A Variable Threats Based Self-Organization Scheme for Wireless Sensor Networks

Similar documents
Parallelism for Nested Loops with Non-uniform and Flow Dependences

RAP. Speed/RAP/CODA. Real-time Systems. Modeling the sensor networks. Real-time Systems. Modeling the sensor networks. Real-time systems:

Simulation Based Analysis of FAST TCP using OMNET++

Efficient Distributed File System (EDFS)

OPTIMAL CONFIGURATION FOR NODES IN MIXED CELLULAR AND MOBILE AD HOC NETWORK FOR INET

Evaluation of an Enhanced Scheme for High-level Nested Network Mobility

Wireless Sensor Network Localization Research

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

Cluster Analysis of Electrical Behavior

Overview. Basic Setup [9] Motivation and Tasks. Modularization 2008/2/20 IMPROVED COVERAGE CONTROL USING ONLY LOCAL INFORMATION

An Optimal Algorithm for Prufer Codes *

Performance Comparison of a QoS Aware Routing Protocol for Wireless Sensor Networks

Internet Traffic Managers

DECA: distributed energy conservation algorithm for process reconstruction with bounded relative error in wireless sensor networks

Constructing Minimum Connected Dominating Set: Algorithmic approach

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

Extending Network Life by Using Mobile Actors in Cluster-based Wireless Sensor and Actor Networks

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Positive Semi-definite Programming Localization in Wireless Sensor Networks

DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT

Adaptive Energy and Location Aware Routing in Wireless Sensor Network

Video Proxy System for a Large-scale VOD System (DINA)

Learning-Based Top-N Selection Query Evaluation over Relational Databases

A Load-balancing and Energy-aware Clustering Algorithm in Wireless Ad-hoc Networks

A Decentralized Lifetime Maximization Algorithm for Distributed Applications in Wireless Sensor Networks

A KIND OF ROUTING MODEL IN PEER-TO-PEER NETWORK BASED ON SUCCESSFUL ACCESSING RATE

DEAR: A DEVICE AND ENERGY AWARE ROUTING PROTOCOL FOR MOBILE AD HOC NETWORKS

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Load Balancing for Hex-Cell Interconnection Network

SPEED: A Stateless Protocol for Real-Time Communication in Sensor Networks

Mobility Based Routing Protocol with MAC Collision Improvement in Vehicular Ad Hoc Networks

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters

FAHP and Modified GRA Based Network Selection in Heterogeneous Wireless Networks

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Fusion Performance Model for Distributed Tracking and Classification

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

A Semi-Distributed Load Balancing Architecture and Algorithm for Heterogeneous Wireless Networks

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices

Base Station Location Protection in Wireless Sensor Networks: Attacks and Defense

Simulator for Energy Efficient Clustering in Mobile Ad Hoc Networks

Study of Data Stream Clustering Based on Bio-inspired Model

Long Lifetime Routing in Unreliable Wireless Sensor Networks

X- Chart Using ANOM Approach

Vol. 4, No. 4 April 2013 ISSN Journal of Emerging Trends in Computing and Information Sciences CIS Journal. All rights reserved.

TECHNIQUE OF FORMATION HOMOGENEOUS SAMPLE SAME OBJECTS. Muradaliyev A.Z.

A Spatiotemporal Communication Protocol for Wireless Sensor Networks

Analysis of Continuous Beams in General

A Robust Method for Estimating the Fundamental Matrix

Study of Wireless Sensor Network Location Algorithm and Data Collection Protocol

Private Information Retrieval (PIR)

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers

sensors ISSN

A New Transaction Processing Model Based on Optimistic Concurrency Control

Mobile Adaptive Distributed Clustering Algorithm for Wireless Sensor Networks

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS

Connection-information-based connection rerouting for connection-oriented mobile communication networks

Module Management Tool in Software Development Organizations

IMPROVEMENT of MULTIPLE ROUTING BASED on FUZZY CLUSTERING and PSO ALGORITHM IN WSNS TO REDUCE ENERGY CONSUMPTION

Machine Learning: Algorithms and Applications

Determining the Optimal Bandwidth Based on Multi-criterion Fusion

A Novel Fuzzy Stochastic Routing Protocol For Mobile AdHoc Network

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

EFT: a high throughput routing metric for IEEE s wireless mesh networks

Bayesian Model for Mobility Prediction to Support Routing in Mobile Ad-Hoc Networks

Cost-efficient deployment of distributed software services

CS 268: Lecture 8 Router Support for Congestion Control

Virtual Machine Migration based on Trust Measurement of Computer Node

Using Particle Swarm Optimization for Enhancing the Hierarchical Cell Relay Routing Protocol

Performance Improvement of Direct Diffusion Algorithm in Sensor Networks

Optimal Fault-Tolerant Routing in Hypercubes Using Extended Safety Vectors

Life Tables (Times) Summary. Sample StatFolio: lifetable times.sgp

Research Article Energy Efficient Interest Forwarding in NDN-Based Wireless Sensor Networks

A Unified Framework for Semantics and Feature Based Relevance Feedback in Image Retrieval Systems

VFH*: Local Obstacle Avoidance with Look-Ahead Verification

Performance Evaluation of Information Retrieval Systems

Cost-Effective Lifetime Prediction Based Routing Protocol for Wireless Network

IP mobility support is becoming very important as the

S1 Note. Basis functions.

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

K-means and Hierarchical Clustering

A STUDY ON THE PERFORMANCE OF TRANSPORT PROTOCOLS COMBINING EXPLICIT ROUTER FEEDBACK WITH WINDOW CONTROL ALGORITHMS AARTHI HARNA TRIVESALOOR NARAYANAN

IJCTA Nov-Dec 2016 Available

Solution Brief: Creating a Secure Base in a Virtual World

Distributed Grid based Robust Clustering Protocol for Mobile Sensor Networks

HYMN: AN INTEREST-BASED MULTIMEDIA PROVIDING SYSTEM FOR HYBRID WIRELESS NETWORKS

Priority-Based Scheduling Algorithm for Downlink Traffics in IEEE Networks

Classifier Selection Based on Data Complexity Measures *

Efficient Content Distribution in Wireless P2P Networks

Lobachevsky State University of Nizhni Novgorod. Polyhedron. Quick Start Guide

Scheduling Remote Access to Scientific Instruments in Cyberinfrastructure for Education and Research

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS

Reducing Frame Rate for Object Tracking

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour

CS 534: Computer Vision Model Fitting

A Finite Queue Model Analysis of PMRC-based Wireless Sensor Networks

Resource-Efficient Multi-Source Authentication Utilizing Split-Join One-Way Key Chain

A Topology-aware Random Walk

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Transcription:

009 Thrd Internatonal Conference on Sensor Technologes and Applcatons A Varable Threats Based Self-Organzaton Scheme for Wreless Sensor Networks Jan Zhong School of Computer Scence and Informaton Technology Royal Melbourne Insttute of Technology Melbourne, Australa E-mal: J.Zhong@student.rmt.edu.au Peter Bertok School of Computer Scence and Informaton Technology Royal Melbourne Insttute of Technology Melbourne, Australa E-mal: peter.bertok@rmt.edu.au Abstract Self-organzaton s an mportant ssue n wreless sensor networks because of the nherent unrelablty of the network. Besdes, varable threats n the networks can not be gnored. Extendng battery lfe and enhancng robustness under varable threats are two essental aspects whch need to be consdered when a self-organzaton scheme s explored. In order to address these ssues, a Redundant Nodes Selecton scheme and a varable Threats Probablty Estmaton scheme are proposed n ths paper. RNS s able to select redundant nodes that can be swtched off wthout affectng overall sensng coverage. TPE s able to help a sensor node to choose the most sutable path and avod hgh-threat neghbors n order to reduce packet loss. The scenaro wth RNS extended battery lfe by 30% to 50%, and postponed the occurrence of frst parttonng n the network by 7% to 40%. TPE decreased packet loss by 5% to 400% when a hgh threat level was nvolved. Keywords-wreless sensor networks; self-organzaton; varable threats; battery lfe; robustness I. INTRODUCTION For wreless sensor networks (WSNs), organzng typcally begns wth neghbor dscovery []. Nodes send rounds of messages (packets), buld local neghbor tables and organze clusters centered around a cluster head. The tables nclude nformaton on each neghbor s ID and locaton. However, durng operaton some sensors become nactve due to battery exhauston whch may result n network parttonng, and packets can be lost due to varous threats. Extendng battery lfe, postponng the occurrence of frst parttonng and reducng packet loss are sgnfcant aspects of self-organzng. A self-organzaton scheme supported by a redundant nodes selecton algorthm (RNS) and varable threats probablty estmaton (TPE) s proposed here to extend battery lfe and reduce packet loss. RNS s desgned to scan all sensor nodes and select redundant nodes that can be swtched off so that the whole area wll stll be covered. The redundant nodes wll be used as backups and replacements to extend the effectve network lfetme wthout any coverage loss. The second method TPE, mproves the scheme orgnally proposed n [6], by allowng nodes to choose a safer neghbor as a default path to the data snk and blockng hghthreat nodes to reduce packet loss. The whole proposed scheme provdes a soluton for WSNs to extend battery lfe and avod varable threats. II. BACKGROUND A. Topologes and Network Archtecture We consder a clustered organzaton, when ndvdual nodes are connected to a cluster head, and data from the cluster s relayed by the head towards the destnaton. However, n a number of cases there s no guarantee that all cluster heads can drectly connect to the data snk or all sensor nodes can drectly connect to a cluster head. For such cases, a random non-unform archtecture has been mentoned n [7], whch s shown n Fg.. In Fg. [7], a small number of hgh-end nodes, called Aggregaton and Forwardng Nodes (AFNs), are deployed together wth numerous low-end sensor nodes, called Mcro Sensor Nodes (MSNs). In addton, the network ncludes a globally trusted base staton (BS), whch s the ultmate destnaton for data streams from all the AFNs. The BS has powerful data processng capabltes, and s drectly connected to an outsde network. Each AFN s equpped wth a hgh-end embedded processor, and s capable of communcatng wth other AFNs over long dstances. A two-tered wreless sensor network model wll be used n our proposed method, and AFN wll be called Cluster Head (CH), BS wll be called Data Snk (DS) and MSN wll be called Sensor Node (SN) n the rest of the paper. Fgure. A two-tered archtecture 978-0-7695-3669-9/09 $5.00 009 IEEE DOI 0.09/SENSORCOMM.009.57 37

B. Self-Organzaton For WSN organzaton, a self-organzaton mechansm for non-unform dstrbuton of sensor nodes was proposed n [3], and a locaton-based scheme for both unform and nonunform was descrbed []. The mechansm n [3] offers good performance for non-unform dstrbuton, and an solated part of the network can be reconnected by moble sensors/devces. The method can fnd the optmum path to connect to the data snk effcently. The soluton n [3] reles on moble robots to mantan connectvty n the network, but n some cases a moble robot s not avalable. The locatonbased mechansm [] s relyng on a specal node named server node. Ths wll rase the cost of the whole network and f ths server node s compromsed or damaged, all nodes under ts control wll be affected. If ths functon s embedded n each cluster head, the power consumpton wll ncrease, due to most of the key management and delvery beng performed by the server node. To reduce the memory overhead as well as mantan securty for the network, a new approach s proposed n [7], whch s called Survvable and Effcent Clustered Keyng (SECK). However, a new ssue, default path selecton becomes a problem n hgh-threat networks. To reduce WSN power consumpton swtchng off some nodes has been ntroduced. To mantan sensng coverage, an nformaton coverage concept [4] was proposed, n whch a deactvated sensor s data s estmated by usng nformaton from the surroundng actve sensors. In [4], a balance between coverage and sensor densty was explored. In some cases data can be estmated relably, n other cases estmates cannot be a replacement of actual data. WSNs are often deployed n hostle envronments, and reslence to falures/attacks s mportant. The attack probablty estmaton algorthm n [5] has a good expermental result for statc attack probablty estmaton and connectvty mantenance. However, a varable attack and falure probablty s more realstc for WSNs. Ths paper wll focus on sensor node organzaton, and a neghbor-orented self-organzaton mechansm wll be proposed. These aspects wll be descrbed respectvely. III. PROPOSED METHOD A WSN self-organzaton scheme s proposed here to address two man aspects, namely battery lfe extenson and network reslence. The self-organzaton mechansm facltates network mantenance, and s based on a Redundant Nodes Selecton scheme (RNS) and a varable Threats Probablty Estmaton scheme (TPE). RNS s employed to select the redundant nodes that can be deactvated n order to save power. TPE s used to help a sensor node to choose the most sutable path to the cluster head and data snk, and avod hgh-threat neghbors n order to reduce packet loss. Noton : It s a redundant node, f and only f there are no changes n the covered area when t has been swtched to sleep mode. Fgure. The redundant node A. Redundant Node Selecton Assume that every node has ts own locaton nformaton whch wll be the coordnate n ths algorthm and all rado range (radus) wll be R. Assume that all sensor nodes are n the same -dmensonal area. In ths paper, assume a pont u s covered (montored) by a node v f ther Eucldan dstance s equal or less than the sensng range,.e. uv R. Defne the sensng crcle C(u) of node u as the boundary of u s coverage regon. The RNS algorthm has two steps. The frst one s to select redundant nodes and the second s to check whether the redundant nodes can be swtched off. The algorthm has the followng steps:. Put a node SN at the orgn of the coordnate system.. Assume there s a node SN n C(SN ) whch means node SN s n node SN s rado range, vce versa. Wthout loss of generalty, let SN be on the X-axs, shown n Fg. 3. 3. The X-axs and crcle C(SN ) ntersect at P. C(SN ) and C(SN ) ntersect at Q and Q. 4. To fnd the next crcle. 4. If there s a node SN 3 n the area Q SN W and SN Q R, then go to step 3 and replace SN by SN 3. If SN could be the next crcle, then node SN can be swtched to sleep mode, as shown n Fg. 4. 4. If there s a node SN 3 n the area Q SN W and SN Q > R, as shown n Fg. 5, there wll be a small area A B Q whch s not covered. If there exsts a node SN E that SN E A R, SN E B R, SN E Q R, then goto to step 3 and replace SN by SN 3. If there s no such SN E, the node SN wll not be swtched off, as shown n Fg. 6. 4.3 If there s not any nodes n the area Q SN W: Assume that there s a node SN A whch s outsde the area Q SN W, as t s shown n Fg. 7, and SN A SN < R, SN A SN R, and SN A Q < R. The C(SN A ) and the C(SN ) ntersect at C and C s n the area Q SN W. Then, go to step 4.3 and replace Q and SN by C and C. If Q s n the next crcle, then node SN can be swtched to sleep mode. If there s no such node SN A, the node SN can not be swtched off. 38

and all SN mp are defned as redundant related nodes (RR nodes or RRNs) and SN ml s defned as redundant related seed (RRS). Especally, there may be only one SN mp related wth SN ml and both of them are RRS for each other. Then, these two nodes are defned as twn redundant related nodes (TRRNs), shown n Fg. 8. Fgure 3. Put both nodes n the coordnate system Fgure 7. Step 4.3 Fgure 4. Step 4. Fgure 8. Twn redundant related nodes Fgure 5. Step 4. Fgure 6. Step 4. Noton : If there exsts a set of nodes R = {SN m, SN m, SN mn },n N, for any SN mk ( k n) that SN mk s a redundant node, but f any node SN ml ( l n) from among them s swtched off, there wll be at least one node SN mp ( p<n) that s no longer redundant, then the node SN ml The algorthm of swtchng the RR nodes s to separate them nto TRRNs. Frst, select a RRS from RR nodes and search all of the SN mp that whether there s a twn node for the RRS. If there s, swtch one of them off and put the other one back. Besdes, the selecton algorthm, from step one to step four, does not cover all coverage probabltes because the computaton overhead stll need to be consdered. The selecton for the TRRNs s based on the Varable Threats Probablty Estmaton algorthm whch s specfed n the followng secton. The hgher threat probablty estmaton node wll be swtched off. Besdes, for any sleepng node SN s, let a pont n the deployment area be Z=(x, y). If there exsts Z, ZSN > R, ZSNs R, s, SN s wll be swtched on. B. Varable Threats Probablty Estmaton There s no sure and effcent way to readly detect a node capture. The authors mentoned n [9] and [0]. Accordngly, a threats probablty estmaton algorthm s employed n the self-organzaton scheme. In my proposed method, a new deployment model s used Compared wth that n [5], the attack and falure probablty 39

does not focus on groups but on ndvdual sensors. A new algorthm s proposed to keep the whole network connected and avod key threats. For securty of Trusted Neghbors, I refer to Reputaton-based Framework for Hgh Integrty Sensor Networks [8]. Also, a sensor node S ( N) s assocated wth a basc falure coeffcent b s representng the threat from the envronment and b s s predstrbuted value by experence. Let w be the varable attack and falure weght, whch s obtaned by experence. Let ϕ be threshold threat level and let D be the detected attack level. Then w s defned ϕ as w, Dϕ = ϕ w, Dϕ = ϕ w =, n, wn Dϕ = ϕ n Here n denotes the number of threat levels. Let { ϕ, ϕ,, ϕn} be a set of threshold values and N T shows how many tmes a certan threat level has been detected. For a certan threat level, the falure estmate s F = wn. The T dfference between the falure estmates at tme k and that at tme (k-) can be defned as e = FS F ( k ) S () whch s the frst correcton factor. Here FS ( denotes the estmate falure of S at tme k. Then the second correcton factor M c, relatve threat value, can be defned as FS M c = θ (3) FC ( S ) where θ s a constant coeffcent. FC ( S )( denotes the sum of falure estmate of S 's neghbors, whch can be defned as F = F + F ( k (4) C( S ) ) S S j S j C ( S ) The thrd correcton factor s connectvty detecton, whch s the number of sensors that have at least an avalable path to a cluster head. It can be defned as CNTS CDS = ( )( CNT CNT ) S (5) S CNTS Here CNTS ( denotes the connectvty of S at the tme k and CNT denotes the connectvty when the network frst S deployed. If CDS > 0, t means some nodes are compromsed or unavalable at the tme k. The frst correcton factor measures the dversty between dfferent tme ntervals. The second factor measures the relatve threat. The thrd factor measures the connectvty for a sensor and ts neghbors. The total correcton measurement derved from (), (3) and (5) can be defned as M A( = ω ( e( + M c( + γcds + ( e( ) (6) α d + M c( ) dk + β ( e( + M c( ) dk Here α, β, γ are constants and wll be set based on experence. Let δ, a constant, be the decreasng threat value, () whch means f no falure/attack problems are detected, b s wll gradually drop down. Then b s can be defned as bs = bs ( k ) δ + M A (7) If bs < 0, then let bs = 0. The bs ( s the real tme threat estmate and ths wll also be used n RNS algorthm for TRRNs. After real tme threat estmate b s s calculated, the degree of a sensor node can be derved. A degree of a sensor node denotes the number of avalable connectons for a sensor node (number of shared keys wth neghbors). Gven a set of threshold { ϖ, ϖ, ϖ n }, n for any sensor nodes, the degree can be calculated by κ, bs = ϖ κ ϖ (8), bs = K( S) =, n, κ bs = ϖ n Here K(S ) denotes the degree of a sensor node n a deployment group. In (8), for, j, j, f ϖ ϖ, then j κ. In my proposed method, dfferently from that n [5], κ j the hgher attack/falure probablty a sensor has, the fewer keys t has. C. Organzng To dscover a prmary cluster head (CH), each sensor node (SN) wants to dscover the ID of the closest CH. In my proposed scheme, let SNID be the ID of a SN, DP be Default Path, DD be the Depth of Default Path, DTD be Dstance to Default Path, SLP be the status (sleepng or actve), TL be Threat Level, NL be a Neghbor Lst. Then, we gve each SN an expresson (9): SN = { SNID, DP, DD, DTD, SLP, TL, ω)} (9) Where ω s the ndex of SN s neghbors. SN s parameters are expressed wth dotted notaton, for nstance, SN.DP denotes the SN s Default Path. After deployment, the RNS algorthm (detaled n A) s actvated. SNs whch are redundant wll be set to SLP =. At the same tme, all SNs dscover all ther neghbors and store n NL. Then they wll wat for connecton from ther neghbors whch can connect to cluster heads and the algorthm s descrbed as follows. At frst, each CH wll search SNs wthn ts sensng range. Each SN wthn CH s sensng range wll update ts expresson (0). SN = update{ DP = CH, DD = 0, DTD, SLP = 0, ω)} (0) Then these SNs contnue to tell ther neghbors they can communcate wth a cluster head by sendng a path message (): PathInfo = { SNID = SN, DP = CH, DD = 0, DTD, SLP = 0} () The SNs who receve the path message () wll update ther expresson (). SN = update{ DP, DD, DTD, SLP)} () If a SN receves more than one path message, t wll calculate the power consumpton (PC) on these dfferent communcaton paths. As mentoned above, the proposed 330

scheme s neghbor-orented. The default path selecton algorthm s descrbed n (3). SN = update{ PC( DP, DD, DTD)} (3) Here PC denotes the power consumpton functon. Before gve the expresson for total power consumpton, assume power consumpton s proportonal to the square of dstance wth a coeffcent θ and each hop wll consume λ. Thus, the power consumpton functon can be descrbed as PC( SN, ndex)) = θ ( DTD + ndex). DTD ) + λ DD (4) ndex).dtd denotes the dstance between a neghbor s default path and the neghbor. The paths from a SN to a CH are called communcaton lnks. For nstance, SN need to send messages to SN j, and SN j need to forward to SN k then fnally to the CH, the lnk j k CH s called a communcaton lnk. Theorem : For any SNs, f there exsts PC (SN,ndex )) < PC (ndex )), the power consumpton of the communcaton lnk on ndex ) s less than that on ndex ) s true. Proof: PC( SN, ndex )) = θ ( DTD + ndex ). DTD ) + λ DD = ( θ DTD + λ) + ( θ ndex ). DTD + ( λ DD )) = PC( SN) + PC( ndex )) (5) PC(SN) denotes the power consumpton between the SN and ts default path and PC(ndex )) denotes the power consumpton between the node of SN s default path and the node of SN s default path s default path. For any avalable communcaton lnks, we defne a searchng functon (SF), that SF(SN) = SN.ndex ) f for any n ( < n N, N s the number of SN s neghbor), there s PC (SN,ndex )) < PC n (ndex n )), where SN.ndex ) denotes the SN s neghbor wth ndex. We defne SF(SF(SN)) = SF (SN), then the most power-savng path wll be the communcaton. DD+ lnk SN SF( SN) SF ( SN) SF SN. Thus, for any SNs, f there exst PC (SN,ndex )) < PC (ndex )), that the power consumpton of the communcaton lnk on ndex ) s less than that on ndex ) s true. Lemma : For any SNs, f there exsts PC (SN,ndex )) < PC n (ndex n )),( < n N, N s the number of SN s neghbors), the ndex ) s the mnmum power-savng lnk for SN. Based on Lemma, all SNs can fnd the mnmum power path. If a sleepng node s on a mnmum power path, then t wll be swtched on. IV. SIMULATION RESULTS In ths secton, smulaton results are gven for total energy consumpton and total packages lost n the network. The latter ndcates network robustness, that s, the ablty of the network to contnue operatng after varable threats. In the smulaton, a JAVA based wreless sensor network smulator was used. Compared wth other smulators, such as OMNeT++, the JAVA smulator proved to be more flexble for envronment confguraton and mplementaton of the proposed solutons. A. Power Consumpton Unform sensor dstrbuton and unform traffc dstrbuton was used when examne the total energy consumptons between before and after mplementng proposed method. In the smulator, each hop costs 0.0005% battery lfe, around 0.00006% per smulaton meter and 0.00% per workng sensor. Smulatons wth dfferent parameters produced smlar results. A typcal set of results s presented here. There were 00 sensors wth three cluster heads and sensor densty s 8. The sensors were arranged n a 0 0 matrx. Fg. 9 shows the total battery consumpton wthout proposed scheme and Fg. 0 shows the total consumpton under the same crcumstance but wth self-organzaton scheme actvated. There are smlar smulaton results n the normal dstrbuton scenaros. Fgure 9. Total power consumpton n unform dstrbuton wthout RNS Fgure 0. Total power consumpton n unform dstrbuton wth RNS The charts prove that the RNS scheme could reduce energy consumpton of the network, and thereby extend the network s lfetme. B. Robustness Ths paper descrbes a Threats Probablty Estmaton (TPE) based self-organzaton scheme to support the key management. For comparson, the default key management scheme s shortest path frst (SPF). There are total 00 sensors were deployed n such area and average sensor densty s 6. One fourth of the sensors are set as hgh threats and random threats are launched sx tmes n each scenaro. In the smulator the process of detectng threats was not modeled, but rather the threat event was drectly passed on to the sensors. ) Normal sensor dstrbuton wth normal traffc dstrbuton Frst, we look at the scenaro when sensors are deployed n a normal dstrbuton (Gaussan dstrbuton). The traffc 33

dstrbuton s also normal, to smulate centralzed events, such as fre n a forest. The Fg. shows the robustness of the network wth SPF scheme and Fg. shows that wth TPE based self-organzaton scheme. The curves llustrated n Fg. ndcate the TPE scheme could enhance the robustness of the network, by reducng packet loss from 5% to %. ) Unform sensor dstrbuton wth unform traffc dstrbuton In unform dstrbuton, the TPE scheme mproves the robustness by reducng packet loss from 4% to 4%. Fgure. Robustness of the network wth SPF n normal dstrbuton Fgure. Robustness of the network wth TPE n normal dstrbuton V. DISCUSSION AND CONCLUSION The smulaton results showed that battery lfe was extended. The results prove that the RNS scheme could reduce energy consumpton of the network, and thereby extend the network s lfetme. Varable threats, such as nodes falure n hostle envronments are nevtable. A sensor wth hgh threat level ndcates a hgh packet loss probablty when packets are receved or sent. The proposed TPE method scans all neghbors and helps sensors to avod hgh-threat nodes to reduce packet loss. TPE also showed a dstnct mprovement n network behavor. Wth hgh threats n the network, TPE can be consdered as an effectve scheme to mprove robustness by reducng packet loss. In my proposed method, the mprovement by RNS depends on sensor densty, the hgher the densty, the more mprovement. Low sensor densty networks wll not beneft sgnfcantly from RNS. TPE s desgned to counter varable threats and there wll not be much mprovement on the scenaro wthout varable threats. possble f t was not for hs contnuous gudance at every stage. A/Prof. Peter Bertok s supports, suggestons and supervson on the whole progress of the project were apprecated. I would also lke to thank hm for proofreadng the drafts and helpng me come up wth ths work. I would also lke to thank my parents and my fancée Ru Ru Zhang for ther understandng and support. REFERENCES [] B. Karp, Geographc routng for wreless networks, PhD Dssertaton, Harvard Unversty, October 000. URL http:// actcomm.dartmouth.edu/papers/karp:paper.pdf [] R. D Petro, L. V. Mancn and A. Me, Energy effcent node-tonode authentcaton and communcaton confdentalty n wreless sensor networks, Wreless Netowrks, Volume, Number 6, pp 709-7, 006. URL http://www.sprngerlnk.com/content/y50w8uk8748g60/fulltext.pdf [3] D. Lu and P. Nng, Effcent dstrbuton of key chan commtments for broadcast authentcaton n dstrbuted sensor networks, In Proceedngs of the 0th Annual Network and Dstrbuted Sensor Networks (ISOC), 003. URL http://www.soc.org/soc/conferences/ndss/03/proceedngs/papers/7. pdf [4] B. Wang, K. C. Chua and V. Srnvasan, We Wang, Sensor densty for complete nformaton coverage n wreless sensor networks, In Proceedngs of ACM SenSys, Sngapore, 005. [5] C. Yu, C. L, C. Lu, D. Lee and S. Kuo, Attack probablty based determnstc key predstrbuton mechansm for non-unform sensor deployment, In Proceedngs of the 7th Internatonal Conference on Dstrbuted Computng Systems Workshops(IEEE), pp 8, 007. [6] C. Karlof and D. Wagner, Secure routng n wreless sensor networks, Unversty of Calforna at Berkeley, 00. URL http://webs.cs.berkeley.edu/papers/sensor-route-securty.pdf [7] M. W. Chorzempa, Key management for wreless sensor networks n hostle envronments, Paper submtted to the Faculty of the Vrgna Polytechnc Insttute and State Unversty, 006. URL http://scholar.lb.vt.edu/theses/avalable/etd-050006-740/unrestrcted/chorzempapaper_v.pdf [8] Saurabh Ganerwal and Man B. Srvastava, Reputaton-based Framework for Hgh Integrty Sensor Networks, SASN 04, October 5, 004, Washngton, D.C., USA [9] G. Jolly, M. Kusu, and P. Kokate. A herarchcal key management method for low-energy wreless MSN networks. In Proc. of the 8th IEEE Symposum on Computers and Communcaton (ISCC), pp. 335 340. Turkey, July, 003. [0] S. Zhu, S. Seta and S. Jajoda. LEAP: effcent securty mechansms for large-scale dstrbuted MSN networks. In Proc. of the 0th ACM Conference on Computer and Communcaton Securty (CCS), pp. 6-7, Washngton DC, October 003. ACKNOWLEDGMENT I would lke to thank my supervsor, A/Prof. Peter Bertok, for hs contnuous support and useful feedback throughout the project. Ths paper would not have been 33