Proceedings of
International Conference on advanced Networking,
Distributed Systems and Applications
(Short Papers)
INDS 2014
Bejaia – Algeria
Copyright@INDS’2014
Forward
On behalf of the technical program committee, it is our pleasure to welcome you to the
International Conference on advanced Networking, Distributed Systems and Applications
(INDS'2014) in Bejaia, Algeria.
INDS'2014 is intended to be an excellent forum where both academia and industry experts
will discuss recent advances in the broad and quickly-evolving fields of computer and
communication networks, and to highlight key issues, identify trends, and develop visions for
advanced networking, distributed computing, Security, and their contribution in the emergent
smart systems such as eHealth, smart cities, smart grids, etc.
The INDS 2014 technical program is spread across three days. There will be seven keynote
speeches, tutorials and ten technical sessions: Networking, Peer-to-Peer Networks, Parallel
and Distributed Systems, Network Security, Wireless Communication, Wireless Sensor
Networks, Data Mining, Software and Knowledge and Access Control and Privacy. All
submitted papers were carefully screened to ensure high quality. INDS 2014 received 90
paper submissions from 9 countries. Of these, 18 were accepted as regular papers, 24 as
short papers and 3 as tutorials, appearing in these proceedings. All papers underwent a
rigorous review process: every accepted paper was reviewed by at least two experts, with
more than 55% receiving three or more reviews.
We would like to thank the members of the technical program committee, the anonymous
expert reviewers and the keynote speakers. Without their help, this conference would not be
possible.
On behalf of Committees,
Abdelmadjid Bouabdallah
Yacine Challal
Abdelkamel Tari
Committees
Honorary Chair
Boualem SAIDANI, Rector, University of Bejaia, Algeria
Program Co-chairs
Yacine Challal (UTC, France)
Mohamed Ahmed Nacer (USTHB, Algeria)
Pascal Lorenz, University of Haute Alsace, France
Peter Mueller, IBM Zurich, Switzerland
Publicity Chair
Walid Bechkit, INSA-Lyon, France
Touazi Djoudi, University of Bejaia
Finance Chair
Abdelmadjid BOUABDALLAH, UTC
General Chairs
Abdelmadjid BOUABDALLAH (UTC, France),
Abdelkamel TARI, University of Bejaia (Algeria),
Vahid TAROKH, Harvard university (USA)
Advisory Committee
Nadjib BADACHE (CERIST)
Abdelmadjid BOUABDALLAH (UTC)
Mouloud KOUDIL (ESI)
Abdelkamel TARI (Bejaia University)
Local Arrangements Chairs
KHELFAOUI Youcef, University of Bejaia
Mawloud OMAR, University of Bejaia
Publication Chair
Enrico Natalizio, UTC
Technical Program Committee
Marie-Hélène Abel, UTC, France
Mohamed Achemlal, Orange Labs, Caen, France
Kamel Adi, Université du Québec en Outaouais, Canada
Mohamed Ahmed Nacer, USTHB, Algeria
Abdelouhab Aloui, Bejaia University, Algeria
Hamada Alshaer, KUSTAR, UAE
Youssef Amghar, INSA Lyon, France
Mohamed Anane, ESI, Algeria
Fayçal Azouaou, ESI, Algeria
Abdelmalik Bachir, Biskra University, Algeria
Nadjib Badache, CERIST, Algeria
Amar Balla, ESI, Algeria
Kamel Barkaoui, Cedric CNAM, France
Walid Bechkit, INSA de Lyon, France
Chakib Bekara, CDTA, Algeria
Abdelkader Belkheir, USTHB, Algeria
Karima Benatchba, ESI, Algeria
Nadjia Benblidia, Blida University, Algeria
Mahfoud Benchaiba, USTHB, Algeria
Ahcene Bendjoudi, CERIST, Algeria
Brahim Bensaou, Hong Kong University of Sci & Tech., China
Chafika Benzaid, USTHB, Algeria
Thouraya Bouabana-Tebibel, ESI, Algeria
Malika Ioualalen, USTHB, Algeria
Farouk Kamoun, Tunisia
Tahar Kechadi, University College Dublin, Ireland
Mounir Kellil, CEA, France
Tayeb Kenaza, EMP, Algeria
Djamel Khadraoui, Henri Tudor Research Ctr, Luxembourg
Hamamache Kheddouci, University of Lyon 1, France
Lyes Khelladi, CERIST, Algeria
Mouloud Koudil, ESI, Algeria
Maryline Laurent, Telecom SudParis, France
Pascal Lorenz, University of Haute Alsace, France
Pietro Manzoni, Universitat Politècnica de Valéncia, Spain
OMAR Mawloud, Bejaia University, Algeria
Ali Melit, Jijel University, Algeria
Abdelhamid Mellouk, UPEC, France
Tomasso Melodia, State University of NY at Buffalo, USA
Mohamed Mezghiche, Boumerdes University, Algeria
Achour Mostefaoui, Nantes University, France
Claude Moulin, UTC, France
Peter Mueller, IBM Zurich, Switzerland
Hassina Nacer, Bejaia University, Algeria
Enrico Natalizio, UTC, France
Omar Nouali, CERIST, Algeria
Congduc Pham, Pau University, France
Abdelmadjid Bouabdallah, UTC, France
Mahmoud Boufaida, Constantine University, Algeria
Zizette Boufaida, Constantine University, Algeria
Abdellah Boukeram, Bejaia University, Algeria
Ahcene Bounceur, Brest University, France
Pedro Castillo, UTC, France
Yacine, Challal, UTC, France
Lin Chen, University of Paris Sud 11, France
Isabelle Chrisment, University of Nancy 1, France
Raphael Couturier, IUT Belfort-Montbéliard, France
Ernesto Damiani, Universita' degli Studi di Milano, Italy.
Abdelouahid Derhab, King Saud University, Saudi Arabia
Djamel Djenouri, CERIST, Algeria
Ahmed El Kamal, Iowa State University, USA
Mohamed Fahem, Tlemcen University, Algeria
Said Gharout, Orange Labs, France
Shuai Han, Harbin Institute of Technology, China
Mohamed Ibnkahla, Queens University, Canada
Antonio Iera, UniversitaMediterranea Di Reggio Calabria, Italy
Hani Ragab Hassen, University of Kent, United Kingdom
Michel Raynal, University of Rennes, France
Imed Romdhani, Edinburgh Napier University, UK
Yves Roudier, Eurecom, Campus SophiaTech, France
Ahmed Serhrouchni, Telecom Paris Tech, France
Fatima Si tayeb, ESI, Algeria
Abderahmane Sider, Bejaia University, Algeria
Renaud Sirdey, Commissariat à l’Energie Atomique, France
Djamel Tandjaoui, CERIST, Algeria
Abdelkamel Tari, Bejaia University, Algeria
Zahir Tari, RMIT University, Australia
Bin Tong, Iowa State University, USA
Abderezak Touzene, Sultan Qabus University, Oman
Tarokh Vahid, Harvard University, USA
Veronique Veque, Supelec, France
Wei Wei, Xi'an University of Technology, China
Belhassen Zouari, University of Carthage, Tunisia
Table of Content
Chapter 1 : Peer-to-Peer Networks
- “Improved P2P Routing Protocol CHORD for Mobile Networks”, Sara Cherbal (University of Bordj Bou Arreridj, Algeria), Abdellah Boukerram (University of Bejaia, Algéria), Abdelhak Boubetra (University of Bordj Bou Arreridj, Algeria) …………………………………………………………p. 1-4
- “Cross-layer Design of Clustering Scheme for Peer to Peer over Manet”, Moufida Rahmani (LSI, USTHB, Algeria), Mahfoud Benchaïba (LSI, USTHB, Algeria)……………………………………p.5-10
- “Toward a match between the P2P overlay and the MANET underlay”, Manel Seddiki (LSI, USTHB, Algeria), Mahfoud Benchaïba (LSI, USTHB, Algeria)…………………………………..p.11-16
Chapter 2 : Networking and Wireless Communication
- “Green Power Management for Wireless Mesh Networks”, Sarra Mamechaoui(University of Tlemcen, DZ), Fedoua Didi(University of Tlemcen, DZ), Guy Pujolle(University of Paris 6, FR)..p17-20
- “Reliable Hierarchical Cluster-Based Routing Protocol for Wireless Sensor Networks”, Mohamed A. Eshaftri (Edinburgh Napier University, UK), Alsnousi Essa (Edinburgh Napier University, UK), Mamoun Qasem (Edinburgh Napier University, UK), Imed Romdhani (Edinburgh Napier University, UK), Ahmed Al-Dubai (Edinburgh Napier University, UK)……………………………………..p.21-26
- « Mobile User Authentication System for e-Commerce Applications », Rania Molla (King Abdulaziz University, Saudi Arabia), Imed Romdhani (Edinburgh Napier University, UK), Bill Buchanan (Edinburgh Napier University, UK) and Etimad Y. Fadel (King Abdulaziz University, Saudi Arabia)………………………………………………………………………………………………p27-34
- “An overview of the emergency and infotainment data dissemination in VANET”, Sedjelmaci Amina (University of Tlemcen, DZ), Fedoua Didi (University of Tlemcen, DZ)…………………p35-39
- “Printed Antennas for UHF RFID Passive Tags”, Mohammed Zakarya Baba-Ahmed (LTT, Tlemcen, Algeria), Fatima Zahra Marouf (LTT, Tlemcen, Algeria), Nawel Seladji (LTT, Tlemcen, Algeria): Mohamed Ryad El-Mansour Fekkar (LTT, Tlemcen, Algeria).......................……………………...p40-45
- “New Pilot Pattern Arrangement for Channel Estimation in LTE Downlink”, Fatma Zohra Bouchibane and Khalida Ghanem …………………………………….…………………………...p46-50
- “INTRODUCTION OF SPATIO-TEMPORAL ORTHOGONAL CODES IN WIRELESS COMMUNICATIONS TRANSMISSION”, Bendimerad Mohammed Yassine (University of Tlemcen, Algeria), Bendimerad Fethi Tarik (University of Tlemcen, Algeria), Ferouani Souhila (University of Tlemcen, Algeria)……………………………………………..…………………………………….p51-57
- “Get Back the Ownership of Your Calls - A Transparent Approach for Protecting VoIP Calls”, Markus Gruber (Vienna University of Technology, Austria), Martin Karl Maier (Vienna University of Technology, Austria), Michael Schafferer (Vienna University of Technology, Austria), Christian Schanes (Vienna University of Technology, Austria), Thomas Grechenig (Vienna University of Technology, Austria)……………………………………………………………………………….p58-61
- “Energy Detection Performance of Cognitive Radio Networks”, Haroun Errachid Adardour (STIC Laboratory, University of Tlemcen, Algeria), Maghnia Meliani (STIC Laboratory, University of Tlemcen, Algeria), Mohammed Feham (STIC Laboratory, University of Tlemcen, Algeria)…….p62-67
- “Combining Cooperative Diversity and Turbo Coding in wireless communications”, Hakim Tayakout (CDTA, ENP, Algiers, Algeria), Khalida Ghanem (CDTA, Algiers, Algeria), H. BOUSBIA-SALAH (ENP, Algiers, Algeria)………………………………………………………………………………………………p.68-71
Chapter 3 : Network Security
- “Tailoring Mickey-Ticket to e-Health Applications in the Context of Internet of Things”, Mohammed Riad Abdmeziem (LSI, USTHB, Algeria) Djamel Tandjaoui (CERIST, Algiers, Algeria)…………………………………………………………………………………………….p.72-77
- “Hardware AES IP for Embedded Cryptosystem on FPGA”, Nadjia Anane (CDTA. Algiers, Algeria), Mohamed Anane (ESI, Algiers, Algeria)………………………………………………...p.78-81
- “A reputation-based approach using collaborative indictment/exculpation for detecting and isolating selfish nodes in MANETs”, Lotfi Zaouche (Heudiasyc Lab, Compiègne, France), Sofiane Ait-Arab (LISSI Lab, Paris Est Créteil, France), Anfel Khireddine (LIMED laboratory, Bejaia, Algeria), Mawloud Omar (LIMED laboratory, Bejaia, Algeria), Enrico Natalizio (Heudiasyc Lab, Compiègne, France), Abdelmadjid Bouabdallah (Heudiasyc Lab, Compiègne, France)…………………….…p.82-85
Chapter 4 : Software and Knowledge - “Survey of Change Impact Analysis Approaches for Software Evolution”, Adenane Hidouci
(Constantine 2 University, Algeria), Youssef Amghar (LIRIS, INSA Lyon, France)…………….p.86-91 - “Using ontology as prior conceptual knowledge in an ILP system for fault diagnosis”, Samiya
Bouarroudj (LIRE Laboratory Constantine 2 University, Algeria) Zizette Boufaida (LIRE Laboratory Constantine 2 University, Algeria)………………………………………………………………...p.92-95
- “New Approach for the Construction of User Profile: E-recruitment”, Taous Iggui (University of Bejaia, Algeria), Hassina Nacer (University of Bejaia, Algeria), Youcef Sklab (University of Bejaia, Algeria) Taklit Ait Radi (University of Bejaia, Algeria)……………………………………………p96-99
Chapter 5 : Parallel and Distributed Systems
- “A coalition model for resource management in Green Cloud Computing”, Nassima Bouchareb (LIRE Laboratory, University Constantine 2, Algeria), Nacer Eddine Zarour (LIRE Laboratory, University Constantine 2, Algeria) Samir Aknine (LIRIS Laboratory, University Lyon 1, France)…………………………………………………………………………………………..p.100-103
- “Using Mobile Agent approach for Balancing the Load of a Parallel Simulation application into Cloud Computing”, Selma Chenni (University Bordj Bou Arreridj, Algeria), Abdelhak Boubetra (University Bordj Bou Arreridj, Algeria), Saad Harous (United Arab Emirates University, UAE)……………………………………………………………………………………………p.104-107
Chapter 6 : Wireless Sensor Networks
- “Distributed Detecting Boundaries of Coverage Holes in Wireless Sensor Networks”, Lynda Aliouane (LSI, USTHB, Algeria) Mahfoud Benchaïba (LSI, USTHB, Algeria)…………….....p.108-111
- “Fault-Tolerance in LEACH Protocol (MC-LEACH)”, Chifaa Tabet Hellel (STIC, University of Tlemcen, Algeria), Mohammed Lehsaini (STIC, University of Tlemcen, Algeria) and Herve Guyennet (LIFC, University of Franche Comté, France)…………………………………………………...p112-115
An Improvement of CHORD Routing Protocol For
Mobile P2P Networks
Sara CHERBAL
Computer Science Department
University of Bordj Bou Arreridj
34000 Bordj Bou Arreridj, Algeria
Abdellah BOUKERRAM
Computer Science Department
University of Bejaia
06000 Bejaia, Algeria
Abdelhak BOUBETRA
Computer Science Department
University of Bordj Bou Arreridj
34000 Bordj Bou Arreridj, Algeria
Abstract— Nowadays, with the unlimited evolution of mobile
devices such as PDAs and tablets, as well as the fast and the
interesting development of their computing resources, with
performances comparable to that of PCs, they became
competent of storing gigabytes of digital content, allowing them
an ideal platform for peer-to-peer (P2P) content sharing. On the
contrary, their energy consumption presents a critical element,
as their work relies on a battery that has a limited-energy level.
The objective of this work is to develop an approach to
adapt P2P routing to this type of dynamic environment, taking
into account the energy levels of nodes and also the optimization
of the energy consumption. It is based on distributing the load
which is the zone of responsibility of a node, in a uniform
manner, taking into account the energy level of nodes.
Keywords— P2P; P2P Routing; Chord; Energy
Consumption; Load Balancing.
I. INTRODUCTION
Scaling, dynamic nature and heterogeneity provided by
P2P overlays and networks present high demand aspects in a
large scale distributed systems which make them well known
and very used over the recent years. P2P overlays approaches
lead to easier networks to operate, that respect the high-level
application requirements.
P2P overlay is a logical virtual network built on a real
physical network through which participating peers (nodes in
the physical network) are organized in a distributed manner
without hierarchy or centralized control [1].
P2P system does not respect the traditional Client/Server
paradigm, because of peers that each of them is a client and a
server at the same time. They communicate to establish self-
organized overlay structure on top of physical networks, in
order to support and maintain a variety of service and
application level [2].
There are two P2P architectures, the unstructured one
which is the first to appear and the structured one with
prominent improvements.
Random propagation of queries between peers in
unstructured architectures was replaced in structured
architectures by a system called DHT “Distributed Hash
Table”, based on hash function that gives each node and each
resource a unique identifier with a specific format, which
helps to improve the lookup function by organizing the
storage space and placing resources on nodes.
In energy consumption field, when talking about
networks, achieving energy efficiency should involve
benefits and advantages, namely, hosts, devices and network
protocols, as well as the communication costs.
Regarding the latter, researches performed on
classification and quantification of different traffic types
found that the most important traffic due to P2P overlays and
networks (40–73%) [3]. Therefore, P2P energy efficiency
must be taken into consideration, as it presents the large part
of internet traffic. The processing of incoming messages [4]
and the necessary computation of P2P protocols functions are
the reason of the exacerbation of the energy consumption.
II. ENERGY ASPECTS WITHIN P2P OVERLAYS
First of all, it is necessary to note that the energy
consumption is not only about P2P systems that are designed
to consume less energy, but also about systems that use
energy-limited devices and must consume less energy to
avoid exhaustion of their batteries [2].
In the latter case, the longer the battery life increases, the
charge level drops, which requires low power consumption.
The energy optimization in P2P networks can be
achieved by optimizing the lookup function, since quickly
research saves time and therefore it gains energy, it can also
be achieved by balancing the load between nodes according
to their energy level, i.e. giving overload to the node with the
highest level of energy in order to ensure the continuity of
work and thus increase the lifetime of the network.
III. RELATED WORK
By studying research review in the domain of energy
efficiency, we met a significant number of approaches which
are divided into categories based on the followed methods in
their solutions.
The authors of [2] mentioned some works on P2P energy
efficiency, according to an arrangement of some of the most
used methods in this field, the table below shows this
organization:
TABLE I. ENERGY EFFICIENT P2P RESEARCH WORKS
ORGANIZATION[2]
Method Description
Proxying The mobile node delegates its tasks to a
proxy and it goes to sleep mode.
Sleep-and-Wake Advanced mechanisms to switch a node
between on and off state.
Task allocation
optimization
Allocation of tasks between nodes
depending on the lifetime of each.
Message
reduction
By minimizing the number of required
messages for P2P operations.
Overlay structure
optimization
Modify or introduce P2P overlays that
suit the energy requirement.
Each method has shown its impact on energy efficiency
in many aspects and specifically in P2P overlay networks, as
we talk about DHTs in this paper, we will focus on works
that have applied these methods on P2P DHT-based
protocols. Some of them tried to apply these methods on
DHTs, as the proxying one on Kademelia DHT-based
protocol [4] but there were no experimentations results, a
sleep and wake approach was applied on DHT-based
protocol called Chimeria [5], in which they propose the
concept of idle time when there is no transmission, during
this time the mobile peer can go to sleep, until he wakes up
because of a query or messages sent by another peer.
The energy-efficiency approach the most applied in DHTs
is the “Message reduction approach”, wherein we reduce the
number of required messages for protocol operation and
maintenance. Since in structured P2P approaches the number
of maintenance messages is great, therefore the energy
consumption is more considerable.
The authors of [4] find that the most significant reason of
the energy consumption is the processing of incoming-
messages; hence they propose to execute mobile peers in
client mode and block all their incoming-messages. This
approach increases the lifetime of mobile peers, but as limit,
we note that it cannot be successfully applied on living
network because of the no ability of controlling the behavior
of nodes; therefore the need of this execution environment
requires further Researches.
Another Message reduction approach with Kademelia-
based DHTs [6], which is based on deleting messages
according to a dynamic probability, its value varies
depending to the energy state of the peer. These protocol
changes are applied only on the mobile network nodes, such
idea of deleting messages helps to reduce the energy
consumption of limited-energy devices, but it affects the
cooperation of peers, this problem can be solved by
increasing number of replication values stored in Kademelia
to increase the chances of operations success.
As limitation of this approach, the determination of the
probability value, for example where there are 90 mobile
nodes in the network, the probability takes a value between
0,3-0,4; therefore the determination need to know the rate of
mobile nodes in the network, and as DHTs are completely
distributed, make this function more difficult with methods
that require a lot of time and bandwidth, which present an
open problem need further research.
An improvement of Chord was proposed in [7] wherein
the authors find that rather than focusing at minimizing the
number of messages in the process of sharing files (by setting
TTL) and decreases the chances of finding the requested file,
it will be better to focus on minimizing the size of these
messages using the BF algorithm which is based on hash
functions.
This compression has shown its benefits by decreasing
the size and the number of messages sent, for achieving a
quick lookup function; it provided an improvement in the
structure of the finger-table which allows reducing the
number of hops, and thus a preserved bandwidth. With these
advantages, there is always a false-positive ratio in the
compressed messages by the BF algorithm, and thus it also
exists in the lookup function, and it mostly increases with
increasing number of nodes despite trying to optimize, so it
is not preferred for large-scale networks.
IV. CHORD OVERVIEW
CHORD [8] provides a scalable and efficient protocol for
dynamic lookup in a P2P system with frequent arrivals and
leaving of nodes. It presents a ring of 2m
nodes with
clockwise directional links.
A hash function (e.g. SHA-1) is used to assign to each
node and resource an identifier with m bits length, this hash
function must have good properties (low risk of collision,
reversal impossible).
Node identifier (ID) = hash (IP address)
Resource identifier (key) = hash (resource name)
The nodes are distributed on the CHORD ring according to
their IDs in ascending order.
To improve research in chord, they use a more complete
routing table called finger table, which contains m entries,
each entry has a key K and the responsible node (the finger)
which is the first successor in the ring whose identifier is
equal to or greater than k.
The area of responsibility of a node is the set of keys that
belong to the identifiers space:
]id.previousNode, id.actuelNode]
When a node wants to find a key, it checks if the key is
between the start node and its successor. Otherwise it
searches through its fingers node whose identifier is the
largest while being less than the key, and transform the
message.
V. BASIC CONCEPTS OF THE PROPOSED APPROACH
In the structure of the DHT, the finger-tables have a fixed
number of outgoing messages that is the number of fingers of
each node [8], but there is no known distribution of incoming
messages. These depend on the zone of responsibility of the
node, i.e., the number of keys that the node is responsible for.
This zone differs from one node to another due to the random
placement of identifiers (IDs) on the ring. The balancing of
these zones helps to balance the routing protocol itself.
Among the works that have applied improvements to
DHT-based protocols we mention the one of [9] named e-
chord in which the authors aim to improve this balance by a
simple enhancement in the finger selection algorithm of
Chord, but instead of pointing the finger to the node N
chosen by the Chord standard algorithm, they choose one of
his successors to be the finger according to some probability,
as in the structure of Chord a node knows s1 of his
successors, the authors also mention that this choice is made
in a random manner.
This random choice to share the load of a node cannot
provide good results for the performance of the system
because it can increase the load of the chosen node at least.
Therefore, we propose to improve the Chord finger
selection algorithm by choosing the finger node in terms of
some criteria that we define to help us to better balance the
load of nodes and also take into account the energy
consumption.
A. The defined criteria
The following points present an overview of the criteria
that are applied in our approach to select a finger from the set
E which represents N and its s successors, so the choice will
be based on:
1. The energy level of the node, where the node with a
higher level of energy is the most preferable, with a
property to remain active for a longer time.
2. The actual distance (geographic) between the node
who looking the best finger to place in its finger-
table and the node chosen to be the finger, here it
prefers the one closest to accelerate the send of
query, as the most short distance implies the speed
of lookup function.
3. Workload of the node or the size of the zone of
responsibility, the one with the low workload level
is most preferred to receive the overload.
In more detail, a node with an identifier id when creating
his table fingers, it starts by applying this formula:
key = (id + 2k-1
) mod 2m, 1 <= k <= m (1) [8]
to calculate the keys, then for each key it applies the search
function to find the node with an identifier id greater than or
1 S = a value shows the number of successors, in this work we give it the
value 3, may be in future work we will change it with certain criteria.
equal to the calculated key, which is the responsible (as in
basic Chord), it sends a query to this node, afterwards the
latter instead of responding with its IP address, it responds
with an IP address of a chosen node of the set E according to
the criteria we have already define above.
B. Overview of proposed energy-based approach
As our first goal is to take into account the energy
consumption of nodes, we focus now on the first criterion
regarding the balancing in terms of the energy level of nodes
in order to increase the lifetime of the network. For this, we
will give more details on how it works by explaining its
operating system using the figure below (Figure 1).
Parameters we will meet in the following:
N: the node chosen by the chord standard algorithm to be the
finger node.
C0: the percentage that represents the energy level of the
node N.
i: a value in (1,2,3) ,as s=3, presents the classification of the
successor in the ring.
Ci: the percentage that represents the energy level of the
successor whose classification is equal to i.
IP0: the IP address of the node N.
IPi: the IP address of the successor whose classification is
equal to i.
Fig. 1. Energy-based approach overview (Example)
The figure shows the operating process of this approach
as follows:
- The node N6 is creating its routing table (finger-table),
- It executes the calculation of the formula (1) to find the
values of the keys, - It finds a key with the value 9, the standard algorithm
Chord gives N5 as responsible for this key,
- Then N6 sends a query to N5 (N5 knows their
successors N4, N10 and N7 respectively), - As it appears in the ring, N4 with i=1 as it is the first
successor, N10 with i=2, N7 with i=3, - N5 sends its C0 + IP0 to N4, - N4 compares his C1 with received C0; in the example:
C1 is the smallest,
- N4 sends C1 + IP1 to N10,
Key resp
… …
… …
… …
9 N? C0=85%
C1=65%
C2=35%
C3=55%
0
6
5
N1
15
4
8
10
N5 9 7
1
3
2
12
11
13
14
N6
N8
N3
N9
N2
N4
N7
N10
- N10 compared its C2 with received C1 ;following the
example it sends C2 + IP2 (the smallest) to N7,
- N7 compared its C3 with received C2 ; according to
Example C2 remains the smallest, - N7 sends a request to N5, contains IP2 (address of the
successor with the smallest load),
- N5 in its turn sends this IP2 to N6, At the latest, the node selected by our algorithm to be the
finger in N6 finger-table is N10, which is the most powerful
one that give us a long lifetime of the network.
The algorithm of this approach is follows:
pred N
succ successor (N)
C0 get.load // get the energy load of the actual node
IP0 get.address // get the IP address of the actual node
Info ← (C0,IP0)
For i=1 to 3 do
send (pred, succ, Info) // the pred node sends the energy
load and IP address contained in Info to the succ node
Ci-1 Info.load // get the energy load contained in Info
Ci get.load
IPi get.address
If (Ci-1 < Ci) then
Info (Ci-1, IPi-1)
else
Info (Ci, IPi)
end if
pred succ
succ successor (succ)
end for
IP Info.address // get the IP address contained in Info
Send (pred,N,IP) // the pred node sends the IP address
of the successor which has the smallest energy-load to node N
C. Analysis of energy-based approach
In this process, each mobile node knows its energy load,
and has no information about other nodes, for not
overloading the nodes.
The requests sent contain only one IP address and a small
number indicates the percentage of energy level, so they are
not heavy when sending.
We can implement this process by a mobile agent to
avoid these queries.
In this way, we give the workload to the node that has a
longer life, to ensure the continuity of the work of the
network, and as another advantage we don’t make changes in
the structure of the routing tables as we don’t put any data or
information in it, therefore we avoid its overloading.
VI. CONCLUSION AND PERSPECTIVES
The two fields of lookup protocols optimization and
energy consumption have known a big tendency by
researchers regarding networks and distributed systems. In
our work, we tried to focus on these two domains on P2P
overlay networks but we gave more priority to the one that
takes into account the energy consumption to have an
approach that adapts better to dynamic environments with
limited-energy devices.
In this paper, we mention our idea to realize this
approach, which is improving the chord finger selection
algorithm without any involvement of the overloaded of
routing tables or nodes, we have given a small theoretical
description on the three criteria that we defined to make this
choice, but we gave more details in the explanation of
energy-based approach and how it balances the load
according to energy level of nodes.
In our future work, we aim to achieve the following points:
Apply each criterion according to the needs of the
network to be simulated.
Validate this approach by implementing these changes
on the Chord protocol, under the Open Chord
platform using the Java language.
Use a simulator (of networks in general or of P2P in
specific) and to compare our approach with that of the
standard Chord and the e-chord.
Apply our approach to other P2P DHT-based
protocols as Pastry, in order to show the impact it
adds compared to standard protocols.
REFERENCES
[1] S. Khan, and A. Gani, “Model of Complex Searching Over Structured P2P Overlay under Dynamic Environment”, International Journal of Information and Electronics Engineering, Vol. 2, No. 2, 2012.
[2] A. Malatras, F. Peng, and B. Hirsbrunner, “CHAPTER X: Energy-efficient peer-to-peer networking and overlays”, handbook of green information and communication systems, eds. M. S. Obaidat, A. anpalagan, I. Woungang, by elsevier, Departements of Informatics, University of Fribourg, Switzerland, 2011.
[3] J. Erman, A. Mahanti, M. Arlitt, and C. Williamson, “Identifying and discriminating between web and peer-to-peer traffic in the network core”, In Proc. of the 16th International Conference on World Wide Web (WWW), 2007.
[4] I. Kelenyi, and J. K. Nurminen, “Energy Aspects of Peer Cooperation – Measurements with a Mobile DHT System”, In Proc. of IEEE International Conference on Communications (ICC), 2008.
[5] S. Gurun, P. Nagpurkar, and B. Y. Zhao, “Energy Consumption and Conservation in Mobile Peer-to-Peer Systems”, In Proc. of 1st International workshop on Decentralized resource sharing in mobile computing and networking (MobiShare), ACM, 2006.
[6] I. Kelenyi, and J. K. Nurminen, “Optimizing Energy Consumption of Mobile Nodes in Heterogeneous Kademlia-based Distributed Hash Tables”, In Proc. of 2nd International Conference on Next Generation Mobile Applications, Services and Technologies (NGMAST), 2008.
[7] S. Wang, H. Ji, and Y. Li, “BF-Chord: An improved lookup protocol to Chord based on Bloom Filter for wireless P2P”, In Proc. of 5th International Conference on Wireless Communications, Networking and Mobile Computing (WiCom), IEEE, 2009.
[8] Ion Stoica, Robert Morris , David Karger , M. Frans Kaashoek , Hari Balakrishnan. “Chord: A Scalable Peer-to-Peer Lookup Service for Internet Applications”. In Proceedings of SIGCOMM. San Deigo, CA, 2001.
[9] R. Cuevas, Manuel Urueña, Albert Banchs. “Routing fairness in Chord: Analysis and Enhancement”. IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2009 proceedings, 2009.
Cross-layer Design of Clustering Scheme for Peer toPeer over Manet
Moufida Rahmani and Mahfoud BenchaıbaLSI, computer-science departement
University of science and technology Houari [email protected]
Abstract—During the last years, multiple peer-to-peer (P2P)overlay networks over mobile ad-hoc network (MANET) areproposed but most of them do not consider the physical proximity.In this case, a single overlay hop usually incurs many physicalhops. This increases physical network traffic and latency whichis not suitable in such environment with very limited bandwidth.Additionally, the message overhead increases with the growthof network size. This often leads to the problem of scalabilityand decreases the search success rate. In order to improvescalability and the performance of mobile peer to peer (MP2P)systems, reducing the message overhead is the main challenge.In this paper, we propose Cross-layer design of ClusteringScheme for MP2P (CCSM). Our proposed algorithm considersthe number of physical hops between peers and their clusterhead,the number of the connected peers of each clusterhead andthe peer’s availability. CCSM groups peers that are physicallyclose to each other into common overlay clusters and limitsmost traffic between some particular peers called clusterheads.Thus, it reduces physical network traffic generated by overlayenhancing the system performance. This paper presents alsotopology construction which consists of the bootstrapping processand establishment of the neighbor relationship process.
Keywords—Mobile peer to peer, Clusterhead, Cross layer, Avail-ability.
I. INTRODUCTION AND MOTIVATION
Peer-to-peer (P2P) is a form of overlay network allowingdirect sharing of resources like CPU, bandwidth, files etcamong a large number of users in a decentralized manner. It isnowadays widely adopted on the Internet and is used for manydifferent application types, such as file sharing (Gnutella [1]and KaZaA [2]), communication as Skype [3] and distributedcomputing as SETI@home [9]. Another network concept isthat of mobile ad-hoc network (MANET) which is a set ofautonomous mobile nodes that form a temporary network andcommunicate directly using wireless links.
P2P and MANET networks are both decentralized and self-organizing networks with dynamic topology. This commonnature has attracted a lot of attention in the research communityto deploy P2P applications for MANET. Compared to P2Poverlay networks in the Internet, P2P overlay over MANET(MP2P) faces many constraints such as the limited bandwidth,limited energy and node mobility. Therefore, the applicationsof traditional P2P can not be directly applied in MP2P withoutmodifications and improvements.
Multiple MP2P architectures are proposed in the recentyears but most of them do not consider the underlying physicalnetwork topology in the construction and maintenance of theiroverlay. Two overlay neighbor nodes may not be physicallyclose to each other. In this case, a single overlay hop usuallyincurs many physical hops and a message is not routed inthe shortest physical path. This increases physical networktraffic and latency which is not suitable in such environmentwith very limited bandwidth. Flooding is a typical lookupalgorithm. This method is simple but generates overhead. Themessage overhead induced by flooding and maintaining theoverlay network increases with the growth of network size.This often leads to the problem of scalability and decreasesthe search success rate. The main issue is thus to reduce thisoverhead by considering physical proximity to construct theoverlay network and limit most traffic between some particularpeers. A well known techniques to solve this issue is clustering.
In this paper, we propose cross-layer design of clusteringscheme for MP2P (CCSM). In CCSM, the peers in the samephysical proximity (in terms of number of hops) are grouped inthe same cluster. A new join peer chooses the suitable clusterto join, based on the physical hop count with clusterhead,the cluster size and the clusterhead’s availability. The char-acteristics of Manet such as dynamic topology due to nodemobility, limited energy make configuration and maintenanceof clustering very difficult. Therefore, a careful choice is to bemade regarding the clusterhead selection and the cluster size.CCSM based on availability to select the clusterhead whichis calculated by the peer’s online time and the energy amountof peer’s device. The clusterhead with larger availability valuemeans that it can remain for a long time. Therefore, its failurerate is reduced thus can improve stability and availability ofoverlay and the performance of MP2P systems. In order tobalance the load of clusterheads, the reasonable number of theconnected peers of each clusterhead (cluster size) is limited.
In this paper, we also present a simple and efficient topol-ogy construction. When a node wants to join a P2P overlay,it first passes by the bootstrapping process which consists tofind the existing peers in the network. After that, it choosesamong them the overlay peers that are also likely to be ”close”to it in the physical network in order to establish the neighborrelationship. Finally, it chooses a cluster to join.
The rest of this paper is organized as follows: The relatedwork is briefly described in Section II. Section III introduces
the proposed system in detail. Section IV provides conclusionand presents future work.
II. RELATED WORK
Clustering is a technique organizing nodes into differentvirtual groups called clusters. Each cluster is managed bya clusterhead which would have to handle the most part ofnetwork traffic. Clustering is an effective technique to reducemessage overhead in MP2P network. Thus, scalability andperformance of these systems can be improved. This sectionintroduces some related works to situate and motivate ourapproach
ORION [4] (Optimized Routing Independent Overlay Net-work) is one of the most typical MP2P which is completelyimplemented at the application layer and does not depend onsupport of a MANET routing protocol. In [5], the authorspropose an approach to construct an efficient unstructured P2Poverlay over MANET by considering the underlying physicalnetwork topology. One of the peers in the P2P network isused as a root-peer to connect all peers. Each peer maintainsconnection with physically closer peers such that it can reachthe root-peer. In these works, during file discovery process,peer utilizes flooding to send a request to all peers until targetfile is found which generates much overhead.
The approaches presented in [6][7] proposed clusteringfor structured (DHT-based) peer-to-peer overlays. They useclustering to regroup the nodes that are physically close toeach other and assign them the same overlay ID prefix. Con-sequently, the amount of physical traffic generated by overlayscan be significantly reduced. There is no cluster ”head” in theseworks and they use bootstrap node during the bootstrappingprocess. Our approach is specific for unstructured MP2P, theutilization of cluster head is crucial in order to ensure scalabil-ity [8]. In bootstrapping process, the utilization of centralizedsource (bootstrap node) in order to join the P2P overlay is notsuitable for MANET which is fully decentralized.
In [9], the authors proposed a single hop clustering ap-proach from clusterhead for unstructured MP2P. Each nodethat joins the network sends HELLO message to all of itsneighboring nodes (the peers within its communication range).When receiving reply from multiple clusterheads, this nodechecks energy level of those clusterheads and sends JOINmessage to the higher one.
In [10], two other single hop clustering approaches areproposed. In the first one, the peer with the highest degreein the network is selected as the first super-peer and its neigh-boring nodes become its sub-peers. After the selection, superpeer and its sub-peers are removed virtually from the network.Then the method selects and removes repeatedly super peerand sub-peers until there is no more peer in the network.In the second, each peer chooses a random number and thencompares it with its neighboring nodes. A peer with the largestnumber among its adjacent peers becomes a super peer. Thesecond method can be implemented in a distributed mannerunlike the first that needs a server to maintain the super-peerswhich is difficult to achieve in such a system. However, bothmethods do not consider the characteristics of mobile peer-to-peer network. A super peer may immediately leave thenetwork after the selection because its availability is limited
and can not remain for a long time. Therefore, the failurerate of super peer increases which leads to longer query delayon searching files. Our CCSM tries to solve these problemsby selecting the peer with high availability as clusterhead ina distributed manner. These approaches apply the clusteringafter the topology construction whereas our approach functionsin an incremental manner from the beginning of networkdeployment.
The approaches proposed in [9][10] do not limit the numberof cluster members, thus the aspect of the load balancingbetween the clusterhead is not treated. Clusterhead can notremain for a long time because its energy is exhausted. There-fore, overhead message can be increased by the maintenanceprocess. Our CCSM treats this aspect by limiting the numberof members in each cluster at a reasonable number.
In [11], the authors have proposed an efficient MP2Parchitecture which considers the maximum connection timeof connected peers, the hop count with the super peer, and thenumber of connected peers of the super peer. The connectiontime between the connected peers can be determined by thelocation, velocity and communication range of each mobilepeer.
III. THE PROPOSED ALGORITHM
In this section we describe our proposed scheme CCSM.It considers the number of physical hops between peers andtheir clusterhead, the cluster size and the peer’s availability.Additionally, we describe how peer joins a P2P overlay andhow it establishes the neighbor relationship.
A. MP2P network environment
We consider the MP2P scenarios, the same as used in [5]where not all nodes share and access the files, i.e. some arepeers and others are no-peers. We called the node that joins theP2P network for sharing and accessing the files as a peer. No-peers are called simple nodes. The nodes cooperate to forwardthe data for the other peers. Figure 1 shows our MP2P networkenvironment. When we apply CCSM (figure 1), There are twotypes of peers: Normal peer and Clusterhead. Clusterhead isthe peer which manages its cluster and maintains the files meta-data that its normal peers are sharing. We believe that thechoice of clusterhead is crucial that’s why in this paper wetried to choose clusterhead which remains for a long time aspossible. Thus, the failure rate of clusterhead is reduced whichgives higher hit ratio for file searches.
In this network, each peer Pi has its own information (NP,NC, nature, DPi, EPi, AV AILPi), where NP is IP addressof peer Pi, NC is IP address of its clusterhead, nature is todistinguish whether the peer is a clusterhead or normal peer,DPi is Pi’s online time, EPi is the energy amount of peer Pi’sdevice and AV AILPi is peer’s availability and is used duringthe clusterhead selection. The bigger the AV AILPi value, themore available is peer Pi and can perform tasks. Pi, wheneverit joins the P2P file sharing network, calculates AV AILPi asfollows:
AVAILPi = EPi/ DPi.(1)
Fig. 1. The proposed MP2P architecture.
AV AILPi means that a peer may wish to remain DPi
time for the current session but its energy EPi may allow itor not. EPi is estimated whenever a peer joins the P2P and itis the energy amount of Pi’s device for a session. DPi is Pi’sonline time and is calculated as follows:
DPi=∑n−1
0 TPi / n.(2)∑n−1
0 TPi is peer’s online lifetime during the last n times,n is the frequency of a peer joining the network.
B. Structure
Each peer (clusterhead or normal) maintains naturally therouting table at the network layer like simple nodes. It main-tains also file table and neighbor-peer table at the applicationlayer. The file table stores the files provided and shared bythe peer. Peer stores in the neighbor-peer table informationabout its neighbor peers like their IPadress and the physicaldistance (hops) that separates it from them. When a peerbecomes a clusterhead, it will maintain also cluster-membertable. The cluster-member table contains information aboutthe peers which belongs to it. There will be 4 entries relatedto each member peer: IPadress, filename, physical distance andAVAIL.
C. Variables
Table 1 summarizes the notations of variable, which weshall use in this paper.
D. Peer joining process
In this section, we describe how new peer: joins P2Poverlay (the bootstrapping process), creates neighborhood andjoins a cluster. When a new peer wishes to join a P2P overlay, itbroadcasts Discovery request (Hello message) with TTL equalto Rn. Each node receiving this message, decrements the TTLand forwards the message if the TTL does not reach 0. Eachpeer (normal or clusterhead) receiving this message, respondsalso by message which contains NUM value. Additionally,
TABLE I. SUMMARY OF NOTATIONS.
Notation SignificanceRc number of physical hops tolerated for searching
a clusterhead to join its clusterRn number of physical hops tolerated for searching
neighborsMAXN maximum number of logical neighbors
that a peer may haveMINN minimum number of logical neighbors
that a peer may haveMAXC maximum number of members
that a clusterhead may haveNUM current number of logical neighborsTTL Time To Livecluster-size number of peers in the clusterknown-peers list a list of known peerscandidate-neighbors list a list of peers with whose NUM is lower than MAXNcandidate1-clusters list a list of peers with whose NUM cluster-size is lower then
MAXC and that exist at distance lower than Rccandidate2-clusters list a list of peers which have AVAIL greater than
the AVAIL of a particular peerneighbor-peer table a list of neighbor peerscluster-member table contains information of members of a cluster
The message must contain AVAIL value and cluster size if thereceiving peer is clusterhead. The new peer stores all the peerswhich reply in known-peers list.
1) How new peer chooses its neighbors: If known-peerslist is not empty, the new peer chooses among this list peersthat have their NUM lower than MAXN as candidates for theneighborhood and stores them in candidate-neighbors list. Thecandidate-neighbors list is ordered from smallest to largestaccording to physical distance (number of physical hops).
After that, the new peer sends Neighborhood-Peer requestto the peers from its candidate-neighbors list to ask themto become their neighbors. Each peer receiving this message,must reply, stores the sending peer as its neighbors peer in itsneighbor-peer table. The sending Neighborhood-Peer requestis stopped when the NUM of this new peer is equal to MAXNor candidate-neighbors list becomes empty. Each peer alsoperiodically sends a Heartbeat message (Hello message) to itsneighbors to inform them that is alive. When the NUM of apeer is equal to MINN, in this case a peer must search newneighbors.
2) how peer chooses a cluster: The Algorithm proposedin this paper is Rc-hops clustering i.e. the number of physicalhops between the clusterhead and its members must notexceed Rc. If known-peers list is empty, the peer becomesa clusterhead and new cluster is created. Otherwise, the peerchooses the clusterheads among this list that exist at Rc hopsor lower than and whose cluster-size is lower than MAXC. Itstores them in candidate1-clusters list. To balance the load ofeach clusterhead, the number of the connected peers of eachclusterhead is limited to MAXC. If candidate1-clusters list isempty, in this case the new peer becomes a clusterhead andnew cluster is created. Otherwise, there are two methods tochoose a cluster:
• The first method is a simple method and it will reducethe traffic of re-clustering when the clusterhead leavesthe network. For each clusterhead in candidate1-clusters list, the peer will calculate for them a weightnoted W. The peer chooses the clusterhead which hasthe max value of W and sends to it Join message.
Fig. 2. An example of peer join.
While Receiving Join message, clusterhead stores thesending peer as member in its cluster-member table.Join message contains the information of IPadress,filenames, physical distance and AVAIL. The W takesinto account the cluster-size in order to balance loadbetween the clusterheads. The W is calculated asfollows:
W=number of physical hops∗p+( AVAIL /cluster-size+1)∗q,
where p+ q = 1.(3)
• The second method is based on the principle that theclusterhead is the peer that has the greatest value ofAVAIL than its members. Therefore, the failure rateof clusterhead is reduced improving stability and theperformance of MP2P systems. The peer chooses fromcandidate1-clusters list the clusterheads with AVAILgreater than its AVAIL. It stores them in candidate2-clusters list. If candidate2-clusters list is empty, inthis case the new peer will become a clusterhead andnew cluster is created. Otherwise, for each clusterheadin candidate2-clusters list, the peer will calculatefor them W as function 3. The peer chooses theclusterhead with max value of W and sends to it Joinmessage.
The first method avoids having a large number of clusterswith a minimum number of members, i.e. it gives an equilib-rium between the number of clusters founded and the numberof members in each cluster. The second one improves stabilityof network.
A simple example of a new peer join is illustrated in Figure2. The new peer e chooses the peer h and c as neighbors(Figure 1.a). The physical hop count to clusterhead a andclusterhead g is lower or equal to Rc. According to the firstmethod (Figure 1.b), the new peer e calculates the W of each
clusterhead. The W of a is 2.5 and W of g is 1. As the W ofa is greater than W of g, the new peer e selects cluster of a tojoin (cluster 1). According to the second method (Figure 1.c),the new peer e compares its AVAIL with AVAIL of a and g.As AVAIL of new peer e is greater than AVAIL of each one,peer e becomes clusterhead and new cluster is created (cluster3).
When the new peer becomes clusterhead, it sends cluster-info message with its IP address and its AV AIL to notifynormal peers about its existence. The new clusterhead mustfind other clusterheads in the near vicinity. Each cluster-head periodically sends a Heartbeat message to its neighborclusterheads to maintain connectivity and to its member toinform them that is alive. Each peer also periodically sends aHeartbeat message to its clusterhead to inform it that it is alive.If the peer shares a new files, it sends information about newfiles to its clusterhead by piggybacking them to the periodicHeartbeat message.
E. Clusterhead leaving process
According to the method of clusterhead selection, there aretwo methods of re-clustering when the clusterhead leaves thenetwork.
1) The first method: there are two cases
• voluntary: The current clusterhead selects its succes-sor from its members. The successor is the physicallyclosest member. After that, the current clusterheadsends a leaving message to its members and itsneighbor to inform them about its leaving. The newclusterhead sends cluster-info message to notify nor-mal peers about its existence. Each peer receiving thismessage, sends Join message to this new clusterhead.Some members do not receive this message becausethe physical distance that separates them compared tothe new clusterhead is surely greater than Rc. In thiscase, they must search a new cluster to join.A simple example of clusterhead leave is illustratedin Figure 3. According to the first method, whenclusterhead a disappears, it chooses peer m to becomeclusterhead because it is the closest to it and it has
Fig. 3. An example of clusterhead leave.
the largest value compared to peerc. Peer m broad-casts cluster-info message. Peer v do not receive thismessage because it is to 3 physical hops compared tonew clusterhead m. Peer v must find new cluster tojoin. In this example, it joins the cluster 2.
• involuntary: If each member does not receive aHeartbeat message from its clusterhead after a periodof time (e.g. three times the period of transmission ofthat message), the member assumes that the cluster-head is disconnected and searches a new clusterhead.
2) The second method: there are two cases
• voluntary: From the cluster-member-table, the currentclusterhead informs a member which has the largestAVAIL to be the new clusterhead. After that, it sendsa leaving message to its members and its neighborsclusterhead to inform them about its leaving. Thenew clusterhead sends cluster-info message to notifynormal peers about its existence. Each peer receivingthis messages, sends Join message to the new clus-terhead to join its cluster if its AVAIL is lower thanthe AVAIL of this new clusterhead, else it must searchanother clusterhead. Some members do not receivethis message because surely the physical distance thatseparates them compared to the new clusterhead isgreater than Rc. In this case, they must search a newcluster to join.As shown in Figure 3.c, clusterhead a chooses peer bto become clusterhead because it has the largest AVAILvalue compared to other members. Peer v does notbecome a member of cluster 1 because it is to 3 hopscompared to new clusterhead b. Peer v must find newcluster to join. Since there are no clusters which havethe AVAIL value of its clusterhead greater than to theAVAIL value of peer v. Peer v becomes a clusterheadand generates a new cluster 3.
• involuntary: Is the same as the previous method.
F. Connectivity changes with clusterhead
Peer can change its clusterhead in two cases:
• First case: When the peer receives cluster-info mes-sage, according to the first method of clusterheadselection, it calculates the W of this new clusterhead.If its W is greater than that of its clusterhead, inthis case the peer informs its previous clusterhead toindicate that it will not be connected to it from thatmoment. If the second method of clusterhead selectionis applied, in this case the peer compares the AVAILof new clusterhead with the AVAIL of its clusterhead.If it is greater then the same process is applied.
• Second case: If the peer receives a Heartbeat messagefrom its clusterhead with a physical distance greaterthan RC, it leaves this cluster and searches for a newone.
IV. CONCLUSIONS AND FUTURE WORK
In this paper, we proposed a cross-layer clustering schemefor P2P Over MANET. CCSM groups nodes that are physicallyclose to each other into the same cluster. Two methods areproposed during clusterhead selection, the first is simple,decreases the overhead during the maintenance and givesa balance between the number of clusters founded and thenumber of members in each cluster. The second one decreasesthe failure rate of clusterhead which improves stability andthe performance of MP2P systems. CCSM reduces physicalnetwork traffic generated by overlay which enhances thesystem performance. Additionally, it balances load betweenclusterheads. In our future work, we will add various simula-tions to validate our methods.
REFERENCES
[1] “Gnutella.” [Online]. Available: http://www.gnutella.com/[2] “Kazaa.” [Online]. Available: http://www.kazaa.com[3] “Skype.” [Online]. Available: http://www.skype.com/[4] O. Waldhorst, C. Lindemann, and A. Klemm, “A special-purpose peer-
to-peer file sharing system for mobile ad hoc networks,” in Proc. IEEESemiannual Vehicular Technology Conference (VTC). IEEE, Oct. 2003,pp. 2758–2763.
[5] N. Shah and D. Qian, “An efficient overlay for unstructured p2p filesharing over manet using underlying cluster-based routing,” TIIS, vol. 4,no. 5, pp. 799–818, 2010.
[6] T. Zahn, R. Winter, and J. Schiller, “Simple, efficient peer-to-peer over-lay clustering in mobile, ad hoc networks,” in Networks, 2004.(ICON2004). Proceedings. 12th IEEE International Conference on, vol. 2.IEEE, 2004, pp. 520–524.
[7] C. Gottron, A. Konig, and R. Steinmetz, “A cluster-based locality-awaremobile peer-to-peer architecture,” in Pervasive Computing and Commu-nications Workshops (PERCOM Workshops), 2012 IEEE InternationalConference on. IEEE, 2012, pp. 643–648.
[8] E. Meshkova, J. Riihijarvi, M. Petrova, and P. Mahonen, “A surveyon resource discovery mechanisms, peer-to-peer and service discoveryframeworks,” Computer networks, vol. 52, no. 11, pp. 2097–2128, 2008.
[9] T. B. NOOR, M. R. SALEHIN, and S. R. ISLAM, “A clustering schemefor peer-to-peer file searching in mobile ad hoc networks,” internationaljournal of advanced research in computer and communication engineer-ing (ijarcce), vol. 1, pp. 487–493, October 2012.
[10] J. S. Han, K. J. Lee, J. W. Song, and S. B. Yang, “Mobile pee-to-peersystems using super peers for mobile environments,” in InformationNetworking, 2008. ICOIN 2008. International Conference on. IEEE,2008, pp. 1–4.
[11] H. Li, K. Bok, Y. Park, and J. Yoo, “An efficient mobile peer to peerarchitecture in wireless ad hoc network,” in Convergence and Hybrid In-formation Technology, ser. Lecture Notes in Computer Science, G. Lee,D. Howard, and D. Slezak, Eds. Springer Berlin Heidelberg, 2011,vol. 6935, pp. 1–8.
Toward a match between the P2P overlay and theMANET underlay
Manel SeddikiUniversity of Sciences and Technology Houari Boumediene
Computer Science Department, LSIAlgiers, Algeria
Email: [email protected]
Mahfoud BenchaıbaUniversity of Sciences and Technology Houari Boumediene
Computer Science Department, LSIAlgiers, Algeria
Email: [email protected]
Abstract—In recent years, many researchers have focusedon the deployment of P2P overlays over MANETs motivatedby the common characteristics shared between them such asdecentralization, self organization and dynamic topology. How-ever, conventional P2P overlays are usually constructed withoutconsidering the network underlay. This causes a considerablemismatch between the P2P overlay and the network underlay andleads to inefficient paths between P2P neighbors and unnecessarynetwork overhead. In this paper, we propose to build an efficientunstructured overlay with is aware of the MANET underlay.The main of our proposal is to select and maintain physicallynearest and fresh P2P neighbors. The combination of elementarystrategies makes our overlay building likely-matches the MANETunderlay. Our initial simulation results show that our approachperforms better than an overlay which is unaware of the networkunderlay.
Keywords—P2P overlay, mobile ad-hoc networks, physical prox-imity, topology construction, overlay build, cross-layer
I. INTRODUCTION
Peer-to-peer (or P2P) paradigm has been firstly known inthe internet. The basic idea is linking entities called peersin order to exchange ressources and services without usingany other intermediate entities.The P2P overlay is a set ofpeers which are both clients and servers and it is used formany applications such as file sharing, streaming and parallelcomputing.
P2P overlays are divided into two categories: Structuredoverlays and unstructured overlays. In structured overlays,the P2P overlay is structured into specific topologies suchas ring or tree and the information about the resources isdistributed among the peers in a specific way. This kind ofoverlays relies on a technique called DHT (distributed hashtable) [1], [2] and [3] where a key is assigned to each peerand each resource in the overlay. Each peer is responsibleof a set of resources whose keys are closest to this peer’skey. The resources are stored as <key,value> entries. Thekey represents the resource and the value represents resourcesprovider. Searching for a resource is like to search for thepeer responsible of the resource. Nevertheless, unstructuredoverlays are spontaneous systems where the overlay is builtrandomly and the information about resources is unknown.Flooding-based techniques [4] and [5] are used to lookup forthe resources.
On the other hand, mobile ad-hoc network (or MANETs)is a form of spontaneous communication and access to in-
formation for mobile users. This network consists of a largepopulation, relatively dense of mobile units which move freelyin a given territory. The only way for these mobile units tocommunicate is the use of wireless interfaces, without usingany pre-existing infrastructure or centralized administration.
In recent years, many researchers have focused on thedeployment of P2P overlays over MANETs motivated bythe common characteristics shared between these two sys-tems such as decentralization, self organization and dynamictopology. However, classical P2P overlays such as Chord,Can or Gnutella are unaware of the network underlay (wiredand wireless) because they are completely independent ofits characteristics. This means that the peer makes a totalabstraction of the lower layers characteristics and constructsits neighborhood according to the upper layers characteristicsonly. In structured overlays like Chord for example, the ringoverlay is built only according to ID assignation and inunstructured overlays like Gnutella for example, the overlayis built randomly. Thus, there is a mismatch between theoverlay and the network underlay specially for the wirelessunderlay such as MANETs because this mismatch generatesa significant traffic redundancy and overhead which stronglyaffects the resources of the mobile nodes such as energy.
In this paper, we are focused on unstructured overlays ontop of MANETs because they share the same spontaneousnature and this makes unstructured overlays more flexible tobe close to the MANET underlay. To have a better view ofthe mismatch issue, Figure.1 illustrates an unstructured P2Poverlay and its corresponding mobile ad-hoc underlay. Asseen, node A has two physical neighbors B and D within itsunderlay communication range and three P2P neighbors L, Iand C in the overlay. The overlay path from A to D is A-C-Dand the physical path corresponding to it is A-B-C-B-D. Thisphysical path is not efficient because it generates unnecessaryand redundant traffic as D is direct physical neighbor of A anda preferred path is A-D. So, it is more interesting if D wasP2P neighbor of A.
In the main, the redundant traffic can be greatly reducedif the peer establishes neighborhood relationship with thephysically close surrounding peers instead of random peers.Considering the physical proximity of peers makes the P2Poverlay close to the physical underlay and the mismatchbecomes less important. In this paper, we propose to buildan unstructured P2P overlay that likely-matches the MANETunderlay. The main idea is to select and maintain physically
Fig. 1: Mobile ad-hoc underlay VS P2P overlay
nearest and fresh P2P neighbors.
The rest of this paper is organized as follows: Section IIdescribes some underlay-aware overlay techniques. Section IIIpresents our approach and explains all details related to ourP2P overlay construction such as Neighborhood relationshipestablishment, neighborhood selection, exchanged messagesbetween neighbors, cache structures and so on. Section VIpresents our initial obtained results while comparing ourapproach with another one which is unaware of the networkunderlay. We finalize this paper with expectations and futurework.
II. RELATED WORK
An efficient P2P overlay construction has to consider thefollowing points:
• Overlay creation: It consists on explaining how thepeer establishes neighborhood relationship? Whendoes the peer have to send neighborhood request?What happens when a new peer join the P2P overlay?How does the peer select its neighbors? And so on.
• Overlay maintenance: It consists on explaining howdoes the peer maintain its neighborhood relationshipduring its participation in the P2P overlay and howdoes the overlay system act when a peer leaves theP2P overlay (voluntarily or not).
• Underlay awareness: An efficient P2P overlay has tobe aware of the network underlay to optimize P2Ppaths and to reduce network network.
Several underlay-aware techniques have been proposed inthe literature. Most of them exploit the physical proximity
during the overlay building process or the P2P routing process.Landmarking [6], [7] and [8] is one of the most importanttechniques to detect physical proximity of peers in structuredoverlays. It consists on fixing some peers called landmarks inthe P2P network as references for the other peers. Each peermeasures its RTT (Round time trip) to those landmarks andsorts them according to this parameter. then, it assigns an IDwhich is close to the ID of the landmark with the smallest RTT.Peers with the similar landmark ID are physically consideredclose to each other. However, conventional landmarking arespecially designed for fixed networks and suffers from thefixed landmarks constraint witch makes them not suitablefor a dynamic underlay. [9] and [10] cope with this andpropose two approaches: the Random landmarking (RLM) andthe closest neighbor prefix assignment (CNPA). The RandomLandmarking replaces fixed landmarks by dynamic landmarks.A new peer locates some peers called temporary landmarkswhich are responsible of a fixed set of landmark keys then,it measures the distance to those temporary landmarks andassigns itself an ID based on its landmark ordering. In closestneighbor prefix assignment (CNPA) technique, the ExpendingRing Search [5] is used to lookup for physically near peersthen, it assign itself an ID which is close to the near peer’sID. The Expanding Ring search consists on broadcasting arequest with a small Time To Live value at the beginningand waiting for reply. If the reply is not received after atime threshold, the request is sent again with an increasedTime To Live value and so on until a reply is received. Whenthe reply is received, the sender is considered as physicallyclose peers. Conventional and random landmarking are moresuitable for structured overlays because they are based on IDassignation to keep the physical proximity information unlikeERS technique which can be used both for structured andunstructured overlays.
In [11], a MANET-aware bootstrapping process is pro-posed. Instead of discovering near peers by the ERS, they usedmulticast mechanism to discover the peers then, selects nearpeers for neighborhood request. When a peer is discovered, itis recorded with other information such as its physical distancefrom the current peer obtained from the MANET routingprotocol and its remaining energy in the cache. Discoveredpeers are sorted according to utility parameter in the cachewhich is calculated based on physical distance and energy.When the peer wants to establish neighborhood relationshipwith the discovered peers, it selects the n top peers from thecache and unicasts a neighborhood requests to them. Utility uis calculated according to Cobb-Douglas utility function [12]represented as u= hα ∗ e(1−α) . h is the physical distancebetween the corresponding peer in the cache entry and the peerholding the cache, e is the remaining energy of correspondingentry and α is a parameter indicating the preference of a nearerpeer to a longerlived one. When α = 1, this means that thepeer want to select physically nearest neighbors and when α=0, this means that the peer prefers longer-lived neighbors.
The bootstrapping process is performed with the consid-eration of the MANET underlay but the maintenance of theP2P overlay is not addressed. This is an important phase ina mobile environment due to mobility of nodes and frequentchanges. Moreover, this approach provides physical distancefrom the MANET routing protocol which means that onlyproactive routing protocols are suitable.
[13], [14], [15] and [16] proposed to build an unstructuredoverlay over MANET underlay and they used the ERS todetect physically nearest peers then, they apply a minimumspanning tree algorithm [17] to improve the overlay in orderto match it with the MANET underlay. A root peer is usedas a reference and it is connected to all the other peers. Eachpeer maintains a file-cache for its 1-hop away p2p neighborsand its 2-hops away p2p neighbors. When a peer receivesthe discovery request, it replies to P and provides its P2Pneighbors. When P receives a reply, the replys sender andits P2P neighbors are considered as Ps surrounding peersthen, neighborhood requests are sent to them. When thosesurrounding peers reply to P, it records them as 1-hop awayP2P neighbors and records their neighbors as 2-hop awayP2P neighbors. Physical distances between P and its 1-hop-away is obtained from the routing agent and the physicaldistance between P and its 2-hop away P2P neighbors areeither provided by the MANET routing protocol in case ofreactive routing or calculated as follows in case of proactiverouting:
Dp−s = min(Dp−q +Dq−s) (1)
Where Dp−s is the distance between P and its 2-hop awayneighbor S , Dp−q is the distance between P and its 1-hop away neighbor Q, Dq−s is the distance between Q andits 1-hop away neighbor S. P builds an undirected weightedconnected graph consisting of P, P’s direct neighbor peers andP’s 2-hops away neighbor peers. Using this graph, MinimumSpanning Tree (MST) algorithm with P as a source is appliedto remove redundant links and obtain new graph close to thephysical topology. For the maintenance, probe messages aresent between peers based on the root peer distance.
This approach may be efficient while using the MSTalgorithm. However, peers exchange too much messages forthe overlay building and the root peer has to be maintainedto keep coherence of the approach; this leads to unnecessaryand redundant communications between peers. Moreover, thecalculated distance of Equation(1) is not efficient and assumesthat the physical distance between a peer and its direct P2Pneighbors is provided by the routing agent; this is not consis-tent in case of reactive routing protocols.
We are interested on [14], [15] and [16] research worksbecause they focused on building a complete unstructuredoverlay over MANET. They proposed the overlay creation,the overlay maintenance and the MANET-aware concept. Weare going to use some of their principles to propose our newapproach which is efficient, more flexible and light-weight.
III. OUR PROPOSAL
The goal of our proposal is to build and maintain anunstructured P2P overlay which closely matches the MANETunderlay. The MANET is composed of mobile nodes whichmay join the P2P overlay at any time. Thus, two types ofnodes can be distinguished: ad-hoc node which is not part ofthe P2P overlay but it participates only in forwarding messagesand peer which is part of the P2P overlay. In the initial state,each joining peer is blind and has to discover and maintain itsneighbors according to the MANET underlay. Considering thephysical underlay while building the overlay allows to likely-match between the two topologies.
TABLE I: Messages of our proposal
Message name SignificanceNeighbor-request Neighborhood lookup+ relationship requestNeighbor-reply Neighborhood relationship replyHello Maintain connectivity with neighbors (heart beat)Hello-reply Reply to Probe messageNeighbor-remove Remove neighbor from list of neighbors
TABLE II: Technical terms of our proposal
Term SignificanceMin-neighbors Minimum number of neighbors for a peern Maximum number of neighbors for a peerNum-neighbors Current number of neighbors for a peerAd-hoc node Mobile node which is not part of the P2P overlayPeer Mobile node which is part of the P2P overlayNeighbor The overlay neighborPhysical neighbor The ad-hoc neighbor
Each peer has a list of neighbors. Each entry of this listis associated to a neighbor and contains its IP address, itsphysical distance from the current peer and its state which willbe used in the overlay maintenance strategy. Peers exchangedifferent types of messages defined in Table I during theoverlay creation and maintenance.
Our proposal is cross-layer which means that the over-lay interacts with the MANET underlay to exploit physicalproximity of nodes. The proposed approach includes bothoverlay creation and maintenance. During the overlay creation,the peer discovers the physically nearest peers expressed byphysical hops and asks for neighborhood relationship at thesame step. It receives also neighborhood requests from theother peers. In order to cope with the dynamic nature ofthe MANET underlay, a maintenance strategy is proposed tokeep the overlay updated and a neighbor selection strategyis designed so that each peer always favors the physicallynearest peers as neighbors. To make our proposal more flexible,the physical hop between two peers is independent from theMANET routing protocol. It is calculated from the messages’paths. For a better understanding of the next sections, we definesome terms in Table II.
1) Neighborhood relationship establishment: In [11], [14],[15] and [16], the peer discovers physically nearest peers ina first step then, it asks for a neighborhood relationship ina second step. This leads to waste of time and increasesnetwork traffic. In our proposal, the joining peer discoversthe physically nearest peers and asks for a neighborhoodrelationship in the same step in order to remove redundancyand gain time. This is done as follows: the peer floods aNeighbor-request in the MANET to lookup for the physicallyclosest peers. When the Neighbor-request is received by anad-hoc node, it simply forwards the packet to all its physicalneighbors if TTL is not expired. When the Neighbor-requestis received by a peer, it checks if it can accept the sender asneighbor. If so, it adds the sender to its neighbors and sendsNeighbor-reply to the sender according to the MANET routingprotocol.
When the Neighbor-reply is received by an ad-hoc node ora peer which are not the destination, they simply forward itto the next hop according to the MANET protocol. When the
Fig. 2: Schemes of neighborhood relationship establishmentbetween P1 and P2
Neighbor-reply is received by the destination peer, it checks ifit can accept the sender as neighbor. If so, it adds the senderto its neighbors. Otherwise, it sends Neighbor-remove to thesender according to the MANET routing protocol.
When the Neighbor-remove is received by an ad-hoc nodeor a peer which are not the destination, they simply forward thepacket to the next hop. When the Neighbor-remove is receivedby the destination peer, it removes the sender from its list ofneighbors.
Figure.2 briefly summarizes the neighborhood relationshipestablishment between two peers.
2) Neighbor selection strategy: The overlay connectivitymay frequently change and the list of neighbors becomes out ofdate. The neighbor selection strategy is dynamic and designedsuch a way that each peer favors the physically nearest peers asneighbors. Each peer P can accept only n neighbors. When Preceives a Neighbor-request, it checks if there is room in its listof neighbors (n neighbors is not reached). If so, the requesteris directly accepted as neighbor. Otherwise, it checks if therequester is physically closer than a subset of its neighbors.If so, the requester is accepted and the physically furthestaway neighbor in the subset is replaced by it. Otherwise,it is rejected. The replaced neighbor is prevented by P forchanges through Neighbor-remove message so it removes Pfrom its list of neighbors. To understand this strategy, we givea scenario where P is a peer which has three P2P neighbors:P1 with physical distance of 5, P2 with physical distance of2 and P3 with physical distance of 4. Let’s put n representingthe max neighbors equal to 3. P receives a Neighbor-requestfrom peer P4. P can know the physical distance of P4 fromthe Neighbor-request’s path length, then assuming that P4 hasphysical distance of 3. P can’t directly accept P4 as neighborbecause it reaches three neighbors. However, P4 is physicallynearer to P than P1 and P3 so the furthest away neighborwhich is P1 here will be replaced by P4 and the new P’s listof neighbors P becomes: P2 with physical distance of 2 , P3with physical distance of 4 and P4 with physical distance of 3.P sends a Neighbor-remove message to P1 so that P1 removesP from its list of neighbors.
3) Overlay maintenance strategy: The overlay maintenanceis very costly for the network but essential to keep the overlay
consistency. Our maintenance strategy tries to balance betweencost and consistency in order to keep a fresh connectivitywithout increasing the network traffic. Neighbors periodicallyexchange Hello messages between them to maintain connectiv-ity and detect involuntary peer departures or moves. However,the messages redundancy can be avoided as in [14], [15] and[16] where the peer don’t necessary send Hello messages toall its neighbors but only to peers which are closer to the rootpeer then it but this strategy presents a drawback; the networktraffic gained will be lost on root peer maintenance. In ourproposal, a root peer is not used. Instead, a state field is storedin the list of neighbors of the peer P. The state value is 1 or 0 :1 indicates that the neighbor is originator of the neighborhoodrequest to P and 0 indicates the opposite. The strategy worksas follows: according to this state field, the peer sends Hellomessages only to its neighbors to whom it sends neighborhoodrequest (state=1) and receives Hello messages from neighborswhich sent neighborhood request to it (state=0). We give anexample: P receives neighborhood request from P1 and Paccepts then, P adds P1 as neighbor with a state=0 becauseP1 is the originator of the neighborhood request. P sends aneighborhood request to P2 and P3 and they accepts so, Padds P1 with state=1 and P2 with state=1 because P is now theoriginator of the neighborhood request. For the maintenance,P will send Hello messages only to P1 and P2 (because theirstate=1) and receives HELLO messages from P1 (because itsstate=0). As it is a distributed system, we expect that P1 forexample sends HELLO message to P because it adds P withstate=1.
Each peer that receives the Hello message from its neighborhas to reply to it by Hello-reply to confirm its existence.Otherwise, it will be removed from the Hello sender’s listof neighbors after a number of retries. When a peer, don’treceive Hello message from a neighbor with state=0 aftera time threshold, this neighbor is removed from the list ofneighbors.
In case of voluntary departure, the peer sends a Neighbor-remove message to all its neighbors to prevent them so theycan remove it from their list of neighbors.
Maintaining connectivity is not enough when the networkunderlay consists of mobile nodes because the physical dis-tance may changes. In our proposal, the physical distance isstored in the list of neighbors and has to be updated. Thephysical distance of a neighbor is initially defined as the lengthof the Neighbor-request’s physical path then; a regular updateis performed by exploiting the Hello messages. Each time apeer receives a Hello message from one of its neighbors, thenetwork layer sends the length of the Hello’s physical path tothe application layer. The physical distance of the concernedneighbor is accordingly updated.
A. Discussion
Our proposal is an MANET-aware overlay which usesphysical proximity of peers to build an unstructured overlayclose to the MANET underlay. Neighborhood discovery andrequest are performed at the same time to gain time and reducenetwork traffic. Neighborhood selection strategy ensures thateach peer keeps only physical nearest peers as neighbors andoverlay maintenance strategy ensures that each peer keeps
TABLE III: Simulation parameters
Parameter ValueNumber of mobile nodes in the P2P network 50Minimum number of neighbors 02Maximum number of neighbors 05Hello Message load 0.2/sDefault physical ttl 10Request frequency uniform(0.0, 10.0)Mobility model Random Way PointManet routing protocol AODV
Fig. 3: Overhead
always fresh neighbors. Combining all these elementary strate-gies provides an efficient light-weight overlay building thatlikely-match the MANET underlay.
IV. SIMULATION
Simulations are done on omnet++ [18], a discrete eventsimulation environment which provides both Ad-hoc modeland P2P model. The ad-hoc network is composed of 50 mobilenodes which may join and leave the P2P network at any time.The reactive routing protocol AODV [19] is used here butany other MANET routing protocol can be used. Simulationparameters are defined in Table III. When a mobile nodejoins the P2P network, it becomes a peer and starts to lookupfor neighbors according to the neighborhood relationship es-tablishment explained above (Section III.1). For our initialsimulations, we firstly compare our overlay construction withmGnutella, an unstructured overlay which is unaware of theMANET underlay [20] to show the impact of the mismatch onMANET performances. We define the peer ratio parameter asthe ratio between number of peers and number of all nodes inthe ad-hoc network. We vary the peer ratio in the networkand analyze three metrics: the network overhead generatedduring the overlay construction which represents total numberof packet sent for each peer, the average number of neighborsper peer and the average physical distance (physical hops)between two neighbors in the overlay.
Figure.3 presents the obtained results for the overheadanalysis. It is shown that the overhead increases both for ourapproach and mGnutella when the peer ratio increases becausemore relationship establishments and updates between peersare done with the increase of peers in the network. However,our approach generates less overhead than mGnutella. Thisis because in mGnutella, the routing protocol broadcasts adiscovery route request for each target peer and waits for thediscovery route reply before sending a join request to it while
Fig. 4: Average physical distance between two neighbors
Fig. 5: Average number of neighbors per peer
in our approach; the same message is sent to discover manysurrounding peers and routes at the same time. Moreover, fewermessages are used in our approach to maintain connectivitythanks to our maintenance strategy.
It is demonstrated in Figure.4 that the average physicaldistance in our approach is lower than one of mGnutellafor all ratio peers. This is because our replacement strategyis designed so that peers keep only nearest neighbors. Theaverage physical distance of our approach decreases when thepeer ratio increases because the probability of finding nearestneighbors is high when there is more peers in the network. FormGnutella, the peer ratio variation doesnt have an impact onthe average physical distance because the peers are selectedrandomly.
Figure.5 shows that the average number of neighbors perpeer increases both for our approach and mGnutella whenthe peer ratio increases. This is because more replies aresent to neighborhood requestors with the increase of peersin the network. However, the average number of neighborsin our approach is higher than mGnutella which implies thatour approach have better neighborhood reply rate. The reasonis that in a mobile environment, packets are lost more oftenwhen the physical path is long. With the random selection ofmGnutella, neighbors may be physically far away from therequestor which implies a long path contrary to our approachwhere short paths are used because only physically nearestpeers are targeted.
From these initial results, it is shown that our approachperforms better than the random overlay construction ofmGnutella in the mobile ad-hoc network. Indeed, Our approachallows to discover fresh and physically near neighbors while
keeping the overhead reduced.
V. CONCLUSION
The mismatch between the P2P overlay and the MANETunderlay is a great challenge to tackle. In this paper, weproposed to build an MANET-aware overlay. Our cross-layerdesign and the combination of efficient elementary strategiesmakes the P2P overlay likely matches the MANET underlay.The intial simulation results show that our approach per-forms better than mGnutella. This reinforces our proposaland prompts us to investigate more in simulation and resultsanalysis. In the next contribution, we are going to compare ourresults with [14], [15] and [16] and study MANET mobilityimpact on our proposal.
REFERENCES
[1] I. Stoica, R. Morris, D. Karger, M. F. Kaashoek, and H. Balakrishnan,“Chord: A scalable peer-to-peer lookup service for internet applica-tions,” in Proceedings of the ACM SIGCOMM ’01 Conference, 2001.
[2] A. Rowstron and P. Druschel, “Pastry: Scalable, distributed object lo-cation and routing for large-scale peer-to-peer systems,” in Proceedingsof ACM/IFIP Middleware, 2001.
[3] B. Y. Zhao, J. D. Kubiatowicz, and A. D. Joseph, “Tapestry: Aninfrastructure for fault-tolerant wide-area location and routing,” inTechnical Report, 2001.
[4] M. Ripeanu, A. Iamnitchi, and I. Foster, “Mapping the gnutella net-work,” Internet Computing, IEEE,Volume 6, pp. 50–57, 2002.
[5] J. Deng and H. Zuyev, “On search sets of expanding ring search inwireless networks,” Journal of Ad Hoc Networks, vol. 6, no. 7, pp.1168–1181, 2008.
[6] T. S. Eugene and H. Z. Ng, “Towards global network positioning,” inACM SIGCOMM Internet Measurement Workshop, 2001.
[7] S. Ratnasamy, M. Handley, R. Karp, and S. Shenker, “Topologically-aware overlay construction and server selection,” 2002.
[8] Z. X. Chunqiang and T. Z. Zhang, “Building topology-aware overlaysusing global soft-state,” in Proceedings of FTDCS, 2003.
[9] R. Winter, T. Zahn, and J. Schiller, “Topology-aware overlay construc-tion in dynamic networks,” in Proc. IEEE ICN04, 2004.
[10] R. Winter and al, “Random landmarking in mobile, topology-awarepeer-to-peer networks,” in Proceedings of FTDCS, 2004, pp. 319–324.
[11] A. Mawji and H. Hassanein, “Implementation of bootstrapping for p2poverlays in manets,” in Communications (QBSC), 2010 25th BiennialSymposium on, May 2010, pp. 14–17.
[12] H. Varian, Intermediate Microeconomics: A Modern Approach, 2003.[13] N. Shah and D. Qian, “A new cross-layer unstructured p2p file sharing
protocol over mobile ad hoc network,” in Proceedings of the interna-tional conference on Advances in computer science and informationtechnology, 2010.
[14] N. Shah and al, “An efficient overlay for unstructured p2p file sharingover manet using underlying cluster-based routing,” KSII Transactionson Internet and Information Systems, vol. 4, no. 5, pp. 799–818, 2010.
[15] N. Shah, “An efficient unstructured p2p overlay for file sharing overmanets using underlying reactive routing,” Journal of Internet Technol-ogy (SCIE), vol. 12, no. 3, pp. 517–530, 2011.
[16] N. Shah and D. Qian, “An efficient overlay for unstructured p2p filesharing over manet using underlying proactive routing,” in Proceedingsof the seventh International Conference on Mobile Ad-hoc and SensorNetwork (MSN), 2011, pp. 248–255.
[17] R. L. Graham and P. Hell, “On the history of the minimum spanningtree problem,” Annals of the History of Computing, vol. 7, no. 1, pp.43–57, Jan 1985.
[18] (2001) Omnetpp. [Online]. Available: www.omnetpp.org[19] C. Perkins and E. Royer, “Ad-hoc on-demand distance vector routing,”
in Mobile Computing Systems and Applications, 1999. Proceedings.WMCSA ’99. Second IEEE Workshop on, Feb 1999, pp. 90–100.
[20] D. N. da Hora, D. F. Macedo, L. B. O. c, I. G. Siqueira, A. A.Loureiro, J. M. Nogueira, and G. Pujolle, “Enhancing peer-to-peercontent discovery techniques over mobile ad hoc networks,” ComputerCommunications, vol. 32, no. 13-14, pp. 1445–1459, 2009.
Green Power Management For Wireless Mesh
Networks
Sarra Mamechaoui
STIC Laboratory
Abou Bekr Belkaid University
Tlemcen, Algeria
tlemcen.dz
Fedoua Didi
Dept of Computer engineering
Abou Bekr Belkaid University
Tlemcen, Algeria [email protected]
tlemcen.dz
Guy Pujolle
Pierre et Marie Curie University
Paris 6, FRANCE
Abstract—Wireless mesh networks are the next step in the
development of wireless architecture, delivering wireless services
for a large variety of applications in personal, local, campus, and
metropolitan. The recent years have seen a rise in energy
consumption in communication technologies and information.
Obviously, energy efficiency solutions are becoming more vital
for both local access networks and wireless networks. Green IT
has attracted considerable attention newly. However, the
application of green network of wireless mesh networks (WMN)
has seldom been described in the literature. In this paper, we
propose a novel green approach for power management in WMN
to enhance energy efficiency and network operation. We suggest
energy-aware design and energy-aware routing protocol to study
the performance of WMN over dynamic traffic profiles. The
combination of this two Algorithms point that, with suitable
design parameters (number of shutdown Mesh Routers, average
path length), we can efficiently reduce energy consumption in
WMN without significantly impacting the network performance.
Keywords-WMN; Energy Saving; Power Management; Routing;
Sleep State; Link State.
I. INTRODUCTION
Reducing unnecessary energy consumption is becoming a major concern in a wireless network, because of the potential economic benefits and its impact on the environment expected [1]. Indeed, the Information and Communication Technology (ICT) consumes only 3% of the worldwide energy consumption and CO2 emissions is approximately 2%, which is the same as the emission aircraft [2]. Access networks represent 10% of this amount. To solve this difficult problem, energy-efficient communication has emerged as a promising solution. Green computing aims to reduce the environmental footprint, economic and social information and communication technology (ICT). Green networking solution provides global environmental network architectures to improve energy efficiency, performance and reduce the cost savings, it also allows companies to gain competitive advantage and maintain a sustainable environment. To solve this problem, it is essential that we aim to improve and sustain performance of ICT and minimize their energy consumption and carbon footprint. An overall solution to energy-efficient ICT should take into consideration [3].
The network devices connected to the Internet has a significant impact of energy costs. It has been considered that
the access networks consume approximately 70% of energy costs of global telecommunications networks and this percentage should increase over the next decade [4]. This is mainly because the wireless access networks (WANs) are generally dimensioned to meet the demands in terms of peak traffic constraints, resulting in over-provisioning in low-demand periods wasting a significant amount of power. In this respect, trying to minimize the energy consumption of deployed access elements is an important goal.
The objective of our work is to associate the flexibility of wireless mesh networks (WMN) with the need to optimize the energy consumption obtaining benefit of low demand periods and dynamic reconfigurations that are possible in WMNs. Some introductions on WNM are presented in section II and detailed discussion of the energy efficiency issues is beyond the scope of this paper.
This paper mainly focus on the power consumed by technology networking products, such as wireless access networks, communication protocols and routing protocols because these are the most widely utilized and deployed over the world as part of the Internet infrastructure. The key objective of the proposed work is to optimize network energy consumption by putting under-utilized MRs to sleep state and using the routing protocol which aims to use the already-used paths. The rest of the paper is organized as follows. In Section II, introduces wireless mesh network and their architecture. Section III, briefly reviews some related work on energy management in access networks that have been proposed to enable energy-efficient. Section IV; proposes an efficient algorithm to optimize network energy consumption by putting under-utilized MRs to sleep state. Section V, develops an energy-aware routing. Finally, we make some concluding remarks and future work in Section IV.
II. WIRELESS MESH NETWORK
As various wireless networks evolve into the next generation to provide better services, a key technology, wireless mesh networks (WMNs) has emerged recently [5], which is a variant of Mobile Ad hoc Networks (MANET), and is an IEEE 802.11-based infrastructure network made up of Mesh Routers (MRs) and mesh clients (MCs), where mesh routers have a minimum mobility and form the backbone of WMNs. The integration of WMNs with other networks such as
the Internet, cellular, IEEE 802.11, IEEE 802.15, IEEE 802.16, sensor networks, etc., can be achieved through the gateway and bridging functions in the mesh routers. Clients mesh may either be fixed or mobile, and can form a client mesh network between themselves and with mesh routers [6] as shown in Figure. 1.
Figure 1. Wireless Mesh Network architecture
However, WMNs infrastructure devices are always active. Thus, during lower traffic periods the energy consumption is the same as in busy hours, while it would be possible to save a large amount of power by turning off unnecessary nodes. Concerning this issue, an energy-aware approach for such kind of networks was tackled in Section III.
III. RELATED WORK
Face the fact that the cost of power continues to rise, and the necessity for providing broadband in rural areas, green network has become one of the most important research areas of industry ICT. To address this difficult challenge, effective communication power appeared to be a promising solution.
The wireless networking community has developed several techniques for wireless technologies with high energy efficiency. In the literature, different approaches have been proposed to reduce the power consumption in different kinds of access networks [7, 8]. A survey of energy-efficient protocols for wireless mesh networks can be found within [9]. Here, we concentrate on efforts aimed to enhance energy efficiency in access networks.
In this paper, we are facing the challenge of reducing the energy consumption in backbone networks through an approach in sleep mode. The intuition has already been reported in the literature, starting from the seminal work of Gupta et al. [10]. Particularly the approaches from the traffic engineering [11], that proposed to minimize the energy in a time varying context by selecting dynamically a subset of mesh routers to switch on considering coverage issues of the service area, traffic routing, as well as capacity limitations both on the access segment and the wireless backhaul links, to routing protocols [12] with it Optimal Green Routing and Link Scheduling, called O-GRLS, and efficient algorithm based on Ant Colony, called Ant Colony Green Routing and Link Scheduling (AC-GRLS), and novel architectures [13], the authors proposed a comparison between a sleeping-mode and a rate switch algorithm, both applied at a network infrastructure
level. In [14], the authors proposed a novel distributed algorithm, called GRiDA, to put into sleep mode links in an IP-based network.
In [4] paper, the authors developed energy-aware design techniques and routing protocols in hybrid wireless-optical broadband access network (WOBAN) green broadband access. They exploited to enable energy savings in the optical part but not in wireless part which is wireless mesh network. Adopting the same paradigm as in [4], in this paper, we propose energy-aware design and energy-aware routing protocol to study the performance of WMN over dynamic traffic profiles.
These studies deal with the reduction of energy consumption of the network by switching off elements, such as routers and links, and major savings can be achieved when the sleep mode States are exploited. Despite all these efforts, there remain major challenges to deploy energy efficiency in the access networks.
IV. ENERGY-AWARE DESIGN
1)- Energy Savings in WMN:
Various aspects of WMN should be considered for its energy-aware design. Present conception of WMN, deployment and management approaches provide fault tolerance, reliability and robustness along with a high degree of performance. Thanks to the mesh backbone, where the traffic can be rerouted through alternative paths in the case of failures such as the failure of a wireless router or gateway. At a specified moment, it is possible to find multiple WMN topologies which can meet the capacity requirements and reliability objectives. This is possible through the densely interconnected wireless mesh that has many redundant paths to send the traffic. The flexibility offered by the WMN can be exploited to allow energy savings.
A further significant feature of the access network is its traffic profile. The traffic load on the access network arriving directly from clients, and it is well known that there are daily fluctuations of this load. During WMN (as well as other network) deployment, the common practice is deploying network equipments so that they may support the peak traffic load. Consequently, during low load hours, some parts of the network can be underutilized.
So, to design WMN topologies with reduced power consumption, we must take to consideration the following points: First, WMN topology may anticipate multiple redundant paths for a packet to reach its destination; and the traffic load fluctuation during different hours of the day. Consequently, we can selectively put some nodes to sleep state during low load hours, thereby reducing network wide power consumption.
2)- How to Put a MR to Sleep State:
A general problem that we are considering aims to manage the mesh nodes in order to conserve energy when some of the nodes in the network are not required and can be deactivated. Energy efficiency is therefore calculated to be the number of nodes that may be turned off without compromising the network performance. This model of energy conservation is
called ON-OFF model, was first introduced by Restrepo et al. [15] and reused in the context of wireless networks in [16, 17]. Capone et al. [11] presented the basic planning model introduced in [18] and described how the model is modified to achieve a formulation of optimal planning who does not take into consideration the time variations of the application and the dynamics of coverage which are necessary for the effective operation systems of energy management. From an operational point of view, this may be easily integrated in centralized network management platforms currently used for carrier grade WMNs and to the centralized and remote control of all devices and their configuration.
Algorithm 1 Mesh Router Sleep Algorithm
Input: WMN topology, Low Traffic (LT), and High
Traffic (HT).
Output: Set of MRs that can be put to sleep state.
• Initialization: Initialize LT and HT.
• Measurement: At different hours of the day, the
centralized management quantifies traffic load at different
MRs by measuring the length of corresponding input
queues.
• Decision: MRs,
—If load < LT, put to sleep state MR.
—else if load > HT, keep MR active and turn on another
inactive MR.
—else keep MR active.
In WMN, the centralized network management platform can manage a centralize sleeping mechanism and maintain two different traffic profiles for the traffic load at MRs low traffic (idle) and high traffic (Busy). During different hours of the day, the centralized network management platform will observe the traffic load at different MRs by measuring the length of corresponding input queues. If a MR is operating under low traffic, MR can put sleep and the wireless mesh backbone will reroute the affected traffic due to MR shut-down to alternate paths. Otherwise put back sleeping MRs into active state when traffic load increases above high traffic in the currently active MRs.
V. ENERGY-AWARE ROUTING
Due to dynamic and Ad hoc topology in Wireless mesh networks the routing problem emerges as a challenging task in such networks [19]. Routing protocols provide the necessary paths through a WMN, so that the nodes can communicate on good or optimal paths over multiple wireless hops [20]. Only a few protocols have been developed specifically for WMNs. Several approaches have been considered. MIT designed new protocols tailored for WMNs for example Hybrid Wireless Mesh Protocol (HWMP) which is the default routing protocol for WLAN mesh networking, Microsoft mesh networks [21] are built based on dynamic source routing (DSR) [22], and many other companies are using ad hoc on-demand distance vector (AODV) routing [23]. Since WMNs share common
features with wireless ad hoc networks, the routing protocols developed for MANETs can be applied to WMNs.
Several routing protocols have been proposed for wireless network like GRiDA [14] and OLSR [24]. These algorithms are link-state (LS) protocols which are based on Shortest Path First (SPF) algorithm to find the best path to a destination also known as Dijkstra algorithm, since it is conceptualized by Dijkstra. Link-state routing always try to maintain full networks topology by updating itself incrementally whenever a change happen in network. Each router constructs a map of the complete network. Another approach that achieves a good performance load balancing that tries to use fairly all parts of the network. However it can result in an under-utilization of some segments of the network during times of low load. Over the times of low load, the traffic may be covered by using a small number of devices in the network. Our routing algorithm is an energy-aware LS protocol which aims to reduce the network wide energy consumption by putting underutilized nodes (mainly MRs) of the network to sleep. For routing the traffic, the goal will be to use the already-used paths. Thereby, MRs with no load can be put to sleep state. In addition, we can find some other MRs with very low load. Being more aggressive, we can set these MRs to sleep state and allow the wireless mesh reroute their traffic across other active MRs. When traffic load increases, sleeping MRs can be activated to carry the increased traffic. To achieve this, we can modify the LS routing algorithm in WMN so that link weights are assigned to satisfy our energy objective. So, we use residual capacity as the link weight. When traffic flows through a link, its next link weight will be the remaining capacity on that link. To route traffic from source to destination, we find the lowest residual capacity path. A formal description of the different steps of our Algorithm will be explained in the following:
•The first step the initialization: each MRs establishes a relationship an adjacency with each of its neighbors and for each link we attribute an initial capacity as residual capacity.
•The second step Link State Flooding: After the adjacencies are established, the MRs may begin sending out Link State Advertisement (LSAs). As the term flooding implies, the advertisements are sent to every neighbor. In turn, each received LSA is copied and forwarded to every neighbor except the one that sent the LSA and it advertise periodically the residual capacity as link weight and time stamp.
•The third step Link Weight Assignment: For each link, update the link weight found from the Link State Advertisement (LSA) by adding Hop Offset (HO) the purpose of this term is to reduce average path length.
•Finally the Path Computation step: After each MRs calculates its own path and find the lowest residual capacity path between source and destination finally update the residual capacity of links on the selected path.
For example, let us consider the small network in Figure. 2. The link metrics are their residual capacities. To send traffic from Source to Destination, energy-aware routing algorithm will route traffic through the path S-A-B-C-D, which has the lowest residual capacity (1+2+1+2+=6).
Figure 2. Residual capacity as link weights.
This approach, however, has its shortcomings. The algorithm selects the path with four hops although that is not the shortest path while using other metrics (such as hop length or delay). This will increase the average path length and path delay in the network. To deal with this problem, we can introduce a term called hop offset the purpose of this term is to reduce average path length. If we have a hop offset m, we add m to the path cost for each hop, i.e., for a path of n hops, the cost of the path will be: Residual capacity of the path + n x m
The example in Figure. 2, for a hop offset 1, the path costs are: 6 + 4 x 1= 10, 17 + 3 x 1= 20, and 11 + 2 x 1= 13 for paths S-A-B-C-D, S-E-F-D, and S-G-D respectively. The path which will be select is S-A-B-C-D. But, for a hop offset 3, S-G-D will be the chosen path in our algorithm. Selecting the optimal hop offset depends on how much delay the network connections can tolerate.
VI. CONLUSION
Energy consumption has become an important issue for the industry, economic and environmental. To solve this problem, it is essential that we aim to improve and sustain performance of ICT and minimize their energy consumption and carbon footprint. In this paper, we propose a green approach for power management in WMN to improve energy efficiency and network operation. The key objective of the proposed work is to aid the design of efficient algorithms to optimize network energy consumption by putting under-utilized MRs to sleep state and allows to easily determining which MRs to switch off, and therefore the energy-aware routing which aims to use the already-used paths. Thereby, MRs with no load can be put to sleep state and reduce average path length and path delay. The future work will focus, on the optimization of the numbers of sleeping MRs. Developing a mathematical model which will investigate the power consumption in WMN over dynamic traffic profiles. Along with analyze of the impact of energy-aware design on the performance of WMN.
REFERENCES
[1] A. Bianzino, C. Chaudet, D. Rossi and J. Rougier, ―A Survey of Green Networking Research‖, Communications Surveys & Tutorials, IEEE, décembre 2010 .
[2] R. Bolla, R. Bruschi, F. Davoli and A. Ranieri, ― Energy-Aware Performance Optimization for Next-Generation Green Network Equipment‖, PRESTO’09, August 21, 2009, Barcelona, Spain.
[3] B. Fargo, D. MacAvoy, ―A practical approach to greening the electronics supply chain results‖, Electronic industry citizenship coalition (EICC), 2009 June.
[4] P. Chowdhury, M. Tornatore, S. Sarkar, and B. Mukherjee. ―Building a green wireless-optical broadband access network (woban)‖. Journal of Lightwave Technology, 28(16):2219–2229, August 2010.
[5] I.F. Akyildiz and X. Wang, (2005) ―A survey on wireless mesh networks‖. IEEE communications Magazine 43(9), s23–s30.
[6] I.F. Akyildiz, X. Wang and W. Wang, (2005) ―Wireless mesh networks: a survey‖. Computer Networks Journal (Elsevier) 47(4), pp 445–487.
[7] S.Mamechaoui, F. Didi and G. Pujolle, ― Energy Saving For Wireless Mesh Network‖, MESH 2012 : The Fifth International Conference on Advances in Mesh Networks, IARIA 2012.
[8] S.Mamechaoui, F. Didi and G. Pujolle, ―Power Conservation for Wireless Mesh Network‖, ICM 2012, 24th International Conference on Microelectronics (ICM). 2012 IEEE.
[9] S.Mamechaoui, F. Didi and G. Pujolle, ―A Survey On Energy Efficiency For Wireless Mesh Network‖, International Journal of Computer Networks & Communications (IJCNC) Vol.5, No.2, March 2013
[10] M. Gupta and S. Singh, Greening of the Internet, SIGCOMM’03, August 25–29, 2003, Karlsruhe, Germany. Copyright 2003 ACM.
[11] A. Capone, F.Malandra, B. Sansò, ―Energy Savings in Wireless Mesh Networks in a Time-Variable Context‖. Mobile Networks and Applications, Volume 17, Issue 2, 2012.
[12] A. Amokrane, R, Langar, R, Boutaba and G. Pujolle, ―A Green Framework for Energy Efficient Management in TDMA-based Wireless Mesh Networks‖; IEEE/ACM CNSM 2012 ,Las Vegas, USA.
[13] S. Nedevschi, L. Popa, G. Iannaccone, ―Reducing Network Energy Consumption via Sleeping and Rate-Adaptation‖, Proceedings of the 5th USENIX Symposium on Networked Systems Design and Implementation (2008).
[14] A. Bianzino, L. Chiaraviglio, M. Mellia, ―GRiDA: a Green Distributed Algorithm for Backbone Networks‖ Online Conference on Green Communications (GreenCom), 2011 IEEE.
[15] J. Restrepo, C. Gruber, and C. Machuca, ―Energy profile aware routing,‖ in Communications Workshops, 2009. IEEE International Conference on, june 2009.
[16] S. Bhaumik, G. Narlikar, S. Chattopadhyay, and S. Kanugovi, ―Breathe to stay cool: adjusting cell sizes to reduce energy consumption,‖ in Proceedings of the first ACM SIGCOMM workshop on Green networking, ser. Green Networking ’10. New York, USA: ACM, 2010,
[17] A. de la Oliva, A. Banchs, and P. Serrano, ―Throughput and energy-aware routing for 802.11 based mesh networks,‖ Computer Communications, vol. 35, no. 12, 2012.
[18] E. Amaldi, A. Capone, M. Cesana, I. Filippini, and F. Malucelli. Optimization models and methods for planning wireless mesh networks. Computer Networks, 52(11):2159–2171, 2008.
[19] S.Mamechaoui, F. Didi and G. Pujolle, ―Energy Aware Cross Layers Design for Wireless Mesh Networks‖, ICNCER 2013, The First International Conference on Nanoelectronics, Communications and Renewable Energy 2013.
[20] Y. Zhang, J. Luo, and H. Hu, ―WIRELESS MESH NETWORKING Architectures, Protocols and Standards‖, 2007 by Taylor & Francis Group, LLC.
[21] J. Mitola III, ―Software Radio Architecture: Object- Oriented Approaches to Wireless System Engineering‖, Wiley Inter-Science, New York, 2000.
[22] D.B. Johnson, D.A. Maltz, Y.-C. Hu, ―The dynamic source routing protocol for mobile ad hoc networks (DSR)‖, IETF Internet-Draft: work in progress, July 2004.
[23] C. Perkins, E. Belding-Royer, S. Das, ―Ad hoc on-demand distance vector (AODV) routing‖, IETF RFC 3561, July 2003.
[24] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler, a. Qayyum et L. Viennot, ―Optimized Link State Routing Protocol‖ IEEE INMIC Pakistan 2001.
Reliable Hierarchical Cluster-Based Routing Protocol for Wireless Sensor Networks
Mohamed A. Eshaftri, Alsnousi Essa, Mamoun Qasem, Imed Romdhani, Ahmed Y. Al-Dubai School of Computing, Edinburgh Napier University, 10 Colinton Road, Edinburgh EH10 5DT, UK
Email: {M.Eshaftri ; A.Essa ; M.Qasem ; I.Romdhani; A.Al-dubai}@napier.ac.uk
Abstract- Wireless sensor networks (WSNs) have recently become
an integral part of a wide range of systems such as surveillance, target tracking, security and health management systems. WSNs deploy a large number of sensor nodes that cooperate to complete a sensing task. Large scale WSNs usually suffer from energy efficiency and quality of service issues. Although different hierarchical cluster-based approaches have been proposed by the research community, reliability, heavy load and load balancing are still open issues to be addressed. This paper proposes a new hierarchical and cluster-based topology formation and routing protocol. This protocol aims to guarantees reliability and avoids the heavy load issue by adapting the EIGRP wired protocol to the sensor context.
Keywords: Wireless sensor network, Cluster-based routing, load
balancing.
I. INTRODUCTION
Research on Wireless Sensor Networks (WSNs) has
attracted a lot of interest given their potential feature to digitise
the physical environment [1]. Although WSNs inherit a lot of
common features and attributes from traditional wireless
networks, they have unique constraints and issues. In general, a
WSN is composed of a large number of wireless sensor nodes
that may be densely or sparsely deployed in an indoor or an
outdoor hash environment. The sensor nodes have limited power
and processing capabilities and therefore they are highly
sensitive to failure. This failure leads to communication
interruption and unwanted and frequent network topology
changes. In fact, in a large-scale WSN scenario, the sensor
nodes that are located furthest from the base station (BS) require
relying on the efforts of a set of intermediate nodes to transfer
their data, or alternatively have to use high transmission power
to forward their data directly to the BS. Whatever approach is
adopted, the latency and power consumption of the entire
network will be affected [3][1]. Different solutions and
algorithms have been proposed by the research community to
overcome these issues. Among these solutions, hierarchical
cluster-based routing approaches are judged to be one of the
efficient solutions that can contribute to system scalability,
lifetime longevity and energy efficiency. In this type of solution,
the sensing field is subdivided into a number of administrative
entities called clusters [2] [3]. Each cluster has an organisation
leader or a root node called Cluster Head (CH). The primary
aim of the CH is to collect the data from its attached and
associated downstream nodes and forward it to the best
well-known hierarchical higher level upstream neighbour
(node) in a hop-by-hop manner until the data reaches the
processing point called the Base Station (BS). The BS may then
send the data to the end user located outside the sensing field.
This hierarchical and cluster-based approach may contribute to
solve many issues such as : reducing the total transmission
power, balancing the energy-exhausting load between all
nodes, reducing the bandwidth demand and efficient use of
limited channel bandwidth, reducing routing and topology
maintenance overheads, eliminating the redundant and highly
correlated data in aggregation process, reducing data collision
and interference in data transmission process, localising the
route setup within the cluster boundaries and thus generating
small-size routing tables, increasing the manageability and
scalability of the network.
In this paper, we propose a new hierarchical and
cluster-based protocol to optimise the topology formation and
achieve fair and reliable load balancing inside the sensor
network.
The rest of the paper is organized as follows. In Section II, we
review a set of up-to-date clustering algorithms proposed for
WSNs. Section III presents the specification of our new
protocol. In Section IV, we compare and contrast our protocol
with similar approaches with respect to common network
features. Finally, we present the future work and the limitations
of our protocol in section IV.
II. RELATED WORK
Different cluster-based approaches have been proposed by
the research community to address the challenging issues of
WSNs. Authors in [4] proposed the first well known clustering
protocol : Low Energy Adaptive Clustering Hierarchy
(LEACH). This protocol targeted to prolong the lifetime of
WSNs and reduce the energy consumption of sensor nodes.
From an algorithmic point of view, LEACH is a hierarchical,
probabilistic, distributed and single-hop protocol. It forms
clusters based on the received signal strength and CH nodes act
as default gateways to the BS as illustrated in Figure 1.
Figure 1. the basic topology of LEACH[5].
In LEACH protocol, nodes make autonomous decisions
without any centralised control. In addition, they have an equal
opportunity to become CHs. Initially a node decides to be a CH
by generating a random number between 0 - 1 and compares it
with a threshold value T (n), calculated by the Equation (1).
The nodes with a random number lower than T (n) then become
CHs. Each elected CH broadcasts an advertisement to the
non-CHs to form a cluster. A non-CH chooses the CH that can
be reached using the least communication energy.
(1)
Where is: P is the desired percentage of nodes to be cluster heads.
r is the current round. G is the set of nodes that have not been cluster heads during the last 1/P rounds.
Generally, LEACH provides a good model of energy
consumption as it gives sensor nodes an equal chance to
become a CH. Once chosen as a CH a sensor node cannot be
reselected in a subsequent round. Furthermore, LEACH avoids
unnecessary collisions from CHs because it uses the Time
Division Multiple Access (TDMA) protocol. Despite its
generally good performance, LEACH has also some clear
limitations. By using single-hop communication LEACH
cannot scale well and the probabilistic election of CHs may
lead to either a concentrations of CHs in one part of the
network or leaving some nodes without CHs in their
neighbourhood.
To address the shortcomings of LEACH with regard to the
CHs’ location and number, a centralised version of LEACH
called LEACH-Centralized (LEACH-C) was presented by
Heinzelman et al.[6]. In this protocol the BS takes the decision
regarding which sensor nodes become CHs and form a cluster.
Each node transmits information on its location and energy
level to the BS which then calculates the average energy level
of the network, and eliminates nodes with remaining energy
levels which are less than this average from selection as CHs
for that round. The centralised algorithm ensures that the
energy load is equally distributed among all the nodes by
selecting a predefined number of CHs and dividing the network
into optimum equally sized clusters. However, the formation of
clusters with an equal number of nodes in each cluster is not
guaranteed in this protocol, not least because it is not always
possible for nodes that are far from the BS to send information
on their status.
Another significant probabilistic clustering algorithm was
proposed by Bandyopadhyay et al. [7]. This protocol (Energy
Efficient Hierarchical Clustering (EEHC)) is a distributed,
randomized clustering algorithm for WSNs that aims to prolong
the network lifetime. The EEHC protocol consists of two
clustering stages: initial and extended. In the initial stage, also
called single-level clustering, each sensor node become CH
with probability p and advertises itself within its
communication range as a volunteer CH to its neighbor nodes.
Any node that receives such advertisement and is not itself a
CH becomes the member of the closest cluster. The node that it
is not within communication range of all volunteer CHs will
become a forced CH.
At the second stage, the same mechanism is extended from
bottom - up to allow multi-level clustering. The algorithm first
selects the level-1 CHs, then level-2 CHs, and so on. At the
level-1 CHs each sensor decides to become a level-1 CH with a
certain probability 1p and advertises itself as a CH to the
neighbor nodes within its radio range. This advertisement is
forwarded to all the sensors within 1k hops (where k is the
number of hops) of the advertising CH. Each sensor that
receives an advertisement joins the cluster of the closest level-1
CH, the remaining sensors become forced level-1 CHs. Level-1
CHs then elect themselves as level-2 CHs with a probability 2 p
and broadcast their decision of becoming a level-2 CH. This
decision is forwarded to all the sensors within 2k hops. The
level-1 CHs that receive the advertisements from level-2 CHs
joins the cluster of the closest level-2 CH. CH at level 3 are
chosen using similar mechanism. The algorithm ensures
multi-hop connectivity between CHs and the BS. In
inter-cluster communication this algorithm ensures that the
energy consumption by CHs that are located far from the BS is
reduced, because these CHs do not need to transmit directly to
the BS. However, the problem with this protocol is that the
close nodes to the BS consume more energy than the other
nodes do in the network, which causes an energy-hole problem.
Hybrid Energy-Efficient Distributed (HEED) clustering
was introduced by O. Younis et al. [8]. Here the authors
improve on the LEACH protocol by using two basic parameters
to elect the CHs. The first main parameter is the remaining
energy of each node, and intra-cluster “communication cost” as
a secondary clustering parameter. For example, cost can be a
function of neighbor proximity or cluster density calculated by
equation (2). Thus, unlike LEACH, in HEED the CH nodes are
not selected randomly. Only sensors that have high remaining
energy have the chance to become CH nodes. Also, the
probability of two nodes within each other cluster range are
both clusterheads is small. In comparison to the LEACH
protocol in the HEED protocol the CH nodes are well
distributed throughout the network. However, HEED is not able
to fix the cluster count in each round, the energy consumption
is not balanced because more CHs are generated than the
expected number and it also creates massive overheads because
of the multiple rounds.
(2)
Where is:
Cprob is an initial percentage of cluster heads among all n nodes. Eresidual is the estimated current energy of the node.
Emax is a reference maximum energy (corresponding to a fully charged battery).
Li Qing et al. [9] proposed a distributed multilevel
clustering algorithm for WSN (DEEC) to improve upon HEED.
In DEEC the CHs are selected by a probability based on
residual energy of each node and the average energy of the
network. The authors of this algorithm assumed that the nodes
have different amounts of energy. With these adaptive values
the sensor nodes decide probabilistically on their role in each
round. The main drawback of DEEC is that each node demands
global knowledge of the network, which increases the
overheads.
Hierarchical control clustering algorithm (HCC) [10] is
particularly useful for networks containing a large number of
nodes where scalability is the main area of concern. In these
applications, load balancing, energy efficiency and data fusion
are the main routing performance criteria. The HCC protocol
can reduce the energy consumption and provide scalability. It
was design to be effective in unicast, multicast and broadcast
communication environments. The algorithm proceeds in two
phases: the Tree Discovery and Cluster Formation. Firstly, the
tree discovery phase is basically a distributed creation of a
Breadth-First-Search (BFS) tree rooted at the initiator node.
Each node broadcasts the information about its shortest hop
distance to the root once every units of time. The neighbouring
node that chooses to be its parent will update its hop distance to
the root if the route through it is the shortest. Therefore, the
broadcast signal carries the parent ID, the root ID, and the sub
tree size. Each node updates its sub tree information when its
children sub tree size changes. Secondly, the cluster formation
phase starts when a sub tree on a node crosses the size
parameter. The node starts cluster formation on its sub tree. It
will form a single cluster for the whole sub tree if the sub tree
size is less than two hops; if not it will form multiple clusters.
The cluster size and the degree of overlap are also considered.
Figure 2 shows the proposed multi-level hierarchy. This
clustering scheme has proven to be effective in dynamic
environments. However, it is not a strictly localised routing
protocol since the spanning tree is a global data structure and
the whole network must be traversed before it can be
computed.
Figure 2. The basic topology of HCC
Most existing clustering protocols consume large amounts
of energy, incurred by cluster formation overheads and fixed-
level clustering, especially when sensor nodes are densely
deployed in WSNs. To address this problem Sangho Yi et al.
[11] introduced the Power-Efficient and Adaptive Clustering
Hierarchy protocol (PEACH), where the main objective is to
minimise the energy consumption of each node, and prolong
the network lifetime. In the PEACH protocol, a node becomes a
CH when it receives a packet destined for the node itself. When
the packet is destined for a different node, the node that
received the packet joins the destination node cluster. The
simulation results showed that PEACH resulted in lower energy
consumption and a higher network lifetime when compared
with LEACH, HEED and PEGASIS algorithms. However the
network is not very scalable because all the nodes must have
global knowledge of the network.
P Ding et al. [12] proposed a distributed weight based energy
efficient hierarchical clustering protocol (DWEHC) which has
proven to be an improvement on the HEED protocol. Its main
achievement is its high energy efficiency achieved by creating
balanced cluster sizes and improving the intra cluster topology.
Each sensor node start broadcast its (x, y) coordinates to find
it’s neighbouring. After finding the neighbouring nodes in its
area each node calculates its weight. The weight will be the
only parameter calculated locally used for CH election, and is
represented by Wweight in DWEHC is define in Equation 3.The
node with the largest weight is selected as a CH and the
remaining nodes become child nodes. At this stage the nodes
are considered as first level members because they have a direct
link to the CH. With the child nodes further divided into levels
(level 1, level 2, etc.) the total number of levels depends on the
clusters’ range and the minimum energy of the CH. Like
HEED, DWEHC is a fully distributed clustering protocol. It has
a more balanced CH distribution. Also its clustering process
does not rely on network size. However, this protocol is not
able to increase its energy efficiency because of its inter-cluster
communication function and it has large control message
overheads.
(3) Where is:
R is the cluster range.
d is the distance from node s to neighboring. Eresidual(s) is the residual energy in nodes.
Einitial(s) is the initial energy in nodes.
The Threshold-sensitive Energy Efficient sensor
Network (TEEN), proposed by Anjeshwar et al.[13], is a
hybrid of hierarchical clustering and data-centric protocols
designed for time-critical applications. After forming the CH
and received the attribute from the user. The CHs then
broadcast two thresholds to the nodes in their clusters those are
hard (HT) and soft (ST) thresholds. The hard threshold (HT)
refers to threshold value for the sensed attribute. It is the
absolute value of the attribute beyond which, the node sensing
this value must switch on its transmitter and report to its cluster
head. Soft Threshold (ST) is a small change in the value of the
sensed attribute which triggers the node to switch on its
transmitter and transmit. It stimulates the node to switch on its
transmitter and report the sensed data. A node will report data
only when the sensed value is override the HT or the change in
the value is greater than the ST .By varying the two thresholds
the protocol has significantly reduced the amount of data
transfer. The main drawback is that data may be lost if CHs are
not able to communicate with each other. Furthermore,
whenever the thresholds are not met, the node will not
communicate.
To address the shortcomings of TEEN the Adaptive
Periodic-TEEN (APTEEN) protocol was presented by Arati
Manjeshwar et al. in[14]. APTEEN is a hybrid clustering-based
routing protocol in which the nodes react to time-critical
situations. The nodes must have global knowledge at periodic
intervals in an energy efficient manner.
APTEEN uses a similar mechanism to LEACH-C in CH
selection. The structure of APTEEN is similar to that of TEEN,
where both protocols use the concept of hierarchical clustering
for energy efficient communication between source nodes and
the BS. The APTEEN protocol is based on three different
queries:
Historical query, to analyse past data values.
One-time query, to take a snapshot view of the
network.
Persistent queries, to monitor an event for a period of
time.
The main drawbacks of APTEEN are centred on the control
overhead on the topology formation of multiple-level clusters,
the method of implementing threshold functions and the fact
that it does not exploit spatial and temporal data correlation in
order to improve efficiency.
The most of the protocols presented in the literature are focused
on balancing the intra-cluster traffic load while forming the
clusters, in order to make more efficient use of critical
resources such as battery power and network lifetime combined
with the need for a fast convergence time and reduced energy
consumption. However, balanced intra-cluster traffic results in
a highly skewed load distribution on cluster heads. In
single-hop communication type where CHs assumes a direct
link to reach the BS, the farther CH consumes more energy and
dies earlier. Also if multi-hop inter-cluster communication is
adopted, the nodes close to the BS are overloaded with heavier
traffic load leading to energy-hole problem [15]. This is caused
by the many-to-one traffic mechanism. The nodes located in the
heavier traffic area exposed to a fast reduction of their energy
and may die quicker than faraway nodes. This may cause
serious coverage and connectivity problems nearness to the
base station. Therefore, both intra-cluster and inter-cluster
traffic have to be considered jointly when designing a
cluster-based routing algorithm. Moreover, an increase in the
number of sensor nodes used in real applications leads to large
scalability of WSNs, requiring a careful extension of multi-hop
communication protocols. Multi-level hierarchies cluster
protocols have also been regarded as efficient in conserving
energy efficiency independently of the network size. In order to
address the above issues we propose a new cluster-based
topology formation and routing protocol. This protocol aims to
load balance in both intra-cluster and inter-cluster traffic, and
guarantees reliability while solving the heavy load problem.
Table 1 shows the comparison between various clustering
protocols used in wireless sensor networks.
Clustering
protocols
Clustering method CH selection
Hie
rarch
ical
Lev
el
Cluster
communication Cluster objective
cen
tra
lized
dis
trib
ute
d
hy
brid
Pro
ba
bil
ity
ba
sed
No
n-P
rob
ab
ilit
y
ba
sed
Intr
a-c
lust
er
Inte
r-c
lust
er
En
erg
y
co
nsu
mp
tio
n
Clu
ster S
tab
ilit
y
scala
bil
ity
Loa
d b
ala
nci
ng
Deli
ver
y D
ela
y
Alg
ori
thm
com
ple
xit
y
LEACH - √
- √
- - Single Hop
Direct
Link Very poor
Medium Very poor
Medium small Low
LEACH-C √ -
- √
- - Single
Hop
Direct
Link Good high Very
poor Medium high
Low
EEHC - √
- √
- Three Single
Hop
Direct
Link Medium Medium Medium Very
poor high
Low
HEED - √
- √
- - Single
Hop
Direct
Link Medium high Medium Medium Medium Medium
DEEC - √
- √
-
Two/
Multi
Single
Hop
Direct
Link Good Medium Medium Medium Very
small
high
HCC - √
- -
√ Multi Single
Hop
Multi
Hop Very
poor Medium Good Medium high
Medium
PEACH - √
- -
√ Multi Single
Hop
Direct
Link Very
high high Medium Medium Medium
high
DWEHC - √
- √
- - Single Hop
Direct
Link Very high
high Medium Very good
Medium Medium
TEEN - √
- √
- Multi Single
Hop
Multi
Hop Medium Medium Good Medium high high
APTEEN - -
√ √
- Multi Single
Hop
Multi
Hop Good Medium Good Medium Very
small
high
Table 1. Comparison of the Clustering Protocols for Wireless Sensor Networks
III. NEW HIERARCHICAL PROTOCOL MODEL
This section presents the vision for the new hierarchical
protocol mode, which is designed to address and improve the
limitations of the protocols that have been discussed in the
previous sections. Our protocol is composed of three main
phases: cluster configuration phase, route discovery and
updates phase, data transmission phase.
A. Cluster configuration phase.
We apply IEEE802.15.4 standard, to discover neighbours and
start forming a cluster. As figure 3 shows, the first node (BS)
initiates the cluster formation process by broadcasting Hello
messages to all neighbour nodes. The Hello message includes:
node Id (Idn), clusters Id (Idcls), depth of the cluster (TTLmax),
maximum association threshold (Assmax) and the node metrics.
As soon as the neighbour node receives the Hello message, it
will first check whether it is already a member of another
cluster (Idcls = 0) within (TTLmax) or not. If the node is not yet a
member of any other cluster, it sends an AssociationRequest
message to the BS, and it will wait for an AssociationReply
message to confirm itself as cluster member. Each neighbour
node, which receives the AssociationReply message, will then
forward a Hello message to the next neighbours. This process
continues until (TTLmax) is reached. As Figure 3 illustrates,
after the neighbour nodes (n1, n2) received the Hello message,
they join the cluster. Then the Hello message will be forwarded
again to n3. However, n3 which is within 3-hops (i.e. depth of
the cluster reached) from the BS can be a CH candidate.
Figure 3. Topology formation.
The node, which has a prospective to be candidate for CH, will
first check whether it is already a CH. If not, it will send a
ClusterHeadRequest message to the BS indicating its interest to
become a CH. The BS starts to make a decision process for all
candidates for CH based on their metrics such as (energy,
distance, delay, bandwidth, priority) by using the diffusing
update algorithm (DUAL) [16] used in EIGRP protocol
(Enhanced Interior Gateway Routing Protocol) and the
maximum association threshold. Compared to EIGRP, used in
wire networks, here we don't represent the composite metrics
by their real values but by an indicative level. In our protocol
the current level of energy and the bandwidth can be
represented by 2 bits due to the limited frame and packet size in
WSN and in 802.15.4 in particular. The delay is identified by
sum of the outgoing of the CHs delay (in microseconds) to the
destination. The distance is identified by number of hops to the
BS. Also each FFD node has a priority and a unique identifier,
that can be assigned by the network administration, however if
it is 0 the FFD is not eligible to be a CH. Table 2 shows we
represent energy and the bandwidth levels (2 bits each).
value Level
00 Very Low
01 low
10 Medium
11 High
Table 2. Composite metrics identification.
The CH candidate (an FFD node in accordance to 802.15.4)
with the lowest composite metric and the highest priority is
considered to be the best choice for becoming a CH if the
maximum association is not reached.
The BS will install the new CH in its routing table and sends a
ClusterHeadReply message to acknowledge that elected node
became a new CH for the first level, which will consequently
start forming a new hierarchical cluster. In our protocol, the
ClusterHeadRequest and ClusterHeadReply messages include the
following informations: node Id (Idn), cluster Id (Idcls), hold
time (holdtime) used to detect the loss of a neighbour CH
metrics. The new CH will select two new CHs for the second
level: One as a main CH and one as a back-up (Feasible
successor in accordance to EIGRP protocol mechanism). The
process of the algorithm continues till all the possible clusters
are formed. When this process ends all CHs should form the
neighbour table and routing table and keep maintaining them
using the reachability messages (Hello message).
B. Route discovery and updates phase.
When any changes occurred in the metrics parameters to the
CH node (mentioned previously), the CH node will start to
advertise update message to the neighbour CHs in order to
maintain the routing table. If no Hello message has been
received from a neighbour node within a hold time (holdtime),
then it should be declared as dead. If a previous association was
established with this same node, the back-up node (feasible
successor), if it exists, should be selected as the default
next-hop toward the BS. If a Hello is expected to be received
from a CH and is it not received within the Holdtime interval, an
FFD node will try to connect with another CH and therefore
change his cluster Id or participate in the election process again
if he has a priority that is different from 0.
To maintain the reachability and validity of the cluster head
information between the BS and its CHs or between CHs of
different levels, a CH waiting time should be used (Lifetime).
This defines for how long any node should wait before
attempting to be a candidate for a CH. Such kind of processs
improves the stability of the sensor network. The BS or any
elected CH should send a new cluster head request before the
expiry of his election. The BS (equally the CH decider of an
upper level) may decline the renewal of the election and
chooses another CH based on the metric parameters and the
maximum association threshold.
C. Data transmission phase
Our protocol is based on multi-hop communication among the
cluster heads and the BS. When a node detects a phenomenon
in the environment, the sensed data have to be sent to the
corresponding CH, by using their clusters Id (Idcls) that were
learned during the configuration phase. After receiving data
from the sensor nodes within its cluster, each CH forwards its
data to the upper level CH through the best path till it reaches
the BS. Table 3 shows the type of message in proposed
protocol.
Message Type Function
Hello
Used for neighbour discovery and start forming
a cluster, and used to convey reachability of
discovery
AssociationRequest Sent when the node is interest to become as
cluster member.
AssociationReply Sent to confirm the node as cluster member.
ClusterHeadRequest Sent when the node is interest to become as
cluster head.
ClusterHeadReply Sent to confirm the node as cluster head.
Updates
Used to convey reachability of destinations, and
when the metrics value are change, update
message are sent so the CH can build up its
routing table.
Table 3. Message types in the proposed protocol.
IV. CONCLUSION
Several hierarchical clustering protocols have been
proposed to address several limitations in WSN protocols, such
as scalability and quality of service (QoS) in real time
application. However, most of the algorithms proposed early
randomly select CHs, and assume that each CH uses a
single-hop communication with the BS, which usually results
in low cluster quality. Therefore the sensor nodes cannot be
deployed over a large region. Hierarchical clustering
approaches have been proposed to address these limitations.
However the increase in the number of the hops in a cluster
increases also the latency and the energy consumption when
transferring the data. In this paper we proposed a new
technique to improve the reliability and the load of a sensor
network by: adapting the Enhanced Interior Gateway Routing
Protocol to the sensor context, changing the metric value and
introducing two new metrics (maximum association and
priority). Despite its unique features and advantages, the
performances of our protocol require to be extensively
evaluated and validated. A dynamic tuning of the new
introduced parameters should be also considered and studied.
Therefore, we plan to simulate our protocol using NS2 and
address its limitations.
REFERENCES
[1] K. Pawar and Y. Kelkar, “A Survey of Hierarchical
Routing Protocols in WSN.” 2012.
[2] S. Gupta, N. Jain, and P. Sinha, “Clustering Protocols in
Wireless Sensor Networks : A Survey,” IJAIS, vol. 5,
no. 2, 2013.
[3] D. J. Dechene, A. El Jardali, M. Luccini, and A. Sauer,
“A Survey of Clustering Algorithms for Wireless Sensor
Networks.” 2013.
[4] W. R. Heinzelman, A. Chandrakasan, and H.
Balakrishnan, “Energy-Efficient Communication
Protocol for Wireless Microsensor Networks,” vol. 00,
no. c, pp. 1–10, 2000.
[5] X. Liu, “A survey on clustering routing protocols in
wireless sensor networks.,” Sensors (Basel)., vol. 12, no.
8, pp. 11113–53, Jan. 2012.
[6] W. B. Heinzelman, A. P. Chandrakasan, S. Member, and
H. Balakrishnan, “An Application-Specific Protocol
Architecture for Wireless Microsensor Networks,” vol.
1, no. 4, pp. 660–670, 2002.
[7] S. Bandyopadhyay and E. J. Coyle, “An Energy
Efficient Hierarchical Clustering Algorithm for
Wireless Sensor Networks,” vol. 00, no. C, 2003.
[8] O. Younis and S. Fahmy, “HEED : A Hybrid ,
Energy-Efficient , Distributed Clustering Approach for
Ad-hoc Sensor Networks,” vol. 0238294, pp. 1–36.
[9] L. Qing, Q. Zhu, and M. Wang, “Design of a distributed
energy-efficient clustering algorithm for heterogeneous
wireless sensor networks,” Comput. Commun., vol. 29,
no. 12, pp. 2230–2237, Aug. 2006.
[10] Y. Zhang, L. T.Yang, and J. Chen, RFID AND
SENSOR. Taylor & Francis Group, 2010.
[11] S. Yi, J. Heo, Y. Cho, and J. Hong, “PEACH:
Power-efficient and adaptive clustering hierarchy
protocol for wireless sensor networks,” Comput.
Commun., vol. 30, no. 14–15, pp. 2842–2852, Oct.
2007.
[12] P. Ding, J. Holliday, and A. Celik, “Distributed
Energy-Efficient Hierarchical Clustering for Wireless
Sensor Networks 2 Related Work,” 2003.
[13] A. Manjeshwar and D. P. Agrawal, “TEEN : A Routing
Protocol for Enhanced Efficiency in Wireless Sensor
Networks,” vol. 00, no. C, 2001.
[14] A. Manjeshwar and D. P. Agrawal, “APTEEN : A
Hybrid Protocol for Efficient Routing and
Comprehensive Information Retrieval in Wireless
Sensor Networks ∗ ,” 2002.
[15] M. K. Watfa, H. Al-hassanieh, and S. Salmen, “Journal
of Network and Computer Applications A novel
solution to the energy hole problem in sensor networks,”
J. Netw. Comput. Appl., vol. 36, no. 2, pp. 949–958,
2013.
[16] C. Zhao, Y. Liu, and K. Liu, “A More Efficient
Diffusing Update Algorithm For Loop-Free Routing,”
2009 5th Int. Conf. Wirel. Commun. Netw. Mob.
Comput., pp. 1–4, Sep. 2009.
Mobile User Authentication System for e-Commerce
Applications
Rania. A. Molla
Department of Computer Science
Collage of Computing and
Information Technology
King Abdulaziz University
Jeddah, Kingdom of Saudi Arabia
Imed Romdhani, Bill
Buchanan
School of Computing
Edinburgh Napier University
Edinburgh, UK
Etimad. Y. Fadel
Department of Computer Science
Collage of Computing and
Information Technology
King Abdulaziz University
Jeddah, Kingdom of Saudi Arabia
Abstract— E-commerce applications provide on-line clients
and merchants with a quick and convenient way to exchange
goods and services. However, the deployment of these
applications is still facing many problems such as security
threats; and on-line attacks. These often cause users to be
concerned about their own privacy and encourage them to stop
using on-line methods. Thus, a number of on-line authentication
technologies and methods have been developed in order to
authenticate users and merchants, verify their identities, and
therefore overcome e-commerce security threats. Although
stand-alone authentication solutions have been successful in
authenticating legitimate clients and in defeating on-line attacks,
they are often weak in overcoming the Man-In-The-Browser
(MITB) attack, which is a type of Internet threat that infects a
web-browser in a concealed fashion, and is invisible to both client
and host applications. This paper presents a Mobile User
Authentication System (MUAS) that uses QR code technology to
authenticate on-line users, through a challenge/response protocol.
Based on this mechanism, the system integrates different
authentication technologies and methods to provide an improved
and secure on-line user and merchant authentication system that
overcomes MITB attack, without compromising usability and
ubiquity.
Keywords— Authentication; Man-In-The-Browser attack
(MITB); QR code; Out-Of-Band communication channel (OOB).
I. INTRODUCTION
With the rapid development of the Internet, e-commerce systems have grown to become more popular for on-line transactions, and this has led to increasing on-line fraud and security issues, which cannot be solved by a single technology. Thus, many researchers have proposed different on-line authentication solutions, by combining various technologies and methods such as: multi-factor authentication[1-3], biometrics [4-6], and smart card technologies [7, 8]. However, these solutions still suffer from some limitations, such as: cost; hardware dependency; lack of mobility, scalability and interoperability [9]. To overcome these limitations, other researchers have taken advantage of mobile phone technology features as a means for: on-line identity authentication [10-15]; secure virtual private networks login [16-18]; and for on-line financial transactions [15, 19-23]. Nevertheless, on-line banking continues to present
challenges to the user’s financial security and personal privacy. On-line attackers and fraudsters are often using advanced methods to target on-line clients. One of the latest and most dangerous methods is the Trojan to launch a Man-In-The-Browser (MITB) attack [24]. MITB is a Trojan that embeds in a user's browser application and can be programmed to trigger when a user accesses specific on-line sites, such as an on-line banking site. Once activated, it can intercept and manipulate any information a user submits on-line in real-time [25]. MITB is hard to detect and, in many cases, it succeeds in causing complete, concealed damage, where neither the client nor the bank is aware of any irregularity [26]. This type of attack highlights the need for a solution that securely authenticates users, and ensures the integrity of transactions in the face of an evolving threat environment context.
This paper introduces a new technique called Mobile User Authentication System (MUAS) that employees the widespread QR code technology, through a convenient challenge/response protocol. The proposed system minimizes the risk of MITB attack by using mobile devices, along with its cellular network as an Out-Of-Band (OOB) communication channel. OOB works by using the web-based application screen as the primary channel, while the cellular network is a second channel used to pass confidential information.
This paper makes the following contributions:
1) Client/Server Mutual Authentication
The technique’s first step is to secure the one and only
web-based communication channel, through TLS/SSL
protocol. It authenticates the client and the server in
order to ensure message integrity and confidentiality.
2) Minimal User Input.
This technique is a user-friendly challenge response
authentication system, requiring the client to only
submit his mobile number, and to capture the QR code.
3) Overcome the risk of Man-In-The-Browser (MITB)
attack. This technique is designed in a way to overcome
the risk of MITB attack, by restricting the use of PC
browsers to client’s mobile number submission only.
And using an OOB communication channel for passing
confidential information, required for user
authentication. Even if MITB exists within the client’s
browser, the type of submitted information cannot be
useful for the attacker.
4) Prevent Replay Attack.
The technique generates two similar random numbers
on the merchant-side and the mobile phone-side, which
will be compared against each other to ensure client and
mobile phone legitimacy.
5) Protect client’s confidentiality.
This technique requires no website registration nor
memorizing any passwords, due to its vulnerability to
server attacks. Therefore, no personal information is
required to be submitted by the client, except for his/her
mobile number.
This paper is divided into seven sections. Section 2 presents the threat addressed by this paper. Section 3 describes the core algorithm of the MUAS, along with its components and architecture. Section 4 presents the related work. Next, section 5 that presents the evaluation criteria and design. Section 6 discusses the security analysis. Finally, section 7 presents conclusion and future work.
II. THE MAN-IN-THE-BROWSER ATTACK (MITB)
MUAS is an authentication system designed to provide a secure and easy-to-use system, while overcoming Man-In-The-Browser (MITB) attack. MITB has been defined by Almeida et al. [24] as “A Trojan that embeds in a user’s browser application and can be programmed to trigger when a user access specific on-line sites, such as an on-line banking site”.
MITB attack consists of two phases: The first phase is the infection of a target computer by means of phishing e-mails, requests to install updates, downloading interesting PDF documents, and so on, which causes malware-infected software to be downloaded into the user’s browser. Whenever the user restarts the browser, the Trojan installs its extension into the browser’s configuration, and start to load the extension. Finally, the extension registers a handler for every page load [24]. The second phase is the transaction takeover. Whenever a page is loaded, its URL is searched for by the extension. Next, all data submitted by the user is extracted from the browser and modified by the attacker, then the browser continues to submit the modified data to the server. The server performs the transaction on the modified data, generates a receipt, and send it back to the browser. Finally, the extension replaces the modified data in the receipt with the original data submitted by the user in the beginning. The browser displays the modified receipt with the original data, while the user thinks that the original transaction has been authorized correctly [24].
Numerous strong authentication techniques such as biometrics, grid card, OTP token, Out-of-Band OTP, EMV-CAP, smart cards and digital certificates are available, and effective against a wide range of threats [27]. However, most of these techniques require user interaction with the browser. Therefore, MITB can intercept them or wait until the user passes the challenge before taking over [24, 27]. Even Anti-
Virus or Anti-Malware applications which are deployed to end-user computers in order to detect and disable malwares cannot be fully trusted. Malwares are rapidly changing and evolving making it difficult for client software to keep up [24, 27]. Therefore, a solution that uses an Out-of-Band (OOB) communication channel can be a powerful weapon against advanced threats, especially Man-In-the-Browser (MITB) attack, since it avoids using the communication channel often used by attackers. Moreover, this technique leverages devices such as mobile phones that are already in-use by most clients, and enables reviewing authentication details away from the influence of malware on the client’s computer.
A. Problem Statement
In developing an e-commerce authentication system that overcomes Man-In-The-Browser (MITB) attack, the following questions arise:
Who will conduct the authentication process?
What kind of information must be provided by the
client?
How to verify the identity of the client and merchant
through a QR code?
What kind of technologies will be used to overcome
the risks of Man-In-The-Browser (MITB) attack?
Does the system satisfy the e-commerce security
requirements?
III. MOBILE USER AUTHENTICATION SYSTEM
(MUAS)
Our Mobile User Authentication System (MUAS) authenticates the user via a mobile phone. The system is based on a challenge/response protocol that generates and displays an encrypted QR code as a challenge, in which the client uses the mobile phone camera to capture, in order to generate a response that ensures user’s authenticity. In this section we present the core algorithm of the MUAS.
We will explain the basic workflow, from Peuedo-Random
Number Generation (PRNG), QR code cryptography, to user
authentication process.
1) Psuedo-Random Number Generation (PRNG)
The system is based on generating two random-
numbers. One on the merchant-side, and the other is on
the mobile phone-side. Both numbers are generated using
the same seed. The merchant generates its PRN (M-PRN)
by using the mobile-number submitted by the client,
while the mobile phone generates its PRN (MP-PRN) by
using the mobile-number embedded in the International
Mobile Subscriber Identity (IMSI), which is stored on the
SIM card.
2) QR code cryptography
The MUAS QR code generation process is based on
encoding the merchant’s public-key, provider’s name,
merchant-generated PRN (M-PRN), and the response
end-point (URL) into a QR code, after encrypting it with
the client’s public key.
The reason behind including the provider’s name and
the response end-point into the QR code is to help the
client ensures the QR code authenticity, and to specify
where the mobile phone will respond to the challenge.
3) Authentication process The authentication process is composed of two layers. The
first layer is performed on behalf of the mobile phone-side, in which the client launches a PIN-protected application on the phone, called “DGCH”. The second layer is performed on behalf of the merchant-side, in which verifies user’s authenticity.
3.1) DGH application layer
Once the client launches the DGCH application
(Decrypt, Generate, Compare, and Hash) on the mobile
phone, and selects the “start” button, the camera
becomes activated to capture the QR code. The
application extracts the encrypted contents by applying
the client’s private key to get the merchant’s public-
key, provider’s name, merchant-generated PRN (M-
PRN), and the response end-point.
Next, DGCH compares the recovered merchant-
generated PRN (M-PRN) against the mobile phone-
generated PRN (MP-PRN). If the comparison is a
success, the application computes a response by
comprising of the SHA256 hash of the mobile phone-
generated PRN (MP-PRN), and sends the hashing result
to the response end-point, along with the mobile phone-
generated PRN (MP-PRN), after encrypting them with
the merchant’s public key. Otherwise, the DGCH gives
an “Unsecure Transaction” message and exits the
application.
3.1) Verification layer
The merchant decrypts the received hash and the
mobile phone-generated PRN (MP-PRN) by applying
its private key. The received hash is verified by
applying SHA256 to the received mobile phone-
generated PRN (MP-PRN). If both hashes are equal, the
user is authenticated.
A. MUAS Assumptions
The MUAS assumes that TLS/SSL protocol is the standard communication channel in the proposed MUAS. The following table shows the notations used in the MUAS framework.
Notations Meaning
MN Mobile number
KC Client’s public key
K-1C Client’s private key
KM Merchant’s public key
K-1M Merchant’s private key
end-point Merchant’s URL address
PRN Psuedo-Random Number
M-PRN Merchant- Psuedo-Random Number
MP-PRN Mobile Phone- Psuedo-Random Number
B. MUAS Architecture
The client in the MUAS initiates the process with the merchant through sending his mobile number. The authentication process performs a challenge/response protocol to authenticate the client. Figure 1 shows the proposed MUAS.
Fig 1. Mobile User Authentication System
The MUAS process works as follows (Figure 2):
1. The client initiates the process with the merchant, by
sending his Mobile Number (MN).
2. The merchant generates a random number (M-PRN) from
the original Mobile Number, provided by the user in the
initialization process.
3. The merchant appends M-PRN with the merchant’s
public-key, Provider’s Name (PN), and merchant’s end-
point (URL), in order to be encrypted with the client’s
public key (KC) to represents the QR code contents.
4. The QR code is displayed on the client’s PC.
5. The client captures the QR code with his mobile phone.
6. DGCH application decrypts the QR code contents by
using the client’s private key (K-1C) to get M-PRN.
7. A random number (MP-PRN) is generated from the
mobile number embedded in the IMSI.
8. Next, DGCH compares the recovered M-PRN with MP-
PRN. If the comparison is a success, then:
9. An SHA256 hashing algorithm will be performed on
MP-PRN, and get Hash result (H1).
10. Hash result (H1) will be appended with the original
MP-PRN, and encrypted with the merchant’s public
key (KM) to be sent to the merchant.
11. The merchant decrypts the received Hash result (H1)
and the original MP-PRN by using its private key (K-
1M).
12. The merchant verifies the received H1 by applying
the same hashing algorithm (as in step7) on the
recovered MP-PRN, and gets a Hash result (H2).
13. The merchant compares H1 and H2. If both hashes
are equal, then the merchant authenticates the user.
Otherwise, DGCH quits.
Fig 2. MUAS Authentication process diagram
Figure 3 shows the MUAS algorithm, which explains how to
verify the identity of the client and merchant through a
cryptographic QR code.
//Merchant-SIDE IF (Mobile_Number)
1. Client_cert=Pass_ClientCert_to_merchant();
2. M_PRN=Generate_MPRN(Mobile_Number);
3. Merchant_PK=Extract_PublicKey_from_MerchantCert();
4. QRcode=Append(Merchant_PK,ProviderName,M_PRN,
EndPoint);
5. Client_PK=Extract_PublicKey_from_ClientCert(Client_cert);
6. Encrypted_QR=Encrypt_QR(QRcode,Client_PK);
7. Pass_Encrypted_QR_to_PCbrowser(Encrypted_QR);
//Mobile Phone-SIDE 8. Captured_QR=Capture_QR_by_camera();
9. Client_PrivKey=Extract_PrivateKey_from_ClientCert
(Client_cert);
10. Decrypted_QR=Decrypt_QR(Encrypted_QR,Client_
Privkey);
11 .QR_Array=SplitString(Decrypted_QR);
12. MP_PRN=Generate_MPPRN(IMSI_MobileNumber);
13. Compare_Result=Compare(M-PRN, MP-PRN);
14. IF (Compare_Result==true)
15. Hash1_MPprn=SHA256(MP_PRN);
16. Hash_and_OrigMsg=Append(Hash1_MPprn,MP_PRN);
17. Encrypted_Hash_OrigMsg=Encrypt_Hash(Hash_and_
OrigMsg,QR_Array[0]);
18. Pass_EncryptedHash-to-Merchant(QR_Array[3]
Encrypted_Hash_OrigMsg);
//Merchant-SIDE 19. Merchant_PrivKey=Extract_PrivateKey-from Merchant
(Merchant_cert);
20. Decrypted_Hash_OrigMsg=Decrypt_RecievedHash
(Encrypted_Hash_OrigMsg,Merchant_PrivKey);
21. Hash_Array=SplitString(Decrypted_Hash_OrigMsg);
22. Hash2_MPprn_by_Merchant=SHA256(Hash_Array[1]);
23. Verify_Hash(Hash2_MPprn_by_Merchant,Hash_
Array[0]);
ELSE
Exit
END IF
END IF
Fig 3. MUAS Algorithm
IV. RELATD WORK
In e-commerce, all authentication evidence between the client and the merchant is exchanged through a single communication channel, usually the internet, making it prone to eavesdropping attacks. Given the large penetration of mobile phones, some recent studies considered extending the usage of SIM card authentication, along with its wireless mobile network, to web services for identity authentication. In addition, the mobile industry began to pay more attention to QR code applications in m-commerce, because of its simplicity and inexpensiveness in presenting diverse
commerce data. Moreover, it effectively improves mobile user experience through reducing mobile inputs [28]. Therefore, a lot of research work and technology studies have been done on mobile-based QR code applications to provide a secure and reliable e-commerce authentication system.
A. SIM-based Identity Authentication
The use of mobile phones and their SIM cards for identity authentication has been the focus of many studies, in order to provide secure, reliable, and easy-to-use on-line authentication solutions. Van Thanh et al.[11, 14] proposed some solutions tailored for Single-Sign-On (SSO) authentication, in which users can access services from their PC/laptop. The user’s browser is redirected to the identity provider (IdP) for logging in, using some extra hardware devices, such as SIM-card readers, USB dongle, or cables to connect the mobile to the PC. In some cases, Bluetooth connection is required. SSO Authentication can also be used for users accessing servers from their mobile phones, where, in this case, verification is done through SMS messages.
Al-Qayedi et al. [10] and Me et al. [12] proposed other authentication approaches that combines traditional web-based username/password approach, with mobile-based challenge/response authentication, and an OTP, in order to provide privacy and eavesdropping security.
Abe et al. [13] proposed a similar, yet better solution requiring an (IdP) software running on the mobile phone. Whenever a user requests a service, the service provider redirects the user to the IdP on his mobile phone. The IdP software generates a digital signature, and sends it to the service provider for verification.
In summary, most SIM-based identity authentication approaches meet the requirements of mobility, usability, availability, and security. On the other hand, it increases the risk of MITB attack, due to the use of PC browsers as the main communication channel. While MUAS decreases the risk of MITB attack by employing QR code in an OOB communication channel.
B. SIM-based QR Code Authentication
The simplicity and cost effectiveness of QR code technology have caused a number of new approaches to use it in on-line authentication solutions. QR code works quickly by establishing a secure connection between the server, desktop, and the mobile phone [29]. Dodson et al. [29] proposed Snap2Pass, a technique in which a client creates an account, and logs into a website from a PC browser. The technique works by the client capturing the displayed QR code, which encodes a cryptographic challenge, and sending the cryptographic response to the server for verification. Vapen et al. [30] proposed a similar approach called 2-clickAuth which is based on implementing an Identity Provider in the OpenID federated identity management system, making the authentication solution available to all users of sites that support OpenID. Both approaches provide high levels of security, availability, and usability. Compared to MUAS, our approach does not require an account creation with any website, nor requires an Identity Provider, since it can be applied to any e-commerce website. Moreover, both approaches do not overcome the risk of MITB attack. Where
Snap2Pass account creation phase lacks QR code cryptography, and 2-clickAuth uses an untrusted PC to capture the response generated by the mobile phone.
Choi et al. [31] proposed a similar approach to 2-clickAuth, yet it employed an extended authentication server, to prevent phishing server attacks in Single-Sign-On (SSO). The extended server generate a QR code to be captured by the mobile, which in turn generates another QR code to be captured by a webcam. This two-way QR code generation can be effective against MITB. But on the other hand, it is less practical.
Kim and Jun [32] proposed an authentication technique by using a registered mobile phone. Whenever a registered user requests a service from the service provider, the service provider extracts the user information and generates a QR code. The mobile phone analyse the QR code contents, and send them to the service provider for validation. In comparison to MUAS, our approach requires the client to only submit his mobile number. While the proposed approach requires no user input, which in turn helps minimize the risk of MITB attack. But on the other hand, the client’s mobile phone needs to be registered with the service provider.
C. SIM-based Encrypter QR Code Authentication
Ease of QR code capturing and contents viewing are considered as the technique’s main shortcomings, especially in authentication systems, where security is essential. Therefore, a trend toward combining QR code technology with cryptography can help ensure recipient’s authorization, and increase QR code contents security and confidentiality, required for authenticating users. As a result, a number of techniques have been proposed to use the QR code in the field of cryptography. Dey [33] proposed SD-EQR, an algorithm to store messages in an encrypted format with a password and sends it to the destination hidden in QR code. The algorithm works by generating a secret code from a chosen password, which will be added to each letter of the message, in order to generate the first phase of encryption. Followed, is reversing the encrypted message. Finally, an exclusive-OR is performed. Compared to MUAS, our QR code cryptography requires less computations than SD-EQR, since MUAS algorithm only performs one level of encryption.
V. SYSTEM DESIGN AND EVALUATION CRITERIA
In this section, we present the system design and initial evaluation criteria of the proposed MUAS.
A. System Design
Figure 4 shows the MUAS architectural design. The components of the proposed system are described as follows:
Fig 4. MUAS architecture
1) Client (PC)
Submission Process: it initiates the MUAS through submitting
the client’s mobile number.
2) Merchant
Generate Random Number Process: it uses the mobile number
as a seed to generate the merchant-random number.
QR-code Encode Process: it appends the merchant-random
number with the merchant’s public-key, provider’s name, and
merchant’s URL address, encrypt them with the client’s public
key, and generate the QR-code.
Hashing Process: it performs an (SHA256) hashing algorithm
on the received mobile phone-random number, generated by
the mobile phone.
Verify Process: it compares both hashes to verify user’s
authenticity.
3) Mobile Phone
QR-code Decode Process: it decodes the captured QR image,
and decrypt it using the client’s private key.
Generate Random number Process: it generates a random
number, from the mobile number embedded in the IMSI.
Compare random Numbers: it splits the appended QR code
contents to get the merchant-random number, and compares it
against the mobile phone-random number
Hashing Process: it perform an (SHA256) hashing algorithm
on the mobile phone-random number.
B. Evaluation Criteria
The trend towards designing an on-line authentication system have extended beyond solving security issues to systems which are available and user-friendly.
The MUAS employment of a secure personal device, such as a
mobile phone is considered an ideal solution for on-line
authentication systems, since the phone is always with us and
always switched on, which makes it available at all times.
Moreover, using mobile phones eliminates the need to carry
around extra hardware for authentication, which in turn help
satisfy the mobility and usability features.
MUAS can be considered as a user-friendly system, due to the
requirement to submit the minimum amount of information,
which is the mobile number.
The proposed system satisfy the following e-commerce
security requirements:
1) Integrity
The generation of random number processor at the
merchant-side and mobile phone-side ensures integrity, in
which both processors generate the same random number,
due to using the same seed. Even if an intruder
manipulated the mobile number submitted by the client,
which is unlikely to happen, due to the use of TLS/SSL,
the mobile phone-side will notice the manipulation and
will stop the authentication process immediately.
2) Confidentiality
The QR code encoding/decoding processors at the
merchant-side and mobile phone-side ensures
confidentiality through encapsulating and encrypting
significant information into a QR code. These information
can only be revealed to the legitimate mobile phone
holder, who must use his PIN-protected private key to
decrypt the information, which will be used in the
authentication process. Therefore, even if the mobile
phone was stolen, the private key cannot be obtained, due
to the use of a PIN.
3) Non-repudiation
The hashing processor at the mobile phone-side satisfies
the non-repudiation security requirement through hashing
the random number generated from the embedded IMSI.
The hash result is sent to the merchant to be compared
against another hash generated from the submitted mobile
number. This can be considered a legal proof that the
client has placed an order.
C. Implementation
The basic prototype of MUAS framework has been implemented using C++ Builder XE3. The next step will be to simulate the framework’s protocol using NS2 to test its speed, reliability, and scalability.
VI. SECURITY ANALYSIS
Assume a secure SSL/TLS communication channel between the client (PC), and the merchant. TLS/SSL protocol has been developed by the Internet Engineering Task Force (IETF) as the standard protocol for providing security services in the context of e-commerce over the internet [34]. The primary goal of employing SSL/TLS in the MUAS is the client/server mutual authentication, and their encryption algorithm and cryptographic keys negotiation which ensures authenticity, privacy and integrity for data being sent between the communicating parties. Moreover, the MUAS employs an Out-Of-Band (OOB) communication channel (cellular network) for QR code capturing, to reduce the use of PC browsers, which in turn helps overcome the risk of Man-In-The-Browser (MITB) attack.
In the proposed system, the merchant generates a random number from the received mobile number, appends it with the merchant’s public-key, provider’s name, and merchant’s end-point. Next, it performs RSA cryptography, and form the QR code. RSA is an asymmetric algorithm, which ideally suited the real-world use, as the secret key does not have to be shared, the risk of being known is much smaller than symmetric algorithms. In terms of security, RSA is considered the most secure algorithm. Finally, the mobile phone performs a hashing algorithm (SHA256) on the random numbers, and sends the hash result along with the original message to the merchant for verification. SHA256 was chosen due to its high speed and its reasonable digest size.
VII. CONCLUSION AND FUTURE WORK
Initially, on-line fraudsters used to steal customer’s personal information by sending phishing e-mails, in order to steal money from their internet banking account. Nowadays, fraudsters are using newer and more advanced methods to defraud on-line clients. One of the most dangerous methods is Man-In-The-Browser (MITB) attack. Therefore, this paper proposes a Mobile User Authentication System (MUAS) that integrates a number of technologies to help achieve a stronger authentication solution, such as mobile phones, cryptography, hashing, and QR code. The MUAS mutually authenticates the client and the merchant, while overcoming the MITB by reducing the use of PC browsers, and using mobile phones along with their cellular network instead. Moreover, the MUAS satisfies e-commerce security requirements: authenticity, integrity, confidentiality, and non-repudiation.
This research is part of a larger scope, in which the future work will extend the MUAS client and merchant authentication to on-line transaction authentication, and verifying legitimate mobile phone holders through GSM Mobile Network (GMN). In other words, a mobile electronic transaction layer will be integrated to MUAS. Where the mobile phone will verify the integrity of the client-generated confidential information by passing it to the merchant, bank, and GSM authentication server, all the way back to the mobile phone. The future framework will authenticate all parties involved in the transaction, while ensuring data confidentiality, non-repudiation, and defeating MITB attack.
REFERENCES
[1] S. Hallsteinsen, I. Jorstad, and T. Do Van, "Using the mobile
phone as a security token for unified authentication," in Systems and Networks Communications, 2007. ICSNC 2007. Second
International Conference on, 2007, pp. 68-68.
[2] X. Yin, J. Zou, C. Fan, and P. Zhou, "An Improved Dynamic Identity Authentication Scheme Based on PKI-SIM Card," in
Wireless Communications, Networking and Mobile Computing,
2009. WiCom '09. 5th International Conference on, 2009, pp. 1-4. [3] T. Fan Yu and P. Su Gui, "Design of Two-Way One-Time-
Password Authentication Scheme Based on True Random
Numbers," in Computer Science and Engineering, 2009. WCSE '09. Second International Workshop on, 2009, pp. 11-14.
[4] C. Li, Y.-x. Yang, and X.-x. Niu, "Biometric-based personal
identity-authentication system and security analysis," The Journal of China Universities of Posts and Telecommunications, vol. 13,
pp. 43-47, 2006.
[5] M. Xiaoming, "Study on the Model of E-Commerce Identity Authentication Based on Multi-biometric Features Identification,"
in Computing, Communication, Control, and Management, 2008.
CCCM '08. ISECS International Colloquium on, 2008, pp. 196-200.
[6] M. Yildiz and M. Gokturk, "Combining Biometric ID Cards and
Online Credit Card Transactions," in Digital Society, 2010. ICDS '10. Fourth International Conference on, 2010, pp. 20-24.
[7] N. Li Huang and D. T. Tan, "A novel JavaCard-based authentication system for secured transactions on the Internet," in
Networks, 2000. (ICON 2000). Proceedings. IEEE International
Conference on, 2000, pp. 262-266. [8] T. Qian, Z. Junwei, F. Chunxiao, and Z. Xiaoying, "A mobile
identity authentication scheme of e-commerce based on Java-SIM
card," in Information Networking and Automation (ICINA), 2010 International Conference on, pp. V2-114-V2-118.
[9] T. Do van, A. Nachef, J. D. Aussel, I. Jørstad, R. Perlman, J.
Vigilante, et al. (2006). Offering SIM Strong Authentication to Internet Services. Available:
http://projectliberty.org/liberty/content/download/397/2750/file/SI
M_Strong_Authentcation_Whitepaper.pdf [10] A. Al-Qayedi, W. Adi, A. Zahro, and A. Mabrouk, "Combined
Web/mobile authentication for secure Web access control," in
Wireless Communications and Networking Conference, 2004. WCNC. 2004 IEEE, 2004, pp. 677-681 Vol.2.
[11] T. Do van, T. Jonvik, T. Do van, and I. Jorstad, "NETp1-09:
Enhancing Internet Service Security Using GSM SIM Authentication," in Global Telecommunications Conference, 2006.
GLOBECOM '06. IEEE, 2006, pp. 1-5.
[12] G. Me, D. Pirro, and R. Sarrecchia, "A mobile based approach to strong authentication on Web," in Computing in the Global
Information Technology, 2006. ICCGI '06. International Multi-
Conference on, 2006, pp. 67-67. [13] A. Tsuyoshi, I. Hiroki, and T. Kenji, "Implementing identity
provider on mobile phone," presented at the Proceedings of the
2007 ACM workshop on Digital identity management, Fairfax, Virginia, USA, 2007.
[14] D. van Thanh, T. Jonvik, F. Boning, D. van Thuan, and I. Jorstad,
"Simple Strong Authentication for Internet Applications Using Mobile Phones," in Global Telecommunications Conference, 2008.
IEEE GLOBECOM 2008. IEEE, 2008, pp. 1-5.
[15] M. Ashraff, M. L. Kabir, S. M. Aziz, and B. k. Dey, "A Conceptual Framework for a SIM-based Electronic Transaction
Authentication System," in Network and Parallel Computing
Workshops, 2007. NPC Workshops. IFIP International Conference
on, 2007, pp. 65-71.
[16] D. van Thanh, I. Jorstad, T. A. Johansen, E. Bakken, and D. van
Thuan, "Pervasive service access with SIM-based VPN," in Mobile Adhoc and Sensor Systems, 2009. MASS '09. IEEE 6th
International Conference on, 2009, pp. 836-841.
[17] W. Audun, L. Lars, J. Ivar, and D. Than van, "Secured enterprise access with strong SIM authentication," in Enterprise Distributed
Object Computing Conference, 2006. EDOC '06. 10th IEEE
International, 2006, pp. 463-466. [18] T. Do Van, T. Jnvik, I. Jrstad, B. Feng, and T. Do Van, "Strong
authentication using dual SIM," in Intelligence in Next Generation
Networks, 2009. ICIN 2009. 13th International Conference on, 2009, pp. 1-4.
[19] C. Le-Pong and C. Jyh-Yen, "SIM card based e-cash applications
in the mobile communication system using OTA and STK technology," in Wireless, Mobile and Multimedia Networks, 2006
IET International Conference on, 2006, pp. 1-3.
[20] R.-J. Hwang, S.-H. Shiau, and D.-F. Jan, "A new mobile payment scheme for roaming services," Electronic Commerce Research and
Applications, vol. 6, pp. 184-191, 2007.
[21] J. T. Issac and J. S. Camara, "An anonymous Account Based Mobile Payment Protocol for a Restricted Connectivity Scenario,"
presented at the Database and Expert System Applications, 2007.
DEXA, 07. 18th international workshop on, 2007. [22] A. A. Tabandehjooy and N. Nazhand, "A Lighweight and Secure
Protocol for Mobile Payments Via Wireless Internet in M-
commerce," in e-Education, e-Business, e-Management, and e-Learning, 2010. IC4E '10. International Conference on, 2010, pp.
495-498.
[23] A. Bottoni and G. Dini, "Improving authentication of remote card transactions with mobile personal trusted devices," Computer
Communications, vol. 30, pp. 1697-1712, 2007.
[24] M. Almeida, E. Dimogerontakis, U. Buyuksahin, and A. Tarhan, "Man-In-The-Browser Attacks," 2011.
[25] (2011). MAKING SENSE OF MAN-IN-THE-BROWSER ATTACKS - Threat analysis and Mitigation for Financial Institutions.
Available:
http://twimgs.com/darkreading/authentication/Making_Sense_of_Man-in-the-Browser_Attacks.pdf
[26] M. Almeida, E. Dimogerontakis, U. Buyuksahin, and A. Tarhan,
"Man-In-The-Browser Attacks," Decemebr 20, 2011. [27] (2010). Defeating Man-in-the-Browser: How to prevent the Latest
Malware Attacks against Consumer and Corporate Banking.
[28] J. Z. Gao, L. Prakash, and R. Jagatesan, "Understanding 2D-BarCode Technology and Applications in M-Commerce - Design
and Implementation of A 2D Barcode Processing Solution," in
Computer Software and Applications Conference, 2007. COMPSAC 2007. 31st Annual International, 2007, pp. 49-56.
[29] B. Dodson, D. Sengupta, D. Boneh, and M. S. Lam. (2009,
Snap2Pass: Consumer-Friendly Challenge-Response Authentication with a Phone. Available:
senguptas.org/Documents/secure2pass_www2009.pdf
[30] A. Vapen, D. Byers, and N. Shahmehri, "2-clickAuth Optical Challenge-Response Authentication," in Availability, Reliability,
and Security, 2010. ARES '10 International Conference on, pp. 79-
86. [31] C. Kyeongwon, L. Changbin, J. Woongryul, L. Kwangwoo, and
W. Dongho, "A mobile based anti-phishing authentication scheme
using QR code," in Mobile IT Convergence (ICMIC), 2011 International Conference on, pp. 109-113.
[32] K. Young-Gon and J. Moon-Seog, "A design of user authentication
system using QR code identifying method," in Computer Sciences and Convergence Information Technology (ICCIT), 2011 6th
International Conference on, 2011, pp. 31-35.
[33] S. Dey, "SD-EQR: A New Technique To use QR Codes in Cryptography," International Journal of Information Technology
& Computer Science (IJITCS), vol. 3, pp. 11-21, May/June 2012.
[34] "The Secure Sockets Layer (SSL) Protocol Version 3.0 ", ed: Internet Engineering Task Force (IETF), RFC: 6101, August 2011.
An Overview of the Emergency and Infotainment Data Dissemination in VANET
Amina Sedjelmaci STIC Laboratory
University of Tlemcen, Algeria [email protected]
Fedoua Didi Dept of Computer engineering
University of Tlemcen, Algeria [email protected]
Abstract— In this paper, we survey recent research of the emergency and infotainment data dissemination in Vehicular Ad-hoc Networks. Vehicular networks allow the development of new applications that can provide not only information about the safety but also information about comfort to passengers. The question that arises is how to propagate data to vehicles efficiently throughout the vehicular environment and to alleviate numerous issues concerning information dissemination in VANETS
Keywords- VANET, ITS, Information dissemination protocols.
I. INTRODUCTION Vehicular communication is considered as a key
technology for improving safety, efficiency and comfort of our travels through Intelligent Transportation Systems (ITS). VANETs perform the communication between the vehicle-to-vehicle (V2V) and vehicle-to-roadside (V2R).Figure 1.
Figure 1: V2V and V2R communication
The growing interest toward several and possible applications led to develop technologies, protocols and efficient approach for data dissemination in cooperative vehicular networks. Communications standards and protocols have to consider specific conditions existing in vehicular networks, i.e. high velocity, long distances, constrained movement and various density determined by traffic demand that changes over time and space. The 802.11 task force group has been working on the development of a new communication standard known as IEEE 802.11p. This new standard is based on the 802.11a technology and is also referred as the Dedicated Short-Range Communications (DSRC) standard. DSRC uses the 5 GHz frequency spectrum that is divided into seven channels (10 MHz each): one control channel (CCH) and six service channels (SCHs) as shown in Figure 2.
Figure 2: Channels available in 802.11p DSRC evolved into Wireless Access in Vehicular Environment (WAVE). WAVE supports high-speed V2V and V2I communications and has major applications in ITS, vehicle safety services, and Internet access. WAVE operates at 5.850–5.925 GHz and adopts Orthogonal Frequency-Division Multiplexing (OFDM) and achieves data rates of 6–27 Mbs/s. The manner in which pertinent information is disseminated throughout the vehicular environment is also an important
aspect in vehicular networks. However, dissemination is usually confronted with two major problems: (i) on one hand, in case of dense traffic, bandwidth proves to be insufficient and it is difficult to limit the packet losses, (ii) on the other hand, if the traffic density is low, temporary disconnection in vehicular network will be unavoidable. The main challenge in vehicular ad hoc networks is the collection of information like accident, speed limit, any obstacle on road, road condition, traffic condition, commercial advertisement, etc, for the safety and non-safety aim. In many dissemination techniques, the vehicle carries the packet until it finds any other vehicle in his range and then it forwards the packet to that vehicle. Data dissemination in VANETS is a complex subject that is linked to the MAC and routing issues and brings additional possibilities such as the use of the infrastructure support and the possibility of aggregating data. Our purpose is to review the latest techniques of safety and non-safety messages dissemination in VANETS
The remainder of this paper is organized as follows. Section II showcases several mechanisms of emergency data dissemination in VANETS. Section III provides various approaches of non-emergency message dissemination in VANETS. Finally, the paper is concluded in Section IV.
II. EMERGENCY MESSAGE DISSEMINATION There are three major scenarios in which safety applications could be very useful: • Accidents: safety application could be used to warn
cars of an accidents that occurred further along the road.
• Intersections: Driving near and through intersections is one of the most complex challenges that drivers face because two or more traffic flows intersect, and the possibility of collision is high
• Road congestion. Many dissemination protocols have been proposed to perform road safety services, they have to respect the delay and delivery ratio constraints.
In [1], authors propose a zone forwarding scheme for
information dissemination in VANETS (ZBF) which provides both a spatial and temporal retention of information and using less communication overhead among the nodes. ZBF divides the effect area of the alert message into segments of length R (R is the transmission range) where an elected vehicle which is a forwarder in each segments is in charge of broadcasting the information to other neighboring vehicles. When the forwarder leaves the zone, a new forwarder will be elected. Simulation results show that ZBF outperforms other information dissemination protocols.
Y. Wang et al. in [2] propose an improvement of the
GPSR protocol by enhancing the decision-making of data delivery. In fact, they use the concept of vector, called also Greedy mode, to choose the next relay to enhance the
accuracy of the routing scheme. For intersections scenarios, they add a predictive mode to predict the motions of neighboring vehicles. Simulations reveal that their technique outperforms GPSR protocol in terms of packet delivery ratio and routing overhead.
In [3], the dissemination protocol proposed should be
able to disseminate different types of events (an accident, an emergency braking, an available parking slot, etc.) in the intervehicle network. The dissemination approach considers the relevance of the data, represented by what they call encounter probability, to decide when rediffusion is needed. The protocol is able to disseminate data about any type of event in the network by setting appropriate weights for the different factors that affect the computation of the encounter probability. when a vehicle receives a message, it computes the EP for the event received. If the EP computed is bigger than a certain diffusion threshold, the message is relevant enough to be rediffused by the receiving vehicle. Otherwise, the message is ignored.
In [4], Chu et al designed an overlay solution that
considers the challenges of propagating multisource information caused by vehicle mobility and multihop forwarding. They used an overlay concept to design an information reduction (IR) scheme and a vehicle adaptive cluster-to-cluster multihop forwarding method (VAC). The IR scheme creates a multihop propagation framework that is composed of a mobility-adaptive clustering protocol and an effective IR method. The mobility-adaptive clustering protocol works to simplify the diverse neighboring relationships among vehicles and groups of vehicles of similar trajectories in short intervals on highways to form a transient clustering infrastructure and designs the IEC (Information Equivalence Class) objects to reduce broadcast information volume. Based on the IR scheme, the VAC method uses a bidirectional and cluster to cluster selective forwarding method to multihop propagated traffic information along highways. The authors explored protocol overhead, broadcasting reduction ratio, and technological deviations to verify the advantages and efficiencies of the solution.
The authors in [5] propose an algorithm which identifies the Multi-Point-Relay (MPR) of OLSR that are geographically behind a particular vehicle. They focuses on directed traffic information propagation in order to avoid network congestion and delay. The authors utilize the idea of multipoint relays (MPRs) to propagate accident information in a restricted way. MPR’s utility can be realized through reduced control message overhead to generate routes for OLSR. An ad-hoc node’s control message is only broadcasted by its selected MPRs (all other nodes do not retransmit the message), thereby reducing network traffic.VANET mobility models generated by SUMO into NS-3 is used for the experiments.
A novel mechanism, called the Density-aware Emergency message Extension Protocol (DEEP) is proposed in [6] to disseminate emergency messages in VANETs. DEEP resolves the broadcast storm problem, achieves low dissemination delay, and provides high reliability over a realistic multi-lane freeway scenario. The freeway is divided into three segments when an accident happens. The hot spot area is defined as the rectangular area before the exit. The exit is a branch of the freeway and it can split the traffic for avoiding the traffic jams. The mechanism delivers emergency messages to a hot spot in a timely manner and guarantees that all relevant vehicles in that area will receive the messages. Drivers can then change their routes and avoid getting caught in a traffic jam caused by the accident.
The protocol presented in [7], termed SIFT: SImple
Forwarding over Trajectory; requires the source node to encode a geographical curve, referred to as a trajectory, into the packet header. In SIFT, the authors define a new routing protocol category: Timed Trajectory-Based Forwarding (TTBF) protocols. Based on timers, they allow the most-suitable node to forward the packet and to suppress other potential forwarders. The principle is to assign shorter retransmission timers to better relays. Hence, the best relay’s retransmission timer will expire before all other possible relays and it will retransmit the packet first. This retransmission will be overheard by the other potential relays which will stop the retransmission timers and give up retransmission. This technique is compare with a pure flooding scheme. For the evaluation, they examine various parameters such as inter-vehicle distance, transmission radio range and traffic load. Experimental results, obtained by several simulations implemented on VanetMobiSim/NS-2 platform, show the effectiveness of trajectory-based solutions for efficient dissemination of emergency messages in VANETs. They also conclude that a simple flooding technique is not always considered harmful, and can outperform more complicated schemes in high density VANETs environment.
III. INFOTAINMENT MESSAGE DISSEMINATION User applications can provide road users with information, advertisements, and entertainment during their journey. The dissemination protocols used in such class of services have no strict constraints in terms of delay and delivery ratio; however, they have other constraints related to the bandwidth use. Several protocols for non-safety message dissemination in VANETS have been proposed.
In [8], the authors proposed a new data dissemination
protocol for vehicular networks in both highways and urban environment. This protocol called ROD (Road Oriented Dissemination) and prior simulation studies have shown excellent performances. the GPS position of the vehicles, the outgoing and the ingoing intersections’ positions are encoded in the same header of the broadcast message to distinguish the dissemination direction. Then, timing is used to select, in a distributed way, the best re-transmitter vehicle.
TABLE 1: COMPARISON OF VARIOUS EMERGENCY DATA DISSEMINATION PROTOCOL IN VANET
Name of Protocol Environment
Simulation Platform
Used
Metrics Used for Evaluation
ZBF Highways and Urban
ns-2 and VanetMobiSim
Total no. of vehicles informed
improve version of GPSR protocol
Urban area MOVE and SUMO
Packet delivery ratio Routing overhead Throughput
VAC Highways ns-2
overhead, broadcasting reduction ratio, and technological deviations
directed traffic information propagation
urban NS-3 Number of back MPRs (Multi-Point-Relay)
DEEP Highways NS-2
the average propagation delay delivery ratio the saved broadcast ratio
SIFT high density VANETs
VanetMobiSim/NS-2
inter-vehicle distance, transmission radio range and traffic load
This mechanism is used to optimize data dissemination in
road sections and in intersections. If no retransmitting vehicle is found, the vehicle in charge of the message uses the Store and Forward module to keep data until binding a better re-transmitter. The function of ROD is described briefly in Figure 3.
Figure 3: Dissemanation process
In [9], DHVN is proposed. It is a distance-based protocol that aims to support an effective and optimized way to propagate infotainment information in both highway and urban environments and takes into account roads’ structure and vehicles’ heterogeneity to provide a higher chance for
vehicles with good dissemination properties (buses, trucks, etc.) to be elected as relays. On the same road, DHVN disseminates the packet in the two directions. Each receiver on the same road triggers a timer based on the distance from the sender. It retrieves the sender position information from the packet header and calculates the backoff timer; this timer takes into account the vehicles’ height as higher vehicle covers a large area and therefore improves the communication range compared to a regular vehicle. Once the relay arrives to the intersection zone and broadcasts the message, all vehicles receiving the message take it into account. One relay is elected for each road and each direction to propagate the message. This should enhance the delivery ratio and latency and avoid packet losses if obstacles are around roads.
(SODAD): In this protocol [10], roads are divided into
segments of known length. Each vehicle generates new information for all segments in transmission range and collects the data on its current segment either by sensing the information itself or observing what the other vehicles report. A data-abstraction process leads to the scalability of the information system and the nature of the aggregation function depends on the application. An adaptive broadcast scheme is used by vehicles to adapt their transmission behavior based on segment information broadcast by other nodes in order to reduce the number of redundant rebroadcast packets. In fact, the information received will be characterized as one of these two events: (i) provocation and (ii) mollification. A provocation event is defined as an event that will reduce the time until the next broadcast, whereas a mollification event is defined as an event that will increase the time until the next broadcast. When a vehicle receives a packet, it will determine whether a provocation or a mollification event has occurred. This is done by assigning a weight to the received packet. A weight is computed from the discrepancy between the received data and those in the vehicle’s knowledge database. If the newly received information is considered newer than that in the database, then the assigned weight will be high. Based on the weight of the packet, a node determines whether a provocation or mollification event has occurred by comparing it to a threshold. The time until the next rebroadcast is increased or decreased based on the weight. In this study, the performance of the interval adaptation scheme is compared with the static scheme in both simulation and in a prototype system. The results confirm that using the adaptive scheme can reduce the number of packet collisions caused by the static periodic broadcast scheme.
In [11], the hybrid data dissemination scheme that
capitalizes on the strengths of both store-and-forward and multi-hop broadcasts and mitigates the weaknesses by combining these two categories.is proposed. The field relative to each vehicle is partitioned into a broadcast zone and a store-and forward zone. Within the broadcast zone, data is disseminated via multi-hop broadcast. At the end of the broadcast zone, the message is no longer retransmitted and instead it progresses away from the broadcast zone via
store-and-forward. The results of the simulation confirm the expected behavior of the algorithms. To find the ideal multi-hop broadcast propagation radius R (broadcast zone radius) for the hybrid data dissemination scheme and to allow applications to choose acceptable levels of latency and data bytes and an acceptable level of tradeoffs, the hybrid protocol behaves like a slide rule that allows you to select a point between the two protocol choices (MHB and SF) where a point is selected between zero (pure SF) and infinity (pure MHB). With the capability of dynamically setting R, the hybrid protocol can give the application a broad range of performance choices based on current network parameters.
In [12], authors introduce a periodic data dissemination
protocol for non-safety applications which distributes data utility fairly among vehicles with conflicting data interests FairDD: (Fair Data Dissemination). FairDD disseminates data through periodic data messages according to a defined application cycle and provide means for keeping the network load under a defined value by suppressing only the least relevant data messages, it consists of two main components: (i) a distributed fair data selection mechanism which depends on the current contextual knowledge acquired by each vehicle and (ii) a synchronized suppression mechanism to cancel only the least relevant data messages. The same authors propose FairAD (Fair and Adaptive Data Dissemination) in [13]. FairAD aims to achieve a fair distribution of data utility throughout the network while controlling the network load. The protocol relies only on local knowledge to achieve fairness with concepts of Nash Bargaining from game theory. Simulation results show that our algorithm presents a higher fairness index and yet it maintains a high level of bandwidth utilization efficiency compared to other approaches. In addition, the rate of transmissions is adaptively controlled as new information about the environment is collected. It consists of two main components: (i) a distributed fair data selection mechanism based on FairDD and (ii) an adaptive periodic protocol based on ATB (Adaptive Traffic Beacon) to control the rate at which messages are broadcast into the network. It is designed to ensure a congestion-free channel by preventing packet loss (collisions) while reducing the messages’ end-to-end delay.
In [14], the authors define a new beaconless
dissemination techniques, based only on the position of the vehicles (without the use of extra relay devices) and able to extend the Road side unit service (RSU which is the device located near the road) area up to several tens of times the radio coverage area of RSUs and to assess the proposed solutions under a general geometrical analysis; thereafter to verify the proposed solutions in dynamic scenarios with realistic vehicle and traffic conditions simulated via SUMO under a continuous message flow.
In [15], W-HCF (WAVE-based Hybrid Coordination
Function) protocol is proposed. W-HCF relies on a
controlled channel access procedure, that differently from other attempts to introduce a centrally-controlled access in VANETs, considers the multi-channel WAVE operation, and leverages on the vehicles’ position-awareness for estimating the connection lifetime and for tuning the resource reservation duration accordingly. Second, W-HCF makes use of the simplified Basic Service Set initiation procedure envisioned in the 802.11p/WAVE specifications that is however enriched with new features in order to advertise channel reservation information through a fully distributed gossiping scheme.
TABLE 2: COMPARISON OF VARIOUS INFOTAINMENT DATA DISSEMINATION PROTOCOL IN VANET
Name of Protocol Environment
Simulation Platform
Used
Metrics Used for Evaluation
ROD Highway and Urban
VehicleMobiGen and NS-3
Saved rebroadcast ratio Packet delivery ratio
DHVN Urban NS--3
Packet Delivery Ratio Transmission duplication
SODAD Highway NS-2
Packet Drop Ratio, End to End Delay
Hybrid data dissemination
Highway and Urban JiST/SWAN
the amount of physical layer bytes the age of the data
FairDD Highway and Urban OMNeT++
Sum of utility gains Utility per data messages received Delay
FairAD: Highway and Urban Environments
OMNeT++ and SUMO
Utility per data messages received Total number of transmissions Delay
Infotainment traffic flow dissemination
Urban NS-2, SUMO and MOVE
Information Coverage the overall number of packets
W-HCF Highway and Urban Environments
NS-2 and VanetMobiSim
throughput, goodput, and delay
IV. CONCLUSION Data dissemination is the transportation of information
from the source to the intended destination while meeting and satisfying some requirements such as delay, reliability, and constraints related to the bandwidth use, and so on. These requirements vary, depending upon the data being disseminated. Moreover, data dissemination in VANETS usually cannot use group addresses but have to rely on the locations of nodes to determine the nodes in the delivery group.
In this paper, we focus on how ITS and, specifically, VANETs can contribute via several research to the development of protocols and mechanisms that improve or solve the problems related to transportation systems. We present an overview of some of safety and emergency messages dissemination in VANETS in section II and non-safety messages dissemination in VANETS in section III.
REFERENCES [1] R. P. Singh et A. Gupta, « Information Dissemination in Vanets
using zone based forwarding », in Wireless Days (WD), 2011 IFIP, 2011, p. 1–3.
[2] Y.-B. Wang, T.-Y. Wu, W.-T. Lee, et C.-H. Ke, « A Novel Geographic Routing Strategy over VANET », in 2010 IEEE 24th International Conference on Advanced Information Networking and Applications Workshops (WAINA), 2010, p. 873‑879.
[3] N. Cenerario, T. Delot, et S. Ilarri, « A Content-Based Dissemination Protocol for VANETs: Exploiting the Encounter Probability », IEEE Trans. Intell. Transp. Syst., vol. 12, no 3, p. 771‑782, sept. 2011.
[4] Y.-C. Chu et N.-F. Huang, « An Efficient Traffic Information Forwarding Solution for Vehicle Safety Communications on Highways », IEEE Trans. Intell. Transp. Syst., vol. 13, no 2, p. 631‑643, juin 2012.
[5] A. K. M. Hossain, P. Mekbungwan, et K. Kanchanasut, « Directed information dissemination in VANET », in Proceedings of the 7th Asian Internet Engineering Conference, 2011, p. 30–37.
[6] M. Chuang et M. Chen, « DEEP: Density-Aware Emergency Message Extension Protocol for VANETs », IEEE Trans. Wirel. Commun., vol. Early Access Online, 2013.
[7] N. Ababneh et H. Labiod, « Safety message dissemination in VANETs: Flooding or trajectory-based? », in Ad Hoc Networking Workshop (Med-Hoc-Net), 2010 The 9th IFIP Annual Mediterranean, 2010, p. 1‑8.
[8] M. O. Cherif, S.-M. Secouci, et B. Ducourthial, « How to disseminate vehicular data efficiently in both highway and urban environments? », in 2010 IEEE 6th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), 2010, p. 165‑171.
[9] S. Mehar, S. M. Senouci, et G. Rémy, Dissemination Protocol for Heterogeneous Cooperative Vehicular Networks. .
[10] L. Wischhof, A. Ebner, et H. Rohling, « Information dissemination in self-organizing intervehicle networks », IEEE Trans. Intell. Transp. Syst., vol. 6, no 1, p. 90‑101, 2005.
[11] M. Rathod, I. Mahgoub, et M. Slavik, « A hybrid data dissemination scheme for VANETs », in Wireless Days (WD), 2011 IFIP, 2011, p. 1‑7.
[12] R. S. Schwartz, A. E. Ohazulike, et H. Scholten, « Achieving Data Utility Fairness in Periodic Dissemination for VANETs », in Vehicular Technology Conference (VTC Spring), 2012 IEEE 75th, 2012, p. 1‑5.
[13] R. S. Schwartz, A. E. Ohazulike, C. Sommer, H. Scholten, F. Dressler, et P. Havinga, « Fair and adaptive data dissemination for traffic information systems », in Vehicular Networking Conference (VNC), 2012 IEEE, 2012, p. 1–8.
[14] P. Salvo, M. De Felice, F. Cuomo, et A. Baiocchi, « Infotainment traffic flow dissemination in an urban VANET », in Global Communications Conference (GLOBECOM), 2012 IEEE, 2012, p. 67–72.
[15] M. Amadeo, C. Campolo, et A. Molinaro, « Enhancing IEEE 802.11p/WAVE to provide infotainment applications in VANETs », Ad Hoc Networks, vol. 10, no 2, p. 253‑269, mars 2012.
Mohammed Zakarya Baba-ahmed, Fatima Zahra
Marouf, Nawel Seladji Dept. Electrical and Electronic Engineering
Laboratory of Telecommunication of Tlemcen, LTT Tlemcen, Algeria
{zaki.babaahmed, Fatimazahra.marouf}@gmail.com [email protected]
Abstract — Radio Frequency IDentification (RFID) is a technology which allows identification of objects by exchanging data through radio frequency signals. This technology is also known as contactless identification. The RFID system is characterized by a base station or reader seeking to identify objects and a label (tag) which is inserted at these objects. Through this work, we are interested only in RFID tags. These last contain an antenna which insures communication with the reader, and an electronic chip in which data will be stored. This paper is limited to study and analyze some printed antennas embedded in passive RFID tags operating in UHF (Ultra High Frequency) band by two means IE3D ( Integral Equation Three-Dimensional) and CST(Computer Simulation Technology). We show in this work the importance of ensuring a match between the input antenna impedance and the chip impedance to improve efficiency of the link. All these parameters are elements to consider when designing an antenna for an RFID system. Keywords — RFID; Printed antennas; Tag; Reader; Chip; UHF band; Matching.
I. INTRODUCTION
To date, contactless identification is a mature industrial technology; however only few applications such as panels RFID (Radio Frequency Identification) in Super and Ultra High Frequency (UHF and SHF) are known [1].
RFID tags have a key organ, ensuring its operation. This element is a chip antenna. It is charged to make communication between RFID tag and RFID base station (Reader).
The printed antennas are designed in order to facilitate their implementation. These antennas have emerged in the 50s but the use of these antennas was limited to scientific or military experiences (in the 70s). We had to wait the 90s to see what type of antenna switch to industrial, including the launch of RFID technology.
In this paper, we are interested to analyze PIFA (Planar Inverted-F Antenna) patch antennas applied to RFID tags. We will firstly explain the RFID technology, we define its functionalities, explain also the communication between the RFID reader and the tag then we will give an architectural detail on RFID tags in the UHF band. In an other hand we reveal the results obtained by the analysis of some types of printed antennas using two powerful software:
Mohamed Ryad El-Mansour Fekkar
Dept. Electronic Laboratory of Signal and Image, LSI
Oran, Algeria [email protected]
The IE3D and CST, based on the Method Of Moments (MOM) and Finite Integration Technic (FIT) respectively. The obtained results show the importance of ensuring the match between the antenna and the chip for the RFID tag, and the influence of surrounding materials (background) on the behavior and performance of the tag. We make a comparison of results between two software’s IE3D and HFSS (Hyper Frequency Structure Simulation) in the first part and between two antennas (found on literature) simulated by CST software in the second part.
II. RFID TECHNOLOGY
A. components of an RFID system An RFID system consists of two entities witch communicate with each other
A tag or intelligent label (a transponder): it is associated with the element to identify. It is able to respond to a request from a reader.
A base station or RFID readers: its mission is to identify the tag. The reader sends an electromagnetic wave in the direction of the element to be connected. In return, it receives the information sent by the tag. [2]
The following figure shows the overall operation between a reader and RFID tag.
Figure 1. Operation of an RFID system [2]
. B. RFID tags
RFID tags are small miniaturized mobile devices which include an antenna associated to an electronic chip which
Printed Antennas for UHF RFID Passive Tags
can respond to requests sent from a transceiver. These tags exist in the form of self-adhesive labels or may be included in objects molded plastic. In its various forms, the principle is the same: the tag communicates with a reader using radio frequency short pulses [3]. An RFID tag is composed of several parts: the chip, the carrier or substrate (paper, cardboard or plastic) and an antenna. Figure 2 shows the three basic components of a label (tag) RFID.
Figure 2. RFID TAG UHF [4]
1) The chip To date, the technology of integrated circuits is required for RFID tags, it is mature and industrialized. New technologies for non-volatile memories are utilized to store a few bits of data to several kb. Researches continue on more efficient low-power silicon technologies like Silicon On Insulator (SOI). A wide range of products exist, the easiest one is to read a unique identifier, to very complex with multiple communication interfaces and different types of sensors (temperature, pressure, etc ...) [3]. 2) The antenna The antenna is a fairly straightforward circuit designed to produce a given radiation. In the low and high frequency bands, the coupling is magnetic. The situation is different in the UHF and SHF bands above where the coupling involves both a magnetic field and an electric field. The operation of the antennas is more difficult to implement on these frequencies. The design of an antenna will depend on the intended application and environment existing during the operation of the drive system. An important parameter is the quality factor which determines the bandwidth of the antenna. It should not be too large, which causes degradation of the received signal due to noise, or too narrow because the risk of not receiving the signal [3]. 3) The substrate The substrate is a non conducting material. This is where the waves propagate transmitted or received by the antenna, it is characterized by its relative permittivity and loss tangent tan . Material may be paper, cardboard or plastic, etc. .... C. UHF RFID Tags
Bands UHF offer less signal penetration through obstacles that the HF band, but they achieve distances greater speed reading and exchange important information. We can differentiate RFID UHF systems into two types:
operating systems without chip (chipless) and systems operating with tags possess a microchip. [4] 1) Systems without chip These are tags that do not have electronic circuit, using physical or chemical principles for generating an identification code (ID). An example of chipless tag is the tag SAW (Surface Acoustic Wave). This type of tag is constituted of the reflector that transforms acoustic waves in the radiofrequency waves (and vice versa). [2] This technique works very well at 2.4 GHz. At higher frequencies, the losses are too great for this principle to be used. 2) Systems with chip We can differentiate UHF RFID systems consisting of a chip depending on the power source of the tag. We distinguish three categories: active tags, without battery assisted passive tags and battery assisted passive tag. [2]
Active tags Active tags have a power supply, a transmitter and a radio receiver which their own. They are mainly used in applications ranging, for a large number of communicating information over large distances. Their cost is very high. This type of transponder has an on-board battery which provides power to the integrated circuit and the transmission / reception of a signal. It can achieve working distances of the order of several hundred meters. [2]
Passive tags without battery assisted Passive transponders have a simpler design and allows for much lower unit costs. They have no battery or radio frequency transmitter and thus does not generate radio frequency wave. Their circuits feed from radiant electromagnetic energy they receive from RFID readers and they collect their antennas. They use the principle of retro - reflection modulation to transmit their data to readers. As against, reading distances of such tag are much shorter than in the case of active transponders: 3 to 10 meters depending on the transmission power of the RFID reader. [2]
Passive tags with battery assisted In the same series, for information, the "semi- active" tags or "semi- passive" do not exist. How many times have we heard these words that make no sense? If under these terms we want to talk tag ensuring their descendants communications passively (without transmitter) and equipped with a battery, it is very simple, it is simply assisted passive tags "battery assisted" battery. [1]
D. Passive UHF RFID Tags
We distinguish two main protocols for communication between a tag and an RFID reader: TTO protocol (Tag Talk Only, meaning that only the tag transmits data) and RTF protocol (Reader Talk First, meaning that the drive is master in communication). The selection of a protocol over another depends on the intended application. In TTO protocol, there is no uplink. Using this procedure a tag transmits its data on a regular basis when it is powered. This procedure is very fast; it can read a large number of RFID tag and at speeds of nearly 250 km / h.
In the protocol RTF, RTF tag when entering the field of a reader, it waits for a request before transmitting its identifier. [2]
III. ANTENNAS GEOMETRY In this article, we propose to study two types of antenna namely circular Antenna PATCH and Dipole Antennas printed tri-band which are highly recommended by passive RFID tags for UHF band.
A. Printed Antenna (Patch Antenna)
With the miniaturization of radio systems, we need now to develop less bulky antennas possible with a sufficiently high yield. Patch antennas (or called printed antennas) meet these requirements. In its simplest version , patch antenna , or printed antenna , shown by Figure 3 , is composed of a plate substrate entirely metalized on one side , while a metal film of variable shape ( here square) and size adjusted is placed on the other side thereof. The latter constitutes the radiating element; the dimensions and characteristics of the substrate on which it is deposited attach, inter alia, the frequency of resonance. The metal ground plane is sufficiently large relative to the radiating element so as to limit the effects of surface waves which radiate on the ends of the plate. As a substrate, there are composite bases of glass fibers and Teflon (polytetrafluoroethylene 2 < r < 3, tan = 10-3), polypropylene ( r = 2.2, tan 3.10-4) but also foams containing many air pockets ( r = 1.03, tan 10-3) [5].
Figure 3. STRUCTURE OF A PRINTED ANTENNA [5]
Typical dimensions are a patch antenna has a length L, width W and the thickness h of the substrate. From a practical point of view , the latter is usually very thin and less than the working wavelength ( h < 0 , 0.05 0 ) , the 0 representing a wavelength in vacuum A printed antenna may be considered as a resonant cavity made up of four magnetic open side walls and two horizontal walls electrical . The radiation is caused by the leakage field at the ends between the metal patch itself and the ground plane. The operation of the antenna is then illustrated by the equivalent of two slots radiating edges and separated by the distance L. [5]
Circular Antenna PATCH
We describe now a passive RFID tag in the UHF band, designed to be placed on different metallic surfaces, mainly metal containers. The label is designed to cover the ETSI band ( 865-870 MHz ) and the band ( 902-928 MHz) used in USA for providing RFID and reading ability in normal or lateral position with a scope that s' extends up to 7 m in the
normal direction and up to 4 m for the lateral directions[6] . Figure 4 shows an example of this application.
Figure 4. THE DIFFERENT POSITIONS FOR READING A RFID TAG [6]
Figure 5 shows the actual label used in this sample application. [6]
Figure 5. REAL STRUCTURE OF THE RFID TAG [6]
The tag is composed of a circular patch antenna fed by a coplanar waveguide. At resonance, this structure has been adjusted to have complex conjugate equality between the input impedance of the antenna and the impedance of the chip, which allows a broadband of tag performance.
For reasons of simplicity, this structure is made by metallization which is a flexible folded about a rectangular dielectric (the dielectric permittivity of HDPE 2.5) as shown in Figure 6. The patch antenna is then placed on a block of rigid plastic (made of Lexan 945 with a permittivity of 2.7). The RFID chip used (G2XM NXP) has an input impedance Zc = 26 - j 150 ohms at 900 MHz Analysis of this structure will be the tool IE3D electromagnetic simulation. We seek to determine the input impedance of the antenna to ensure its adaptation to the chip. We will consider the case where the tag is mounted on a metal surface of 30 cm x 30 cm, as shown in Figure 6. [6]
Figure 6. Geometry of the proposed tag for printed antenna [6]
Dipole Antennas printed tri-band
The second structure to which we are interested is a tri-band printed dipole designed for RFID tags. This antenna can operate at frequencies 0.92 GHz, 2.45 GHz and 5.8 GHz. The simulation of this structure will be completed by CST software. For the tri-band nature, the dipole has two legs, serving as the two additional resonators involving both 5.8 GHz and 2.45 GHz frequencies. We consider two configurations, one with straight legs and the other with folded arms. [7] They are represented by the following figure:
Figure 7. Real structures of the tri-band dipole antenna [7]
The tri-band dipole is printed on a substrate of Taconic (dielectric constant r = 3.5, thickness = 0.508 mm and loss tangent = 0.0019). The dimensions of the substrate are 145 mm x 20 mm for the right dipole (Figure a) and de105 mm x 20 mm for the meandered dipole (Figure b). As we can see, the bend of branches serves to reduce the size of the antenna. An area of 2.0 mm in the center of the antenna is reserved for the insertion of the chip. In our analysis, we will replace it with a discrete power 50 port. The following figure shows the data for the first configuration (a) (l1 = 67.0 mm, l2 = 18.1 mm, l3 = 11.6 mm, br = 4.5 mm and h = 23.0 mm) and those of the second configuration (b) with: (77.3 mm l1 = l2 = l3 = 18.1 mm and 10.6 mm, br = 3.5 mm and h = 23.0 mm, the width w = 1.2 mm ). [7]
Figure 8. Geometry of the proposed tri-band dipole antenna [7]
IV. ANTENNAS RESULTS AND DISCUSSIONS
A. Case of a circular antenna PATCH The following figure shows an outlet 3D structure studied by IE3D software.
Figure 9. The patch antenna seen in 3D
Since the impedance of the chip is defined by Zc = 26 - 150 j, then we should estimate the input impedance of the patch antenna Ze = Zc = 26 + j * 150. The figure 10 shows respectively the real and imaginary part of the parameter Z11 found with IE3D software.
Figure 10. Clipping the real and imaginary part of the parameter based on
the Z11 frequency
We now see the results presented by [6] with the HFSS software. The latter is illustrated by the following figure.
Figure 11. The Z11 parameters obtained by HFSS software [6]
Interpretation and comparison between the two
results
We have seen in the second part, the RFID tag has an antenna and a chip on board. Consistency between these two entities is their impedance inputs that achieve an adaptation or not. In this example we have demonstrated that consistency between the circular patch antenna and the integrated chip. The results by the software IE3D coincide with that found in HFSS where the input impedance of the antenna is
approximately equal to the conjugate of the impedance of the chip (26.8 to 159.8 + j and IE3D 25 + j 90 for HFSS). With regard to the resonance frequency, it is in the vicinity of 950 MHz for IE3D, which is different from that found in HFSS (at around 900 MHz). This frequency shift is due to the RAM of our computer which does not support the study of an antenna deposited on a large area (more than 200 mm) with a high frequency of the mesh. B. Case of two tri band printed antennas Figure 12 shows us the first structure built with CSE software.
Figure 12. Structure of Figure (a) taken by CST Software
The reflection coefficient S11 presented in Figure 13 shows the result obtained by the simulation of the antenna (a).
Figure 13. Comparison plots of S11 parameter as a function of frequency
for the first Antenna The result presented in the article [7] shows a first frequency band 920 MHz band a second resonance at 2.45 GHz and a third frequency band is 5.8 GHz, as can be seen by the figure13 to the right.
We did the same for the antenna (b) carried out by the CST software illustrated in Figure 14.
Figure 14. Antenna structure (b) CST software
For the second structure (Antenna b), the simulation parameter S11 is shown in Figure 15. We give in the following figure the result of reference [7] (right figure) the parameter S11. Here too, we notice the presence of the three frequency bands used by RFID technology: 0.92 GHz, 2.45 GHz and 5.8 GHz.
Figure 15. Comparison plots of S11 parameter as a function of frequency
for the second Antenna
Interpretation and comparison between the two results
For the first antenna (a), we have found in the first band, an adaptation of -25.94 dB at a frequency of 915 MHz, an adaptation of -21.03 dB at a frequency of 2.425 GHz and an adaptation of -20.39 dB for a frequency of 5.75 GHz. These values are very close to those found in reference [7]. Similarly for the second antenna, we found in the first band, an adaptation of -22.69 dB at a frequency of 955 MHz and is disconnected from the one presented by the article [7]. For the second band, we found an adaptation of -12.67 dB at a frequency of 2.52 GHz which is somewhat similar to that found in article [7]. For the third band, we found an adaptation of -27.75 dB at a frequency of 5.8 GHz which is the same frequency but better adaptation than the article [7]. Comparison between the two structures, we found that the first figure gives a better fit for the first two bands, while the second figure gives a better fit for the third frequency band.
V. CONCLUSION In this article, we simulated and analyzed some types of printed antennas by two means. The choice of the analyzed structures was made to show the influence of some parameters on the performance of an RFID tags antenna. Indeed, it is not enough to determine the performance of the tag antenna in an isolated environment, as it is meant to be filed on a particular platform whose presence can affect the operation and the radiation characteristics of the antenna. On the other hand, the antenna carries at its end a load which is the chip. It is essential to ensure a match between them to receive the maximum power provided by the RFID reader and to get better quality of the link. Finally, we presented two tri- band antennas that could serve multiple applications simultaneously while reducing both manufacturing cost and size of the structure to facilitate its integration into a system RFID. All the presented simulations in this paper have been confronted to those of the literature. The results are satisfying and consistent.
References
[1]: Dominique Paret, "RFID super and ultra high frequency UHF - SHF" DUNOD, PARIS, 2008.
[2]: A. Ghiotto, T.N.H. DOAN, T.P. VUONG , L. Guilloton, G. FONTGALLAND and S. TEDJINI «Étude d’une Antenne Planaire Compacte pour Lecteurs Embarqués RFID UHF», March, 2007.
[3]: Jean-Claude Sirieys, "Light", Opticsvalley, May-June, 2006.
[4]: Anthony Ghiotti, "Conception d’antennes de Tag RFID UHF, application a la réalisation par jet de matière " Institut Polytechnique de Grenoble, November 26, 2008.
[5]: Odile Picon, "Les antennes" DUNOD, PARIS, 2009.
[6]: K. V. S. Rao, F. Sander Lam, Pavel V. Nikitin, "UHF RFID Tag for Metal Containers", Proceedings of Asia-Pacific Microwave Conference 2010.
[7]: M. Abu and K. A. "Rahim, TRIPLE-BAND PRINTED DIPOLE ANTENNA FOR RFID TAG," Progress In Electromagnetics Research C, Vol. 9, 145 {153, 2009.
New Pilot Pattern Arrangement for
Channel Estimation techniques in the Downlink of
the LTE standard
F.Zohra Bouchibane1,2
, Khalida Ghanem1
1Centre for Development of Advanced Technologies
Algiers, Algeria
Messaoud Bensebti2
2Saad Dahlab University, Algeria
Abstract— One of the crucial challenges in wireless
communication systems is to guarantee excellent performance
particularly in time-frequency varying channel. Channel
estimation has proven to be a pillar stone to achieve such a
performance and is therefore still a very hot research topic.
The aim of this paper is to highlight the role of pilot pattern
design in channel estimation stage in the performance of LTE
downlink scheme. Pilots arrangement in time-frequency grid
has a significant impact on both spectral efficiency and the
reliability of an OFDM system. The paper mainly proposes a
new pilot pattern by adjusting both time and frequency
spacing. It also provides a comparison between the proposed
design and the 3GPP pattern. Several simulation results are
described to demonstrate the effectiveness of the proposed
pattern through different propagation channels and using
different interpolation techniques.
I. INTRODUCTION
Orthogonal Frequency Division Multiplexing (OFDM) is a multicarrier modulation scheme which is incorporated to combat the non-desired effects of frequency selectivity of the wireless channel. It reaches this by converting it into a set of nearly flat narrowband orthogonal channels. Frequently, OFDM systems use coherent detection at the receiver that needs precise channel state information.
In radio mobile communication system, the quality of the received signal is degraded by the time-frequency selective nature of the channel. In order to overcome this, channel estimation must be performed. It consists on finding the time and frequency dispersive characteristics of the medium, thus allowing that the transmitted signal can be recovered with less error. This can be critical and could have a deep impact on the equalization procedure and the receiver performance [1].
Channel estimation techniques can be categorized into two types, namely training-based and blind techniques. In training- based techniques, a sequence known to the receiver is embedded within the same frame as the data and sent over the channel. In blind methods, no training sequence is required, and instead certain fundamental mathematical properties of the sent data are exploited. Either for the former or the latter technique, a large number of approaches has been proposed in that perspective in OFDM systems [2].
In training-based techniques, channel estimation using periodically inserted pilots instead of appending a long training sequence at the beginning of the transmitted signal is a thriving method for channel estimation and prediction in rapidly fading environment. In such a case, an efficient interpolation technique is required for estimating channel in frequency domain.
Actually, the interpolation technique can considerably impinge the channel estimation and should be selected adequately. When targeting the complexity reduction at the receiver level, the interpolation step could be achieved using a simple interpolator such as linear Lagrange and spline interpolation scheme [3]. In such case, the performance of channel estimation is affected by the pilot pattern (i.e., the shape and the spacing of the pilots) as well as the pilot density[4]. For instance, when the channel is experiencing fast variation with a small multipath delay spread, it would be more efficient to insert an increasing number of pilot symbols in the time domain rather than in the frequency domain. Nevertheless, pilot insertion imposes an overhead on the transmitted packet since pilot symbols do not carry information and consume a part of the power budget and frequency resources.
Fortunately, it is widely known now, that there should be a tradeoff between the number and the power of the pilot symbols and the system performance [5]. To design the pilot pattern, authors in [6] chose the minimization of the Mean
Squared Error (MSE) of a channel estimator as a cost function. They showed that equi-powered and equi-spaced pilot-symbols lead to the lowest MSE. [7] suggested a pattern design that maximized the channel capacity. In [8], authors propose adaptive pilot symbol patterns that follow channel statistics in MIMO-OFDM systems. They used the 3GPP pattern and optimize both the pilot spacing and the power without modifying the overall shape.
This paper considers the design of novel pilots arrangement for a SISO-OFDM system, by maintaining the same average power in pilot and data symbols while varying the spacing between the pilots in both time and frequency domains. The performance of the designed pattern will be evaluated in different propagation channels and when using two types of interpolators.
This paper is organized as follows. In section II, the LTE downlink system model is depicted with a particular emphasis on the pilot pattern. Simulation results are presented and commented in section III. Concluding remarks and future work are provided in the last section.
II. SYSTEM MODEL
The architecture of the downlink transmission system using LTE standard, as shown in Fig. 1, is mainly based on the OFDMA technology [9] [10] [11]. Its main objective is to reduce the receiver complexity when using high data communications while efficiently combatting channel impairments. Fig.1 depicts a typical block diagram of the transceiver system based on the OFDM technique. Complying with the 3GPP specifications for the LTE downlink scheme, the data flow is first OFDM-modulated at the base station, thus forming the well-known LTE frame which is composed of 10 subframes: each subframe is composed of two slots of 0.5ms duration. Depending on the cyclic prefix (CP) length, being either extended or normal,
each slot consists of 6sN or 7sN OFDM symbols,
respectively. By fixing the subcarrier spacing at 15 kHz, twelve adjacent subcarriers of one slot are grouped into a so-called resource block. Choosing the bandwidth of 1.4 MHz herein, the number of resource blocks will accordingly be 6. The pilot symbols, known also as reference signal are inserted at specific positions in each resource bloc as depicted in Fig.2.
For the SISO case, the pilot symbols are transmitted during
the first and the fifth OFDM symbols of each slot when the
normal Cyclic Prefix is used (NCP) and during the first and
fourth OFDM symbols when the extended CP (ECP) is
used. In this paper, only the NCP is of interest. Binary
information is then grouped and mapped according to the
chosen modulation scheme into an entity ),( knX ,
where n is the subcarrier index and k is the symbol index.
Fig. 1. LTE downlink transceiver system model.
Fig. 2. SISO Downlink reference signal in LTE standard.
Pilot symbols are inserted with data in the so-mentioned specific intervals. Afterwards, the modulated data
),( knX is converted into a time domain signal by
performing the N point IDFT and the normal cyclic prefix is added to the OFDM signal. Afterwards, the resulting signal is sent to the channel through an omnidirectional multiband antenna. At the receiver, the guard interval is removed and the DFT is performed. The received signal can then be expressed as;
.1,...,0,1,...0
),(),().,(),(
sc NkNn
knWknXknHknY
(1)
Where ),( knH represents the complex channel
coefficients for thethn subcarrier and the
thk symbol,
),( knW is the Additive White Gaussian Noise (AWGN),
and cN is the number of OFDM symbols per frame. The LS
estimator of the channel pertaining to the pilot symbols is given as the solution to the minimization problem:
ppppp
LS
P yXhXyh 12
ˆminargˆ (2)
The channel coefficients at the data symbols have to be acquired by means of interpolation. In this paper, a linear and spline interpolations are used to acquire the channel vector of the whole number of subcarriers in an OFDM symbol.
Linear interpolation is the easiest way to estimate the channel coefficients on the data positions. Two adjacent channel estimates are connected using a linear function.
12
12
11
,,
,,
ˆˆ
)(ˆˆ
pp
kpnkn
pdknknkk
hhkkhh
sps
psds
(3)
where 1pk and
2pk are the indices of adjacent
subcarriers on which the pilot symbols are located and dk
is the index of the subcarrier, on which just the data symbols are located, with the understanding that
21 pdp kkk .
Spline interpolation employs a low degree polynomial to connect the LS estimates, whereby the continuity is preserved [12].
III. SIMULATION, RESULTS AND DISCUSSIONS
In order to examine the effect of the pilot symbol pattern employed in the LTE downlink transmission system, a set of simulations has been performed in different scenarios. The most relevant simulation parameters are given in Tab. 1. Perfect synchronization is assumed and SSD (Soft Sphere Decoding) equalizer is used.
TABLE I. SIMULATION PARAMETERS
Parameters Simulation parameters
Bandwidth 1.4MHz
Channel model PedB
Sampling frequency 1.92MHz
Subcarrier spacing 15KHz
Slot duration 0.5ms
OFDM symbols per slot 7
Antenna scheme SISO
In the first set of simulations, the frequency spacing (SF) is fixed whereas the time spacing (ST) is varied while in the second set, the ST is maintained constant while the SF is adjusted. The system performance in terms of Block Error Rate (BLER), studied for the first case as reported in Fig. 3 by averaging the results over 1000 subframes, shows that, changing the ST has no effect on BER performance. This is predictable in a block fading scenario.
In Fig.4 the BER performance is reported for the second set where SF has been expressed in terms of the number of subcarriers involved. As seen from this figure, the BER performance is enhanced with the increase of SF.
0 5 10 15
10-1
100
BLER, CQI 7, PedB, 1000 subframes, fixed SF
BLE
R
SNR [dB]
LTE standard
ST=3 symbols
ST=2 symbols
ST=1 symbol
Fig. 3. BLER versus SNR for a fixed SF and variable ST.
0 5 10 15
10-1
100
BLER, CQI 7, PedB, 1000 subframes, fixed STB
LE
R
SNR [dB]
LTE standard
5 subcarriers
4 subcarriers
3 subcarriers
2 subcarrier
proposed design
Fig. 4. BLER versus SNR in the case of fixed ST and variable SF.
In the same figure, the results of applying the novel design are presented; this design consists in using two different frequency spacing (SF1=3sub, SF2=9sub) per subframe as illustrated in Fig. 5. Clearly, the proposed pilot pattern performs approximately similarly to the standard LTE 3 GPP pattern.
Fig. 5. The proposed pilot pattern design.
The effectiveness of the proposed design could be improved by varying the interpolation scheme. Aiming that, the estimation reliability of the Pedestrian B channel has been investigated using the two aforementioned linear and spline interpolation techniques and the corresponding MSE performance in terms of SNR is plotted in Fig. 6. From this figure it can be seen that the proposed pattern performs similarly to the 3GPP pattern.
0 5 10 150
0.1
0.2
0.3
0.4
0.5
0.6
0.7
MSE, CQI 7, PedB, 1000 subframes
MS
E
SNR [dB]
linear 3GPP pattern
linear new pattern
spline 3GPP pattern
spline new pattern
Fig. 6. MSE versus SNR using three types of interpolators.
Furthermore, the linear interpolator shows the best performance since the MSE levels are low compared to the spline interpolator. This is probably due to the fact that in the former the pilot symbols are located approximately close to each other.
Fig.7 investigates the BLER performance reached when incorporating the new pilot pattern as compared with the 3GPP conventional pattern, with different channel models (Pedestrian B, Pedestrian A, Rayleigh and Vehicular A).
Note that the assumption of block fading is fulfilled for the four previous channels. The results show that our proposed pattern achieves just the same performance as the 3GPP pattern. The two schemes give better results for the case of Pedestrian A and Rayleigh channels in low SNR range while they exhibit better performance in Pedestrian B and Vehicular A channels with high SNR levels.
0 5 10 1510
-2
10-1
100
BLER, CQI 7, 1000 subframes, for different channel
BL
ER
SNR [dB]
VehA 3GPP pattern
VehA new pattern
PedB 3GPP pattern
PedB new pattern
PedA 3GPP pattern
PedA new pattern
Rayleigh 3GPP pattern
Rayleigh new pattern
Fig. 7. BLER versus SNR for different type of channel applied on block fading scenario.
Next figure presents the throughput performance in the same environments as in Fig. 7 of the proposed pattern along with the 3GPP pattern. The results consolidate the same conclusions drawn in the previous case since the same trend is noted.
0 5 10 150
0.2
0.4
0.6
0.8
1
1.2
1.4
THROUGHPUT, CQI 7, 1000 subframes, for different channel
TH
RO
UG
HP
UT
SNR [dB]
VehA 3GPP pattern
VehA new pattern
PedB 3GPP pattern
PedB new pattern
PedA 3GPP pattern
PedA new pattern
Rayleigh 3GPP pattern
Rayleigh new pattern
Fig. 8. Throughput versus SNR for different type of channel applied on block fading scenario
IV. CONCLUSION
In this work, a full review of different pilot arrangements based on time and frequency spacing adjustments has been presented. A new pilot pattern is proposed and its reliability and impact on the BLER and throughput performances is investigated with two types of interpolation techniques applied in the estimation stage of different propagation channels. The simulation confirm that the proposed design allows to achieve the same performance as the 3GPP pattern. Furthermore, it has been shown that the combination of LS estimation method with linear interpolation technique offers a strong robustness to channel conditions while yielding quite viable performances and a moderate architecture complexity.
V. REFERENCES
[1] L. Ye, L. J. Cimini, Jr., and N. R. Sollenberger, "Robust channel estimation for OFDM systems with rapid dispersive fading channels," in Communications, 1998. ICC 98. Conference Record. 1998 IEEE International Conference on, 1998, pp. 1320-1324 vol.3.
[2] M. Shin, H. Lee, and C. Lee, "Enhanced channel-estimation technique for MIMO-OFDM systems," Vehicular Technology, IEEE Transactions on, vol. 53, pp. 261-265, 2004.
[3] S. Coleri, M. Ergen, A. Puri, and A. Bahai, "Channel estimation techniques based on pilot arrangement in OFDM systems," Broadcasting, IEEE Transactions on, vol. 48, pp. 223-229, 2002.
[4] M. F.-G. Garcia, S. Zazo, and J. Paez-Borrallo, "Pilot patterns for channel estimation in OFDM," Electronics Letters, vol. 36, pp. 1049-1050, 2000.
[5] E. Alsusa, M. W. Baidas, and Y. Lee, "On the impact of efficient power allocation in pilot based channel estimation techniques for multicarrier systems," in Personal, Indoor and Mobile Radio Communications, 2005. PIMRC 2005. IEEE 16th International Symposium on, 2005, pp. 706-710.
[6] I. Barhumi, G. Leus, and M. Moonen, "Optimal training design for MIMO OFDM systems in mobile wireless channels," Signal Processing, IEEE Transactions on, vol. 51, pp. 1615-1624, 2003.
[7] B. Hassibi and B. M. Hochwald, "How much training is needed in multiple-antenna wireless links?," Information Theory, IEEE Transactions on, vol. 49, pp. 951-963, 2003.
[8] M. Simko, P. S. R. Diniz, W. Qi, and M. Rupp, "Adaptive Pilot-Symbol Patterns for MIMO OFDM Systems," Wireless Communications, IEEE Transactions on, vol. 12, pp. 4705-4715, 2013.
[9] A. Ghosh, R. Ratasuk, B. Mondal, N. Mangalvedhe, and T. Thomas, "LTE-advanced: next-generation wireless broadband technology [Invited Paper]," Wireless Communications, IEEE, vol. 17, pp. 10-22, 2010.
[10] S. Parkvall, A. Furuskar, and E. Dahlman, "Evolution of LTE toward IMT-advanced," Communications Magazine, IEEE, vol. 49, pp. 84-91, 2011.
[11] S. Morosi, E. Del Re, and L. Vettori, "Advanced receiver and MIMO schemes for LTE communications system," in Wireless Communication, Vehicular Technology, Information Theory and Aerospace & Electronic Systems Technology, 2009. Wireless VITAE 2009. 1st International Conference on, 2009, pp. 217-221.
[12] N. Sun, T. Ayabe, and T. Nishizaki, "Efficient spline interpolation curve modeling," in Intelligent Information Hiding and Multimedia Signal Processing, 2007. IIHMSP 2007. Third International Conference on, 2007, pp. 59-62.
INTRODUCTION OF SPATIO-TEMPORAL ORTHOGONAL CODES IN
WIRELESS COMMUNICATIONS TRANSMISSION
Bendimerad Mohammed Yassine
Department Of Electrical Engineering
University Of Tlemcen
Bendimerad Fethi Tarik
Department Of Electrical Engineering
University Of Tlemcen
Ferouani souhila
Department Of Electrical Engineering
University Of Tlemcen
Abstract—Telecommunication systems using multiple antennas
become increasingly popular during this decade through the
results conducted by the fundamental work of Alamouti
(1998), Foschini, Gans (1998),Kammeyer and Kǘhn (2004).
The reason for this interest is justified by the spectral
efficiency afforded by deploying these types of antennas for
mobile radio systems. Mainly two independent objectives can
be distinguished: The first is to improve the credibility of the
connection, this can be achieved by enhancing the signal to
noise ratio. The second objective discusses the perfectly
throughput by transmitting multiple replicas of information on
different antennas. This paper mainly deals with various
techniques used for antennas, from traditional SISO systems to
diversity level called MIMO technology and thereafter a
comprehensive description of the encoding techniques are
presented to provide more clarity on the increased need for
these techniques in a system exploiting the spatial diversity as a
basic parameter for the transmission.
Keywords- ISI; QoS; MIMO; VBLAST; DBLAST; STBC;
OSTBC.
I.INTRODUCTION
Mobile radio communications are evolving from pure telephony systems to multimedia platforms offering a variety of services ranging from simple file transfers, audio and video streaming, to interactive applications and positioning tasks. Naturally, these services have different constraints concerning data rate, delay, and reliability (QoS). Hence, future mobile radio systems have to provide a large flexibility and scalability to match these heterogeneous requirements. Additionally bandwidth has become extremely a valuable resource emphasizing the need for transmission, schemes with high spectral efficiency. To cope with these challenges, a key area have been the focus of research in the last decade and is addressed in this paper: Multiple antenna systems.
II.ARRAY ANTENNA
An array antenna is a system consisting of elementary antennas, distributed in space, whose outputs are weighted in amplitude and (or) in phase before being summed together. A feeder can control the amplitudes and (or) the relative phases of these sources. Thanks to this control source, amplitude and phase of array antennas can produce radiation patterns with singular forms in the desired directions, as the
creation of several lobes simultaneously or a single lobe in the incident direction of the signal and zeros in the direction of interference. The antenna arrays can take different geometries: linear arrays, planar and circular arrays [1]. The total field radiated by the system is determined by the vector addition of the fields radiated by the different antenna elements.
Smart antennas, based on antenna arrays can dynamically
combine different signals and optimize the transmission
link. These antennas can be oriented according to the
preferred directions of the mobile position throughout the
communication, in order to limit interference signals. The
intelligence of the antennas is translated by:
An electronic controlled by a software radio system
associated with antennas capable to not transmit
anywhere and anyhow.
Interactions pronounced between electromagnetism
and signal processing.
A system based on the technical array antennas and
on an adaptive real-time receiver processor, which
allow modification of the radiation diagram and the
optimization of the signal output according to
predefined control algorithms.
The advantage of these systems is the ability to
automatically respond, in real time to changes in the channel
propagation. In the conventional antenna systems, each
transmitter sends its signal broadly because of the position
of the receiver is unknown. This causes pollution of the
electromagnetic environment. Instead, a smart antenna
system determines the location of the mobile to focus
energy and emit only in the desired direction [2].
A smart antenna is composed of M antenna elements, the
outputs are weighted by a complex term w before being
summed together. The weighting control unit allows to form
a "smart" radiation pattern by adjusting the amplitude (or)
the phase. The overall output of the network is:
(1)
(2)
Where is conjugate of the transpose of the weighting
vector and denotes the received signal by the ith
antenna. If we assume that the first element of the array is
the reference phase, the relative phase of the signal received
at the nth element will be:
(3)
We denote by θs the direction of the incident signal and E
the elementary diagram. Then the signal of the nth element
is:
(4)
From previous equations, we obtain:
(5)
III.MIMO ANTENNA
For a single antenna system, increasing the size of
modulation or the frequency band are the alone solutions to
increase throughput. "Talatar" and "Foschini" demonstrated
independently that the capacity of multi-antenna systems
increased linearly with the number of transmitting antennas,
significantly exceeding the theoretical Shannon limit. The
ability of these systems to resist to fading and interference is
consider as a definite advantage [3]. During transmission,
the amount of information transmitted is limited by the
channel capacity in the case where only one antenna is used
at both the transmitter and the receiver. This capacity is
given by the following formula:
(6)
To make this capacity independent from the channel
bandwidth W, the notion of spectral efficiency η has been
introduced, it follows from the quotient of the channel
capacity and bandwidth:
(7)
Physically this means, the amount of information that can
be sent through each Hertz of bandwidth. The spectral
efficiency inform about the limit at which we can exploit the
available bandwidth. This physical limit imposed by the
channel on the amount of information leads us to test
solutions that improve throughput, while maintaining a
reasonable complexity of the equipment available.
The above equation shows that an increasing in the power
of the transmitted signal is reflected by a logarithmic
increase in spectral efficiency. For example, a gain of one
bps/Hz requires double of power radiated by the transmitter.
Another case arises where we want to increase the spectral
efficiency of 10 bps/Hz, the transmitted power must
therefore be multiplied by 1000, which in some cases is not
possible. Other solutions further the transmitted power, have
been developed, among them, those that focus on the
techniques of spatial diversity in transmission (MISO),
spatial diversity in reception (SIMO) or the joint diversity
(MIMO). Spatial diversity has the potential to improve the
spectral efficiency of the system for a power transmission
initially predefined (fixed) and this goal is achieved by
introducing additional channels operated codes.
A MIMO system typically consists of m transmit
antennas and n receive antennas. The use of the same
transmission channel, allows an antenna to receive not only
the direct sent component, but also other indirect
components transmitted in favor of other antennas. The path
from the transmitting antenna one to the first receiving
antenna is denoted by , as one that connects the first to n-
th receiving antenna by , a resultant matrix is obtained by
an n × m:
(8)
The output of this system is described in terms of the entry
according to this matrix H by the following expression:
(9)
The information to be transmitted is divided into several
independent packets. the number of packets is generally
less than or equal to the number of transmitting antennas. In
the case where an asymmetric constellation antenna is in use
(m ≠ n) this number must always be less than or equal to the
minimum of the two values representing respectively the
number transmitting and receiving antenna. Theoretically,
the capacity has a linear dependence on the number of
packets M:
(10)
IV.MIMO CHANNEL
The correlation of a MIMO channel depends primarily on
the space between the antennas, the angular spread and the
mean angle between the transmitting and receiving antenna
array. Two types of representation of the MIMO channel
propagation exist: The deterministic propagation channel.
This deterministic approach retains all ray paths arriving at
the receiver. The second type of representation is the
stochastic propagation that considered independent and
identically distributed channels, and correlates them with a
correlation matrix. This stochastic approach needs the
orientation of the transmission and reception array antenna
[4].
V.CAPACITY OF MIMO SYSTEMS
The first studies on the capacity of multi-antenna systems
date back to 1987 when Winters studied the theoretical
limits of these systems in a Rayleigh environment. We mean
by Rayleigh environment or Rayleigh MIMO channel,
channel in which all paths between the transmitting and
receiving antennas are non-selective frequency, uncorrelated
and follow a Rayleigh distribution. Winters demonstrated
that with M transmit antennas and M receive antennas, M
independent channels can be established in the same
frequency band, demonstrating the great potential of these
systems. The capacity of a MIMO system is higher than the
capacity of SISO systems under the assumption of channel
fading and reception noise are uncorrelated [5]. For a
system with transmitting antennas and receiving
antennas; Consider the vector x compounds of symbols:
(11)
Figure 1. MIMO radio communication System
Corresponding to the symbols to be transmitted on each
transmit antennas for a symbol period Ts, H is the
matrix [6].
Consider another vector b, noise vector at receiving
antennas such as:
(12)
Then the received vector Y can be expressed as follows:
(13)
If Q is the matrix of size × corresponding to the
expectation of the covariance matrix of the transmitted
signal on transmit antennas:
(14)
Then the covariance matrix of the signal is normalized
such that:
(15)
If B is the matrix × corresponding to the expectation
of the covariance matrix of the noise receiving antennas:
(16)
Then by the same way, this matrix is normalized such that:
(17)
Consider that the noise vector consists of independent
Gaussian additive noise with zero mean and variance on
each division of the receiver then:
(18)
The MIMO channel capacity can be expressed as follows:
(19)
(20)
The transition from one to the other formulas is done by
applying the rule of determining the identity of:
(21)
VI.TECHNIQUES OF EMISSION FOR MIMO SYSTEMS
Some of multi-antenna transmission techniques can
significantly increase the capacity of the multi-antenna
system with comparison to the capacity of a SISO system.
Emission techniques also depend on knowledge of the
channel at the reception and / or transmission, the so-called
CSI (Channel State Information) is necessary to recall these
different transmission techniques to compare the capacities
and diversities obtained [7]. In multi- antenna systems with
CSI in reception only, as the case of spatial multiplexing,
the transmission is more robust and allows an improved data
throughput. The information to transmit, is divided into
independent packets transmitted on either the various
transmitting antennas. In the case where the appropriate
matrix H is known by the receiver system, the identification
of transmitted information is systematically. Otherwise a
series of methods are used to allow the receiver to get an
estimation on the canal. Among those the " open- loop "
method that derives its importance from the sending test
packets, said information packets, which when there are
received allow to estimate the influence of the environment
of the channel by comparison to the original information
known by the receiver. Another method also important to
mention, is the closed–loop where the receiver report the
estimated channel state to the transmitter via a special
channel said return channel[8]. The main techniques for
multiplexing space-time are developed by Bell Labs as
Dblast (Diagonal Bell Labs Layered Space- Time) or
VBLAST (Vertical Bell Labs Layered Space- Time). For
Dblast, the transmitted data stream is demultiplexed into
channels; each one treats separately, the resulting stream
allocated to the transmission antennas according to a
periodically varying order by circular permutation. This
distribution, giving the signal a diagonal structure in time
and space where the name DBLAST, this aims to ensure
equitable distribution of information over the MIMO
channel. On the VBLAST, the difference lies in flows where
the assignment is frozen in time.
VII.SPATIAL TIME CODING
The space-time coding takes part of the multi-antenna
systems with CSI at the reception only. When designing a
space-time coding coder and decoder associated therewith,
one of the fundamental questions that arise is the
performance criteria on which the study focuses [9]:
The spectral efficiency: Is expressed in bits per
second per Hertz, we aim to have the highest possible
throughput. The ultimate limit where the flow rate is
as high as possible is defined by the capacity, and the
capacity is pushed to a maximum value by adopting a
construction criterion of code, including a
maximization of mutual information between the
input and the output of system.
Robustness: the robustness of the transmission varies
often in opposite to the increased flow. It can be
measured by the bit error rate (BER); in this context,
diversity retains importance in generating codes.
Complexity: it is usually quite possible in wireless
networks that both transmitter and receiver are
powered by a battery; therefore it is important that the
complexity of the mapper is the lowest possible.
Thus, it is desirable to have a non-symmetrical design
of transmission and reception so that complexity is
less intense at the side powered by batteries.
Generally, these performance criteria are contradictory,
and the role of engineers is to face tradeoffs depending on
the application and economic aspects presented.
Space-time coding techniques can be classified into two
categories: space-time codes in Trellis (STT), the spatial-
temporal block codes (STB), and even add a third type, the
spatial-temporal frequency codes (STF). STT codes are a
generalization of trellis-coded modulations for MIMO
channels. Although the performance obtained by these
codes are excellent, the decoding complexity is exponential
with performance. The study in this paper focus specifically
on the ST block codes called STBC, which are most
interesting in practice.
VIII.SPATIAL TIME BLOCK CODING
The different replicas sent over the MIMO channel to
operate at maximum diversity, are generated by a space-
time encoder, allowing a flow of information to be encoded
in spatial dimension by the use of all the transmitting
antennas and in time by sending each symbol at several time
intervals. This form of coding is called STC (Space Time
code). Qualified by their simplicity of decoding the STC’s
are based on a coding STBC block [10].
The scheme introduced by Alamouti represents the initial
transmission technique by block coding, simplicity and
structure of this scheme led to its inclusion in the standard
W-CDMA and CDMA-2000[11]. Alamouti STBC uses two
transmitting antennas and receiving antennas with a
diversity gain of 2× , on the other hand it transmits two
symbols each two time intervals (full rate). The matrix of
this STBC coding is given by the following formula:
(22)
On all code matrix, lines represent the different time
intervals, while columns represent the transmission through
each antenna symbols. In this case at time t, the and symbols are respectively transmitted from antennas 1 and 2.
Assuming that a symbol have a length of T, at time T+ , the
denoted and symbols are then transmitted on the
antennas 1 and 2. The decoding and reception of the signal
depends on the number of receiving antennas, when the
number is reduced to one, the received signal is on the form:
+ + (23)
+ + (24)
The received signals before transmission to the decoder
will be:
= (25)
(26)
Replacing and by their values given by the
precedent equation, we get:
= (27)
= (28)
The terms represents the square of the amplitude
transfer function . Reaching this stage the calculated
symbols and allow the decoder ML (Maximum
Likelihood) to estimate transmitted symbols and
respectively. Statistical Decision ML decoder for the two
sent symbols and goes in accordance to criterion of the
two following equations, to make their most minimal
possible:
(24)
(25)
The emblem ψ is given by the following dependence:
(26)
IX.ORTHOGONAL CODES
The Alamouti scheme mentioned above is part of a
general class of spatial temporal block codes known as the
Orthogonal STBC (OSTBC). Orthogonal STBC are divided
into two classes: the real OSTBC and complex OSTBC, in
the case of complex OSTBC, the acquisition of diversity
gain and a maximum yield of code, is not certain for a
number of receiving antennas more than 2, unlike the case
of real OSTBC. For a number of transmit antennas equal to
3 Tarokh and All build block codes with yield equal to ½ ,
¾ respectively and a full diversity of [12].
X.SIMULATION RESULTS
In this section, we validate the BER performance of the
MIMO systems associated to the cases when spatial block
codes are used. The next Table summarizes major MIMO
system parameters for performance evaluation. We simulate
the effect of code type used, on the performance of data
transmission. We consider that the simulation is done with
the following assumptions:
The channel is considered narrowband, non-
frequency-selective.
Each sub-channel is a Rayleigh channel described
by the probability density and the different
moments.
The Doppler Effect is excluded when modeling the
propagation environment.
The noise at the receiver appears on the model of
additive white Gaussian noise.
Figure 4. Comparative study SISO-SIMO-MIMO
The first figure shows the different capacities depending
on the SNR for multiple values of transmitting and
receiving antennas. In the SISO case, the capacity varies
approximately from one to five bps / Hz. It remains low and
increases slowly with the SNR, which illustrates the
limitation of SISO transmissions. MIMO capacity increases
much more quickly, to finish with a gain of more than 50%
at 40 dB of SNR. Exactly the same comments can be made
on SIMO and MIMO systems for a number of eight
antennas. We can also check that the capacity of a MIMO
system for eight antennas is nearly equal to 3/2 than the
same MIMO system for four-antenna. Generally we
conclude that the capacity of transmission systems increase
with the number of antenna use for transmitting information.
We note also from the last graph that an increase in the
value of the signal to noise ratio can contribute to the
augmentation of the capacity, but this solution is always
avoided.
Figure 5. Diversity Code effect for Nr=1
Figure 6. Compare between Real and Complex STBC
The next figure shows three space-time codes for the
same number of receiving antennas, which allows us, on one
hand to compare the different space-time codes and on the
other hand to show the influence of the number of
transmitting antennas. Putting the number of transmit
antennas in this test, makes in indirectly, yield included in
the study; note that the yield of a code is calculated by the
ratio of the number of symbols sent on the number of time
intervals and do not admit values greater than unity. The
three-plotted curves visualize the difference entails
introduced by the increasing in diversity of code on its
effectiveness in terms of bit error rate, allowing to admit the
following results: more than the higher efficiency value of
the code tends to unit, lower than is the quality transmission
provided. In the end, it is interesting to note that only the
Alamouti coding can achieve yield equal to one.
The figure 6 shows a simple comparison between two
kind of orthogonal codes: the red curve represents a real
orthogonal code; the blue curve is that of the complex code,
in the last case we have chosen for the simulation the
Alamouti code. We can show from the figure that there is no
high difference between the two curves. However, in theory
it is proved that in some cases complex codes are more
efficient than the real ones.
XI.CONCLUSION
The study presented in this paper presents the outcomes
of spatial temporal block codes on the quality of data
transmission on wireless multiple antenna systems. The
whole curves and overall results agree on the fact that the
implementation of diversity in new digital transmission
systems is a necessity for these systems to face challenges in
terms of high flow data and better quality.
REFERENCES
[1] Volker Kuhn, "Wireless communications over MIMO channels: Application to CDMA and Multiple antenna systems", University Rostock Germany, 1sted, vol.2, WILEY EDITION 2006, pp. 275–328.
[2] Daniel W. Bliss, Keith W. Forsythe, and Amanda M. Chan, "MIMO wireless communications",Lincoln Laboratory Journal vol. 15, number 1, 2005.
[3] I. E. Talatar, “Capacitu of multi-antenna gaussien channel,” Eur. Transl. Telecommun, 1999, pp. 585-595.
[4] Didier Le Ruyet and Berna Ozbek, "MIMO systems and space-time coding ",Transl, Journal of electricity and electronics, 2005,vol.4,pp.69-78.
[5] S.M.Bahri "Adaptive MIMO antennas associated with multiplexing techniques and multi-carrier modulations" Doctoral Thesis ,May 2011, unpublished.
[6] John Fitzpatrick, "Simulation of a multiple input multiple output MIMO wireless system",B.Eng inTelecommunications Engineering Dublin city university school of electronic engineering , April 2004, pp.07-26.
[7] Olivier BERDER," Optimization and power allocation systems multi-antenna transmission strategies," PhD thesis, University of Western Brittany , December 2002 .
[8] S.Ferouani "Study and implementation of a MIMO OFDM system in Rayleigth envirement ", Thesis of Magister, July 2010, unpublished.
[9] Rohde & Schwarz,"Introduction to MIMO Channel, Application Note" Schwarz edition ,note 07-1MA142_0e July 2009.
[10] Gordon Stuber," MIMO Space time block coding: simulations and results", Personal and mobile communications journal, Cortes - Pena , ECE6604 , April 2009.
[11] Pascal Djiknavorian, " Advanced digital communications", PhD thesis, Laboratory of radio telecommunications and signal processing , revised edition, January 2007.
[12] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Space time block codesfrom orthogonal designs,” IEEE Trans. Inf. Theory, vol. 45, no.5, pp.1456-1467, Jul. 1999.
Concept and Design of a Transparent Security Layer
to Enable Anonymous VoIP Calls
Markus Gruber, Martin Karl Maier, Michael Schafferer, Christian Schanes, Thomas Grechenig
Vienna University of Technology
Industrial Software (INSO)
1040 Vienna, Austria
{markus.gruber, martin.maier, michael.schafferer
christian.schanes, thomas.grechenig}@inso.tuwien.ac.at
Abstract—Voice over IP (VoIP) has gained widespread acceptanceand is used for many business communications already. However,voice calls in traditional phone services, as well as in VoIPsystems, have some security flaws and therefore can be easilyintercepted, which can cause high damage by, e.g., industrialespionage. To establish secure and private phone calls, additionalnon-invasive measures are required to protect the signalingand voice channel between the parties for existing and wellknown VoIP applications. We propose an approach for secureand privacy sensitive VoIP communication by introducing anadditional security layer. The introduced security layer can beapplied to known VoIP solutions on different channels (e.g., softphones or mobile phones) and is independent from the deployedVoIP implementation in order to improve security and privacyof VoIP calls for company systems.
Keywords—Security, Internet telephony, Communication systemsecurity
I. INTRODUCTION
The wide usage of VoIP systems increases the interest of at-tackers to misuse these systems. Additionally, communicationsystems in general are of high interest for different stakehold-ers to trace connections between parties and to get to know thecontent of the communication. Often less security measuresare used for performance reasons. Another security criticalproblem concerns with the circumstance that a third-party, butalso other untrusted VoIP provider, can detect who is talking towhom, even if Transport Layer Security (TLS)/Secure SocketLayer (SSL) is used.
We propose an approach to secure VoIP communication tosatisfy privacy by establishing an non-invasive security layerbetween the network and the application layer of the Internetprotocol stack. The usage of well-known VoIP protocols suchas Session Initiation Protocol (SIP) and Real-Time TransportProtocol (RTP) ensures compatibility with a broad range ofexisting VoIP applications. The combination of state of theart cryptographic mechanisms ensure security and privacyof the communication. The presented approach provides anadditional security layer to guarantee eavesdropping-safety andprotection against caller and callee tracking and identificationvia signaling data by extending a traditional VoIP system(based on SIP and RTP). By using state of the art cryptographicmechanisms the approach provides a performant and securesolution. This allows to protect closed communication systems,e.g., companies or projects where multiple companies areinvolved.
II. RELATED WORK
Communication security of VoIP systems using encryptionhas been addressed by multiple authors (e.g., Palmieri et al.[10], Gurbani et al. [5] and Perez-Botero [11]), identifyingthat most current security mechanisms for VoIP do not protectfrom rogue proxy servers. Many security mechanisms (e.g.,TLS) do not use end-to-end encryption, allowing intermediaryproxies to have access to the unencrypted payload. To over-come this issue, Palmieri et al. [10] suggest to introduce anadditional encryption and authentication layer to SIP and RTP.Other authors add additional encryption techniques to VoIPsystems to strengthen not only security but also to addressmechanisms concerning authentication in Phil Zimmermann’sReal-Time Transport Protocol (ZRTP) (Hlavacs et al. [6]),mitigate weaknesses of key exchange mechanisms (Gurbaniet al. [4]), or enhance security by introducing point-to-pointmutual authentication as described by Yu et al. [14]. Whiteet al. [13] presented an approach for unmasking parts ofan encrypted VoIP communication, where the interaction ofvariable bit-rate codecs and length-preserving stream ciphersleaks information. Furthermore, one of the biggest disadvan-tage of ZRTP compared to this approach results from the useof the Diffie-Hellman key exchange, which does not offerprotection against Man in the Middle (MitM) attacks, andtherefore must be extended with e.g., Short AuthenticationString (SAS). SAS again does have the drawback of needing averbally cross-check by the communicating parties. Due to itsWeb-integration, Web Real-Time Communication (WebRTC)as an increasingly popular media exchange technology utilizesSSL/TLS for security and consequently inherits its securityconcerns. It is therefore not considered a viable alternative fora highly secured VoIP applications.
Gurbani et al. [5] present a security evaluation comparingprotocols, which allow to establish a shared secret usedby Secure Real-Time Transport Protocol (SRTP) for me-dia encryption, namely Session Description Protocol SecurityDescriptions (SDES), ZRTP and Datagram Transport LayerSecurity (DTLS)-SRTP. In terms of their security features,SDES has shown to be the weakest, due to transferring themedia encryption key in the SIP protocol’s Session DescriptionProtocol (SDP) sections. This ends up by exposing the key toSIP servers. Contrary, DTLS-SRTP and ZRTP come with sig-nificant computation and communication costs, being a majorcause for their lag in deployment [4]. Perez-Botero and Donoso[11] compare not only media keying protocols, but also suggest
Component Trusted Comment
VoIP Server X For this approach the VoIP Server components are
seen as trustworthy.
User Agent
(UA)
X For this approach the UA is seen as trustworthy.
Provider for
Data Commu-
nication
- Data communication providers are not considered
trustworthy. They might be compromised or forced
to give out data.
External VoIP
Provider
- 3rd Party VoIP providers are not considered trust-
worthy. They might be compromised or are victim
of abuse etc.
Other Institu-
tions
- Institutions outside the system cannot be seen
as trustworthy, as they might compromise the
communication.
TABLE I. TRUST MODEL UNDERLYING THE DESIGN OF A SECURE
VOIP SYSTEM.
Secure/Multipurpose Internet Mail Extensions (S/MIME) tobe ideal for VoIP environments, although S/MIME is not yetwidely supported in popular VoIP software. The effects ofpassive attacks such as eavesdropping are analyzed by Zhanget al. [15], who is also providing a mitigation proposal.
III. PROPOSED APPROACH FOR A SECURE VOIPCOMMUNICATION SOLUTION
When VoIP was introduced, concerns were about costs, func-tionality and reliability of the systems, and less in securityconsiderations. Since then, this changed because VoIP hasgained widespread acceptance and significance, but even moreafter the revealing of mass-surveillance and various privacyviolations, such as authorities capturing and analyzing all datain the Internet. Current and future VoIP solutions have to pro-tect users from these privacy breaches, as well as dealing withnew kinds of issues and flaws concerning secure and privacysensitive messaging and communication systems. The aim isto explicitly give data sovereignty back to the communicatingpartners. Every user in the system must be constrained toa minimal set of essential tasks and functionalities neededto meet their responsibilities. Thus, reducing the scope formalicious behavior can be achieved.
As the Guardian and Spiegel Online reported, it must beassumed that various network components (e.g., router orfirewalls), cryptographic algorithms, end-devices, as well asservice providers (e.g., email or VoIP services) or Internetproviders have been compromised. This is critical for privatecommunication, but also for companies transferring confi-dential information. Even the existence of a communicationbetween persons can uncover a lot of information, e.g., aperson calls a specific kind of medical doctor. The architectureof the approach is based on a closed VoIP system. Table Ishows the trust model underlying the design of the securitylayer for VoIP systems.
The approach described in this work, based on the trustmodel, takes the whole communication into account. Thereforesignaling as well as the voice data itself is part of the secu-rity considerations. Conventional VoIP systems are based onperformance considerations to the detriment of security, hencemost of the time only the media data are encrypted. Our pro-posed approach of a secure VoIP communication solution takesextensive security and privacy precautions for users by addinga non-invasive security layer, based on approved cryptographicimplementations, to conventional VoIP systems. In addition tothe encrypted voice data, communicating partners can not beidentified by third-parties analyzing the captured VoIP data,
as signaling data and metadata is encrypted as well. Thesecurity layer also takes into account limiting factors of VoIPsystems, as limited resources (e.g., on mobile devices for thecryptographic algorithms) or the usage in slow communicationnetworks (e.g., slow mobile data connections) require avoidingunnecessary overhead (e.g., the packet size or number ofrequests in the handshake).
A. Security Architecture Using a Non-Invasive Security Layer
We introduced the concept of an additional non-invasive secu-rity layer, located between VoIP application layer protocolsand network transport protocols. This allows independentoperation of VoIP protocols such as SIP or RTP and enablescomprehensive coverage of communication data generated andreceived by VoIP UAs. Additional to the dedicated encryptionbetween the UAs and the VoIP server, the media traffic will beend-to-end encrypted for the caller and the callee. Therefore,not even the VoIP server or any other third party can interpretthe communication data.
However, by using cryptography in the context of VoIP systemswe need a mechanism for authenticating the communicatingparties, high performant cryptography algorithms for the voiceand signaling data and a key exchange mechanism for simpledistribution of cryptography keys. Therefore, each encryptionkey used is negotiated by a key exchange mechanism thatalso ensures the authenticity of the communicating partnersand the integrity of data exchanged. Consequently, for eachconnection a dedicated symmetric encryption key, only knownby the involved communicating partners, is used. For signaling,the dedicated key is only known by the UA and the VoIPserver. For the communication data, the shared key is onlyknown by the caller and the callee. For an UA the VoIP securitylayer is required to participate in the secure communication.Without the security layer participation is not possible, asthe authenticity of the parties can not be guaranteed and theVoIP system does not interpret any VoIP packet of a non-authenticated party. This also protects from simple Denialof Service (DoS) attacks. Our proposed approach integratesPerfect Forward Secrecy (PFS) [3] for the signaling commu-nication between the UA and the VoIP server as well as the(direct) media communication between the UAs. Therefore,any communication that happened prior to a disclosure oflong term private keys can not be decrypted or successfullycrypto-analyzed by an untrusted third party. This ensuresconfidentiality of the transferred signaling and communicationdata.
When looking at the approach and the underlying VoIP system,the overall security gains can be described as follows:
• A VoIP server should only be accessible after asecure channel (considering confidentiality, integrityand availability) has been established [2] [9].
• The authentication does not rely on probable weak cre-dentials, since cryptographic material for the purposeof authentication is used. Authentication relies on thepossession of the users’ private key, and additionallycredentials transmitted by the requests.
• A VoIP server should only exchange signalling andmedia data in a confidential way, with the content notbeing disclosable to third-parties.
Fig. 1. Communication flow using the proposed security layer approach
• To protect the server from DoS the clients have toexecute cryptographic challenges to prove their work(similar to the Hashcash system [7]).
B. Security Enhanced VoIP Protocol
The security protocol implemented by the security layer isbased on a key-exchange phase before any data can be trans-mitted (as seen in Figure 1). Once the key is available on theUA and the VoIP Server, the signaling data can be encryptedand transferred to the VoIP server. Every information requiredto establish a shared secret between the UAs is transmittedthrough the secured signaling protocol. With these informationthe key for the communication encryption is derived. The keywill remain valid until the session is finished. The presentedapproach does not require an additional certificate authority,but instead the VoIP Server can be used to establish a trustrelationship between systems.
A secure VoIP communication is established by the followingsteps:
1) The signaling encryption key is generated using a keyexchange protocol (we use Fully Hashed Menezes-Qu-Vanstone (FHMQV) [12]) initiated by the UA.This enables the UA to communicate with the signal-ing server until the signaling encryption key expires(e.g., after one hour). After a signaling key hasexpired, an UA has to trigger the key exchange again,or the UA will be disconnected.
2) The conventional VoIP signaling is encrypted usingthe established cryptographic key (i.e., SIP packets).
3) The UA initiating a media session passes itsephemeral public key to the signaling protocol inorder to have the information transported to the com-municating partner, utilizing the secured signalingconnection. The static public key from the callerwill be added by the already trusted VoIP server inorder to transfer only verified public keys. Both keys(ephemeral and static public) are required to generatea shared secret.
4) The call acknowledge packet from the called UAcarries its ephemeral public key and the verified staticpublic key (added by the VoIP server). With this twokeys the shared secret, namely the communicationencryption key, can be generated. The informationexchanged includes all required information in orderto calculate a shared secret, which enables bothcommunicating partners to implicitly authenticate
Fig. 2. Structure of an encrypted data packet
each other. The communication encryption key isdiscarded once the concerned communication sessionhas been finished.
5) From now on, the communication stream is transmit-ted encrypted directly or via a RTP proxy betweenboth UAs.
The system allows to interconnect more closed VoIP systems,by establishing a trust relationship through exchanging theirstatic public keys for validation reasons. The VoIP clientsprovide a transient trust relationship to each other without theneed for additional verification by using static public keys.The distribution and the management of the static public keysis the responsibility of the VoIP servers, and therefore anadditional Certification Authority (CA) is not required. Theapproach provides a simple mechanism to extend for examplethe network to new members, e.g., project teams.
C. Cryptographic Protocol for VoIP Protection
To protect the communication between UAs and the VoIPserver as well as between UAs from both passive and activeattackers, we first establish a secure communication channelbefore starting to exchange VoIP data.
Figure 2 shows the structure of an encrypted data packet,created by the encryption function of our security layer im-plementation. This structure is tailored for usage with oursecurity layer implementation, making it light-weight andreducing unnecessary overhead. Consequently, bandwidth andresource requirements can be kept low. It consists of: KID (KeyIdentifier), IV (Initialization Vector), Chk (Check), DL (DataLength), SP (Separator), Dta (Data), Rnd (Random Data).
As the fields “KID”, “IV” and “Chk” provide data, which thecommunicating partner requires to be able to decrypt properly,those three fields are not part of the encrypted data. Due topacket size restrictions of the transport network media (Max-imum Transmission Unit (MTU) size), the allowed maximumpacket size of 1400 bytes has been defined. To reduce thepacket size, the payload data is optionally compressed usinggzip before it is encrypted (only in case the compressed data isin fact smaller than the original payload data). Especially forsmall data packets, the gzip headers may cause the compressedpacket to become bigger.
1) Key-Agreement Phase: For the key-agreement phase, weuse the elliptic curve FHMQV protocol, which is an au-thenticated key-agreement protocol based on Diffie-Hellmanscheme. FHMQV is based on carefully analyzed predecessorprotocols and the security of the algorithm was evaluatedseveral times. For this purpose, FHMQV provides strongersecurity guarantees than its predecessors and moreover ispatent-free to use which is an important aspect for building awide spread security protocol. Its implicit authentication allowsa very efficient handshake avoiding unnecessary messages.Therefore, the handshake is a lot less cumbersome than thehistorically grown SSL handshake. Due to the simplicity of
the design FHMQV avoids unnecessary complexity and error-proneness during design and implementation.
One of the main advantages of Elliptic Curve Cryptography(ECC) compared to other public-key cryptography algorithmsis its high performance and efficiency, as Bos et al. [1] pre-sented. It provides comparable security level with considerablysmaller key-size compared to Rivest-Shamir-Adleman (RSA).RSA e.g., requires for 128 bits of cryptography strength a key-size of 3072 bits. On the contrary, ECC requires only a max-imum key-size of 383 bits to provide the same cryptographystrength level.
2) Secure Data-Exchange Phase: After the key-agreementphase, the two communicating parties are in possession of acommon shared key, so that we can now switch to a moreefficient symmetric data encryption. Concretely, we use anauthenticated block cipher, namely the Advanced EncryptionStandard (AES) in Galois/Counter Mode (GCM) mode [8]with a key-length of 256 bit, which provides data integrity,authenticity and confidentiality. GCM is a very efficient high-performance mode of operation for symmetric block ciphersand it also can be easily pipelined or parallelized to boost theperformance even more. For encryption the input parametersfor the AES-GCM are the plaintext Tp, the initialization vectorIV and the additional authentication data AAD. The outputsare the encrypted ciphertext Tc and the authentication tagATAG. The authentication tag ATAG provides authenticationof the transmitted messages. Considerable advantages of GCMtowards comparable authenticated encryption modes is that itcan be used without limitations (as it is patent free, contrary toOffset Codebook Mode (OCB)) and rather simple (especiallycompared to Counter with CBC-MAC (CCM), which is knownto be overly complex).
3) Authentication of Server and Client: Our proposed solutionrelies implicit on authentication of the VoIP server as wellas the UAs using the security layer. The VoIP server aswell as the UA has to know the static public key of thecommunicating partner and the own private and public keypair (ephemeral and static). Only if the UA static public keyis known by the VoIP server the server will communicate withthe UA. If the client static public key is not known by theserver, the messages from the UA will be rejected and notprocessed by the VoIP server. Therefore the system providesstrong protection against application-level DoS attacks, thusenhancing system availability.
IV. CONCLUSION AND FURTHER WORK
We propose an approach for securing VoIP communicationsby installing an additional security layer to existing VoIPcomponents. With this generic concept it can be applied todifferent VoIP solutions with different UAs (e.g., soft phonesor mobile phones) and is independent from the respectiveimplementation of VoIP. The approach is based on trustworthyclients and VoIP Server and is therefore designed for closedsystems like company networks.
The focus of the approach is to develop a system in simplicityby combining existing cryptographic primitives and focus onhigh security, while maintaining acceptable quality of serviceof the phone calls by using high performant cryptographymechanism. In case of signaling only this special UA and the
VoIP server know the key for encryption and decryption of thepackets. In case of communication data only the caller and thecallee know the key, and not even the VoIP server is able todecrypt the communication packages. Thus, in the case of apassive eavesdropping VoIP server only the signaling data canbe interpreted, but not the communication data.
Further work concerning the presented approach resides in theoptimization of packet sizes on the media channel. This isrequired to decrease network based delays. The approach givesback the control of phone communication to it’s parties andensures therefore privacy of phone calls on the Internet. Thismight be very important for critical communications, e.g., toprotect company information against industrial espionage.
REFERENCES
[1] J. W. Bos, J. A. Halderman, N. Heninger, J. Moore, M. Naehrig, andE. Wustrow. Elliptic Curve Cryptography in Practice. IACR Cryptology
ePrint Archive, Report 2013/734, 2013.
[2] R. Canetti and H. Krawczyk. Analysis of key-exchange protocols andtheir use for building secure channels. In B. Pfitzmann, editor, Advances
in Cryptology - EUROCRYPT 2001, volume 2045 of Lecture Notes in
Computer Science, pages 453–474. Springer Berlin Heidelberg, 2001.
[3] W. Diffie, P. C. Van Oorschot, and M. J. Wiener. Authentication andauthenticated key exchanges. Des. Codes Cryptography, 2(2):107–125,June 1992.
[4] V. K. Gurbani and V. Kolesnikov. Work in progress: A secure andlightweight scheme for media keying in the session initiation protocol(sip). In IPTComm ’10 Principles, Systems and Applications of IP
Telecommunications, IPTComm ’10, pages 32–41. ACM, 2010.
[5] V. K. Gurbani and V. Kolesnikov. A survey and analysis of media keyingtechniques in the session initiation protocol (sip). In Communications
Surveys and Tutorials, IEEE (Volume:13 , Issue: 2 ), pages 183–198.IEEE, 2011.
[6] H. Hlavacs, W. Gansterer, H. Schabauer, J. Zottl, M. Petraschek,T. Hoeher, and O. Jung. Enhancing zrtp by using computational puzzles.In Journal of Universal Computer Science, vol. 14, no. 5, pages 693–716, 2008.
[7] B. Laurie, , and R. Clayton. ”proof-of-work” proves not to work. InIN WEAS 04, 2004.
[8] D. A. McGrew and J. Viega. The galois/counter mode of operation(gcm). NIST Modes Operation Symmetric Key Block Ciphers, 2005.
[9] W. Nagao, Y. Manabe, and T. Okamoto. A universally composablesecure channel based on the kem-dem framework. In J. Kilian, editor,Theory of Cryptography, volume 3378 of Lecture Notes in Computer
Science, pages 426–444. Springer Berlin Heidelberg, 2005.
[10] F. Palmieri and U. Fiore. Providing true end-to-end security inconverged voice over ip infrastructures. In Computers and Security,
Volume 28, Issue 6, pages 433–449. Science Direct, 2009.
[11] D. Perez-Botero and Y. Donoso. Voip eavesdropping: A comprehensiveevaluation of cryptographic countermeasures. In Networking and Dis-
tributed Computing (ICNDC), 2011 Second International Conference
on, pages 192–196, Beijing, China, 2011. IEEE.
[12] A. P. Sarr, P. Elbaz-Vincent, and J.-C. Bajard. A secure and efficient au-thenticated diffie-hellman protocol. In Proceedings of the 6th European
Conference on Public Key Infrastructures, Services and Applications,EuroPKI’09, pages 83–98, Berlin, Heidelberg, 2010. Springer-Verlag.
[13] A. White, A. Matthews, K. Snow, and F. Monrose. Phonotacticreconstruction of encrypted voip conversations: Hookt on fon-iks. InSecurity and Privacy (SP), 2011 IEEE Symposium on, pages 3–18, 2011.
[14] R. Yu, J. Yuan, G. Du, and P. Li. An identity-based mechanism forenhancing sip security. In Software Engineering and Service Science
(ICSESS), 2012 IEEE 3rd International Conference on, pages 447–451,Beijing, China, 2012. IEEE.
[15] G. Zhang and S. Berthold. Hidden voip calling records from networkingintermediaries. In IPTComm ’10 Principles, Systems and Applications
of IP Telecommunications, IPTComm ’10, pages 12–21. ACM, 2010.
Energy Detection Performance of Cognitive Radio Networks
Haroun Errachid ADARDOUR STIC Laboratory, University of Tlemcen, Algeria
E-mail: [email protected]
Maghnia Meliani and Mohammed FEHAM STIC Laboratory, University of Tlemcen, Algeria
E-mail: [email protected]
Abstract— Nowadays, wireless networks show limits in terms of spectral resource, data rate, and quality of service. The cognitive radio is a new paradigm in the history of wireless communications. Cognitive radio is a promising technology which increased and improved the using of the available electromagnetic spectrum of a way dynamic and opportunist. In this paper, survey of Cognitive radio technology and spectrum sensing techniques is presented. Thus, the advantages and drawbacks of spectrum sensing techniques are discussed and compared in detail of various methodologies. One of the problems of cognitive radio systems is the ability to detect the presence of the primary user with fast speed and high accuracy. This work exposes simulation performances of spectrum sensing over Rayleigh channel environment, and future research directions for cognitive radio networks.
Keywords- Cognitive Radio; Spectrum Sensing; Energy detector;IEEE 802.22.
I. INTRODUCTION In recent years, the wireless communication and the static
of the spectrum are evolving rapidly (in terms of resource spectral, data rate and quality of service) [1]. It has been observed that most of the allocated spectrum is under-utilized, such as IEEE 802.11, Bluetooth, Zigbee, and to some extent WiMAX [2]. And to solve the spectrum scarcity problem, the cognitive radio (CR) technology has been presented to tackle this problem [1]. This technology has improved the spectrum agility and system capacity of wireless systems. Particularly, she plays a pivotal role in fourth generation (4G) cellular networks, such as Long Term Evolution (LTE)-Advanced and WiMAX. A many researchers consider TV white space (TVWS) as the answer to spectrum scarcity, because a large portion of the spectrum in the UHF/VHF bands is becoming available on a geographical basis after analog-to-digital TV switchover [3]. The cognitive radio (CR) uses two major tasks which are exploration and exploitation of the spectrum. Spectrum exploration is used to identify the available and free spectrum unused of the licensed users. Spectrum exploitation is used to access the available spectrum [1].
The cognitive radio network and spectrum sensing techniques used in the IEEE 802.22 WRAN standards. The 802.22 standard is the first standard to adopt a cognitive radio spectrum sensing as a means of gaining greater use of the radio spectrum. The 802.22 network is responsible for ensuring that it creates no undue interference to other users of the relevant spectrum [4]. The technique of spectrum sensing
has become increasingly important in this technology (CR), there are two ways in which cognitive radio is able to perform spectrum sensing, and these ways are made up of two categories: Non-cooperative spectrum sensing and cooperative spectrum sensing. Non-cooperative spectrum sensing: This form of spectrum sensing, occurs when a cognitive radio acts on its own. The cognitive radio will configure itself according to the signals it can detect and the information with which it is pre-loaded [5].
Cooperative spectrum sensing: Cognitive radio cooperative spectrum sensing occurs when a group or network of cognitive radios share the sensitive information they gain. This provides a better picture of the spectrum usage over the area where the cognitive radios are located. The policy of cooperative spectrum sensing can be classified into two great approaches: centralised approach and distributed approach [6].
The rest of this paper is organized as follows. The section II gives an overview of the concept cognitive radio technology. Section III specifies policy of spectrum sensing. The performance of sensing is described in section IV. The section V provides simulation study with performance results for the energy detection over Rayleigh channels. Finally, the section VI offers concluding and outlines future possible research.
II. COGNITIVE RADIO The term ‘cognitive radio’, described by Mitola [7], was
recently raised by the FCC[8] to define a class of terminals which are able to modify their transmission parameters based on interaction with their environment. Cognitive radio is the promising technique for utilizing the available spectrum optimally. The important aspect of cognitive radio is spectrum sensing and from that identifying the opportunistic spectrum for secondary user communication [9].
Figure 1. (a) Time-Frequency sensing, (b) Geographical space sensing, (c)
Angle sensing
The cognitive radio is a communication system intelligent that is aware of its environment. Cognitive radio is used the spectrum in a dynamic manner. She enables to use spectrum in various domains (Frequency, Temporal, Geographical space, Angle), as shown in Figure 1 and in Table I [10], [11].
TABLE I. MULTI-DIMENSIONA RADIO SPECTRUM SENSING[12]
Dimension What must be sensed?
Frequency Holes
The sensing in the frequency domain (some frequency bands might be free for opportunistic (white space) usage).
Temporal Holes
The sensing in the time domain (some spectrum in time might be free for opportunistic(white space) usage).
Geographical space Holes
The sensing in the location domain depends latitude, longitude, elevation and distance between the areas.
Angle Holes The sensing by directions of primary users’ beam depends azimuth angle , elevation angle and locations of primary users.
Among the main functions of cognitive radio are
introduced in the following way (as shown in Figure 2):
Spectrum Sensing: A cognitive radio oversees the available spectrum bands and then sensing the spectrum holes.
Spectrum Management: A cognitive radio is chosen the best available spectrum to meet user communication requirements according to the spectrum characteristics and the needs are mentioned in Table II.
Spectrum Sharing: The third phase is to share white spaces (Spectrum Hole) between the secondary users (CR).
Spectrum Mobility: The US users should evacuate the spectrum immediately and continue their communications in another empty part of the spectrum, without cutting the communication during the transition.
Figure 2. Classification of function
The environment cognitive is referred to as a cognitive cycle (ODA) formed by three tasks: Observation, Decision and Action. The main steps for the simplified cognitive cycle are described in Table II [10], [13], [14] and [15].
TABLE II. THE MAIN STEPS FOR CYCLE COGNITIVE (ODA):ELECTROMAGNETIC CONSIDERATIONS
Observe → Sensing
Electromagnetic examples of sensors (EM): ED, MFD, CsFD, WD, DP.
Decide → The decision is an implicit device in our proposal. An intelligent agent is obligatory for the cognitive operation of the CR.
Examples of decision making and training in bond with the EM:
DRT, TM, BT, ORWB, RBPU, CSU.
Act → Action.
Examples of reconfiguration (radio software) in bond with the EM: FE/FR, CSR, MoCF, MuCF.
ED: Energy Detection -MFD: Matched Filter Detection -
CsFD: Cyclostationary Feature Detection -WD: Wavelet Detection -DP: Detector of Positioning -DRT: Data Rate of Transmission-TM: Transmission Mode-BT: Bandwidth of Transmission -ORWB: Occupancy Rate of Wave Bands -RBPU: Release of Band for a Primary User -CSU: Coordination with other Secondary Users -FE/FR: Frequency of Emission and/or Reception -CSR: Conversion of Sampling Rate -MoCF/MuCF: Mono/Multi Channel Filtering.
Moreover, the cognitive radio can be divided into two types of spectrum being used:
Licensed-bands: Radio spectrum has been assigned to licensed users having the primary rights to use these bands, such as: spectrum in the 900MHz and 2.1GHz bands for mobile telecommunications (currently used for GSM and UMTS), Radio Broadcast (both analog and digital T-DAB), C2000 or DVB-T.
Unlicensed-bands: are currently assigned for various purposes and in various bands, such as: Wi-Fi, Bluetooth, Zigbee, and other protocols using the 2.4 GHz band; RFID; Private GSM in a part of the 1800 MHz band [16].
III. THE POLICY OF SPECTRUM SENSING Obviously, the policy of the detection of spectrum is the
first step for cognitive radio technology. The role of the policy spectrum sensing is to detect whether a signal is present or absent in the monitored spectrum band, which can be estimated as a binary hypothesis testing problem:
𝑋𝑋 = � 0 → (𝑤𝑤ℎ𝑖𝑖𝑖𝑖𝑖𝑖 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑖𝑖 channel) 𝑯𝑯𝟎𝟎 1 → (𝑏𝑏𝑏𝑏𝑠𝑠𝑏𝑏 channel) 𝑯𝑯𝟏𝟏
� (1) H0: the scenario of Primary user is absent,
H1: the scenario of Primary user is in operation, as shown in Figure 3.
Figure 3. The policy of Sensing
A. Performance Analysis of spectrum sensing The performance in spectrum sensing is evaluated by the
probability of correct detection (𝑃𝑃𝑃𝑃), the probability of false alarm (𝑃𝑃𝑃𝑃𝑠𝑠) and probability of miss detection (𝑃𝑃𝑃𝑃𝑃𝑃) of the secondary user, which are given by respectively,
𝑃𝑃d = Prob �𝜀𝜀 = H1H1
� (2)
𝑃𝑃fa = Prob �ε = H0H0� (3)
𝑃𝑃md = Prob �ε = H0H1� (4)
Where ε is a decision threshold [13].
B. Spectrum Sensing Techniques There are several spectrum sensing techniques which can
be found in the literature. In [11], [13], [14] and [13] various spectrum sensing techniques have studied. The mostly used spectrum sensing techniques is given as ED, MFD and CsFD. In this section we describe these sensing techniques in Table III.
TABLE III. SPECTRUM SENSING METHODES FOR COGNITIVE RADIO
C NC WB NB References ED * * * [13],[18],[19]-[20]
MFD * * [9],[11],[20],[21]-[22] CsFD * * [9][11],[19]
C: Coherent -NC: NonCoherent -Wb: Wideband -Nb:
Narrowband.
The advantages and drawbacks of the aforementioned spectrum sensing techniques are summarized and compared in Table IV [13].
TABLE IV. ADVANTAGES AND DRAWBACKS OF SPECTRUM SENSING TECHNIQUES
Approach Advantages Drawbacks ED -does not need any prior
information -low computational cost
-cannot work in low SNR -cannot distinguish users sharing the same channel
MFD -optimal detection performance -low computational cost
-requires a prior knowledge of the primary user
CsFD -robust in low SNR -robust against interference
-requires partial information of the primary user -high computational cost
IV. THE PERFORMANCE OF SENSING
A. The mathematical model Throughout this paper, we assume that energy detection is
applied at each CR user (Figure 4). The energy detector consists of a square law device followed by a finite time Integrator. The output of the Integrator at any time is the energy of the input to the squaring device over the interval T. The noise pre-filter serves to limit the noise bandwidth (Figure 5) [22]. The energy detection is performed by measuring the energy of the received signal in a fixed bandwidth 𝑊𝑊 over an observation time window 𝑇𝑇 [23]. The performance analysis of the energy detector has been studied for AWGN channels in [11], [21] and for Rayleigh fading channels in [21].
A binary hypothesis model for transmitter detection, i.e., the model of signals received by the PU, is defined as
𝑟𝑟(𝑖𝑖) = � 𝑛𝑛(𝑖𝑖) → 𝐻𝐻0ℎ 𝑠𝑠(𝑖𝑖) + 𝑛𝑛(𝑖𝑖) → 𝐻𝐻1
� (5)
𝑟𝑟(𝑖𝑖) and 𝑠𝑠(𝑖𝑖) represent the signal observed in the cognitive radio and the signal transmitted from the PU, respectively. 𝑛𝑛(𝑖𝑖) And ℎ represent the additive white Gaussian noise (AWGN) and the complex channel gain of the sensing channel between the PU and the cognitive radio, respectively [9][13].
Figure 4. The structure of Energy Detector
Figure 5. Energy detection
As shown in Figure 4, the energy collected in the frequency domain is 𝐷𝐷 = ∑ |𝑌𝑌(𝑃𝑃)|2 , which serves as a decision statistic with the following distribution:
𝐷𝐷~ �𝑥𝑥2𝑏𝑏
2 → 𝐻𝐻0𝑥𝑥2𝑏𝑏
2 (2𝛾𝛾 ) → 𝐻𝐻1� (6)
The probability density function (PDF) of 𝐷𝐷 can then be
written as
𝑃𝑃𝐷𝐷(𝑟𝑟) =
⎩⎨
⎧1
2uΓ(𝑏𝑏) 𝑟𝑟𝑏𝑏−1𝑖𝑖−𝑟𝑟2
12 (
𝑟𝑟2𝛾𝛾)
𝑏𝑏−12 𝑖𝑖
−2𝑏𝑏+𝑟𝑟2 𝐼𝐼𝑏𝑏−1(�2𝛾𝛾𝑟𝑟)
� (7)
B. Detection and False alarm probabilities over AWGN channels The probability of detection 𝑃𝑃𝑃𝑃 and probability of false
alarm 𝑃𝑃𝑃𝑃𝑠𝑠 of energy detector over AWGN channels are approximated by [21], [11]:
𝑃𝑃𝑃𝑃 = Prob{𝐷𝐷 > 𝜀𝜀|𝐻𝐻1} = 𝑄𝑄𝑏𝑏 ��2𝛾𝛾 ,√𝜀𝜀 � (8)
𝑃𝑃𝑃𝑃𝑠𝑠 = Prob{𝐷𝐷 > 𝜀𝜀|𝐻𝐻1} =Γ �𝑏𝑏, 𝜀𝜀2�Γ(𝑏𝑏) (9)
𝑃𝑃𝑃𝑃𝑃𝑃 = 1 − 𝑃𝑃𝑃𝑃 (10)
Where Γ(. ) and Γ(. , . ) are complete and incomplete gamma functions, respectively. 𝑄𝑄𝑏𝑏 ( . , . ) is the generalized Marcum 𝑄𝑄 -function, 𝛾𝛾 is instantaneous signal-to-noise ratio (SNR), 𝑏𝑏 is time bandwidth product and 𝜀𝜀 is decision threshold of energy detector.
C. Detection and False alarm probabilities over Rayleigh channels If the signal amplitude follows a Rayleigh distribution, the
SNR γ follows an exponential PDF given by (11):
𝑃𝑃(𝛾𝛾) =1��𝛾 exp(−
𝛾𝛾��𝛾 ), 𝛾𝛾 ≥ 0 (11)
𝑃𝑃𝑃𝑃 gives the probability of the detection conditioned on instantaneous SNR as follows:
𝑃𝑃𝑃𝑃 = ∫𝑥𝑥 𝑄𝑄𝑏𝑏��2𝛾𝛾 ,√𝜀𝜀 �𝑃𝑃𝛾𝛾(𝑥𝑥)𝑃𝑃𝑥𝑥 (12) 𝑄𝑄𝑏𝑏(𝑠𝑠,𝑏𝑏) − is a function defined as follows:
𝑄𝑄𝑏𝑏(𝑠𝑠, 𝑏𝑏) = �𝑥𝑥𝑏𝑏
𝑠𝑠𝑏𝑏−1
∞
0
𝑖𝑖−𝑥𝑥2−𝑠𝑠2
2 𝐼𝐼𝑏𝑏−1 (𝑠𝑠𝑥𝑥)𝑃𝑃𝑥𝑥 (13)
Where 𝐼𝐼𝑏𝑏−1 (. ) is the modified Bessel function of (u−1) order.
For the cognitive radio with the energy detector [13], the average probability of detection over Rayleigh fading channel 𝑃𝑃𝑃𝑃 may be obtained by (12).
𝑃𝑃𝑃𝑃 = 𝐸𝐸𝛾𝛾 [𝑃𝑃𝑟𝑟𝑃𝑃𝑏𝑏{𝐷𝐷 > 𝜀𝜀|𝐻𝐻1}]
= 𝑖𝑖−𝜀𝜀2 �
1𝑛𝑛!
𝑏𝑏−2
𝑛𝑛=0
�𝜀𝜀2�
𝑛𝑛+ �
1 + ��𝛾��𝛾 �
𝑏𝑏−1
∗ � 𝑖𝑖−𝜀𝜀
2(1+𝛾𝛾�) 𝑖𝑖−𝜀𝜀2 �
1𝑛𝑛! �
𝜀𝜀��𝛾2(1 + ��𝛾)�
𝑛𝑛
𝑏𝑏−2
𝑛𝑛=0
� (14)
The average probability of false alarm and the average
probability of miss over Rayleigh fading channels are given by, respectively,
𝑃𝑃𝑃𝑃𝑠𝑠 =Γ(𝑏𝑏, 𝜀𝜀2)Γ(𝑏𝑏) (15)
𝑃𝑃𝑃𝑃𝑃𝑃 = 1 –𝑃𝑃𝑃𝑃𝑅𝑅𝑠𝑠𝑏𝑏 (16) In (14), ��𝛾 is the average SNR of the PU signal on the CR.,
and 𝐸𝐸𝛾𝛾 is the expectation over the random variable 𝛾𝛾 [21].
V. SIMULATION AND RESULT Simulation was done in MATLAB over Rayleigh channel
by the energy detection. We have generated a signal waveform with a required SNR (Figure.6) and we have used it to plot ROC curves (Pd Vs Pfa) or complementary ROC curves (Pmd Vs Pfa) for different SNR values according to (15) and (16). We can now define two possibilities for r(t). It may equal either a noise (AWGN) or a sinusoidal signal plus a noise. Thus we define two hypothesis:
�𝐻𝐻0: 𝑟𝑟(𝑖𝑖) = 𝑛𝑛(𝑖𝑖) 𝐻𝐻1: 𝑟𝑟(𝑖𝑖) = 𝑛𝑛(𝑖𝑖) + 𝐴𝐴 ∗ sin(2 ∗ 𝑠𝑠𝑖𝑖 ∗ 𝐹𝐹𝑠𝑠 ∗ 𝑖𝑖 + 𝜃𝜃)
� (17)
Where 𝑛𝑛(𝑖𝑖) ~ 𝐺𝐺(0,𝜎𝜎𝑛𝑛2) , Fc = carrier frequency, A =
amplitude (peak-to-peak), 𝜃𝜃 = random receive phase, 𝜎𝜎𝑛𝑛2 = variance of noise signal, G( ) = Gaussian (normal) distribution,
And in order to calculate the SNR we define the following two equations
𝑆𝑆𝑆𝑆𝑅𝑅 =𝐴𝐴2
2𝜎𝜎𝑛𝑛2 𝑃𝑃𝑟𝑟 𝑆𝑆𝑆𝑆𝑅𝑅 = 10𝑙𝑙𝑃𝑃𝑙𝑙10 �
𝜎𝜎𝑠𝑠2
𝜎𝜎𝑛𝑛2� (18)
With 𝜎𝜎𝑠𝑠2 = variance of the signal.
(a) (b)
Figure 6. Generating a signal with required SNR/ (a) and (b)
Figure 7. Pmd Vs Pfa for ≠ SNR
0 500 1000 1500 2000 2500-5
0
5Generated Noise, n(t) @ H0
0 500 1000 1500 2000 2500-10
0
10Signal + Noise, r(t) @ H1 for SNR= 4.9983 dB
0 500 1000 1500 2000 2500-5
0
5Input Signal, s(t) the signal transmitted from the PU
0 500 1000 1500 2000 2500-20
-10
0
10
20Input Signal, s(t) the signal transmitted from the PU
0 500 1000 1500 2000 2500-5
0
5Generated Noise, n(t) @ H0
0 500 1000 1500 2000 2500-20
0
20Signal + Noise, r(t) @ H1 for SNR= 19.9984 dB
0.0385 0.039 0.0395 0.04 0.0405 0.041 0.0415 0.042 0.04250
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Pfa: Probability of Fals alarm
Pm
: P
robabili
ty o
f m
issed d
ete
ction
SNR= 4.9983 dBSNR= 19.9984 dB
Figure 8. Pd Vs Pfa for ≠ SNR
Figures 7 and 8 show the variation of the 𝑃𝑃𝑃𝑃𝑃𝑃 and 𝑃𝑃𝑃𝑃 with 𝑃𝑃𝑃𝑃𝑠𝑠 for different 𝑆𝑆𝑆𝑆𝑅𝑅 values, with 𝑏𝑏 = 5 . These curves illustrate the impact of the 𝑆𝑆𝑆𝑆𝑅𝑅 on the 𝑃𝑃𝑃𝑃𝑃𝑃 and 𝑃𝑃𝑃𝑃. More the 𝑆𝑆𝑆𝑆𝑅𝑅 increases, more 𝑃𝑃𝑃𝑃𝑃𝑃 and 𝑃𝑃𝑃𝑃 have good results. For example when 𝑆𝑆𝑆𝑆𝑅𝑅 = 4.9983 𝑃𝑃𝑑𝑑 ≈ 5 𝑃𝑃𝑑𝑑 with 𝑃𝑃𝑃𝑃𝑠𝑠 = 0.041, the 𝑃𝑃𝑃𝑃 = 0.7646 but when the 𝑆𝑆𝑆𝑆𝑅𝑅 = 19.9984 𝑃𝑃𝑑𝑑 ≈ 20𝑃𝑃𝑑𝑑 with the same 𝑃𝑃𝑃𝑃𝑠𝑠 , the 𝑃𝑃𝑃𝑃 = 0.915.
Figure 9. Di Vs Pfa
𝐷𝐷𝑖𝑖 = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑠𝑠 𝑖𝑖(𝑠𝑠𝑛𝑛𝑟𝑟 = 19.9984) − 𝑃𝑃𝑃𝑃𝑠𝑠𝑃𝑃𝑠𝑠 𝑖𝑖(𝑠𝑠𝑛𝑛𝑟𝑟 = 4.9983) (19)
Where Di is the uncertainty between two Pd.
From equation (19), we obtain the graph illustrated in Figure 9, it is apparent from the figure that the uncertainty decreases when the two probabilities Pd (SNR=19.9984 dB) and Pd (SNR=4.9983 dB ) are close to 1. In other hand, from the 𝑃𝑃𝑃𝑃𝑠𝑠 = 0.04167 whatever the value of SNR, the uncertainty Di tends to 0.
Figure 10 shows Pd for different SNR over Rayleigh channel. This figure shows the performance of detection probability under different values of SNR taken between -30 dB and 30 dB, where 𝑃𝑃𝑃𝑃𝑠𝑠 = 0.99 and time bandwidth factor 𝑏𝑏 = 5. We can see that when SNR increase, the detection also increase and For 𝑆𝑆𝑆𝑆𝑅𝑅 = 17 𝑃𝑃𝑑𝑑 the detection of probability Pd tends to 1. Table 5. shows the results of Pd.
Figure 10. Pd Vs SNR
TABLE V. SNR AND DETECTION OF PROBABILITY
SNR (dB) Pd -30 to 08 (Bronze) Bad results
08 to 17 (Silver) Acceptable results 17 to 30 (Gold) Excellent results
VI. CONCLUSION With the increasing demand of radio spectrum, it is noticed
that the use of the spectrum is ineffective in terms of resources. The Cognitive radio is a promising solution for better using the spectrum at each moment and in each place. The capacity of the cognitive radio relies on detection, the decision and the adaptation of its electromagnetic environment. The cognitive radio takes into accounts other information on its environment namely the position and the speed of displacement of a secondary user (CR). In this paper we have performed several simulations, in which: First, we have generated a signal waveform with a required SNR. Second, we studied the impact of the SNR on the Pmd and Pd versus the probability of false alarm. Finally, we analysed the probability of detection versus SNR ranging between −30 dB and 30 dB at a constant (Pfa = 0.99). Currently, the researches on cognitive radio are based mainly with several areas: the detection of the free resources, the dynamic distribution of the frequencies between terminals CR and the mobility of a terminal RC (spectral handover). In future work, we aim to study the impact of mobility on spectrum detection.
REFERENCES [1] Ying-Chang Liang, Kwang-Cheng Chen , Geoffrey Ye Li, Petri
Mahonen and Dusit Niyato, “ Guest Editorial Advances in Cognitive Radio Networking and Communications (II),” IEEE Journal on selected areas in communications, Vol. 29, NO. 4, APRIL 2011, pp 673–675.
[2] Alfred Asterjadhi, Nicola Baldo and Michele Zorzi, “ A Distributed Network Coded Control Channel for Multi-hop Cognitive Radio Networks ,” IEEE Network • July/August 2009 ,pp.26–32.
[3] Songlin Sun, Yanhong Ju, Yasushi Yamao, “Overlay cognitive radio OFDM system for 4G cellular networks,”next generation cognitive cellular networks, IEEE Wireless Communications, April 2013,pp 68-73.
[4] Http://www.radio-electronics.com/info/wireless/ieee-802-22/cognitive-network-spectrum-sensing.php.
[5] Http://www.radio-electronics.com/info/rf-technology-design/cognitive-radio-cr/spectrum-sensing.php.
0.0385 0.039 0.0395 0.04 0.0405 0.041 0.0415 0.042 0.04250.7
0.75
0.8
0.85
0.9
0.95
1
Pfa: Probability of Fals alarm
Pd:
Pro
babili
ty o
f dete
ction
SNR= 4.9983 dBSNR= 19.9984 dB
D5
D4
D3
D2
D1
0.0385 0.039 0.0395 0.04 0.0405 0.041 0.0415 0.042 0.04250
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
Pfa: Probability of Fals alarm
Di
Pfa = 0.04167
-30 -20 -10 0 10 20 300
0.2
0.4
0.6
0.8
1
SNR : Signal-to-Noise Ratio dB
Pd:
pro
bability o
f dete
ction
Pfa = 0.99
SNR = 17 dB
[6] Http://www.radio-electronics.com/info/rf-technology-design/cognitive-radio-cr/cooperative-spectrum-sensing.php.
[7] J. Mitola, “Cognitive Radio:an Integrated Agent architecture for Software Defined Radio,” Dissertation, Royale Institude of Technology,2000.
[8] Federal Communications Commission (FCC), “Facilitating Opportunities for Flexible, Efficient, and Reliable SpectrumUse Employing Cognitive radio technologies,” ET Docket No.03-108;March 2005.
[9] R. Suresh Babu and M. Suganthi, “Review of energy detection for spectrum sensing in various channels and its performance for cognitive radio applications,” American Journal of Engineering and Applied Sciences, 2012, 5 (2), pp 151-156.
[10] Lamiaa Khalid and Alagan Anpalagan, “Emerging cognitive radio technology: Principles, challenges and opportunities,” in WINCORE Laboratory, Ryerson University, Toronto, Canada, Computers and Electrical Engineering 36 (2010), pp 358-366.
[11] Ian F. Akyildiz, Won-Yeol Lee, Mehmet C. Vuran and Shantidev Mohanty, “NeXt generation/dynamic spectrum access/cognitive radio wireless networks: A survey,” in Broadband and Wireless Networking Laboratory, School of Electrical and Computer Engineering,Georgia Institute of Technology, Atlanta, GA 30332, United States, Computer Networks 50 (2006) , pp 2127-2159.
[12] Tevfik Yucek and Huseyin Arslan, “A Survey of Spectrum Sensing Algorithms for Cognitive Radio Applications,” IEEE communicatios SURVEYS surveys and tutorials, vol. 11, no. 1, first Quarter 2009 ,pp 116-130.
[13] Khaled Ben Letaief and Wei Zhang, “Cooperative Spectrum Sensing,” Department of Electronic & computer Engineering, Hong Kong University of Science & Technology Clear Water Bay, Kowloon, Hong Kong, in Cognitive Wireless Communication Networks , 2007, Springer, pp 115-138.
[14] I. Budiarjo, M. K. Lakshmanan and H. Nikookarled, “Cognitive Radio Dynamic Access Techniques,” Published online: 29 February 2008,Wireless Pers Commun, Springer, pp 293-324.
[15] Christophe Moy and Jacques Palicot, “Mieux analyser les ondes pour mieux communiquer : la radio intelligente,” URSI-France, Journées scientifiques 26/27 mars 2013, pp 35-37.
[16] Sietse van der Gaast, Alexander ter Haar, Hind Abdulaziz and Paul Brand, “ MOB-12-07a Trend report fixed/wireless networking: Trends and options in efficient spectrum use ,” Project : Gigaport3, February 2013, Version : 1, © Stratix 2013,pp 1-33.
[17] Ian F. Akyildiz , Won-Yeol Lee and Kaushik R. Chowdhury, “CRAHNs: Cognitive radio ad hoc networks,” in Broadband Wireless Networking Laboratory, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, 30332, United States, Ad Hoc Networks (2009), pp 1-27.
[18] Seung-Hoon Hwang and Jun-Ho Baek, “Multiple Antenna-Aided Spectrum Sensing Using Energy Detectors for Cognitive Radio,” Vehicular Technologies: Increasing Connectivity, Dr Miguel Almeida (Ed.), ISBN: 978-953-307-223-4, InTech, Available from: http://www.intechopen.com/books/vehicular-technologies- increasing-connectivity/multiple-antenna-aided-spectrum-sensing-using-energy-detectors-for-cognitive-radio2011, pp 239-260.
[19] Waleed Ejaz, Najam ul Hasan and Hyung Seok Kim,“ SNR-Based adaptative spectrum sensing for cognitive radio networks,” International Journal of Innovative Computing, Information and Control Volume 8, Number 9, September 2012, ISSN 1349-4198, pp 6095-6105.
[20] Fadel F. Digham, Mohamed-Slim Alouini and Marvin K. Simon,“ On the Energy Detection of Unknown Signals over Fading Channels,” Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN 55455, USA, IEEE 2003, pp 3557-3579.
[21] Mohammad Alamgir Hossain, Md. Shamim Hossain and Md. Ibrahim Abdullah,“ Performance Analysis of Cooperative Spectrum Sensing in Cognitive Radio,” International Journal of Innovation and Applied Studies, ISSN 2028-9324 Vol. 1 No. 2 Dec. 2012, http://www.issr-journals.org/ijias/ , pp 236-245.
[22] Ying-Chang Liang, Yonghong Zeng, Edward Peh, and Anh Tuan Hoang,“ Sensing-Throughput Tradeoff for Cognitive Radio Networks,” ISSN 1-4244-0353-7/07 ©2007 IEEE, pp 5330-5335.
[23] Waleed Ejaz, Najam ul Hasan and Hyung Seok Kim,“ SNR-Based adaptative spectrum sensing for cognitive radio networks,” International Journal of Innovative Computing, Information and Control Volume 8, Number 9, September 2012, ISSN 1349-4198, pp 6095-6105.
Combining Cooperative Diversity and Turbo
Coding in wireless communications
H. TAYAKOUT 1,2
, K. GHANEM 1
1 Center for Development of Advanced Technologies
CDTA, City 20 aout 1956 Baba Hassen,
Algiers, Algeria.
H. BOUSBIA-SALAH 2
2 National Polytechnical School ENP
10, Avenue Hassen Badi BP 182- 16200 El Harrach,
Algiers, Algeria.
Abstract— Cooperative diversity has gained a great
research interest because of its performance and usage
potential for future and flexible architectures in wireless
communications. In this paper, a scheme combining
turbo coding and cooperative diversity techniques is
proposed and studied. The performance is investigated
with both Binary and Quadratic Phase-Shift Keying
modulations, namely BPSK and QPSK, and with the
amplify and forward (AF) strategy adopted at the relay
level. The impact of coding gain is highlighted by
comparing the performance achieved with two different
codes at the source, and the signals emanating from the
source and destination are simultaneously exploited
using maximum ratio combining technique (MRC) at the
destination node. The impact of channel estimation
errors are beyond the scope of this paper, and it is thus
assumed that source-destination and all the relay-
destination channels are well estimated at the final
receiver.
Keywords— Turbo coding Cooperatif; Multiple relays;
Amplify and forwad AF; MRC combining.
I. INTRODUCTION
Cooperative communications, in which multiple single-antenna relay terminals assist the source in transmitting information to the destination, allow to increase the spatial diversity and the reliability of the communication provided by the several paths between the source node and the destination node, through the relay nodes. When the direct link is in deep fade or blocked by an obstacle, the communication is maintained through the available relay nodes [1], [2].
Cooperative techniques in wireless communications have gained so much interest after the presentation of this new concept by Sendonaris et al. [1]. In [2], Laneman et al. introduced various cooperative protocols and evaluated their performance in both ergodic and quasi-static channels. A couple of relaying protocols were introduced such as decode and forward (DF), amplify and forward (AF), and some adaptive methods that switch between the two techniques. Because of its low complexity, AF relaying strategy is the most popular cooperative diversity schemes and, therefore will be adopted in this paper[2].
Another paradigm that has significantly changed the communications field and made Shannon capacity limit no more an utopia, was the introduction of turbo coding by Berrou et al. which leads afterwards to the enhancement of either the system performance or throughput [3].
Our proposed scheme(see fig.1) which is a cooperative turbo coded system in AF multiple-relays channels can be viewed as an extension of the work presented in [4] in which only the case of a single relay was introduced. In this work, the source, relays and destination are assumed equipped with omni-directional antennas. Furthermore, we compare the performance when adopting BPSK and QPSK modulation with two turbo codes generated at the source node with a rate of 1/3and ½, respectively, assuming both Rayleigh and Gaussian channels cases. The rest of this paper is organized as follows. Section II introduces the system and channel models as well as the relaying and the combination techniques. In Section III, some simulation results are provided and discussed, and finally a conclusion finalizes the paper.
II. SYSTEM AND CHANNEL MODEL
A. System Model
Fig. 1 depicts an AF multiple-relays system which consists of one source, one destination and N relays.
Fig. 1. System Model.
During the first stage, the source broadcasts the desired coded symbol to the relays and destination with a transmitted power . Then, throughout the second stage, each relay amplifies its received signal and retransmits it to the destination.
(a) Source Node.
(b) Destination Node.
Fig. 2. Block diagram with a Turbo coding and mapping procedure at the source node and with MRC combining, soft metric and Turbo decoding at the
destination node.
We assume all channels are subject to frequency non-selective fading. Two types of channels are investigated in this work, namely Rayleigh and Gaussian cases. In the former case, the fading coefficients of the different links between the source and destination are modeled as a zero-mean, independent, circularly complex Gaussian random variables with unit variance. In the latter, the channel is fixed at unity. We denote the variable encompassing additive channel noise and interference, and modeled as a
complex Gaussian random variable with variance
.
Furthermore, refer to channel coefficients
between the source and the ith
relay node while
the dual channel coefficients between
relays and destination. The sequence of bits generated at the source is coded by the channel turbo encoder (T.ENC) to produce a sequence of coded bits which are mapped onto to symbols ϵ [+1, -1] for a BPSK constellation and [+1+1, +1-1, -1+1, -1-1] for a QPSK constellation (Fig.1) which is received at the i
th relay node as:
√
(1)
We consider scenarios in which each fading coefficient
is known by the relay , but not known to the source.
The signal received at destination node from the source denoted by can be described as follows:
√ (2)
In the second phase, relays amplify the symbols they receive by a parameter inversely proportional to the channel magnitude. Thus, the signal received at the destination from can be formulated as:
√
√ | |
(3)
Before iterative decoding, the destination performs a MRC of the signals emanating from the source node and the multiple relay transmitters as follows:
∑
(4)
In order to do so, the channels between the relays and the receiver are assumed perfectly estimated at the receiver.
B. soft Detection
The ML detector chooses, as estimated input, the most probably code-word given a received input by maximizing ( | ). The ML solution for is
( | )
|| || (5)
The soft Maximum Likelihood (ML) detector computes bit-level reliability metrics by converting the received signal to log-likelihood ratios of the a posteriori probability (APP) of the encoded bits being +1 or -1. The values for the ML detector are defined as:
[ ( | )
( | )] (6)
where ( ) is the number of bits that constitute the symbol . By employing Bayes’ theorem and assuming statistical independence and equiprobability among the bits , the can be written as [5], [6]:
[∑ ( | )
( )
∑ ( | )
( )]
[∑ ( (|| ||
)
( )
∑ ( (|| || )
( )
]
(7)
Where ( )
( )
are disjoint sets of vector
symbols that have the kth bit of the transmit symbol s equal to +1 and -1, respectively, is the symbol containing the bits and is average signal-to-noise ratio at the receive antenna which is given as Es/N0. In Gaussian fading channel, h is fixed at unity.
In the BPSK modulation case, the number of bits that constitute the symbol is ( ) the calculation becomes:
(8)
In the latter (QPSK modulation), The expression (7) is simplified by applying the max-log approximation:
∑ ( ) ( )
( )
(|| || )
( )
(|| || )
The obtained soft information at the ML output should
be fed into the soft-input Turbo decoder to recover the
transmitted information source bits.
III. SIMULATION RESULTS
The performance of the proposed turbo-coded multiple relays system is studied in both Gaussian and Rayleigh channels when using M-PSK modulation and AF relaying schemes with two different code rates. In the simulation P1 is fixed at unity, the distances between the different nodes at 1m, and the possible number of relays is 1,2 or 4.
The turbo encoder uses two RSCs in parallel with a constraint length and two generator polynomials which octal representation are . A standard interleaver , namely the 1024-bit random interleaver is used between the two RSC codes components. The Log-Likelihood ratios LLR about the encoded received symbols are calculated using the BCJR algorithm. Other adopted parameters in our simulations are summarized in Table 1.
TABLE I. TURBO ENCODEUR AND JTERATIF DECODER PARAMETERS
Frame length 1024 bits.
Channel Rayleigh quasi-static and
Gaussian model
Modulation BPSK end QPSK
Component Enc 2 identical RSC
Enc. parameters
Interleaver 1024-bits random interleaver
Component It. Dec Max-log MAP decoder
Number of iteration 4
Fig 3 shows the BER performance with perfect channel
estimates, using only one relay, the two aforementioned
turbo codes and two types of modulations, namely BPSK
(M=2) and QPSK (M=4). From this figure, it can be seen
that, at very low SNR range, the performance of the two
codes is similar, because the adverse channel condition does
not help in exploiting the channel coding gain offered by the
turbo codes. When the SNR increases, the performance
using the code with lower rate is better. For instance, when
the required BER is 10-6, decreasing the code rate yields a
gain of 3.5 dB for BPSK and 3.2 dB for QPSK.
Fig 4 shows the BER performance in a Gaussian channel
when varying the number of relays. From this figure, it is
shown that, when the SNR level is very low, the
performance of the two codes is similar, whatever the
number of relays is. Thus the combination of turbo coding
and cooperative diversity brings no gain at this range.
Increasing the SNR level allows to improve the BER
performance particularly with the lower rate code. This
performance is improved with the increase of the number of
relays. For a required BER of 10-5
, passing from 2 relays to
4 provides a gain of 2 dB with a 1/2 rated code while this
gain is marginal for the case of a 1/3 rated code.
Fig. 3 BER vs Receive Antenna SNR for AF relay with the perfect CSI BPSK & QPSK channel modulation, N=1024 bits.
Fig.4 BER vs Receive Antenna SNR for AF Multiple relay with QPSK
modulation in Gaussian fading channel, N=1024 bits,
In Fig. 5, the performance of the same system in Rayleigh channel is studied. It first seen that the performance is degraded in this channel compared to Gaussian case. Indeed, the required SNR to achieve the same BER performance as in Gaussian channel is higher. The gain brought by cooperative diversity is notable in this case for both codes. However, the performance gap provided by the coding gain is less than in Gaussian case.
-10 -5 0 5
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
Receive antenna SNR (dB)
B
E
R
QPSK R=1/2
QPSK R=1/3
BPSK R=1/2
BPSK R=1/3
3.5 dB 3.2 dB
-10 -8 -6 -4 -2 0 210
-6
10-5
10-4
10-3
10-2
10-1
100
Receive antenna SNR, (dB)
B
E
R
R =1/2 & Nbre
Relay =1
R =1/2 & NbreRelay
=2
R =1/2 & NbreRelay
=4
R =1/3 & NbreRelay
=1
R =1/3 & NbreRelay
=2
R =1/3 & NbreRelay
=4
For instance, when adopting a 4 relays-scheme, achieving a BER of 10
-5 with a 1/2 rated code requires less than 1 dB
increase in SNR compared to the 1/3 rated code, which is inferior to the 2 dB required in Gaussian channel case.
Fig. 5 BER vs Receive Antenna SNR for AF relay with QPSK
Modulation in Rayleigh fading channel, N=1024 bits,
IV. CONCLUSION
In this paper, the performance of a coded cooperation
scheme using multiple relays with AF relaying protocol has
been investigated for both Gaussian and Rayleigh channels.
We have assumed that the channel information is available
at the receiver. It is shown that increasing the number of
relays allows to enhance the channel link quality. For a
given number of relays, lowering the code rate allows to
improve the system reliability.
REFERENCES
[1] A. Sendonaris, E. Erkip, and B. Aazhang, ―User cooperation diversity— part I and part II,‖IEEE Trans. Commun., vol. 51, no. 11, pp. 1927–1948, Nov. 2003.
[2] J. N. Laneman, G. W. Wornell, and D. N. C. Tse, ―An efficient protocol for realizing cooperative diversity in wireless networks,‖ in Proc. IEEE Int. Symp. Information Theory, Washington, DC, pp. 294, Jun. 2001.
[3] C. Berrou, A. Glavieux, and P. Thitimajshima, ―Near shannon limit error-correcting coding and decoding,‖IEEE International Conference on Communications, vol. 2, May 1993.
[4] Z. Zhang and T. M. Duman, ―Capacity-Approaching Turbo CodingFor Half-Duplex Relaying,‖ in IEEE Trans. On Communication, vol. 55, no. 10, Oct 2007.
[5] Steingrimsson, B. and Zhi-Quan Luo and Kon Max Wong. Soft quasi-maximum-likelihood detection for multiple-antenna wireless channels. Signal Processing, IEEE Transactions on, 51(11):2710 - 2719, 2003.
[6] chwald, B.M. and ten Brink, S. Achieving near-capacity on a multiple-antenna channel. Communications, IEEE Transactions on, 51(3):389 - 399, 2003.
-10 -8 -6 -4 -2 0 2 4
10-5
10-4
10-3
10-2
10-1
100
Receive antenna SNR, (dB)
B
E
R
R =1/2 & Nbre
Relay =1
R =1/2 & NbreRelay
=2
R =1/2 & NbreRelay
=4
R =1/3 & NbreRelay
=1
R =1/3 & NbreRelay
=2
R =1/3 & NbreRelay
=4
1
Tailoring Mikey-Ticket to e-Health Applications inthe Context of Internet of Things
Riad Abdmeziem, LSI, USTHB: University of Sciences and Technology Houari BoumedienneBP 32, El Alia, Bab Ezzouar, Algiers, Algeria.
[email protected] Tandjaoui, CERIST: Center for Research on Scientific and Technical Information
03, Rue des freres Aissou, Ben Aknoun, Algiers, [email protected]
Abstract—In the context of Internet of Things where realworld objects will automatically be part of the Internet, e-health applications have emerged as a promising approach toprovide unobtrusive support for elderly and frail people basedon their situation and circumstances. However, due to the limitedresources available in such systems and privacy concerns thatmight rise from the capture of personal data, security issuesconstitute a major obstacle to their deployment. Key managementprotocols are required to establish secure communication chan-nels. Nevertheless, e-health systems are unable to run existingstandard key management protocols due to their highly con-strained resources. In this paper, we introduce two design ideas inorder to tailor Mikey-Ticket protocol to constained environments.We introduce a new mode exchange to reduce the number ofexchanged messages and a new header compression scheme toreduce the size of the different messages and hence decreasingfragmentation rate.
Keywords—Internet of Things, E-health, Security, Key Manage-ment, Compression, Mikey-Ticket.
I. INTRODUCTION
INTERNET of Things (IoT) is one of the maincommunication development in recent years that makes
everyday sensors and working devices connected to eachother and to the Internet. According to [1], the basic conceptbehind is the pervasive presence around us of variouswireless technologies such as Radio-Frequency IDentification(RFID) tags, sensors, actuators or mobile phones in whichcomputing and communication systems are seamlesslyembedded. Through unique adressing schemes, these objectsinteract with each other and cooperate to reach common goals.
Technology advances along with popular demand willfoster the wide spread deployment of IoT’s services, itwould radically transform our corporations, communitiesand personal spheres. From the perspective of a privateuser, IoT’s introduction will play a leading role in severalservices. E-health is one of the most interesting applicationas it will provide medical monitoring to millions of elderlyand disabled patients while preserving their autonomy andcomfort anywhere. Using sensors planted in or around thebody, physiological data are gathered and transmitted toqualified personal that can intervene in case of emergency.
Nevertheless, e-health applications are unlikely to fulfila widespread diffusion until they provide strong securityfoundations. Making sure that only authorized entitiescan access and modify data is particularly relevant ine-health applications where data are very sensitive andany unwanted modification could lead to dramatic events.Securing communications in e-health systems necessarilypasses through key management protocols that are in chargeof delivering security credentials to the involved entities.However, classical solutions deployment is hindered due tothe scarcity of available resources, either energy power orcomputation capabilities.
Mikey-Ticket [2] is a key management protocolcharacterized by its simplicity and its well adaptation tocentralized architectures based on a central entity. In fact,these architectures are interesting to be considered for resourceconstrained environments. There is no need to pre-distributecredentials as users can request them only when required.Centralized solutions also scale well when the number of usersgrows. Additionnaly, Mikey-Ticket provides different messageexchanges that can be transported over UDP and integratedwithin different security protocols (e.g. IPSEC, DTLS,HIP...etc.). Nevertheless, Mikey-Ticket needs to be tailored forconstrained environments in order to take into considerationresource scarcity. To this end, we introduce two ideas to adaptMikey-Ticket to e-health environments without weakening itssecurity properties. The first one aims to reduce the numberof exchanged messages from six to four by proposing anew exchange mode. The main concern being to involve asless as possible the constrained node in the exchange. Thesecond one aims to save energy by reducing the message sizeand avoiding 6LowPAN fragmentation that occurs when adatagram size exceeds the link layer MTU 1. Fragmentation isundesirable regarding the security aspect too as 6LoWPAN isvulnerable to fragmentation attacks [3]. Hence, we propose anew 6LoWPAN header compression scheme for Mikey-Ticket.
The rest of this paper is organized as follows. Section 2summarizes related work. In section 3, we present a briefoverview of the technologies used in this paper. We outline
1Maximum Transmission Unit of the IEEE 802.15.4 protocol
2
Fig. 1: Mikey-Ticket exchange
our network scenario in section 4. In section 5, we describein detail our improvements for Mikey-Ticket. Finally, section6 concludes the paper and gives future directions.
II. RELATED WORK
In our discussion of related work, we focus on approachesthat are intended to tailor standard security protocols for IoTenvironments. In particular, we distinguish three researchdirections: i) compression schemes for the IP-based IoT, ii)delegation procedures of protocol’s primitives, iii) furtherdesign improvement approaches.
Several compression schemes for the IP-based IoT havebeen proposed. In [4] and [5], the compression of IPV6headers, extension headers along with UDP headers has beenstandardized through 6LoWPAN. Authors in [6] and [7] havepresented 6LoWPAN compressions for IPsec payload headers(AH and ESP). Their standardization is proposed in [8]. In[9], an IKE compression scheme has also been proposed inorder to provide a lightweight automatic way to establishsecurity associations for IPsec. Likewise, header compressionlayer for DTLS and HIP DEX was respectively introduced in[10] and [11]. However, to the best of our knowledge, no priorcompression scheme have been proposed for Mikey-Ticket.
Apart from packet compression schemes, several delegationprocedures of protocol’s primitives have been proposedthat aim to offload the computational load to third entities.Authors in [12], [13] and [14] have introduced collaborationfor HIP. The idea is to take advantage of more powerful nodesin the neighborhood of a constrained node to carry heavycomputations in a distributed way. Likewise, IKE sessionestablishment delegation to the gateway have been proposedin [15]. Furthermore, authors in [16] introduce a delegationprocedure that enables a client to delegate certificate validationto a trusted server. While delegation approaches reduce thecomputational load at the constrained node, they break theend to end principle by requiring a third trusted party.
Further design improvement approaches have been intro-duced to tailor end to end security protocols to the IoT, authorsin [17] have proposed complementary lightweight extensionsto HIP DEX that could be generalized to DTLS and IKE.
Likewise, in [18], authors have introduced design ideas toreduce the overhead of the DTLS handshake. Their goal is tomake the use of certificates for authentication purposes viablein IoT contexts.
III. BACKGROUND
A. Mikey-Ticket overview
Mikey-Ticket [2] is a key distribution protocol designed toenhance the Multimedia Internet KEYing protocol (Mikey)[19]. It defines new modes of key distribution which are welladapted to centralized based scenarios where a third trustedentity is available. Mikey-Ticket considers two entities that aimto establish a shared secret. One of the two entities assumesthe Initiator role and the second assumes the Responder role.The key establishment relies on a Key Management Serverto generate and deliver the needed credentials. Such designspares the peers from a pre-distribution phase that wouldrequire credentials storing. Instead, peers can request suchcredentials only when required. Scalability, when the numberof users grows, is then improved. In this work, we onlyconsider the Pre-Shared Key mode (PSK) of Mikey Ticket asthe Public Key (PK) mode is ruled out due to its inadequacywith IoT constrained environments.
We provide a brief description of Mikey-Ticket messageexchanges and the general Mikey header (HDR) format. Table1 summarizes the used notations.
1) Message exchanges: Mikey-Ticket is composed of sixmessages to establish a new key between the Initiator I and theResponder R (See Figure 1). The protocol relies on the KeyManagement Server (KMS) which delivers the generated key.The Initiator and the Responder do not share any credentials.Instead, they share a secret master key with the KMS. Thekey is used to derive an authentication key and an encryptionkey. The generated keys are used to secure communicationsfor I and R with the KMS providing data authenticity, dataintegrity and confidentiality. We briefly describe the contentof each exchanged message:
REQUEST INIT: Through this message, I expresses itswillingness to establish a shared key with R. The messagecontains information about the responder’s identity. To ensure
3
Notation DescriptionI InitiatorR ResponderKMS Key Management ServerXID The Identity of XNX Nonce generated by XKX,Y Shared key between X and YaKX,Y Shared authentication key between X and YeKX,Y Shared encryption key between X and Y[data]K Data encrypted with the key KTicket Object used to identify and deliver keys
TABLE I: Terminology Table
Fig. 2: Mikey Common Header Format (RFC 3830)
authenticity, a Message Authentication Code (MAC) computedwith aKI,KMS is included.
REQUEST RESP: After successful verifications, the re-quest is authorized and the KMS generates the requested keyK and encodes it in a ticket.
TRANSFER INIT: Upon the reception of REQUEST RESPmessage, I derives an authentication key (aK) and an encryp-tion key (eK) to secure data transmission between I and R. Itransfers the ticket to R through TRANSFER INIT message.Also, a MAC is computed using aK and included in themessage.
RESOLVE INIT: Through this message, R asks the KMS toreturn the key encoded in the ticket. The message is protectedby a MAC based on aKKMS,R.
RESOLVE RESP: If R is authorized to receive the gen-erated key encoded in the ticket, the KMS sends RE-SOLVE RESP message that includes the generated key K. Themessage is protected through encryption and a MAC messagebased on aKKMS,R.
TRANSFER RESP: R is in possession of the generatedkey K. TRANSFER INIT’s MAC can thus be checked.The exchange is concluded through TRANSFER RESPmessage proving the correct reception and derivation of thegenerated session key. It is worth noticing that the differentmessages contain a nonce for protection against replay attacks.
2) Common Header Format (HDR): The Common Headerpayload (See Figure 2) describes the exchanged messages.It is always present as the first payload in each message. Inthe following, we present a succinct description of each fieldcontained in the Mikey Ticket header. We refer to RFC3038[19] and RFC6043 [2] for a more detailed description:
- Version (8 bits): Mikey’s version.
- Data type (8 bits): Message’s type.- Next Payload (8 bits): Identifies the payload added after
the current payload.- V (1 bit): Flag to indicate the use of a verification
message.- PRF func (7 bits): Indicates the key derivation function.- CSB ID (32 bits): Crypto Session Bundle (CSB) is a
collection of one or more Crypto Sessions (CS). CSBID field identifies the CSB.
- ] CS (8 bits): A Crypto Session refers to a data steamprotected by a single instance of a security protocol. ]CS field indicates the number of Crypto Sessions withinthe CBS.
- CS ID map type (8 bits): Specifies the method ofuniquely mapping crypto sessions to the security pro-tocol sessions.
- CS ID map info (variable length) Identifies and mapscrypto sessions to the security protocol sessions.
B. 6LowPANThe 6LoWPAN standard defined in [5] aims to make IPv6
practical on IEEE 802.15.4 networks. 6LoWPAN is based onIPV6 header compression mechanisms of IPv6 datagrams.Compression mechanisms are motivated by the limited spaceavailable in 802.15.4 frame for IPV6 packets. In fact, thesize of the 802.15.4 frame payload (102 bytes) leaves limitedspace for an IPV6 packets as 48 bytes are required only for itsheader. 6LoWPAN defines encoding formats for compressionbased on shared state within contexts. The main idea behindis to take advantage of the fields that are implicitly known toall nodes in the network or can be deduced from lower layers.The compression scheme consists of IP Header Compression(IPHC) and Next Header Compression (NHC).
4
Fig. 3: IPHC
IPHC encoding describes how an IPv6 header iscompressed. As depicted in Figure 3, 13 bits of the 2bytes long IPHC are used for compression. The IPv6 headerfields that are not compressed are placed immediatelyafter IPHC. Moreover, NH field in IPHC indicates whetherthe following header is encoded using NHC. If so, NHCencoding follows immediately the compressed IPV6 header.Compression formats for different next headers are identifiedby a variable ID bits plus the specific header compressionencoding bits. The NHC to encode IPV6 extension headersand UDP header are already defined. For more details on6LoWPAN, we refer to the corresponding RFC: RFC6282[5].
IV. NETWORK SCENARIO
We assume an e-health application scenario consistingof smart objects (contextual sensors), gateways and remoteentities (See Figure 4). IP-enabled smart objects are in chargeof sensing health related data (e.g. blood pressure, bloodglucose level, temperature level, etc.). They are planted inor on the human body. Gateways connect these objects to abackend infrastructure such as Internet. It is worth noticingthat user’s smartphones could be used as gateways. It isalways close to the user and thus spares constrained objectsfrom transmitting data over long range radio transmission.Energy consumption can then be reduced. Remote entities arein charge of further processing the received data and makingthe results available to caregiver services. Appropriate actionscan then be carried out according to the received information.
Securing e-health applications relies on efficient keymanagement schemes that ensure a reliable key distribution.We do believe that the best approach to tackle securitychallenges in the evolving IoT is to focus our efforts onstandard based protocols. We have chosen Mikey-Ticket forits simplicity and its adaptation to centralized scenarios whichsuits well our e-health application. However, standard-basedprotocols were designed to be used in an unconstrainedenvironment which does not take into consideration resourcelimitations.
Smart objects have limited computational power, memoryand energy resources, whereas gateways are much less resourceconstrained and are comparable to standard routers. Remoteentities can take the form of a server hardware or beingdistributed in a Cloud infrastructure with dynamic resources.The mapping with Mikey-Ticket concepts is defined as follows:Remote Entity: Initiator,
Gateway: Key Management Server,Smart object: Responder.
V. REDUCING MIKEY-TICKET OVERHEAD
In this section, we propose to tailor Mikey-Ticket to ournetwork scenario by reducing the overhead on the constrainedentities. We focus on two aspects: i) Reducing the messageexchanges overhead. ii) Reducing the packet size of theexchanged messages.
A. Novel exchange modeOur novel exchange mode for Mikey-Ticket is designed to
involve the constrained node as less as possible. We considerthe remote entity as the Initiator of the protocol and theconstrained node as the Responder. The remote entity is incharge of requesting the establishment of a session key withthe constrained node and periodically launching updates. Weassume that I and R are sharing security credentials with theKMS that is in charge of generating, deriving and deliveringthe required keying materials. Our exchange mode is depictedin Figure 5 and Table 1 summarizes the different notationsused.
REQUEST INIT: The Initiator starts the exchange bysending a REQUEST INIT message to KMS. The messagecontains the identities of I (IID), KMS (KMSID) and R(RID). The message also contains a nonce NI generatedby I which will be used as a session identifier. A MAC iscomputed using aKI,KMS to ensure message authenticity.The message has the following structure:{[IID, RID,KMSID, NI ]eKI,KMS
, MAC}
REQUEST RESPONSE: KMS receives REQUEST INIT,it checks the freshness of NI and validate the MAC usingaKI,KMS . Upon successful verifications, KMS decrypts themessage using eKI,KMS and retrieve the different identities.If the request is authorized, KMS generates the requestedkey KI,R and use the Key Derivation Function defined inRFC3830 [19] to derive aKI,R and eKI,R. R constructstwo versions of REQUEST RESPONSE. The first versionis intended to I. It is thus authenticated and encrypted usingthe shared credentials with I. The second version is intendedto R. It is thus authenticated and encrypted using the sharedcredentials with R. REQUEST RESPONSE has the followingstructure: {[IID, RID,KMSID, aKI,R, eKI,R, NI ]eKKMS,X
,MAC}. The used encryption key depends on the recipient ofthe message. REQUEST RESPONSE is then sent to I and R.
TRANSFER END: After receiving RE-QUEST RESPONSE, R checks the freshness of NI
5
Fig. 4: Network Scenario
and validate the MAC using aKKMS,R. Upon successfulverifications, R decrypts the message and retrieve both aKI,R
and aKI,R. I proceeds the same way as R to retrieve aKI,R
and eKI,R upon REQUEST RESPONSE reception.R constructs TRANSFER END as a verification message. Itgenerates a nonce NR and computes a MAC on IID, RID
and NI with aKI,R. The message is then sent to I. It has thefollowing structure:{[IID, RID, NI , NR]eKI,R
, MAC}.I checks the freshness of NR to avoid any replay attacks andvalidates the MAC. A successful verification is considered asa proof of R’s knowledge of both aKI,R and eKI,R.
Our new mode exchange reduces the number of exchangedmessages from 6 to 4 messages compared to RFC 6043 [2].The constrained node is only involved in 2 messages. Weoffload the derivation of the authentication and encryption keyto the KMS, thereby further reducing the overhead.
B. Novel header compression scheme
In this section, we describe our proposed 6LoWPAN headercompression scheme for Mikey-Ticket. The Mikey-Ticketcommon header is contained in the UDP payload. However,despite 6LoWPAN has defined header compression forUDP, no NHC is defined in case where headers containedin UDP payloads are compressed. We therefore, proposeto use the 6LoWPAN extension proposed in [10]. Theextension indicates that the UDP payload is compressed with6LoWPAN-NHC. Specifically, our UDP payload containsMikey-Ticket Common header payload compression.
Figure 6 shows our proposed 6LoWPAN-NHC-HDR forMikey-Ticket. 13 bits are required to encode the differentfields. Nevertheless, in order to remain standard compliant(The size of NHC encodings is multiple of bytes), 6LoWPAN-NHC-HDR is two bytes long. We present in detail eachfield compression: To comply with 6LoWPAN-NHC encodingschemes, we set aside the first four bits as the ID field touniquely identify our NHC encoding. We set the ID bits to
1100. To the best of our knowledge, the 1100 bits are currentlyunused as NHC identifiers.
- Version (V): If 0, the version is the default Mikey-Ticketversion defined in [19]. The field is skipped. If the bitis set to 1 the version number is carried in line after the6LoWPAN-NHC-HDR header.
- Data type (DT): The data type field describesthe type of the exchanged message. Based on ournew exchanged mode (See section 4.1), we onlyconsider three types of messages (i.e. REQUEST INIT,REQUEST RESPONSE, TRANSFER END) plus theERROR type. We propose to encode the data type fieldwith two bits using the following encoding:
DT = 00 : REQUEST INITDT = 01 : REQUEST RESPONSEDT = 10 : TRANSFER ENDDT = 11 : ERROR
- Verification V (VF): The VF field encoding is similar tothe non-compressed header. If set to 0, no verificationmessage is used. If set to 1 a verification message isrequired.
- PRF func (PRF): If 0, the default PRF function definedin [19] is used. If set to 1, the PRF function value iscarried in line.
- CSB ID (CSB): The CSB ID is chosen by the Initiatorand needs to be unique between each Initiator-Responderpair. Instead of carrying its 32 bits size inline, wepropose its derivation from the concatenation of lowerlayer identifiers (e.g. IPV6 addresses). 1 bit is sufficientfor the encoding. If set to 0, the CSB ID is derivedinstead of being carried in line. If set to 1, the 32bits CSB ID are carried after the 6LoWPAN-NHC-HDRheader.
- ] CS: We assume only one CS in each CSB. It allowsus to encode the ] CS with one bit. If set to 0, only oneCS is considered. If set to 1, the number of CS is carriedin line.
- CS ID map type(MT): If 0, the default GENERIC-ID
6
Fig. 5: Novel message exchange mode
Fig. 6: 6LoWPAN-NHC-HDR encoding for Mikey-Ticket header
map type defined in [2] is used. If set to 1, the CS IDmap type is carried in line.
- CS ID map info (MI): The CS ID map info size is keptvariable in [2]. If we assume only one CS in each CSB,we are able to use 1 bit for the encoding. If 0, the uniqueCS is identified with its corresponding mapping to thesecurity protocol for which security associations shouldbe created. If 1, the map info field is carried in line.
The next payload field is always carried in line. The threelast bits are used as padding bits to remain standard compliant(NHC size is defined as 2 bytes long).
VI. CONCLUSION AND FUTURE WORK
In this paper, we have introduced the idea of tailoringMickey-Ticket for the constrained environment of e-healthapplications. We have therefore proposed a new exchangemode to reduce the number of exchanged messages fromsix to four along with a new header compression scheme toreduce the message size. As future work, we plan to formallyvalidate our proposed mode exchange. Additionally, we alsoplan to implement our extensions on real test-beds to quantifyenergy cost savings compared to the current Mickey-Ticketspecification.
REFERENCES
[1] L. Atzori, A. Iera, and G. Morabito, “The internet of things: A survey,”Computer Networks, p. 19, May 2010.
[2] J. Mattsson and T. Tian, “Mikey-ticket: Ticket-based modes of keydistribution in multimedia internet keying (mikey),” RFC 6043, IETF,2011.
[3] R. Hummen, J. Hiller, H. Wirtz, M. Henze, H. Shafagh, and K. Wehrle,“6lowpan fragmentation attacks and metigation mechanisms,” Proc. 6thACM Conf. Security Privacy Wireless Mobile Netw, pp. 55–66, Apr2013.
[4] G. Montenegro, N. Kushalnagar, J. Hui, and D. Culler, “Transmissionof ipv6 packets over ieee 802.15.4 networks,” RFC 4944, IETF, 2007.
[5] J. Hui and P. Thubert, “Compression format for ipv6 datagrams overieee 802.15.4-based networks,” RFC 6282, IETF, 2011.
[6] J. Granjal, E. Monteiro, and J. S. Silva, “Enabling network-layersecurity on ipv6 wireless sensor networks,” Proc. of IEEE GLOBECOM,2010.
[7] S. Raza, S. Duquennoy, T. Chung, D. Yazar, T. Voigt, and U. Roedig,“Securing communication in 6lowpan with compressed ipsec,” in Proc.of IEEE DCOSS, 2011.
[8] S. Raza, S. Duquennoy, and G. Selander, “Compression of ipsec ah andesp headers for constrained environments,” draft-raza-6lowpanipsec-00(WiP), IETF, 2013.
[9] S. Raza, T. Voigt, and V. Jutvik, “Lightweight ikev2: A key manage-ment solution for both compressed ipsec and ieee 802.15.4 security,”IETF/IAB workshop on Smart Object Security, 2012.
[10] S. Raza, D. Trabalza, and T. Voigt, “6lowpan compressed dtls for coap,”in Proc. of IEEE DCOSS, 2012.
[11] R. Hummen, J. Hiller, M. Henze, and K. Wehrle, “Slimfit - a hip dexcompression layer for the ip-based internet of things,” WiMob, IEEE,2013.
[12] Y. B. Saied and A. Olivereau, “D-hip: A distributed key exchangescheme for hip-based internet of things,” in Proc. of IEEE WoWMoM,2012.
[13] Y. B. Saied and A. Olivreau, “(k, n) threshold distributed key exchange-for hip based internet of things,” in Proc. of ACM MobiWac, 2012.
[14] Y. B. Saied and A. Olivereau, “Hip tiny exchange (tex): A distributedkey exchange scheme for hip-based internet of things,” in Proc. ofComNet, 2012.
[15] R. Bonetto, N. Bui, V. Lakkundi, A. Olivereau, A. Serbanati, andM. Rossi, “Secure communication for smart iot objects: Protocol stacks,use cases and practical examples,” In Proc. of IEEE WoWMoM, 2012.
[16] T. Freeman, R. Housley, A. Malpani, D. Cooper, and W. Polk, “Server-based certificate validation protocol(scvp),” RFC 5055, IETF, 2007.
[17] R. Hummen, H. Wirtz, J. H. Ziegeldorf, J. Hiller, and K. Wehrle,“Tailoring end-to-end ip security protocols to the internet of things,”in Proc. of IEEE ICNP, 2013.
[18] R. Hummen, J. H. Ziegeldorf, H. Shafagh, S. Raza, and K. Wehrle,“Towards viable certificate-based authentication for the internet ofthings,” HotWiSec ’13 Proceedings of the 2nd ACM workshop on Hottopics on wireless network security and privacy, 2013.
[19] J. Arkko, F. Lindholm, M. Naslund, and K. Norrman, “Mikey: Multi-media internet keying,” RFC 3830, IETF, 2004.
Hardware AES IP for Embedded Cryptosystem on FPGA
Anane Nadjia
CDTA (Centre de Développement des Technologies Avancées) Algiers, Algeria
Anane Mohamed ESI (Ecole nationale Supérieure d’Informatique)
Algiers, Algeria
Abstract—AES (Advanced Encryption Standard) is a symmetric-key algorithm where a same key is used for both encrypting and decrypting data. In this paper, we present three AES hardware architectures with several performances levels: low, moderate and high encryption rates. These architectures can be used as IPs in a Hybrid cryptosystem RSA-AES implemented on an FPGA PSoC. In this work, the S-Box memories are implemented on FPGA circuit Slices which offer generation functions of 8 inputs in one Slice. This has advantageously improved the performance and occupied area of S-Box memories. The Xtime() function has been used in the implementation of the MixColumn transformation. The resulting architectures are adaptable to several applications and offer a good compromise between performance and area.
Keywords—AES-RSA, Cryptography, FPGA.
I. INTRODUCTION The rapid and continuous development of communication
via open networks such as Internet has created a growing need to secure sensitive or confidential data before their transfer. This is often ensured by using encryption methods whose key sizes are increasing with the technology development, which provides to code breakers more important computing means to find information without having the decryption key.
To ensure high security level and execution performance, encryption/decryption algorithms require more computing and more memory which are provided by hardware implementation or combined hardware/software implementations on SoC (System on Chip).
Two encryption algorithms are commonly used: symmetric and asymmetric. The first ones suffer of the key exchange and their management [1], whereas the others suffer of the encryption slowness due to their complex calculations. Hybrid applications can solve problems of each other, where an asymmetric encryption algorithm is used to exchange the key whereas a symmetric algorithm encrypts/decrypts the information. In these applications, the most used combinations are RSA-AES [2] and ECC-AES [3].
In this work, we focus on hardware implementation of an IP (Intellectual Property) for the AES, dedicated to a hybrid RSA-AES crypto-system on a Programmable SoC (PSoC).
In this PSoC application, the AES is implemented on FPGA, but the RSA and the AES KeyExpansion are implemented in software and run on MicroBlaze.
The reminder of this paper is organized as follows. In Section 2, we detail the AES principles. In Section 3, we summarize the architectures developed in this work, where we focus on the hardware implementation of the two transformations SubByte and Mixcolumn. Section 4 gives the
simulation and implementation results on FPGA circuit of Virtex-5. Finally, we end with a conclusion given in section 5.
II. THE AES STANDARD The AES is used in many applications related to security
[4] as it combines performance, efficiency, implementation ease, flexibility and it is often the sole associated with an asymmetric algorithm for hybrid encryption. The AES is a block cipher algorithm based on one symmetric key used for encrypting/decrypting data. The data block size is 128 bits partitioned in a matrix state of 4 ×4 bytes and the key size is of 128, 192, or 256 bits.
AES operates on several transformations called Rounds that convert the plaintext in an encrypted one or vice versa. The number of Rounds is 10, 12 and 14 for respective key sizes 128, 192 and 256 bits. The ten rounds of the AES-128 for encryption/decryption are shown on Figure 1.
subbytes
addroundkey
addroundkey
subbytes
shiftrows
mixcolumns
addroundkey
9 rounds
ciphertext
rk[i]
shiftrows
rk[i]
rk[0]
Encryptionplaintext
128
invsubbytes
addroundkey
addroundkey
invsubbytes
shiftrows
invmixcolumns
addroundkey
9 rounds
plaintext
rk[10]
invshiftrows
Decryptionciphertext
128 roundkeysrk[0]
128
rk[10]
Keyexpansion
Key
Operations involved by AES are:
• KeyExpansion: where the Round-keys are generated. • Initial Round executes the AddRoundKey transformation by
a simple XOR between the state matrix and the Round-key 0. • The ten identical Rounds: from the first round to the last one,
the following four transformations are executed sequentially: 1) SubByte – a linear substitution where each byte is
replaced by another one found in a specific substitution table related to AES, the S–Box.
Fig. 1. AES encryption/decryption
2) ShiftRow – a transposition where each row of the state matrix is cyclically shifted by 0, 1, 2 or 3 bytes respectively.
3) MixColumn - apply a linear transformation on the Galois field GF (28) for each column (4 bytes), which diffuses the value of each byte on the others.
4) AddRoundKey calculates a (XOR) between the state matrix at the input and a Round-key. It is the sole operation that involves the Round-key value. • Final Round (where the MixColumn step is skipped).
The AES decryption operation is the inverse of transformations: InvShiftRow(), InvSubByte(), InvMixColumn() and AddRoundKey () in the reverse order.
III. HARDWARE ARCHITECTURES OF AES AES hardware architectures are divided in three main types
ranging from the less expensive in resources and less efficient to that consuming more resources with presenting the best encryption/decryption rate:
1-Serial/Serial Architecture, where the message block to be encrypted is introduced in the architecture in serial and rounds are executed sequentially.
2- Parallel/Serial Architecture, where the message block is inserted in parallel while rounds are executed serially.
3- Parallel/Pipeline Architecture, where the message block is introduced in parallel and rounds are executed in pipeline. This architecture presents the best throughput with more resources.
What is common in these AES architectures is the execution of the four transformations: SubByte, ShiftRow, Mixcolumn and Addroundkey.
A. SubByteTransformation In this transformation, each byte of the State matrix is
replaced by another byte found in the "S-Box" table. It involves at least one byte and at most all the state or 16 bytes. The hardware implementation of this function requires a memory of 28×8 bits (2 Kbits). The number of S-Box depends on the data size to be processed in parallel which may be up to 16 bytes and requires 16 S-Boxes with a total memory of 32 Kbits. In FPGA implementation, we can use SRAM blocks or LUTs (Look-Up-Table). In the first approach, we use a SRAM block for each S-Box or a block of 18 Kbits for a memory of 2 Kbits which presents a great waste of resources. Hence the second approach is more economical when we use LUTs as generation functions. Each bit of the S-Box memory will be considered as a generation function of 8 address inputs of the S-Box, which is the byte to be substituted. Then, the implementation of this function requires 4 LUTs or one slice of Virtex-5 FPGA [5]. The implementation of an S-Box requires 8×4 LUTs representing 8 slices on the FPGA circuit of Virtex-5.
B. Mix Column transformation This transformation operates on State matrix columns after
the S-Box and ShiftRows. It consists in multiplying the column by an AES pre-defined matrix in the Galois field GF(28).This multiplication is illustrated on figure 2.
= ×
The MixColumns results are obtained by using operations
in GF(28). Every element of GF(28) is a 7 degree polynomial with coefficients in GF(2) (which is equivalent to Z2). Thus, coefficients of each polynomial term can take the value 0 or 1. As a 7 degree polynomial consists of 8 terms, an element of GF(28) can be represented by a string of 8 bits, where each bit represents a coefficient. For example, the binary string 10101011 represents the polynomial: x7 + x5 + x3 + x +1.
The term xi is in the polynomial expression if the corresponding coefficient is equal to 1. The term is omitted of the expression if the coefficient is equal to 0.
The addition of two elements in GF(28) is performed by a simple XOR of the two elements bits to be added. While the multiplication of two elements in GF(28) requires more work. The multiplication of two elements of Z2 is similar to an AND gate.
The multiplication in GF(28) can be performed by multiplying each term of the second polynomial by all terms of the first polynomial. Then, these products are summed. If the new polynomial degree is higher than 7, then, it must be reduced modulo a certain irreducible polynomial. In AES, the irreducible polynomial is x8 + x4 + x3 + x +1. This can be accomplished by multiplying the latter by xi-8, where i is the polynomial degree to be reduced. Then, this result is added to this polynomial. We continue this process until the degree of the resulted polynomial is lower than 7. In this work, multiplicands are elements of the two AES matrices of encryption/decryption and are split into simple elements (02) and (01) such as the example (03) = (02) + (01). Multiplication by (03) becomes multiplying by (02) and (01), then summing these products. It should be noted that (01) is the identity element of the multiplication in GF(28). While the multiplication of one byte by (02) is denoted "Xtime ()" and may be considered as a multiplication of the polynomial representing this byte by x. This will result in incrementing degrees of polynomial elements by 1.
If the resulting polynomial degree is 8, it is reduced by adding the AES irreducible polynomial, otherwise nothing is done. The function Xtime(A) = 02•A is as a left shift of A by one bit if its MSB is 0 , otherwise it is equal to the same value shifted by one bit to the left, followed by a XOR with the value {1B}. The hardware implementation of this function is illustrated on figure 3.
Fig 2. AES Mix Columns operation
Fig.3 Hardware implementation of Xtime()
C. ShiftRow transformation It is a shifting of the state matrix bytes as shown on figure 4.
In parallel processing, 16 bytes of the state matrix are considered at once and this transformation is performed by a simple wired shifting. Nevertheless, for a serial implementation, a strategy is considered concerning the order of bytes to be processed to avoid a bottleneck met before the MixColumns transformation; we have addressed at least one column and the resulted column after ShiftRows. The proposed solution for serial architectures is that the ShiftRows operation is performed firstly in the round. D. Serial/Serial Architecture
In this architecture, the block to be encrypted is introduced in serial. The data size must be a multiple of bytes, i.e. 1, 2, 3 or 4 bytes. To avoid waiting at the MixColumns, a column of 4 bytes is considered and the operation is immediately launched after the S-Box. Hence, the data path size of the serial/serial architecture is 32 bits as shown in figure 5.
Columns w1, w2, w3 and w4 are loaded into a latch register of 128 bits; data are selected via a switcher to undergo the round transformations: SubByte, MixColumn and Adroundkey. We note that the ShiftRow is done by wiring at the beginning of the round during the loading and into the loop from the output to the input. Table 1 summarizes hardware and temporal characteristics of the Serial/Serial architecture for encrypting a block of 128 bits.
TABLE 1. CARACTERISTICS OF AES SERIAL/SERIAL ARCHITECTURE
One round 10 rounds S-Boxs 4 4 Xor Gates 32 32 Mixcolumn cellules 1 1 Iterations 4 4×10 Clock cycles 4 4×10
E. Parallel/Serial Architecture In this architecture, data of 128 bits enter in parallel, i.e. the four columns at once. This architecture is shown in figure 6.
This architecture has the same components as the Serial/Serial one with the absence of switchers but with duplication by four of the round module. This is a sequential architecture with a parallelism processing in the state block. This will reduce the number of iterations by four hence the execution time. In this architecture the iteration number is reduced by four, while resources are multiplied by four with an average rate quadrupled compared to the Serial/Serial architecture. Table 2 summarizes hardware and temporal characteristics of the Parallel/Serial architecture encrypting a block of 128 bits.
TABLE 2. CARACTERISTICS OF AES PARALLEL/SERIAL ARCHITECTURE
One round 10 rounds S-Boxes 16 16 Xor Gates 128 128 Mixcolumn cellules 4 4 Iterations 1 10 Clock cycles 1 10
F. Parallel/Pipeline Architecture
This Architecture allows to the four 32-bits words to be performed together and rounds are executed in pipeline, which gives a higher encryption/decryption rate. The Parallel/Pipeline architecture is shown on figure 7. The pipeline concept is inspired of the mechanical engineering operation of an assembly line. This concept is used in this architecture to increase the throughput [6]. Our architecture is a chain of 11 stages which work in parallel, where each stage is dedicated to a specific round and the
Fig. 4. ShiftRows transformation
Fig. 5. Serial/Serial Architecture
Fig.6. Parallel /Serial Architecture
encryption/decryption of a 128 bits message block is performed at each clock cycle.
State(Cipher text)
word1'’ word2'’ word3'’ word4'’
Clk
32
Subbyte
Adroundkey
32
Subbyte
Adroundkey
32
Subbyte
Adroundkey
32
Subbyte
Adroundkey
32
SubbyteMixcolumn
32
Subbyte
Mixcolumn
32
Subbyte
Mixcolumn
Subbyte
Mixcolumn
word1‘ word2‘ word3‘ word4‘
AdroundkeyAdroundkey Adroundkey Adroundkey
32
Rkey(i)
word1 word2 word3 word4Clk
State (Plaintext )
word1 word2 word3 word4Clk
AdroundkeyAdroundkey Adroundkey Adroundkey
32323232
Rkey(0)32323232
Wired ShiftRows
Wired ShiftRows
Rkey(10)
Table 3 summarizes the hardware and temporal characteristics of the Parallel/Pipeline architecture.
TABLE 3. CARACTERISTICS OF AES PARALLEL/PIPELINE
ARCHITECTURE One round 10 rounds S-Boxs 16 160 Xor Gate 128 1280 Mixcolumn cellules 4 40 Latency 10 Clock Cycles Throughput 128 bits/clock cycles
V. IMPLEMENTATION RESULTS The AES architectures have been implemented on an
FPGA circuit of Virtex-5, the xc5vlx330t-2ff1738. The ISE 12.2 (Integrated Software Environment) Design suit and VHDL were used for the design of the three architectures. The functional verification of architectures has been realized with the ISim (ISE Simulator).
The Proposed implementations are based on optimizations at the lowest level (slice level) of the two transformations SubByte and MixColumn. These optimizations have resulted in high frequency operation combined with a small occupied area. The implementation results of the architectures have shown that a period of 4.34 ns is sufficient to perform the four transformation of the round, which corresponds to a frequency
of 230.57 MHz. The throughput of the three encryption architectures are given in Table 4.
TABLE 4. THROUGHPUTS OF THE THREE AES ARCHITECTURES
Optimization at the lowest level has also resulted in a
reduction of the area in terms of number of slices consumed by our three architectures, for example the most voluminous architecture is parallel/Pipeline which occupies only 795 slices.
VI. CONCLUSION In this paper, we have developed three architectures for the
AES standard. These architectures can be used as IPs in reconfigurable PSoC cryptosystems, where the AES encryption/decryption is needed. These architectures are adaptable to various applications ranging from the low encryption rate with low area resources to high encryption rate with a relatively large area resource.
During the design of these architectures, a particular interest was granted to the optimization of the two transformations SubByte and MixColumn, which are the most expensive transformations of AES. This optimization has contributed significantly in increasing the AES execution performances by achieving frequencies around 230 MHz. These performances can be easily improved by the introduction the pipelining in the round execution.
REFERENCES [1] C. Paar, J. Pelzl, “Understanding Cryptography“, Springer-
Verlag 2010. [2] M.N. Praphul, K.R.Nataraj, “FPGA Implementation of Hybrid
Cryptosystem”, International Journal of Emerging Science and Engineering (IJESE), Volume-1, Issue-8, June 2013, pp. 14-19.
[3] Rajpal Amit Jayantila ,K.Brindha, G.Ramya, “Secured Data Transfer in Wireless Networks Using Hybrid Cryptography”, International Journal of Advanced Research in Computer Science and Software Engineering, Volume 3, Issue 10, October 2013, pp. 379-381.
[4] N.F. Standard. Announcing the Advanced Encryption Standard, Federal Information Processing Standards Publication 197, 2001.
[5] Bin Liu & Bevan M. Baas, “Parallel AES Encryption Engines for Many-Core Processor Arrays”. IEEE Trans on Computers, November 2011.
[6] Santhosh Kumar, K.Navatha, Syed mushtak Ahmed, “Implementation of AES Algorithm on MicroBlaze Processor in FPGA”, International Journal of Advanced Research in Computer and Communication Engineering, Vol. 2, Issue 10, October 2013.
[7] Ankita Nampalliwar, Sheeja Suresh, “Design and Implementation of AES Algorithm Using FPGA”, International Journal of Advance Research in Computer Science and Management Studies,Volume 2, Issue 1, January 2014.
Architecture Throughput (Gbits/s) Serial/Serial 0.737
Parallel/serial 2.94 Parallel/Pipeline 29.49
Fig. 7. Pipeline/Parallel Architecture
A reputation-based approach using collaborative indictment/exculpation for detecting and isolating selfish
nodes in MANETs Lotfi Zaouche(1), Sofiane Ait Arab(2), Anfel Khireddine(3),
Mawloud Omar(3), Enrico Natalizio(1), Abdelmadjid Bouabdallah(1) (1) Heudiasyc Lab - UMR CNRS 7253, Université de Technologie de Compiègne, <name.surname>@hds.utc.fr
(2) LISSI Lab - Université de Paris Est Créteil, [email protected] (3) LIMED Lab – Université de Bejaia, Algérie, {[email protected], [email protected]}
Abstract—Collaboration between nodes in Mobile Ad hoc Networks (MANETs) is very important for the proper functioning of the network. This is an assumption that has to be fulfilled in the design of routing protocols. However, this is not always true since some nodes could misbehave in order to have some benefits or simply avoid wasting resources. In this paper we question this assumption that does not take into consideration the bad behavior of nodes involved in the routing protocols. We analyze the characteristics of the existing solutions, and we propose a reputation-based mechanism that isolates selfish nodes based on control packets generated as a result to nodes’ observations on the behavior of other nodes. We propose a mathematical framework to increase/decrease the reputation of a node depending on the situation and the observation condition. We show via simulation that our solution achieve remarkable improvements in the delivery rate of packets, more than satisfying results concerning false positive and false negative, and it shows that the overhead caused by our system is negligible.
Keywords—MANET; selfish behavior; reputation system;
I. INTRODUCTION
In Wireless Sensor Networks (WSN), collaboration between nodes is essential. If a source and destination of data flow are not in line of sight, the information should be transmitted along intermediate nodes, along a path established and maintained by the network. Routing in such conditions becomes a complex task, especially as energy resources are limited, and nodes can legitimately become selfish and refuse to route other nodes’ packets to preserve their energy. The need for cooperation between nodes to ensure the functioning of the network conflicts with the individual interest of each node to spend its energy solely for data for which they are the source or the destination. We can identify two types of non-cooperative nodes: faulty/malicious nodes and selfish nodes. Faulty/malicious nodes belong to the class of nodes that are either defective and therefore cannot follow a well-defined protocol, or intentionally malicious, thus trying to attack the system [5].
Although the problem of selfishness is a form of passive attack, it still causing a negative impact on network’s performances. Numerous studies have been performed to evaluate the impact of the presence of selfish nodes in an ad hoc network [6, 7, 9]. The non-cooperation of a node implies that packets passing through this node will be lost. The mentioned problem calls for solutions that force the selfish nodes to cooperate in the network and if necessary, excluding them. Such solutions would greatly increase the network performance. In this work, we are interested in the study of the proposed problem of selfishness in mobile ad hoc networks.
Our contribution is summarized in the following points: We propose an improvement of the TWOACK scheme [3], aimed at considerably decreasing the number of control packets. Also, the TWOACK scheme is only able to detect a selfish link, whereas our solution detects the selfish node. We propose
reward/punishment model for cooperative/selfish nodes by taking into consideration all nodes that participate in the deliverance of a packet.
The rest of the paper is organized in four sections. In Section II, we present the state of the art of solutions for the problem of selfishness. In Section III we present our solution approach. Section IV is devoted to the presentation of the simulation results. Section V concludes the paper.
II. RELATED WORK
Since the transmission of a message imposes a cost (energy and other resources) to the nodes, a selfish node will need an inducement or reward for transmitting messages from others [8]. There are two types of solutions to encourage selfish nodes in ad hoc mobile network to cooperate: credit-based systems and reputation-based system.
A. Credit-based system
Systems based on credit provide incentives to nodes to ensure network functionality. To achieve this virtual goal, a payment system may be implemented. Nodes are paid to rely other nodes’ packets. This kind of system can be implemented using two models: the Packet Purse Model (PPM), and Packet Trade Model (PTM) [2].
In [8], the authors proposed an interesting solution called SPRITE. When a node receives a message, it keeps a receipt of this message, then when it has a fast connection to the Credit Clearance Service (CCS) it reports its receipts of the messages it received/transmitted. CCS then determines the charge and the credit of each node involved in the transmission of the message.
In this type of solution, we are facing new problems such as the centralization/decentralization of the paying authority, false receipts and sometimes the solution needs to address not only software issues but also hardware ones.
B. Reputation-based System
A reputation-based system relies on the observations of nodes to other nodes. Since one observation does not allow a direct and objective measure of malicious nodes, it is necessary that each node maintains a degree of confidence in respect of all the nodes it has observed. The value of this confidence is influenced by observations on the behavior of nodes. In this type of system, the reputation calculation is either performed locally at each node, or by the distribution of reputations stored in the nodes within the network.
TWOACK scheme [3] is based on reputation. A node that transmits/broadcasts a message, is informed that the following node has completed its task by forwarding the message at his turn, by receiving from the two hops next node a special acknowledgment called TWOACK packet. Each node that receives a message must send an acknowledgment to the node two hops back in the message path. The message path is the path that has been given by the routing protocol. To detect a misbehaving node, the source maintains a list of IDs of
messages that he has not received TWOACK packet yet, and each node maintains a unique list of data structure for each transmission link that it uses.
S-TWOACK (Selective-TWOACK) [3] and 2ACK [4] schemes aims at reducing network congestion caused by the large number of TWOACK packets sent. The first is inspired by the principle of the sliding window, acknowledging a certain number of well received messages. The second one acknowledges only a part of them, and includes a certification mechanism for the security of its packets.
In respect of the presented solutions, our proposal consists in optimizing the control packets. We do not generate control packet until something does not work in the message delivery, whereas TWOACK generates control packet during all transmissions. Consequently, the more selfish nodes are discovered, the less control packets are generated.
III. PROPOSED APPROACH
Since it is the less constraining in the architectural design, and there is no need to have special hardware, we choose to work on Credit-based systems to force the nodes to participate with other nodes, in order to keep a good reputation and keep being well served by others.
A good solution should: (i) guarantee the detection of selfish nodes, (ii) penalize the selfish nodes, and avoid, in the routing phase, those excluded because they do not cooperate anymore (iii) be able to know if it is necessary to give a second chance to a node who wants to repent.
A. Assumptions
We assume that the links are bidirectional. We also assume that a certification service public key is set up and used to encrypt the messages circulating in the network, including messages that are unique to our system, and guarantee data integrity.
B. Operating details of our approach
The principle of our approach is quite simple, and relies on multi-hop acknowledgment. Several studies justify and prove that two hops is an efficient number of hops for the acknowledgments [3, 4]. Based on these studies, we choose to make two hops acknowledgment, because it will make the indictment of a node more precise than in the case with higher number of hops. Messages used by the system are described in Table I.
TABLE I. MESSAGES LIST Message Definition
2HopAck Message sent by a node Ni to Ni-2 node. SelfExculpation
Packet sent to the source by the last node that tried to transmit the message to report the refusal of a node to transmit the message,
Selfish_Detection
Packet sent by a node that does not receive the exculpation of his successor, thus accusing him of being selfish. It must be said here that if for example the message is sent by a node Ni, then the node Ni-1 will not accept to transmit that packet only if Ni+1 has exculpated Ni by sending a 2HopAck packet to him.
SelfishAlert
Packet sent in order to report the detection of a selfish node.
Knowing that we have made the assumption that a certification service is implemented in the network, 2HopAck and SelfExculpation messages will be encrypted to ensure their integrity and authenticity.
Let consider that a node Ns wants to send a message to Nd. Ns builds a path to Nd by using any routing protocol, but avoiding the known selfish nodes. By sending this packet, all nodes that are on its path will wait for an acknowledgement of the destination for an Ack_Delay time. Upon the reception of the acknowledgment is received, each node increases the
reputation of following nodes in the path. If the timeout Ack_Delay expires and no acknowledgment is received, each node Ni involved in the message transfer, send a 2HopAck message, to the node Ni-2, who was two hops back in the message path to prove that the node Ni-1 who was the intermediate is innocent. When the node Ni-2 receives this message, it will increase the reputation of the nodes Ni-1 and Ni. When after a delay, Exculpation_Delay, the node Ni sees that the node Ni+1 did not send the 2HopAck packet to node Ni-1, it exculpates itself by sending the previous nodes a package SelfExculpation and hold Ni+1 responsible for the failure of the transmission of the message, and decreases its reputation. Nodes receiving this message start increasing the reputation of nodes that transmitted the message to the node Ni, and since they do not know which node caused the problem, they will penalize the two nodes Ni and Ni+1, by decreasing the reputation of the node that seems most selfish. If after a delay, a node Ni does not receive the 2HopAck packet from node Ni+2, and the node Ni+1 does not proclaim its innocence, then it will be indicted by the node Ni, and will be signaled to the source node. All nodes in the path receiving this indictment will reduce the reputation of node Ni+1, and increase the reputation of other intermediate nodes. In addition, when a node receives a packet, it will check the message path, and will increase the reputation of all intermediate nodes from the source to him.
Each node holds a table named TrustTable that stores the values of reputation he has for other nodes. Whenever a node obtains an observation about a node, it updates the value of its reputation if it has an entry in the table for this node, if no entry in the table corresponds to this node, then it will create a new entry for this node and save the value inside. Initially, each node gives an initial reputation to neighboring nodes. In addition, each node holds a data table called PostTable, which stores the identifier of the messages it transmits, and the path it takes. When the destination sends an acknowledgment to the source to confirm the receipt of a message, all nodes that receive this acknowledgment, in their PostTable checks if there is a message that matches the packet acknowledged, then deletes the corresponding line. And nodes do the same when they receive SelfExculpation or SelfishDetection packets. If a timeout after no acknowledgment is received for a message, the line for this message will be deleted after it has sent the message to the node 2HopAck two hops back, as explained above to exonerate one hop rear's node.
In order to more quickly detect selfish nodes, it is preferable that the network nodes collaborate by exchanging knowledge on the behavior of other nodes. In order not to clutter the network messages, the nodes transmit the values of the reputation of a node only if it changes by a certain threshold. The value of the threshold should be well studied; a value too small will make the collaboration stronger and therefore more effective, but will increase the load on the network, and will be a waste of energy to the nodes. Also, taking a bigger value will reduce the network load, but it will also reduce nodes collaboration, making the detection of selfish nodes slower.
C. Reward and punishment computation
Formulas’ parameters and functions are detailed in Table II. TABLE II. PARAMETER LIST
Parameter/function Definition RSD Positive value to add to the reputation of a
node having transmitted a message. PNSD Positive value to subtract of the reputation
of a node as a punishment for failing to transmit a message.
APNSD Positive to subtract also the node that seems to be the selfish node value.
RSM Positive value to add to the reputation of a node as a reward for reporting the bad behavior of a node.
RST Positive value to add to the reputation of a node as a bonus for sending 2HopAck packet.
lastBroadcastedTrusti(j) The last value of reputation had broadcast the node i about the node j.
Ack_Delay
The time that a node has to wait for acknowledgment of the message he collaborated to transmit. After that, the node Ni launches Exculpation_Delay.
Exculpation_Delay
The time node Ni must wait for the 2HopAck packet from Ni+1 toward Ni-1. Beyond this period, the node Ni sends a packet SelfExculpation.
Others_Exculpation_Delay
The time a node Ni must wait 2HopAck packet from node Ni+2 exculpating the node. Ni+1.
��������� The initial reputation of each node. ��������� Change threshold that must wait before
broadcasting the new reputation of a node. Threshold The threshold for which a node is
considered selfish if it goes below it. WitnessRate The required rate of nodes accusing a node
to be selfish, to be considered as such throughout the network.
MPS(i, j) A function that returns 1 if the reputation of the node j is greater than those of node i, and returns 0 otherwise.
Trusti(j) A function that returns a reputation of node j that holds node i, which is stored in the TrustTable.
Upon receipt of a reputation value of a node, a new reputation value is calculated for this node, taking into account the value received and the value we already had. To avoid defamatory values that distort the reputation of the nodes, we can take into account the values received with a low impact factor. We apply the following formula to calculate a new value of reputation taking into account the values received:
� ������� =�� ∗ � ������� + ∑ ��������� ��!��"#$
$%& '� + #
Where ( is the factor that is given to the reputation we have already calculated previously, receivedReputaion is the value of reputation received from any node.
When a node detects that the reputation of another node has fallen below a certain threshold, it broadcasts a packet named SelfishAlert, containing its identity and the identity of the accused node. Each node receiving this message will save it. If at a certain time you get a number of accusation for a given node, equal to WitnessRate, then this node will be considered selfish by all network nodes. This technique speeds up the process of detecting selfish nodes in the network.
The WitnessRate parameter is very important. Indeed, high levels reduce the rate of false accusations caused by defamatory information sent by malicious nodes, but the detection of selfish nodes becomes longer. A smaller rate ensures that the detection of selfish nodes is faster, but false positive rate could increase.
Ni rewards Ni+1 for sending the message to Ni+2 as follow: � �����)�*&� = � �����)�*&� + �+, (1)
And rewards Ni+2 for confirming it as follow: � �����)�*-� = � �����)�*-� + �+� (2)
When Ni receives a message from Ni-1, it applies: � �����)$' = � �����)$' + �+,, ∀01 ∈ {04, 06,… ,0894} (3)
A node penalizes his successor after it did not send 2HopAck packet, using the following formula: � �����)�*-� = � �����)�*-� − <)+, (4)
When a node receives a SelfExculpation packet of a node k, it executes: � �����)$' = � �����)$' + �+,, ∀01 ∈ {08*6, 08*=, … , 0>94} (5)
� �����)?� = � �����)?�–�1 −� �����)?�' ∗ <)+, −B<+�)? , )?*&� ∗ C<)+, + B<+�)?*&, )?� ∗ �+B
(6)
� �����)?*&� = � �����)?*&� − �1 − � �����)?*&�' ∗ <)+,−B<+�)?*&, )?� ∗ C<)+,
(7)
If we receive an acknowledgment of the destination these will be applied to each node Ni path: � �����)$' = � �����)$'+ �+,, ∀01 ∈ {08*4, 08*6, … , 0D94} (8)
� �����)�*-� = � �����)�*-�+ �+� (9)
If no exoneration is sent neither by the node Nk itself nor by the node that follows it, a Selfish_Detection packet is sent, and the reputation of node Nk will be reduced by all nodes in the path receiving this message by applying formula (4).
IV. PERFORMANCE EVALUATION
In this section we will describe the simulation environment and we will show the simulation results.
A. The simulation environment
For our simulation campaigns, we implemented TWOACK and our protocol on Java. We simulated an ad hoc network with 50 mobile nodes. Selfish nodes are randomly selected from these 50 nodes with a percentage ranging from 0 to 40% in steps of 5%. These nodes move according to the Random Waypoint model. The maximum speed of a node is 15 m/s and the maximum idle time is 3 s. Network nodes have the same range that is equal to 150 meters. The deployment surface of nodes is 1000 * 1000 m². Each node sends a message according to Poisson distribution with parameter E=10 ms. Finally, the simulation time is 300 seconds, and we repeat it 50 times each time we increment the selfish nodes rate in the network. To show that we are tolerant to collisions, we assume that the packets loss rate is 5%. As shown in [10], this packet loss rate is sufficient. Table III. summarizes these parameters.
TABLE III. SIMULATION PARAMETERS Parameter Value
Number of nodes 50 Rate selfish nodes 0-40 % (step 5%) Maximum Speed 15 m/s Maximum idle time 3 s Radius increased 150 m Width of the area 1000 m Length of the area 1000 m Simulation time 300 s ��������� 0.75 Threshold 0.25 RSD 0.1 PNSD 0.15 APNSD 0.05 RST 0.05 RSM 0.03 ��������� 0,2 WitnessRate 5 %
The important thing to show here as an input, is the rate of selfish nodes in the network, and the time to. Concerning the output, first, we will show the delivery ration of messages. Right after, we will present the overhead caused by the control packet generated by our protocol.
B. Results and discussion
Impact of selfish nodes on message delivery
As shown in Fig. 1, our approach gives excellent results; indeed the lower bound of the success rate of message delivery exceeds 86%. Initially, in the absence of selfish nodes in the network, the rate of rescues message reaches 97%, then increasing the rate of selfish nodes in the network, the delivery rate decreases to 86% when the selfish node rate reaches 40%. And we can see that TWOACK degrades significantly when the selfish rate increases. This is due to the fact that they do not discover the selfish note, but the failing link. So whenever a selfish node moves and create a new link with another node, it can behave as it likes.
When we fix the selfish rate to 40% as shown in Fig. 2, we have at the beginning a message delivery rate of 86% because we have not yet detected the selfish nodes, and as and as these nodes are selfish and discovers that avoids, we the message delivery rate increases till 95%, and it corresponds exactly to the 5% of assumed collision. Meanwhile, TWOACK stabilizes between 40 to 45%.
Fig.1: Packet delivery ratio
Fig.2: Packet delivery ratio over time
Control packets overhead on the network
The Fig. 3 shows the evolution of the protocol overhead over time. This is important to show to argue if our solution is too heavy to be interesting or not. Unfortunately, we did not see this kind of figure in the studied work. We can see that our solution is a very light one comparing to TWOACK. We do not exceed 4% of the network traffic, and we can do less than 1% in favorable condition. In fact, what makes control packets in our solution does not reach 0% is only caused by failing links, false detections, and collisions. Concerning TWOACK, we see that the control packet decreases, but this is only because the packets do not reach them destination and therefore no control packets are generated.
Fig.3: Routing overhead
Fig.4: Routing overhead over time
In the Fig. 4, we can see that the two protocols have stabilized over time. Our solution stabilizes in 1.3% and TWOACK in 8.7%. Another time, as we said before, we do not 0% only because of false detections, collisions and link failing.
V. CONCLUSION AND FUTURE WORK
In this paper, we proposed a new approach for detecting selfish nodes. The simulation results that we obtained were largely sufficient, and have proven the effectiveness and robustness of our approach. As noted in the previous section, our approach goes beyond 90% regarding the detection rate of selfish nodes, which avoids these and offer a delivery rate high enough messages (more than 91%). The evaluation of our approach with respect to time clearly demonstrates its advantages. In fact as the time passes, the rate of successful delivery of messages increases, and the message loss decreases. When 40 of network nodes are selfish, the initial packet delivery ratio is 86% of all the sent packets, which increases to 95% over time.
REFERENCES [1] H. Miranda and L. Rodrigues "Preventing selfishness in open mobile ad
hoc networks". 23rd International Conference on Distributed Computing Systems Workshops, May 2003.
[2] S. D. Khatawkar, U. L. Kulkarni, K. K. Pandyaji "Detection of Routing Misbehavior in MANETs", International Conference on Computer and Software Modeling IPCSIT vol.14 IACSIT Press. Singapore 2011.
[3] Kash yap Balakrishnan, Jing Deng, Pramod K. Varshney. "TWOACK: Preventing Selfishness in Mobile Ad Hoc Networks", Wireless Communications and Networking Conference, IEEE, Vol. 4, March 2005.
[4] Kejun Liu, Jing Deng, Pramod K. Varshney, and Kashyap Balakrishnan «An Acknowledgment-based Approach for the Detection of Routing Misbehavior in MANETs". IEEE Transactions on Mobile Computing, Vol. 6, Issue: 5, May 2007.
[5] S. Marti, T. Giuli, K. Lai, and M. Baker, "Mitigating routing misbehavior in mobile ad hoc networks». In Proceedings of the Sixth International Conference on Mobile Computing and Networking. Boston 2000.
[6] Satyanarayana Vuppala, Alokparna Bandyopadhyay, Prasenjit Choudhury, Tanmay De. "A Simulation Analysis of Node Selfishness in MANET using NS-3". Int. J. of Recent Trends in Engineering and Technology, Vol. 4, No. 1, November 2010.
[7] Shailender Gupta, C. K. Nagpal, Charu Singla. “Impact of selfish node concentration in manets”. International Journal of Wireless & Mobile Networks (IJWMN) Vol. 3, No. 2, April 2011.
[8] Shen Zhong, Jiang Chen, Yang Richard Yang, "Sprite: A Simple, Cheat-Proof, Credit- Based System for Mobile Ad-Hoc Networks". Twenty-Second Annual Joint Conference of the IEEE Computer and Communications. IEEE Societies. vol. 3, April 2003.
[9] Sundararajan, A.Shanmugam. "Modeling the Behavior of Selfish Forwarding Nodes to Stimulate Cooperation in MANET", International Journal of Network Security & Its Applications (IJNSA), Vol 2, No. 2, April 2010.
[10] Constantinos Dovrolis, Parameswaran Ramanathann, "Proportional differentiated services, part II: loss rate differentiation and packet dropping", Eighth International Workshop on Quality of Service, June 2000.
0 5 10 15 20 25 30 35 400
10
20
30
40
50
60
70
80
90
100
Percentage of Misbehaving Nodes (%)
Pac
ket D
eliv
ery
Rat
io
Our solutionTWOACK
0 0.5 1 1.5 2 2.5 3
x 105
40
45
50
55
60
65
70
75
80
85
90
95
Time (ms)
Pac
ket D
eliv
ery
Rat
io P
rogr
ess
Our solutionTWOACK
0 5 10 15 20 25 30 35 400
5
10
15
20
25
30
Percentage of Misbehaving Nodes (%)
Pro
toco
l Ove
rhea
d
Our solutionTWOACK
0 0.5 1 1.5 2 2.5 3
x 105
0
1
2
3
4
5
6
7
8
9
10
Time (ms)
Pro
toco
l Ove
rhea
d P
rogr
ess
Our solutionTWOACK
Survey of Change Impact Analysis Approaches for
Software Evolution
Hidouci Adenane
Constantine 2 University
Constantine, Algeria
Amghar Youssef
LIRIS, UMR 5205
INSA Lyon, France
Abstract— Information system is implemented through
many software systems of which maintenance is recognized as
being the most expensive phase of system life cycle. These
software systems continuously change for several reasons such
as: correction of anomalies, adaptation to the context, adding
functionalities or improving performances. However the
implementation of a single change can impact many parts of
the system. The high cost of maintenance leads researchers and
professionals to the investigation of change impact analysis
approaches which can reduce the maintenance cost each time
the software system evolves. This article aims at providing an
exploratory study of existing software change impact analysis
approaches while describing the techniques used, the
advantages and the limits of each type of approaches.
Keywords— Software maintenance; software evolution;
change impact analysis; classification.
I. INTRODUCTION
In order to automate daily tasks and to remain competitive, the companies must install multiple software tools such as ERP (Enterprise Resource Planning), CRM (Customer Relationship Management), Workflow systems, etc. These systems do not meet all the needs for the company and many other applications or software systems are also developed. These software systems are frequently modified for various reasons: to correct anomalies, to answer new requirements or to improve the performances. Independently of its causes, the changes within a software system are generally not limited to one component only but they can propagate towards several other software components of the system. This is called, "Ripple effects" [24]. So change impact analysis is the determination process of the set of software artifacts likely to be changed each time one artifact has been modified [18].
When changing software, impact analysis is required to preserve the quality of the whole system [15] . However, making software changes without visibility of their effects can lead to bad estimations, delay in release schedules, degradation in software design, unreliable software products, and premature retirement of the software system [8,18].
The high cost of maintenance imposes that research and business investigate the field of change impact analysis which aims at averting avoidable changes, capturing change effects in early stages and rendering possible prediction of future effects from past time experiences [5].
Several approaches were proposed to evaluate the change impact while being often based on the analysis of the dependency graph or on the traceability relations between the software components of the system at different abstraction levels. Furthermore, it was noted that the document the most used as bases knowledge in the majority of the impact analysis approaches is mainly the source code because this latter is easy to access and provided up to date information contrary to the conceptual documentation or specifications, for example, which is non-existent, incomplete or obsolete.
Considering the importance of the work already undertaken in this field and the numerous approaches which have been proposed, it appeared interesting to us to gather most relevant works as a comprehensive state of art. Thus, this article established a survey on software change impact analysis approaches currently used in the field of the software engineering and in particular in software maintenance. For each type of approaches, we will present the principle, the techniques and tools allowing their implementation, the advantages and the limits of such approaches.
The remainder of the paper is organized as follows:
section Ⅱ presents the work already achieved in this field. In
section Ⅲ, we will stress the types of impacts of a change
before exposing in section Ⅳ the various software change
impact analysis approaches. This paper concludes with some directions for future works.
II. RELATED WORKS
Although it is not new field, the change impact analysis was widely discussed in the literature and remains an active field of research which is in perpetual evolution. In 1979, Weiser [11] introduced the technique of "Program Slicing" which aims at extracting the parts of a program impacted by a given change. The concept of "Ripple effects" was introduced in 1980 by Yau and Collofello [11]. Their model allows the evaluation of the impact due to the propagation of a change in the source code. In 1996, Arnold and Bohner published a series of research papers entitled: "Software Changes Impact Analysis" by gathering principal works relating to the analysis of change impact [3]. In their approach of "Document-Driven Development", Turver and Munro claim that during the development of a new version of software, various types of documents (design documents, user's manual) must be updated as well as the source code. The impact on the source code also relates to the associated
documents [23]. In order to simplify the representation and the analysis of the networks of dependencies in the complex systems, Bohner and Gracanin [11] proposed to combine the impact analysis techniques with 3D visualization techniques.
Since the 1980s, there have been many investigations on change impact analysis, especially in the field of code-based techniques [4]. Wilkerson proposes in [24] taxonomy of the types of impacts that can result from source code changes in both procedural and object-oriented code. In [14] Kilpinen identifies three groups of impact analysis approaches: dependency impact analysis, traceability impact analysis and experimental impact analysis.
In traceability impact analysis, traceability links are the basic mean to express relationships between system components, and they are used to connect entities of different levels of abstraction, e.g. requirements and source code. Traceability links can be utilized to propagate changes, or the impact of changes to related elements.
Fig. 1. Types of change impact analysis techniques [13].
In dependency impact analysis group, impact analysis can be conducted within the internal links of a software system, such as dependencies between objects, methods, and variables. Several techniques which utilize program dependencies have been proposed, for example program slicing.
Although the source code based techniques are the most used to asses change impact (see fig. 2.) because they have the advantage of being more precise in their calculus by identifying the impact at the surest state (i.e. on the final product); however, they have the disadvantage of being limited in the range, time consuming and they require the implementation of the change before the impact can be computed.
Fig. 2. Impact analysis target in research papers, distribution over time
[11].
In contrast to source code based techniques like dependency analysis, experimental approaches are focused on manual analysis and therefore labor intensive techniques, such as audits, reviews, and code inspections [16].
III. TYPES OF CHANGE IMPACT
An assessment of the impact of certain change requires explicit information about the nature of the change, about the type of the involved artifacts, as well as about dependency relations between the artifacts [15]. That change impact can be primary or secondary according to the distance from the origin of the change.
A. Primary impact
Primary impact, also referred to as direct impact, corresponds to the components of the system that are identified by analysing how a proposed change affects the system. This analysis is typically difficult to automate because it is mainly based on human expertise.
B. Secondary impact
Secondary impact or indirect impact is complementary changes following a primary change due to design constraints, system’s structure or functional dependencies. The indirect impact can take two forms:
Side effects: are unintended behaviors resulting from the modifications needed to implement the change. Side effects affect both the stability and function of the system and must be avoided.
IV. SOFTWARE CHANGE IMPACT ANALYSIS APPROACHES
Based on a thorough literature review, we distinguish five main classes of software change impact analysis approaches:
Static impact analysis approaches.
Dynamic impact analysis approaches.
History based impact analysis approaches.
Information retrieval (IR) based impact analysis approaches.
Probabilistic based impact analysis approaches.
A. Static impact analysis approaches
1) Automatable approaches Automatable impact analysis approaches often employ
algorithmic methods in order to identify change propagation and indirect impact. Nevertheless, these approaches require a structured specification of the system such as the object model, the control flow diagram, data flow, and dependency graph .etc.
The automatable approaches are more suitable to evaluate the secondary impacts than to determine the primary impacts. Several approaches were developed for the analysis of dependencies but three main lines of techniques have been described in the literature for implementing such analysis approach:
a) Dependency analysis
The dependency analysis consist in exploring, capturing and defining the dependence relationships that exist between the software artefacts being in the same level of abstraction, generally on the level of the source code: modules, functions, classes, methods or variables.
A network of dependencies can be represented in two manners: the written way (the dependency graph) [7] or a matrix way (dependency structure matrix) [19]. It is sometimes possible to enrich the matrix by semantic information such as the type of relationships in order to improve and simplify its interpretation.
b) Traceability analysis
Traceability is the capacity to describe and trace the life cycle of software artefacts from requirement specification to source code and reciprocally [22].
The traceability analysis, on the other hand, can provide information on the existing relationships between the requirements, the design and the implementation artefacts, being thus a support for studying the impact of changes. The quality of the analysis of traceability in addition, is determined by the completeness of the whole of the relations which were identified gradually during the development of the system.
The traceability analysis has the advantage of being applicable at first stages of the software development cycle and thus allows identifying various types of impacted artefacts. Consequently, the field of application of the traceability analysis is much broader than that of the dependency analysis which completely relies on the availability of the source code.
Among the tools for traceability analysis: DOORS, Rational Requisite Pro and TOOR (Traceability of Object-Oriented Requirements). The weak point of all these tools is that they do not support automatic or semi-automatic maintenance of traceability links and do not take into account the new versions of the artefacts. Thus, the detection and the maintenance of the bonds of traceability must be operated manually, generating high costs. The investment is all the more heavy if one takes into account the iterative nature of the development processes [22].
c) Program slicing
The program slicing technique seeks to delimit wide change in a program by breaking it up into two independent parts: the part affected by the change and the part representing the remainder of the program. To locate the impacted elements, the technique of program slicing employ the data flow, the control flow or the dependency graph. The whole of these algorithms are thus based on the analysis of the source code of the programs.
Architectural slicing is an alternative which was introduced by Zhao in 1998 [11]. The advantage of architectural slicing is that it can be applied independently of the availability of the source code.
2) Semi automatable approaches Semi automatable approaches or interactive approaches
rely on collaboration between the man (the analyst) and the machine (an impact analysis support tool). The analysis process consists then in alternating control between these two entities as follows:
At each iteration, the tool explores the dependency relationships; compute a whole of elements likely to be affected by the change which are proposed to the analyst. This last examines then determines among the identified elements those which he considers relevant.
These approaches allow, on a side, to accelerate the calculation part of the process of impact analysis and on the other, to improve quality of the predictions by giving the possibility to the analyst of intervening to correct the possible errors made by the tool support.
Among the tools most used in this category, we can find: JRipples (Ripples Java) which was integrated thereafter in the Eclipse development environment.
3) Manual approaches Contrary to the automatable approaches, the manual ones
are easy to integrate in a process of change management. However they miss some precision of calculation.
The manual approaches, generally, are intended to determine primary impacts generated by a change i.e. to identify the whole of the elements of starting impact set.
a) Design document based manual approaches
A design document (an architectural model, UML diagram, a simple textual specification describing the components of the system .etc) is characterized by a whole of attributes like: the context in which it was published, its version, the goal of its creation, its frequency of update and the data which it provides. The precision of the predictions and the quality of the impact analysis depend primarily on the quality of the design documents as well as the following factors:
The availability and the quality of the documentation used during the analysis.
The homogeneity of the vocabulary, terminology and used notations.
The clearness and coherence of documentation.
The experiment and preliminary knowledge of the analysts on the studied system.
b) Interview approaches
According to a study on the analysis of impact (Lindval, 1997), the interview with the developers is considered as the fastest means for the capturing information on a software and the possible consequences of a given change [11].
In [10] authors have conducted an empirical study using interviews in order to understand how issues associated with impact analysis are seen at different levels and under different perspectives.
Indeed, the analysts find that it is more practical to question the developers which are quite informed on the system than searching in documentation, if it exists, needed information.
B. Dynamic impact analysis approaches
Dynamic impact analysis approaches consist in collecting, through a series of executions, information on the behavior of a program such as: order of the functions (procedures or methods) calls, the results turned over, the end of execution of the program… etc. The whole of the impacted elements is the whole of the parts of the program which are executed simultaneously with the initially modified parts.
1) Simulation Simulation is a dynamic approach which proposes also
mechanisms to estimate quantitatively and qualitatively the impact due to a change.
By means of simulation tools one can, for example, assign values to the various components of the system, simulate a whole of changes and analyze the results obtained. That means we can anticipate the undesirable repercussions of a possible change, study the various aspects of the change: time, cost, resources and provide estimates concerning the impact of the change on the system performances.
In spite of the advantages mentioned above, simulation presents some limits:
Failure to gather a large amount of data necessary to simulation what is sometimes difficult to achieve.
The quality of the collection of simulation data which must be consistency in order to cover all possibilities.
The simulation of complex systems comprising several processes in interaction requires more resources.
C. History based impact analysis approaches
With the availability of the historical data which allow to store and exploit the versions management systems as well as the bugs tracking systems, a new prospect of change
impact analysis was born, namely: History based Impact Analysis approaches. In fact, these approaches come to supplement those previously mentioned in bringing extra information’s relating to the types of the frequent changes, the causes of these changes, elements changing simultaneously… etc.
1) Association rules based approaches The approach suggested by Ying & Al [6], consists in
representing the simultaneous changes (Co-changes) which were met in the past by a set of association rules which will be used thereafter to determine the impact of a future change. That means that the change of an artifact A will be probably followed by a change of an artifact B if these artefacts changed jointly in the past.
The advantage of this approach is its aptitude to capture the coupling relationship of change which exists between the entities of the system without there being explicit dependency relationship between them. However, the coupling and causality relationships from which the repercussions appears later in time are difficult to capture.
2) Mining Software Repository approaches MSR for short treats system evolution and software
impact change prediction from a historical point of view. The historical data analysis allows, in fact improving comprehension of the software evolution and the prediction of future changes.
Contrary to the techniques of static impact analysis where only the current version of the software is taken into account, MSR adopts a multi versions view to software change prediction [12].
MSR techniques are similar to those used in knowledge extraction and data mining. By analyzing the historical data of several versions of the source code, MSR techniques allow [1]:
Detect the coupling and dependency relationships.
To define patterns of modification which can be taken into account to predict the future changes of the later versions.
To envisage trends of change.
However, the quality of the results obtained with this approach depends on the relevance of the historical data to take into account: neither too old thus become obsolete, nor too recent because it is likely to be based on unstable short period of the system [21].
D. IR-based impact analysis approaches
The software development is often accompanied by the production of various types of documents expressed in natural language like requirements specification, certain documents of design or simple user’s guide. The fact of identifying dependency relationships and establishing traceability links between these informal documents and the source code (text-to-code traceability links) can play a crucial role in the change impact analysis process as well as
Fig. 3. Text to code traceability impact analysis method [2].
in several other activities like program comprehension and software maintenance.
The impact analysis based on information retrieval techniques allow us to trace the text of a maintenance request onto the set of system components initially affected by the maintenance request [2].
The first stage consists in sorting the documents according to their relevance to the text of the maintenance request. In the second phase, all the existing documents are "mapped» towards the source code components to which they refer. Finally, this "mapping» serve to find the impacted elements of the code corresponding to the relevant documents identified in the first stage.
In order to help counter the precision or recall deficit of individual techniques and improve the overall accuracy, a novel approach which combine information retrieval with dynamic analysis and mining software repositories techniques (MSR) has been proposed in [17].
E. Probabilistic based impact analysis approaches
1) Bayesian networks-based approaches As there exists always a part of uncertainty difficult to
manage for the comprehension of the propagation process of the effects of a change, the Bayesian networks, by integrating uncertainty, then offer a particular quantitative approach which can provide suitable predictions in the presence of subjective judgments of experts or missing information that is inherent in the field of the software engineering [1].
The other advantage of Bayesian networks is their capacity of incremental learning from historical data. This capacity will contribute to the improvement, with time, of the structure of Bayesian networks and its parameters, and this, by the acquisition of new data.
Nevertheless, other probabilistic approaches have been developed such as CPM (Change Propagation Method) (for more details refer to [9]) and machine learning based approach for predicting maintainability.
V. DISCUSSIONS
The ability to identify the change impact or possible effect will greatly help a maintainer or manager to determine appropriate actions to undertake with respect to change decision, schedule plans, cost and resource estimations [20].
The amount of proposed approaches for change impact analysis is vast and a comprehensive paper is required to enable their evaluation and comparison. We provided through this article a repertory gathering the whole of the most relevant change impact analysis approaches while we have emphasized advantages, underlined the limits and exposed the techniques used.
The manual approaches are hard and error prone since it relies on the human factor while the automatable approaches are costly and more resources consuming. The tools used by the majority of impact analysis approaches are based on source code or need a substantial human interaction to achieve their task.
We project in our future works to extend the present work by considering new change impact analysis approaches, then comparing them empirically with standardized metrics and common benchmarks in order to provide some features and recommendations which could facilitate software maintainer’s tasks .
In addition, one of interesting application domain of the approaches described in this paper is the information system evolution according to business processes. Indeed, business processes being the way to synchronize business rules, they have to be managed when these rules change. The impact can be important on information system such as the modification of database structure or evolution of services and programs. Before applying changes, it is necessary for manager to have an idea of the cost of changing. This application domain will necessitate both the investigation of new techniques for analyzing the impact of rule’s change and revisiting existing ones.
REFERENCES
[1] Abdi, M.K., Lounis, H., Sahraoui, H. “Analyse et prédiction de
l’impact de changements dans un système à objets : Approche probabiliste”. In COMPSAC '09 proceedings of the 33rd IEEE International Computers, Software and Applications Conference, (pp. 234-239). Seattle, USA, 2009.
[2] Antoniol, G., Canfora, G., Casazza, G., & De Lucia, A. “Identifying the starting impact set of a maintenance request: A case study”. In Software Maintenance and Reengineering, 2000. Proceedings of the Fourth European (pp. 227-230). IEEE, 2000.
[3] Bohner, S.A. & Arnold, R.S., “Software Change Impact Analysis”, IEEE Computer Society Press, Los Alamitos, California, US, 1996.
[4] Bixin, L. & al. “A survey of code-based change impact analysis techniques”. Software Testing, Verification and Reliability. Vol 23, issue 8, (pp 613-646). December 2013.
[5] Breivoll, J & al. “Change Impact Analysis – A Case Study”. In 8th Conference on Systems Engineering Research, 2010.
[6] Ceccarelli, M., Cerulo, L., Canfora, G., & Di Penta, M. “An eclectic approach for change impact analysis”. In ICSE '10 proceedings of the 32nd ACM/IEEE International Conference on Software Engineering-Vol 2 (pp. 163-166).ACM, 2010.
[7] Chetna, G., Yogesh, S. and Durg, S.C. “Dependency based Process Model for Impact Analysis: A Requirement Engineering Perspective”. In International Journal of Computer Applications. Vol 6– No.6, September 2010.
[8] Chetna, G. & al. “An Efficient Dynamic Impact Analysis using Definition and Usage Information”. International Journal of Digital Content Technology and its Applications Vol. 3, No 4, December 2009.
[9] Clarkson, J., Simons, C. and Eckert, C. “Predicting change propagation in complex design”. In Proceedings of the 2001 Design Engineering Technical Conferences and Computers and Information in Engineering Conference. 2001.
[10] Hassan, O.A & al. “Approaches of Impact Analysis Assessment and Classification towards Projects Changes”. In World Journal of Social Sciences. Vol. 2. No. 7, (pp 167-173). November 2012 Issue.
[11] Jönsson, P., “Exploring Process Aspects of Change Impact Analysis” .Doctoral dissertation. 2007.
[12] Kagdi, H., & Maletic, J. I. “Software-Change Prediction: Estimated+Actual”. In Software Evolvability, 2006. SE'06. Second International IEEE Workshop on (pp. 38-43). IEEE, 2006.
[13] Kilpinen, M.S. and Clarkson, P.J., “Exploiting change impact analysis to support sustainability”. In: 7th Annual Conference on Systems Engineering Research, CSER' 09, Loughborough, UK, 2009.
[14] Kilpinen M.S., “The Emergence of Change at the Systems Engineering and Software Design Interface: An Investigation of Impact Analysis”, PhD Thesis, University of Cambridge, Cambridge, UK, 2008.
[15] Lehnert, S. & al. “A Taxonomy of Change Types and its Application in Software Evolution”. In ECBS '12 Proceedings of the 2012 IEEE 19th International Conference and Workshops on Engineering of Computer-Based Systems (pp 98-108), 2012.
[16] Lehnert, S., “A Taxonomy for Software Change Impact Analysis”. In IWPSE-EVOL '11 Proceedings of the 12th International Workshop on Principles of Software Evolution and the 7th annual ERCIM Workshop on Software Evolution (pp. 41-50), 2011.
[17] Malcom, G. “Integrated Impact Analysis for Managing Software Changes”. In Proceedings of the 2012 International Conference on Software Engineering (pp. 430-440 ). 2012.
[18] Michele, C. & al. “An Eclectic Approach for Change Impact Analysis”. In 32nd ACM/IEEE International Conference on Software Engineering. Vol. 2, (pp 163-166). 2010.
[19] Rutka & al. “Methods for engineering change propagation analysis”. In 25th international congress of the aeronotical sciences. 2006.
[20] Suhaimi, I. & al. “Integrating Software Traceability for Change ImpactAnalysis”. In The International Arab Journal of Information Technology, Vol. 2, No. 4. October 2005.
[21] Sunny, W., Yuanfang, C. and Michael, D. “Change Impact Analysis with Stochastic Dependencies”. In the 33rd International Conference on Software Engineering. 2011.
[22] Vasutiu, O., “Gestion des connaissances pour la maitrise de la relation entre patrimoine documentaire et système d’information” . Doctoral dissertation, 2009.
[23] Vasutiu, O., Jouve, D., Amghar, Y. “Gestion des changements et étude d'impact dans un système d'information réglementaire”. In INFORSID (pp. 1007-1022), 2006.
[24] Wilkerson, J.W. “A Software Change Impact Analysis Taxonomy”. In ICSM '12 proceedings of the 28th IEEE International Conference on Software Maintenance (pp. 625 - 628), 2012.
Using ontology as prior conceptual knowledge in an ILP system for fault diagnosis
Samiya Bouarroudj, Zizette Boufaida LIRE Laboratory
Constantine 2 University Constantine, Algeria
[email protected], [email protected]
Abstract—Fault detection and identification (FDI) has
received significant attention in literature. Intelligent systems that are built upon an integration of FDD methods can enhance plant performance and enable the operator to make more appropriate decisions. This paper presents the development process of a decision support system for monitoring and diagnosis of a gas plant. The system can monitor operations at the plant based on the input data, detect abnormalities and suggest diagnostic actions to the operator. For this purpose an ontology was developed and considered as a prior conceptual knowledge in Inductive Logic Programming (ILP) for inducing diagnosis rules. The learning examples are event sequences obtained by simulation of a model of an industrial steam boiler.
Keywords— inductive logic programming (ILP); SHIQ+log; hybrid reasoning; semantic web technologies; control system; knowledge management.
I. INTRODUCTION
Process supervision methods can be classified based on the type of information used for model development into data-driven, analytical, and knowledge based approaches [1][2]. Methods of faults detection and diagnosis mentioned above have their strengths and weaknesses. Thus the combination of complementary methods is an effective way to achieve high performance. However, knowledge of these control systems are not available in structured formats. For this reason, the new generation of decision support systems needs to tap into knowledge that is very broad combining learning, structured representations of domain knowledge such as ontologies and reasoning tools. It is in this context that joins our reflection.
The use of large steam boilers is quite common in industry due to their advantageous features [3][4]. However, such facilities are subject to several operating failures that could expose the system structural integrity to serious hazard and huge economic and human life losses. Early detection of such faults ensures safety of the systems.
Since supervision models are dependent on disparate information drawn from distributed sources, shared semantics based on a common ontology offers a way to develop these linkages. We address this critical need in this paper. Ontologies are a suitable formal representation able to convey this complex knowledge, but their use in learning algorithms is still a research issue.
Our objective is the construction of a steam boiler ontology by integrating and merging existing databases. Methods and tools have been proposed to generate ontologies from such structured input. For example, the DataMaster Protégé plugin [5] is a tool that allows importing schema definition and data into Protégé, but the target populated models are simply based on ontologies of the relational model. Ontologies with flat structure is the typical result of learning techniques that exclusively exploit information from the schema without considering the data. In this paper, we show how the RDBToOnto tool [6] can be used to derive accurate ontologies by taking advantage of both the database schema and the data, and more specifically through identification of taxonomies hidden in the data.
Inference rules may be crafted by the domain expert as part of the ontology design, or automatically learned by machine learning techniques. We focus on this latter case as a generic component to easily adapt them to new domains. However, as opposed to previous approaches, learning takes place in the ontology language to produce deductive diagnosis rules which is possible with inductive logic programming (ILP). We propose also a framework allowing the cohabitation of rules acquired by induction and the ontology as well as their exploitation for reasoning.
The paper is organized as follows: initially, a discussion of knowledge acquisition is presented. So, the proposed fault diagnosis system is developed. After discussing system structure, the main steps of the methodology designed are described in detail.
II. KNOWLEDGE ACQUISITION AND ONTOLOGICAL
ENGINEERING
Relational databases are valuable sources for ontology learning. In this paper, we describe an approach for ontology construction using heterogeneous databases. The main data and information constituting our system come from disparate data bases for equipment characteristics.
The database-to-ontology mapping approaches are usually classified into two main categories: (i) approaches which create an ontology from a database and, (ii) approaches which map a database to an existing ontology. In the former, the objective is the creation of ontology from a database. The mappings, in this case, are the correspondences between each created ontology
component (e.g., concept, property) and its original database schema concept (e.g., table, column). In the latter, the goal is to create a set of mappings between the existing ontology and the database. In our approach, we focus on the former, i.e.,we build an ontology from databases for equipment characteristics.
RDBToOnto [6] is a highly configurable tool that makes easier the design and implementation of methods for ontology learning from relational databases. It is also a user oriented tool that supports the complete transitioning process from access to the input databases to generation of populated ontologies. The settings of the learning parameters and control of the process are performed through a full-fledged dedicated interface.
One of the main motivations behind the RDBToOnto tool is to implement a process that allows to learn populated ontologies with rich taxonomies by exploiting both the schema and the data in the identification of the ontology structure.
A. Application problem domain
The industrial steam boiler, an ABB ALSTOM type[3][4], installed in the complex of natural gas liquefaction, generates a nominal steam capacity of 374 tons/h at superheated steam conditions of 73 bars and 487 ◦C. It is composed of three main parts: the main feedwater line, the steam generator and the main superheated steam line.
B. Ontology modeling
In the application domain, after all the classes, attributes, values, tasks, and strategies have been identified using the RDBToOnto tool, they are configured into a skeletal framework of the ontology which consisted of two models, the Domain Specific Model (An application ontology describes concepts dependent on a steam boiler) and the task model (Some samples of tasks include monitoring, diagnosis, selection or control). This part of taxonomy was extracted using the techniques described in sections 2.
Message sources water-level-measuring-unit
steam-level-measuring-unit pump-controller
Potentially failing hardware water-level-measuring-unit steam-level-measuring-unit pump control-unit pump-controller
Messages signal message message-stop message-steam-boiler-waiting
Message-open-pump
message-close-pump message-pump-failure-detection . . .
Actuators valve pump
Failures failure pump-failure pump-controller-failure water-level-measuring-unit-failure steam-level-measuring-unit-failure
Fig. 1. Part of the extracted taxonomy
This taxonomy was also enriched by associations. Every association involves two concepts and has the following form
• transmission-failure CAUSES emergency-stop-mode • pump-controller CONTROLS pump • message-pump-control-repaired-acknowledgement
IS-SENT-BY control-unit • . . .
III. INDUCTIVE LOGIC PROGRAMMING
Inductive Logic Programming (ILP) [7] was born at the intersection of Concept Learning and Logic Programming. The background knowledge in ILP is often not organized around a well-formed conceptual model. This practice seems to ignore the growing demand for an ontological foundation of knowledge in intelligent systems.
Induction in ILP generalizes valid hypotheses from individual instances/observations in presence of background knowledge. In Concept Learning, generalization is traditionally viewed as a search through a partially ordered space of inductive hypotheses [8]. According to this vision, an inductive hypothesis is a clausal theory and the induction of a single clause requires (1) structuring, (2) searching and (3) bounding the space of clauses.
IV. ONTOLOGIES AND RELATIONAL LEARNING IN ILP
An ontology formally represents knowledge as a set of concepts of a specific domain and the relationships between these concepts. The widely accepted definition of ontology is “a formal, explicit specification of a shared conceptualization” [9].
Hybrid KR systems combining DLs and (fragments of) HCL have very recently attracted some attention in the ILP community. Three ILP frameworks have been proposed which adopt a hybrid DL-HCL representation for both hypotheses and background knowledge:hooses Carin-ALN [10], resorts to AL-log [11], and builds upon SHIQ+log [12].
V. ARCHITECTURE OF OUR SYSTEM
Our system is developed to identify the causes and provides operation suggestions when abnormal situations occur. The structure of the proposed system is presented in Figure 2. It shows the main architecture of our system. This architecture consists of two distinct parts, one used offline which includes a module for generating examples and a module for learning discriminative patterns, and the other is used on line which includes a module for semantic reasoning.
Large and complex industrial processes such as chemical plants and petroleum refineries are typically equipped with distributed control systems (DCS) which allow users to vary the number of alarms for the purpose of better monitoring process-variables. A Honeywell Distributed Control System (DCS) operates in the Production Unit, which serves the data via the so-called Process History Database (PHD) module.
The data stored by DCS definitely have the potential to provide information for product and process design, monitoring
and control. They not only may be thousands of individual alarms, nuisance alarms could also distract the operator’s attention from more important problems.
A. Off-line analysis
The main objective of our work is the automatic learning of diagnosis rules. We are interested in cases of plant malfunctions. Our system allows extracting knowledge about the production facility during malfunction situations. Later, this knowledge will be used on line during a decision-aid step.
1) Data acquisition and Data Warehouse: The first phase in the knowledge discovery process is the generation of simulated data sets and data preprocessing. It is observed that the injection of disturbances will cause changes in plant measurements, while the measurements outputs of the model under normal operation conditions remain unchanged. Process DataWarehouse is a data analysis-decision support and information process unit, which operates separately from the databases of the DCS. It is an information environment in contrast to the data transfer-oriented environment, which contains trusted, processed and collected data for historic data analysis. The data collected into DW directly provide input for different data mining, statistical tools, like classification, clustering, association rules, etc.
During the data preprocessing phase it is extremely important to carefully investigate and prepare alarm data since real plant and simulated data tends to be inconsistent, with errors, missing values, outliners and duplicate values.
2) Generator of examples: The generation process is an ontology population task. It lies in the acquisition from simulator of new extensional knowledge, i.e. ontology’s
instances. Each example of the base of the sequences alarms follows the following syntax: Example (I, C, O)
The parameter I is the identifier of example. We number the examples by integers from 1 to 200. The parameter C is a list of class that the example belongs. The last parameter O corresponds to the object describing the sequence.
For instance a positive example is defined in the following way:
Example (1, default (V4-1, TRC 317), sequence ([increase (V4-1, pressure)], [closed (V4-1)], [increase (V4-1, flow)], [after (V4-1, P1)]))
3) Diagnosis rules learning by PLI: We consider the problem of learning rules from ontologies and relational data. We assume that the predicate in the rule head represents a concept to be characterized (characteristic induction).
The data are represented as a SHIQ+log knowledge base B where the intensional part K (i.e., the TBox T plus the set R of rules) plays the role of background knowledge and the extensional part (i.e., the ABox A plus the set F of facts) contributes to the definition of observations.
Therefore ontologies may appear as input to the learning problem of interest. The observations are represented as a finite set of logical facts E. E could generally discomposed into the positive examples E+ and the negative ones E- .
The background knowledge is supposed to be insufficient to explain the positive observations but there is no contradiction with the negative ones. So an ILP machinery with input E and B, will output a program H. So H constitutes a kind of explanation of observations E+.
Fig. 2. Architecture of the proposed system
The language L of hypotheses must allow for the generation of SHIQ+log rules. More precisely, we consider defined clauses of the form:
p(X) ← r1(Y1), . . . , rm(Ym), s1(Z1), . . . , sk(Zk).
Where m ≥ 0, k ≥ 0, each p(X), rj(Y j ), sl(Zl) is an atom, and the literal p(X) in the head represents the target concept.
Figure 3 reports the main procedure of an algorithm analogously to FOIL [13] for learning onto-relational rules. The outer loop learns new rules one at a time, removing the positive examples covered by the latest rule before attempting to learn the next rule. The inner loop searches a second hypothesis space, consisting of conjunctions of literals, to find a conjunction that will form the body of the new rule.
Hset := Ø Pos:= E+ while Pos ≠ Ø do h := {p(X ) ←}; Negh:= E- while Negh ≠ Ø do add a new literal L to specialize h end while Hset := Hset∪{h}; Posh:= {e ∊ Pos ∣ B∪h e}; Pos:= Pos\Posh End while return Hset Fig. 3. General algorithm for learning onto-relational rules
If we run the system on the complete set of positive examples that describe the problem, the system induces for example the following definition for the predicate default/1: Default (A) � increase (A, pressure), after (A,P1), closed(A), which means that if the pressure increase in A and A is located after the pump P1 and A is closed then the component A causes a malfunction of the plant.
B. On-line analysis
The system is applied as a real-time computer aided decision support system, providing operation suggestions to help operators when abnormal situations occur. It consists of the monitor, knowledge base, inference machine, and integrated database.
1) Monitor: It monitors data streams obtained from the control system, e.g. temperature, pressure, and flow. If a situation is judged to be abnormal by the module, the data are automatically transferred to the inference machine to solve the problem. At the same time the data are stored in the integrated database.
2) Inference machine: We suggest a combination method of reasoning to improve the search of results. The method is based on the principle of hybrid combination of two reasoners in which each treats a distinct party of knowledge base: a logical description raisoner for the structural part (OWL-DL) and a rule engine for the deductive part (RULES).
VI. CONCLUSION
In this paper, a real-time system is proposed for monitoring and diagnosing of chemical processes. The representation of knowledge base, inference machine and the relations among them are considered in this paper according to the characteristics of chemical processes.
This system helps the operators a lot to eliminate potential and disasters faults. The system also decreases the loss brought by unstable process situations and the loss if the time used for eliminating faults is too long. When new fault occurs, the stored data helps the domain expert to analyze the reason of the fault, and give earlier prediction of the trend. Our system design approach can be exploited to develop and rapidly prototype real time distributed multi-agent systems.
REFERENCES [1] R. Srinivasan, “Artificial intelligence methodologies for agile refining:
an overview.” Knowledge and Information Systems 12 (2), 129–145. 2007
[2] H.L Chiang, L.E. Russell and D.R. Braatz, “Fault Detection and Diagnosis in Industrial Systems”. Springer, London, Great Britain. 2001
[3] Manuel technique d’exploitation de la Chaudière ABB ALSTOM (2000).
[4] D. R. Tucakovic, V. D. Stevanovic and T. Zivanovic, “Thermal hydraulic analysis of a steam boiler with rifled evaporating tubes.” Applied Thermal Engineering, 27, 509–519. 2007
[5] C. O’CONNOR NYULAS, S. M. TU, ”DataMaster – a Plug-in for Importing Schemas and Data from Relational Databases into Protégé”. 2007 (http://protege.stanford.edu/conference/2007/presentations/10.01_Nyulas.pdf).
[6] F. CERBAH, “Learning Highly Structured Semantic Repositories from Relational Databases - RDBtoOnto Tool”, in Proceedings of the 5th European Semantic Web Conference (ESWC 2008), Tenerife, Spain, June, 2008.
[7] S.Muggleton, “Inductive Logic Programming: Issues, Results and the Challenge of Learning Language in Logic”, Artificial Intelligence, 114(1-2), pp.283--296,1999.
[8] T. Mitchell, “Generalization as search. Artificial Intelligence” 18, 203–226, 1982.
[9] A. Borgida, “On the relative expressiveness of description logics and predicate logics.”, Artificial Intelligence 82(1–2), 353–367, 1996.
[10] C. Rouveirol, V. Ventos, “Towards Learning in CARIN-ALN.” In: Cussens, J., Frisch, A.M. (eds.) ILP 2000. LNCS (LNAI), vol. 1866, pp. 191–208. Springer, Heidelberg , 2000.
[11] F. Lisi, “Building Rules on Top of Ontologies for the Semantic Web with Inductive Logic Programming.”, Theory and Practice of Logic Programming 8(03), 271–300, 2008.
[12] F. Lisi, F. Esposito: “Building Rules on top of Ontologies? Inductive Logic Programming can help!” SWAP 2007
[13] J.R. Quinlan: “Learning logical definitions from relations.” Machine Learning, 5:239–266. 1990.
New Approach for the Construction of User Profile:E-recruitment
Taous Iggui(a), Hassina Nacer(b), Youcef Sklab(c), and Taklit Ait Radi(d)Department of Computer Science, University of Bejaia, Algeria
Email(a): [email protected], Email(b): sino [email protected](c): [email protected], Email(d): [email protected]
Abstract—Within the framework of the works of SemanticWeb, this one comes to exploit the information’s extractiontechnics to extract and build a user profile. We were interestedin the texts of job applications to determine the conveyed userprofile. In order to achieve this, we took advantage of the formalpower of the statistic, ontologies as well as techniques of NLP(Natural Language Processing).
Index Terms—Information Extraction, User Profile , NaturalLanguage Processing, Ontologies, Statistic, Semantic Web.
I. INTRODUCTION
Information systems are, increasingly, accessible via In-ternet or Intranet, they allow users to reach an enormous massof information from several sources. This fast increase engen-dered the problem of how to find information we are lookingfor in this big mass of data [9].The most important challengeof the current researches, tends to optimize the performanceand the accuracy of the results returned by applications andcomputer systems, in order to better satisfy the needs of users.This objective is according to several constraints, in which theuser profile is a decisive one.
The user profile is a crucial concept for various do-mains, such : e-recruitment, e-commerce, criminal domain,information retrieval, education and security [2], [17]. For this,we referred to a research topic which is at the intersectionof several disciplines : Artificial Intelligence (AI), LanguageProcessing, Text-mining and Semantic Web, with the aim ofextracting the relevant information concerning a user profile.
In this context, we propose an information extractionsystem, from electronic e-mails of job applications, toconstruct the user profile, by combining between severalstatistical metrics, inspired by [6], [4], semantic Webtechnologies, as well as the technics and methods of NLP(Natural Language Processing), such kind in the fact as oursystem can take into account all the relevant informationwhich can represent different preferences and user activities.
The remainder of the document is organized as follows:
We Synthesize in Section II the most important works,as well as the existing standard solutions in the field ofinformation retrieval, and others about building a user profile.In Section III the proposed system architecture, entitledBPS (Builder Profile System), an automatic system for the
construction of a user profile, and we detail the variouscomponent modules. Section IV involve the evaluation of theproposed approach, by presenting results and performancemeasures of this one. The conclusion and future extensionworks are described in Section V.
II. RELATED WORKS
With the aim of exploring and exploiting the variousformats of data, there was a birth of a whole discipline called” Information Extraction ”(IE), using a set of tools, methodsand technics, to extract the relevant information that can beconveyed by an information source. Several research studieshave emerged, according to the application field, we present :
A. The Field of Automatic Summarization
An automatic summarization system consists in returninga condensed representation of the original text while keepingits semantic. In other words, it means extracting the relevantinformation. Several summarization technics have beendeveloped allowing the construction of several summarizationsystems, such [5], [4], [11], [3], YACHS2, CORTEX andROUGE2.
They use a set of graph theory technics and statisticalmetrics (the sum of the frequencies of words in a sentence,the sum of the weights of words in a sentence, the positionof a sentence in the document, the similarity with the title,the interaction between sentences etc.) which, once assembled,allow us to assign a relevance score to each sentence of thedocument, and in this case, sentences having the highest scoresare selected for the summary (the number of sentences dependson the size of the required summary).
B. The Field of Information Retrieval
The problem that arises in the field of informationretrieval, is the satisfaction of the user’s needs related to itsdifferent interests and preferences. It is where the user profileintervenes for a personalized information search as [18], wherethe user profile is used in the results ranking step. In this area,the user profile is a represented by a set of key words, extractedfrom the consulted documents by using the tf.idf measure [12],ontologies [13], and other knowledge bases like Wikipedea in[14].
C. The Field of Information Extraction (IE)
Actually the intersection of methods and technics, offeredby IE discipline and the notion of the user profile, gives riseto the topic of building this user profile. Technics of IE allowsus to construct the user profile with different ways, from a setof relevant key words extracted from applications titles [1],using a set of named entities belonging to a specific domain[8], from its comments in Web forums [15] by using a set ofclustering technics (K-means, EM), methods of stylometry aswell as ontologies of speech.
D. Reviews and Discussion
During the study and the analysis of the previous works,we have been able to identify the characteristics of eachmethod, we notice that the extraction of relevant informationeither the construction of a user profile bases itself on a setof statistical metrics, which provide the highly-rated formal ofthe obtained results, but we also underline the independencebetween these results and treated texts, in other words ” theabsence of semantic constraint ”. In fact, the statistical metricsreturn approximate results, but require a semantic validation,for example, a too frequent term may not be relevant. There isalso, works treating the constraint of the user profile evolution,based on the history of his interactions with the system whichstill do not reflect the interest of a user. A user profile is moreimportant than a set of key words and history interactions, it’sa set of semantic dimensions.
III. PROPOSED APPROACH: THE CONSTRUCTION OF AUSER PROFILE : BPS (Builder Profile system)
Having for purpose, saving time, improving the qualityof recruitment service, with the great impact of the userprofile in the area of e-recruitment, we propose an automaticsystem for the construction of this user profile from texts ofemails job application. The Fig.1 shows the proposed systemarchitecture :
Fig. 1. The proposed system architecture.
The user profile will be represented according to the userprofile model of [16], illustrated in Fig.2 :
Fig. 2. The user profile model.
A. Pre-processing
The treatment involves set of operations such the : segmen-tation of the body of the email into sentences, normalizationand terms lemmatization.
B. Hybrid Summary
At the level of this module, we propose a summarizationmethod method A, which is a combination of a set of statisticalmetrics, having already proved their worth in the automaticsummarization field, with an ontology. This combination en-sures us a summary oriented by the ontology’s domain. Themethod A will be coupled with method B of [3]. The resultsof the method A will be fused to the results of method [3]and the union of the two sets will be the output of the moduleof hybrid summary.
C. Named Entities Determination
Once, the most relevant sentences are selected, we drawon [7], to determine the set of Named Entities (NE) thatincludes the email. The goal of the determination of namedentities is to locate and classify text elements into prede-fined categories corresponding to the attributes of the userprofile. For this, we use a set of Named Entities DetectionRules(NEDR).
D. Matching with User’s Profile Information
This module consists in associating each named entitydetected to the corresponding user profile attribute. It’s donein three phases :
1) Determination of Information’s Units : Inspired by[10], we assume a processing algorithm of dependenciesbetween words, witch are represented by predicates as PRED-ICAT(Mot1, Mot2) having true value, if the relation betweenthe two words exists, else false. This, in order to have a setof basic units, that can provides information about the userprofile, called ”Information Unit” (IU).
2) Selection of Candidate Attributes: The aim of this stepis to select the most appropriate attribute to receive theinformation conveyed by the IU.
3) Validation: The attribute value conveyed by a IU mayhave multiple candidate attributes, from the user porfile model,hence, the need for the validation phase. To do this, we proposea set of generic rules called ”Contextual Exploration Rules”.
E. Updating of the User Profile
Considering the evolution characteristic versus time, thatcharacterizes the user profile, we suggest the updating module.We define :
• Static attributes: This kind of information doesn’t change,as personal data (name, birthday, nationality).
• Dynamic attributes: In this case, we associate to eachattribute the most current value, in other words, the ith attribute of the J th execution will receive the valuereturned by the (J+1)th execution of the system, such as: age, family situation.
• Scalable Attributes: In this case, the attribute values areobtained by the union of the values returned by eachpartial profile, such as : experience, function, diploma.
IV. SYSTEM’S EVALUATION
In order to evaluate our system, we propose some exe-cutions, taking into account a set of hypotheses:
• The emails are written in English.• The email’s body is not an attachment.• The information shall be expressed by a single sentence,
ie, the information is not dispatched on several sentences.• We suppose that the sentences are not in the negative
form.We will also use an ontology in the computing recruitmentfield.
A. Scenario N01: Running the Approach Without the Moduleof ”Hybrid Summary”
We start the execution with the second module of oursystem architecture (Named Entities Determination module ).Fig.3 show the result of an email example, after applying aset of Rules of Determination of Named Entities (RDNE) :
Fig. 3. The detection of named entities results.
After the application of the algorithm of IU acquisition,while based on Stanford 1 dependencies between terms, theselection of the attribute candidate and the validation of theresults, with a set of contextual exploration rules, the result ofthe system is shown in Fig.4 as a partial profile :
1http://nlp.stanford.edu/software/stanford-dependencies.shtml
Fig. 4. The UML diagram of the user profile.
We note from our executions that the accuracy of thesystem depends, in large part, on the step of detecting UIand on the similarity measure, indeed, we have distinguishedsome special cases, some of which are outlook, and othersanomalies to which we proposed solutions, as the following:
• The attributes StartDate and EndDate, being of the samekind (Date) and having the same verbs triggers and thesame semantic indicators, it is possible that the measureof similarity is insignificant.Proposed solution : we propose as solution, the compar-ison of the two dates.
• ” I am Bill I am Algeria I am a Master degree ” .We notice in these examples, the semantic ambiguitiesbetween the information of the same type, as the attributeName, which undergoes an error attribution.Proposed solution : We propose the use of a thesaurus,which will be in the form of lists of attribute values (de-gree, domain, residence, company, nationality, jobTitle).
B. Scenario N 02: Execution of the Approach with the Moduleof the ” Hybrid Summary ”
The email of Fig.3, has 8 sentences, numbered accordingto their appearance rank. The returned results, by the hybridsummary module, are as follows :
• Summary A = {2, 3, 4, 5, 6, 7}.• Summary B = {1, 5, 4, 6, 2, 7}.• Summary hybrid ={1, 2, 3, 4, 5, 6, 7}.
We notice that the summary method, we have proposed,by using ontologies of domain, gives us, indeed, a better result,it eliminated the sentences 1 and 8, which are sentences ofgreeting. While the summary B, by absence of semantics,eliminated the sentence 3, in spite of its relevance, and keptthe sentence 1.
C. Metrics of Evaluation of the Proposed System
The implementation of the approach is complex andtakes time, for this, we improvised a manual execution ona base of five e-mails consisted of complex sentences, soallowing to validate the expected results. The results of oursystem on this set of emails, were quite satisfactory and
convincing, we take for reference, the obtained values of theevaluation’s metrics. Among the metrics which we were ableto define, we note :
• Precision = Number of attributes well detected / Numberof attribute detected.
• Error rate = 1- Precision.• Efficiency = Number of attributes well detected (in an
email) / Number of attributes conveyed by the email.• The relation between the number of IU and the accuracy
of the system.TableI presents the precision values, obtained during the ex-
ecutions of the approach, according to the number of sentencesand UI that contains each e-mail :
TABLE ITHE EVALUATION OF THE PROPOSED SYSTEM.
Number UI’s Precision Errorof sentences Number per E-mail rate
E-mail 1 8 14 100% 0%
E-mail 2 4 5 40% 60%
E-mail 3 4 6 50% 50%
E-mail 4 4 6 66% 34%
E-mail 5 3 7 71% 29%
The System accuracy 64.4%
V. CONCLUSION
We proposed in this work, an automatic system toconstruct a user profile from textual documents, called BPS(Builder Profile System), but also, a new automatic summarymethod directed domain, by favoring semantics and minimiz-ing the time of processing. We executed the approach proposedon a set of electronic emails, in our case, we took the exampleof the applications in the IT domain, the returned results wererather satisfactory, concerning the results of every module andthe approach generally. We have, indeed, a summary methoddirected domain, we have detected a convincing set of NamedEntities, so allowing the construction of a partial user profile(by the module of Matching), which will become global afterone or several updates. During, the execution of this approach,we were able to distinguish a few scenarios that we envisageas perspectives and among:
• The treatment of documents sent as an attachment.• The construction of a user profile from a multi-domain
email.• The integration of a lexical analyzer based on semantics.
REFERENCES
[1] J. Seguela and G. Saporta. E-recrutement : recherche de mots-cles perti-nents dans le titre des annonces d’emploi. 10eme Journees internationalesd’Analyse statistique des Donnees Textuelles, Paris, France, pp. 47, 2010.
[2] T. Kaczmarek and D. Zyskowski. Information extraction from web pagesfor the needs of expert finding. Studies in logic, grammar and rhetoric,vol. 22, pp. 141-157, 2010.
[3] J. M. Torres-Moreno and J. R. Rodrigez. Reg : Un algorithme glouttonapplique pour le resume automatique. In Prceedings of 10th InternationalConference on Statical Analysis of Textual Data, Montral Canada, 2010.
[4] A. Bossard. Contribution au resume automatique multi-documents. Thesede Doctorat en informatique presentee a l’Universite Paris 13 ,Paris-Nord,2010.
[5] F. Boudin. Exploration d’approches statistiques pour le resume automa-tique de texte. These de Doctorat en informatique presentee a l’Universited’Avignon et des Pays de Vaucluse, 2008.
[6] F. Boudin and J. M. Torres-Moreno. Resume automatique multi-documentet independance de la langue : une premiere evaluation en franais. Pro-ceeding de la 9ieme Confrence annuelle sur le Traitement Automatiquedes Langues (TALN’09), Senlis, France, Juin 2009.
[7] A. Bouhafs. Utilisation de la methode d’exploration contextuelle pourune extraction d’informations sur le web dediee a la veille,realisation dusysteme informatique java veille. These de Doctorat en Mathematiques,Informatique Appliquees aux Sciences Humaines presentee a l’UniversiteParis IV Sorbonne, 2005.
[8] N. Omrane, A. Nazarenko, and S. Szulman. Le poids des entites nommeesdans le filtrage des termes d’un domaine. In proceedings of 9emeinternationale de Terminologie et Intelligence Artificielle, Paris : France,Septembre 2011.
[9] F. Harrathi. Extraction de concepts et de relations entre concepts apartir des documents multilingues : Approche statistique et ontologique.These de Doctorat en InfoMaths : Distribution et Recherche d’InformationMultimedia, presentee a L’Institut Nationale des Sciences Appliquees deLyon, Septembre 2009.
[10] P. E. Genest and G. Lapalme. Framework for abstractive summarizationusing textto-text generation. Proceedings of the 49th Annual Meetingof the Association for Computational Linguistics, Portland, Oregon, vol.32(1), pp. 64-73, 24 June 2011.
[11] G. Erkan and D. RRadev. Graph-based centrality as salience in textsummarization. Journal of Artificial Intelligence Research (JAIR), pp. 39,2004.
[12] L. Tamine-Lechani, N. Zemirli, and W. Bahsoun. Approche statistiquepour la definition du profil d’un utilisateur de systeme de recherched’information. Publie dans ”Information - Interaction - Intelligence”,Cepadues Editions, vol. 7, pp. 5-25, 2007.
[13] M. Daoud and L. Tamine-Lechani. Construction des profils utilisateurs abase d’ontologie pour une recherche d’information personnalisee. Journalof Laboratoire IRIT Universite Paul Sabatier, 2008.
[14] M. Nicoletti, S. Schiaffino, and D. Godoy. Mining interests for userprofiling in electronic conversations. Journal of Isistan Research Institute,Conicet-Unicien, Campus Universitario, Paraje Arroyo Seco, 2012.
[15] S. El manar el bouanani and I. Kassou. Vers une methodologie demodelisation dune signature unique des profils web : Module de detectiondes auteurs des forums web. These de Doctorat presente lUniversiteMohammed V Souissi - Rabat, 2012.
[16] Y. Sklab, H. Nacer and A. K. Tari. Modeling User Context for PervasiveComputing. In Proceedings of the International Workshop on Require-ments Engineering, pp. 40-46, Constantine, Algeria, 2013.
[17] H. Jeong, E. Choi. User Authentication using Profiling in Mobile CloudComputing. AASRI Conference on Power and Energy Systems, AASRIProceding 2, pp. 262-267, Hong Kong, September 2012.
[18] A. Sieg, B. Mobasher, R. Burke. Learning Ontology-Based User Pro-files: A Semantic Approach to Personalized Web Search. IEEE IntelligentInformatics Bulletin, vol. 8, No. 1, pp. 7-18, Novembre 2007.
1
A coalition model for resource management in Green Cloud Computing
Nassima Bouchareb, Nacer Eddine Zarour LIRE Laboratory, Department of Software Technologies
and Information Systems, Faculty of New Technologies of Information and Communication, University Constantine 2,
Algeria [email protected]
Samir Aknine LIRIS Laboratory,
Department of Computer Science, University Lyon 1,
France [email protected]
Résumé— Cloud coalition is a recent paradigm that helps providers to overcome resource limitation by outsourcing requests to other coalition members. In this paper, we propose an agent-based mechanism to automatically manage Cloud resources to simultaneously achieve the maximum of accepted requests with minimum cost and energy consumption. We present new strategies that help in the decision-making process. These strategies increase provider's profit (selecting suitable resources to consume less energy and choosing coalitions to get more gain). Finally, we present a case study to illustrate our resource management mechanism in Cloud Computing. Also simulation results indicate that there are just a smaller number of rejected requests with the proposed mechanism.
Keywords- Cloud Computing; resource management; Green Computing; virtualization; multi-agent systems; coalition.
I. INTRODUCTION
Cloud provides a pool of resources which includes storage, computing platforms, data centers and software services "a dynamic allocation of computing resources (hardware, software, etc.) of tiers over a network" [1]. So, one of the essential features of the Cloud Computing (CC) is the availability and sharing of resources,. The resource management in CC encounters difficulties. The selection of suitable resources and provider cost are among these problems.
It has been argued that energy costs are among the most important factors impacting on provider total cost [2]. Some studies have shown that CC can use some technologies to minimize energy consumption. The key technology is "virtualization" [5]. When virtualization is used, it reduces the number of physical servers and increases the utilization rates [3], in order to minimize the energy consumption. In this paper, we use the coalition formation concept to increase the availability of services. As the CC is a distributed and a complex system, when it comes to design this type of system, the agent technology proves suitable, because agents don’t only allow the sharing or distribution of knowledge, but also the fulfillment of a common goal. The benefit of their use in resource allocation in CC, is in their features "autonomy, flexibility, etc" [10]. We use agents to represent different Clouds, and a multi-
agent system is needed to handle their interactions. The concerned agents select the appropriate resources, and especially cooperate and negotiate with other Clouds. Another benefit of multi-agent systems is their scalability. Since they are inherently modular, it is much easier to add agents to the system. For example, more agents that represent new coalition members. So in this article, we propose an agent-based mechanism that introduces the coalition concept in Green CC to handle the problem of resource management, when the provider disposes sufficient resources and also when it can't satisfy the new request (insufficient resources). The remainder of this paper is organized as follows. In the next section, we present the proposed architecture and mechanism. In section III, we give a case study to illustrate our solution. An evaluation of the mechanism is presented in section IV. Finally, we discuss the similar works in section V, before concluding in Section VI.
II. AN AGENT-BASED CLOUD ARCHITECTURE
The proposed Cloud architecture contains three cognitive agents. The architecture of each agent and the negotiation model between customers and agents were detailed in [19]. In this paper, we go beyond the negotiation stage. Provider has to satisfy the request even by forming or leaving coalitions. A. Cloud Agent (CA)
When it receives the request; it detects the quantity of VMs (Q) needed for the request, the duration of use (D), price (P) and the customer's country. If P >= min price, (Q) and (D) are sent to the AIA. Otherwise, the request is rejected. B. Allocator Agent (AlA)
When AlA receives Q and D, it selects available resources according to Green Computing "a fully charged resource consumes less energy than many resources with low load" [20]. Also, it migrate VMs from a physical resource to another, but it should know when migrating them, because the migration operation consumes energy [9]. We detail some
2 cases of resource allocation (see table I for denotations and table II for examples):
TABLE I. DENOTATIONS 1
1st case: There's at least one resource with free VMs >= X, choose the one with min number of free VMs >= X. 2nd case: No resource with free VMs >= X, share the request on resources with unoccupied VMs, starting with the one which has the max of free VMs. 3rd case: If the number of unoccupied VMs in resources is the same, select first the resource that will be released as soon as possible to not stay started up and consumes energy for a negligible number of VMs. 4th case: Migrate the occupied part to another machine to not share the request; it is preferred when the unoccupied part is much greater than the occupied part. 5th case: The free VMs sum < X; activate a new resource (Rx). If the selected resource is still insufficient, look for a resource with State =0, by always choosing the one with the minimum number of free VMs >= the rest of the request Y. 6th case: There is no unoccupied resource, or even with the activation of resources, the request can't be accomplished. The AIA sends a message to the CA to contact the CoA and obtain more VMs from a coalition.
TABLE II. EXAMPLES OF THE ABRM MECHANISM
Simulation of these cases were detailed in [8]. In this work we detail also the CoA role. C. Coalition Agent (CoA)
When AlA doesn’t find sufficient resources. The CoA decides to form a new coalition with other Clouds to use their resources, or leave an old coalition to release its own resources. The decision is made according to the gain of each
proposal, knowing that the gain is computed as follows (for denotation see table III:
Gain = Revenue – Cost Revenue = Rcust + Rcoal + Pcust + Pcoal
Cost = Ccust + Ccoal + Cout + Ncust + Ncoal So: Gain (G) = Rcust + Rcoal + Pcust+ Pcoal - Ccust - Ccoal - Cout -
Ncust - Ncoal (1)
TABLE III. DENOTATIONS
From equation (1) we can compute the gain in different cases. If the Cloud has received a coalition offer: Accepting this coalition involves more coalition revenues (R'
coal > Rcoal). - If the Cloud forms a new coalition with other Clouds to
use their resources and satisfy the received request, it must pay the solicited Cloud for their resources (VMs) which increase the outsourcing costs (C'out> Cout).
So (1) becomes (2): Gain (G) = Rcust + R'
coal + Pcust + Pcoal - Ccust - Ccoal - C'out
- Ncust - Ncoal (2) - If the Cloud leaves a coalition to free up its internal
VMs and accept the new request, it must pay penalties to Clouds of the canceled coalition. So coalition penalties are increased (N'
coal > Ncoal). Coalition costs are also increased because the Cloud is a member of the new coalition (C'
coal > Ccoal). So (1) becomes (3):
Gain (G) = Rcust + R'coal + Pcust + Pcoal - Ccust - C'
coal -Cout - Ncust - N'
coal (3) If the Cloud has received a customer request: Accepting this request involves more customer revenues (R'
cust> Rcust). We have exactly the same cases of a coalition offer:
- If the Cloud forms a new coalition with other Clouds to use their resources. The outsourcing costs are increased (C'out> Cout).
So (1) becomes (4): Gain (G) = R'
cust + Rcoal + Pcust + Pcoal - Ccust - Ccoal - C'out
- Ncust - Ncoal (4) - If the Cloud leaves a coalition to free up its internal
VMs. Ncoal and Ccust are increased. So (1) becomes (5):
1 Power consumption initially, cooling costs, hardware and software acquisition, staff salary, physical space, amortization of facilities, etc [12].
Symbol Signification X Number of VMs required by the request. nbr (Ri) Number of unoccupied VMs in the resource Ri. min (nbr) The minimum number of unoccupied VMs >= X State (R) (-1) unoccupied resource, (0) partially occupied resource, (1)
occupied resource. D (R) The use duration of occupied VMs (in months)
Case X Resources states Best allocation 1 X = 3 nbr (R1) =10, nbr (R2) =3,
nbr (R3) =7, nbr (R4) =2 Allocate R2
2 X = 9 nbr (R1) = 7, nbr (R2) = 2 Allocate R1 and R2 3 X = 3 nbr (R1) = 2, nbr (R2) = 2,
D (R1) = 2, D (R2) =1 Allocate R2 and R1
4 X = 9 nbr (R1) =8, nbr (R2) =2 (8>>>2)
Migrate the two VMs from R1 to R2 and allocate R1.
5 X = 12 nbr (R1) =10, nbr (R2) =2, nbr (R3) =3, nbr (R4) =4, State (R1) =1, State (R2, R3, R4) =0
Allocate R1 (State (R1) = 1), and R2 because Y = 2
6 X = 12 nbr (R1) = 0, nbr (R2) = 2, nbr (R3) = 3, nbr (R4) = 4
Look for 3 VMs.
Symbol Signification Revenue Request revenues Cost Request costs Rcust Customer's offered price Rcoal Coalition's offered price Pcust Customer's penalties Pcoal Coalition's penalties Ccust Customer satisfaction costs0F
1 Ccoal Coalition satisfaction costs1 Cout Outsourcing costs Ncust Penalties paid to consumers in cancellation case Ncoal Penalties paid for coalitions in cancellation case
3
Gain (G) = R'cust + Rcoal + Pcust + Pcoal - C'
cust - Ccoal - Cout - Ncust - N'
coal (5) In this paper, provider does not share the request between several Clouds to avoid inconvenient of data sharing [6]. So, even if the Cloud has some free resources, it can’t use them and form coalition just for the lacking number of resources.
III. CASE STUDY
We present in this paper the following case study: Our Cloud (CA1) has received a new request from a Spanish costumer. It requires eight VMs for two months, with 2100$. Supposing that CA1 has just two free VMs. To satisfy the new customer request, CA1 compares between (4) and (5) presented in section II. 3. (4): it contacts another Cloud (CA3) to form a new coalition
(CO2) to obtain eight VMs. (5): it leaves an old coalition (CO3) to release more VMs and
use them in the new request. R'
cust, C'cust, N'
coal, C'out are the only attributes that change in
these equations. R'cust is the same in (4) and (5), so we
compare (C'out) with (C'
cust + N'coal):
C'out = 1000$ (4), C'cust + N''coal = 600$ + 2000$ = 2600$ (5) We remark that cost in (4) is the smallest, which means the greatest gain. So the CA1 decides to form a coalition with CA3 to use its eight VMs and pay 1000$ for two months (table IV).
IV. EVALUATION To test the performance of the proposed ABRM mechanism, we have compared it with three policies proposed in [12], which differ in how providers handle requests when they cannot be served by available resources: • Non-Federated Totally In-house (NFTI): provider
terminates spot VMs2
• Federation-Aware Outsourcing Oriented (FAOO): provider outsources the request. If outsourcing is not possible, spot VMs are terminated as a last resort.
with lower bids. If it doesn’t release enough resources for the request, request is rejected.
• Federation-Aware Profit Oriented (FAPO): provider compares outsourcing with termination of spot VMs.
• The proposed mechanism (ABRM): provider compares the profit of leaving and forming a coalition before deciding.
We intend to study the behavior of providers in different situations. For this purpose, effects of two input parameters are investigated “system load and the number of providers”. It is an important parameter because it increases the chance of forming coalitions with lower costs to outsource requests. Impact of the load: Fig. 1 shows the impact of varying the number of requests on the proposed policies. Since these policies are triggered when the provider is fully utilized, load is the most influential parameter in our experiments. According to Fig. 1 “FAPO, FAOO and ABRM” which support outsourcing, have minimum number of rejected 2 Spot VMs are VMs that can be terminated whenever their current value for running exceeds the value that the client is willing to pay [4], [16].
requests by increasing load. They outperform the NFTI policy. The difference between FAPO, FAOO and ABRM is more observable at higher loads. The proposed strategy gives the best results. However, we remark that there is not a big difference between our strategy and FAPO; this is due to gaps of the suggested mechanism (paying penalties when leaving a coalition involves more costs than terminating spot VMs). In Fig. 1, FAOO and FAPO give the same results according to [12]. By decreasing the number of requests, the number of on-demand VM rejection also decreases, because providers are subject to a lower load. However, a higher number of rejected requests is seen when NFTI is applied. Impact of number of providers in the coalition: This experiment evaluates the impact of number of participants in the coalition on the results delivered by each policy. The experiment was repeated with 3, 5, and 7 providers. The results are presented in Fig. 2. By increasing the number of providers, policies with an outsourcing option have a smaller number of rejected on demand VMs, because it is more likely that a provider can found who could serve the outsourced request. Increasing the number of providers does not have any impact in NFTI, as in this policy there is no interaction with federation members. At seven providers level, strategies with outsourcing option have no rejected request. FAPO and FAOO also give the same results according to [12], but the suggested ABRM mechanism represents the best results because when there are more providers, there are more coalitions to form and therefore more coalitions to leave.
Fig. 1. Impact of load on number of rejected VMs for a provider with different policies
Fig. 2. Impact of number of providers on number of rejected VMs for a provider with different policies
V. RELATED WORK
4 Very few studies have treated the resource management in coalition CC. In [12], authors just compare between liberating resources occupied by customers and outsourcing the new request. However leaving a coalition may be the best solution. This case is treated in the proposed ABRM mechanism (The difference between these two works was detailed in section IV). In [13], authors allow the provider to contact only its neighbours Clouds. While a non neighbour Cloud can be more beneficial than another neighbour Cloud. So in this mechanism we just give priority to neighbors Clouds when calculating their trusts. Also in [17], authors study the cooperative behavior of multiple Cloud providers to obtain stable coalition structures, and the paper presented in [18] extends the work in [17], where the cooperative game is also used to analyze the resource and revenue sharing in CC but without taking into account the GC. Mashayekhy and Grosu propose in [14] a mechanism that enables Cloud providers to dynamically form a Cloud federation. But, the Cloud providers do not have incentives to break away from the current federation and join some other federation. Among works that treat GC in CC: [20] where authors propose efficient green enhancements in CC, using power-aware scheduling techniques, and live migration of VMs. Our contribution is based on their results, but also takes into consideration what has been proven in [7, 9]. In [7]; authors have evaluated the cost of migration of VMs. In [9]; authors experimentally show that migration has an energy overhead. For energy conservation in federated CC, there is a distributed coalition formation algorithm proposed in [15]. Each provider decides whether to leave the current coalition to join a different one according to his preference. Unlike in [14], that rely on a centralized architecture in which a trusted third party computes the federation set, in [15] and in our work, a distributed approach is adopted in which each Cloud provider autonomously and selfishly makes its own decisions, and the best solution emerges from these decisions without the need of synchronizing them, or to resort to a trusted third party. According to these insufficiencies, we have proposed an agent-based architecture that introduces coalitions to handle the problem of resource management in Green CC.
VI. CONCLUSION The aim of this paper is to propose a new mechanism for Cloud resource management, in order to simultaneously maximize the number of accepted requests and reduce the provider’s costs and energy consumption. We have proposed an agent-based architecture and new strategies to help in the decision-making process for increasing provider's profit. Finally, we have given an example and we have compared it with similar work. This work opens many perspectives that seem interesting “. We hope first finishing the evaluation part. Testing the scalability of the mechanism, with more requests, more allocator agents and Clouds. Then for future research, we have: definition of the gain strategy when provider has some free VMs and he can use them, we will also give an offer strategy to compute Clouds trusts.
REFERENCES
[1] P. Brébion, “Computing in the Cloud, decryption,”archIMAG website, a monthly magazine published by SERDA group, June 1, 2010.
http://www.archimag.com/article/l%E2%80%99informatique-dans-le-nuage-d%C3%A9cryptage
[2] M. Guazzone, C. Anglano and M. Canonico, “Energy-efficient resource management for Cloud Computing infrastructures,” in the 3rd IEEE International Conference on Cloud Computing Technology and Science, pp. 424- 431, 2011.
[3] H. Li , J. Jeng, “CCMarketplace: A marketplace model for a hybrid Cloud,” in the 2010 Conference of the Center for Advanced Studies on Collaborative Research, pp. 174-183, 2010.
[4] Amazon Elastic Compute Cloud (Amazon EC2) website http://aws.amazon.com/ec2
[5] S., Kumar Garg, R., Buyya, “Green Cloud Computing and environmental sustainability,” Harnessing GIT: Principles and Practices.Wiley, 2012.
[6] D. Yuan, Y. Yang, X. Liu and J. Chen, “A data placement strategy in scientific Cloud workflows,” in Future Generation Computer Systems, 2010.
[7] W., Voorsluys, J., Broberg, S., Venugopal, R., Buyya, “Cost of virtual machine live migration in Clouds: A Performance Evaluation,” in the 1st International Conference on Cloud Computing, pp. 254 –265, 2009.
[8] N. Bouchareb, N. E. Zarour and S. Aknine, “An agent-based architecture for resource allocation in Cloud Computing”, in the 2nd European Conference on Service-Oriented and Cloud Computing, pp. 54-59, 2013.
[9] A. Strunk and W. Dargie, “Does live migration of VMs cost energy?,” in the 27th IEEE International Conference on Advanced Information Networking and Application, pp. 514-521, 2013.
[10] S. Touaf, “A logic diagnostic of complexe and dynamic systems in a multi-agent context,” Doctoral thesis in Computer Science, Joseph Fourier University, Grenoble 1, France, 02 Mar, 2005.
[11] Cisco, AMC and VMware, “Virtual computing environment coalition services at-a-glance,” 2009.
[12] A. Toosi, R. Calheiros, R. K. Thulasiram, and R. Buyya, “Resource provisioning policies to increase IaaS provider’s profit in a federated Cloud environment,” in proceddings the IEEE International Conference on High Performance Computing and Communications, pp. 279- 287, 2011.
[13] D. Ye, M. Zhang and D. Sutanto, “Integrating self-organisation into dynamic coalition formation,” in the International Foundation for Autonomous Agents and Multiagent Systems, pp. 1253- 1254, 2012.
[14] L., Mashayekhy, and D., Grosu “A coalitional game-based mechanism for forming Cloud Federations”, in the 5th International Conference on Utility and Cloud Computing, pp. 223–227, 2012.
[15] M., Guazzone, C., Anglano, and M., Sereno, “A Game-Theoretic Approach to Distributed Coalition Formation in Energy-Aware Cloud Federations (Extended Version)”, Cornell University Library. http://arxiv.org/abs/1309.2444v5 (Accessed 9 April 2014).
[16] J. Varia, “Best practices in architecting Cloud applications in the AWS Cloud,” in Cloud Computing: Principles and Paradigms, R. Buyya, J. Broberg, and A. Goscinski, Eds. Wiley Press, ch.18, pp.459–490, 2011.
[17] D. Niyato, A.V. Vasilakos, and Z. Kun, “Resource and revenue sharing with coalition formation of Cloud providers: Game theoretic approach,” in the 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, pp. 215- 224, 2011.
[18] D. Niyato, Z. Kun, and P. Wang, “Cooperative virtual machine management for multi-organization Cloud Computing environment,” in the 5th International ICST Conference on Performance Evaluation Methodologies and Tools ,pp. 528-537, 2011.
[19] N. Bouchareb, N.E. Zarour and S., Aknine, “An agent-based model for negotiation and resource allocation in Cloud Computing,” in the International Conference on New Technologies and Communication, pp. 42-48, 2012.
[20] J. Younge, G. Laszewski, Lizhe Wang, S. Lopez-Alarcon and Warren Carithers, “Efficient resource management for Cloud Computing environments,” in the IEEE International Conference on Green Computing, 2010.
Using Mobile Agent Approach for Balancing the Load of a Parallel Simulation
Application into Cloud Computing
Selma Chenni
Computer Science Department
Mohammed El-bachir El-Ibrahimi
University Bordj Bou Arreridj,
Algeria
Abdelhak Boubetra
Computer Science Department
Mohammed El-bachir El-Ibrahimi
University Bordj Bou Arreridj,
Algeria
Saad Harous
College of Information Technology
United Arab Emirates University
P.O Box 15551Al Ain, UAE
Abstract— Storage capacity and computing power are very
important requirements to model and simulate complex
systems because simulation process is time consuming, uses
and generates a large amount of data which is suitable for the
use of cloud computing technology. Cloud computing might
provide new opportunities in the area of modeling and
simulation especially in the case of parallel and distributed
simulations “PADS” because it owns a fully parallel and
distributed architecture that allows a typical execution of such
kind of computer simulation. One of the challenges faced by
PADS into cloud computing is the dynamic load balancing [1]
because of the unpredictable behavior of both: simulation
model of a complex system and the execution environment of
cloud computing so rearranging the load dynamically fits
better than the static one. In this paper we will explore
important impact of cloud computing on computer simulation
by reviewing some existing research works and introduce an
idea on how to make dynamic load balancing of a parallel and
distributed simulation possible by using mobile agent
paradigm.
Keywords- Cloud Computing; simulation; SaaS (Simulation
as a Service); load balancing; mobile agent.
I. INTRODUCTION
As defined by the National Institute of Standards and
Technologies (NIST) [2], cloud computing is:
“a model for enabling ubiquitous, convenient, on-demand
network access to a shared pool of configurable computing
resources (e,g networks, servers, storage, applications, and
services) that can be rapidly provisioned and released with
minimal management effort or service provider
interaction.”
Most of us are usually using cloud computing services in
our daily activities without knowing. A good and well
known example is the set of services provided by Google
that are part of Google’s Cloud (e.g. Gmail, Google Earth,
Google Docs…). Other examples are social networks like
Facebook and web based email service. For this purpose
cloud computing is a “general term for anything that
involves delivering hosted services over the Internet”. Those
services belong to a cloud computing service model called
SPI model. It contains the following three layers:
Software as a Service
Platform as a Service
Infrastructure as a Service
Similar to grid computing and clusters technologies,
cloud computing provides an important storage capacity and
computing resources virtually and on demand [3]. It costs
less for the users to use the cloud. It also frees them from
complex management and maintenance of hardware and
software.
In this paper we discuss one of the services that cloud
computing could provide to users which is Simulation.
Simulations in general use and generate large amount of
data and need computing power. Both requirements are
provided by the cloud computing. So it is a suitable
platform for simulation to run and store the huge amount of
data generated. Even simulators might be hosted on the
cloud computing which allows users to access simulation
services much easier. As mentioned before, parallel
simulation could be run efficiently over cloud computing
and to obtain simulation results faster, a load balancing
strategy has to be implemented.
This work focuses on simulation process and cloud
computing technology. It does not only address their
relationship and how can simulation make use of cloud
computing it also presents how to improve the execution
service of a parallel simulation by using a load balancing
technique.
II. CLOUD BASED SIMULATION
Simulation over the cloud is also called cloud based
simulation. As we mentioned cloud computing provides
software / applications as a service accessible via Internet,
running such applications on the cloud helps to release a
large part of user’s local resources. An example of those
applications may be a simulator hosted into the cloud. This
simulator benefits from cloud computing resources in order
to run simulations that require a higher computing power
than available locally at the user’s site.
Cloud based simulation is described as follow in [4]
“Simulation models become services which can be accessed
from many different locations by different stakeholders”.
Another form of cloud based simulation is mentioned in [5]
when a user wants to perform simulations remotely:
1. He/she sends a request to the service provider,
2. A virtual machine will be assigned
3. Simulation starts running on it,
4. Output data could be stored in the cloud ,
5. Results are sent to the user.
Thus the service provided is Simulation as a Service
accessible by everyone through the Internet.
Figure 1: Running simulation on the cloud
But before discussing the reason behind the adoption of
cloud computing, we have to distinguish between
Simulation in the cloud and simulating a cloud which are
two distinct notions.
Cloud based simulation refers to the fact when simulation
services are available in the cloud computing. The users can
access them from any location. On the other hand,
simulating a cloud means the use of a simulator in order to
simulate cloud computing infrastructure and services, an
example of such simulator is CloudSim.
Simulation in the cloud can take advantage of both
infrastructure and software services supplied by its provider
as follows:
Cloud computing is an excellent environment for
hosting simulation systems that require a high
performance computing power.
There is no need to purchase or maintain hardware
and software, which may save a lot of resources to
support simulation.
Cloud computing provides a fully parallel and
distributed architecture that allows a typical
execution of parallel simulations. For example cloud
computing is a suitable environment for running
parallel and discrete event simulation.
When the number of stakeholders using the same
simulation service increases, new servers could be
assigned with the aim to enhance computing and
storage capacity. This leads to a multitenant
architecture which allows a simultaneous access to
the same service by different users.
No local installation is required; the simulation
service is hosted over the cloud and is accessed
remotely from any location.
When a group of users working on the same
simulation project are geographically dispersed.
They can collaboratively work together by accessing
the service simultaneously in order to develop their system
or make modifications via the Internet.
After discussing some of the benefits that result from
adopting Cloud Computing technology, the next section
examines existing works in the area of cloud computing and
simulation.
III. RELATED WORK
A variety of studies have been conducted in computer
simulation and cloud computing field. Most of them are
trying to transit traditional parallel simulation to cloud
computing [6] because of cloud’s parallel architecture as
discussed previously. In the literature there are an increasing
number of works related to the association of cloud and
simulation issues due to the novelty use of cloud computing
in modeling and simulation areas. A brief review of some
works is summarized bellow.
Malik et al. [7] proposed a mechanism allowing the
synchronization of parallel and discrete event simulation in
cloud computing environment. It helps to reduce
considerably execution time of parallel simulation.
In the same context of parallel simulations, Fujimoto et
al. [8] listed the advantages, drawbacks and suggests
solutions related to the adoption of cloud computing to run
parallel and distributed simulations.
Shao and McGraw’s [9] discussed a design of a service
oriented simulation software framework used for military
services where simulation allows commanders to make
improved battlefield decisions.
In [10], Seo et al. proposed a cloud computing
environment for modelling and simulating discrete event
systems where a service oriented architecture (SOA) is used
for implementing a cloud based simulation. The reason
behind the use of SOA and cloud technology is to reduce
development costs.
A comparison between cloud computing and grid
computing from networking and also simulation point of
view is given in [11].
Tsai et al. [12] highlighted the benefits of running
simulation over cloud. The SimSaas framework was
proposed whose main feature was to support a multi-
tenancy architecture.
Yamazi et al. [5] described a successful Simulation
Platform hosted on the cloud. This platform includes
various development tools and preinstalled simulators such
as: GENESIS, NEURON, NEST, GNUPLOT…etc. This
simulation project is accessible at: http://sim.neuroinf.jp/.
D’Angelo [13] reviewed important challenges that
could be faced by parallel and distributed simulation with
the new improvement in parallel architectures like cloud
computing and many cores architecture.
A concurrent simulation platform, mJADES, was
implemented by Rak et al [14]. It is a cloud application that
allows users to run concurrent simulation. The authors
argued how this environment could help in the performance
prediction of parallel programs in order to run the same
simulation software at the same time for different
parameters values.
Liu et al. [6] introduced their attempt to transit existing
simulation software into the cloud by presenting the design
and implementation of the CSIM simulation software into
cloud computing. Also a recent work is proposed by the
same authors in order to run parallel simulations with CSIM
[15].
Rescale is a cloud simulation platform that helps expert
and non-expert researchers [16], engineers and scientists to
build, compute and analyse simulations into cloud
computing by bringing them an important and on demand
computing power (see Figure 2).
Figure 2: Rescale’s Pricing Strategy
Another real existing simulation project integrated in
cloud is the Scenario Navigator Web (SNWeb) [4]. It is a
platform that allows simulation to be run in the cloud
instead of running simulation models locally. To use
SNWeb, the user needs only access to a web browser.
Scenario Navigator supports a set of simulation engine
(Arena, Simul8, Enterprise Dynamics and ExtendSim) and
different prototype models such as: gas station, call centre,
transporter, assembly line and emergency department that
are available for a direct use without implementing them
again.
In the reminder of this section, a comparison between
some projects of simulation in the cloud will be represented
in table 1. The comparison is based on some criteria:
supported simulation type (parallel simulation, concurrent
simulation, and sequential one), simulation process service
(modeling service, simulation run service, analyzing and
visualization of results service), availability and
accessibility (accessible from everywhere, free or payable)
and objective of the project.
Simulation
project
Simulation
kind
Simulation
process
service
Availability
and
accessibility
objective
SNWeb [3] All
simulations
are
supported
-Modeling
-Execution
-Analysis
visualization
- payable
-online
access
-Running
simulation
models
everywhere
and
anytime
Simulation Platform
[4]
All simulations
are
supported
-Modeling -Execution
-Analysis
visualization
-making online tests
for free
-Running neural
simulation
models
CSim [5] Parallel
discrete
event
simulation
-Modeling
-Execution
-Analysis
visualization
-Laboratory
work
-no access
-Running
existing
parallel
discrete
event
simulation
software in
the cloud
mJADES
[13]
Concurrent
simulations
-Execution -Laboratory
work
- no access
-Running
concurrent
simulation
SimSaaS
[11]
All
simulations
are
supported
-Modeling
-Execution
-Analysis
-Laboratory
work
- no access
-Running
simulation
with multi-
tenancy
feature.
Rescale
[15]
All
simulations
are
supported
-Modeling
-Execution
-Analysis
- online
access
-payable
-Running
all
simulations
in the cloud
TABLE 1: Comparison of Different Simulation Project
IV. LOAD BALANCING OF PARALLEL SIMULATION
In order to increase the efficiency of any distributed
application and reduce its response time, a load balancing
mechanism has to be implemented.
In this section, we introduce a dynamic load balancing
strategy based on the mobile agent paradigm for parallel and
distributed simulation over cloud computing. The idea of
using mobile agent is not new, since it has been discussed
before in [17] [18] [19] [20]. The reason behind its adoption
is that those autonomous entities have the ability to migrate
from an overloaded node to an under loaded one. They
continue their execution easier especially in the case of a
strong migration where the agent continues its execution at
the point before its transfer. Moreover, mobile agent reduces
the network traffic which decrease the communication cost
noticed in the existing message passing mechanisms and
helps to keep a global knowledge of the load of the system
under study.
Parallel agent based simulation is a typical example of
parallel simulation over cloud computing: “Agent based
simulation due to their computational power requirements
appear to be a natural application for parallel
architectures” [21]. The idea of load balancing used is a
combination between two existing works [19] [20] in an
attempt to address the problem observed by the modeling
and simulation community [1] [6] which is automatic
dynamic load balancing of PADS into cloud computing.
In [19], the proposed architecture is composed of two
agents. The first one is a mobile agent that collects load
information of all the nodes. Based on this information
determines a threshold according to which, the imbalance
state of the system is defined. The second one is a stationary
agent that orders tasks migration (if needed) to nodes that
are under loaded. In order to find the lightly loaded nodes,
messages are exchanged between nodes which results in
high communication latency. Meanwhile the solution
proposed by Patel [20] also comprises a stationary agent and
a mobile agent, but in this case the mobile agent is not only
responsible of information gathering but also for finding a
partner for its loaded node which makes its script complex
(takes long time to run). Thus the script of the mobile agent
has to be simplified as much as possible in order to get
accurate load information before that the load of the system
changes. Also network traffic should be reduced.
Our proposed idea for balancing the load of a parallel
agent based simulation contains three agents:
1. Controller Agent (CA): CA is a stationary agent. It
measures the local load of the node where it has been
created and also detects if the node is overloaded or not.
2. Collector Mobile Agent (CMA): CMA is a mobile
agent. It gathers load information measured by CA, and
calculates the average load of the system.
3. Locator Mobile Agent (LMA): when the load exceeds
the average load calculated by CMA, CA creates a mobile
agent called LMA “Locator Mobile Agent” in order to seek
a partner for its heavily loaded node.
This technique could be simply applied on a parallel
agent based simulation. In this case the agents composing
the model could move easily from an overload node to a less
loaded one to reduce simulation time.
V. CONCLUSION
This study is the first step towards enhancing our
understanding of cloud based simulation. In this work we
have presented the importance and the benefits of adopting
cloud computing into modelling and simulation field. We
have also proposed an idea on how to balance the load of a
parallel simulation using mobile agent paradigm in order to
speed up its execution.
VI. REFERENCES
[1] Richard Fujimoto, distributed simulation challenges in sensor networks and the cloud:
http://www2.isye.gatech.edu/~skim/NSF%20Workshop/fujimoto.pdf
[2] R. Buyya, J. Broberg, and A. Goscinski, Cloud Computing Principles and Paradigms, Canada, John Wiley & Son, 2011, 674 p.
[3] Q. Zhang. L. Cheng and R. Boutaba, “Cloud Computing: state-of-the-art and research challenges” J Internet Serv Appl. Springer, vol. 1, Apr. 2010, pp. 7-18, doi:10.1007/s13174-010-0007-6.
[4] http://scenarionavigator.systemsnavigator.com/
[5] T. Yamazaki. et al. “Simulation Platform: A Cloud-Based Online Simulation Environement” Neural Networks. Elseiver, vol. 24, Jun. 2011, pp. 693-698, doi: 10.1016/j.neunet.2011.06.010.
[6] X. Liu. et al. “Cloud-based computer simulation: Towards planting existing simulation software into the cloud” Simulation Modelling Practice and Theory.
Elseiver, vol. 26, May. 2012, pp 135-150, doi: 10.1016/j.simpat.2012.05.001
[7] A. Waqar Malik A. Park, and M. Fujimoto, “Optimistic Synchronization of Paralle Simulations in Cloud Computing Environments,” Proc. International Conference on cloud Computing, 2009, doi:10.1109/Cloud.2009.79.
[8] R. Fujimoto A. Waqar Malik A. Park. “Parallel and Distributed Simulation in the Cloud” SCS M&S Magazine, Jul, 2010.
[9] Gary Shao. Robert McGraw, “Service-Oriented Simulations for Enhancing Situatiion Awreness”, 2009.pp
[10] C. Seo. et al. “Implementation of Cloud Computing Environment for Discrete Event System Simulation using Service Oriented Architechture,” Proc. International Conference on Embedded and Ubiquitous Computing, 2010, doi:10.1109/EUC.2010.60
[11] H. Rajaei, Jeffrey Wappelhorst, “Clouds & Grids: A Network and Simulation Perspective”, Proc. Communications and Networking Symposium (CNS’11), 2011, pp. 143-150.
[12] W. Tsai et al. “SimSaaS: Simulation software-as-a-service”, Proc. Annual Simulation Symposium (ANSS’11), 2011, pp. 77-86.
[13] Gabriele D’Angelo, “Parallel and Distributed Simulation from Many Cores to the Public Cloud”, Proc. International Conference on high Performance Computing and Simulation (HPCS’11),Jul 2011, pp. 19
[14] Massimiliano Rak et al. “Cloud-based Concurrent Simulation at Work: Fast Performance Prediction of Parallel Programs” Proc. International WETICE, 2012, doi:10.1109/WETICE.2012.74.
[15] X. Liu. et al, “Cloud-based Simulation: the State-of-the-art Computer Simulation Paradigm,” Workshop. Principles of Advanced and Distributed Simulation, 2012, doi:10.1109/PADS.2012.11.
[16] www.rescale.com/platform
[17] Thomas Djotio Ndié, et al, “MAMID: A Load Balance Network Diagnosis Model Based on Mobile Agent”, journal of information security, vol 3, Octobre 2012, pp 281-294.
[18] VijayaKumar G; et al, “Using Mobile Agents For Load Balancing In Peer-To-Peer Systems Hosting Virtual Servers”, journal of computer scinece, 2014, doi: 10.3844/jcssp.2014.948.960. pp 948-960
[19] Salah EL-falou. Programmation répartie, optimisation par agent mobile. Doctoral thesis CAEN university , 2006.
[20] R.B Patel and Neetu Aggarwal, "Load Balancing on Open Networks: A Mobile Agent Approach", Journal of Computer Science, vol, 2 (4): pp 337-346, 2006.
[21] Cosenza. B et al, “Distributed Load Balancing for Parallel Agent-Based Simulations”, Conf Parallel, Distributed and Network-Based Processing (PDP), 2011.
Distributed Coverage Hole Boundary Detection inWireless Sensor Networks
Lynda Aliouane and Mahfoud BenchaıbaLSI, Department of Computer Science
University of Sciences and Technology Houari Boumedienne (USTHB)Algiers 16111, Algeria
Email: [email protected], [email protected]
ABSTRACT
Coverage holes can be caused by random deployment,depletion energy of sensors or by natural events such asfire. Detecting boundaries of coverage holes is interestingto repair the network or to route information outside thehole. In this paper, we present an algorithm that identifiesall boundary nodes. The main contribution of this paper isthe cooperation between nodes to compute the boundariesof coverage holes. Our algorithm is distributed, localizedand operates in two phases. In the first phase, each nodeidentifies locally if it is in the boundary of the hole. In thesecond, boundary nodes cooperate to detect the coveragehole. Several types of coverage holes are considered in thisalgorithm.
I. I NTRODUCTION
A rapid advances in sensing technology and network-ing have led to the emerging of Wireless Sensor Networks(WSNs). A WSN is composed of a large number of sensornodes characterized by tiny size and limited energy [1]. Eachnode can sense events, collect data and communicate withother nodes. A WSN have a wide range of applications sucha health monitoring, scientific applications, environmental andmilitary.
Coverage is one of the major concerns in WSN. Sensornodes have sensing coverage range of radiusRswithin whichsensor nodes can sense or observe events. Sensor nodes havealso communication range of radiusRc within which sensornodes can directly exchange messages. A relationship betweencoverage and connectivity is given in [2]. When transmissionrangeRc is at least twice the sensing rangeRs (Rc ≥ 2Rs),coverage implies connectivity In this paper, we focus on thedetection of coverage holes. A coverage hole is an area whichis not covered by any sensor in the network. A coveragehole occurs when several adjacent nodes in network fail dueto natural events such as fire or others such as crushing byanimals, vehicles or even depletion of battery. Coverage holescan be also caused by random deployment. Detecting coverageholes is important for many reasons; interesting events cannot be observed in these regions. So, it’s essential to detectcoverage holes in order to repair the network. Also for routingdata, it is important to detect the coverage holes to rerouteinformation outside the holes.
In this paper, we present a distributed algorithm that enablessensor networks to detect the borders of uncovered regionswithin a wireless sensor networks. The main contribution ofthis paper is the cooperation of nodes to detect coverage hole.Several complicated types of holes are considered.
This paper is organized as follows. In Section II, wepresent related work on detecting coverage holes in a WSN.Our distributed algorithm is detailed in Section III. A briefdiscussion is given in Section IV. Finally Section V concludesthe paper and gives some future work.
II. RELATED WORK
Several authors have proposed different algorithms to detectcoverage holes. A boundary detection algorithm [2] basedon the perimeter of each node’s sensing disk is proposed.In this algorithm, a node is considered as boundary if thereexists a point in the sensing disk which is not covered byany other node. Another algorithm [4] is also based on thesensors perimeter, but rather than verifying all the perimeter,it verifies only the crossings points of two sensing disks.These two works [2], [4] detect boundary nodes but the exactboundary of coverage hole is not identified. Another recentwork [5] based also on the perimeter of sensing disk, presentsa centralized hole detection algorithm by finding first boundarycritical points (BCPs). BCP is defined as an intersection pointnot covered by any other node. Then, the coverage hole isdetected by connecting each consecutive intersection pointsalong the border of the hole.
Other works on coverage hole detection presented in [6],[7], [8], [9], [10], [11], are based on Voronoi diagrams. TheVoronoi diagram of a set of nodes in the space is the partitionof the plane into Voronoi polygons [12]. Each sensor has itscorresponding Voronoi polygon, such that all the points insidethe polygon are closer to the corresponding node than to anyother node. If some portion of the polygon is not covered bythe sensor lying inside the polygon, it will not be covered byany other sensor and is considered as a boundary node.
The works [14], [15] introduce a new technique for detect-ing coverage holes by algebraic topology with minimal geo-metric data and no localization of nodes. But these solutionsare centralized and can’t detect all coverage holes.
A computational geometry-based algorithm is presented in[16]. It discovers coverage holes by verifying a triangle formed
by a sensor node and its two neighbors. The existence ofcoverage holes depends on the type of the triangle formedby three neighbors. Another work [17] using also triangularstructure, can detect a coverage hole and calculate its size. Thetriangular based works are simple for constructing and havelower calculation than Voronoi diagram.
The proposed works either based on Voronoi diagram ortriangular or even perimeter, can only detect if a node is onthe boundary of coverage hole. They determine the existenceof coverage holes, but they don’t discover the exact boundariesof the coverage holes. Also, these works don’t detect all typesof holes.
In our work we detect locally boundary nodes using perime-ter sensing disk. And unlike previous works, the nodes arounda hole cooperate to compute exact borders of coverage hole.
III. PROPOSED ALGORITHM
We start first by defining some terms used in our algorithm.
A. Definitions
A neighbor of a sensor in our case is a sensor within thesensing rangeRsof the sensor.A sensing diskof a sensor is a disk of radiusRs, centered atitself. A sensor can sense all events inside this disk.Intersection point is the point where two sensing disks (or thesensing disk and the field of interest) cross. Two sensing disks(or the sensing disk and the field of interest) can intersect inone or two distinct points.Boundary arc is an arc on sensing disk sensor that is notcovered by any other sensor node.Boundary sensor is a sensor which is adjacent to coveragehole. Boundary sensor has one or more boundary arcs.Coverage holeis the area which is not covered by any activenodes.Boundary of coverage holeis a list of boundary sensors thatare adjacent to the coverage hole and that has boundary arcs.
B. Problem statement
Consider a wireless sensor network, where nodes are de-ployed randomly, consider that each node has a sensing range,that is circular disk of radiusRscentered at its location. Anypoint within the sensing range is detected by the sensor. Asensor has also a communication range that is a circular diskof radius Rc which is centered at its location. A node cancommunicate with each node within its communication range.The communication radiusRc of a sensor node is at leasttwice of its sensing radiusRs (Sensor can transmit data to itsneighbors directly sinceRc>= 2Rs).We suppose that each node had a unique identification numberID. We assume that the sensor nodes know their exact locationand the neighboring nodes positions within the network. Wesuppose also that we know the limits of the field of interest.
The main goal of this contribution is the cooperation ofsensors to detect coverage holes in a distributed way. Also,we give the type and the number of coverage holes.
C. Proposed algorithm
Our algorithm detects nodes that enclose the coverage holein a distributed manner. The algorithm gives also the types ofthe detected hole at the end of discovery. We define three typesof coverage holes : closed holes, open holes and composedholes. We define first each type of hole.
Closed hole: a hole is closed when boundary nodes aroundthe hole form a cycle, as shown in Figure 1(a).
Open hole: a hole is open when there is not a cycle, anexample in Figure 1(b), shows an open hole.
Composed hole: a hole is composed when it consists ofmore than a hole, it can be composed from open holes, closedholes or combined of the two types of holes, see Figure 1(c).
(b) (c)(a)
Fig. 1. Examples of types of holes. (a) Closed hole, (b) Open hole, (c)Composed hole.
Our algorithm operates in three steps. The first detects theboundary nodes. In the second, each node discovers boundaryneighbors. In the third step, the boundary and the type ofcoverage hole is identified.
First step: : Locally detection of boundary nodes.Each node determines locally whether it is adjacent to anycoverage hole ”whether it is a boundary node”. We considerthat a node is boundary if there exists an intersection pointon its sensing disk that is not covered by the sensing disk ofany other node in the network. So, each node calculates theintersection points of its sensing disk with neighbors’ sensingdisk or with the limits of field of interest. If at least oneintersection point is not covered by another node, the nodecan conclude that it’s a boundary node. In this case, boundarysensor creates a list of boundary arcs : list-arcs that will containthe uncovered arcs in its sensing disk. The boundary arcs aredefined by two intersection points of the sensing disk.
An example is shown in Figure 2. A nodem is not aboundary node because all its intersection points are coveredby another node. Whereas, nodej is a boundary node becauseits intersection pointsa and b with nodes i and k are notcovered by another node. When nodej discovers that is aboundary node, it creates a list-arcs that contains the uncoveredarc ab.
i
jk
ab
m
Fig. 2. Detection of boundary nodes by checking intersection points.
Second step:: Discovery of boundary neighbors.When a nodei discovers that it is a boundary node, it sends
a Hello message to all its one-hop neighbors. This messageinforms the neighbors that nodei is a boundary node. Whenthe nodei receives a Hello message, it creates a list of itsboundary neighbors : list-boundary, that contains for eachnode, its boundary neighbors and at each receipt of Hello,node i inserts the ID received in this list. At the end of thisstep, each node knows all its boundary neighbors. In Figure 2,the boundary nodej, creates a list-boundary that containsi’IDandk’ID.
Third step: : Detecting the boundaries of coverage hole.For detecting the coverage hole, an initiator node must starta request message including its ID to initiate the detectionof coverage hole. The request message includes a list-holethat contains identities of boundary sensors limiting eachhole.When a sensor node receives this request message, it appendsits ID to the list-hole and choose a boundary neighbor fromits list-boundary which does not appear in the list-hole. Then,it forwards the message to its chosen boundary node.
The request message must end when it returns to theinitiator node. List-hole will contain all identities of boundarynodes limiting the hole. Thus, the initiator has the boundaryinformation of coverage hole. In Figure 2, considering thatiis the initiator, it creates a list-hole and sends it to nodej andso on until the request message returns to nodei.
This procedure is valid when the hole isclose. But theremay be cases where it’s not enough that the message returnsto the initiator. For example, in the case where the hole isnot close, see Figure 3. If nodei is the initiator, when therequest message returns to this node, the discovery of hole isnot finished.
Therefore, it is necessary to add a condition that the list-hole must contains all identities of list-boundary. So, whenthe request message returns at the initiator, it must verifyifthe resulting list-hole contains all its boundary neighbors inlist-boundary, else it must resend the request message to theboundary neighbor which does not appear in the list-hole. InFigure 3, when the request message returns to initiator nodei, it must verify if all its boundary neighbors are included inlist-hole which is not verified in this case, because m’ identitydon’t belong to list-hole, thus the sensori concludes that it is
anopen hole and must send the request message to sensorm.A node can be also boundary of two holes or more depend-
ing on number of boundary arcs, show example in Figure 3. Inthis case, a node creates another list-hole, each list identifies adifferent hole. In certain cases, two list-holes can identify the
i
j
k
m
H1H2
Fig. 3. Discovering open hole.
same hole. It’s at the end of request that we can discover if it’sthe same hole. Example in Figure 3 shows that the initiatornode i must creates two list-holes since it has two boundaryarcs, arriving at nodek which has only one boundary arc,we can discover that the two list-holes are for the same hole(H1 = H2).
When a sensor receives a request message containing twolist-holes, it appends its IDs in the two lists and verifies ifithas more than a boundary neighbor, it sends for each neighborthat shares the same intersection point a request messagecontaining one list-hole. If it finds only one boundary neighbor,it sends the request message with the two list-holes to theboundary neighbor. In Figure 3, sensori sends to nodej twolist-holes, each list identifies a different hole, nodej sends tok also the two lists. But nodek finding one boundary arc inhis list of boundary arcs, can detect that it’s the same hole,and appends its ID in the two list-holes and sends it to nodej.
The procedure detecting boundaries of coverage holes de-scribed in the third step, does not detect all types of hole.For example, in Figure 4 the request message initiated byican return to the initiator, which verifies that all boundaryneighbors are in list-hole but the holeH3 is not discovered.For this reason, we introduce a notion ofrelay node whoserole is to discover a portion of coverage hole not detected byinitiator and transmitting the result to the initiator. A relaynode uses a boolean initiated at false value and when it endsits discovery change the value of boolean and transmit it withits discovery to the initiator.
A node is identified as relay if it has more than twoboundary nodes or has a number of boundary arcs higher thanthe number of list-hole received. Example in Figure 4 shows anoder1 which has four boundary neighbors and receives twolist-holes. In this case, it identifies itself as relay and sendseach list-hole to the boundary which has the same intersection
H1
H2
H3H2
i
r1
r2
k l
1
1
1
1
2
2
2
2
Fig. 4. A relay nodes.r1 and r2 are two relay nodes.
point. Relayr1 sends List-hole 2 to nodel and List-hole 1 tonodek.
When the initiator creates the list-hole to initiate the discov-ery, it checks if it has more than boundary node, it sends at oneof them (for example who has maximum ID). It sendsk list-holes; such ask represents the number of intersection pointswith the boundary node to which it must send the requestmessage.
When a boundary node receives the message withk list-holes, it checks its list-arc. If it has the same number ofboundary arcs, it sends alsok list-holes, each list to a neighborwhich has the same intersection point.
The relay modifies value of boolean to True and sendsit with request message, since the request message may bereturned to initiator before the discovery of the entire hole iscompleted. When the initiator receives a list-hole with booleanat True, it discovers that a portion of hole is not yet discovered.The boolean is reset at False when the relay ends his request.Then, it must sends it to the initiator.The initiator must receive at the end a list-hole containingall boundaries node. Thus at each reception, the initiator mustverify that list-hole contains all list-boundary, also must verifythat all Boolean of all relays are False, else it must waitbecause the request is not finished.
IV. D ISCUSSION
Our coverage hole detection algorithm can find exact bound-aries of coverage hole even if the network is not connected.In each connected portion, an initiator starts the discovery.
There may be more than one initiator, so we suggest to usea back-off timer to avoid collision, providing an opportunityto prioritize different nodes. When the back-off timer expires,the nodes can start transmitting the request message, so thenode with the smallest back-off delay becomes a local initiatorand transmits the request. But after a certain time, we can findanother initiator. In this case, when the request message returnsto the first initiator, it can ignores each other request message.
Detecting Boundaries of coverage hole is interesting topatch the uncovered area. The boundary sensors can be used
for patching since these sensors know all the boundaries ofcoverage hole.
V. CONCLUSION
We have presented an algorithm detecting coverage holes ina wireless sensor network. The coverage hole detection is ac-complished by distributed algorithm. Each node independentlydetermines whether it is a boundary by verifying its intersect-ing points. Our approach is unique because sensor nodes cancooperate to discover the exact boundary of coverage hole.In future work, a hole recovery can be proposed to repairthe holes after implementing our algorithm of locating thecoverage holes.
REFERENCES
[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A surveyon sensor networks,”IEEE Communication Magazine, vol. 40, no. 8,pp. 104–114, August 2002.
[2] C.-F. Huang and Y.-C. Tseng, “The coverage problem in a wireless sen-sor network,”Proceedings of the 2nd ACM WSNA’03 (2003), September2003.
[3] N. Ahmed, S. Kanhere, and S. Jha, “The holes problem in wirelesssensor networks: A survey,”ACM SIGMOBILE Mobile Computing andCommunications Review, vol. 9, no. 2, pp. 4–18, 2005.
[4] H. Zhang and J. C. Hou, “Maintaining sensing coverage andconnectivityin large sensor networks,”journal of Ad Hoc Sensor Wireless Networks,vol. 1, no. 1, pp. 89–124, 2005.
[5] Z. Kang, H. Yu, and Q. Xiong, “Detection and recovery of coverageholes in wireless sensor networks,”Journal of Network and ComputerApplications, vol. 8, no. 4, pp. 822–828, April 2013.
[6] Q. Fang, J. Gao, and L. J. Guibas, “Locating and bypassingholes insensor networks,”Journal of Mobile Networks and Applications (2006),vol. 11, no. 2, 2006.
[7] G. Wang, G. Cao, and T. L. Porta, “Movement-assisted sensor deploy-ment,” IEEE INFOCOM 2004, Hong Kong, China (2004), June 2004.
[8] A. Ghosh, “Estimating coverage holes and enhancing coverage in mixedsensor networks,”29th Annual IEEE International Conference on LocalComputer Networks (LCN04)(2004), 2004.
[9] C. Zhang, Y. Zhang, and Y. Fang, “Detecting coverage boundary nodesin wireless sensor networks,”In Proc IEEE Int Conf Networking, Sensingand Control (ICNSC ’06)(2006), April 2006.
[10] B. Carbunar, A. Grama, J. Vitek, and O. Carbunar, “Redundancy andcoverage detection in sensor networks,”ACM Transactions on SensorNetworks (TOSN)(2006), vol. 2, no. 1, February 2006.
[11] B. Carbunar, A. Grama, and J. Vitek, “Coverage preserving redundancyelimination in sensor networks,”In Proceedings of IEEE InternationalConference on Sensor and Ad hoc Communications and Networks(SECON04), pp. 377–386, 2004.
[12] F. Aurenhammer, “Voronoi diagrams - a survey of a fundamentalgeometric data structure,”ACM Computing Surveys., vol. 23, no. 3, pp.345–405, September 1991.
[13] C. Zhang, Y. Zhang, and Y. Fang, “Localized algorithms for coverageboundary detection in wireless sensor networks,”journal of WirelessNetworks (2009), vol. 15, no. 1, January 2009.
[14] R. Ghrist and A. Muhammad, “Coverage and hole-detection in sensornetworks via homology,”The Fourth International Conference on Infor-mation Processing in Sensor Networks (IPSN05), UCLA, Los Angeles,pp. 25–27, April 2005.
[15] J. Kanno, J. Buchart, R. Selmic, and V. Phoha, “Detecting coverage holesin wireless sensor networks,”MED ’09 Proceedings of the 2009 17thMediterranean Conference on Control and Automation, pp. 452–457,2009.
[16] W. Qinq-Sheng and G. Hao, “Coverage holes discovery algorithm forwireless sensor network,”Journal of Theoretical and Applied Informa-tion Technology (2013), vol. 48, no. 2, February 2013.
[17] S. Babaie and S. S. Pirahesh, “Hole detection for increasing coveragein wireless sensor network using triangular structure,”InternationalJournal of Computer Science Issues (2012), vol. 9, no. 2, January 2012.
Fault-Tolerance in LEACH Protocol (MC-LEACH)
Chifaa TABET HELLEL, Mohamed LEHSAINI STIC Laboratory, Faculty of Technology
University of Tlemcen Tlemcen, Algeria
[email protected], [email protected]
Herve GUYENNET LIFC, Laboratory of Computer Science
University of Franche-Comte Besançon, France
Abstract— As the wireless sensor networks (WSNs) continue to grow, the number of failures becomes important and unavoidable due to energy depletion, environmental hazards, communication link errors, etc. In this paper; we propose an improved version of LEACH protocol that suffers from some limitations. Our goal is to make LEACH (Low Energy Adaptive Clustering Hierarchy) a fault tolerant protocol in order to achieve high energy efficiency and increase the network scalability and that by considering that each cluster head (CH) aggregates its data received from its members and instead of transmitting data aggregated to the base station, it sends it to one of its members called Member-CH which is close to him and it remaining energy is high, then this Member-CH sends this data to the base station, and if the CH fails this Member-CH becomes the new CH. The simulation shows that the proposed work is better than the LEACH protocol in terms of fault tolerance and energy consumption.
Keywords-component; WSN, Leach protocol, CH, Fault-tolerance.
I. INTRODUCTION
Wireless sensor networks (WSNs) [1] have received significant attention in recent years due to their potential applications in military sensing, wildlife tracking, traffic surveillance, health care, environment monitoring, building structures monitoring, etc. WSNs can be treated as a special family of wireless ad hoc networks. A WSN is a self-organized network that consists of a large number of low-cost and low powered sensor devices, called sensor nodes, which can be deployed on the ground, in the air, in vehicles, on bodies, under water, and inside buildings. Each sensor node is equipped with a sensing unit, which is used to capture events of interest, and a wireless transceiver, which is used to transform the captured events back to the base station, called sink node. Sensor nodes collaborate with each other to perform tasks of data sensing, data communication, and data processing. The use of clusters [2] for transmitting data to the base station leverages the advantages of small transmit distances for most nodes, requiring only a few nodes to transmit far distances to the base station. Each cluster has a representing node called cluster head that aggregates all data received from its members of cluster to send it to the destination node. Data aggregation is an efficient mechanism as it not only reduces the energy consumption of packet transmission but also lowers the traffic load and therefore reduces the contentions and collisions.
Since the sensor nodes are prone to failure [3], fault tolerance should be seriously considered in many sensor network applications to keep functioning network without any interruption in the presence of faults. The rest of the paper is organized as follows: Section II presents the protocol LEACH; in Section III, we have presented some protocols that extend LEACH; in Section IV, we propose the improved version of LEACH; Section V illustrates performance analysis of LEACH and the proposed scheme. Finally, we conclude our paper in Section VI.
II. LEACH PROTOCOL DESCRIPTION
Low-Energy Adaptive Clustering Hierarchy [4.5] is a distributed clustering protocol that utilizes randomized rotation of local cluster heads to evenly distribute energy consumption among sensors in the network. The main objectives of LEACH are:
• Extension of the network lifetime
• Reduced energy consumption by each network sensor node
• Use of data aggregation to reduce the number of communication messages.
The basic idea of LEACH protocol is that nodes elect themselves as cluster heads according to some probability. Nodes that are not cluster head join the cluster of the closest cluster head, as shown in figure 1. After the clusters are formed, nodes send their packets to their cluster heads; the cluster head aggregates their packets and sends the aggregated data to the sink directly. The cluster head selection process is performed once every round, and nodes that have been selected as cluster heads will not become cluster heads again in the following rounds before all nodes in the network have been selected as cluster heads. This protocol ensures that every sensor node in the network has the same probability to be selected as a cluster head, thus evenly distributing energy load to all nodes. Therefore it can increase the network life time compared with conventional routing protocols.
Fig.1. LEACH protocol model
In LEACH, nodes have the capability to adjust their transmission power such that they can use low transmission power to communicate with their neighbors, and use high transmission power to send packets directly to the sink. If all nodes send packets to the sink directly, nodes will die out very soon, especially for those nodes that are far away from the sink. If nodes use the Minimum Transmission Energy (MTE) routing protocol [4] to forward packets to their closest neighbor, nodes closer to the sink will handle more traffic than nodes farther away from the sink, and therefore will deplete their energy quicker. LEACH eliminates these problems by selecting some nodes, which are the cluster heads, to send packets to the sink directly, and let other nodes only send packets to the cluster heads. However, if the cluster heads are fixed, they will consume more energy than other nodes and die quickly because they have to participate in all communications. Therefore LEACH randomly selects different nodes as cluster heads in each round to avoid this problem. The operation of LEACH is divided into rounds, where each round contains a set-up phase and a steady phase. In the step-up phase, each node decides whether or not to become a cluster head by choosing a random number between 0 and 1. If the number is less than a threshold, than the node becomes a cluster head. Node n calculates the threshold as follows:
P
1- P * (r mod 1/P) if n ∈ G
T (n) = 0 otherwise Where: P is the percentage of nodes that are cluster heads, r is the current round, G is the set of nodes that have not been selected as cluster heads in the last 1/p rounds.
Using this threshold function, all nodes take turns to be cluster heads in a random order. After all nodes have been cluster heads exactly once, i.e., after 1/P rounds, all nodes start over to participate in the cluster head selection process again.
The node that elects itself as cluster head broadcasts an advertisement message to notify other nodes, nodes that are not cluster heads receive the cluster head advertisement messages. During the set-up phase, each non-cluster heads receive multiple advertisement messages. Assuming the radio channel is symmetric; choosing the cluster head with the strongest signal strength minimizes the required energy to transmit packets. Therefore nodes join the cluster of the cluster head from which they received the advertisement with the highest radio signal strength. Nodes send a “joining” message to the cluster head they want to join, and the cluster heads assign a TDMA time slot for each node to transmit its cluster. After the clusters are formed and the TDMA schedules are assigned, the network enters steady-phase. In this phase, nodes transmit their packets to the cluster head in their scheduled time slot. The cluster heads receive all packets from its cluster members, compress the data into one packet, and send the packet to the sink directly. However, nodes in different clusters may be scheduled to transmit in the same time slot, which will cause packet collisions and affect transmission in neighboring clusters. To reduce interference, each cluster chooses one CDMA code that is different from other clusters to communicate within the cluster. The CDMA codes are broadcast when the cluster heads broadcast their advertisement messages.
III. ENHANCED PROTOCOLS BASED-LEACH TAXONOMY
Number of hierarchical routing protocol has been proposed. LEACH is considering as a basic energy efficient hierarchical routing protocol. Many protocols have been derived from LEACH with some modifications and applying advance routing techniques. In the following, we discuss some derived protocols and explain the modifications added in each protocol.
A. LEACH-C
LEACH-Centralized uses a centralized clustering algorithm and same steady-state protocol. During the set-up phase of LEACH-C [5, 8], each node sends information about its current location (possibly determined using GPS) and residual energy level to the sink. In addition to determining good clusters, the sink needs to ensure that the energy load is evenly distributed among all the nodes. To do this, sink computes the average node energy, and determines which nodes have energy below this average. Once the cluster heads and associated clusters are found, the sink broadcasts a message that obtains the cluster head ID for each node. If a cluster head ID matches its own ID, the node is a cluster head; otherwise the node determines its TDMA slot for data transmission and goes sleep until it’s time to transmit data. The steady-state phase of LEACH-C is identical to that of the LEACH protocol.
B. V-LEACH
In this new version of LEACH [6], each cluster contains cluster head (CH) which is responsible only for sending data
received from its members of cluster to the base station, vice-CH the node that will become the CH of cluster in case of CH dies, cluster members that gathering data from environment and send it to the CH.
In the original version of LEACH, the CH is always on receiving data from cluster members, aggregates these data and then sends it to the base station that might be located far away from it. The CH will die earlier than the other nodes in the cluster because of its operation of receiving, sending and overhearing. When the CH dies, the cluster will become useless because the data gathered by cluster nodes will never reach the base station. The proposed V-LEACH protocol besides having a CH in the cluster, there is a vice-CH that takes the role of the CH when the CH dies.
C. K-LEACH
The proposed protocol K-LEACH [7] uses the K-medoids clustering algorithm to obtain highly uniform clustering of nodes and very good choice of cluster heads and it is a very well known fact that energy retention of a WSN is highly dependent on the grouping or clustering of transmitting and receiving nodes. K-LEACH considers least distant from the center of cluster as a criterion for a node to be chosen as a cluster head (CH) during cluster head selection procedure. Moreover, K-LEACH improves the clustering and cluster head selection process. For the first round of communication, in setup phase, the K-medoids algorithm is used for cluster formation, which ensures uniform clustering. The cluster formation by Kmedoids algorithm ensures best clustering and selection of cluster head using Euclidian distance at the nearer or at the center of cluster always gives most energy efficient solution in WSN. From second round onwards cluster heads are selected based on the next nearest node to the first round cluster head and so on. Then, they applied MRE till getting unique cluster heads, but as soon as cluster heads are duplicated due to dynamic clustering, they switch to random selection of cluster head nodes from amongst the alive nodes.
IV. THE PROPOSED WORK
A. MC-LEACH description
LEACH exhibits several properties which enable the protocol to reduce energy consumption, but after an analysis of LEACH, it has been shown that the direct communication between the cluster head and the base station consumes more energy and so the CH can depletes its energy quickly, so our objective is to prolong the network lifetime and to avoid packet loss.
In LEACH, at the time of cluster formation, when the cluster head is elected, it broadcasts an advertisement message to the other nodes, and each node that receives this message and it is not a cluster head and has a high RSSI (signal strength of message received), it considers itself as a member of this CH, and in this time, the member who is closest to the CH and its remaining energy is high is appointed Member-CH.
After that the CH receives data from its members, it aggregates this data, and instead of transmitting the data
aggregated to the base station directly, it forwards it to the Member-CH, and this latter transmits this data to the base station and this is done in a time interval T, and if this time T is exceeded and the Member-CH has not received anything, it indicates that the CH stops working and then the CH fails, and in this time the Member-CH is the new CH so, it will announce to the other members that it becomes a CH as shown in figure 2 .
Fig.2. MC-LEACH protocol model
V. SIMULATION RESULTS
A. Simulator:
TOSSIM, the TinyOS simulator, is a very useful tool for debugging and testing TinyOS programs. This tool is especially important because of the large number of possible problems that could be encountered by the motes which may not have anything to do with the NesC program. It would help a great deal to know that your code works fine, before you actually upload it to the motes [9].
Failure is not necessarily linked to energy depletion, in our simulation context, we assume that there is a rate of nodes that may stop working, or the probability that CH and its corresponding fall both down is almost negligible.
In LEACH protocol, the CH may fails during the round whereas, it members are alive and sensing data. If the CH fails, the collected data from members will be lost because it will never reach the base station. Also members consume their energies for sensing data while there will be no actual data transmission, so the entire cluster will be useless. But in our context, there may be several solutions such as: if the CH fails then its correspondent (Member-CH) will send a message to all cluster members informing them that the Member-CH became the new CH. And if the latter is crashed then the CH will send the data aggregated to the base station, and in the case that both CH and Member-CH fail is almost negligible.
B. Simulation parameters:
Parameter Value
Simulation time 500 sec
Number of nodes 200
Packet size 29 octets
Initial node power 1 J
Nodes distribution
Nodes are randomly distributed
02468
101214161820
50 100 150 200En
erg
y c
on
sum
pti
on
(J)
number of nodes
MC-LEACH
LEACH
Fig.3. variation of energy consumption with number of nodes
Fig.3 shows that the energy consumption in MC-LEACH is lesser than in LEACH because in LEACH, the unbalanced energy distribution tasks cause CH to consume its energy faster than other members. But in MC-LEACH, the CH and the Member-CH will divide the task.
50
60
70
80
90
100
5%
10
%
15
%
20
%
25
%
Ra
tio
of
succ
ess
ful
rece
pti
on
Fialure rate
MC-LEACH
LEACH
Fig.4. Ratio of successful reception with failure rates
Fig.4 shows that the ratio of successful reception of packets to the base station is higher than in LEACH, and this is because in LEACH, if CH fails the data packets will be lost while in MC-LEACH the Member-CH can be the new CH, and then transmits the packet to the base station.
VI. CONCLUSION
In this paper, we have proposed an enhancement of LEACH, a clustering based routing protocol called MC-LEACH to deal with fault tolerance in WSNs.
This proposal work takes into consideration the energy consumption which is the critical problem in Wireless Sensor Network and in the same time the fault tolerance, which minimizes the loss of packets in the presence of faults and then ensures the reliability of network.
REFERENCES [1] S. Misra et al. (eds.), Guide to Wireless Sensor Networks, Computer
Communications and Networks, DOI: 10.1007/978-1-84882-218-4 10,
_c Springer-Verlag London Limited 2009.
[2] Kasem Sohraby, Daniel Minoli, Taieb Znati, “Wireless Sensor Networks Technology, protocols, and Applications,” ISBN 978-0-471-74300-2, 2007Wiley-Interscience.
[3] Ajay, N.Tarasia, S. Dash, S.Ray, A.R.Swain “Fault Tolerant Multilevel Routing Protocol with Sleep Scheduling (FMS) for Wireless sensor Networks” European Journal of Scientific Research ISSN 1450-216X Vol.55 No.1, pp.97-108, 2011.
[4] Naveen Kumar, Mrs. Jasbir Kaur, “Improved LEACH (I-LEACH) Protocol for Wireless Sensor Networks” , IEEE 2011.
[5] W. Heinzelman, A. Chandrakasan and H. Balakrishnan, ”Energy-efficient communication protocols for wireless microsensor networks” Proceedings of the Hawaii International Conference on Systems Sciences, Jan. 2000.
[6] M. Bani Yassein, A. Al-zou'bi, Y. Khamayseh, W. Mardini ,“Improvement on LEACH Protocol of Wireless Sensor Network (VLEACH)”, International Journal of Digital Content Technology and its Applications Volume 3, Number 2, June 2009.
[7] Parul Bakaraniya#1, Sheetal Mehta*2, K-LEACH: An improved LEACH Protocol for Lifetime Improvement in WSN, International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 5- May 2013.
[8] M. Aslam, N. Javaid, A. Rahim, U. Nazir, A. Bibi, Z. A. Khan$, Survey of Extended LEACH-Based Clustering ,Routing Protocols for Wireless Sensor Networks, arXiv: v1 [cs.NI] 11 Jul 2012.
[9]
http://cs.ecs.baylor.edu/~donahoo/classes/5321/projects/SensorNet/tossim.html