (in case of timeval)

or in nanoseconds (in case of timespec):

struct timeval {

long tv_sec; /* seconds */

long tv_usec; /* microseconds */

}

struct timespec {

long tv_sec; /* seconds */

long tv_nsec; /* nanoseconds */

};

Assignment to these structures is also possible with

HighPerTimer objects through copying or moving:

const HighPerTimer & operator= (const struct

timeval & TV);

const HighPerTimer & operator= (const struct

timespec & TS);

const HighPerTimer & operator= (const

HighPerTimer & Timer);

HighPerTimer & operator= (HighPerTimer &&

Timer);

This way, the HighPerTimer library provides a fast

and efficient way to handle time values by operating

main counter value and seconds and nanoseconds

values only on demand. It also relieves users from the

manual handling of specific two-value structures such

as timeval or timespec.

However, for the whole routine of handling time

values, some central parameterization of the library

must be performed at the initialization time of the

library. Primarily, this is the initialization of the

HighPerTimer source, which is accomplished on the

basis of the appropriate method calls from the

TimeHardware file. Especially, InitHPTimeSource()

calls InitTSCTimer() and InitHPETTimer() methods,

which attempt to initialize respective time hardware

and return true on success or false on failure ( Fig. 1).

Before using any timer object, the following global

parameters must be measured and set: the frequency

of the main counter as a double precision floating

point value and as a number of ticks of the main

counter within one microsecond, the value of the shift

of the main timer counter against Unix Epoch, the

maximum and minimum values of HighPerTimer for

the given hardware-specific main counter frequency,

and the specified HZ frequency of the kernel. The

value of HZ is defined as the system timer interrupt

rate and varies across kernel versions and hardware

platforms. In the context of the library, the value of


1081

HZ is used for the implementation of an accurate sleep

mechanism, see Section 6. The strict sequence of the

initialization process is determined within an

additional HPTimerInitAndClean service class (Fig. 1)

by invoking corresponding HighPerTimer

initialization methods through their “friend”

relationship. A strict order of initialization of the

given global variables must be assured, which is

somewhat tricky since all the variables must be

declared static and must be initialized before entering

the main routine of the application.

Despite the advantage of automatic detection of

the appropriate time source, situations sometimes

arise when an application programmer prefers to use a

different time source than the one automatically

chosen at library initialization time. To account for

this, a special ability to change the default timer is

provided. This change causes a recalculation process

for most of the timer attributes:

// create variable for a new value of time source

TimeSource MySource;

MySource = TimeSource::HPET;

HighPerTimer::SetTimerSource (MySource);

However, since this change leads to invalidation of

all the already existing timer objects within the

executed program, this feature should be used with

caution and only at the system initialization time, and

definitely before instantiation of the first

HighPerTimer object.

5. Time Fetching Performance Results

Table 1 shows the performance results when getting

the time values using the HighPerTimer library as

measured on different processor families. The mean

and standard deviation values of the costs of setting a

HighPerTimer object are shown. For this investigation,

time was fetched in a loop of 100 million consecutive

runs and set to a HighPerTimer object. Since we are

interested here in measuring the time interval between

two consecutive time fetches only, without any

interruption in between, we filter out all outlying

peaks. These peaks are most probably caused by

process interruption by the scheduler or by an

interrupt service routine. Thus, filtering out such

outliers allows us to get rid of the bias caused by

physical phenomena, which are outside the scope of

this investigation.

The following two examples demonstrate the

behavior of HighPerTimer sources in more detail and

allow a comparison of their reliability and costs

depending on the particular processor conditions.

Although Table 1 shows the results for all three

processors, later investigations are shown only for less

powerful systems. It makes sense to examine in more

depth those systems, where for example, TSC is

unstable or does not possess Invariant TSC flag

(Section 3).

In the first case, processor VIA Nano X2 has TSC

as a current time source. Costs of time fetching here

are about 38 ns. Since TSC source has the highest

priority and has been initialized successfully, the

HPET device check is not necessary and so omitted

here. Moreover, on this processor, the Linux kernel is

also using TSC as its time source and so, within the

clock_gettime() call, the kernel is also fetching the

TSC register of the CPU. Fig. 2 shows the relation

between the TSC, OS and HPET timers on this

processor. Similarity between TSC and OS costs are

seen very clearly. As seen in Table 2, the difference

between the mean value of time fetching between OS

Timer and TSC Timer is 64 ns. Each system call with

a context switch would last at least ten times longer,

Table 1 Costs of setting timer on different processors.

Processor (CPU) Time source Mean, ns St. deviation, ns

Intel ® Core ™ i7-2600, 1600 MHz TSC 16.941 0.1231

VIA Nano X2 U4025, 1067 MHz TSC 38.203 0.3134

Athlon ™ X2 Dual Core BE-2350, 1,000 MHz HPET 1,063.3 207.92


1082

Fig. 2 Measurements of TSC, HPET and OS Timer costs on the VIA Nano X2 processor.

Table 2 Mean and standard deviation values of HPET, TSC and OS Timer costs on the VIA Nano X2 processor.

Timer source Mean, ns. Standard deviation, ns.

TSC Timer 38.23 0.3134

HPET Timer 598.72 76.015

OS Timer 102.20 0.5253

thus we can conclude that, on this system, a virtual

system call is issued by clock_gettime() instead of a

real system call with a context switch. HPET source

for the library can be set by the static method

HighPerTimer::SetTimerSrouce. However, we would

expect here much slower time operations, as seen in

Table 2.

The next example illustrates another case of a

dependence on the OS Timer from the current time

source. For the processor AMD Athlon X2 Dual Core,

the TSC initialization routine fails because TSC is

unstable here. However, since the HPET device is

accessible, there are two more options for the time

source for HighPerTimer–HPET or OS Timers - and it

is necessary to check the mean costs of getting the

ticks of both timers.

Although the mean value of time fetching for TSC

can be significantly lower than for HPET, the

HighPerTimer library considers the TSC to be a

non-stable, unreliable time source since the Invariant

TSC flag (Section 3 above) is not available and the

TSC constancy is not identified by additional library

checks. So, it must be assumed that TSC frequency

changes from time to time due to power saving or

other techniques of the CPU manufacturers. In the

next step, HPET and OS Timer characteristics must be

considered. The difference between the mean values

of HPET and OS Timer is about 54.1 ns, which is not

enough for a system call with a context switch. Thus

we conclude that clock_gettime() also uses the HPET

timer and passes it to the user via a virtual system call.

However, to provide an appropriate level of reliability,

we also evaluate numbers through their deviation

values. For this evaluation, a threshold for the

difference of mean values was chosen. When the

difference of the mean values of HPET and OS Timer

is no more than 25%, we also take into account

standard deviation values of time fetching and so

check the temporal stability of the considered time

source. Consequently, when the mean time fetching

value of the two time sources is similar, the

HighPerTimer library would give precedence to the

time source with a less standard deviation of the time

fetching costs.

Table 3 Mean and standard deviation values of HPET, TSC and OS Timer costs on the AMD Athlon processor.

Timer source Mean, µs Standard deviation, µs

TSC Timer 0.0251 0.0015

HPET Timer 1.0633 0.2079

OS Timer 1.1174 0.3743


1083

Fig. 3 Measurements of TSC, HPET and OS Timer costs on the AMD Athlon processor.

6. Precise Process Sleeping Aspects

For process sleeping or suspension, Linux provides

the sleep function (implemented in the standard C

library). Dependent on the sleep duration, the function

either suspends from the CPU or waits in the

busy-waiting mode (sometimes also called spinning

wait). However, measurements performed in this work

revealed that the sleep function of the standard C

library misses the target wake-up time by more than

50 ms on average. Such an imprecision however is

unacceptable for high-accuracy program sleeps. By

comparison, pure busy wait implementations within

an application miss the target return time by about 100

ns, but keep the CPU busy throughout the wait time.

Unlike the C library’s sleep call, the sleep of the

HighPerTimer library combines these two ways of

sleeping. It has very short miss times on waking up

with a minimum CPU utilization at the same time.

This improvement provides a big competitive

advantage over the predecessor solutions.

HighPerTimer provides a wide range of functions

for making a process sleep. For example, the user can

define the specific sleep time, given purely in seconds,

in microseconds or nanoseconds. A process

suspension with a nanosecond resolution can be done

as follows:

HighPerTimer timer1;

uint32_t SleepTimeNs(14500);

// sleep in nanoseconds

timer1.NSecSleep(SleepTimeNs);

Alternatively, the time value of a HighPerTimer

object can be set to a specific time value at which the

process shall wake up. On the call of SleepToThis(),

the process will then be suspended till the system time

has reached the value of that object :

//declare timer object equaled to 10 s, 500 ns

HighPerTimer timer2 (10, 500);

timer2.SleepToThis();

Table 4 shows the precision of sleeps and

busy-waits using different methods. Miss values are

here the respective differences between the targeted

wakeup time and real times of wakeups measured in

our tests. However, the miss values of sleep times

heavily depend on the fact, whether target sleep

interval was shorter or longer, than time between

Table 4 The comparison of miss values of different methods of sleeping, performed with TSC on the Intel Core –i7 processor.

Sleep time >= 1/HZ Sleep time < 1/HZ

Mean miss, µs Mean miss, µs

System sleep 61.985 50.879

Busy-waiting loop 0.160 0.070

HighPerTimer sleep 0.258 0.095


1084

Fig. 4 Dependency of miss on the target time from sleep time, performed with TSC on the Intel Core –i7 processor, HZ = 1,000.

two timer interrupts. So, Table 4 consists of two

parts—one where sleep time is longer than 1/HZ, and

one where it is less than 1/HZ. Thus, the left column

shows results for waits lasting longer than a period of

two kernel timer interrupts. The right column shows

the results for the scenario, in which the sleep call

lasts less than the interval between two kernel timer

interrupts. These measurements have been performed

on the Intel Core–i7 processor. Other than in

measurements from Section V, in this case it makes

sense to show results on a more stable and powerful

system. Moreover, it was expected that the accuracy

of sleeps would be higher on the newer Linux kernel

versions where time handling has been changed

significantly. However, as the measurements below

show, these kernel changes are still not sufficient.

In this test scenario, we have issued the respective

sleep method within a loop of 100000 sleeps with

different sleep times between 0.25 s and 1 µs, and

then the mean value of the sleep duration miss has

been calculated.

The above experiment took about 830 min, so the

upper limit of the range for sleep time value was

reduced to 0.25 s. The chart in Fig. 4 demonstrates

more detailed results of this experiment and shows the

dependency of miss against the target sleep time in

dependence from sleep duration. To track this

dependency more deeply, here the range of sleep time

value was increased and is taken between 10 s and 1

µs

In the next step, we measured the CPU

consumption of the respective sleep routine. In the

busy-waiting loop, the total CPU consumption during

the test achieves, as expected, almost 100%. For the

sleep function of the standard C library, it tends to

zero. In the case of sleeping using the HighPerTimer

library, the overall CPU consumption during the test

was 1.89%, which can be considered as a good

tradeoff between precision of waking up time and

CPU consumption during the sleep.

7. Conclusion and Future Work

In accordance with the requirements of advanced

high-speed data networks, we showed an approach for

the unified high performance timer library that

successfully solves two significant problems. Firstly,


1085

HighPerTimer allows identification of the most

efficient and reliable way for time acquisition on a

system and for avoiding system calls invocation on

time acquisition. Secondly, it solves the problem of

precise sleeping aspects and provides new advanced

sleeping and resuming methods.

The HighPerTimer library has the potential to

become widely used in estimation network packet

dynamics, particularly when conducting

high-accuracy and high-precision measurements of

network performance. At this stage, the integration of

the suggested solution into the appropriate tool for

distributed network performance measurement [22] is

in progress. Moreover, to the next steps, the better

support of the ARM processor will be addressed.

Since the ARM processor possesses neither HPET nor

TSC, the only way to support ARM at this stage is to

select OS Timer. Presumably, an invocation of the

initial ARM system timer can afford to save several

additional microseconds and improve the timer

accuracy.

References

[1] R. Takano, T. Kudoh, Y. Kodama, F. Okazaki, High-resolution timer-based packet pacing mechanism on the linux operating system, IEICE Transactions on Communication E94-B (2011) 2199-2207.

[2] J. Micheel, S. Donnelly, I. Graham, Precision time stamping of network packets, in: Proc. of the 1st ACM SIGCOMM Workshop on Internet Measurement, San Francisco, CA, USA, Nov.1-2, 2001, pp. 273-277.

[3] H. Hu, Untersuchung und prototypische implementierung von Methoden zur hochperformanten Zeitmessung unter Linux, Bachelor Thesis, Anhalt University of Applied Sciences, Koethen, Germany, Nov., 2011.

[4] Performance monitoring with the RDTSC instruction [Online], http://www.ccsl.carleton.ca/~jamuir/rdtscpm1.pdf (accessed: Jan., 2013).

[5] E. Corell, P. Saxholm, D. Veitch, A user friendly TSC clock, in: Proc. PAM, Adelaide, Australia, Mar. 30-31, 2006, pp. 141-150.

[6] S. Siddha, V. Pallipadi, D. Ven, Getting maximum mileage out of tickles, in: Proc. of the 2007 Linux Symposium, Ottawa, Jun. 27-30 2007, pp. 201-208.

[7] Intel IA-PC HPET (High Precision Event Timers)

Specification [Online], http://www.intel.com/content/dam/ www/public/us/en/documents/technical-specifications/software-developers-hpet-spec-1-0a.pdf (accessed: Jan., 2013).

[8] D. Kang, W. Lee, C. Park, Kernel Thread Scheduling in Real-Time Linux for Wearable Computers [Online], DOI: 10.4218/etrij.07.0506.0019 ETRI Journal 29 (2007) 270-280.

[9] P. Orosz, T.Skopko, Performance evaluation of a high precision software-based timestamping solution, International Journal on Advances in Software 4 (2011) 181-188.

[10] T. Gleixner, D. Niehaus, Hrtimers and beyond: transforming the Linux time subsystems, The Linux Symposium, Ottawa, Canada, Jul. 19-22, 2006, pp. 333-346.

[11] D. Kachan, E. Siemens, H. Hu, Tools for the high-accuracy time measurement in computer systems, in: 6th Industrial Scientific Conference, “Information Society Technologies”, Moscow, Russia, 2012, pp.22-25.

[12] Intel Trace Collector Reference Guide [Online], p. 5.2.5. http://software.intel.com/sites/products/documentation/hpc/ics/itac/81/ITC_Reference_Guide/ITC_Reference_Guide.pdf (accessed: Jan., 2013).

[13] D. Grove, P. Coddington, Precise MPI Performance Measurement Using MPIBench, in: Proc. of HPC Asia, Gold Coast, Australia, Sep. 24-28, 2001, pp. 1-14.

[14] The Dag project [Online], http://www.endace.com, (accessed: Jan., 2013).

[15] Pasztor, D. Veitch, PC based precision timing without GPS, in: The 2002 ACM SIGMETRICS International Conference on Measurement and Modeling of Systems, Marina Del Rey California, USA, 30 (2002) 1-10, DOI: 10.1145/511334.511336.

[16] Intel 64 and IA-32 Architectures, Software Developer‘s Manual [Online], vol. 3B 17-36, http://download.intel.com/ products/processor/manual/253669.pdf, (accessed: Jan., 2013).

[17] J. Dike, A user-mode port of the Linux kernel, USENIX Association Berkeley, California, USA, 2000, pp. 63-72,.

[18] K. Jain, R. Sekar, User-Level Infrastructure for System Call Interposition: A Platform for Intrusion Detection and Confinement, in: Proc. of the ISOC Symposium on Network and Distributed System Security, Feb. 3-4, 2000 pp. 19-34.

[19] J. Corbet, On vsyscalls and the vDSO, Kernel development news, Linux news site LWN [Online], http://lwn.net/Articles/446125/ (accessed: Jan., 2013).

[20] Online C++11 standard library reference, URL: cppreference.com (accessed: Jan., 2013).


1086

[21] GNU Operating System Manual, “Elapsed Time” [Online], http://www.gnu.org/software/libc/manual/html_node/Elapsed-Time.html (accessed: Jan., 2013).

[22] E.Siemens, S.Piger, C. Grimm, M. Fromme, LTest–A

Tool for Distributed Network Performance Measurement, in: Consumer Communications and Networking Conference, Las Vegas, NV, USA, Jan. 11-14, 2004, pp. 234-244, DOI: 10.1109/CCNC.2004. 1286865.

Journal of Com

New H

Adjust

Mansour Zu

Department of

Arabia

Received: May

Abstract: Theunderground lattracting consgiven cellular essential perfolimitations of scheme that camaximizing thadjusting QoS Key words: F

1. Introdu

Femtocell

capacity an

transmission

macrocell lo

and small c

femtocell is

maximum c

basic structu

The most

future wirel

(quality of

promising te

huge increa

various wir

network [4,

Femtocell

CorresponM.Sc., [email protected].

mmunication an

Hybrid A

ting Qo

uair, Abdul M

f Computer Eng

y 09, 2013 / Ac

e increasing numlocations; one osiderable attent

and to enhancormance promowireless comman manage the he system utilizS of the connect

Femtocell, hybri

uction

l provides in

nd provides

n of wireless

oad balancing

apacity base

between 13-2

overage is ab

ure of femtoce

t important p

less commun

service) an

echnologies

ase in wirele

reless applic

5].

l base station

nding author: arch field: .sa.

nd Computer 10

Access

oS

Malik Bacheer

gineering, Colle

ccepted: June 08

mber of mobileof suggestion sion in mobile cce user’s capac

otion issue. Devmunication are c

femocell resouzations, we provted users.

id access metho

ndoor, small

s high qua

communicat

[1]. Femtoce

station. The

20 dBm, in th

bout 15 to 50

ell access [2,

arameters for

nications are

nd cost. On

is to meet th

ess capacity

cations of th

n is installed i Abdul Malikcomputer n

0 (2013) 1087-1

s Meth

r Rahhal and M

ege of Compute

8, 2013 / Publis

e users raises isssolution to accocommunicationscity. Call admiveloping methocapacity and ranurces between svide a mechanis

od, QoS.

office cover

ality and h

tions services

ell is a low-po

e power rang

he same floor

0 m. Fig. 1 is

3].

r the current

e data rate, Q

ne of the m

he demand of

requirement

he femtocell

in homes as m

k Bacheer Ranetworks. E-m

1091

hod for

Mohamad M

er and Informat

shed: August 31

sues about coveommodate this s, it is a small bission control ads for femtocelnge. The main subscriber (femsm that leads to

rage,

high

s for

ower

ge of

r, the

s the

and

QoS

most

f the

t by

lular

mall

ahhal, mail:

priv

part

fem

own

call

F

con

hyb

In

serv

e-H

Fig.

Femto

Mahmoud Alra

tion Sciences, K

1, 2013.

erage extensiongrowing of mobase station thaand resource alll utilization is contribution of

mtocell user) ando serve more us

vate station

ticular subsc

mtocell may al

ner to use it,

led a hybrid a

From access

nfigured to be

brid access [6

n the past

vices increas

Health, online

. 1 The basic

ocell th

ahhal

King Saud Univ

n in some areas obile user is femat install to imprllocation infemvery comparatif our paper is dd non-subscribesers by admittin

which is ow

cribers can

llow users wh

for cell cove

access mode.

methods view

e either open

].

few years, t

sed (e.g., d

gaming, etc.

structure of fe

hrough

versity, Riyadh 1

such as rural zomtocell. Femtorove the indoor

mtocell’s hybridive nowadays. T

developing resouer (macrocell u

ng more subscri

wned by use

access it. H

ho are not of t

erage optimiz

wpoint femto

n access, clos

the demand

data, voice,

). 66% of voi

emtocell access

h

11543, Saudi

ones, indoor or cell have been r coverage of a d mode are an The two major urce allocation

user in order to ibers basing on

er. Therefore

However, the

this femtocell

zation, this is

ocells can be

sed access or

of wireless

multimedia,

ice and 90%

s.

e

e

l

s

e

r

s

,

New Hybrid Access Method for Femtocell through Adjusting QoS

1088

of data traffic occurs indoors [7], and due to high

penetration losses, macrocells are not very efficient

when delivering indoor coverage, providing such

coverage has become a challenge for operators this is

showing why the use of FAPs (femtocell access points)

seems a promising approach for coping with this

coverage problem.

Main goal of proposed scheme in Ref. [8] is to

transfer macrocell calls to femtocellular networks as

many as possible. They divide the proposed model into

three parts. The first one for the new originating calls,

the second one for the calls that are originally

connected with the macrocellular BS, and the last one

for the calls that are originally connected with the FAPs.

They offered the bandwidth degradation policy of the

QoS adaptive multimedia traffic to accommodate more

number of macrocell-to-macrocell and

femtocell-to-macrocell handover calls [8, 9].

One model has been proposed in Ref. [10] is based

on guard channel which is a mechanism to reserve

some portion of resources in advance for important

event. The propped model assumed that there are two

kinds of event, call for CSG (closed subscriber group)

members and call for non-CSG members. The call

will be new call in the cell or handover call from other

cells.

The remainder of this paper is organized as follows:

Section 2 explains access method; Section 3 talks

about our proposed model; Section 4 presents the

simulation results; Section 5 gives conclusions.

2. Access Method

Femtocells can be configured to be either open

access or closed access [6, 7]. Open access allows an

arbitrary nearby cellular user to use the femtocell,

whereas closed access restricts the use of the femtocell

to users explicitly approved by the owner. Seemingly,

the network operator would prefer an open access

deployment since this provides an inexpensive way to

expand their network capabilities, whereas the

femtocell owner would prefer closed access, in order to

keep the femtocell’s capacity and backhaul to himself.

These access methods (open, closed) suffer from

advantages and drawbacks. In order to overcome those

drawbacks, a hybrid access method reach can be used

to compromise between the impact on the performance

of subscribers and the level of access granted to

nonsubscribers. Therefore, the sharing of femtocell

resources between subscribers and nonsubscribers

needs to be tuned. Otherwise, subscribers might feel

that they are paying for a service that is to be exploited

by others. On the other hand, the impact on subscribers

must be minimized in terms of performance or via

economic advantages (e.g., reduced costs) [7].

The advantage and disadvantages of each access

method can be summarized as follows:

Open access: although open access method increases

the throughput and improve QoS but it also increases

number of handoff due to movement of outdoor users,

security issue, reduce the performance for femtocell

owner due to the sharing of femtocell resources with

nonsubscriber [6, 7, 11].

Closed access: in this case it is not allowed for

nonsubscriber to access to the femtocell even if

femtocell signal is stronger than that of macrocell,

which would cause strong cross tier interference but

also it increases the security, in addition to increasing

co-tier interference between neighboring femtocell in

dense deployment [6, 7].

Hybrid access: It needs to be analyzed carefully,

otherwise subscribers might feel that they are paying

for a service to be exploited by others.

3. Proposed Model

In this system there is one femtocell with capacity up

to M = 10 channels, time is slotted, arrivals are Poisson,

average service time is equal to slot time, probability to

transmit is P, and there is finite number of user equal to

20 users (10 subscribers and 10 nonsubscribers) with

no queuing. We can consider this work as extension of

our previous work [12]. The proposed model has these

enhancements: (1) subscriber users can use any

available ch

specific par

channels; (2

with all cha

users is redu

shown in Fig

4. Simulat

We used m

two differen

4.1 Hybrid A

In this ca

Fig. 2 Propo

Fig. 3 Hybr

N

annel, where

rt of capacity

2) in the case

annels reserve

uced by 10% t

g. 2 below.

tion Results

matlab to sim

nt cases.

Access Metho

ase, new sub

osed model dia

rid access with

New Hybrid A

as nonsubscr

y in this cas

when first s

ed, the QoS f

to make room

s

mulate our mo

od with Full Q

bscriber call

agram.

full QoS.

Access Metho

riber can use o

se equal to 7

subscriber arr

for all conne

m for new user

odel, we simu

QoS

will be reje

od for Femtoc

only

70%

rives

ected

r. As

ulate

ected

if

QoS

W

of b

exp

4.2

T

sub

thro

by

T

the

cell through A

all channel

S.

Whenever the

blocking will

pected and sho

Hybrid Acce

The goal is to

scriber with

ough reduci

10%.

There is very

system, as sh

Adjusting Qo

ls are busy

e arrival rate is

increase, as s

ould be enhan

ess Method wi

o reduce bloc

all channel

ng QoS fo

good improv

hown in Fig. 4

oS

y, no com

s increased th

shown in Fig.

nced.

ith 90% QoS

ck rate. In thi

s busy will

or all conn

vement in per

4.

1089

mpromise in

he probability

3, and this is

is case a new

be admitted

nected users

rformance of

9

n

y

s

w

d

s

f


1090

Fig. 4 Hybrid access with 0.9 QoS.

5. Conclusions

Femtocells have been attracting considerable

attention in mobile communications [13] which cover a

cell area of several tens of meters. Femtocells use

common cellular air access technologies [14].

Femtocells have the potential to provide high quality

network access to indoor users at low cost, while

reducing the load on the macrocells. Call admission

control in hybrid mode femtocell is an essential

performance promotion issue. In this paper, the

authors developed a mechanism for improving the

utilization of femtocell capacity and reducing rejection

rate of call through slight customization QoS.

Acknowledgment

This work was supported by the research center of

college of computer and information system, King

Saud University, Riyadh, Kingdom of Saudi Arabia.

The authors are grateful for this support.

References

[1] D. Lopez-Perez, A. Valcarce, Guillaume De La Roche, E. Liu, J. Zhang, Access methods to WiMAX femtocells: A downlink system-level case study, in: 11th IEEE Singapore International Conference on Communication System, Guangzhou, Nov. 19-21, 2008.

[2] D.N. Knisely, F. Favichia, Standardization of femtocells in 3GPP2, IEEE Com. Mag. 47 (2009) 76-82.

[3] S.J. Wu, A new handover strategy between femtocell and

macrocell for lte-based network, in: 4th International Conference on Ubi-Media Computing, Sao Paulo, Jul. 3-4, 2011.

[4] R.Y. Kim, J.S. Kwak, K. Etemad, WiMAX femtocell: requirements, challenges, and solutions, IEEE Communication Magazine 47 (2009) 84-91.

[5] M.Z. Chowdhury, Y.M. Jang, Z.J. Haas, Cost-effective frequency planning for capacity enhancement of femtocellular networks, Wireless Personal Communications 60 (2011) 83-104.

[6] J. Zhang, Guillaume de la Roche, Femtocell: Technologies and Deployment, John Wiley & Sons Ltd., UK, 2010.

[7] T. Chiba, H. Yokota, Efficient route optimization methods for femtocell-based all IP networks, in: IEEE International Conference on Wireless and Mobile Computing Networking and Communications WIMOB, Marrakech, Oct. 12-14, 2009.

[8] M.Z. Chowdhury, Y.M. Jang, Call admission control and traffic modeling for integrated macrocell/femtocell networks, in: 2012 4th International Conference on Ubiquitous and Future Networks (ICUFN), Phuket, Jul. 4-6, 2012.

[9] F.A. Cruz-Perez, L. Ortigoza-Guerrero, Flexible resource allocation strategies for class-based QoS provisioning in mobile networks, IEEE Transaction on Vehicular Technology 53 (2004) 805-819.

[10] S.Q. Lee, R.B. Han, N.H. Park, Call admission control for hybrid access mode femtocell system, in: 2011 IEEE 7th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), Oct. 10-12, 2011.

[11] G. de la Roche, A. Valcarce, D. Lopez-Perez, J. Zhang, Access control mechanisms for femtocells, Communications Magazine IEEE 48 (2010) 33-39.


1091

[12] M. Zuair, Development of an access mechanism for femtocell networks, JATIT Journal 51 (2013) 434-441.

[13] S. FarazHasan, N.H. Siddique, S. Chakraborty, Femtocell versus WiFi-A survey and comparison of architecture and performance, in: 1st Internet Conference on Wireless Communication, Vehicular Technology, Information

Theory and Aerospace & Electronics Systems Technology Wireless VITAE, Aalborg, May 17-20, 2009.

[14] D. Lopez-Perez, A. Valcarce, G. de la Roche, J. Zhang, OFDMA femtocells: A Roadmap on Interference Avoidance, Communications Magazine IEEE 47 (2009) 41-48.


Design of an Information Connection Model Using

Rule-Based Connection Platform

Heeseok Choi and Jaesoo Kim

National Science & Technology Information Service Center, Korea Institute of Science and Technology Information, Daejeon

306-806, Korea

Received: August 06, 2013 / Accepted: August 18, 2013 / Published: August 31, 2013.

Abstract: NTIS (National Science & Technology Information Service) collects national R&D information through the connection system in real time with specialized institutions under government ministries for R&D information service. However, because the information connection between the research management systems in each ministry (institution) and the NTIS is different, it is not easy to operate the connection system, and immediate data collection is thus not ensured. This study aims to propose an information connection model to be applied on the NTIS-like systems. To do this, we examine methods or styles of information connection and compare strength and weakness of connection methods. In this paper we also understand issues or characteristics of the methods through analyzing current information connection methods applied on the NTIS. Therefore, we design a rule-based information connection platform to minimize the information connection issues. Based on the platform, we also propose an information connection model.

Key words: Information connection, information sharing, rule-based connection.

1. Introduction

NTIS (National Science & Technology Information

Service) was developed for improving efficient

research and development throughout the cycle from

planning R&D to using the outcome thereof [1]. For

this purpose, each representative institution under the

government ministries and agencies comprehensively

manages its R&D information for connection with the

NTIS for the purpose of real-time collection of

national R&D information (periodical update when

collecting and changing the information at the time of

project agreement). The NTIS has currently built a

real-time connection system with representative

institutions under government ministries and agencies

[2].

However, since the connection criteria of each

Corresponding author: Heeseok Choi, Ph.D., KISTI senior

Researcher, research fields: information integration, information management, big data and data mining. E-mail: [email protected].

institution with external systems are different, the

NTIS is connected with the research management

system of each institution in various manners, to

collect R&D information. Accordingly, immediate

data collection is not ensured, and various manners of

connection contribute to inefficient

operation/maintenance.

This study aims to propose an information

connection model to be applied on the NTIS-like

systems. To do this, we examine methods or styles of

information connection and compare strength and

weakness of connection methods. In this paper we

also understand issues or characteristics of the

methods through analyzing current information

connection methods applied on the NTIS. Therefore,

we design a rule-based information connection

platform to minimize the information connection

issues. Based on the platform, we also propose an

information connection model to be applied on the

NTIS-like systems.

Design of an Information Connection Model Using Rule-Based Connection Platform

1093

The paper is organized as follows: Section 2

examines methods or styles of information connection

and compares strength and weakness of connection

methods. Section 3 contains issues or characteristics

of the methods through analyzing current information

connection methods applied on the NTIS. Section 4

presents our approach to design of an information

connection model using rule-based connection

platform. Section 5 gives conclusions.

2. Related Works

2.1 Technologies of Information Connection

For information connection between systems, P2P,

EAI, ESB or combinations thereof are currently

applied [3]. Features and characteristics of each

method are described below.

P2P: 1:1 connection between individual systems,

can not extended or reused. However, because

connection between individual systems is simple, a

connection system is easily built in conformity with

features of each system.

EAI: Individual applications are connected to the

central hub by means of an adapter, and are connected

to other applications through the central hub. This

significantly improves typical complex connections.

However, this method uses vender-dependent

technology and adapter costs should be paid for each

connected application.

ESB: This method was developed to avoid

weakness of non-standard EAI (Hub & Spoke) and

SPOF (single point of failure). However, because the

current ESB solution market is controlled by the EAI

solution providers, the tendency is that the previous

EAI solutions are supplemented and developed. In

general, this method is used for service connection.

Table 1 describes the strength and weakness of each

technology in terms of complexity, extensibility,

flexibility, and integration cost, etc..

2.2 Styles of Data Provision (Collection)

When systems of other organizations are connected,

independency of system operation and system/data

security of the organizations can be also an important

issue. Therefore, the subject of data provision

(collection) can be considered as an important factor

in building and operating a connection system.

Connection depending on the subject of data provision

(collection) is divided into the push method and the

polling method [4].

Push method: an information source (information

provider) that owns and manages original information

pushes data into information targets (information

consumers) in a given cycle according to the

information provision policy. Therefore, information

connection in this method is by a subject of

information provision who leads information

connection. This method is in favor of operation and

security of data and institution systems for

information providers, but does not ensure immediate

information collection for information providers.

Polling method: this is to access information

providers (systems for information sources) when an

information collector requires the information to bring

required information. Therefore, this is a method of

Table 1 Comparison of connection technologies.

Strength Weakness

P2P -Easy application to simple interworking between systems in a non-complex environment.

-As the number of systems to be connected increases, the cost for maintenance may sharply increase. -Low extension capability and flexibility.

EAI -Easy extension in introducing new applications. -Increased productivity and convenience for development and maintenance.

-High cost of establishment and maintenance. -Central hub failure affects the entire system. (Single Point of Failure)

ESB

-Reduced integration cost because standard technology is used and service units can be reused. -Loose connection in a bus type contributes to high extension capability and flexibility.

-High cost of initial establishment.


1094

connection in which the subject of information

collection leads information connection. This method

ensures immediate information collection, but is not

preferred by information providers in terms of

operation independency and security of internal

systems because information sources are directly

accessed from external systems.

Table 2 describes the strength and weakness of

provision styles in terms of security, maintenance,

performance, and dependency, etc..

3. Analysis of the NTIS Connection System

We analyzed the case of NTIS. The NTIS

established an information connection system with

research management systems of representative

institutions under government ministries and agencies

in order to connect and collect national R&D

information. In this case, the push method is applied

as a connection method to enable each representative

institution to have the right of providing data

appropriate for the connection policy and system

environment thereof with external systems, and to

have the ownership of the data owned by each

representative institution. That is, data are provided by

means of DB connection according to a given cycle

for data items defined in the national R&D

information standard. However, the principle of

providing data on a daily basis if data are created is

not ensured by adding an approval procedure for data

provision to the automatic connection process in

actually operating the connection system. Also,

although data are provided in the push method, the

DB link method and the unidirectional EAI method

are applied to each institution depending on each

institution environment. More details are shown in

Table 3 to show implemented connection systems in

various types. The reason of application is because the

method of data provision is determined and

implemented in various types according to the

preference of each institution (for tasks or system

environment) when the push method is applied. This

results in no assurance of immediate data provision,

and also makes monitoring and operation of the

connection system difficult.

Table 2 Comparison of data provision styles. Strength Weakness

Push

-High security in system connection between systems of institutions. -Independent connection with external systems in system operation. -Information providers lead connection.

-Because information providers lead information connection, integrated management of systems to be connected is not easy. -Information connection varies with institution by institution, maintenance is not easy.

Polling

-Because information collectors lead information connection, integrated management is easily implemented according to information collection policies. -Immediate data collection is ensured.

-Security is vulnerable in system connection. -Because performance may be affected by connection with external systems in system operation, information providers do not like this method.

Table 3 Information connection styles in the NTIS.

Method Type Description

DB link

View Direct connection to a DB in a connection server of an institution. The procedure Inquiry/Send is used.

Snapshot The copy of institution DB table refreshes the connection server DB in a given cycle.

DB trigger Trigger is established to reflect relevant changes on the connection server DB in updating the institution DB.

DB trigger + JDBC Trigger is established to reflect relevant changes on the connection server DB in updating the institution DB (based on Java, etc.).

DB script Uses script to send data from an institution DB to the connection server in a given cycle.

Procedures Transmits data from an institution to the connection server DB table by means of the procedure.

EAI Program Transmits data by means of the EAI program of each ministry (institution) (unidirectional).


1095

4. Design of an Information Connection Model Using Rule-Based Connection Platform

The current NTIS information connection system

has been established in the P2P style through the DB

link or the unidirectional EAI method as described in

the Section 3. However, it is necessary to improve the

connection system in a standardized method for

immediate provision, integrity, efficient connection

system monitoring and operation of data connection.

It is also important to design the method based on

easily managable rules. In this study, parts which can

be functionally standardized in information

connection between the NTIS and specialized

institutions under ministries and agencies are

identified to design them as a major functional module

of the connection platform. We design a rule-based

information connection platform to be applied on the

NTIS-like information connection systems. Based on

the platform, we also propose an information

connection model.

4.1 Rule-Based Connection Platform

Rule-based connection platform should be basically

designed on standardization for information

connection among heterogeneous systems. The

sections from which data are first acquired in

connection with each institution are divided into

variation areas for the purpose of information

connection based on standardization, and the next

phases are identified as standard functional areas. That

is, as shown in Fig. 1, the R&D information is first

transferred to the connection DB from the institution’s

system DB. This section is defined as a variation area.

Subsequent data processing in the connection DB and

data transmission to the integration DB is defined as a

major function of a common area which can be

standardized.

Therefore, it is necessary to build the connection

platform as a function to overcome variability of each

institution in the variation areas and to process the

functions in the common area. To this end, the

connection platform includes functions of connection

rule processing, mapping, connection error processing,

and creation of update information and monitoring

information. Fig. 2 shows system architecture of a

connection platform which includes the

aforementioned functions.

Fig. 2 shows the connection platform which is

composed of rule processing (preActors), connection

processing (Workers), and parallel task processing

control (Controllers), and which performs connection

of the common areas according to the rule predefined

in the Rule File.

Schema mapping (Mapper): This is carried out

between an institution DB schema and the NTIS

standard collection schema according to the schema

mapping rule defined in the Rule File for the data

transferred from the institution. Code mapping is also

carried out from the code value in the institution DB

to the code value in the NTIS integration DB

according to the code mapping rule defined in the

Rule File.

Comparison of data (Comparer, I/U/D Handler):

This is to compare the data transferred from an

institution with the data transferred in the previous

cycle to decide whether to update the data. On the

basis of comparison result, the system in the

connection institution system displays I (Insert), U

(Update), or D (Delete) to indicate that the relevant

data is new, updated or deleted data.

Connection error processing (Error Checker,

Error Handler): This is to check errors in connection,

for example, key errors, errors in essential connection

Fig. 1 Basic steps of information connection.


1096

Fig. 2 Rule-based connection platform.

items, code conversion errors, data conversion errors,

data format errors, data length errors, etc.. Details of

the checked connection errors are created to be an

error DB. Normal data are then stored as an OrgDB to

be transmitted to the NTIS integration DB.

Connection monitoring information creation

(Monitor): Information is created about whether

connection normally operates, for example, the

number of connection data, details of updated data,

execution of the connection module according to the

schedule, or how much new data have been provided.

Rule Parser: This enhances data mapping through

rule based processing. The Rule Parser interprets the

Rule File which specifies schema mapping, code

mapping and rules that should be observed when data

are provided from an institution to the NTIS.

Scheduler: Information connection is performed

periodically or in real time depending on information

type. Therefore, the function to control the

information connection execution cycle is provided.

Three types of execution scheduling is provided,

including manual execution (immediate execution) by

an operator, periodical execution and execution after

standby for a given period of time in consideration of

the features of NTIS connection.

Data Cacher: Information frequently used, for

example, schema information and code mapping

tables, is internally cached to improve connection

capability. Data are deleted after a given period of

time.

Operation environment (Controllers): This

controls listing tasks to be processed for optimized

resource management and function processing by

means of multi threads, and to carry out the tasks

according to the processing sequence. Controllers also

provide access to storages to store the connection

result in the DB or a file.

The connection platform designed as such can

improve data processing speed through internal data

caching. It can enhance data mapping through

rule-based data processing, and can operate and

maintain the connection system by

producing/changing rules. It sorts data sources from

targets to manage data history, and systematically

checks data errors. It can address difficulty in

connection monitoring due to different connection

methods between institutions, and processes data

update information.

4.2 Information Connection Model Using Rule-Based

Connection Platform

On the basis of the rule-based connection platform

designed in Section 4.1, information connections in

the push method, and in the EAI method or the polling

method by agents can be implemented. Therefore, two

types of information connection models were

designed and the two types of connection models were

compared with respect to the important issues

considered in information connection by the NTIS

with the representative specialized institutions.

(1) P2P & Push method using the connection

platform:

For standardized information connection, the

connection platform was defined, which processes the

information connection rule, performs encryption, and

creates connection error and monitoring information.

The connection standard platform contributes to

addressing limitations by different information

connection methods of each institution. That is,


1097

although each institution provides data in a different

method (entire relevant data or some of changed data),

it is possible to identify details of data change, and to

create consistent error and monitoring information.

Consistent connection also contributes to easy

management of information connection. This

connection method, however, can provide data to a

connection system at times desired by an institution.

Fig. 3 shows this method as described above.

(2) Agent & Polling method using the connection

platform:

This is a method of connection to apply the Polling

method which uses an agent to bring institution data

while information connection is based on

standardization. This method enhances the efficiency

of using connection server resources, and ensures the

initiative of data collection to ensure immediate data

collection. Integrated connection system management

can be also implemented. The system for jointly using

administrative information employs this method. Fig.

4 shows this method as described above.

Finally, Table 4 shows the strength and weakness

of two connection methods in information connection.

Of course, most external connection institutions prefer

connection by the “P2P & Push” methods because

security is a key factor to determine their connection

method. However, in consideration of the strength

described in Table 4, it is necessary to employ a

connection method which implements the

aforementioned strength. Therefore, in this study, the

method of “Agent & Polling” using rule-based

connection platform is suggested for future NTIS

information connection. To this end, it is necessary to

establish schemes for strengthening security, and to

establish access supported by policies and strategies.

5. Conclusions

This study examined methods or styles of

information connection and compared strength and

weakness of connection methods. We also analyzed

the current NTIS information connection system

established with representative specialized institutions

under government ministries and agencies. On the

basis of this, the variation area and the common area

were identified to design a connection platform for the

Fig. 3 Connection in P2P & Push method.

Fig. 4 Connection in Agent & Polling method.

Table 4 Comparison of the connection method.

P2P & Push Agent & Polling

Connection speed Same Same

Storage capacity

Great *Because each institution uses its own data provision method, data pre-processing is thus required.

Not great

Management efficiency Because information providers are the subject of information connection, integrated management is not easy.

Because information collectors are the subject of information connection, integrated management is easy.

Immediate connection Not ensured Ensured

Security Relatively high (in terms of institution systems)

Relatively low (in terms of institution systems)


1098

functions of the common area. We also examined the

methods. In this paper we also understand issues or

characteristics of the methods through analyzing

current information connection methods applied on

the NTIS. Therefore, we designed a rule-based

information connection platform to be applied on the

NTIS-like information connection systems. Based on

the platform, we also proposed an information

connection model.

It is necessary to expand rule-based connection

platform to establish a flexible and extensible

connection system. In addition, it is necessary to

develop a connection guideline for standardizing

information connection.

References

[1] NTIS (National Science and Technology Information Service) Home Page, www.ntis.go.kr.

[2] H. Choi, etc., A study on real-time integration system extension of national R&D information, in: Korea Computer Congress, 2010.

[3] Y. Nah, ESB-based Data Service Connection [Online], 2010, www.dator.co.kr.

[4] H. Choi, etc., Technology Trends on Information Connection, Technical reports, Korea Institute of Science and Technology Information, Korea, 2012.


Communication Methods: Instructors’ Positions at

Istanbul Aydin University Distance Education Institute

Kubilay Kaptan and Onur Yilmaz

Disaster Education, Application and Research Center, Istanbul Aydin University, Kucukcekmece 34290, Turkey


Abstract: In the world of higher education, several issues are converging: (1) advances in computer technology; (2) rapidly growing enrollments; (3) changing student demographics; and (4) continued cost containment requirements. Work based research involves action plan for the improvement of Distance Education Institute that requires a change based on quality mission of Istanbul Aydin University and EUA norms. This research is part of learning process in Doctorate of Professional Studies that has significant contributions on online pedagogy and teaching process for online tutors in enhancing their professional growth. The work based research aims to explore the importance of communication process within the work setting and to investigate roles of tutors in facilitating communication to overcome social barriers in constructing knowledge during online learning-teaching process. The research has qualitative and inductive nature that action research approach is chosen to change professional practices through collaborative activities. Therefore, focus group, trainings, in-depth interviews, self-reports and researcher diary are used as data collection techniques for each action. This proposal is part of ongoing research project that process is aimed to be shared with academic community.

Key words: Communication, online pedagogy, roles, work based project.

1. Introduction

Work based project is one of the achievements of

workers as researchers for their work settings and

their developments. It is part of lifelong learning

process that learning and reflection are incorporated to

propose a change and innovation regarding to the

vision and missions of work settings. In this respect,

researcher as worker play a great role to evaluate

environment of work and facilitate proposed actions

for the development of working practices.

Work based learning is a type of experiential

learning, in the sense that work based learning is

gained mostly through what people do at work. It

focuses on individuals’ work practices and on

experience gained from their work roles, cumulatively

throughout their careers. In essence it requires

Corresponding author: Kubilay Kaptan, Ph.D., lecturer,

research fields: disaster management and online education. E-mail: [email protected].

questioning and reflecting upon what researcher have

learnt about and from their own work based learning.

Work based research project provides researchers to

reflect concrete experiences and develop concepts, be

in active experimentation. It is about becoming a good

practitioner, about choosing the best option for better

working practice. It is about respecting humanity for

supporting others’ through researchers’ experiences

and actions. It provides to focus on being the

researcher as worker seeking to improve aspects of

own and colleagues’ practices [1].

1.1 Focus of Work Based Research Project and Its

Significance

In constructing continuous quality improvements in

higher education, there is intensified need to turn

attention in having innovative strategies in order to

gain competitive advantage. With this respect,

strategic plans regarding collective vision, culture,

Communication Methods: Instructors’ Positions At Istanbul Aydin University Distance Education Institute

1100

climate of the organizations and higher education

practices need to meet distance education practices as

part of innovation in order to gain differentiation and

competitive advantage [2]. In addition, higher

education institutions can gain competitive advantage

by having service differentiation which distance

education could be the most effective service and

strategy for improving institutional performance and

reputation in line with global standards in competitive

environment. The study of Thomas [3] put initial

perspective that distance education practices for the

universities which have dual mode, are the innovative

strategy for quality. According to Kamauin [4],

universities can have dual mode model of Mugridge

[5] which have both distance education and traditional

applications in their organizational structure. In other

words, universities use distance education that has

gained popularity as an alternative mode of delivery

because of its ability to address issues of equity to

people who did not go on with their education for one

reason or other. It enables higher institutions to train

staff, upgrade peoples’ academic and professional

qualifications and impart new skills without

withdrawing them from their duties. Its flexibility has

made it a feasible alternative since it utilizes available

physical, human and material resources. In order to

sustain quality in standards for higher education,

distance education practices need to be developed in

its both pedagogical and organizational aspects. In this

respect, communicative practices become crucial

element within the higher education institutions that

this unit of the institution need to expand its

connection with internal and external environments

and develop new pedagogy by considering changing

roles of the tutors based on change and development

activities regarding strategic action plans. Academic

staff who teach courses in distance education

programmes are invited to be aware on changing roles

and new pedagogy in online context by putting

emphasis on impact of social presence, collaborative

learning on students’ learning and satisfaction.

The study of Srikanthan and Dalrymple [6]

provided in-depth insights on the features of quality

management in education based on the approaches

spelt out in four well-articulated methodologies for

the practice of quality in higher education. The study

reflected the necessity of the quality improvements

and performance assessment in the higher education

by using various activities and quality action plans.

Additionally, Frahm and Brown [7] explored the

understanding on change and development of the

practices by the impact of communication roles and

the collaboration. Parka et al. [8] supported the idea

that communication and the collaboration are the

critical factors to implement change and development

within the organisations for the quality improvements.

Furthermore, this study put forward to the nature of

interaction among teachers and how that collegial

interaction influenced teachers’ professional

development.

Pallof and Pratt [9] defined quality as practicing

learner focused online education. In this respect,

online tutors need to be facilitator encourage online

learner to take their own learning process toward

acquisition of the knowledge without loosing the

facilitation of collaborative learning and social

interaction. Hubbard and Power [10] provided a

ground that online education practices could be

changed and developed through action research

approach based on collaboration and facilitation in the

process. In this respect, Saito et al. [11] underlined the

action research approach as tool to do reform in

educational practices. Gilbert [12] suggested criteria

of being effective online tutors who are designing

their courses specific for distance learning rather than

in-class courses, planning activities carefully by

letting students to know agenda in advance, being

comfortable with technology and seeing it as a tool,

guide students about technology and the structure of

the system. Online tutors need to consider interaction

which is the critical element for the lifelong learning.

Berge [13] listed the roles and functions of the


1101

online tutors and simply stated these roles in various

types as direct human-human communication;

transaction router; storage and retrieval functions.

Berge [13] pointed out that most important role of the

tutors is the responsibility of keeping discourse,

contributing special knowledge and insights, weaving

together various discussion threads and course

components; maintaining group harmony. In addition,

Berge [13] categorized four roles of successful online

tutoring which are pedagogical (intellectual, task);

social (social presence to overcome social barrier);

managerial (organizational, administrative); technical

(technology transparent). Coppola et al. [14] provided

insights on pedagogical roles and changing roles of

the tutors within online learning and teaching process.

Lim and Cheah [15] provided positive insights to the

Berge [13] study on the roles of the tutors and

explored these roles in online discussion within a

Singapore case. Lim and Cheah [15] categorized roles

of the tutors as managerial, facilitating and

pedagogical roles. The study examined the

pre-discussion, during the discussion and the

post-discussion evaluations on the role of the tutors by

focus groups and discussion record analysis.

Significantly, Maor [16] provided a framework for

the participatory action research process in relation to

research focus where focus on dialogue, instructor

c-learning and the joint construction of knowledge.

The simple metaphor of the “four hats” of pedagogical,

social, managerial and technical actions is used as a

framework to discuss the roles of the tutors and the

construction of knowledge based on the convergence

of social presence and interaction.

In respect to worthwhile reality on the status of

distance education institutes, proposing

change-oriented actions for the development of

working practice is crucial [10]. Research focus

covers the changing roles of the tutors, building and

maintaining the awareness on the necessity of new

pedagogy in online context differing from traditional

context, creating awareness on the importance of

social presence and communication to help student

construct knowledge within the institute’s online

courses and applications.

1.2 Aim of the Research

The proposed project aims to investigate the impact

of communication practices within organizational

change and development. In respect to this broader

aim of the project, the research is taken place at

distance education institute at Istanbul Aydin

University which is the innovative and strategic unit

of higher education to reach out quality and global

standards. In relation to that worthwhile reality,

current roles of the online tutors and changes on the

roles after the training based on participatory action

research is examined.

Additionally, the impact of social presence and

facilitation role of the tutor in the construction of

knowledge is explored within the study. Thereby, the

selected case provides to develop working practice

and academic agenda could gain insights from

proposed action plan and process in order to improve

their performances.

The main aim of our project is to create an action

plan for the development of Distance Education

Institute based on European University Association

Standards by focusing on the roles of tutors in

facilitating communication to overcome social barriers

in constructing knowledge. The research is significant

with its action plan by aiming to reach out the

following objectives:

(1) To provide in gaining awareness on the

relevance of communication, organizational climate in

Distance Education Institute and online learning and

teaching in order to focus on social interaction;

(2) To provide trainings on the roles of tutors and

create a consciousness on their roles in online learning

and teaching process;

(3) To enhance online socialization of students by

overcoming social barriers;

(4) To create an organizational culture to Distance


1102

Education Institute by focusing on communication

practices between tutors and among students.

2. Method

2.1 Research Design and Approach

Research relies on inductive process that

experiences, meanings of participant are constructed

from social context within a qualitative research

nature [17]. Action research approach was chosen for

this work based project as it allows the researcher to

use as a method in setting where a problem involving

people, tasks and procedures cries out for solution, or

where some change of feature results in more

desirable outcome [18]. Action research can be used

as an evaluative tool, which can assist in

self-evaluation for an individual or an institution. It

was thought that action research approach would

make an environment of improving the rationality and

justice of professional practices within self-reflective

self-critical context that relies on improving practice.

2.2 Data Collection Techniques and Analysis

Qualitative research covers participatory action

research that researcher attempts to use series of

action to change and develop institute’s performance

on new pedagogy based on the collaboration of the

members. During the participatory action research,

focus groups, in-depth interviews, self-report and

researcher diary are used as data collection techniques.

Firstly, focus group is used as a milestone data

collection technique to create awareness on

communication for change and development and gain

the initial perceptions of the members. In addition,

there is intensified need to examine that what the

extent the online tutors have awareness on their roles

in online learning-teaching process differing from

traditional context, online tutors perform facilitation

roles within online learning and teaching process.

Therefore, in depth interview is used to gain in-depth

insights from understandings and experiences of the

online tutors. Training process is done for the online

tutors to make them trained and developed awareness

on contemporary tutoring roles in distance education

practices. Self-reports of the online tutors and students

help to understand the online learning and teaching

process in relation to realize changes on roles of tutors.

Keeping diary about the actions within the research

process verifies the accuracy of findings. In managing

data and increasing the credibility of findings,

collected data are triangulated and content analysis is

used.

2.3 Ethics

Qualities that make a successful qualitative

researcher are revealed through sensitivity to the

ethical issues. The researcher’s role within research

process became essential that researcher enters into

the lives of participants and share participants’

experiences. Therefore, stressing researcher’s role by

technical and interpersonal considerations enhances

the degree of trust, access in the research. Having time

to focus issues, considering resources are not enough

to be qualitative researcher that qualitative researcher

needs to be active, patient, thoughtful listener, have

emphatic understanding and respect (Hubbard, Power,

1993).

In this respect, ethics in work based project is

crucial that there is an intensified need to concentrate

on conditions and guarantees proffered for school

based research project. In this project, the principles

which are remaining anonymous, treating with the

strictest confidentiality, verifying statements when the

research draft form, submitting final copy of final

report, benefiting report to school were the initial

considerations before making research into practice

[18]. Feedback was guaranteed by researchers in order

to increase confidentiality and building trust between

researchers and participants.

3. Results

At the end of the research project, the following

outcomes are expected to have:


1103

(1) Different tutors from different departments and

backgrounds will gain pedagogical knowledge and

reflection about online education;

(2) Having consciousness on the roles of online

tutors and students through training for their personal

and professional development;

(3) Collaboration and negotiation can be created

among online tutors in order to develop their

collegiality and critical friendship for organizational

knowledge and development;

(4) Strong communication can be created within

Distance Education Institute to create an

organizational climate and culture;

(5) Enhancing quality mission of the Istanbul Aydin

University regarding the Distance Education Institute

practices based European University Association;

(6) Enhancing the reputation of Distance Education

Institute within and outside of the university;

(7) Providing a handbook for online program and

courses of Distance Education Institute at Istanbul

Aydin University;

(8) Being beneficial to our institution and other

departments within the university.

4. Discussion

4.1 Self-evaluation on Research Process

Work based study is the reflective process which

requires practitioners engage with those we work with

and the way we see the world. In other words, it is the

engagement with problem where researcher act for the

solution, attempt to create difference, critically

understand the context and learn within the research

process.

Learning and the reflection are the critical factors

which bring practitioner to the success. In this respect,

reflecting on actions and learning are the significant

part of the research process. In order to implement

research, researcher needs to be secure on subject

knowledge, expertise, and choice of approach and data

collection techniques. Therefore, engaging various

events, negotiation with others, working hard on

writing up crucial chapters of project, preparing

research package and process guideline were the

evidences to justify that researcher has struggled with

challenges and gained confident to implement

process.

Having a worthwhile research topic and its

significance to the work context is the influential

factor to be succeeded. Investigation on literature

underlined the gap in relation to research focus. In line

with the literature analysis, EUA report on distance

education practices justified the necessity of work

based project in relation to new roles of the tutors in

online education for sustainable change and

development within work setting. In this respect,

having consciousness and confidence on worthwhile

topic provided destination to implement practical

project.

Educational background and positive relationship

within work environment draw attention to be

confident for implementing research process by

minimizing challenges of access and the specialization.

Because, access and being highly expertise on the

subject field and process may create challenge to

implement research process. Knowing more about

context, subject field helped overcome these

challenges.

In addition, creating a voluntarism to be part of the

research and increasing the sensitivity on ethics are

also milestones of successful process. Therefore,

preparing research package which consisted of aim,

importance of research, role of the researcher, process

with approach and data collection techniques, consent

forms provided significant evidence to justify that

researcher has rationale to create awareness about

research in order to increase participation for change

and innovation based on ethical understanding.

Also, preparing research process guideline which

gives detailed information on trainings, data collection

techniques is the justification of the research questions

and process which proposed to implement. Reviewing

guideline by experts and piloting increased the


1104

credibility of the process and created confidence for

the researcher.

As the reflection is central essence of the work

based learning, having high level of responsibility on

pre-planned actions provided confidence for the

researcher and reflection, negotiation on these

activities increased the learning.

5. Conclusions

It is work based project which requires

collaboration of the colleagues to propose change for

better working practice. Reflections and learning

experiences of researcher from research process are

aimed to share with academic community.

High risks and questionable rewards are the reality

for most complex organizations experiencing rapid

change. Work, even in higher education, is shifting

toward greater interdependence among individuals to

create collective and synergistic products and services

using advanced technology. As the boundaries

between traditional positions blur, role clarification

becomes increasingly important. In this learning

environment, the role of the ODL instructor requires

the merging of multiple roles. The convergence of

advances in computer technology, rapidly growing

enrollment needs, and cost cutting measures for higher

education suggest that innovative solutions are

required. The findings of this study illustrate the

complexity of the role of the online instructor through

a unique perspective in which two types of roles were

examined in great detail.

References

[1] Middlesex University Module Guide Handbook, (2008). [2] T.V. Eilertsen, N. Gustafson, P. Salo, Action research and

micropolitics in schools, Educational Action Research 16 (2008) 295-309.

[3] H. Thomas, An analysis of the environment and competitive dynamics of management education, Journal

of Management Development 26 (2007) 9-21. [4] J. Kamau, Challenges of course development and

implementation in a dual mode institution in Botswana, in: Pan Commonwealth Forum on Open Learning: Empowerment through Knowledge and Technology, Darussalam, Brunei, Mar. 1-5, 1999.

[5] I. Mugridge, Response to Greville Rumble’s article “The competitive vulnerability of distance teaching universities”, Open Learning 7 (1992) 59-62.

[6] G. Srikanthan, J. Dalrymple, A Synthesis of a quality management model for education in universities, International Journal of Educational Management 18 (2004) 266-279.

[7] J. Frahm, Developing communicative competencies for a learning organization, Journal of Management Development 25 (2006) 201-212.

[8] S. Parka, S.T. Oliverb, T.S. Johnsonc, P. Grahamd, N.K.

Oppongb, Colleagues’ roles in the professional

development of teachers: Results from a research study of

National Board certification, Teaching and Teacher

Education 23 (2007) 368-389.

[9] R.M. Pallof, K. Pratt, The Virtual Student, John Wiley &

Sons, San Francisco, 2003.

[10] R.S. Hubbard, B.M. Power, The Art of Classroom

Inquiry, Heinemann, USA, 1993.

[11] E. Saito, P. Hawe, S. Hadiprawiroc, S. Empedhe,

Initiating education reform through lesson study at a

university in Indonesia, Educational Action Research 16

(2008) 391-407.

[12] S.D. Gilbert, How to be a Successful Online Student,

McGraw-Hill, San Francisco, 2001.

[13] Z. Berge, The role of the online instructor/facilitator，

Educational Technology 35 (1995) 22-30.

[14] N.W. Coppola, S.R. Hiltz, N.G. Rotter, Becoming a

virtual professor: Pedagogical roles and asynchronous

learning networks, Journal of Management Information

Systems 18 (2002) 169-189.

[15] P.C. Lim, P.T. Cheah, The role of tutor in asynchronous discussion boards: A case study of a pre-service teacher course, Education Media International 40 (2003) 33-47.

[16] D. Maor, The teachers’ role in developing interaction and

reflection in an online learning community, Education

Media International 40 (2003) 127-137.

[17] D. Silverman, Doing qualitative research, SAGE, London, 2000.

[18] H. Altrichter, P. Posch, B. Somekh, Teachers Investigate Their Work, Routledge, London, 1993.


Coordination in Competitive Environments

Salvador Ibarra-Martinez, Jose A. Castan-Rocha and Julio Laria-Menchaca

The Engineering School, Autonomous University of Tamaulipas, Victoria 87000, Mexico

Received: August 02, 2013 / Accepted: August 19, 2013 / Published: August 31, 2013.

Abstract: Despite several researches in autonomous agents important theoretical aspects of multi-agent coordination have been largely untreatable. Multiple cooperating situated agents support the promise of improved performance and increase the task allocation problems in cooperative environments. We present a general structure for coordinating heterogeneous situated agents that allows both autonomy of each agent as well as explicit coordination of them. Such situated agents are embodied for taking into account their situation to solve any action. Indeed, organizational features have been used as metaphor to achieve highest levels of interactions in an agent system. Then, a decision algorithm has been developed to perform a match between the situated agent knowledge and the requirements of an action. Finally, this paper presents preliminary results in a simulated robot soccer scenario showing an improvement of around 92% between the worst and the best cases. Key words: Multi-agent coordination, e-institutions, interactive norms, soccer robotics.

1. Introduction

Coordination depends on how autonomous agents

make collective decisions to work jointly in real

cooperative environments [1]. Nowadays, several

researchers have proposed that autonomous agent

systems are computational systems in which two or

more agents work together to perform some set of

tasks or satisfy some set of goals. Research in

multi-agent systems is then based on the assumption

that multiples agents are able to solve problems more

efficiently than a single agent does [2]. Special

attention has been given to MAS developed to operate

in dynamic environments, where uncertainty and

unforeseen changes can happen due to presence of

other physical representation (i.e., agents) and other

environmental representations that directly affect the

agents’ decisions. Such coordination allows agents to

reach high levels of interaction and increase their

successful decisions, improving the performance of

complex tasks. Agents must therefore work in some

Corresponding author: Salvador Ibarra-Martinez, doctor,

research fields: intelligent systems, autonomous robots and coordination algorithms. E-mail: [email protected].

way and under a wide-range of conditions or

constraints. In fact, an agent system will have to be

handled with a great level of awareness because the

failure of a single agent may cause a total degradation

of the system performance. For thus, this paper aims

to introduce a decision algorithm based on the e-I

(electronic Institution) features [3], which it represents

the rules needed to support an agent society.

Specifically, such algorithm uses knowledge of the

agent situation regards to three perspectives:

interaction with social information and other relevant

details to entrust in other agents or humans; awareness

representing the knowledge of the physical body

reflecting in the body’s skills; and world including

information perceived directly from the environment.

But each type of agent reacts to its perception of the

environment in different ways, modifying the overall

system performance. In particular, a match function

has been formulated to reach a suitability rate based

on the situated agents’ capabilities and the actions’

requirements. In fact, agents can select those actions

for which they are the best qualified. The

effectiveness of this work is illustrated by developing

several examples that analyze cooperative agents’


1106

behavior considering different situations in a real

cooperative environment. Section 2 introduces the

formal coordinated structure introduced in this

approach. In Section 3 an example of the

implementation is presented. Finally, the results and

conclusions are showed in Section 4.

2. Our Approach

A group of situated agents are here presented as

cooperative systems constituted by a group of

autonomous agents who must cooperate among

themselves in order to reach specific goals within real

cooperative environments. When agent interaction

exists, each element of the agent group must be able

to be differentiated from the others. These agents

require a sense of themselves as distinct and

autonomous individuals obliged to interact with others

within cooperative environments (i.e., they require an

agent identification) [4]. This identification refers to

the property of each agent to know who it is and what

it does within the group. In this sense, this work

proposes two agent classifications: CA (coach agents)

and SA (situated agents).

2.1 Adopting e-Features

In order to imitate the ideology of the e-I (i.e., e-I

uses a set of rules to manage the action performance

in groups of agents), the paper describes how agents

that work in temporal groups are able to achieve

collective behaviour. Such behaviour is possible by

using communication among agents. Let us suppose a

scene sα as a spatial region where a set of actions must

be performed by a group of situated agents sα.

},...,,,{where|, 321 njiji ssssSssSss

S is the set of all possible Scenes.

Let us define a coach agent caα in charge of

supervising the execution of the actions in a particular

sα.

CAGcacaGcaca CAjiCAji |,

},...,,,{where 321 ncacacacaCA

where CA is the set of all possible supervisor agents.

When saα has identified its s, it must claim

information in order to know which actions must be

achieved in such s. It is possible, then, to define a saα

as sensitive to the events that happen in real

cooperative environments based on the agent

paradigm [5].

Let us define a situated agent sai as an entity that

has a physical representation on the environment and

through which the systems can produce changes in the

world.

SAGsasaGsasa SAjiSAji |,

},...,,,{where 321 nsasasasaSA

SA is the set of all possible situated agents.

In this sense, sai could be represented in many ways,

(i.e., one autonomous robot with arms, cameras,

handles, etc.) but for the scope of our proposal; sai is

embodied as an entity which is characterized by the

consideration of three parameters: interaction,

awareness and world.

In fact, the paper argues coordination at two

meta-levels (cognitive level—supervision of the

intentions; physical level—execution of the action in

the world, Fig. 1), where the coach agent coordinates

among them to allocate of a set of actions for a group

of situated agents.

Let us define a norm ni that is denoted as a rule that

must be respected or must fix the behavior that a sai

must keep at trying to perform an action in a sα. We

indicated the conception of a norm within a scene

Fig. 1 Levels of interaction.


1107

following a set of rules such that:

if (ni) do/dont {action}

NsNnnsNnn jiji )(|)(,

},...,,,{)(where 321 annnnsN

Let us define an obligation obl as the imposition

given to some sai to perform some action, which it is

established following a set of rules. In order to denote

the notion of obligation obl the predicate [3] is present

as follows:

),,( spaobl i

where a sai is obligated to do in sα.

2.2 Cooperative Actions

Studies about which actions are involved in

determine scene are needed to perceive knowledge

that make possible the organization of any determined

scene. Once a coach knows in which scene it will

develop its function, it must identify the goals to be

accomplished in such spatial region, indicate the tasks

that must be performed to achieve these goals, and

what roles are necessary for the task achievement.

Then, a coach is defined in its knowledge base

KB(caα) by the consideration of a set of goals G, a set

of tasks T and a set of roles R.

)()()()( sRsTsGcaKB

where )( saKB is the information of all the issues

regarding to a specific scene sα, such that: )( sG is

the set of goals, )( sT is a set of all tasks, and )( sR

is the set of all roles involved in determined scene sα.

Indeed, it is necessary to propose a priority index pi

that represents the importance of every action. A saα

will know both the order in which the goals and the

tasks must be performed and the order of the role

allocation process regarding its supervised sα. Such

priority index will be established according to system

requirements (i.e., timeline) in order to achieve the saα

aims.

Goals then embody the overall system purpose;

however, a caα could achieve a particular goal without

the necessity of performing another goal at the same

sα.

GsGggsGgg jiji )(|)(,

},...,,,{)(where 321 oggggsG

1)(0|)()( )( isGii gpPgpsGg

where G is the set of all possible Goals and )( sG is

gβ involved in sα.

Let us to define a set of tasks T which represent the

issues that must be performed to achieve a specific gβ.

Goal then could be achieved without the implicit

necessity of performing all its involved tasks.

Therefore, the tasks selected are independent, but their

development could affect in a positive or negative

way the development of other tasks.

TsTgTttgTtt jiji )()(|)(,

},...,,,{)(where 321 pttttgT

1)(0|)()( )( igTii tpPtpgTt

where T is the set of all possible Tasks.

Let us define a set of roles R which represent the

actions that a pai must fulfil to perform a tγ within a sα.

RsRtRrrtRrr jiji )()(|)(,

},...,,,{)(where 321 qrrrrtR

1)(0|)()( )( itRii rpPrptRr

where R is the set of all possible Roles.

In order to illustrate how this process is performed,

let us suppose a scene s1 which is supervised by the

coach ca1 performing a decision process to define

which goal must be attended firstly (Fig. 2).

2.3 Embodying Situated Agents

Supposing that a situated agent lives in a real

environment, therefore, it has the ability to consider

Fig. 2 The coach ca1 defines which goal must be performed first.


1108

its physical representation in such world. Although

these characteristics could supposedly take a lot of

“things” regarding the environment our proposal takes

three kinds of knowledge that seek to reference all the

information that characterize the perception of

particular sai.

2.3.1 Interaction

Interaction I refers to the certainty that an agent

wants to interact with other agents to assume a

specific behavior with successful and high reliability

to achieve any action proposed within any determined

scene. Such information is useful in the interaction

process of the agents because they can trust in other

agents based on the result of their previous

interactions. Obviously, if a sai has a positive

performance of its actions, its interaction level

increases; but if the outcome of the action is negative,

its interaction level decreases. Such knowledge is

obtained when a sai has a direct relationship with a

caα.

)()(, iisrSAi saIsaIGsa

where )(, isr saI

is the interaction level of a sai to

perform rγ in the sα. 2.3.2 Awareness

Awareness A refers to the set of physical

self-knowledge that a physical agent has represented

about its skills and physical characteristics to execute

any proposed action. Such physical representation is

considered as the embodiment of the physical features

that constitute all the information that physical agents

can include in their decision-making.

Physical agents could be any physical objects

“handled” by an intelligent agent (i.e., an autonomous

robot, a machine or an electric device). Such pai has

features that consider their physical body properties

(i.e., their dynamic, their physical structure) usually

when they commit to perform some task or to assume

a specific behaviour within a cooperative environment.

This fact represents the skill of the physical agents to

know that actions will be performed based on the

knowledge of the physical agents’ bodies, which is

achieved through representation of them on a

capabilities basis.

)()(, iisti paApaAPApa

where )(, ist paA

is the Awareness of pai to perform

t in the sα. 2.3.3 World

World W refers to the set of environmental

knowledge that physical agents have to perform the

proposed set of actions. Such domain representation is

considered as the embodiment of the environment

knowledge that represents all the physical information

that has influence in the physical agents’ reasoning

process.

Let us define a set of world conditions that

represent information about empirical knowledge of

the environmental state, such that:

)()(, iisti paWpaWPApa

where )pa(W is,t is the environmental condition of

pai to perform t in a sα; saα uses the above

information to know the physical situation of each pai. All knowledge of a particular pai )pa(KB i is then

constituted by the information provided for the three

modules, such that:

)]()()([)( iiiii paWpaApaIpaKBpa

In particular, all knowledge related to a specific tγ

in sα is given such that:

)]()()([)( ,,,, istististsti paWpaApaIpaKB

2.4 Communication Process

The humans have a communication process that

allows transmit information or ideas in a common

language to make sure and reliable commitments

between us. Likewise, artificial intelligence has

several approaches showing the same process [5-6] to

exploit the advantages of expressing communication.

To accomplish an action, a group of agents must

establish communication (to coordinate them). On

such coordination agents must “converse” among

them to agree who is who within the group (Fig. 3).

Then, a communication with three simple dialogues


1109

based on the KQML specification is presented as

follows:

),,,(Request nssasa

where saα asks to saβ its θ in the scene sα

),,(Reply sasa

where saβ responds to saα its decision based on the

information dispatched.

),,,(Inform ssasa

where saα informs saβ its state in the scene sα.

This process helps to the saα to communicate

among them and with a pai.

Otherwise, some concepts have been explained

throughout this research work, but none of them has

clarified how a saα could decide who is the pai (or

group) that will take part in the action of its

responsible sα. saα then considers an ID (influence

degree) to all these actions involved in a sα by the

tupla ID(sα) based on the consideration of the

aforementioned parameters to generate an utility

function that helps them in their decision making

structure.

)]s()s()s([)s( TVPKEC idididID

where )s( ECid , )s( PKid and )s( TVid are

values that establish the relevance of each parameter

related to a sα. These values are in the range [0, 1]. In

this sense, the sa responsible in s uses the

stipaKB ,)( and the )s( ID to perform a match

function by means of Eq. (1).

))(1(3

)()(

))(),((3

1)(

3

1)(,)(

,

jj

jjstij

sti

sID

paKBsID

paKBsIDmatch

(1)

A sa uses the match to determine which pai must

perform rq in a s, assigning the higher pai for the

most prior rq in s. In addition, Fig. 4 shows an

example of the match process.

Fig. 3 Conversation between the sa1 and sa2.

Fig. 4 Empirical example of a match process.


1110

2.5 Decision Algorithm

An important criterion for the development of

collective actions within real cooperative

environments is the traffic of the information available

from the perception of the intentions to the execution

of them. We have therefore determined a particular

decision algorithm of four simple stages.

Stage 1. Refers to the property of a saα to perceive

which sα must manage, therefore, a saα then knows its

goals, tasks, roles (the priority of every item is also

perceived) and ID involved in its sα. Hence, the

knowledge base of each saα could be achieved.

Stage 2. All the sa (of the entire SA) must organize

them to define which will be the order in that they

could begin the recruitment of pa to perform the

actions within its sα. For thus, the sa must converse

among them using the developed dialog.

Stage 3. Based on the order obtained above, a saα is

approved to start the communication with the entire

PA to determine that pai will be the selected to

perform every action. For thus, a saα must obtain the

physical knowledge of each pai by means of directly

communication with them; the environment

conditions and trust value of each pai are obtained

when the saα uses the modules aforementioned

(respectively for each parameter). Once a saα completes the )pa(KB i of the entire PA,

it takes such information to perform the match using

the Eq. (1), considering the priority index of all the

roles. Then, saα has a list detailed (form higher to

lowest coefficient) of the entire PA. After, saα knows

that pai must perform that role; therefore, it is able to

obligate a determined pai to perform a role which

represents that action must be performed.

Hence, the best pai (of the entire PA) will choose

the prior role to perform and then others successively,

until all the roles finish in such sα. Such process

guaranty allocates us a suitable role because the rq

always allocated to the best pai. Indeed, a saα knows

how many PA needs because it needs the same

amount of PA, such as R(sα). Suppose that the system

has enough amount of PA to take all the defined roles.

To know, every saα is able to exclude a pai that

presents a lowest action capability.

Stage 4. Show-time. A pai knows the rq that must

perform. This involves physical changes in the

environment. Now, the environment has been

modified. So, a new consensus among the SA could be

performed to adjust it to the current changes in the

environment.

3. Implementation

In our implementation, each physical agent has a

different movement controller which differentiates

from others. Then, we have segmented the scenario

into three spatial regions (Fig. 5) to represent each sα.

For sake of simplicity, we only have defined one

goal per scene G(s1) = g1; G(s2) = g2 and G(s3) = g3.

The consensus to define the execution order of the

scenes is derived as as shown in Fig. 6.

The cbp is the current ball position on the

environment. So, the spatial regions are limited

according to the simulator dimensions (axis x: [0 220];

Fig. 5 Geographic segmentation of the experimental environment.

Fig. 6 Supervisor agent consensus.


1111

axis y: [0 180]). Moreover, specific tasks are defined

in order to accomplish each gi such that:

},{)( },{)(},{ 653432211 ttg Tttg Ttt)T(g

where t1 is make-pass, t2 is shooting, t3 is player-on, t4

is kick-ball, t5 is protect-ball and t6 is covering a

position.

Following the rule presented for the goals, the tasks

also use the cbp as a reference to determine its

execution order.

Then, using the ranges above, a saα may decide the

task to perform at any time. But, to attempt to achieve

such tasks a saα must define which roles it must

perform and the priority order of such roles. Therefore,

by means of human analysis we have proposed four

roles that could be used to perform any task such that:

},,,)(t 4321 rrrrR

where, r1 goes to the ball, r2 kicks the ball, r3 covers a

zone and r4 takes a position to be used in each t.

In addition, we have performed a combination with

the information involved in the environment-based

knowledge. Such combination is used by saα to

perform the match process considering the

aforementioned parameters. Then, a binary

combination lets us generate eight influence degrees

(Table 1). We present a review to show how we have

implemented these parameters in the robot soccer

testbed.

Interaction here called Trust TV represents the

social relationship among agents taking into account

the result of past interactions of a sa with a pai. Eq.

(2) shows the trust calculation if the aim is reached.

Otherwise, using Eq. (3) shows the trust calculation if

the aim is not reached.

)2(),()()( ,, sApatvpatv istist

)3(),()()( ,, sPpatvpatv istist

where the ]1,0[)(, ist patv

and higher )(, ist patv

represents the best pai to perform t in s, ),( sA

and ),s(P are the awards and punishments given

in s respectively and is from 1 to )( sQ and ω is

from 1 to )(' sQ ; that are the number of awards and

punishments in s.

Awareness here called Physical Knowledge, PK

represents the knowledge of the agents about their

physical capabilities to perform any proposed task. In

particular, the introspection process is performed by

using neural networks taking into account the

knowledge that a pai has related to perform t in sα.

Consider that a high ]1,0[)(, ist paPK

by

representing a suitable pai.

World here called Environmental Conditions, EC is

a value related to the distance between the current

location of a pai and the location of the ball. Eq. (4)

shows the calculation: ,

,

( ) (1 ( , ( , ))

/ max(( )) ( ) [0,1]

y a

y a i

t s i i y a

a t s

ec pa d pa r t s

d s ec pa

(4)

where )(, ist paec

is the value of a pai to perform a

tγ in s; )),(,( strpad i is the distance between the

pai with ),( str and )max( sd is the maximal

distance of all pa in s. Then, Eq. (5) shows the )smax(d calculation where m is the total number of

pa in IAS. max( ) max( (1, ),..., ( , )) max [0, 1]a a ad s d s d m s d

(5) In order to show how our approach performs the

role allocation process we present a possible situation

(Fig. 7) where the ball is within the s2 and we use all

Table 1 Influence degree consideration (0: is not considered; 1: is considered).

Influence degree TV PK EC

ID0 0 0 0

ID1 0 0 1

ID2 0 1 0

ID3 0 1 1

ID4 1 0 0

ID5 1 0 1

ID6 1 1 0

ID7 1 1 1


1112

Fig. 7 Possible situation for the PA in the environment.

the influence degrees generated to perform the pa

selection. Then, we only showed the allocation for one

action (kick the ball). In Table 2 we present the values

of a pai regarding to the proposed action. In Table 3

we show the match values obtained by means of Eq.

(1). Then, is possible to see it will be the pai selected

by the sa2 to perform the proposed action.

Additionally, the remained physical agents follow a

fix strategy which was defined to consider actions to

the entire PA.

4. Results and Conclusions

We ran two experimental evaluations to validate the

proposed approach. In particular, in the experiments

our IAS uses all the binary combination of the ID to

perform the match process. In Exp. 1, our IAS

competed against a blind opponent in 30 games. Here,

the IAS performance is improved when all the

parameters are considered. So, IAS(ID7) shows a

better average (improvement rate: +81% better) than

IAS(ID0) (any parameter considered). Then, in the

Exp. 2, a league of 28 games was performed to

confront the IAS among them. So, the IAS

performance increases when using jointly all the

parameters. In fact, the IAS(ID7) shows a better

average (improvement rate: +92%) than IAS(ID0).

As conclusions we argue the need of agent

meta-coordination to exploit the advantages of the

abstract environment knowledge (by the supervisor

agents) and use it to influence the reasoning process

of the physical agents.

In addition, a combination (named Influence

Degree) describes the consideration among these

parameters giving to the sa the ability to determine a

decision process to perform a match between the

scene requirements and the physical agent capabilities.

In fact, the best performance is obtained when our

team agent took into account all the parameters in its

decision process. But it is really interesting to analyze

Table 2 Physical agents’ knowledge bases.

pa Trust Intro. Prox.

kickball,s21 tKB( )pa 0.43 0.47 0.31

kickball,s22 tKB( )pa 0.65 0.52 0.46

kickball,s23 tKB( )pa 0.71 0.69 0.79

kickball,s24 tKB( )pa 0.83 0.77 0.63

Table 3 Some examples of physical agent selection.

ID(s2) pa1 pa2 pa3 pa4 pa Selected

ID1(s2) 0.31 0.46 0.79 0.63 pa3

ID2(s2) 0.47 0.52 0.69 0.77 pa4

ID3(s2) 0.39 0.49 0.74 0.70 pa3

ID4(s2) 0.43 0.65 0.71 0.83 pa4

ID5(s2) 0.37 0.55 0.75 0.73 pa3

ID6(s2) 0.45 0.58 0.70 0.80 pa4

ID7(s2) 0.40 0.54 0.73 0.74 pa4

Table 4 Our approach vs. other approaches.

ID T I P VS

0 0 0 0 References take at least one of these parameters.

1 0 0 1 Not references yet.

2 0 1 0 [4-7]

3 0 1 1 [8-10]

4 1 0 0 No references yet.

5 1 0 1 [11-13]

6 1 1 0 [14]

7 1 1 1 [15]


1113

how the cooperative IAS performance increases when

the system takes the parameters into consideration. In

conclusion, the situation matching approach is a

promising method to be used as utility function

between task requirements and physical agent

capabilities in MAS.

In Table 4 we show some approaches regarding

architecture for multi-agent cooperation. In particular,

these architectures express behavior by implementing

different kinds of knowledge, which can be related to

our approach.

References

[1] S. Ibarra, C. Quintero, J. A.Ramon, J. Ll de la Rosa, J. Castan, PAULA: Multi-agent Architecture for coordination to intelligent agent systems, in: Proc. of European Control Conference (ECC’07), Kos, Greece, July 2-5, 2007.

[2] D. Jung, A. Zelinsky, An architecture for distributed cooperative-planning in a behaviour-based multi-robot system, Journal of Robotics & Autonomous Systems (RA&S) 26 (1999) 149-174.

[3] M. Esteva, J.A. Rodriguez, C. Sierra, J.L. Arcos, On the formal specification of electronic institutions, Agent Mediated Electronic Commerce Lecture Notes in Computer Science 1991 (2001) 126-147.

[4] B. Duffy, Robots social embodiment in autonomous mobile robotics, Int. J. of Advanced Robotic Systems 1 (2004) 155-170.

[5] S. Russell, P. Norving, Artificial Intelligence: A Modern Approach, 3rd ed., Ed. Prentice Hall, London, Dec. 2009, p. 1152.

[6] M. Luck, P. McBurney, O. Shehory, S. Willmott, Agent Technology: Computing as Interaction (A Roadmap for Agent Based Computing), AgentLink, 2005.

[7] A. Oller, DPA2: Architecture for co-operative dynamical

physical agents, Doctoral Thesis, Universitat Rovira I Virgil.

[8] C.G. Quintero, J. Ll. de la Rosa, J. Vehi, Self-knowledge based on the atomic capabilities concept—A perspective to achieve sure commitments among physical agents, in: 2nd International Conference on Informatics in Control Automation and Robotics, Barcelona, Spain, Sep. 14-17, 2005.

[9] L. Pat, An adaptative architecture for physical agents, in:

IEEE /WIC/ACM International Conference on Intelligent

Agent Technology, Sep. 19-22, 2005 pp. 18-25.

[10] D. Busquets, R. Lopez de Mantaras, C. Sierra, T.G.

Dietterich, A multi-agent architecture integrating learning

and fuzzy techniques for landmark-based robot

navigation, Lecture Notes in Comp. Science 2504 (2002)

269-281.

[11] C.G. Quinero, J. Zubelzu, J.A. Ramon, J. Ll. de la Rosa,

Improving the decision making structure about

commitments among physical intelligent agents in a

collaborative world, in: In. Proc. of V Workshop on

Physical Agents, Girona, Spain, Mar. 25-27, 2004, pp.

219-223.

[12] R.S. Aylett, D.P. Barnes, A Multi-robot architecture for

planetary rovers, in: 5th ESA Workshop on Advanced

Space Technologies for Robotics and Automation,

ESTEC, Noordwijk, The Netherlands, Dec. 1-3, 1998.

[13] R. Simmons, T. Smith, M. Bernardine, D. Goldberg, D.

Hershberger, A. Stentz, et al., A layered architecture for

coordination of mobile robots, in: Multi-robot Systems:

From Swarms to Intelligent Automata, May, 2002.

[14] C.G. Quintero, J.L. de la Rosa, J. Vehi, Physical Intelligent Agents’ Capabilities Management for Sure Commitments in a Collaborative World, Frontier in Artificial Intelligence and Applications, IOS Press, 2004, pp. 251-258.

[15] S. Ibarra, C. Quintero, J. Ramon, J.L. De la Rosa, J. Castan, Studies about multi-agent teamwork coordination in the robot soccer environment, in: Proc. of 11th Fira Robot World Congress 2006, 2006, pp. 63-67.


Logistics Customer Segmentation Modeling on Attribute

Reduction and K-Means Clustering

Youquan He and Qianqian Zhen

Department of Information Science & Engineering, Chongqing Jiaotong University, Chongqing 400074, China


Abstract: To develop logistics customers’ potential demands for logistics services, and raise the level of logistics enterprise services, the research for customer segmentation has become a primitive work of logistics enterprises in order to run a differentiated customers’ marketing. Through the use of clustering algorithm, this paper presented a segmentation modeling for differentiating customers in logistics industry. Firstly, based on attribute reduction, redundant properties were simplified in the complex data mining under variable parameters in order to improve the quality and efficiency of the modeling, and then the customer segmentation model was constructed through unsupervised clustering K-Means algorithm. It was verified that the logistics users have the obvious differentiation of characteristics by using the cluster model. And a logistics enterprise achieved significant benefits with application of the model in the differentiated data service marketing.

Key words: Customer segmentation, PCA, rough set, K-Means, logistics enterprise.

1. Introduction

Customers are one of the most important resources

of an enterprise, customer retention and satisfaction

drive enterprise profits. With the development of the

internet, the market competition intensifies, customers

become more and more diversified. To better identify

customers, allocate limited enterprise resources and

improve core competitiveness, it is important to do

customer segmentation. Customer segmentation is the

key to successful customer retention. After years of

development, the theory and methods of customer

segmentation are constantly improved. It also has

been used in differentiated marketing in various

industries, such as banking business [1],

telecommunications industry [2], retail business [3],

securities business [4], aviation industry [5] and some

other data-intensive industries. However, it is less

used in logistics industry.

Corresponding author: Youquan He, Ph.D., professor,

research fields: information processing and data mining. E-mail: [email protected].

With the development of logistics informatization,

many logistics enterprises have a lot of information

systems, and also accumulated a large amount of data.

Traditional methods of customer segmentation are not

strong enough to deal with increased enterprise data

and more complicated customer segmentation. The

appearance and development of the data mining

technology makes it possible to find new solutions for

big data based and complicated customer

segmentation cases. Clustering algorithm is one of the

most important data mining algorithms [6].Therefore,

this paper presented a segmentation modeling for

differentiating customers in logistics industry through

the use of clustering algorithm.

2. Problem Description and Analysis

2.1 Problem Description

Customer segmentation is the basis of differentiated

management and improving the level of logistics

service for logistics enterprises. The problem

descriptions of customer segmentation like this:

Logistics Customer Segmentation Modeling on Attribute Reduction and K-Means Clustering

1115

Considering that the quantity of customers is M,

followed by numbered 1, 2,..., M, customers have N

characteristic attributes. Aij represents the value of the

j-th attribute of the i-th customer. Therefore, M

customers based on their different values of

characteristic attributes can be expressed as M vectors

of N × 1 column. Vector distance can be used as the

basis of customer segmentation, which can help us

determine the similarity of the customer. And the

similar customers can be divided into a category,

distinguished from the dissimilar customers.

Sometimes, the customer attributes exist

redundancy, which have an effect on the customer

segmentation. So it can not reach the expected results.

Therefore, reduction eliminating redundant attributes

is necessary for the characteristics attributes of

customers to improve the quality of segmentation.

2.2 Model Formulation

The model is applied to the logistic customers

segmentation as shown in Fig. 1:

According to the above analysis, the customer

attribute values can be seen as enter data, the

dispersed client base can be seen as class, and then

customer segmentation problem can be translate into

cluster problem. Considering the current domestic

logistics enterprise data characteristics and the

characteristics of clustering methods, we can choose

K-Means clustering method to achieve customer base

clustering.

K-Means algorithm is a fast iterative clustering

algorithm [7], it divided the set of n objections into k

clusters, and set distance as the “degree of affinity”

indicators between measurable objections, this can

cause high similarity within a cluster, and low

similarity inter-cluster. Similarity of the cluster is

measured by objective means in cluster, which is seen

as a cluster centroid or the center of gravity.

Suppose there have m-dimensions data set X,

dividing its clusters into k clusters w1, w2, …, wk, their

centroids are c1, c2, …, ck, and then we can obtain:

1

i

ix w

c xn

(1)

where, i is the number of data point and n is the

number of objection in the data set X.

Fig. 1 Logistics customer segmentation model framer.


1116

Attribute reduction is one of core contents in the

rough set theory [8]. It is the transformation of

high-dimensional data into a meaningful

representation of reduced dimensionality that

corresponds to the intrinsic dimensionality of the data

[9]. There have two basic conceptions in attribute

reduction: reduct and core.

Suppose RS is the gens relation of equivalence, R

∈ RS, If ind(RS) = ind(RS – {R}), then we can say R

is unnecessary in RS; conversely, we can say R is

necessary in RS. If every R ∈ RS are necessary in R

∈ RS, then we call Rs is independent; or they are

dependent, and can make reduction. If PS and Qs are

independent, and ind(QS) = ind(PS), then we can say

QS is one of reductions of PS. The set of all necessary

relationship in PS are being called the core of PS,

which is signed as core(PS).

3. Solution Algorithm

3.1 Attribute Reduction Used for Raw Data

Suppose U is the whole data field, Ui is the i-th

element in U, i ∈ 1, 2, 3, …, M; Aj is the j-th

attribute in U, j ∈ 1, 2, 3, …, N; Sij is the attribute

value corresponds to the i-th element, j-th attribute.

According to a specific problem, the importance of

each attribute is not the same, and also different in the

decision making. Therefore, this paper achieves

attribute reduction through calculating the importance

of the attribute. As shown below, it is the calculation

steps of attribute importance.

Algorithm: calculation of attribute importance;

Input: the attributes used for calculation of attribute

importance, and information systems;

Output: the scores of each attribute importance in

information systems;

Begin

(1) calculating core(RS) , RsR , calculating

Rs RR s Rs RR s R

R s R s

DR s R s1

R s

C ard Posr rSig

r C ard Pos D

,

all of the Sig attribute value which is greater than zero

constitute core(Rs), it may be ￠;

(2) Red(Rs) ← Core(Rs);

(3) Estimating whether ind(Red(Rs)) is equal to

ind(Rs). If they are equal, go to step 6; otherwise go to

step 4; (4) Calculating all of the SigRed(Rs)(R) value , R ∈

Rs – Red(RS), R1 accords with:

1 maxRED Rs RED RR Rs RED Rs

Sig R Sig R

(2)

(5) Red(RS) ← Red(RS)∪{R1}, go to step 3.

(6) Output minimum reduction Red(RS).

End

3.2 To Achieve K-Means Clustering Algorithm

The target of K-Means clustering: the data set is

divided into k clusters according to the specified

parameter k. Its core steps are as follows:

(1) Specified the cluster number k and the initial

clustering center;

(2) Cluster according to the principle of the closest;

(3) Redefining k classes centers.

Support i = i + 1, recalculate k classes centers using

Eq. (1), preparing for further iterations.

(4) Determine whether the clustering termination

condition is satisfied.

Calculating 2

i

k

ii p c

E p c

, if 1E i E i is

satisfied, K-Means algorithm is ended, otherwise go

to step 2, operation will continue.

The flow of K-Means algorithm as follows：

Algorithm: K-Means (S, k), S = {x1, x2,…,xn}

Input: n data object collection xi

Output: k cluster centers Zj, and k cluster data object

collection Cj

Begin

m = 1

initial k prototype Zj, j ∈ [1, k]

repeat

for i = 1 to n do

computer D(xi, zj) = ︱xi – zj︱

if D(xi, zj) = min{D(xi, zj)} then xi ∈ Cj

end


1117

if m = 1,then

2

1 i j

k

i jj x C

E m x z

m = m + 1

for i = 1 to k do

1

1 jnj

i iij

Z xn

,

2

1 i j

k

i jj x C

E m x z

Until ( ) ( 1)E m E m

End

4. Case Analysis

HK logistics enterprise is a logistics transport

enterprise that whole business covers the

Jiangsu-Zhejiang-Shanghai area, which is founded in

1997. It is business scope includes cargo

transportation, logistics planning, distribution,

value-added services, information processing etc., and

it still stay in expanding. HK enterprise changes into

an integrated modern logistics company with

transpotation, distribution, storage, circulation process

and information handling services from a traditional

freight company gradually. Through the in-depth

understanding of HK logistics enterprise’s present

situation, searching it is related databases, to collect

and collate customer information that buy it is service.

After eliminating some “problematic data”, get 25

sample raw data. As shown in Table 1, it includes

yearly relative profit margins (D11), yearly profit

contribution rate (D12), the number of customer

distributed (D21), transportation volume (D22),

frequency of transportation (D23), profitability (D31),

scale (D32), business environment (D33), possibility

of cross marketing (D41), business growth rate(D42),

profit margin rate (D43), allocation rate (D51), bank

credit rating (D52), the average debt rate yearly (D53),

repetitive purchase rate (B21), customer share(B22),

customer relationship intensity (B23), customer’s

price-sensitivity (B24) and customer switching costs

(B25), a total of 19 attributes.

4.1 Data Preprocessing

If the data of the 25 groups are clustered through

K-Means, for the more input attribute, K-Means is

unable to identify the redundant attribute, so it is

easier to cause slower clustering speed, the falling in

the quality of clustering, etc.. After reducing 19

attribute using the rules of attribute reduction in rough

set, only use the key attributes as input variables of

the K-Means clustering to solve the problem of

redundant attributes. Take HK logistics enterprise for

example to cluster simulation, first of all, normalize

the raw data, the result is shown in Fig. (2).

Then, using attribute reduction algorithm to reduce

the data set based on attribute importance. First,

calculate the nuclear Core (F) is {F4}. Then, on the

basis of the nuclear, surplus the rest of the condition

attribute gradually according to the attribute

importance, the process is continuing until it meets the

conditions. Finally, the reduction results are shown in

Table 2.

Table 1 Some sample data.

CusId C1 C2 … C25

D11 55 3.4 … 7.55

D12 40 13.3 … 9.4

D21 16 3 … 3

D22 10 4 … 1

D23 13 2 … 8

D31 5 4 … 4

D32 5 2 … 2

D33 5 4 … 4

D41 5 3 … 3

D42 5 3 … 3

D43 26.25 5.25 … 6.2

D51 6.87 25.98 … 55.8

D52 5 4 … 4

D53 58 22 … 45.5

B21 35.4 55.6 … 54.7

B22 60 67 … 65

B23 4 3 … 3

B24 2 3 … 3

B25 2 5 … 5


1118

Fig. 2 Discretization of data set.

Table 2 Data reduction.

X1 X2 X3

C1 0.70985 0.62773 -1.47345

C2 0.12587 0.07789 0.99645

C3 0.62016 -1.62654 0.52526

C4 1.34267 0.64071 -0.3608

C5 0.10486 1.32636 -1.09793

C6 -0.79073 -0.6545 1.77739

C7 -0.70061 0.91469 0.4336

C8 -0.96809 -1.07615 -0.50857

C9 -1.83022 0.21416 -0.459

C10 0.05782 -0.78809 -1.17345

C11 -1.48158 -0.07978 -0.09249

C12 0.04591 0.32934 0.81276

C13 -0.8961 -0.2026 -1.16839

C14 -1.04842 -1.41836 -0.7961

C15 1.63087 0.26136 -0.30325

C16 -0.44634 1.06685 0.39702

C17 0.2321 -1.69705 1.27427

C18 -1.20336 1.40651 -1.40786

C19 0.30905 -1.38533 -0.07175

C20 0.00047 0.04238 1.28661

C21 1.48548 -0.90059 -0.78134

C22 1.72006 1.23757 0.98059

C23 1.4249 -0.24336 -1.17173

C24 -0.3348 1.76434 1.37100

C25 -0.10979 0.16244 1.01117

4.2 The Results of Cluster Analysis

Calculate the data on Table 3 by K-Means

clustering computing, the experiment results are

obtained as shown in Table 3.

Table 3 shows the clustering result of each

customer. Compared with HK logistics enterprise’s

customer segmentation results gained from the ABC

classification method, we can find some changes of

the following:

For example, the customer C22, it is placed on the

second category in the original subdivision which is

placed on the first category. C4 and C18 are placed on

the worst category in the original subdivision which

are placed on the third and the second. C1 and C12 are

placed on the first category in the original subdivision

which are placed on the third and fourth. There is a

big difference between the company only measure

each customer in accordance with the customer’s

current value and the method used in this paper. The

former method only considered the customer’s current

value, but customer’s development potential and

relationships are not been considered. This method is


1119

Table 3 Customer segmentation results.

Category Customer ID

1 C1、C4、C15、C21、C23

2 C8、C9、C10、C11、C13、C14

3 C22

4 C5、C7、C16、C18、C24

5 C2、C3、C6、C12、C17、C19、C20、C25

Table 4 Comparison of result.

The number of attribute

The total error rate (%)

K-Means 19 12.09

RS-K-Means 3 5.24

easy to cause enterprises are unable to identify

valuable customers. We can solve this problem

through the latter method. In customer segmentation

issues, the latter method is more reasonable than the

previous method.

As shown in Table 4, after attribute reduction

processed data the results are even better. This proves

that the proposed methods and models for HK

enterprises are feasible and effective.

5. Conclusions

Many indicators, redundant attributes are the

characteristics of the logistics enterprise data.

Attribute reduction based clustering method

mentioned in this paper could eliminate redundant

attributes, improve the quality of clustering.

Instantiated with HK logistics enterprises have proved

the effectiveness of the method. Customers fine

classification is based on the customer attributes

classified feature class. This method is conducive to

the logistics enterprises to change the business model,

extensive change management as differentiated

marketing, improve competitiveness; manage

customers based on customer demand, and allocate

service resources reasonable.

References

[1] X. Li. Establishment and application of commercial bank customer segmentation model, Statistics and Decision 9

(2008) 144-146.

[2] H. Ahn, J.J. Ahn, K.J. Oh, D.H. Kim, Facilitating

cross-selling in a mobile telecom market to develop

customer classification model based on hybrid data

mining techniques, Expert Systems with Applications 38

(2011) 5005-5012.

[3] V.L. Migueis, A.S. Camanho, J.F. Cunha, Customer data

mining for lifestyle segmentation, Expert Systems with

Applications 39 (2012) 9359-9366.

[4] Y. Wang, The securities industry customer segmentation

model building and empirical research, ShangHai

Management Science 34 (2012) 30-35.

[5] J.J.H. Liou, G.H. Tzeng, A dominance-based rough set

approach to customer behavior in the airline market,

Information Science 180 (2010) 2230-2238.

[6] M. Bottcher, M. Spott, D. Nauck, R. Kruse, Mining

changing customer segments in dynamic markets, Expert

Systems with Applications 36 (2009) 155-164.

[7] J.W. Han, M. Kamber, Data Mining Concepts and Techniques, Mechanical Industry Press, Beijing, Mar. 2005, pp. 253-273.

[8] S.S. Hu, Y.Q. He, The Theory and Application of Rough Decision, Beijing University of Aeronautics and Astronautics Press, Beijing, Apr. 2006, pp. 1-12.

[9] R. Dash, R. Dash, D. Mishra, A hybridized rough-PCA approach of attribute reduction for high dimensional data set, European Journal of Scientific Research 44 (2010) 29-38.


UHF Propagation Parameters to Support Wireless

Sensor Networks for Onboard Trains

B. Nkakanou1, 2, G.Y. Delisle1, 2, N. Hakem1 and Y. Coulibaly1

1. LRTCS-UQAT, Val d’Or, Qc G1V 0A6, Canada

2. Dept. of Electrical Engineering and Computer Engineering, Laval University, Québec G1V 0A6, Canada

Received: July16, 2013 / Accepted: August 7, 2013 / Published: August 31, 2013.

Abstract: This paper reports numerical results for the characterization of the propagation channel in a train. Since the availability of a train to carry out measurements is not always easy, particularly when many changes must be done, a simulation tool provides a useful and reliable mean for the evaluation of the propagation characteristics of this complex and highly fluctuating channel. In order to benefit from previous results, the various existing softwares for complex electromagnetic fields environments simulations were fully searched and one that seems best suited has been retained for these computations. The results presented here are original, preliminaries and our approach provides a basis for study the propagation of waves in a very complex environment consisting of different electromagnetic fields like a train.

Key words: Train, channel propagation, wireless communication, electromagnetic fields, interference.

1. Introduction

Railway transportation has always been one of the

most effective means of transporting passengers and

goods and this being amplified in the last few years.

The passenger numbers has increased progressively

while the volume of goods and equipment increased

significantly. With the passengers and goods

transportation in a modern telecommunication world,

new challenges arise, notably the transmission of

passenger data, sensors data, etc., both for

communications and security purposes.

Wired communications have many drawbacks for

train communication systems (locomotives and

wagons), particularly when a large number of sensors

are in use in many locations. Indeed, most of the

system components must still be connected by cables.

Wireless communications systems in the railway

industry would represent an attractive alternative to

Corresponding author: Nadir Hakem, Ph.D., prof.,

research fields: wireless communications, confined areas, RF characterization and modeling. E-mail: [email protected].

wired and optical fiber communication system but the

reliability of the transmitted and received data must be

secured. Wireless communications systems in train

would allow acquiring information on various physical

parameters of the locomotives (fuel, position, dynamic

load, speed, axle load, wheel flat, etc.), transferring

data to the fixed points of acquisitions (wayside),

warning train drivers of impending dangers,

controlling different wagons and giving the exact

position of the train to the signalers.

Locomotives, wagons and special cars are

constructed mostly of metallic materials such as steel,

copper, aluminum or various alloys that severely

impact signal propagation. Radio waves propagating in

such environments dissipate some of their energy

within the structures and the induced currents generate

EM fields that are reradiated and they therefore alter

the propagation characteristics of the original RF signal.

Wireless communication is generally severely affected

by electromagnetic interferences [1-4].

For a comprehensive evaluation and understanding

UHF Propagation Parameters to Support Wireless Sensor Networks for Onboard Trains

1121

of the main implications of installing wireless

communication system onboard trains, the RF signal’s

propagation characteristics must thoroughly be

analyzed [5-13].

It would be very helpful to characterize the

propagation of electromagnetic fields to ensure the

quality of communications systems, quantify these

electromagnetic fields and then evaluate the most

penalizing electromagnetic noise [1-9].

The fact that our work focuses on the propagation in

an environment where the electromagnetic field

undergoes a lot of variations; the precise

characterization of this environment is crucial [5-13].

One of the aspects that must be considered is also the

communications among different sensors placed at

different locations in the locomotive and cars

[14-19].

Sensing devices must be located very close to the

monitored event sources (such as wheels, brakes,

boxcar doors, refrigerator units, etc.) without impeding

the signal path with obstacles, low ground clearance

and other impairments [14-16].

The objectives of this paper are to present how a

simulation and characterization of the propagation

channel in a train can be handled. The detail of the

procedures and measurement setup used to collect the

data against which simulation results shall be validated

will be reported and discussed in full details as soon as

completed. The strategy and the correct way to process

the acquired data are also explained. Finally, the

relating results to the signal quality, small-scale effects,

large-scale path-loss exponents and time dispersion

parameters are discussed. With the parameters of time

dispersion in particular RMS (root mean square) delay

spread, the coherence bandwidth can thus be

determined.

The paper is organized as follows: in Section 2, the

software environment and simulation setup are

presented. In Section 3, some relevant parameters of

the propagation channel, based on simulation, are

presented. Section 4 presents the conclusions.

2. Setup

For the wave propagation, a 3D ray-tracing tool

based on the UTD (uniform theory of diffraction) and

the theory of GO (geometrical optics) has been used.

The model includes modified Fresnel reflection

coefficients for the reflection and the diffraction, based

on the UTD. One of the major advantages of a 3D

ray-tracing tool is the wideband analysis of the channel.

Also, the frequency selectivity (delay spread) and time

variance (Doppler spread) of the channel can be easily

determined.

2.1 Software and Overview of XGtD

The first step of this study is the simulation of the

radio propagation channel and, for that purpose; a

rigorous simulation is conducted using XGtd®, of

Remcom Inc., a specialized package for EM-Field

simulations [20].

XGtd® is a ray-based electromagnetic analysis tool

for assessing the effects of a vehicle on antenna

radiation, estimating RCS (radar cross section) and

predicting coupling between antennas. This software

also allows us to evaluate the interaction between

electronic circuits and radio waves. Basically, it

simulates the interaction and interrelation between EM

fields and structures, based on classical techniques

such as GO (geometric optics), the UTD (uniform

theory of diffraction), PO (physical optics) and the

MEC (method of equivalent currents).

Performance and memory requirements are less

dependent on the electrical size of objects than full

wave methods. XGtd includes a number of features that

extend the capabilities of ray tracing. These

capabilities provide data suitable for various

applications namely far zone antenna radiation

diagrams for surface-mounted antennas, monostatic or

bistatic RCS, creeping waves diffraction, high-fidelity

field prediction in shadow zones, Doppler for moving

transmitters and receivers, coupling between antennas

and wide range of outputs for predicted EM fields and

channel characteristics [20]. To obtain results that are


1122

the closest possible to the reality, the available 3D

software environment has been fully exploited in

conjunction with 3D objects. Material properties have

been assigned to each individual component to fully

represent the 3D objects and their properties.

Assigning these properties is critical to the entire

simulation and evaluation process.

2.2 3D Train

In order to simulate with real train data, the Siemens

Velaro platform [21] which is one of the fastest

currently operating high-speed train in the world has

been chosen. High-speed trains of this kind are already

reliably operating in several countries around the world

including Germany, Spain and China. The technical

characteristics of this train are supplied in Table 1 [21].

In our simulations, the data for the train (train

Germany) shown in Fig. 1 has been purchased from

3DcadBrowser [22]. It’s composed of 11,940 polygons,

7,660 vertices, 430 sub-objects and 11 materials with a

length of 24 m.

The 3D model is imported in the simulation software.

This import was made very extensive due to the large

number of materials and polygons of the model. Due to

the difficulty to import the 3D train model, the number

of materials and polygons were reduced. Therefore, in

the simulation process, a maximum of six reflections

on the different faces of the train, two transmissions

and zero diffraction were considered. Due to the large

computation time required in simulation process, the

addition of one single diffraction increases the

computation time by about 18 h. Dipoles antennas at

990 MHz with a bandwidth of 20 MHz and several

transmitters were located within the train. The

receptors were located within and outside of the train.

In our scenarios, the green points represent

transmitters while the red ones are receivers. The

transmission power has been fixed to 0 dBm.

Table 1 Technical characteristics of the train.

Technical data

Maximum speed 320 km/h

Length of train 200 m

Length of end car body 25,535 mm

Length of intermediate car body 24,175 mm

Distance between bogie centers 17,375 mm

Width of cars 2,950 mm

Height of cars 3,890 mm

Track gauge 1,435 mm

Empty weight 439 t

Voltage system 15 / 25 kV AC and 1.5 / 3 kV DC–maximum

Traction power 8,000 kW

Gear ratio 2,62

Starting tractive effort 283 kN

Brake systems Regenerative, eddy-current brake, pneumatic

Number of axles 32 (16 driven)

Wheel arrangement Bo’Bo‘+2‘2‘+Bo’Bo‘+2‘2‘+2‘2‘+Bo’Bo‘+2‘2‘+Bo’Bo‘

Number of bogies 16

Axle load < 17 metric tons

Acceleration 0–320 km/h 380 s

Braking distance 320–0 km/h 3,900 m

Number of cars / train 8

Number of seats 485 / 99 / 386 (total / 1st / 2nd Class)


1123

Fig. 1 Train used in our 3-D simulation [22].

3. Results

3.1 Path Loss

The path loss of the channel represents the

attenuation a signal undergoes when transmitted

through the medium, and is an important parameter in

the design of wireless communication systems

regarding the coverage. Path loss is defined as the ratio

of the effective transmitted power to the received

power; a reference measurement is performed at a

distance from the transmitter. Using the log-normal

shadowing assumption, the path-loss exponent, n, is

related to the receiver power at a distance d is given by

[23]:

10 log (1)

where， is the mean path loss at the reference

distance and is a modeling factor for the

shadowing effect. It’s represented by a zero-mean

Gaussian distributed random variable expresses in dB

with the standard deviation σ. The mean path loss at

and the path loss exponent n are determined using a

least square regression analysis. The Gaussian random

variable represents the difference between this

fitting and the simulated data.

By considering antennas located in the middle of the

train (Fig. 2), the path loss exponent obtained is equal

to 1.34.

Fig. 3 presents this result, slightly different from the

path loss exponent in free space (n = 2). This result

indicates that the path loss in the train car is less than

that in free space. The result of path loss exponent

values is due to multipath signal addition and they are

similar to those observed in Ref. [23] for indoor

environments.

However, this lower value of n is explained by the

presence of multipath caused by the complex structure

of the train.

By considering the antennas located on the floor of

the train, the value of n increases to stand at n = 4.3

(Fig. 4). This value is large but can be explained by the

Fig. 2 Rx Inside (reds) and Tx (green) antennas on the middle of the train.

Fig. 3 Path loss in the middle of the train.

Fig. 4 Path loss in the floor of the train.

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14-45

-40

-35

-30

-25

-20

Distance

Path

loss

(dB

)

Path loss vs. Distance

y = - 13.461*x - 22.817

Simulated Data

linear fit

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

-160

-140

-120

-100

-80

-60

-40

-20

0

Path

loss

(dB

)

Path loss vs. Distance

y = - 4.376*x - 4.848

Data simulated

linear fitting


1124

presence of surface waves which introduces

interference.

After obtaining the path loss for a transmitter located

in the middle of the wagon, the impact of this shift on

the transmitter power loss has been evaluated. The

transmitting antenna is placed on the roof of the wagon

in different places (middle and two extremities) as in

Fig. 5 and the data associated therewith has been

recorded.

These have allowed us to trace the path loss as

shown in Fig. 6.

The path loss exponents for receivers located on the

ground are 5.34 and 6.59 for transmitting antennas

located in the middle and at one extremity of the wagon,

respectively.

The position of the sensor is thereby important

because it may have an impact on radio wave

propagation and degrade received signal power in the

wagon. Comparing the path loss exponent of power

loss for transmitters situated in the middle of the wagon,

the losses in the wagon are greater at ground level than

on the edges. In addition, when the transmitting

antennas are located on the roof, these losses are even

more pronounced. Thus, the losses are stronger when

the data from the sensors are collected on the ground

with antennas located on the roof.

3.2 Impulse Response

The impulse response of a system is a useful

characterization of the system. A radio propagation

channel can be completely characterized by its impulse

Fig. 5 Rx Ground (reds) antennas on the ground and Tx roof (green) antennas on the roof of the train.

(a) middle

(b) extremity

Fig. 6 Path loss on the floor of the train.

response. The impulse response of a multipath channel

can be modeled as:

∑ (2)

where, N is the number of multipath components, ,

and are the random amplitude, arrival time and

phase of the kth path, respectively, and δ is the delta

function. The phases are assumed to be statically

independent uniform random variables distributed [23]

over [0, 2π].

The impulses responses are plotted at 21 m (Fig. 7)

then at 10 m (Fig. 8) with Tx roof antenna.

3.3 RMS Delay Spread

The time dispersive properties of broadband

multipath channels are most commonly quantified by

RMS delay spread. RMS Delay spread parameters

highlight the temporal distribution of power relative to

the first components. Delay spreads restrict the

transmitted data rates and could limit the capacity of the

0 2 4 6 8 10 12-120

-100

-80

-60

-40

-20

0

y = - 5.34*x - 9.24

Simulated data

linear fitting

0 2 4 6 8 10 12 14-120

-100

-80

-60

-40

-20

0

y = - 6.593*x - 18.11

Simulated data

Linear fit


1125

Fig. 7 Impulse response at 21 m.

Fig. 8 Impulse response at 10 m.

system when multi-user systems are considered. The

number of multipath in a train is important due to the

reflection and scattering from the surface, ceiling, side

walls and electromagnetics noises. The best parameter

that allows us to measure multipath is the root mean

square delay spread. This parameter determines the

frequency selectivity of channel, which degrades the

performance of digital communication systems over

radio channels. The RMS delay spread also limits the

maximum data transmission rate that can be transmitted

by the channel. Without using diversity or equalization,

the RMS delay spread is inversely proportional to the

maximum usable data rate of the channel.

The formulation of time dispersion parameters is

given in Ref. [23] by

(3)

where, and are the Mean excess delay and the

second moment of the PDP respectively. The mean

excess delay can be computed as:

∑

∑ (4)

∑

∑ (5)

where, and are the power and delay of the

thk path respectively. It is measured relative to the

first detectable signal arriving at the receiver at 0

Delay spreads over distance was computed and Fig. 9

shows the computed results.

The maximum is obtained at a distance of 15 m

within the wagon and is equal to approximately 14 ns.

In light of this result, it can be concluded that there is no

relation between delay spread and distance as in many

complex environments.

The CDFs of delay spread with transmitter in the

middle of wagon obtained are shown in Fig. 10.

Fig. 11 compares the CDF of the delay spread for the

receiving antennas on the ground and inside the car. On

the ground, the distributtion of delay spread is

completely different from that on the edges or inside the

wagon.

In order to measure impact of the position of

transmitting antennas on signal propagation and also to

know the impact of positioning of the receiving

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

4

6

8

10

12

14

16

Distance (m)

De

lay

spre

ad

(ns)

Delay spread vs. Distance


1126

Fig. 9 Delay spread versus distance.

Fig. 10 CDF of delay spread Tx Point.

Fig. 11 CDF of delay spread Tx roof.

antennas, several scenarios have been adopted. The

first scenario consisted of placing a transmitting

antenna in the middle of the train (Tx point). Receivers

were positioned at various locations in the car. These

positions were dictated by different potential positions

of sensors. It was started by positioning receivers

inside and in the middle of the train therefore locating

21 antennas spaced 1 m apart and called them Rx

Inside (Fig. 2).

Then the knowledge of the interactions on the

different compartments of wagons can prove to be

necessary. It is for this reason that it has been decided

to locate the receiving antennas on different sides of the

cars. Thus, 21 receivers have been placed on the left

wall (Rx Route 1) and also on the right wall (Rx Route

2) (Fig. 12).

The surface of the wagons is obviously one of the

areas of interest since several sensors are therein

located. These, positioned under the floor pan are

extremely important. Indeed, they provide most of the

signaling information. For these reasons, a grid (0.5 ×

0.5) of several antennas (282) has been placed on the

surface of the car. These antennas are separated by 0.5

m (Rx Ground) (Fig. 5).

Finally, the information gathered in a particular

wagon should be transmitted to a processing point

which can be located in the leading wagon. Thus, the

train-to-train communication process is equally

important.

The transfer of data from various wagons is through

wagon to wagon communications. The latter are

generally separated from a distance of up to 2 m. To

achieve this, four antennas were located outside for the

simulation. These antennas are named Rx Outside

(Fig. 13).

Fig. 12 Rx Route 1 and 2 receivers.

Fig. 13 Rx Outside receivers.

0 0.5 1 1.5 2 2.5 3

x 10-8

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay spread (s)

Cum

ulat

ive

prob

abili

ty

Delay spread cumulatve probability

Inside

Route 1 or 2Outside

0 1 2 3 4 5 6 7

x 10-8

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay spread (s)

Cum

ulat

ive

prob

abili

ty

Delay spread cumulatve probability

Inside

Route1 or 2

Ground


1127

With this first scenario, different parameters have

thereby been obtained. Depending on the distance, the

different delays spread between them were plotted and

compared. Fig. 14 shows the results and Table 2 gives

the average values in each case.

Careful reading of this table permits to conclude that

the average delay spread does not vary much with

respect to the internal antennas of the train. This result

is important insofar as the positioning of the sensors in

the wagon is critical. In addition, this result tells us that

with sensors located on the left or right sides of the car,

the same signal perturbations occur. At the surface,

these perturbations are in the same order of idea just in

varying slightly. This slight variation can be explained

easily by the path loss associated with distances. Also,

within the first 10 m, the delay spread for Rx Inside

increases and then begins to decrease gradually. For Rx

ground, the fluctuation of delay spread is much greater.

Indeed, the surface waves play a more important role.

For sensors located outside of the train, the variation is

quite large.

To get a deeper insight into the simulation results, a

closer look at the second scenario is given below.

Generally, the sensors can take different positions in

different situations. For example, it can be interesting

to know the wind speed on the roof to adjust the speed

of the train when one wants perform turns. Therefore,

the second scenario involves placing a transmitting

antenna on the roof of the train (roof Tx) (Fig. 5). Here,

only the location of the transmitting antennas has

changed. The reception antennas have also been placed

in the same locations as in the first scenario. These

positions are in particular within the train (Rx inside),

on the left wall (Rx Route1), on the right wall (Rx

Route2) on the surface of the train (Rx ground) and

then outside the train (Rx outside). The obtained results

are presented in Fig. 15 and the average in Table 3.

By looking more closely at these results, it can be

deduced that the significant difference with the first

scenario lies in the outdoor antennas of the wagon.

The positioning of the receiving antennas in the

wagon gives interesting results. Indeed, along the side

walls (left or right), the delay spread is almost the same.

This result is confirmed by the change of position of

the transmitting antenna. The delay spread for the

ground antennas fall significantly from roughly 16 ns

Fig. 14 Delay spread vs. distance at Tx point.

Table 2 Mean delay spread TX point.

Tx Point Inside Route 1 Route 2 Ground Outside

Min (ns) 40.498 3.407 3.786 0.010 6.181

Max (ns) 17.205 15.420 16.714 29.172 36.760

Mean (ns) 8.998 8.359 8.527 7.799 16.720

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 220

4

8

12

16

20

24

28

32

36

40

Distance (m)

Dela

y sp

read (ns)

Delay spread vs. Distance Tx Point

Rx Inside

Rx Route1

Rx Route2Rx Ground

Rx Outside


1128

Fig. 15 Delay spread vs. distance at Tx roof.

Table 3 Mean delay spread TX roof.

Tx Point Inside Route 1 Route 2 Ground Outside

Min (ns) 2.963 0.036 0.583 0.01 2.311

Max (ns) 17.941 11.868 11.564 33.707 4.595

Mean (ns) 8.275 6.51 6.672 7.108 3.244

to 3 ns. Contrary to the first scenario where the delay

spread increased over the first 10 m Rx inside, it

remains fairly constant here and begins to increase

after the first 10 m. Similarly for side’s antennas, the

delay spread is experiencing growth over a few meters

and then begins to decrease. For ground antennas, the

variation is in the same direction as in the first scenario.

3.4 Coherence Bandwidth

Multipath fading has an important impact on

performance of wireless communications system.

Multipath fading channels are globally classified as flat

fading and frequency-selective fading according to their

coherence bandwidth relative to the bandwidth of

transmitted signal. Coherence bandwidth is defined as

the range of frequencies over which two frequency

components retains a strong amplitude correlation. It is

defined as the range of frequencies over which the

channel can be considered flat. The analytic issue of

coherence bandwidth was first studied by many authors;

they concluded that the coherence bandwidth of a

wireless channel is inversely proportional to its RMS

delay spread. It can be easily computed. The coherence

bandwidth is defined by Ref. [23]:

90%

50% (6)

where, τ denotes the rms delay spread.

These results may allow us to determine the

maximum bandwidth that can be used for

communications in a wagon.

Fig. 16 and Fig. 17 respectively present coherence

bandwidth over distance with Tx point and Tx roof

respectively.

By inspecting these results it is found that when the

transmitting antenna is located on the roof of the car,

Fig. 16 Coherence bandwidth vs. distance at Tx point.

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 230

4

8

12

16

20

24

28

32

36

40

Distance (m)

Dela

y sp

read (ns)

Delay spread vs. Distance Tx Toit

Rx InsideRx Route1Rx Route2Rx GroundRx Outside

0 1 2 3 4 5 6 7 8 9 10 1112 13 1415 16 1718 19 2010

15

20

25

30

35

Distance (m)

Coh

enre

nce

Ban

dwid

th (

MH

z)

Inside RxOutside RxRx Route1Rx Route2

Delay spread vs. TxRoof


1129

Fig. 17 Coherence bandwidth vs. distance at Tx roof.

the coherence bandwidth is relatively constant over the

first 15 m for the receiving antennas situated in the

middle, on the edges of the wagon. The average value

of the coherence bandwidth here is 18.47 MHz. For the

transmitter in the middle of the wagon, the results are

different. The coherence bandwidth varies according to

the distance and fluctuates greatly. With receptors

located in the middle of the wagon, the coherence

bandwidth varies from about 10 MHz to 24 MHz at 1

meter to 4 m. It then drops to 19 MHz to 6 m and rising

to 33 MHz at 9 m. This fluctuation is also observed for

edges antennas. The coherence bandwidth for

train-to-train communications is also determined over

a distance of about 3 m and is plotted in Fig 16 and Fig.

17. Therefore, the location of the sensors inside a

wagon is crucial and knowledge of the proper position

is an important factor not to be neglected in any system

design.

4. Conclusions

A train environment was simulated in order to find

the important propagation parameters and unique

results were found. These results could help to deploy

wireless technologies onboard train for various needs

and especially environmental physical parameters

monitoring.

The path loss exponent is smaller than that in the free

space when the receiving antennas are located at a

convenient distance from the ground. On the contrary,

bringing them the floor, the path loss becomes

significant and this will render it more difficult to

acquire data from sensors located on or under the floor.

The delay spread varies slightly depending on the

distance of the first meters. The delay spread is almost

the same for sensors located along the side walls. These

delays spread allowed us to determine the coherence

bandwidth. The coherence bandwidth which is an

important parameter of a propagation channel was

determined and we noticed that it varies greatly

depending on the position of the receiving antennas.

This would determine the ideal locations for the

receivers.

The positioning of sensors is crucial in data

acquisition and that depends on the needed information.

Acquiring the characteristics of the propagation channel

in a train allows us to find the optimal positions for the

sensors.

For future work, these results obtained will be

compared with physical measurements of the

propagation parameters using actual trains. The

measurement campaign to obtain experimental results

is currently underway. Moreover, the simulation case

of a locomotive, where interference plays a major role,

is another step in the study of signal propagation in a

train. This step is part of the work to be developed in

the future.

Acknowledgments

The authors are deeply indebted to M. Denis Roy of

RDG2, Bromont, Qc, Canada for his special

contribution in our understanding of the behavior and

importance of many sensors onboard trains.

References

[1] H. Sharif, M. Hempel, Study of RF Propagation Characteristics for Wireless Sensor Networks in Railroad Environments [Online], June 2010, A Report on Research, http://matc.unl.edu/assets/documents/finalreports/Sharif_PropagationCharacteristicsWirelessSensorNetworks.pdf.

[2] L. Taehyung, C. Hongsik, K. Seogwon, K. Kihwan, Measurement and analysis of consumption energy for Korean high-speed trains, in: 2012 IEEE PES Innovative Smart Grid Technologies (ISGT), Washington, Jan. 16-20,

0 2 4 6 8 10 12 14 16 18 200

20

40

60

80

100

Distance (m)

Coh

eren

ce b

andw

idth

(M

Hz)

Inside RxOutside RxRx Route1Rx Route2


1130

2012, pp. 1-5. [3] C. R. Garcia, A. Lehner, T. Strang, K. Frank, Channel

Model for Train to Train Communication Using the 400 MHz Band, in Vehicular Technology Conference, 2008. VTC Spring 2008, Sigapore, May 11-14, 2008, pp. 3082-3086.

[4] L. Chang-Myung, V.N. Goverdovskiy, A.N. Trofymov, V.V. Babenkov, High-speed train and environment: A system concept of multi-stage vibration isolation, in: 2010 International Forum on Strategic Technology (IFOST), Ulsan, Oct. 13-15, 2010, pp. 299-305.

[5] T. Ito, N. Kita, W. Yamada, T. Ming-Chien, Y. Sagawa, M. Ogasawara, Study of propagation model and fading characteristics for wireless relay system between long-haul train cars, in: Proceedings of the 5th European Conference on Antennas and Propagation (EUCAP), Rome, April 11-15, 2011, pp. 2047-2051.

[6] N. Kita, T. Ito, W. Yamada, T. Ming-Chien, Y. Sagawa, M. Ogasawara, Experimental study of path loss characteristics in high-speed train cars, in: Antennas and Propagation Society International Symposium 2009, APSURSI '09, Charleston, June 1-5, 2009, pp. 1-4.

[7] B. Nkakanou, G.Y. Delisle, N. Hakem, Y. Coulibaly, Acquisition of EM propagation parameters onboard trains at UHF frequencies, in: 2013 7th European Conference on Antennas and Propagation (EuCAP), Gothenburg, April 8-12, 2013, pp. 1493-1496.

[8] Q. Jiahui, T. Cheng, L. Liu, T. Zhenhui, Broadband Channel Measurement for the High-Speed Railway Based on WCDMA, in: 2012 IEEE 75th Vehicular Technology Conference (VTC Spring), Yokohama, May 6-9, 2012, pp. 1-5.

[9] S. Knorzer, M.A. Baldauf, T. Fugen, W. Wiesbeck, Channel Characterisation for an OFDM-MISO Train Communications System, in: Proceedings of 2006 6th International Conference on ITS Telecommunications, 2006, pp. 382-385.

[10] D. Weihui, L. Guangyi, Y. Li, D. Haiyu, Z. Jianhua, Channel Properties of indoor part for high-speed train based on wideband channel measurement, in: 2010 5th International ICST Conference on Communications and Networking in China (CHINACOM), Beijing, August 25-27, 2010, pp. 1-4.

[11] Ghazal, W. Cheng-Xiang, H. Haas, M. Beach, L. Xiaofeng, Y. Dongfeng, A Non-Stationary MIMO Channel Model for High-Speed Train Communication Systems, in: 2012 IEEE 75th Vehicular Technology

Conference (VTC Spring), Yokohama, May 6-9, 2012, pp. 1-5.

[12] S. Knorzer, M. A. Baldauf, T. Fugen, W. Wiesbeck, Channel Analysis for an OFDM-MISO Train Communications System Using Different Antennas, in: 2007 IEEE 66th Vehicular Technology Conference: VTC-2007 Fall, Baltimore, Sep. 30-Oct.3 2007, pp. 809-813.

[13] Y. Lihua, R. Guangliang, Y. Bingke, Q. Zhiliang, Fast Time-Varying Channel Estimation Technique for LTE Uplink in HST Environment, IEEE Transactions on Vehicular Technology 61 (2012) 4009-4019.

[14] M. L. Filograno, P. Corredera Guillen, A. Rodriguez-Barrios, S. Martin-Lopez, M. Rodriguez-Plaza, A. Andres-Alguacil, Real-time monitoring of railway traffic using fiber bragg grating sensors, Sensors Journal, IEEE 12 (2012) 85-92.

[15] T. Alade, H. Osman, M. Ndula, In-Building DAS for

High Data Rate Indoor Mobile Communication, in： 2012

IEEE 75th Vehicular Technology Conference (VTC Spring), Yokohama, May 6-9, 2012, pp. 1-5.

[16] S. Dhahbi, A. Abbas-turki, S. Hayat, A. El-Moudni, Study of the high-speed trains positioning system: European signaling system ERTMS / ETCS, in: 2011 4th International Conference on Logistics (LOGISTIQUA), Hammamet, May 31-June3, 2011, pp. 468-473.

[17] Y. Lihua, R. Guangliang, Y. Bingke, Q. Zhiliang, Fast time-varying channel estimation technique for LTE uplink in HST environment, IEEE Transactions on Vehicular Technology, 61 (2012) 4009-4019.

[18] R. Lo Forti, G. Bellaveglia, A. Colasante, E. Shabirow, M. Greenspan, Mobile Communications: High-Speed train Antennas from Ku to Ka, in: Proceedings of the 5th European Conference on Antennas and Propagation (EUCAP), Rome, April 11-15, 2011, pp. 2354-2357.

[19] Y.Q. Zhou, Z.G. Pan, J.L.Hu, L.L. Shi, X.W. Mo, Broadband wireless communications on high speed trains, in: 2011 20th Annual Wireless and Optical Communications Conference (WOCC), Newark, April 14-15, 2011, pp. 1-6.

[20] Remcom Home Page, http://www.remcom.com/xgtd. [21] http://www.mobility.siemens.com/mobility/global/en/inte

rurbanmobility/rail-solutions/high-speed-and-intercity-trains/velaro/Pages/velaro.aspx.

[22] http://www.3dcadbrowser.com/preview.aspx?modelcode=11784

[23] T.S. Rappaport, Wireless Communications Principles and Practice, Prentice-Hall, New Jersey, 2002.


A Novel Matlab-Based Underwater Acoustic Channel

Simulator

Zarnescu George

Department of Electrical Engineering and Telecommunications, Faculty of Electromechanics, Maritime University of Constanta,

Constanta 900663, Romania


Abstract: An accurate modeling of the UAC (underwater acoustic channel) can facilitate the development of an efficient architecture for an UAM (underwater acoustic modem). The performance comparison of different architectures can be performed rapidly and at a low cost in a simulation environment, compared to testing the modems in sea water. This article presents the development and utilization of an underwater acoustic channel simulator. The simulator can be used by a communications engineer in characterizing the time variability of the physical channel’s parameters or by a hardware engineer in designing an underwater acoustic modem. This tool is programmed in Matlab and is based on the algorithms Bounce and Bellhop. The input parameters of these algorithms must be saved in text files after a specific template and are cumbersome to process manually. To streamline the modeling of an UAC and the simulation of various communication algorithms the simulator automatically creates the input files based on key parameters entered by the user, hiding the algorithmic dependent ones and allows a quick visualization of the simulation results with a few routines specially created. The use of this simulator is emphasized with results obtained from the design of a low-power UAM for long-term monitoring activities.

Key words: Underwater acoustic channel simulator, underwater acoustic channel, underwater acoustic modem.

1. Introduction

When a hardware engineer wants to design a new

architecture for an underwater communications

system he must take into account the variability of the

UAC in the location where the modem is intended to

be placed. Testing the new system in a real

environment is performed at a high cost because a lot

of sophisticated equipment is needed. On the other

hand, the observation period is usually quite short

while the parameters of the underwater acoustic

environment might be constant. Thus the behavior of

the new system must be observed when different

parameters of the environment are changing. In

conclusion the testing must be done in different

periods of the year and this will raise the total cost of

Corresponding author: Zarnescu George, license, teaching

assistant, research fields: underwater acoustics, digital logic circuits. E-mail: [email protected].

the designing process.

To overcome these shortcomings an UAS must be

used to design and test the new architecture. An

accurate modeling of the variation of the parameters

of the underwater acoustic communication channel

can facilitate the development of efficient system

architecture. In Ref. [1] and Ref. [2] it was shown that

the simulated impulse responses obtained after

modeling the parameters of the underwater

environment with real measurements were very close

to those obtained from the ocean or seawater. The

simulated results were obtained with the algorithms

Bounce and Bellhop [3, 4]. It must be emphasized that

these routines represent the core of the simulation tool

described in this article. The input parameters of the

algorithms must be saved in text files after a specific

template and are cumbersome to process manually. To

streamline the modeling of the UAC and the

A Novel Matlab-Based Underwater Acoustic Channel Simulator

1132

simulation of various communication algorithms the

simulator, which is programmed in Matlab,

automatically creates the simulation files based on key

parameters entered by the user and hide the

algorithmic dependent ones. The values of the

dependent features were chosen so that high quality

simulation results can be obtained. The key input

parameters like the acoustical and geophysical

parameters, the location of the emitter and receiver or

the transmission frequency can be easily introduced in

an xlsx file by the user. Afterwards the simulator

processes this file, runs the algorithms and allows the

visualization of the results.

This tool came from the need to simulate the

operation of an underwater communications system at

the physical level in a synthetic environment that can

imitate the real one. The simulator was designed to

enable the rapid configuration of the underwater

environment and the modem’s parameters and this

was possible using Excel and Matlab.

In Ref. [1] the authors propose a simple underwater

acoustic channel simulator which can predict the

quality of a transmission for future field trials. The

channel estimates are obtained in AcTUP (Acoustic

Toolbox User interface and Post processor) [2, 3],

which is a guide user interface written by Amos

Maggi and Alec Duncan and can facilitates the

application of different acoustic propagation codes.

The signal processing scheme attempts to characterize

the operation of an existing commercial modem and is

presented as a block diagram. The authors emphasize

the processing results and provide a brief description

about the simulator’s modules. There are several other

free underwater simulators with which the operation

of a system at the networking layer can be

characterized [4-6].

The article is organized in the following manner.

Section 2 presents the organization of the simulator

emphasizing how the input data are introduced and

how they are processed. Section 3 highlights the

results obtained from the design of a low-power UAM

for long-term monitoring activities. Section 4 presents

the conclusions of this article and how the simulator

can be improved.

2. The Organization of the Underwater Acoustic Channel Simulator

Fig. 1 highlights the modular organization of the

underwater acoustic channel simulator. The simulator

is based on two routines, written in Fortran by

Michael Porter, called Bounce and Bellhop [7] and are

often used to simulation the propagation of

high-frequency sound underwater because they

produce very accurate results [8-10]. These routines

accept text files with the input data.

A text input file, when is created, requires detailed

knowledge of its structure and each parameter. An

input file must be created for each particular scenario

and for each transmission frequency making the file

management process to become cumbersome. These

inconveniences were eliminated by automating the

process of creating the input files. Therefore, the

important input data are entered into an xlsx file and

the specific parameters of the algorithms were hidden

from the user, but carefully chosen in order to obtain

simulation results at a good resolution without hinder

the running time of the algorithms. Two routines have

been developed to characterize the bathymetric profile

and the organization of the sedimentary layer and also

for the sea surface there are two routines to

characterize the surface profile and the reflection loss

at the surface.

Bounce and Bellhop algorithms will process the

input files and will create amplitude-delay profiles

that can be post processed with the communications

module or can be plotted using the plotting module.

Next the detailed structure and functionality of each

module from Fig. 1 is presented.

2.1 The Scientific Databases

The characterization of a certain area in terms of the

propagation of sound underwater requires the knowledge


1133

Fig. 1 The organization of the underwater acoustic channel simulator.

of the acoustical and geophysical parameters. They can

be found in the databases GEBCO [11], NGDC [12]

and NOAA [13]. If the data are complete and have the

necessary resolution, automating the process of

obtaining them is natural [14].

The data on the databases highlighted above were

recorded during scientific expeditions and were useful

in various research projects. Depending on the

purpose of the project there have been recorded only

certain parameters. In some cases the recorded data

are incomplete or do not have the resolution to

characterize in detail the transmission of sound

underwater. In other cases the data are nonexistent. In

conclusion automating the process of obtaining the

scientific data from the above databases is very

difficult to perform and is therefore better that the data

is processed by the user with a few routines.

The steps that define the data collection process are

as follows. The first step is to obtain, for a certain area,

the csv files in which the user can find information

about the parameters. The second step is to import and

process the csv files in Matlab using the routines from

the Scientific Data Preprocessing Module.

Following the above steps the user obtains the

location on the globe, the year, the month, the day and

the time of the day the data were recorded and

information about temperature, salinity and depth. In

case the information about the wind speed and

sedimentary composition do not exist, the user can

choose general values or a thorough documentation in

the scientific literature is required.

2.2 The CTD Processing Routine

The salinity, temperature and depth data, also called

CTD data, are processed with the program named

underwater_sound_speed.m. This routine provides the

sound velocity profile and is based on the underwater

sound speed equation [15], which is highlighted in

Relation 1

3422 109.2105.56.42.1449 TTTc

zST 22 106.1)35)(1034.1( (1)


1134

Relation 1 is an empirical equation that is valid for

CT 350 , 450 S and 000,10 z

m.

The sound velocity profile can be further processed

to obtain velocity averages depending on the season,

the mouth or the day. Afterwards the underwater

sound speed and the data about the wind speed and the

composition of the sedimentary layer can be

introduced by the user in an xlsx file. This file will be

used to automatically create the input text files for the

Bounce-Bellhop algorithms.

2.3 The Xlsx Input File

To allow the user a quick and easy configuration

and modification of the acoustical and geophysical

parameters, these input data will be entered in an xlsx

file. The structure of this file is shown in Fig. 2. The

user can enter in the first sheet a sound profile or a

collection of sound profiles. In the second sheet the

user can introduce the geophysical properties of each

sedimentary layer. It must be emphasized that the

organization of the first two sheets is approximately

identical to the structure of the main input text files

for the Bounce and Bellhop algorithms. Therefore, the

user will enter in the xlsx file only the important

parameters. The algorithm specific parameters were

masked from the user. It must be pointed out that in

case there were entered several sound speed profiles,

they will be processed individually and not as a 2D

surface, where consecutive profiles are considered for

particular locations on the transmission distance

between the transmitter and receiver.

In the third sheet the user can configure the way the

transmission frequencies will be interpreted. If the

user wants to simulate the transmission of the

underwater sound at a single frequency or for a range

of frequencies, defined as maxmin :: fff , can enter

the value 1.

This mode is used when there is a need to know the

Fig. 2 The organization of the xlsx input file.

optimal transmission frequency. If the user wants to

simulate the transmission for a few random

frequencies can enter the value 0. This mode is used

when there is the need to characterize the transmission

for each season or month of the year. For this mode

the number of frequencies must be equal to the

number of sound speed profiles. If the underwater

sound speed is measured at intervals of a few minutes

or hours during several days, the user can enter an


1135

array with the exact moments when the data were

recorded. This mode is used when there is a need to

characterize the transmission during the day and night.

Therefore the must define two transmission

frequencies. It is known that the mean sound speed

profile is different for the two periods of the day

which means that there will be two optimal

transmission frequencies.

In the fourth sheet the wind speed can be

introduced as an average value or as an array of values

for certain moments of time. In the fifth sheet the user

can configured the transmission distance, the depth of

the transmitter and receiver and the values of the

transmission frequencies according to the parameter

set in third sheet.

2.4 The Input Text Files

At this moment the program run_bellhop.m is run

in Matlab, having as input parameter the xlsx file.

This routine imports and processes the data from the

xlsx file and calls the routines that are intended to

create the input text files for the Bounce and Bellhop

algorithms. The routine create_bellhop_file.m

constructs the main input text file for the Bellhop

algorithm.

The program create_ati.m produces a file in which

the surface profile is defined as a sinusoid. The

routine has two input parameters, the wind speed and

the maximum depth, and computes the root mean

square height and the wavelength of the sea surface

with the following relations

g

vhrms

214784.0 (2)

g

vgdv 877.0

2 (3)

In the above relations v is the wind speed in m/s

measured at an altitude of 19.5 m, g is the

gravitational acceleration in 2sm , d is the

maximum depth of the water column in m, rmsh is

the root mean square height and v is the

wavelength of the sea waves both in m. The above

equations are based on the Pierson and Moskowitz

spectrum [16].

The program create_trc.m creates a text file in

which the reflection coefficient at the surface is

defined. This routine has three input parameters: the

transmission frequency, f, measured in Hz, the wind

speed, v, in m/s and the grazing angle, θ, measured in

degrees and is based on the relation 4, which is

describe in detail in Ref. [17] 2422106.8 vfRL (4)

The routine create_brc.m and the Bounce algorithm

create a text file in which the reflection coefficient for

the sedimentary layer is defined. The routine

create_bty.m creates a text file with a flat bathymetric

profile.

At this moment the main text file and the auxiliary

files described above are processed by Bellhop which

will generate output files with the amplitude-delay

profiles for each transmission frequency and for each

sound profile, for the defined transmission distance

and transmission-receiver depth.

2.5 The Post Processing Module

The user can use the routines of this module to

process the amplitude-delay profiles and to obtain

detailed information about the optimal transmission

frequency, the attenuation in the underwater

communication channel and the noise level.

2.6 The Plotting Module

The routines of this module can be used to plot the

amplitude-delay profiles, the frequency response or

the interactions of the sound waves with the sea

surface and the seafloor.

2.7 The Communications Module

With the use of the routines of this module the user

can simulate the transmission of digital signals in the

underwater acoustic communication channel. The

amplitude of the transmitted signals is computed using

a specific amplification value and the transmitting


1136

voltage response of a particular transducer. Then these

signals are BFSK, MFSK, BPSK, QPSK modulated

and are convolved with the amplitude-delay profiles.

Afterwards over the convolved signals is added white

Gaussian noise with a standard deviation that depends

on the wind speed. The resulted signals are passband

processed and the signal-to-noise ratio is computed.

3. The Simulation Results

This section presents the way in which the

underwater acoustic channel simulator was used in

designing a low-power underwater acoustic modem

for long-term monitoring activities in the

north-western part of the Black Sea.

The acoustical data were imported from NOAA

database for the region of interest and were processed

using the routines from the first module from Fig. 1.

The CTD data were processed using the routines from

the second module and the sound speed profiles were

obtained. Afterwards the profiles were averaged

depending on the season and the time of day and are

shown in Fig. 3.

The diurnal sound speed profiles are represented

with blue and the nocturnal sound speed profiles are

shown in red. The geophysical data were

characterized using the information from [18]. The

wind speed was chosen 10 m/s, the transmission

distance was chosen 500 m, the transmitter and

receiver were placed at 0.5 m above the sea floor and

the simulations were done in the range 1-99.9 kHz.

An xlsx file was created using the above data then

the routine run_bellhop.m was run in Matlab to obtain

the simulation results. The routines from the fourth

module were used to process the simulation results.

The programs from the sixth module can be used to

plot the results of the simulations. Thus Fig. 4 shows a

sample impulse response and in Fig. 5 a sample

transmission loss profile (frequency response) could

be seen. Posts processing the simulation results the

optimal transmission frequencies were obtained and

are shown in Fig. 6.

Afterwards using the optimal transmission

frequencies from Fig. 6, a series of data collected in

the region of interest at intervals of three hours, during

several days, for a few months were processed and a

series of amplitude-delay profiles were obtained. A

Fig. 3 Sound speed profiles organized by season and time of day.

Fig. 4 A sample amplitude-delay profile.


1137

Fig. 5 A sample transmission loss profile.

Fig. 6 Optimal transmission frequency organized by season and time of day. Season: Winter (1), Spring (2), Summer (3) and Autumn (4).

sample of these series of profiles is displayed in Fig. 7

using the routines from the sixth module. These series

were post processed with the routines from the fifth

move and a sample bit error rate analysis is shown in

Fig. 8.

4. Conclusions and Future work

This paper presents the organization of a novel

Matlab-based underwater acoustic channel simulator

and the way it was used in designing a low-power

underwater acoustic modem for long-term monitoring

activities in the north-western part of the Black Sea. It

was attempted, by creating this simulator, to fix the

shortcomings of the current simulators and to allow

the user a quick and easy configuration and

modification of the synthetic underwater environment

and the modem’s parameters.

The simulator is based on two routines named

Bounce and Bellhop. This tool allows the user to

introduce only the important parameters and hides

Fig. 7 A series amplitude-delay profile sample.

Fig. 8 Bit error rate analysis for the amplitude-delay profiles from Fig. 7 for BPSK and QPSK modulation schemes.

those that are algorithmic dependent. Using CTD data

from the world’s databases, wind speed values and

geophysical information the user can thoroughly

synthesize an underwater environment similar to the

real one. The CTD data are used to characterize the

variability of the sound speed profile. The wind speed

values are used to create a surface profile and to

compute the reflection loss at the surface and the noise

level. The geophysical information is used to create a

bathymetric profile and to compute the bottom

reflection loss.

The user could characterize the underwater acoustic

communication channel in terms of the

amplitude-delay profile, the frequency response and

the optimal transmission frequency. This simulator

has a communications module which other simulators

do not have, with which the user can compute the bit

error rate for BPSK, QPSK, BFSK, MFSK digital


1138

modulation techniques. The plotting module allows a

rapid visualization of the simulation results.

Currently the simulator implements only a flat

seafloor, but in the future the user will be able to

select from multiple choices. In addition the user will

be able to choose the surface reflection loss scheme

and to use the OFDM technique of transmitting digital

signals, which will be implemented in the

communications module.

References

[1] G. Pusey, A. Duncan, Development of a simple underwater acoustic channel simulator for analysis and prediction of horizontal data telemetry, in: Proceedings of ACOUSTICS 2009, Australia, Nov. 23-25, 2009.

[2] A. J. Duncan, A. L. Maggi, A consistent, user friendly interface for running a variety of underwater acoustic propagation codes, in: Proceedings of ACOUSTICS 2006, New Zealand, 2006, pp. 471-477.

[3] A.J. Duncan, A. L. Maggi, Underwater acoustic propagation modeling software–AcTUP v2.2l [Online], http://cmst.curtin.edu.au/products/actoolbox.cfm.

[4] F. Guerra, P. Casari, M. Zorzi, World ocean simulation system (woss): A simulation tool for underwater networks with realistic propagation modeling, in: Proceedings of the Fourth ACM International Workshop on UnderWater Networks, ser. WUWNet’09, Berkeley, California, USA, Nov. 3, 2009, pp. 1–8.

[5] A.F. Harris III , Michele Zorzi, Modeling the underwater acoustic channel in ns2, in: NSTools'07, Nantes, France, Oct. 22, 2007.

[6] I.A. Sehgal, J. Schonwalder, Aquatools: An underwater acoustic networking simulation toolkit, in: IEEE, Oceans, Sydney, Australia, May 24-27, 2010.

[7] M. B. Porter, The Bellhop manual and user’s guide

[Online], http://oalib.hlsresearch.com. [8] M. Badiey, S. Forsythe, M. Porter, Ocean variability

effects on high-frequency acoustic propagation in the Kauai Experiment, in: High Frequency Ocean Acoustics, New York, 2004, pp. 322-335.

[9] M. Badiey, A. Song, D. Rouseff, H-C. Song, W. S. Hodgkiss, M. B. Porter, High-frequency acoustic propagation in the presence of ocean variability in KauaiEx, in: Proceedings of OCEANS’07, Aberdeen England, Jun. 18-21, 2007.

[10] A. Song, M. Badiey, H.C. Song, W. S. Hodgkiss, M. B. Porter, and the KauaiEx Group, Impact of ocean variability on coherent underwater acoustic communications during the Kauai experiment (KauaiEx), JASA 123 (2008) 856–865.

[11] General Bathymetric Chart of the Oceans [Online], http://www.gebco.net.

[12] National geophysical data center, seafloor surficial sediment descriptions [Online], http://www.ngdc.noaa.gov.

[13] National Oceanic and Atmospheric Administration, NOOA, http://www.noaa.gov.

[14] J. Llor, M. Stojanovic, M. P. Malumbres, An integrated simulation framework for underwater acoustic networks, 2010.

[15] L.M. Brekhovskikh, Fundamentals of ocean acoustics, 3rd ed., Springer, New York, 2003.

[16] W.J. Pierson, L. Moskowitz, A proposed spectral form for fully developed wind seas based on the similarity theory of S. A. Kitaigorodskii, J. Geophys. Res. 69 (1964) 5181-5190.

[17] A.D. Jones, J.Sendt, A.J. Duncan, P.A. Clarke, A.Maggi, Modelling the acoustic reflection loss at the rough ocean surface, in: Proceedings of ACOUSTICS 2009, Adelaide, Australia, Nov. 23-25, 2009.

[18] G. Oaie, D. Secrieru, Black Sea basin: Sediment types and distribution, sedimentation processes, in: Proceedings of Euro-EcoGeoCentre, Ireland, May 10-13, 2004.


Normalized Efficient Routing Protocol for WSN

Rushdi Hamamreh and Mahmoud I Arda

Computer Engineering Department, Al-Quds University, Jerusalem, Palestine


Abstract: WSNs (wireless sensor networks) consist of thousands of tiny nodes having the capability of sensing, computation, and wireless communications. Unfortunately these devices are limited energy devices, that is means we must save energy as much as possible, to increase network life time as long as possible. In this paper we introduce NEER—normalized energy efficient routing protocol that increases network life time through switching between AODV protocol that depends on request-reply routing, and MRPC that depends on residual battery in routing. Key words: WSN, energy-aware routing, routing protocols, meta-data, negotiation, network lifetime, energy threshold.

1. Introduction

A WSN (wireless sensor network) [1, 2] in its

simplest form could be defined as a network of

(possibly low-size and low-complex) devices denoted

as nodes that can sense the environment and

communicate the information gathered from the

monitored field through wireless links; the data is

forwarded, possibly via multiple hops relaying, to a

sink that can use it locally, or is connected to other

networks (e.g., the Internet) through a gateway [1].

Each node has five components shown in Fig. 1:

(1) Communication unit.

(2) Controller unit.

(3) Actuator unit

(4) Memory unit.

(5) Power supply.

The node senses the data from the environment, then,

processes it and sends it to the base station.

These nodes can either route the data to the BS (base

station) or to other sensor nodes such that the data

eventually reaches the base station as shown in Fig. 2.

In most applications, sensor nodes suffer from limited

energy supply and communication bandwidth. These

Corresponding author: Rushdi Hamamreh, Ph.D., lecturer,

research fields: routing protocols, networks security and distributed systems. E-mail: [email protected].

nodes are powered by irreplaceable batteries and hence

network lifetime depends on the battery consumption.

Innovative techniques are developed to efficiently use

the limited energy and bandwidth resource to

maximize the lifetime of the network. These

techniques work by careful design and management at

all layers of the networking protocol. For example, at

the network layer, it is highly desirable to find methods

for energy efficient route discovery and relaying of

data from the sensor nodes to the base station.

The route of each message destined to the base

station is really crucial in terms network lifetime. On

the other hand there are many factors that affect the

network life time such as topology of the network, the

transmission rate, transmission range and routing

protocol.

The simplest forwarding rule is to flood [3] the

network: send an incoming packet to all neighbors. As

long as source and destination node are in the same

connected component of the network, the packet is sure

to arrive at the destination. To avoid packets circulating

endlessly, a node should only forward packets which

have not yet been seen (necessitating, for example,

unique source identifier and sequence numbers in the

packet). Also, packets usually carry some form of

expiration date (time to live and maximum number


1140

Fig. 1 Sensor node.

Fig. 2 WSN structure.

of hops) to avoid needless propagation of the packet

(e.g., if the destination node is not reachable at all).

While these forwarding rules are simple, their

performance in terms of number of sent packets or

delay. Determining these routing tables is the task of

the routing algorithm with the help of the routing

protocol. In wired networks, these protocols are usually

based on link state or distance vector algorithms

(Dijkstra’s or Bellman-Ford [4, 5]). In a wireless,

possibly mobile, multi hop network, different

approaches are required. Routing protocols here should

be distributed, have low overhead, be self-configuring,

and be able to cope with frequently changing network

topologies. This question of ad hoc routing has

received a considerable amount of attention in the

research literature and a large number of ad hoc routing

protocols have been developed.

A commonly used taxonomy [6] classifies these

protocols as either (1) table-driven or proactive

protocols, which are “conservative” protocols in that

they do try to keep accurate information in their routing

tables, or (2) on-demand protocols, which do not

attempt to maintain routing tables at all times but only

construct them when a packet is to be sent to a

destination for which no routing information is

available. In addition to energy efficiency, resiliency

also can be an important consideration for WSNs. For

example, when nodes rely on energy scavenging for

their operation, they might have to power off at

unforeseeable points in time until enough energy has


1141

been harvested again. Consequently, it may be

desirable to use not only a single path between a sender

and receiver but to at least explore multiple paths. Such

multiple paths provide not only redundancy in the path

selection but can also be used for load balancing, for

example, to evenly spread the energy consumption

required for forwarding.

2. Motivation

In WSN, the route of each message destined to the

base station is really crucial in terms network lifetime.

If we always select the shortest route towards the base

station, that will causes the intermediate nodes deplete

faster and decreased network lifetime. We need to

increase WSN life time as long as possible.

3. Related work

In this section, we would study some routing

protocols that are important to get an overview about

routing in wireless sensor network. We would get a

view if ad-hoc on demand distance vector routing

AODV, SPIN, power aware routing protocols and

MRPC.

3.1 AODV

AODV [7] is the widely used algorithm for both

wired and wireless network. Ad-hoc On-demand

Distance Vector is known as one of the most efficient

routing protocols in terms of using the shortest path

and lowest power consumption. AODV is a reactive

protocol that builds routes between nodes on-demand

i.e. only as needed. Messages to other nodes in the

network do not depend on network-wide periodic

advertisements of identification messages to other

nodes in the network.

It broadcasts “HELLO” messages to the neighboring

nodes. It then uses these neighbors in routing.

Whenever any node (Source) wants to send a message

to another node (Destination) that is not its neighbor,

the source node initiates a Path Discovery in which the

source would send a RREQ (route request) message to

its neighbors. Nodes that received the Route Request

could update their information about the sending node.

The RREQ should contain the IP address of source

node. On the other hand, the RREQ contains broadcast

ID that necessary to identify that RREQ. The RREQ

has to have a current sequence number that determines

the freshness of the message. Finally, the RREQ should

keep track of the number of nodes that visited through

path discovery in a variable of Hop Count. When a

node receives a RREQ, it would check whether it has

received the same RREQ earlier (using IP, ID, and

Sequence number), if so, it would discard it. On the

other hand, if the recipient of the RREQ was an

intermediate node that does not have any information

about the path to the final destination, the node

increases the hop count and rebroadcasts the RREQ to

its neighbors. If the node that received the RREQ was

the final destination or an intermediate node that knows

the path to the final destination, it sends back the Route

Reply (RREP). This RREP should keep track of

traverse path of the RREQ but from destination to

source. As shown in Fig. 3, when the source node

receives the RREP, it should then start sending data.

We should take into consideration another control

message; that is it, RERR is used if a node detect that

there is a link break on the next hop of an active route,

or if it gets a data packet destined to a node which does

not have an active route without repairing. Finally if a

node receives a RERR from a neighbor for one or more

active routes it sends a RERR message.

Fig. 3 (a) Timing diagram; (b) Hello packet.


1142

3.2 SPIN Protocols

Heinzelman et al. [8] has proposed a family of

adaptive protocols called SPIN (sensor protocols for

information via negotiation) that passes all the

information at each node to every node in the network

assuming that all nodes in the network to be a potential

BS (base-stations). In this algorithm the user has the

ability to query any node and get the required

information or data immediately. These algorithms

make assumes that nodes in close proximity have

similar data, and hence there is a need to only distribute

the data that other nodes do not posses. The SPIN

family of protocols uses data negotiation and

resource-adaptive algorithms. Nodes running SPIN

assign a high-level name to completely describe their

collected data (called meta-data or meta content). Meta

data in its simplest definitions is describes as data of

data. That is it, meta data should provide data about

one or more aspects of the original data, for example

meta data aspects may be the mean of creation of that

data, purpose of the data, time and date of creation,

Creator or author of data, and location on a computer

network where the data was created) and perform

meta-data negotiations before any data (we means here

original data) is transmitted. Its importance arises from

the fact that we have used to make sure that there is no

redundant data sent throughout the network. That is it

to reduce the overhead on the network and to save

power. The semantics of the meta-data format is

application-specific and is not specified in SPIN. For

example, when sensors want to send meta-data for an

event in certain area, it would use its ID. On the other

hand, SPIN algorithm has the ability to access to the

energy level of the node and monitor the protocol

running according to how much energy it remaining in

a certain node. These protocols are known as a

time-driven fashion and broadcast the information all

over the wireless sensor network, despite the fact that

the user does not request any data at that moment.

SPIN’s meta-data negotiation approach solved the

traditional problems of flooding, and thus achieving a

lot of energy efficiency because you send meta-data,

not all data as used to in flooding. In SPIN, there are

three stages in which sensor nodes use three different

types of messages ADV (advertise) REQ (request) and

DATA to communicate with other nodes. ADV is used

to advertise new data, REQ to request data by the node

or sink or user itself and DATA is the actual message

itself. The protocol starts when a node gets new data

that it is willing to share with other nodes, after that it

broadcasts an ADV message containing meta-data. If

any nodes that receive ADV were interested in that data,

it sends a REQ message for the DATA and the DATA

is sent to this neighbor node. The neighbor sensor node

then repeats this process with its neighbors. As a result,

the entire sensor area will receive a copy of the data.

3.3 Power Aware Routing

Several algorithms had been developed for routing

in wireless sensor network, some of these algorithms

and protocols are energy based algorithms. In these

algorithms we take the network graph, assign to each

link a cost value that reflects the energy consumption

across this link, and pick any algorithm that computes

least-cost paths in a graph. An early paper along these

lines is Ref. [9], which modified Dijkstra’s shortest

path algorithm to obtain routes with minimal total

transmission power.

One of the most important algorithms used is known

as minimum energy per packet or per bit. The most

straightforward formulation is to look at the total

energy required to transport a packet over a multi hop

path from source to destination (including all

overheads). The goal is then to minimize, for each

packet, this total amount of energy by selecting a good

route. Minimizing the hop count will typically do not

achieve this goal as routes with few hops might include

hops with large transmission power to cover large

distances—but be aware of distance-independent,

constant offsets in the energy-consumption model.

Nonetheless, this cost metric can be easily included in

standard routing algorithms. It can lead to widely

differing energy consumption on different nodes [10].


1143

Some researches went to routing considering

available battery energy, as the finite energy supply in

nodes’ batteries is the limiting factor to network

lifetime, it stands to reason to use information about

battery status in routing decisions. Some of the

possibilities are maximum total available battery

capacity choose that route where the sum of the

available battery capacity is maximized, without taking

needless detours (called, slightly incorrectly,

“maximum available power” in Ref. [11]). Minimum

battery cost routing instead of looking directly at the

sum of available battery capacities along a given path;

MBCR instead looks at the “reluctance” of a node to

route traffic [10, 12]. This reluctance increases as its

battery is drained; for example, reluctance or routing

cost can be measured as the reciprocal of the battery

capacity. Then, the cost of a path is the sum of this

reciprocals and the rule is to pick that path with the

smallest cost. Since the reciprocal function assigns

high costs to nodes with low battery capacity, this will

automatically shift traffic away from routes with nodes

about to run out of energy. MMBCR (min-max battery

cost routing), this scheme [10, 12] follows a similar

intention, to protect nodes with low energy battery

resources. Instead of using the sum of reciprocal

battery levels, simply the largest reciprocal level of all

nodes along a path is used as the cost for this path.

Then, again the path with the smallest cost is used. In

this sense, the optimal path is chosen by minimizing

over a maximum. The same effect is achieved by using

the smallest battery level along a path and then

maximizing over these path values [11]. This is then a

maximum/minimum formulation of the problem.

Minimize variance in power levels to ensure a long

network lifetime, one strategy is to use up all the

batteries uniformly to avoid some nodes prematurely

running out of energy and disrupting the network.

Hence, routes should be chosen such that the variance

in battery levels between different routes is reduced.

MTPR (minimum total transmission power routing)

without actually considering routing as such, Bambos

[13] looked at the situation of several nodes

transmitting directly to their destination, mutually

causing interference with each other. A given

transmission is successful if its SINR exceeds a given

threshold. The goal is to find an assignment of

transmission power values for each transmitter (given

the channel attenuation metric) such that all

transmissions are successful and that the sum of all

power values is minimized. MTPR is of course also

applicable to multi hop networks.

3.4 MRPC

Misra and Banerjee [14] used to maximize network

lifetime for reliable routing in wireless environments

(MRPC), they depended on the fact that selecting the

path with the least transmission energy for reliable

communication may not always maximize the lifetime

of the ad-hoc network. On the other hand since the

actual drain on a node’s battery power will depend on

the number of packets forwarded by that node, it is

difficult to predict the optimal routing path unless the

total size of the packet stream is known during

path-setup. MRPC works on selecting a path, given the

current battery power levels at the constituent nodes,

that maximizes the total number of packets that may be

ideally transmitted over that path, assuming that all

other flows sharing that path do not transmit any

further traffic.

4. NEER Algorithm

MRPC algorithm has a problem in which it uses a

path that consumes much power. Simulation results [14]

showed that the transmission power per packet was

higher than that of minimum energy algorithm. Fig. 4

below shows that MRPC algorithm would take path P1

(A-C-F-H) because it would send 3 packets from to

while it would send only 2 packets through P2

(A-B-E-H) despite the fact that sending a packet

through P1 (6 units) consumes much more power than

P2 (only 3 units). We proposed a new algorithm called

NEER (normalized energy efficient routing).


1144

Fig. 4 Graph G and its components.

Our algorithm could be summarized as following:

Let G represent sensor network graph;

u, v represents nodes;

Edge (u, v) is the link between u and v;

ce (u) :residual battery of node u;

w(u, v) is the weighted cost of edge(u, v);

c(u, v) is the total number of packets that could be

sent from u to v. this value is defined as ce(u)/w(u, v).

Step 1: Initialize

Eliminate from G every edge (u, v) for which ce(u) <

w(u, v) this condition is used to ensure we could send at

least one packet through this path.

For every remaining edge (u, v) let

, / , (1)

Let L be the list of distinct c(u, v) values.

Step 2: Binary Search

Do a binary search in L to find the maximum value

max for which there is a path P from source to

destination that uses no edge with:

, (2)

For this, when testing a value q from L, we perform a

depth- or breadth-first search beginning at the source.

The search is not permitted to use edges with

, (3)

Let P be the source-to-destination path with lifetime

max. Simultaneously we should find minimum energy

path using Dijkstra’s algorithm as following:

∑ w u, v w u, v P (4)

Step 3: Wrap Up

If no path is found in Step 2, the route is not possible.

Otherwise, use P for the route.

Also find min(x), x P.

We need to derive a hybrid algorithm that takes the

advantages of both. Here we use the following steps:

We have to add a new condition that represents

threshold value Z in which we check battery level at all

nodes, if one of them was less than threshold value,

then NEER algorithm would switch to run MRPC.

Unless, NEER would continue to run AODV protocol.

In this case we took two factors in consideration.

The total power consumed through that path and the

residual battery in all nodes of that path. But we should

note that we use weight in our new algorithm. The

higher weight is for minimum energy factor. In such

case we guarantee that we use minimum energy

algorithm as long as possible but not to power off these

nodes.

In Fig. 5, we could summarize NEER functionality,

first of all, AODV would work until the first node

becomes about one fifth of its initial energy. In this

case, NEER would switch route algorithm. In this case,

MRPC algorithm would work. NEER would test if

there is any node

Fig. 5 NEER flow chart.


1145

We should also note that if we used threshold value

of the first condition to be (0) then NEER would

behave as AODV protocol. While if we choose

threshold value to be (1) then NEER protocol would be

full MRPC.

5. Analysis

There are some parameters used in simulation. Table

1 below summarize that parameters used in simulation.

We used NS-3.16 installed on Ubuntu as simulator,

after that we used wire shark network analysis to

analyze data gotten from NS-3. Here we would study

the behavior of NEER algorithm in comparison with

MRPC and Min-Energy Protocol in three fields. First

factor is the total number of dead nodes according to

time. In this factor nodes are dying slowly at the

beginning of running for NEER algorithm and would

die suddenly at the end of execution, since it takes the

features of both of (MRPC and Min-Energy). The

behavior of NEER algorithm is shown in Fig. 6.

If we want to compare between these algorithms

according to total sent packets by the network nodes,

our algorithm would send packets more than

Min-Energy and less than MRPC as shown in Fig. 7.

The horizontal axis represents time in seconds, while

the vertical one represents number of dead nodes.

Finally we would study energy per packet; here we

also expect that energy per packet would be also between

MRPC and Min-Energy, more than Min-Energy, less

than MRPC. Fig. 8 explains the idea.

6. Conclusion

In this paper we aim to introduce a new hybrid

algorithm that could increase network life time as long

as possible using minimum energy and residual battery

concepts. In this paper our algorithm depends mainly

on power consumption algorithms that are used in

WSN. This algorithm is based in the fact that most

power is consumed during data transmission not during

computations. This algorithm in fact takes advantages

of two important protocols. It takes the advantage of

consuming least power through minimum energy

protocol. On the other hand the presence of MRPC

protocol would increase network life time as long as

possible. In this way, NEER algorithm would use

minimum energy protocol as long as the residual power

is over a known threshold.

Table 1 Simulation parameters.

Parameter Description

Channel type Wireless channel

Mac protocol Mac/802_11

Number of nodes 40

Routing protocol Proposed Algorithm, MRPC, ME

Grid size 800 × 800

Packet size 64

Simulation time To die

Topology Random , Flat

Initial energy 3 joules

Source node 1

Destination node 1

Fig. 6 Expiration sequence.

Fig. 7 Number of sent packets.

0

10

20

30

40

50

20 40 60 80 100 120 140 160

Time

Expiration time (number of dead nodes)

MRPC Min‐ Energy NEER

0

5000

10000

15000

20000

MRPC Min ‐ Energy NEER

Number of sent packets


1146

Fig. 8 Energy per packet.

References

[1] G. Rodriguez-Navas, M.A. Ribot, B. Alorda, Understanding the role of transmission power in component-based architectures for adaptive WSN, in: Proceedings of the 2012 IEEE 36th Annual Computer Software and Applications Conference Workshops, Izmir, Jul. 16-20, 2012.

[2] F.C. Jiang, D.C. Huang, C.T. Yang, F.Y. Leu, Lifetime elongation for wireless sensor network using queue-based approaches, The Journal of Supercomputing 59 (2012) 1312-1335.

[3] I. Stojmenovic, X. Lin, Loop-free hybrid single-path/flooding routing algorithms with guaranteed delivery for wireless networks, IEEE Transactions on Parallel and Distributed Systems 12 (2001) 1023-1032.

[4] X. Chen, J. Wu, The Handbook of Ad Hoc Wireless Networks, CRC Press, USA, 2003. pp. 2–1-2–16.

[5] J.E. Wieselthier, G.D. Nguyen, A. Ephremides, On the construction of energy-efficient broadcast and multicast trees in wireless networks, in: Proceedings of IEEE Infocom, Tel-Aviv, Israel, March, 2000.

[6] E.M. Royer, C.K. Toh, A review of current routing

protocols for ad-hoc mobile wireless networks, IEEE Prersonal Communications 6 (1999) 46-55.

[7] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficient communication protocol for wireless mi-crosensor networks, in: Proceedings of the 33rd Hawaii International Conference on System Sciences (HICSS '00), Hawaii, Jan. 4-7, 2000.

[8] W. Heinzelman, J. Kulik, H. Balakrishnan, Adaptive protocols for information dissemination in wireless sensor networks, in: Proceedings of the 5th Annual ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom99), Seattle, WA, Aug. 15-19, 1999.

[9] K. Scott, N. Bambos, Routing and channel assignment for low power transmission in PCS, in: Proceedings International Conference on Universal Personal Communications, Cambridge, MA, Sep. 29-Oct. 2, 1996, pp. 469-502.

[10] C.K. Toh, Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc networks, IEEE Communications Magazine 39 (2001) 138-147.

[11] I.F. Akyildiz, W. Su, Y. Sankasubramaniam, E. Cayirci, Wireless sensor networks: A survey, Computer Networks 38 (2002) 393-422.

[12] S. Singh, M. Woo, C.S. Raghavendra, Power-aware routing in mobile ad hoc networks, in: Proceedings of the 4th ACM/IEEE International Conference on Mobile Computing and Networking (MOBICOM’98), Dallas, TX, Oct. 25-30, 1998.

[13] N. Bambos, Toward power-sensitive network architectures in wireless communications: Concepts, issues, and design aspects, IEEE Personal Communications 5 (1998) 50-59.

[14] A. Misra, S. Banerjee, MRPC: Maximizing network lifetime for reliable routing in wireless environments, in: 2002 IEEE Wireless Communications and Networking Conference, Orlando, Mar. 17-21, 2002.

0

2

4

6

8

10

MRPC Min‐Energy NEER

Energy per packet

Journal of Communication and Computer（Issue 8，2013）

Documents

Transcript of Journal of Communication and Computer（Issue 8，2013）