Phd Thesis

Ph.D. THESIS Defended on: March 13, 2014 in Engineering School of Communications, SUP'COM To obtain Diploma of

Doctor In Information and Communications Technology Elaborated by:

Ramzi BELLAZREG

Security, Deployment and Scheduling in WSN-based Applications: Models and Applications Thesis Committee: President:

M.Nabil Tabbane

Maitre de conférences à l'école supérieure de communications de tunis, Tunisia.

Examiners:

M. Yahya Slimani

Professeur à l'Institut Supérieur des Arts Multimédias de la Mannouba, Tunisia.

M. Ahmed Mehaoua

Professeur à l'université Descartes, France

Members:

M. Slim Rekhis

Maitre de conférences à l'école supérieure de communications de tunis, Tunisia.

Thesis Supervisor:

M. Noureddine Boudriga

Professeur à à l'école supérieure de communications de tunis, Tunisia.

This Thesis is elaborated in Communication Networks and Security Research Unit CNAS, SUP'COM www.cnas.org.tn

Paris

Acknowledgment Several people gave me support to achieve this thesis and I would like to use this opportunity to thank them all for their help and assistance. First of all, I would like to express my deep and sincere gratitude to my advisor, Professor Noureddine Boudriga. His wide knowledge and his logical way of thinking have been of great value for me. His invaluable comments, ideas, encouragement, and guidance have provided a good basis for the present thesis. I would also like to express my gratitude to Professor Ahmed Mehaoua and Professor Yahya Slimani for reviewing the manuscript of this thesis, and to Dr. Nabil Tabbane and Dr. Slim Rekhis for evaluating this work. During this work, I have collaborated with many colleagues, from the Communication Networks and Security Research Laboratory (CN&S), for whom I have great regard. It has been a pleasure cooperating with and learning from them. My thanks go also to the faculty and staff of the Engineering School of Communications (Sup’Com), who provided me a great environment and the resources needed to complete this work. Finally, I owe special gratitude to my family. My deepest gratitude goes to my parents, my sister and my wife for their encouragement, care, and sacrifice.

i

Contents 1. Introduction to the Wireless Sensor Networks 1.1. Introduction to the chapter . . . . . . . . . . . . . . 1.2. The WSN architecture . . . . . . . . . . . . . . . . . 1.2.1. The sensors architecture . . . . . . . . . . . . 1.2.2. The WSN network architecture . . . . . . . . 1.3. Applications of the Wireless Sensor Networks . . . . 1.3.1. Military surveillance and target tracking . . . 1.3.2. Border surveillance applications . . . . . . . . 1.3.3. Environmental applications . . . . . . . . . . 1.3.4. Agriculture . . . . . . . . . . . . . . . . . . . 1.3.5. Health Applications . . . . . . . . . . . . . . 1.3.6. Home and smart Applications . . . . . . . . . 1.4. The research issues for Wireless Sensor Networks . . 1.4.1. Infrastructure and network architecture . . . 1.4.2. Network topology . . . . . . . . . . . . . . . 1.4.3. Deployment of the sensor nodes . . . . . . . . 1.4.4. Mobility of the sensors . . . . . . . . . . . . . 1.4.5. Power saving and scheduling . . . . . . . . . 1.4.6. Security . . . . . . . . . . . . . . . . . . . . . 1.5. Thesis statement and contributions . . . . . . . . . . 1.5.1. Deployment techniques for large areas . . . . 1.5.2. Scheduling solutions . . . . . . . . . . . . . . 1.5.3. Secured distributed key management protocol 1.5.4. Border Surveillance applications . . . . . . . 1.6. Outline of Dissertation . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

1 1 1 2 3 5 5 5 6 6 7 7 8 8 9 9 10 11 11 12 12 13 13 14 14

2. Deployment strategies for large areas 2.1. Introduction to the chapter . . . . . . . . . . . . . . . . . . . 2.2. Deployment strategy based on radio irregularities . . . . . . . 2.2.1. Related Works . . . . . . . . . . . . . . . . . . . . . . 2.2.2. Irregular radio propagation . . . . . . . . . . . . . . . 2.2.3. A new coverage control strategy . . . . . . . . . . . . 2.3. Performance evaluation of the radio irregularities deployment 2.3.1. The details of ALUL metric . . . . . . . . . . . . . . . 2.3.2. Evaluation of the developed model . . . . . . . . . . . 2.3.3. Impact of the radio effect components . . . . . . . . . 2.4. Deployment based on geographical patterns . . . . . . . . . . 2.4.1. Architectural issues . . . . . . . . . . . . . . . . . . . 2.4.2. The Deployment Strategy . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

16 16 17 17 18 19 21 21 21 25 26 27 28

ii

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

Contents 2.4.3. The Re-Deployment . . . . . . . . . . . . . . . . 2.5. Performance evaluation of the Geographical deployment 2.5.1. The Simulation Model . . . . . . . . . . . . . . . 2.5.2. The Simulation Scenario . . . . . . . . . . . . . . 2.5.3. The Length of Uncovered Path (LUP) . . . . . . 2.5.4. Simulation Results . . . . . . . . . . . . . . . . . 2.6. Conclusion of the chapter . . . . . . . . . . . . . . . . .

Contents . . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

30 31 31 32 32 33 34

3. Scheduling Schemes for Wireless Sensor Networks 3.1. Introduction to the chapter . . . . . . . . . . . . . . . . . . . . . 3.2. Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. A deployment and energy based sleep scheduling scheme . . . . . 3.3.1. The energy based coverage control strategy . . . . . . . . 3.3.2. The energy based Scheduling algorithm . . . . . . . . . . 3.4. Performance evaluation of the energy based scheduling . . . . . . 3.4.1. The Worst deployment without scheduling . . . . . . . . . 3.4.2. The Round Robin Scheduling . . . . . . . . . . . . . . . . 3.4.3. The proposed scheduling solution . . . . . . . . . . . . . . 3.5. An optimized scheduling scheme based on target mobility . . . . 3.5.1. The architectural issues . . . . . . . . . . . . . . . . . . . 3.5.2. The proposed scheduling strategy . . . . . . . . . . . . . . 3.5.3. Selecting the model parameters . . . . . . . . . . . . . . . 3.6. Performance evaluation of the targets mobility based scheduling . 3.6.1. The Simulation Model . . . . . . . . . . . . . . . . . . . . 3.6.2. The activated sensors . . . . . . . . . . . . . . . . . . . . 3.6.3. The lifetime of the network . . . . . . . . . . . . . . . . . 3.7. Conclusion of the chapter . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

35 35 36 37 37 38 39 40 40 41 43 44 44 49 52 52 52 53 54

4. DynTunKey: A Dynamic Distributed Group Key Tunneling Management Protocol 4.1. Introduction to the chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3. Architectural issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1. Need for encrypted tunnels in WSNs . . . . . . . . . . . . . . . . . . . . 4.3.2. Proposed security architecture . . . . . . . . . . . . . . . . . . . . . . . 4.3.3. Communication exchanges . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Initial and authentication exchanges . . . . . . . . . . . . . . . . . . . . . . . . 4.5. Cluster SA negotiation exchanges . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.1. Sensor clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.2. Establishment of the CSA . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6. Robustness of the algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6.1. Messages Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6.2. Key Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6.3. The effect of a node compromise . . . . . . . . . . . . . . . . . . . . . . 4.7. Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7.1. Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7.2. Comparison of DynTunKey to MAKM, NSKM and RDKM . . . . . . . 4.7.3. Comparison of DynTunKey and IPSec . . . . . . . . . . . . . . . . . . .

55 55 57 61 61 61 62 63 65 65 66 67 67 69 69 70 70 72 73

Contents

Contents

4.8. Conclusion of the chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 5. Border Surveillance using sensor based thick-lines 5.1. Introduction to the chapter . . . . . . . . . . . 5.2. Related Works . . . . . . . . . . . . . . . . . . 5.2.1. Border Surveillance applications . . . . 5.2.2. The adhoc routing protocols . . . . . . 5.3. Architectural issues . . . . . . . . . . . . . . . . 5.3.1. Presentation of the linear networks . . . 5.3.2. Node Hierarchy . . . . . . . . . . . . . . 5.3.3. The Network topology . . . . . . . . . . 5.4. The deployment strategy . . . . . . . . . . . . . 5.4.1. The deployment of the DRNs . . . . . . 5.4.2. The deployment of the BSNs . . . . . . 5.4.3. The deployment of the DDNs . . . . . . 5.5. The routing technique . . . . . . . . . . . . . . 5.5.1. The routing from BSN to DRN . . . . . 5.5.2. The routing from DRN to DDN . . . . 5.5.3. The routing to the BSNs . . . . . . . . 5.6. Performance evaluation . . . . . . . . . . . . . 5.6.1. The simulation model . . . . . . . . . . 5.6.2. The variation of the number of hops . . 5.6.3. The number of non-connected BSNs . . 5.7. Conclusion of the chapter . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

80 80 81 81 82 84 84 84 85 86 86 87 89 89 89 94 94 95 95 96 96 98

6. DWBS: A Distributed Wireless Border Surveillance System 6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2. Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3. The characterization of the paving patterns . . . . . . . . . . 6.3.1. The shape of the paving pattern . . . . . . . . . . . . 6.3.2. The customization of the paving patterns . . . . . . . 6.4. DWBS: paving based deployment technique and architecture 6.4.1. The paving of a monitored area . . . . . . . . . . . . . 6.4.2. Architectural issues . . . . . . . . . . . . . . . . . . . 6.5. DWBS: the deterministic deployment case . . . . . . . . . . . 6.5.1. The deployment model for DRNs . . . . . . . . . . . . 6.5.2. The deployment model for BSNs . . . . . . . . . . . . 6.6. Performance evaluation . . . . . . . . . . . . . . . . . . . . . 6.6.1. Evaluation of the DRNs deployment . . . . . . . . . . 6.6.2. Evaluation of the BSNs deployment . . . . . . . . . . 6.7. Conclusion of the chapter . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

99 99 100 101 101 107 109 109 110 112 112 114 116 116 117 118

7. DWBS: The controlled random deployment case 7.1. Introduction . . . . . . . . . . . . . . . . . . . . 7.2. The DRNs controlled random deployment case 7.3. The DRNs inter strips connectivity . . . . . . . 7.4. The BSNs controlled random deployment case .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

119 119 120 123 126

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

Contents

Contents

7.5. The comparison of the proposed model to the random deployment 7.5.1. The variation of PLC in relation with Rc . . . . . . . . . . . 7.5.2. The variation of PLC in relation with δD . . . . . . . . . . . 7.5.3. The variation of PLC in relation with WED . . . . . . . . . 7.5.4. Interpretation of the simulation results . . . . . . . . . . . . 7.6. The evaluation of DWBS connectivity and coverage . . . . . . . . 7.6.1. The evaluation of DWBS connectivity . . . . . . . . . . . . 7.6.2. The evaluation of DWBS coverage . . . . . . . . . . . . . . 7.7. The DWBS global performance probabilities . . . . . . . . . . . . . 7.7.1. The inter strips connectivity . . . . . . . . . . . . . . . . . 7.7.2. The linear connectivity and coverage . . . . . . . . . . . . . 7.8. Conclusion of the chapter . . . . . . . . . . . . . . . . . . . . . . . 8. Conclusions and perspectives 8.1. Deployment strategies . . 8.2. Scheduling schemes . . . . 8.3. Security Protocol . . . . . 8.4. Future Works . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

128 128 129 130 130 131 131 133 134 135 135 136

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

137 137 138 139 139

Bibliography

141

A. Computation of the measure of a sub ellipse

148

List of Figures 1.1. Sensor nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. The sensor node architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3. The WSN architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 4 4

2.1. Radio irregularities impact on the surface of sensing . . . . . . . . . . . 2.2. Density requirements for target coverage . . . . . . . . . . . . . . . . . . 2.3. ALUL evaluation for the trivial model . . . . . . . . . . . . . . . . . . . 2.4. ALUL evaluation for the log-normal shadowing model . . . . . . . . . . 2.5. Variation of the coverage range . . . . . . . . . . . . . . . . . . . . . . . 2.6. ALUL evaluation for the averaged log-normal shadowing model . . . . . 2.7. ALUL evaluation for the irregular (k,t) coverage model . . . . . . . . . . 2.8. Impact of P L(d0 ) on the ALUL . . . . . . . . . . . . . . . . . . . . . . . 2.9. Impact of χσ on the ALUL . . . . . . . . . . . . . . . . . . . . . . . . . 2.10. Impact of n on the ALUL . . . . . . . . . . . . . . . . . . . . . . . . . . 2.11. A sub zone coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.12. The Image Sensors Deployment . . . . . . . . . . . . . . . . . . . . . . . 2.13. Impact of the number of division per cell on the intrusion percentage . . 2.14. Impact of the mean direction of the targets on the intrusion percentage

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

19 20 22 23 23 24 24 25 26 27 28 29 33 34

3.1. Worst deployment without scheduling . . . . . . 3.2. The Round Robin scheduling method . . . . . . 3.3. Number of sets needed . . . . . . . . . . . . . . . 3.4. Scheduling using the proposed model . . . . . . . 3.5. The Network Architecture . . . . . . . . . . . . . 3.6. The Displacement Prediction . . . . . . . . . . . 3.7. The variation of the velocity . . . . . . . . . . . . 3.8. The variation of the direction . . . . . . . . . . . 3.9. The decomposition of the velocity into intervals . 3.10. The decomposition of the direction into intervals 3.11. The Averaged Number of Sensors . . . . . . . . . 3.12. The Lifetime of the Network . . . . . . . . . . . . 4.1. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7. 4.8.

The The The Key Key The The The

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

40 41 42 42 44 45 46 47 47 48 53 54

communication exchanges . . . . . . . . . . . . . Group Key establishment . . . . . . . . . . . . . . CSA establishment steps . . . . . . . . . . . . . . Storage Space for Random Walk . . . . . . . . . . Storage Space for Gauss Markov . . . . . . . . . . Number of tunnels for the Random Walk Model . Number of tunnels for the Gauss Markov Model . Number of messages for the Random Walk Model

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

62 65 67 72 73 74 74 76

vi

. . . . . . . . . . . .

. . . . . . . . . . . .

List of Figures

List of Figures

4.9. The Number of messages for the Gauss Markov Model . . . . . . . . . . . . . . 76 4.10. The Latency time for the Random Walk Model . . . . . . . . . . . . . . . . . . 77 4.11. The Latency time for the Gauss Markov Model . . . . . . . . . . . . . . . . . . 78 5.1. The nodes hierarchy . . . . . . . . . . . . . . . . 5.2. The network topology . . . . . . . . . . . . . . . 5.3. A DRN broken line . . . . . . . . . . . . . . . . . 5.4. The required density for DRN deployment . . . . 5.5. The determination of the width of the strip . . . 5.6. The routing process of sensed data . . . . . . . . 5.7. An illustration of a broadcast storm . . . . . . . 5.8. The solution to a node failure . . . . . . . . . . . 5.9. The number of hops . . . . . . . . . . . . . . . . 5.10. Percentage of non-connected BSNs for 1-coverage 5.11. Percentage of non-connected BSNs for 2-coverage

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

85 86 86 87 88 90 90 93 96 97 98

6.1. The landing motion and paving pattern . . 6.2. The elliptic wind model . . . . . . . . . . . 6.3. The circular wind model . . . . . . . . . . . 6.4. The rectangular wind model . . . . . . . . . 6.5. The decomposition of the ellipse En . . . . 6.6. A maximal part paving of a strip . . . . . . 6.7. A total paving of a strip . . . . . . . . . . . 6.8. The paving of a large area . . . . . . . . . . 6.9. The topology of DWBS . . . . . . . . . . . 6.10. The aerial deployment of the sensors . . . . 6.11. An area paving for the DRNs . . . . . . . . 6.12. The strip paving for the DRNs . . . . . . . 6.13. The choice of deployment positions of BSN 6.14. BSNs thick linear surveillance . . . . . . . . 6.15. The number of connected points . . . . . . 6.16. The number of deployed BSNs . . . . . . . 6.17. The percentage of covered points . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

103 105 106 107 108 109 110 111 111 112 113 113 115 115 116 117 118

7.1. The DRN neighbors connectivity . . . . . . . . . . . . . 7.2. The DRN positions for Line Connectivity . . . . . . . . 7.3. The decomposition of the ellipse ED . . . . . . . . . . . 7.4. A case where connectivity is impossible . . . . . . . . . 7.5. The needed intersection area for inter strips connectivity 7.6. The intersection with the paving patterns . . . . . . . . 7.7. The intersection between Discx,Rs and paving patterns . 7.8. The variation of PLC in relation with Rc . . . . . . . . . 7.9. The variation of PLC in relation with δD . . . . . . . . . 7.10. The variation of PLC in relation with WED . . . . . . . 7.11. The variation of PLC in relation to WED and Rc . . . . 7.12. The variation of PLC in relation to WED and δD . . . . 7.13. The variation of PLC in relation to δD and Rc . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

121 122 122 123 124 125 127 129 129 130 131 132 132

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

List of Figures

List of Figures

7.14. The Probability of Linear Coverage . . . . . . . . . . . . . . . . . . . . . . . . . 134 7.15. The Probability of Inter Strips Connectivity PISC . . . . . . . . . . . . . . . . 135 7.16. The Probability of Linear coverage and connectivity PLSC . . . . . . . . . . . . 136 A.1. The surface delimited by f(x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 A.2. The measure of a sub ellipse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

List of Tables 2.1. Results of the four models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.2. The new coverage required densities for the Re-deployment . . . . . . . . . . . 31 3.1. 3.2. 3.3. 3.4. 3.5.

Decay of P L(d0 ) in relation with time . . Variation of ALUL . . . . . . . . . . . . . The probabilities of the Velocity intervals The probabilities of the direction intervals The areas probability . . . . . . . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

40 43 47 48 48

4.1. 4.2. 4.3. 4.4.

Comparison of Key Management Solutions . . . . . The standard deviation of the number of tunnels . The standard deviation of the number of messages The standard deviation of the security latency . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

60 75 77 78

5.1. 5.2. 5.3. 5.4.

A comparison of our routing protocol to other The used messages for routes establishment . The non-connected BSNs for 1-coverage . . . The non-connected BSNs for 2-coverage . . .

routing protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

83 94 97 97

6.1. Comparison of the deployment methods . . . . . . . . . . . . . . . . . . . . . . 102

ix

List of Tables

List of Tables

Abstract Ramzi BELLAZREG. TITLE. PhD thesis, Engineering School of Communications (Sup’Com), Networks and Security Research Lab (CN&S), Date 2014. (Under the direction of Pr. Noureddine Boudriga).

The WSNs are used in many fields to build surveillance systems to detect and report intrusion related events. Many research works have focused on the implementation of WSNs and treated many issues related to the Wireless Sensor Networks. In this thesis, we will address research issues related to the wireless sensor networks to provide solutions adapted to different applications and implementation contexts. In the first issue, we took interest to provide new deployment strategies for WSNs. We developed several deployment techniques adapted to many contexts while resolving a set of constraints. We first developed a deployment technique that takes into account the radio irregularities. We also provided a deployment technique taking into account the variation of the sensing range in relation to the energy consumed by the sensors. An other deployment technique based on the geographical nature of the monitored area was developed. As a fourth deployment technique, we developed a heterogeneous network suitable for Border Surveillance applications and presented deployment and routing techniques to ensure both coverage and connectivity. An other framework for Border Surveillance is provided and a new deployment technique based on paving the monitored area with paving patterns is introduced. In this model, we consider an aerial deployment of several types of nodes. The paving patterns correspond to the landing area of the sensors in relation with the environmental factors. In the second issue, we focused on the scheduling schemes to extend the lifetime of the network. We propose two scheduling schemes based on different decision factors. The first scheduling scheme allows the sensors to alternate between active and sleep status is depending on their energy. The proposed scheduling scheme updates in time the number of activated sensors to cover the monitored area with the reduced density depending on the sensors energy. The second scheduling scheme is based on the analysis of the targets mobility. A prediction of the next positions of the targets will give decisions on the sensors that will be activated. The scheduling scheme generates a non uniform coverage of the monitored area to use only the needed sensors while preserving full monitoring of the operation area. In the third issue, we proposed a dynamic tunneling protocol named DynTunKey. The proposed security protocol ensures the authentication of sensor nodes, along with the integrity and the confidentiality of the exchanged data. DynTunKey constructs many-to-many tunnels and establishes group keys. All the trusted nodes contribute in the establishment of the secured tunnels. We introduced in DynTunKey the concept of the CSA (Cluster Security Association) to represent the many-to-many tunnel based on the group key established. DynTunKey is shown suitable to applications that necessitate secured communication between sensors that belong to the same group.

Introduction to the Wireless Sensor Networks

1

1.1. Introduction to the chapter Recent advances in micro-electro-mechanical systems (MEMS) technology and digital electronics have enabled the development of low-cost, low-power, multifunctional sensor nodes that are small in size. In their first implementations, the sensors were deployed to detect specific event in their sensing area. The researchers in many related disciplines were interested in conducting many research works based on the developed sensors because of their low cost and sensing capabilities. They found in the sensors technology, the best solution that permits a continuous and efficient monitoring of events in a given area. For example, based on micro digital electronics, many recent research works have focused on the use of the sensors. The main results of the works have focused on the development of a new concept called the Wireless Sensor Networks (WSN). Using the network technology, the sensors are no longer used in a single manner but are organized in a network and take benefits of the advances of wireless communications to implement larger and more efficient networks. These networks permit a monitoring of a larger area using collaborating sensor nodes. Wireless sensor networks represent a significant improvement over traditional sensing systems. The introduction of the WSN has attracted many researchers to work on this field and has become one of the most investigated research domain during the last years. The design goals and applications related to the WSNs are currently considered in many recent research works. In this introductive chapter we will present the general aspects of the WSNs and the thesis statement. We will introduce in Section 1.2 the Wireless Sensor Networks architecture. Section 1.3 will present the most investigated research issues related to WSNs. Section 1.4 will be devoted to the presentation of the most current applications using the WSNs. In Section 1.5 we will present the statement and contributions of this thesis. Section 1.6 will present the outline of the dissertation.

1.2. The WSN architecture This section represents WSN architectural issues. We will essentially present the sensors architecture and a typical WSN architecture.

1

1.2. THE CHAPTER WSN ARCHITECTURE 1. INTRODUCTION TO THE WIRELESS SENSOR NETWORKS

1.2.1. The sensors architecture A sensor node is a node in a wireless sensor network (WSN). This node is capable of sensing events, performing processing tasks on the detected events, and communicating the gathered data with other sensor nodes belonging to the same network. Example of a sensor node are depicted by Figure 1.1.

Figure 1.1.: Sensor nodes A sensor node is generally composed of four main components. These components are the processing unit, the power unit, the transceiver, and the sensors.

• The processing unit is the device entity that executes all the tasks performed by the sensor node. The processing unit is generally a micro-controller. It processes the sensed data and controls the hardware components of the sensor node. To each processing unit is associated an external memory. Two categories of memory based on the purpose of storage can be distinguished: the user memory is used for storing application related to personal data, and the program memory is used for programming the unit. Program memory also contains identification data of the device. • The power unit is the power source of the sensor. Power is stored either in batteries or capacitors. Batteries, both rechargeable and non-rechargeable, are the main source of power supply for sensor nodes. The sensor node consumes power for sensing, communicating and data processing. More energy is required for data communication than any other process. An important aspect in the development of a wireless sensor node is to ensure that there is always energy available to power the system. • The Transceiver unit is a single communication device of the sensor node and combines both transmitter and receiver. The transceiver is used by the sensor node to send and receive data and communicate with the different neighboring sensors and in case of necessity to the outside of the network. In general cases, the sensor nodes communicate on a wireless transmission media to avoid the installation of heavy wired connections. The wireless transmission media can be either radio frequency (RF) or optical communication (laser) or infrared. The Lasers consume less energy but need direct line-of-sight for communication. The infrared, like lasers, do not need antenna but the broadcasting capacity is limited. The most used wireless media for the sensors is the Radio Frequency based communication, because it satisfies the requirements of most of the sensors based applications.

1.2. THE CHAPTER WSN ARCHITECTURE 1. INTRODUCTION TO THE WIRELESS SENSOR NETWORKS The transceiver can operate in one of four operational states which are transmit, receive, idle, and sleep. The energy consumption level differs from one state to another. For a better utilization of the power, a sensor operates either in idle or sleep mode to consume less energy, if it is not sending or receiving data,. • Sensors are the most important hardware devices in a sensor node. The processing unit, transceiver and power unit are important devices since they are needed to support the main functionality of the sensor nodes, which is the detection of events. The sensing unit measures and controls particular events or parameters and produces a measurable response to a change in a physical condition like temperature, pressure or also objects presence. Each sensor has a certain coverage area in which it can sense efficiently and report the events observed. Sensors measure physical data of the parameter to be monitored and generates an analog signal. The analog signal is digitized by an Analogto-Digital Converter (ADC) and sent to the processing unit for further data processing. Some sensor nodes contain more than one sensor to be able to detect various types of events using the same sensor node. The evolution and development of sophistical sensors units have improved the possibilities of sensor node utilization. In addition to the four essential components, a sensor node may have additional components to ensure some supplementary tasks. • A sensor node can include a location finding system that permits accurate determination of the sensor node position. This information is needed in many WSN applications. Generally, a Global Positioning System (GPS) is attached to the sensor node to operate as a location finding system. • Some sensor nodes can be equipped with energy generators to extend the lifetime of the sensor node. Generally, these devices produce power by energy scavenging and extraction from the environment. An example of energy generators is the solar cellsbased system. • In some implementations and applications of WSN, the sensor node position is not static but it is dynamic and may change its position, when needed. In that case, the sensor node is equipped with an additional device called a mobilizer to move the senor node. Figure 1.2 depicts the architecture of a sensor node including the most common included hardware components.

1.2.2. The WSN network architecture A Wireless Sensor Network consists of spatially distributed sensor nodes to monitor physical or environmental conditions. The WSN is built of nodes from a few to several hundreds or even thousands depending on the scale of the monitored area. The data collected between sensor nodes is sent to a specific node generally referred to as a sink node. The sink node reports the gathered data to a main location for processing. The data relay between the sink node and the analysis center can be done directly or through a gateway node. A generic WSN architecture is depicted by Figure 1.3. Depending on the types of used sensor nodes, the Wireless Sensor Networks can be classified based on two dimensions.

1.2. THE CHAPTER WSN ARCHITECTURE 1. INTRODUCTION TO THE WIRELESS SENSOR NETWORKS

Figure 1.2.: The sensor node architecture

Figure 1.3.: The WSN architecture • Simple vs flat networks. In a flat WSN, all the sensor nodes have the same sensing, communication and processing roles. For this kind of architecture, all the sensor nodes operate as sensing nodes and collaborate to ensure simple routing tasks. This kind of network is used for low cost deployment providing a simple monitoring of a supervised area without need of complex treatments and tasks. • Heterogeneous vs uniform networks. An heterogeneous network integrates various sensor types with different capabilities. Typically, a large number of inexpensive nodes are deployed to perform simple sensing tasks. A few expensive nodes are deployed to provide advanced functionalities such as node management, data filtering, processing, fusion and transport. These nodes have larger battery lifetime, more powerful processing and communication capabilities. The presence of heterogeneous nodes in a sensor network has the advantage of increasing network reliability and lifetime because the energy of the simple nodes will not be used in advanced treatment that consumes a lot of energy. This separation of roles promotes a cost-effective design of the WSNs as well as a more efficient implementation of the overall sensing application.

1.3. APPLICATIONS CHAPTER 1. INTRODUCTION OF THE WIRELESS TO THE SENSOR WIRELESS NETWORKS SENSOR NETWORKS

1.3. Applications of the Wireless Sensor Networks As presented previously, the main goal of a WSN is monitoring a phenomena in an operational area. In their first implementations, the WSNs were designed to ensure detection of objects in an area. In addition to the researchers in the telecommunications field, the researchers and industrials in electrotonic and electric engineering were interested on developing advanced sensor units. We notice that sensors permit nowadays the monitoring of diverse phenomena. Many types of sensor units are implemented such as seismic, low sampling rate magnetic, thermal, visual, infrared, acoustic and radar, which are able to monitor a wide variety of ambient conditions. This progress has enlarged the fields in which can be used the WSNs while permitting the sensing of phenomena that were impossible or hard to be performed. In the following subsections, we summarize the most important WSN based applications.

1.3.1. Military surveillance and target tracking Military applications are considered as the main WSN applications. In general cases, the military fields are not directly accessible and it is not easy to install infrastructures in a hostile environment. The ability to quickly set up a WSN network among military units in hostile territory without any infrastructure support can provide a perfect solution supporting military applications. The self-organizing mechanisms of the WSNs provides a robust and reliable sensing in dynamic battle fields. Monitoring friendly forces: Headquarters can continuously monitor the status of friendly troops and their locations. Small sensors can be installed in troops, soldiers and vehicles to gather informations about their functioning and/or positions and send the information periodically to the analysis center. The reports relative to a given troop can also be sent to the troop headquarter to let it aware of anything going wrong for the troop. Battlefield surveillance and target tracking: Inaccessible regions and critical areas can be rapidly covered with sensor networks and closely watched for the activities of the opposing forces. The self-organizing feature of such networks makes the deployment of them very fast without human intervention. One particular case of battle field surveillance is the targets tracking application. In this application, the deployed WSN will detect the presence of enemies entities and also will permit a tracking of the path followed by this target. Additional advanced analyses permit an anticipation of the targets future movements.

1.3.2. Border surveillance applications One of the most recent monitoring applications of WSNs is the border surveillance application. This kind of applications is becoming critical due to the increase of the risks of intrusion on borders. Because to the global risks near their borders, governments are frightened of the appearance of intruders, either for unauthorized importation of goods or for terrorism actions. Generally, the borders are large and it is costly and impossible to perform a direct human surveillance along all the border. An automated surveillance system is then needed to provide continuous and efficient monitoring of suspect activities. The better solution found for an efficient monitoring is the use of WSNs. Deploying small device sensors near the border will permit the detection of intruders. These sensors are organized in a WSN and send the

1.3. APPLICATIONS CHAPTER 1. INTRODUCTION OF THE WIRELESS TO THE SENSOR WIRELESS NETWORKS SENSOR NETWORKS collected data to an analysis center. This solution provides an easily to implement solution for a real time surveillance of border crossings.

1.3.3. Environmental applications Sensors can be used to monitor events that occurs in the environment. Sensor networks are used to track the movement of animals such as insects, fish, or birds. They can also be used in monitoring environmental conditions and atmospheric contexts. Forest fire detection: An example of environmental application on WSNs is forest fire detection. In this application, the sensors are deployed in a forest and help to determine the exact origin of a fire before it becomes spread uncontrollable. The early detection is crucial for a successful action of the firefighters; thanks to Wireless Sensor Networks, the fire brigade will be able to know when a fire has started and how it is spreading. Flood detection: The WSNs can also be used for flood detection. WSNs composed of many sensing nodes used in this system to detect, predict, and prevent the damages that can be caused by floods. Wireless nodes are distributed in rivers so that changes of the water level can be effectively monitored. This kind of surveillance is difficult to achieve by humans because the monitored regions are large and sometimes not accessible. Monitoring Bio-diversity: Satellite and airborne sensors are used in observing large bio diversity, but they are not fine grain enough to observe small size bio diversity. WSNs can be used to control, monitor and observe the finest details of bio-complexity of the environment. Ground level sensor nodes are used to observe the bio-complexity.

1.3.4. Agriculture Using wireless sensor networks within the agricultural industry is becoming increasingly common; using a wireless network frees the farmer from the maintenance of wiring in a difficult environment. Gravity feed water systems can be monitored using pressure transmitters to monitor water tank levels, pumps can be controlled using wireless devices and water use can be measured and wirelessly transmitted back to a central control center for billing. Irrigation automation enables more efficient water use and reduces waste. Accurate agriculture: Wireless sensor networks let users make precise monitoring of the crop at the time of its growth. Hence, farmers can immediately know the state of the item at all its stages which will ease the decision process regarding the time of harvest. Irrigation management When real time data is delivered, farmers are able to achieve intelligent irrigation. Data regarding the fields such as temperature level and soil moisture are delivered to farmers through wireless sensor networks. When each plant is joined with a personal irrigation system, farmers can pour the precise amount of water each plant needs and hence, reduce the cost and improve the quality of the end product. The networks can be employed to manage various actuators in the systems using no wired infrastructure.

1.3. APPLICATIONS CHAPTER 1. INTRODUCTION OF THE WIRELESS TO THE SENSOR WIRELESS NETWORKS SENSOR NETWORKS Greenhouses: Wireless sensor networks are also used to control the temperature and humidity levels inside commercial greenhouses. When the temperature and humidity drops below specific levels, the greenhouse manager must be notified via e-mail or cell phone text message, or host systems can trigger misting systems, open vents, turn on fans, or control a wide variety of system responses. Recent research in wireless sensor networks in agriculture industry give emphasis on its use in greenhouses, particularly for big exploitations with definite crops. Such micro climatic ambiances need to preserve accurate weather status all time. Moreover, using multiple distributed sensors will better control the above process, in open surface as well as in the soil.

1.3.5. Health Applications Wireless sensor networks for health care have emerged in the recent years. Each patient has small and light weight sensor nodes attached to it. These small sensors allow an efficient and continuous measurement of the patients physiological data. For example, a sensor node may detect heart rate, blood pressure or temperature. The medical applications can be of two types: wearable and implanted. Wearable devices are used on the body surface of a human or just at close proximity of the user. The implantable medical devices are those that are inserted inside human body. The data collected by sensor nodes is sent to a decision center where the doctors can remotely follow the patient. The reception of continuous physiological data allow doctors to identify symptoms earlier. In addition, this wireless sensor network allows greater freedom of movement meaning that the patients are not forced to be in the hospital or the health care center but can move in any place and the physiological data will be sent via wireless communication to a centralized decision center. Sensors networks can also collect vital signs and monitor the health status and behavior of the elderly.

1.3.6. Home and smart Applications Besides the above presented applications, sensor networks can improve the way of life through many home applications such as home automation and smart environment. The sensors are considered to be ubiquitous, their integration can be with all appliances. This application is not very productive when compared with the previously implemented applications but the architects has found in the WSNs a perfect solution to enhance the luxury of the build and designed homes. Smart Homes: In smart homes, sensors make intelligent decisions about how to adapt, what changes to make, what actuation to do on the basis of statements of environmental change. A scenario example could be turning lights on when someone enters the room at night, by controlling the temperature by switching the cooling/heating levels at an air conditioner, etc. Smart sensor nodes and actuators can also be placed into appliances such as refrigerators, furnace and air-conditioner. Thus, users can manage their home via Internet or satellite. Environmental control in office buildings: For many cases, adequate airflow is needed to distribute air throughout the facility in order to maintain uniform air temperatures. The WSNs are implemented in that case to monitor the repartition of the air temperature. If the air temperature is not uniformly distributed, the fans are then automatically ordered to reduce or augment their speed and have an intermittent or constantly operation.

1.4. THE CHAPTER RESEARCH 1. INTRODUCTION ISSUES FOR WIRELESS TO THE WIRELESS SENSOR NETWORKS SENSOR NETWORKS Home and buildings safety: Sensor networks provide an inexpensive way to implement a security system able to monitor both indoor and outdoor. Small sensors can be placed on windows, doors or common areas to generate alerts in the case of intruder presence. A sensor network can also be deployed in parking to provide security monitoring against car thefts. The same WSN can be used for another purpose and inform about the number and position of free parking places. Interactive museums : Some museums use WSNs to bring interactivity between visitors and museum so to let them learn more. Each visitor has a multimedia device that displays comments about he museum contents. The rooms and expositions of the museums are equipped with sensing nodes and when a visitor is near an exposition automatically the audio played in the multimedia device describes the corresponding exposition. This kind of application is very easy to implement using WSNs and enhances the quality of expositions and permits an enjoyable and easier navigation in the museum.

1.4. The research issues for Wireless Sensor Networks The WSN research issue is one of the most investigated field in communications and wireless networks. An efficient design of a WSN needs the fulfilment of a set of constraints to ensure the performances and applications of the conceived network. A study of the required design perspectives is then necessary. With the proliferation of the WSNs applications, the investigated research concerns related to the WSNs are more and more diversified. In this section, we will present the most addressed problems and design goals by the researchers in the WSN field.

1.4.1. Infrastructure and network architecture One of the addressed topics for WSN is the architectural issues of the network. The architecture of the network depends widely on the application that will be implemented on the WSN deployed. Defining the architecture of the networks consists especially at : • defining the kind of network either simple or hierarchical, meaning composed on a unique kind of sensors or many types of sensors. The choose of the network kind depends widely on the needed sensing quality and type. • the choice of the kinds of needed sensors to ensure the monitoring performances and goals of the network. Based on the phenomena that will be monitored, the kind of the sensor node should be selected. • a definition of the relation rules between the different deployed sensors either for processing, communication or hierarchical relations and dependance. Given the importance of the infrastructure and network architecture choice on the ulterior network performances many research works have focused on this aspect.

1.4. THE CHAPTER RESEARCH 1. INTRODUCTION ISSUES FOR WIRELESS TO THE WIRELESS SENSOR NETWORKS SENSOR NETWORKS

1.4.2. Network topology The network topology designs the physical repartition of the sensors in the monitored area. We can distinguish many possible topologies. The most used topologies are presented in the followings. • The sensors can be organized as a star network where all the nodes are connected to a single sink node. • The sensors can be physically organized as a mesh or a single hop network. In this topology any sensor node is able to communicate directly with every other node. • A multiple hop networks form either an arbitrary graph or a tree structure of the sensors deployed. • The sensors can also be physical arranged in the form of linear network for which all the sensors deployed form a line. The topology chosen depends on the application that will be implemented. It also affects many network characteristics such as the latency, the routing, the data processing, the network lifetime and the robustness.

1.4.3. Deployment of the sensor nodes The deployment of the sensors is the physical placement of them in the monitored area. Many research works have focused on the deployment process and we find in the literature several forms of deployment. Deployment can be done in a random or deterministic manner. Also it can be done in one time activity or in a continuous process along all the network lifetime. Regardless to the selected deployment strategy, the locations of the nodes are of great importance because it will directly affect the quality of surveillance. The deployment of sensors generally should ensure two major important requirements of the networks which are respectively the coverage and the connectivity. 1.4.3.1. Coverage The sensors have a sensing range which defines the coverage area that can be sensed by a sensor node. One of the most important goals of a deployed WSN is to ensure the coverage of the monitored area. We can classify the coverage into the following categories. • The total coverage : In that case all the monitored area should be fully covered meaning that any point in the operation area should be in the coverage area of at least one sensor node. • The sparse or partial coverage : Some applications do not need a total coverage of the area but an efficient surveillance of some parts is adequate. • The redundant coverage : For that case, multiple sensors cover the same physical location. • The non homogeneous coverage : In that case, some interesting parts of the monitored area are more densely covered than the others.

1.4. THE CHAPTER RESEARCH 1. INTRODUCTION ISSUES FOR WIRELESS TO THE WIRELESS SENSOR NETWORKS SENSOR NETWORKS Despite the type of coverage, the deployment process should be performed to guarantee the needed coverage. A bad deployment strategy will generate a lack in the coverage, then the sensing goal of the WSN will not be ensured. For that reason, many research works have addressed the problem of coverage oriented deployment. 1.4.3.2. Connectivity In addition to the sensing range, every sensor node has a communication range. The communication range defines the communication area of the sensor node. The sensor node can communicate to any node belonging to its communication area. The network is said to be connected if there is a communication path between any pair of nodes. The pair nodes communication can be done either directly or over multiple hops throw other nodes. The connectivity is of a great importance because if a sensor node is isolated, it will not be able to report its data to the analysis center for processing. In that case, even if the deployment satisfies a full coverage of the monitored area, the final resulting coverage will not be sufficient because some nodes are isolated as if they do not exist. Due to the importance of the deployment process and its impact on the performances of the network deployed many research works have focused on proposing deployment schemes. The deployment strategies proposed aims at ensuring a set of requirements. In particular, the major considered goals to be addressed are coverage or connectivity or both of them.

1.4.4. Mobility of the sensors Consistent research has been introduced assuming the mobility of the sensors. In that case, the sensors are not static and can change their positions when needed. The mobility has been proposed in many research works as a solution to many situations and applications. Adding sensor networks with motion can exploit this surplus to enhance sensing while also improving the network’s lifetime and reliability. In literature, three types of mobility can be distinguished. • Node Mobility: The sensor nodes change their positions to enhance the quality of coverage or connectivity of the network. As a solution to the lack of coverage, one or more sensors move to the non covered portion of area. In some cases, the positions of the sensors may change unintentionally on deployment due to environmental factors such as wind or water. In the last case, the mobility will permit that sensors to get to the required positions to avoid non covered areas. • Event based mobility: In this mobility scheme, the sensors react to the changes or events in their environment. The sensor nodes move to the location of the detected event to ensure a better sensing in the event surroundings. For example, when a major event is detected in an area, additional sensors move to this location to ensure better coverage of the occurred event. • Sink mobility: The data gathered by the sensors is sent to the sink node directly or in multiple hops transmission, one can easily notice that the closest sensor nodes to the sink will deplete their battery power faster due to their heavy overhead of relaying messages. Recent research works have proposed that the sink node moves in the monitored area to reduce and balance energy expenditure among all the network sensors.

1.4. THE CHAPTER RESEARCH 1. INTRODUCTION ISSUES FOR WIRELESS TO THE WIRELESS SENSOR NETWORKS SENSOR NETWORKS The researchers has found in sensors mobility a practical solution to implement a dynamic network. The dynamicity of the network nodes gives the possibility of changing the network operating when needed to adapt the network to additional functional requirements such as for example new coverage or connectivity needs.

1.4.5. Power saving and scheduling In the general cases sensor nodes have poor energy resources. The power consumption is always a critical problem to be resolved for sensor nodes because the sensors may be deployed in a hard-to-reach location and then changing the battery regularly may be costly and for the several applications impossible. A rapid decay of the sensors residual energy reduces the lifetime of the network and then affects the availability of the sensing. Many research works have treated the problem of extending the lifetime of the network based on a control and control of sensor’s power saving. Two power saving policies are distinguished: • Dynamic Voltage Scaling (DVS): In a DVS scheme the power levels used for the sensors are not constant and depend on the non-deterministic workload. By the use of lower power levels when needed will permit a reduction in energy consumption and by consequence extend the lifetime of the network. • Dynamic Power Management (DPM): For the DVS scheme the sensors are always in operating modes but changes the power level used. DPM conserves power by shutting down parts of the sensor node which are not currently used or active. In this case, the sensors may be in idle or sleep mode to reduce the energy consumption. In other cases, some sensors may also be shutdown when needed which extends their lifetime. This solution is referred in the research work to as a scheduling scheme. A scheduling schemes alternate the sensors between sleep and active status for an optimal power consumption. Many research works have focused on proposing scheduling schemes suitable for the applications based on the WSNs.

1.4.6. Security It is imperative to protect the communications done over WSNs because the sensor nodes may communicate sensitive data and operate in hostile environments. However, due to the specific constraints of WSNs adapted security protocols are needed. Currently enormous research works propose security protocols and mechanisms suitable for WSNs context. The security requirements investigated by the researchers are presented in the followings. • Data confidentiality: Data confidentiality is most important security property to be ensured. Data confidentiality ensures that the data exchanged between the different nodes is kept secret and cannot be read by malicious and non authorized entities. Given that the data interacted in WSNs is sensitive the data confidentiality should be ensured. • Data integrity: With data confidentiality an intruder may be unable to steal information. However, it can change the data to prevent the nodes to read the correct content of the sent data. For example a malicious node may add some fragments or delete some parts of the sent data packet. A change in the content of the sent data may also occur due

1.5. THESIS CHAPTER STATEMENT 1. INTRODUCTION AND CONTRIBUTIONS TO THE WIRELESS SENSOR NETWORKS to the harsh communication environment or errors on the transmission channel. Thus, data integrity should be ensured to guarantee that the received data have not been altered in transit. • Authentication: In addition to a trial of reading or altering the data contents, an adversary can inject additional data packets that may affect the correctness of the reported events. The receiver needs to ensure that the data received originates from the claimed source. This security property is the authentication. The authentication is ensured when the system is able to detect the injection of data packets with forged identity source. • Data freshness: Given that the communication on WSNs is done over wireless links, an adversary can easily listen and take copies of some transmitted packets. Even if it cannot read the content of the packets due to the data confidentiality but it can store the packets and send them at ulterior moments. This replayed data packet will disrupt the normal work of the sensing network because false events will be reported and may cause false decisions. Data freshness suggests that the data is recent and ensures that no old messages have been replayed. • Self organization: The WSNs are ad hoc networks and their infrastructures is dynamic. For sensing and management reasons, some sensors may be added and others may change their locations. The dynamics of the whole network inhibits the idea of pre-installation of fixed security mechanisms like fixed shared keys between the sink and all the sensors. When the network infrastructure is not fixed, the security protocols should be able to self organize, adapt the security management and built updated trust relation among the sensors. We can easily notice that the presented research axes for WSNs are complementary and are dependent to ensure a global efficient sensing quality.

1.5. Thesis statement and contributions It comes from the previous sections that the performances of an implemented WSN is evolving towards challenging trends. The objective of this thesis is to provide solutions to resolve the most addressed challenges of WSNs in different contexts. Our contribution is four-fold.

1.5.1. Deployment techniques for large areas In the first axis we proposed two deployment techniques suitable for large monitored areas. The impact of radio irregularities on deployment We investigated the radio irregularity effects. We studied in a first part the impact of radio irregularities on the quality of sensors coverage. We proved that these factors degrade the sensing quality of the network deployed. Based on this analysis, we proposed in a second part a deployment technique that takes into account the radio irregularities to ensure coverage quality, even if these factors occurs. At the best of our knowledge, many research works have addressed the impact of radio irregularities on the coverage quality, but have not addressed or proposed a deployment technique based on radio irregularities.

1.5. THESIS CHAPTER STATEMENT 1. INTRODUCTION AND CONTRIBUTIONS TO THE WIRELESS SENSOR NETWORKS A geographic based deployment technique To better address this type of deployment technique, we consider the particular case of a military target tracking application. In the proposed deployment scheme, the density of deployed sensors depends on the geographic characteristics of the monitored area. Given that the monitored area is large, it can contain more than one geographical category. We will then perform a non-uniform deployment. A redeployment mechanism is also proposed in relation with the targets presence in the monitored area. Then the proposed deployment strategy is a non uniform, dynamic and geographical based strategy which was not addressed previously.

1.5.2. Scheduling solutions In the second axis, we proposed two scheduling solutions for WSNs to extend the lifetime of the deployed network. Energy based scheduling scheme One of the factors that affect the radio range of a sensor is the residual energy. Globally speaking, it can be said that the transmission power of a network device decreases with respect to time. We studied the effect of the power decay of the coverage range of the sensor and by consequence on the global sensing performance of the network. The existing deployment solutions suppose that the sensor radio range is constant and does not take into account its variability in relation to the sensors power. In a first part, we proposed a deployment technique that permits an efficient sensing for any battery power level. Based this deployment, we proposed a scheduling scheme that activates only the required number of sensors. Targets based scheduling scheme The second scheduling scheme presented is based on the analysis of the targets movements. We consider the particular case of a wireless sensor network for military target tracking. The principle of our solution is to predict the positions of the targets. This prediction is calculated relatively to the previously detected positions. Then, the scheduling will appropriately select the sensors that are likely to detect targets and orders them to wake up. The main contribution is that the prediction model reflects the characteristics of the sensed area and also provides an heterogenous coverage of the monitored area to provide a minimal coverage of all the area and the parts that are likely to contain targets are densely covered. The advantage of the proposed solution is that it ensures both a scheduling scheme and a dynamic distribution of sensors in the monitored area.

1.5.3. Secured distributed key management protocol In the third axis we proposed a dynamic security protocol suitable for WSNs. The main contribution of the proposed security protocol is the use of the tunnels in the WSN context. At the best of our knowledge, we are the first that proposed the use of tunnels in WSNs. The tunnels established are many-to-many tunnels to permit communication between many nodes. Any communicating nodes can build the many-to-many tunnels. We also introduced a new concept called the CSA (Cluster Security Association) which is an abstraction of the established many-to-many tunnels and represents shared security attributes between many sensor nodes. The solution permits authentication of the nodes, provides mechanisms for data confidentiality and integrity, and ensures self organization.

1.6. OUTLINE CHAPTEROF 1. DISSERTATION INTRODUCTION TO THE WIRELESS SENSOR NETWORKS

1.5.4. Border Surveillance applications One of the most recent applications of WSNs is the Border Surveillance application. The aim of the deployed WSN is to provide an efficient surveillance system to control suspect activities on borders In the fourth axis we propose two solutions based on WSN for Border surveillance. A Border surveillance framework using a thick linear architecture: In this part we propose a thick linear architecture that permits an efficient sensing and tracking of targets near a border line. We proposed a deployment strategy to ensure coverage, connectivity and routing efficiency. We also proposed a dynamic routing mechanism to ensure an efficient and optimal relay of data between the nodes and a self-organized elaboration of routes when the network topology changes. A dynamic deployment scheme for Border Surveillance In this part, we built a WSN based surveillance system that is able to provide a controllable surveillance of infiltration within a large area neighboring the border. In this deployment model, the sensors are supposed to be thrown from an airplane. The first contribution is the setup of a new deployment method which consists at paving the monitored area with paving patterns of predetermined shapes translating the environment conditions. The paving patterns are not randomly chosen but have predetermined shapes translating the environment conditions. We conducted detailed kinematic studies to characterize precisely the shapes of the paving patterns. In addition to the paving based deployment strategy, we proposed mathematical models that permit the compute of sensing and coverage probabilities. The mathematical models are built to provide a tight control on the quality of sensing and communication and evaluate the network sensing efficiency. In addition, the mathematical models are developed to plan and dimension the deployed network.

1.6. Outline of Dissertation The content of this thesis is organized as follows: • Chapter 2 is devoted to present a deployment technique of WSN for targets tracking in large monitored areas. In this chapter will be presented an analysis of the radio irregularities on the coverage performances of WSNs. Based on this analysis, we will propose a deployment technique that takes into account these factors. We will propose a second deployment technique. This technique provides for a non uniform repartition of the sensors in the monitored area. The non uniformity is determined in relation to the geographical class of the sub parts of the monitored and to the probability of intruders presence in each portion of the land. We also present a redeployment mechanism to enhance the quality of coverage in relation to the recorded events. • Chapter 3 addresses the power consumption control for WSNs. We present two developed scheduling solutions. In the first scheduling scheme, we present an analysis of the impact of power decay on the sensing performances of a deployed network. We present an amelioration of the deployment technique proposed in the second chapter to take into account these factors. Then based on the energy consumption of the sensing nodes we propose a scheduling scheme to provide an efficient control of the power and extend

1.6. OUTLINE CHAPTEROF 1. DISSERTATION INTRODUCTION TO THE WIRELESS SENSOR NETWORKS the network lifetime. We will then present the second developed scheduling solution. This solution is based on a thorough analysis of the targets movements and deduces the sensors that are likely to detect targets. • Chapter 4 takes a particular interest to the WSNs security tasks. We present a new key management protocol proposed for Wireless Sensor Networks that guarantees authentication, data confidentiality, and data integrity. We will present a new concept called the Cluster Security Association (CSA) and the tunnels establishment process. We also provide a mechanism of dynamic integration of newly deployed sensors. We then present a performance evaluation of the proposed protocol. • In Chapter 5, we present at first the border surveillance applications. We then present and detail the linear WSNs. A deployment method based on a thick linear hierarchical WSN and adapted for border surveillance will be presented in the same chapter. We present a routing technique and establishment of routes adapted to the network deployed. • In Chapter 6, we propose a network based on an aerial deployment of sensors and used in surveillance applications called DWBS (Distributed Wireless Border Surveillance System). We propose a new strategy of deployment based on a paving technique of the monitored area with paving patterns determined in relation to the aerial deployment conditions. In the same chapter, we will characterize the paving patterns where a sensor can land in relation to the environmental conditions. In this chapter, we consider the deterministic deployment case of DWBS. • In Chapter 7, we extend the work presented in Chapter 6. In this chapter, we will consider the controlled random deployment case of DWBS. In the same chapter mathematical models will be detailed to evaluate the DWBS connectivity and coverage performances. Some results of simulations will be presented to asses the efficiency of the proposed models. The same mathematical models may be used in dimensioning the deployed DWBS network. • Chapter 8 concludes the thesis and provides a set of perspectives which have been opened up by the achieved results.

Deployment strategies for large areas

2

2.1. Introduction to the chapter One of the most investigated research field for the wireless sensor networks is the deployment of the nodes to ensure the goals of the network implementation. The WSNs can be used in several applications ranging from patient monitoring to battle fields surveillance. A common factor for all these WSN use-cases is that monitoring a phenomenon encompasses coverage requirements to avoid missing valuable measurements. One important issue for being able to develop an efficient WSN is to have an optimal node placement strategy. The WSNs are used either in indoor applications where the goal of the network is to monitor a small area or outdoor applications where the area to be monitored is of large scale. In the first part of the thesis, we propose two strategies of coverage control suitable for large outdoor areas through efficient deployment of sensors. The characteristics of the deployment strategies are summarized in the following. • In the first strategy, we propose a solution to determine the required density for a uniform deployment of sensors to provide k-coverage in a large monitored area. The main contribution in this part is the determination of the radio irregularity impact in the deployment process; • For the second deployment strategy, we propose a non uniform deployment strategy in large areas. The network considered in this part has an heterogeneous architecture composed of several kinds of sensors. In addition to the non uniformity of sensors’ repartition, the main contribution in this deployment process is the integration of the sensors density repartition, as it depends on the geographical nature of the monitored area. The rest of the chapter is organized as follows. Section 2.2 will present the deployment strategy based on the radio irregularities. In this section we will present the existing works related to radio irregularity for the WSNs and present the details of the proposed coverage control strategy. Section 2.3 will present the results of the conducted simulations to compare the proposed deployment scheme with the deployment schemes that do not take into account the radio irregularity. In Section 2.4, we present the second deployment strategy based on geographical patterns of the monitored area. Section 2.5 assesses the efficiency of the geographical based deployment scheme through some conducted simulations. Finally, Section 2.6 concludes the chapter.

16

2.2. DEPLOYMENT CHAPTER STRATEGY 2. DEPLOYMENT BASED ON STRATEGIES RADIO IRREGULARITIES FOR LARGE AREAS

2.2. Deployment strategy based on radio irregularities Research in the field of Wireless Sensor Networks (WSNs) has been plagued by difficulties in performing realistic simulations. For instance, most of the existing coverage optimization techniques presuppose that the region covered by a sensor node is a disc characterized by the radio transmission range. This assumption is generally false because of the propagation phenomena including fading and shadowing. We investigate in this section the impact of the weakness of the aforementioned assumption on the performance of coverage control strategies in a WSN. We consider the log-normal shadowing model introduced in [1, 2] to represent the effect of the environmental features on the WSN performance. We then propose a deployment strategy supporting irregular radio propagation meaning that the sensor coverage area is not a disc. The output of this deployment method is the density of the sensors in the monitored area for uniform distribution considering the radio irregularities effects. Next, we go through theoretical and experimental assessments of radio irregularities on the fraction of area covered by the WSN. We also carry out simulations to evaluate the average distance that can be made by a target without being detected, called the Average Linear Uncovered Length (ALUL) [3] to assess deployment strategies in presence of radio irregularities.

2.2.1. Related Works The analysis of the radio effects on WSN performance have attracted significant research interest. In [4] J. Ma et al. presented an empirical study of radio signal strength in sensor networks. A series of experiments using MICA2 nodes in real environments is performed to investigate the parameters besides distance-frequency, variation of transceivers, antenna orientation, battery voltage, temporal-spatial properties of environment, and environmental dynamics. This work presents in details the impact of radio propagation; but, it does not address how the sensing is affected. This information is necessary to build a new deployment model for WSN. Zhou et al. [5] studied the behavior of MICA2 motes and concluded that the wireless links exhibit a random and unreliable behavior. With empirical data obtained from the MICA2 platform, they established a radio model for simulation, called the Radio Irregularity Model (RIM). It relies on a concept, called DOI (Degree of Irregularity), used to denote the irregularity of the radio pattern. It was originally defined as the maximum range variation per unit degree change in the direction of radio propagation. With this model, they analyzed the impact of radio irregularity on some of the well-known MAC and routing protocols. However, this model does not represent explicitly the probabilistic variation of the coverage range of the sensor in relation with the radio environment parameters. In [6], Woehrle et al. proposed a model for deployment coverage and connectivity for WSNs. The authors introduce a radio model derived from the DOI model [5]. They provide models for the nodes and environment for determining the sensor coverage, the sensing shape, and the connectivity conditions. They also present objectives and constraints for an efficient deployment. However, the authors do not represent a model which gives the number of the required sensors or the 2-dimensional coordinates of each sensor. In this work, the radio propagation model was only used to describe a more realistic operating mode of the sensors. In [7], several measurement campaigns are performed in three different scenarios. The main contribution of this work is that a two-slope log-normal path-loss near ground outdoor chan-

2.2. DEPLOYMENT CHAPTER STRATEGY 2. DEPLOYMENT BASED ON STRATEGIES RADIO IRREGULARITIES FOR LARGE AREAS nel model at 868 MHz is validated, and compared to the widely used one-slope model, which is the log-normal path loss model. The model is validated using three outdoor environments: a ground plain area, a university yard and a green park. This work gives a parameter adjustment model for the three scenarios. Nonetheless, these values are not deduced or resulted in an analytic model and cannot be applied in general. Therefore, it cannot be used to develop a deployment model because these values cannot be generalized to all the radio environment cases. In [8], the authors analyze the impact of radio irregularities and in particular the RIM model on typical localization algorithms. They conducted experiments to show that the radio irregularities has a significant impact on some main evaluation aspects of localization algorithms. Despite of their relevance, these researches only focus on the analysis of the radio irregularity impact but they do not propose deployment models that take into account these irregularities. In the following, we will study the impact of radio irregularities on the sensing quality of the network and propose a deployment model for coverage control supporting irregular radio propagation.

2.2.2. Irregular radio propagation Irregular radio propagation generally results from the intrinsic properties of the transmission device and the characteristics of the environments where the WSN is developed. Three major phenomena related to irregular radio propagation are discussed in the following: 1. Anisotropic transmission range: When a signal propagates within an environment, it is subjected to reflection, diffraction, and scattering. Reflection occurs when the signal encounters an object that is larger than the signal wavelength. Diffraction occurs when the signal encounters an irregular surface. Scattering occurs when the environment contains a large number of objects smaller than the signal wavelength. 2. Heterogeneous transmission power: Wireless network devices may transmit RF signals at different sensing powers, even though they are identical. This is mainly due to some random factors that occur during the manufacture of these devices. 3. Variance of transmission power according to time: After having operated for some time, the energy of the sensor nodes begins to deplete. Consequently, the intensity of the transmission signal is affected. Globally speaking, it can be said that the transmission power of a network device decreases with respect to time. It is noteworthy that, due to the non-uniform activity scheduling of the WSN nodes, the decreasing rate of the transmission range is not identical for all nodes. We rely on the log-normal shadowing model detailed in [9] to represent radio propagation in natural environments. In fact, both theoretical and measurement based propagation models indicate that average received signal power decreases logarithmically with distance. This model represents the path loss in relation with the distance by using path loss and shadowing components. The path loss is expressed by the following relation. P L(d)dB = P L(d0 ) − 10n log(

d ) + χσ, d0

(2.1)

2.2. DEPLOYMENT CHAPTER STRATEGY 2. DEPLOYMENT BASED ON STRATEGIES RADIO IRREGULARITIES FOR LARGE AREAS where d is the Transmitter-Receiver (T-R) separation distance, d0 is the distance associated with a reference measurement of the path loss, n is the path loss parameter standing for the rate at which the path loss increases with regard to distance, χσ denotes a zero mean, Gaussian random variable (in decibels) with standard deviation σ (also in decibels). Generally, the value of n depends on the specific propagation environment. It increases for environments that contain more obstacles. In addition, χσ is site and distance dependent. The received signal power can then be expressed by the following equation. P r(d)[dBm] = P t[dBm] − P L(d)[dB]

(2.2)

The log-normal distribution describes the shadowing effect which occurs over a large number of measurement locations that have the same T-R separation, but have different levels of clutter in the propagation path. This phenomenon is referred to as log-normal shadowing. Log normal shadowing implies that measured signals levels at a specific T-R separation have a Gaussian normal distribution. The close-in reference distance d0 , the path loss exponent n and the standard deviation σ, statistically describe the path loss model for an arbitrary location having a specific T-R separation. In the sequel, we use this model to determine the number of sensors to be deployed in the monitored region to reach a specific coverage degree.

2.2.3. A new coverage control strategy As it has been mentioned above, the proposed sensor deployment models do not take into account the effects of natural environments. These models are based only on the sensor range Rs . Nonetheless, in real contexts, due to the radio effects, every sensor does not have a uniform range distribution according to the propagation direction. We denote by Rsi the sensor range direction i (the value of i varies between 0 and 359).

Figure 2.1.: Radio irregularities impact on the surface of sensing An illustration of the surface covered by a sensor considering radio irregularities is depicted by Figure 2.1. Therefore, we propose a new deployment strategy taking into account the impact of radio effects. The goal of this deployment strategy is to determine the required number of sensors for an efficient detection in a large monitored area. In [10], the authors demonstrated that, when the coverage of a sensor S has a perfect circular form, k sensors should be present in the disc centered in S of radius Rs + Rt , where Rs is the sensor sensing range and Rt is the target radius, so that the k-coverage condition is fulfilled

2.2. DEPLOYMENT CHAPTER STRATEGY 2. DEPLOYMENT BASED ON STRATEGIES RADIO IRREGULARITIES FOR LARGE AREAS in the neighborhood of S. In other terms, the sensor density ρS is given by the following formula. ρS =

k π(Rs + Rt )2

(2.3)

In this work, we adapt this result by considering a disc of radius Rt and expanding this area with the respective Rsi in each direction. We will denote by Sc the surface of this area representing the coverage domain of the sensor. An example is depicted in Figure 2.2, where it is clear that the sensors S1 and S2 detect the target of radius Rt because their coverage surfaces intersect with the area Sc . However, sensor S3 does not cover the target. Based on this reasoning, it can be said that for the target to be k-covered, the area Sc must contain at least k sensors.

Figure 2.2.: Density requirements for target coverage Therefore, the required density is given by the following Equation. ρS =

k Sc

(2.4)

The next step consists in calculating the area of the surface Sc analyzing the statistical properties of the variation of Rsi . As it has been mentioned in the foregoing section, this variation is log-normal Gaussian distributed and characterized by the standard deviation σ, the mean range µ and the maximal range Rsmax . The corresponding probability density function is given by the following equation. 1 (Rs − µ)2 p(R) = √ exp(− ) 2σ 2 σ 2π

(2.5)

Consequently, the area of the surface Sc is then given by the following formula. Sc =

Z

Rs =Rsmax Rs =0

Z

θ=2Π∗pθ (Rs )

(Rt + Rs )dRs dθ, θ=0

(2.6)

2.3. PERFORMANCE EVALUATION OF THE RADIO IRREGULARITIES DEPLOYMENTCHAPTER 2. DEPLOYMENT STRATEGIES FOR LARGE AREAS where pθ (Rs ) is the probability that the range is equal to Rs . Then, we will calculate the integral of (Rt + Rs ) when varying θ between 0 and 2Π ∗ pθ (Rs ). This gives the area of the sector of radius Rt + Rs and angle 2Π ∗ pθ (Rs ).

2.3. Performance evaluation of the radio irregularities deployment In this section, we present the results of the conducted simulations. In the first simulation, we evaluate the deployment rule established in Section 2.2.3 with regard to traditional coverage control approaches. In the second simulation, we assess the influence of the radio propagation parameters on the performance of the proposed coverage control model. For the valuation metric of a given sensor density (i.e., number of sensors per unit of surface), we estimate the distance that can be made by a target without being detected, supposing that the sensor nodes are uniformly deployed. To this purpose, we measure the Average Linear Uncovered Length (ALUL), introduced in [3].

2.3.1. The details of ALUL metric The Average Linear Uncovered Length (ALUL) gives an approximation of the distance that can be made by a target before being detected by the sensor network. The undetected path length of a target traveling from location x with direction θ is given by the the Linear Uncovered Length (LU L), denoted by L(x, θ). The average of target paths before detection at location x and over all directions is the Average Linear Uncovered Length (ALU L(x)) which means the average distance can be traveled by a target at the location x without be detected. Obviously, if x is within the coverage of at least one sensor node then ALU L(x) equals 0. More generally, ALU L(x) is calculated as follows: ( 0 : x is covered (2.7) ALUL(x) ≡ R 2π L(x,θ)dθ 0 : otherwise 2π The ALUL in an area A, denoted by ALU L(A) is the mean uncovered distance that can be traveled by a target without being detected by any of the nodes deployed in the region of interest. The expression of ALU L(A) is given by: R ALU L(x)dx (2.8) ALUL(A) ≡ x∈A , kAk where k A k is the area of A.

2.3.2. Evaluation of the developed model For the simulation environment, we set: • the sensor range Rs to 50 meters; • the target radius Rt to 5 meters; • 3-coverage condition: we require that every target must be located by at least 3 sensors; • the region to be covered is supposed to have an area A = 1000 ∗ 1000 m2 .

2.3. PERFORMANCE EVALUATION OF THE RADIO IRREGULARITIES DEPLOYMENTCHAPTER 2. DEPLOYMENT STRATEGIES FOR LARGE AREAS Under these conditions, we compare the proposed coverage model, called Irregular (k;t)coverage, as based on Equation 2.4, to the traditional (k;t)-coverage based on Equation 2.3. Since this latter is based on an isotropic radio range, we consider the three following cases: 1. The trivial model: In this case we do not take into accounts the radio effects. Hence, the required sensor density depends only on the range Rs . In this simulation, we deployed the sensors uniformly in the monitored area while the sensors density repartition is given by Equation 2.3. As an evaluation metric, we measured the Averaged Linear Uncovered Length (ALUL). Since the length of the uncovered path varies substantially from a deployment configuration to another, we repeated the experiment (i.e., ALUL measurement) until the standard deviation of the ALUL process is less than a threshold, set to 2 m. The evaluation of the ALUL for the trivial model is illustrated by Figure 2.3.

Figure 2.3.: ALUL evaluation for the trivial model 2. The Log-normal shadowing model: The second simulation scenario adds the irregular radio propagation effects to the functioning of the sensor network. We deploy the same number of sensors as for the trivial model using Equation 2.3. But, when we compute the ALUL, we took into account the radio irregularities meaning that we will consider that the sensed area by a sensor is not a disc of range Rs but depends on the radio irregularities. The aim of this simulation is to study the impact of a deployment that does not take into account the radio propagation effects on the coverage quality of the network evaluated as ALUL. To describe the radio environment, we adopted the log-normal shadowing model. These effects are described by the values P L(d0 ), n and χσ. By assumption, P L(d0 ) is equal to 36 dB, n is equal to 4 and χσ has a variation of 4.70. The path loss threshold is equal to 100 dB. The evaluation of the ALUL for the Log-normal shadowing model is illustrated by Figure 2.4. 3. The averaged log normal shadowing model: The density of the uniformly deployed sensors in that case is computed using Equation 2.3. The range Rs used is considered as the

2.3. PERFORMANCE EVALUATION OF THE RADIO IRREGULARITIES DEPLOYMENTCHAPTER 2. DEPLOYMENT STRATEGIES FOR LARGE AREAS

Figure 2.4.: ALUL evaluation for the log-normal shadowing model average range computed as the mean of transmission ranges Rsi for all directions in the case of an irregular radio propagation. An example of the average range is illustrated by Figure 2.5 where the averaged range is plotted in dotted line.

Figure 2.5.: Variation of the coverage range The evaluation of the ALUL for the averaged log normal shadowing model is illustrated by Figure 2.6. We also conducted simulations to evaluate the ALUL value for the proposed deployment technique called Irregular (k; t)-coverage. In this scenario, either in the deployment process or in the computation of ALUL we considered that the area covered by a sensor is not a uniform disc but is function of the radio propagation effects. The results of these simulations are represented by Figure 2.7.

2.3. PERFORMANCE EVALUATION OF THE RADIO IRREGULARITIES DEPLOYMENTCHAPTER 2. DEPLOYMENT STRATEGIES FOR LARGE AREAS

Figure 2.6.: ALUL evaluation for the averaged log-normal shadowing model

Figure 2.7.: ALUL evaluation for the irregular (k,t) coverage model A first major remark is that, between the trivial and the log-normal shadowing model, the ALUL difference is 42% in average. This feature clearly highlights the limitations of traditional density computation approaches that rely on perfect radio propagation. Obviously, this reinforces the pertinence of our idea to develop a new coverage control model. Therefore, the impact of the radio irregularities is foreseeable. Moreover, comparing the first model to the third allows concluding that averaging the radio irregularity would be equivalent, in terms of performance, to the trivial model. Hence, despite its low computational complexity, the averaged log-normal shadowing model does not give satisfactory results. More importantly, it can be noticed that our model, the irregular (k,t)-coverage, returns the lowest ALUL values. More concretely, the consideration of the radio irregularities substantially enhances the ability of the WSN to detect malicious targets. These results are corroborated by Table 2.1 where

2.3. PERFORMANCE EVALUATION OF THE RADIO IRREGULARITIES DEPLOYMENTCHAPTER 2. DEPLOYMENT STRATEGIES FOR LARGE AREAS

The trivial model The Log normal shadowing model The averaged log normal shadowing model The scattering model

Minimum value of ALUL 36.5641 71.4310

Mean value of ALUL 40.1751 74.6685

Maximum value of ALUL 44.9030 78.2628

35.9547

38.5033

43.6484

20.9323

22.4365

24.9638

Table 2.1.: Results of the four models the maximum and minimum ALUL values are given for each of the four models.

2.3.3. Impact of the radio effect components In this subsection, we evaluate the influence of the radio parameters P L(d0), n, and χσ on the performance of our deployment model.

Figure 2.8.: Impact of P L(d0 ) on the ALUL

1. Measure of P L(d0 ) impact: We fix the variance of χσ to 4.70 and n to 4 and vary P L(d0 ) in the interval [30,43]. For every value of P L(d0 ) we measured the correspondent ALUL. These measures are represented in Figure 2.8. It is noticeable that the ALUL is between 31.8031 and 17.0461. Its standard deviation is equal to 5.4375. Therefore, P L(d0 ) has an important impact on the performances of the WSN. In fact, when P L(d0 ) varies from 30 to 43, the ALUL is enhanced by a factor of 0.42.

2.4. DEPLOYMENT CHAPTER BASED 2. ON DEPLOYMENT GEOGRAPHICAL STRATEGIES PATTERNS FOR LARGE AREAS 2. Measure of the impact of χσ: For fixed values of P L(d0 ) and n respectively equal to 36 dB and 4, we vary the variance of χσ in the interval [3.7,4.9] with a step of 0.2. The corresponding ALUL values are given in Figure 2.9. Clearly, the ALUL decreases with regard the variance of χσ. It takes values between 21.0915 and 23.9145. Consequently, we can deduce that the parameter χσ does not considerably affect the ability of the WSN to track malicious intruders.

Figure 2.9.: Impact of χσ on the ALUL 3. Measure of the impact of n: The variance of χσ is fixed to 4.70 and P L(d0 ) to 36. We varied n in the interval [3,5] with a step of 0.2. The evolution of the ALUL is illustrated by Figure 2.10. As for the other components, the ALUL decreases with respect to the path loss parameter n. It varies between 14.4368 and 123.7617. The standard deviation of ALUL is equal to 38.0549. This denotes an important impact on the performances of the WSN. In fact, a variation of 0.2 in the value of n leads to a substantial variation of the ALUL value. Compared to the two other components, the path loss parameter n has the greater effect on the ALUL value.

2.4. Deployment based on geographical patterns In the previous presented deployment method, we presented a method based on radio irregularities that calculates the number of the required sensors to monitor the presence of targets in a large area. The network architecture considered is a homogeneous network composed of a unique kind of sensors. We also opted for a uniform and static repartition of the sensors in the monitored area. In this section, we present the second deployment strategy developed for coverage control in large monitored areas. We introduce a heterogeneous WSN architecture suitable for intruder detecting systems. Then, we will present a deployment method based on this architecture to ensure a good quality of monitoring. The main contributions of the proposed network and deployment scheme are listed as follows:

2.4. DEPLOYMENT CHAPTER BASED 2. ON DEPLOYMENT GEOGRAPHICAL STRATEGIES PATTERNS FOR LARGE AREAS

Figure 2.10.: Impact of n on the ALUL • The deployment scheme adapts the number of needed sensors relatively to the characteristics of the covered area in the sense that the density of deployed sensors depends of the geographic characteristics and class of the monitored area; • We will perform a non-uniform deployment. We will proceed to a conditioned deployment for each portion of the monitored area depending on its geographic class; • We propose a redeployment mechanism based-on the presence of intruders in the monitored area. As function of the occurrence of the targets, we will update the different densities of sensors to enhance the coverage quality of the most vulnerable portions of the controlled zone.

2.4.1. Architectural issues In this section, we discuss the architectural aspects related to the implementation of the deployed WSN. The WSN that will be implemented is a heterogeneous WSNf. In this category, we have various types of sensors having different capabilities. Some sensors may have larger battery lifetime, better communication capability, and more powerful processing resources. In order to have an efficient target tracking, we need several types of sensors with different sensing, communication capabilities and functions. Therefore we consider an heterogenous WSN. The proposed network is three-layered; each layer consists of a type of nodes with a given role. Those nodes are: the Intruder Detecting Sensors, the Image nodes, and the Sink Nodes (or core nodes). The network architecture is depicted in Figure 2.11. In this figure, we represent a snapshot of a sub zone of the area to be monitored. 1. The sensing Layer consists of Intruder Detecting Sensors which are miniature devices whose role is the detection of intruders. After detecting trespassers, those nodes generate real-time alerts and transmit them to the closest Sink Node. In some cases, those nodes can relay the generated alerts through other sensors to the sink;

2.4. DEPLOYMENT CHAPTER BASED 2. ON DEPLOYMENT GEOGRAPHICAL STRATEGIES PATTERNS FOR LARGE AREAS

Figure 2.11.: A sub zone coverage 2. The image Layer contains the Image Sensors which are equipped with an image acquisition card. The main role of those sensors is to take an image snapshot of a given area. With those nodes, through a multimedia analysis of the detected snapshots we can deeply analyze the kind of the present object within the monitored area to check whether it is an intruder or a harmless object; 3. The Core Layer is composed of the Sink Nodes which are equipped with powerful sensing and transmission capabilities. Hereinafter, those nodes will be referred to as Sink Nodes. They are able to acquire and exchange voluminous high-resolution data related to the alerts generated by the low-level sensors. The other goal of implementing such nodes is to manage the other nodes by giving instructions to the sensors and ordering the Image Nodes to take and report snapshots when needed.

2.4.2. The Deployment Strategy As stated before, we have two categories of Sensing Nodes especially the Image nodes and the Sensing Nodes. In this section, we will develop the deployment strategy of those nodes to guarantee the coverage degree needed for a good sensing quality. 2.4.2.1. The Image Nodes Deployment As presented in the architectural issues, each image node will take snapshots of a part of the monitored area. Therefore, the coverage degree needed for those sensors is 1-coverage and they will be statically deployed. In [11], the authors shown that the best distance between √ two nodes is 3RCIm , where RCIm is the sensing range of the image sensors. In our case, we considered sensors needed to capture images, then the coverage range is the size of the captured image. We will deploy the sensors in a strip-based manner in the monitored area respecting the previously presented relationship between the coverage range and the distance between the sensors. An example of the Image Sensors is depicted by Figure 2.12. To determine the number of the required sensors and their placements, we will use the following steps:

2.4. DEPLOYMENT CHAPTER BASED 2. ON DEPLOYMENT GEOGRAPHICAL STRATEGIES PATTERNS FOR LARGE AREAS

Figure 2.12.: The Image Sensors Deployment • Considering the monitored area A(LEN GT H ∗ W IDT H), we will reduce it in all the sides by RCIm ; • We determine the number of the needed Image Sensors to cover the length of the area. CIm ) + 1; This number is given by the following formula: NLEN GT H = LEN GT√H−(2×R 3R CIm

• Accordance with the length, we determine the number of the Image Sensors needed to CIm ) + 1; cover the width of the area. The required number is NW IDT H = W IDT√H−(2×R 3R CIm

• The total number of needed Image Sensors is given by NIm = NLEN GT H ∗ NW IDT H . 2.4.2.2. The Intruder Detecting Sensors Deployment In this section, we will present the proposed deployment strategy for the Intruder Sensors. The Intruder Detecting Sensors are the ones who detect first the intrusions and report the alerts to the Sink Nodes. In our deployment strategy, we focused on the geographic nature of the monitored area, which can have an effect on the target presence probability. Then, our first idea is to establish a relationship between the geographic area and the degree of coverage. The Geographic Grid-Based Classification: Any area can be classified into many types such as mountains, roads, rivers, and forests. It is obvious that the probability of appearance of intruders differs between the geographic categories. The uniform deployment strategies do not take into consideration these characteristics. Therefore, our idea is to proceed to a nonuniform deployment that depends on the geographic nature of the monitored area. At a first step, we have as inputs a classification of the geographic categories which corresponds to each category the probability of appearance of the targets in this kind of geographic areas denoted by PGC . Based on those probabilities, we can determine for each geographic category GCi , a coverage degree Ki . The relationship between the probability PGCi and its corresponding coverage degree Ki is given by equation 2.9 where KM ax is the maximum quality of coverage needed for the most vulnerable areas.

2.4. DEPLOYMENT CHAPTER BASED 2. ON DEPLOYMENT GEOGRAPHICAL STRATEGIES PATTERNS FOR LARGE AREAS

PGCi (2.9) PGCM ax The following step of the deployment scheme is to subdivide the monitored area in a grid manner. Each sub-area corresponds to a geographic category and is Ki − covered. As a result, the sub-areas within the grid have different coverage degrees and therefore our monitored area is said to be non-uniformly K-covered. Ki = f (PGCi ) = KM ax ∗

The number of required Intruder Detecting Sensors: Once the coverage degree of each sub-area has been determined, we will determine the required number of sensor that will ensure this coverage quality. This number will be determined using the sensor density within the considered sub-area. The sensor density ρi is function of the sensor nodes sensing range denoted by Rs and the coverage degree Ki , as stated in the following formula: ρi =

Ki π ∗ Rs 2

(2.10)

The number of the required sensors is given by equation 2.11 where ABlock is the area of the block. NBlock = ρi ∗ ABlock (2.11)

2.4.3. The Re-Deployment In this section, we propose a redeployment technique to ensure that the sensing capabilities and coverage quality of the conceived network are as close as possible to the behavior of the intruders and the characteristics of the monitored zone. In fact, the geographical classification provides only an estimation of the probability of appearance of an intruder in the zone. But in reality, many blocks or sub areas of the same geographic class can have different probabilities of target appearance. Based on this assumption, we will use the statistics of the previously detected events to update the probability of appearance of targets in each block. Then, we will have a new value of coverage degree KIn depending not only on the geographic class but also on the real appearance of intruders in this block. Having the information about the number of the detected intruders in each block, we will elaborate a new value for each block called PIn which represents the new probability of detection of intruders. The expression of the probability PIni for the block i and is given by equation 2.12. number of detected intruders at block i PIni = (2.12) T otal number of detected intruders The relationship between KIn and PIn is given by equation 2.13. KIn = f (PIn ) = KM ax ∗

PIn PInM ax

(2.13)

Then, we will determine the new value KN ew of each block. This value can be an averaged value with the former value of Ki = f (PGC ) and the newly determined KIn = f (PIn ). The new density of coverage will be function of both geographic class and the real detection in each block, which reflects better the real context of the application. Table 2.2 illustrates the value of PIn and the new corresponding value KIn . It illustrates also the determination of the Knew coverage degree for each block. This value will increase, when compared with the

2.5. PERFORMANCE CHAPTER EVALUATION 2. DEPLOYMENT OF THESTRATEGIES GEOGRAPHICAL FOR DEPLOYMENT LARGE AREAS former Ki value, if many intruders has been detected in this block and will decrease if the number of the detected intruders is not high. At this step, we determined for each Blocki the new value of k-coverage denoted by KN ewi , Block Block1 .... Blocki .... BlockN

PIn PIn1 .... PIni .... PInN

The new KIn KIn1 = f (PIn1 ) .... KIni = f (PIni ) .... KInN = f (PInN )

Former Ki K1 .... Ki .... KN

KN ew KN ew1 = KIn12+K1 .... KN ewi = KIni2+Ki .... KN ewN = KInN2+KN

Table 2.2.: The new coverage required densities for the Re-deployment and we will determine the new number of needed sensors at each block. This value is given by equations 2.10 and 2.11. We should now update the number of the deployed sensors in each block: • If the previous number of sensors at the block is less than the newly required one, we have two possible actions; the first possible one is that the Sink Node will order some unused sensors in the other blocks to move to this block to reach the required coverage value. The second option is to deploy new sensors; • If the previous number of sensors is greater than the new required number, the extra sensors will be either placed in sleep mode or displaced to another block if needed. This redeployment is preformed periodically at regular intervals to satisfy at each time a deployment matching the actual behavior of the intruders.

2.5. Performance evaluation of the Geographical deployment This section is devoted to the performance evaluation of the developed deployment strategy. First of all, we will provide a description of the adopted simulation model. Then, we will introduce our proposed intrusion-detection metric in the frame of border surveillance WSNs named LUP (Length of Uncovered Path).

2.5.1. The Simulation Model We consider a WSN deployed in an area divided into sub-regions having the same size but each of which having its own coverage degree. In the simulations, we have considered only the Intrusion detecting sensors. As we have stated before, the values of the coverage degrees is dictated by the number of geographic categories that we have on hand. For the sake of simplicity, we suppose that three geographic categories define the monitored field and the coverage degrees required for these categories are k1 = 1, k2 = 2, and k3 = 3. The values of the coverage degrees attributed to each sub-areas are randomly generated. For all the simulations, the width of the monitored field is set to 300 m and the length is set to 900 m. The parameters that could be tuned within the monitored field are the dimensions the sub-areas which are both set to 150 m.

2.5. PERFORMANCE CHAPTER EVALUATION 2. DEPLOYMENT OF THESTRATEGIES GEOGRAPHICAL FOR DEPLOYMENT LARGE AREAS

2.5.2. The Simulation Scenario The objective of our simulation is to compute the percentage of intruders that can cross the network without being detected. The intruders will be initially randomly distributed in the monitored area. These intruders are supposed to be in motion in the monitored area. We will then periodically change the positions of the targets using mobility models. In literature [12], two mobility models are used to simulate the movements of targets which are respectively the random walk and the Gauss-Markov mobility model. The random Walk model supposes that the targets move in a random manner and have not an homogeneous behavior motion line. For the Gauss-Markov mobility model, the new positions of the targets are function of the previous recorded positions. This model simulates a target that has an homogeneous line of motion meaning that the targets are supposed to follow a logical path. Under the assumption that the targets crossing our WSN move according to the GaussMarkov mobility model, the main objective of the simulation is to assess the efficiency of the developed deployment strategy. This efficiency is measured by the probability of intrusion detection using a new metric called LUP (Length of Uncovered Path) which is inspired from the ALUL (Average Linear Uncovered Length). Details of LUP are provided in the following sub-section.

2.5.3. The Length of Uncovered Path (LUP) The ALUL measures the average distance that can be made by a target before being detected. The major limitation of this metric is that it supposes that the target moves according to a linear path. In order to cope with this limitation, we propose a new metric LUP which will consider paths that are made of several line segments like those generated by the GaussMarkov mobility model. We will not provide the mathematical details about the LUP, only algorithm aspects will be dealt with. Another novelty of our metric is that it is developed to work in k-covered areas and in areas that are non-uniformly k-covered as the case of our deployment strategy. The idea of our metric is to calculate the LUL (Linear Uncovered Length) of each line-segment forming the entire path. The following algorithm defines how to calculate the LUP of a given path. Algorithm 1 Linear Uncovered Path Input: P ath = Pi , ..., Pq : The set of points that define the path; Output: LUP (Length of the Uncovered Path) Description: 1 : LU P := 0; 2 : for(i = 1; i = q − 1; i + +) 3 : for each Line-segment[Pi , Pi+1 ] 4 : L := LU L(Pi , Pi+1 ); // Compute the Linear Uncovered Length of the line-segment defined by points Pi and Pi+1 5 : LU P := LU P + L; 6 : D:=Distance(Pi , Pi+1 ); // Compute the length of the line-segment 7 : if (D ≥ L) then break; // a portion of the line segment is uncovered 8 : return LU P ;

2.5. PERFORMANCE CHAPTER EVALUATION 2. DEPLOYMENT OF THESTRATEGIES GEOGRAPHICAL FOR DEPLOYMENT LARGE AREAS The idea of the above algorithm is to consider a given path and to compute the LUL (Linear Uncovered Length) of each of its line-segments. Initially, the value of LUP is set to 0. This means that the path is totally covered and no intrusion can occur. Then, in each iteration a line segment is considered and its LUL is computed and added to the previous value of LUP. Also, a test is made to check whether the LUL of the segment is less than its length, if the result is true the algorithm stops and this mean that we met the first covered point of the path. The above algorithm is not applied to paths whose origin if sufficiently covered, i.e the obtained coverage degree of the origin is equal or beyond the expected coverage degree. The LUP of such paths is null. The LUL metric that we adopted is similar to the one used in the ALUL, in fact our metric is an enhanced version of the proposed one and which takes into consideration the nonuniformity of the sensor node deployment. The LUL algorithm is detailed in the following. The main idea of the LUL algorithm is to look into the line segment and to find the first point that is sufficiently covered. A point is sufficiently covered if its coverage degree is equal or more than its expected coverage degree. The expected coverage degree is the coverage of the sub-area to which this point belongs. The actual coverage degree corresponds to the number of sensors that are actually covering this point. This algorithm is applied to segments whose origin is not sufficiently covered. In case where the origin is sufficiently covered the LUL is set to 0.

2.5.4. Simulation Results The results of the simulations will be discussed in this subsection. In a first time, we consider the simulation model introduced in the beginning of this Section and we calculate the intrusion percentage.

Figure 2.13.: Impact of the number of division per cell on the intrusion percentage The coordinates of the sensor nodes within a given sub-area are uniformly generated using the function unifrnd. The limitation of this function is that the obtained coordinates are not perfectly uniformly distributed when dealing with high dimensions such as our case (150 ∗ 150m). This will lead to an inefficient deployment which will negatively affect the intrusion detection of our network. In order to solve this problem, we propose a modification of our deployment strategy, in fact we will split each sub-area into small portions and generate the coordinates of the sensor contained in this portion. Through the simulations depicted by Figure 2.13, we

2.6. CONCLUSION CHAPTER OF THE 2. CHAPTER DEPLOYMENT STRATEGIES FOR LARGE AREAS proved that the more portions we have the better the detection percentage is. To reach such deployment, we suppose that the sensors are dropped from an aircraft which is going back and forth and for each sub-area not all the sensors are dropped at the same time but by small quantities. In the second simulations, we intend to check the impact of the mean direction of the targets on the intrusion percentage. The results of this simulation are depicted by the following figure.

Figure 2.14.: Impact of the mean direction of the targets on the intrusion percentage Figure 2.14 shows that whatever the mean directions chosen by the targets, the detection ability of our network remains high.

2.6. Conclusion of the chapter In this chapter, we proposed two deployment strategies for target and intruder surveillance using Wireless Sensor Networks. The proposed frameworks are adapted to large areas. We formally demonstrated that, depending on the intrinsic characteristics of the environment, irregular radio propagation can have an important influence on the performance of the WSN. Based on this analysis, we proposed a coverage control method that takes into account radio irregularities and provides a solution to determine the required number of uniformly distributed sensors in the monitored area. In the second deployment strategy, we proposed a heterogeneous WSN architecture to ensure several kinds of monitoring of the operation area. We developed a deployment solution to have a non-uniform deployment depending on the geographic nature of the monitored area. We also introduced a solution of sensors redeployment based on the probability of intrusions that have been recorded.

Scheduling Schemes for Wireless Sensor Networks

3

3.1. Introduction to the chapter Wireless Sensor Networks (WSNs) are being used to monitor events in an operation area. Many design tasks have to be addressed for an efficient sensing quality of the deployed networks. One of the most critical constraints is that the sensors have poor physical resources such as the energy. Due to that fact, the lifetime of the Wireless Sensor Networks was always a crucial issue to be addressed. For that many research works in the Wireless Sensor Networks have for goal increasing the lifetime of the network. One of the most proposed techniques in the literature are the scheduling solutions. The general concept of a scheduling method consists at placing only some sensors into ACTIVE status and the others into SLEEP status which extends the lifetime of the nodes and by consequence the lifetime of the WSN. This decision can be made in relation to many factors. In this chapter, we present two scheduling solutions that aim at extending the lifetime of the sensors and then the whole network lifetime. • The first scheduling scheme is based on a monitoring of the nodes energy. Based on their current energies, the sensors will be put in active or sleep status. In this scheme, we investigate the impact of the sensors energy decrease on the coverage quality and use this property as a decision factor in the scheduling solution; • The second scheduling method is based on a thorough analysis of the targets’ movements and their variations. Based on the estimated movements, the sensors will be put in sleep or active mode. The proposed scheduling method ensures a minimal coverage of the monitored area that permits the detection of any target present in this area. The rest of the chapter is organized as follows. In Section 3.2, we present a classification of the existing scheduling schemes. Section 3.3 introduces and describes the detailed steps of the energy based scheduling scheme. In Section 3.4, we present the results of simulations to evaluate the performances of the energy based scheduling scheme. Section 3.5 is devoted to the presentation of the scheduling scheme based on the targets mobility. The performances of this scheduling scheme will be presented in section 3.6. Finally, Section 3.7 concludes the chapter.

35

3.2. CHAPTER RELATED 3. SCHEDULING WORKS SCHEMES FOR WIRELESS SENSOR NETWORKS

3.2. Related Works In this section, we present some research works that focused on the sensors scheduling schemes. In those solutions, sensors alternate between being in active and sleep state. The sensor selection problem can be defined as follows: Given a set of sensors S = S1, ..., Sn, we need to determine the ”best subset” S of k sensors to satisfy the requirements of one or multiple missions. In [13], Kumar et al. adopt the Randomized Independent Scheduling (RIS) mechanism to extend network lifetime while achieving asymptotic K-coverage. In this work, the time is divided into cycles based on time synchronization. At the beginning of each cycle, each sensor independently of the others decides either to be active or passive node. The node decides to be in active state with probability p and in passive mode with probability 1 − p. This solution does not require neighborhood table neither a central managing node and has no communication overhead. But, because the sensors do not dynamically evaluate their situation, the algorithm is not robust against unexpected failure of nodes. It is probabilistic and can give some sub areas that are not covered. The selection of the activated nodes is random and does not reflect the characteristics of the targets movements or the monitored area. In addition, this scheduling solution does not take into account the variation of the sensing range. In another work, Perillo and Heinzelman in [14] divides the sensors into sets. Every set gives a complete coverage of the sensed area. At a time, only one subset is active. In this work, the authors proposed a solution to optimally schedule the sets. This is done by selecting which set is active, when and for how long. This solution ensures at any time the required coverage, but does not take into account the probability of the presence of the targets. Then, many sensors can be activated, where no targets are likely to be in their coverage zone. This solution supposes that the sensing range is constant and then may provide a lack in coverage while supposing that the number of sensors activated ensure full coverage. Cardei and Du [15] divide the sensors into disjoint sets. Unlike the previous solution the sets are scheduled using Round Robin. In this work, the focus is on the problem of finding the maximum number of disjoint sets. The main drawback of the proposed work, is that it is difficult to apply it in a distributed manner. Lu et al. [16] proposed a solution of sensor scheduling to provide k-coverage. At the initialization of the network, all the sensors are in passive state. Before being active, each sensor waits a back-off time relative to the amount of contribution they can provide to the coverage. Sensors are turned on one by one. The sensor with the highest contribution turns on first and so on. The contribution or coverage merit is computed based on the probability of detection of an event by that sensor. In the literature, an other category of scheduling solutions are given. These solutions are Grid based algorithms [17–22]. In those algorithms, the covered area is partitioned into rectangular or hexagonal grids. At any time, only one node is scheduled to be active in each grid. Then, for those solutions, the size of the grid have to be monitored by just one sensor. Then, those methods are also based on the round robin technique, but they aim at providing only 1-coverage. The contribution of our proposed protocol, is that it permits k-coverage and also the coverage density is dynamic and changes in function of the behavior of the targets. An other kind of scheduling solutions are the coverage based solutions. Those methods tries to select as little sensor nodes as possible. But those sensors have to ensure a full coverage of the area or same targets points. The other sensors are inactive. Then those solutions are also

3.3. CHAPTER A DEPLOYMENT 3. SCHEDULING AND ENERGY SCHEMES BASED FOR SLEEP WIRELESS SCHEDULING SENSOR NETWORKS SCHEME round robin schemes and focuses on the density of the required coverage. Those solutions can be classified into two categories. The first ones are centralized solutions, where a central sink node has the topology of the network and orders to the sensors to be either in active or sleep status [23–25]. The second ones are distributed approaches [26–32], where each node locally and periodically checks whether it has to be active or no to ensure the required k-coverage. The aim is then to ensure k-coverage of the targets or the area to be monitored. When compared to our solution, the distributed approaches are quite similar because it treats the k-coverage and the dynamic selection of the active sensors. But, the major contribution of our work is that the sensors to be active are selected as result of a thorough analysis of the targets movements.

3.3. A deployment and energy based sleep scheduling scheme The deployment strategies for wireless sensor networks suppose that the transmission range of the sensors is constant during all the network lifetime. In realistic cases the transmission power varies according to time. After having operated for some time, the energy of the sensor nodes begins to deplete. Consequently, the intensity of the transmission signal is affected. Globally speaking, it can be said that the transmission power of a network device decreases with respect to time. The deployment models proposed in literature only focus on the theoretical transmission range of the sensor nodes supposing that the sensor coverage range does not vary according to time, which is not true because the available energy decreases. To overcome these lacks, we will propose a modification of the deployment and coverage control model proposed in section 2.2 and [33]. The developed coverage control model takes into account the variation of the transmission range according to time in relation with the impact of the radio irregularities. Based on the proposed initial deployment strategy, we propose a scheduling solution based on the energy consumption of the sensing nodes.

3.3.1. The energy based coverage control strategy In this section, we will present the deployment process that takes into account the temporal non-stationarity of the coverage process. As mentioned above, the sensed area of a sensor node depends on three radio environment parameters which are respectively: (a) P L(d0 ) which represents the path loss at the reference distance d0 , (b) n is the path loss parameter and (c) χσ is a zero mean Gaussian random variable representing the fading effect on the transmission range. The parameters n and χσ, are site dependent and are not affected by the energy of the sensing node. The parameter P L(d0 ) is related to the transmission puissance of the sensing device which depends on the energy of the sensor. Given that the transmission power of a network device decreases with respect to time, then the value of P L(d0 ) becomes greater as well as the energy of the sensing nodes is consumed. Deploying the sensors using the strategy presented in Section 2.2.3 and considering the minimal value of P L(d0 ) (corresponding to the maximal initial sensor energy) will ensure initially an efficient coverage of the monitored area. But, as well as the energy of the sensing nodes decays, the covered area by the sensors becomes less and the number of deployed sensors will no longer be sufficient to ensure the required coverage of the monitored area.

3.3. CHAPTER A DEPLOYMENT 3. SCHEDULING AND ENERGY SCHEMES BASED FOR SLEEP WIRELESS SCHEDULING SENSOR NETWORKS SCHEME This lack should be studied in the initial step of deployment. To ensure that the conceived network ensures along all its lifetime the coverage of the monitored area, we should not consider the minimal value of P L(d0 ) where the sensor is fully charged. The density and number of sensors should be determined in the case where P L(d0 ) is equal to a threshold value denoted by P L(d0 )T hreshold . The value of P L(d0 )T hreshold represents the maximal value of path loss above which the sensor node cannot detect any target. Determining the number of sensors Ntotal relatively to the value P L(d0 )T hreshold will ensure that the deployed sensors will ensure the required coverage till reaching the sensor minimal energy. These sensors will be deployed in the monitored area using a poisson process.

3.3.2. The energy based Scheduling algorithm In the previous step, the number of deployed sensors Ntotal is determined considering the value P L(d0 )T hreshold . We notice that when the value of P L(d0 ) is less than P L(d0 )T hreshold , the number of needed sensors to cover the monitored area is less than Ntotal . Given that the deployment process provides more sensors than needed, we will profit of this deployment property to extend the lifetime of the network. This extension will be ensured using a scheduling scheme for the sensors. The proposed scheduling scheme will alternate the sensors between active and sleep status. At a particular instant the number of activated sensors is determined in function to the value of P L(d0 ) relative to the current sensors energy. Operating at this manner, only the needed sensors are in active status. We will then ensure that the sensors will not consume their energy when not needed and the lifetime of the network is extended. In the followings we detail the steps of the proposed sleep scheduling strategy. We suppose as hypothesis that the sensor’s level of power decreases after functioning along a time interval equal to τ . τ represents the discharges of the nodes. A sensor node is operational from its first deployment time at the moment 0 until the moment in which the value of P L(d0 ) reaches P L(d0 )T hreshold . • Step 1: In the first step of the scheduling scheme, we should determine at each instant t the required number Nt of sensors to be in active status. We denote by P L(d0 )t the value of P L(d0 ) at the instant t of the sensor node functioning. Relatively to the value P L(d0 )t , we can determine the number Nt of the sensors required to ensure the coverage of the monitored area. Initially we deployed Ntotal sensors which is ≥ to any value Nt . • Step 2: After determining the required number of sensors Nt , we will partition the Ntotal sensors into sets of Nt sensors. The Nt sensors should be selected while ensuring that they are distributed into sets by a poisson process. The number of sets NSets,t at time t are given by the following formula. NSets,t =

 Ntotal  Nt

(3.1)

• Step 3: In the previous step, we decomposed the sensors into NSets,t sets. These sets will not operate at the same time but will be scheduled using TDMA repartition

3.4. CHAPTER PERFORMANCE 3. SCHEDULING EVALUATION SCHEMES OF THE FOR ENERGY WIRELESS BASED SENSOR SCHEDULING NETWORKS meaning that at a duration time equal to τ only the senors belonging to one set are in active status and all the others in sleep status. As we presented previously τ is the slot of time in which the power of the battery decreases. When τ time elapses, the value of P L(d0 ) changes which necessitates a new decomposition into sets because the area covered by a sensor has changed. If the sensors were not divided into sets, all of them will operate at the same time and in one slot time τ all the sensors deployed will be discharged. Using the proposed scheduling strategy, the sensors are divided into NSets,t groups that operates separately in successive NSets,t time slots. The energy of all the sensors will decrease after NSets,t ∗τ instead of τ . Then each division into sets extends the life time of the networks by (NSets,t − 1) ∗ τ . The previous presented three steps are repeated at regular time intervals separated by τ . Those time instants represents a variation from the initial value of P L(d0 ) (at the deployment time before any discharge of the battery power) until we reach the last value of P L(d0 ). When P L(d0 ) reaches the value P L(d0 )T hreshold , all the Ntotal sensors are used and partitioned in a unique set of sensors. At every level of power, the sensors are assigned to sets synchronized using TDMA. A summary of the proposed scheduling algorithm is illustrated by algorithm 2. Algorithm 2 Algorithm of the energy based scheduling Ntotal =deploy(P L(d0 )T hreshold ) //determine the total number of deployed sensors Repeat Nt =deploy(P L(d  total  0 )t ) //determine the number of required sensors at the time moment t NSets,t = NN //determine the number of sets t for i=1 to NSets,t do setactive(Seti ) //sets in ACTIVE status the sensors belonging to the it h set end Until (P L(d0 )t =P L(d0 )T hreshold )

3.4. Performance evaluation of the energy based scheduling In this part, we conduct simulations to evaluate the performances of the proposed scheduling scheme. We will deploy sensors in a monitored area A using the deployment method proposed in [33] and compare the proposed scheduling scheme to two other deployment and scheduling scenarios. An efficient scheduling algorithm must validate some characteristics: • When scheduling the sensors, the active nodes must give a good and efficient quality of coverage without having congestion due to a big number of sensors. To evaluate the quality of coverage we measured the ALUL [3]; • The scheduling algorithm must prolong the lifetime of the network which is the first goal of a scheduling solution. As presented in the previous section, the sensor battery power decays in relation with time. When the power decays, the characteristic value of path loss at a reference distance increases, which means the coverage range of the sensor decreases. We consider that P L(d0 ) increases after one unit slot time τ . Table 3.1 represents the variation of P L(d0 ).

3.4. CHAPTER PERFORMANCE 3. SCHEDULING EVALUATION SCHEMES OF THE FOR ENERGY WIRELESS BASED SENSOR SCHEDULING NETWORKS

Table 3.1.: Decay of P L(d0 ) in relation with time Time slot 1 2 3 4 5 P L(d0 ) 58 65 70 75 80 For all the cases, the number of sensors is computed using the deployment method presented in [33]. This method considers the characteristics of radio environment such as P L(d0 ), n and χσ .

3.4.1. The Worst deployment without scheduling For this technique, the sensors are deployed considering the worst case for which the value of P L(d0 ) is equal to P L(d0 )T hreshold . In this case, we are sure that at the last decay of power, the sensors can maintain a good quality of coverage. For this method, the sensors are not divided into sets and there is no scheduling. This means that at any time all the sensors are in operation. The result of the simulation is shown in Figure 3.1.

Figure 3.1.: Worst deployment without scheduling This solution gives an excellent quality of coverage but its main demerit is that at most cases we have an excessive number of sensors. For example in the time period [0,τ ] all the sensors are used while only a small number of them can ensure the required coverage quality. In addition, this solution does not provide a long lifetime of the sensor network. For this example, the whole life time of the network is only 5 × τ , until we reach the last acceptable value of P L(d0 ) which is equal to P L(d0 )T hreshold .

3.4.2. The Round Robin Scheduling For the second simulation, the total number of sensors Ntotal is measured using the worst value of P L(d0 ) (P L(d0 )T hreshold ). Then, we will have the same number of total sensors deployed than the previous case. Unlike the previous method, we deployed the sensors into sets in the first step. The number of sensors required in each set is computed using the initial value of P L(d0 ) before any discharge of the sensor’s battery. These sets are scheduled using TDMA Round Robin method. The sensors are not dynamically assigned into new sets but the decomposition into sets is

3.4. CHAPTER PERFORMANCE 3. SCHEDULING EVALUATION SCHEMES OF THE FOR ENERGY WIRELESS BASED SENSOR SCHEDULING NETWORKS static and defined from the beginning of the network operation. The sensors are then filled into sets and remain in them till the end of the lifetime of the network which corresponds to complete discharge of the sensors. For the considered simulation case, the sensors are divided into 8 sets of sensors. The result of the simulation is represented by Figure 3.2.

Figure 3.2.: The Round Robin scheduling method As shown in Figure 3.2, the value of the ALUL increases in relation to time. This is logic, because at the first deployment, sensors are sufficient for the case where P L(d0 ) is equal to the first value. But when the time advances, the power of the sensors decreases and relatively the value of P L(d0 ) increases which means that the sensors’ area of coverage decreases. Due to the fact that the number of sensors into the sets is fixed, as well as the value of P L(d0 ) becomes greater, the density of the sensors will not be sufficient to give a good quality of coverage. The main merit of this method is that it enhances the lifetime of the network when compared to the lifetime of the previous case. But besides the first slot time, the values of ALUL are very far from the accepted values which gives a very bad quality of coverage. We can consider that the sensor network is not serviceable because having a long lifetime is not important if the network does not ensure its first goal which is a good quality of coverage.

3.4.3. The proposed scheduling solution In this part, we will conduct simulations to evaluate the proposed scheduling algorithm. The number of total sensors Ntotal is measured considering the worst case of coverage which corresponds to the maximum value P L(d0 )T hreshold . The total number Ntotal of deployed sensors is equal to the number of sensors used in the round robin scheduling scheme. As presented in the description of the algorithm proposed, the deployed sensors are assigned

3.4. CHAPTER PERFORMANCE 3. SCHEDULING EVALUATION SCHEMES OF THE FOR ENERGY WIRELESS BASED SENSOR SCHEDULING NETWORKS into disjoint sets in relation with the current value of P L(d0 ). This means that the number of sets is not already fixed like the round robin scheduling case. As mentioned in the specifications of the scheduling scheme, the number of sensor sets are calculated for each change of the value of P L(d0 ). This is done to conserve a good density of sensors in each step to ensure a good quality of coverage.

Figure 3.3.: Number of sets needed Figure 3.3 represents the number of sets corresponding to each value of P L(d0 ). The evaluation of the ALUL value using the proposed scheduling scheme is represented by Figure 3.4.

Figure 3.4.: Scheduling using the proposed model A recapitulation of the ALUL values is illustrated by table 3.2. As presented in Table 3.2 and Figure 3.4, the value of the ALUL varies between 5.93 and 23.80. These values are considered as accepted values and reflects a good quality of coverage which is quite similar to the results given by the first case and much better than the coverage given by the Round Robin scheduling algorithm. In addition to this merit, the proposed scheduling scheme gives a lifetime much better than

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY

Table 3.2.: Variation of ALUL Value of P L(d0 ) 58 dB 65 dB 70 dB 75 dB Number of sets 8 sets 4 sets 2 sets 1 set Mean ALUL value 22.54 17.03 10.32 5.9303

80 dB 1 set 11.2891

the first simulated solution. We can conclude that the proposed solution have a balance between the two requirements of a scheduling solution. In fact, it enhances the lifetime of the network while giving a good quality of coverage.

3.5. An optimized scheduling scheme based on target mobility In this part, we propose an other scheduling method to extend the lifetime of the network. We consider the particular case of a wireless sensor network for military target tracking. The major application considered is battlefield surveillance and tracking of military vehicles. The scheduling solution should extend the lifetime of the network while providing k-coverage of the sub-areas where the targets will be mostly present. For the first presented method, the status of the sensors is determined in function of the energy of the sensors but for this scheduling scheme the decision factor is the probability of targets presence in the monitored area. This scheduling method places only some sensors into active status and the others into sleep status. The principle of this scheduling solution, is to predict the positions of the targets. This prediction is calculated related to the previously detected positions and the characteristics of the target motion such as velocity and deviation. Then, the scheduling will appropriately select the sensors that are likely to wake up. The sensors that will detect targets are ordered to wake up and be in active mode and the others will be in sleep mode. Our solution is probabilistic because it is based on a probabilistic prediction of the targets movements. The solution is also dynamic because the functioning mode of the sensor (sleep or active) is not initially determined but depends of the the network state, especially the occurrence of target related events. The main contributions of the proposed scheduling scheme are listed in the followings: • The prediction model used takes into account the historic movements of the targets to a certain depth and then reflects the characteristics of the sensed area; • The scheduling scheme adapts the prediction model parameters relatively to the characteristics of the sensed targets behavior and the nature of the covered area in the sense that the variation of the movement characteristics are able to inform about the environment; • The solution presents a scheduling dynamic repartition of the sensors to ensure a better optimization of the sensors consumption. At a given instant, we will not try to ensure a fully k-coverage of the area. The sub areas where the targets will be present with high probabilities will be k-covered. The sub areas where the targets will be present but with less probabilities will be k-covered but for a small interval. The other areas in which no occurrence of targets is estimated will be 1-covered. Then a minimal coverage of the area will be given to permit both the optimization of the activated sensors and the detection of any target in any position of the monitored area.

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY

3.5.1. The architectural issues In this section, we discuss the architectural aspects related to the implementation of the scheduling solution. We will consider an heterogeneous network for military targets tracking using several nodes responsible of different goals. The network is composed of two layers: the core layer and the sensing layer. This network is represented by Figure 3.5.

Figure 3.5.: The Network Architecture • The core layer includes nodes which are equipped with powerful sensing and transmission capabilities. Hereinafter, these nodes will be referred to as core nodes. They are able to acquire and exchange voluminous high-resolution data and perform management tasks of the WSN. The scheduling task will be done by those sensors. They will in a first step analyze the behavior of the targets and establish a probabilistic prediction of the targets presence in the monitored area. Based on this prediction, they will be responsible of informing the sensors to be either in active or sleep state; • The sensing layer consists of elementary sensors, whose role is limited to the detection of the occurred events in the monitored area. The sensing nodes are monitored and controlled by the core nodes. The sensing nodes will be ordered by the core nodes to be either in active or sleep status. Remark. The sensing nodes are considered to be uniformly deployed in the monitored area providing k-coverage. Any deployment method can be used with the proposed scheduling scheme.

3.5.2. The proposed scheduling strategy In the following, we will present the scheduling strategy developed. When the core node receives the reports from the sensors, it can have an evaluation of the targets moving behavior. Based on the global behavior of the targets motion, the core node can anticipate the next positions of the targets. Having the locations of the sensors, the core node can determine at which time each sensor will detect a target. The core node establishes for each ulterior time a list of the sensors that will probably detect events. Based on these information, the core node will send the scheduling information to the sensors that will not have targets in their areas. This information indicates the time duration in which the sensor will be in sleep status

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY and when it should return to active status. For example, if the predicted movements of the detected targets indicate that a sensor Si will have a target in its coverage area at the time t, the core node sends a scheduling message to this sensor and orders him to be in sleep status and wake up at the time t.

Figure 3.6.: The Displacement Prediction An example is depicted by Figure 3.6 and illustrates the predicted positions of the target deduced from the detected positions of the target. The predicted positions indicate that the sensor Si will detect a target only in the time t5 , then the core node orders the sensor to be in sleep mode and wake up at the time t5 . In the literature to follow or predict the movements of a target we have two major strategies [12]. The first one is the random walk model, that supposes that the targets have not a behavior and can follow any random direction and velocity. The second method is the Gauss Markov Model which is more realistic because it determines the deduced movements of the targets in a uniform manner in relation with the previous movements. In our work, we selected the Gauss Markov model because the random walk model does not reflect a realistic behavior of the targets and can give false decisions to the scheduling scheme. We will use the Gauss Markov Mobility model in our scheduling scheme to deduce the movement line of each target depending of the previous detected movements and positions. This model relates the velocity vi (t0 + ∆t ) and the current direction θi (t0 + ∆t ) at the moment t0 + ∆t with the previous velocity vi (t0 ) and direction θi (t0 ) at the moment t0 . The following formula gives the new velocity vi (t0 + ∆t ) relatively to vi (t0 ). p vi (t0 + ∆t ) = a ∗ vi (t0 ) + (1 − a) ∗ v¯ + 1 − a2 ∗ Xv

(3.2)

The following formula gives the new direction θi (t0 +∆t ) relatively to the previous direction θi (t0 ). p θi (t0 + ∆t ) = θi (t0 ) + 1 − b2 ∗ Xθ , (3.3) where:

• a and b: are the tuning parameters ∈ [0..1]; • v¯: is the average velocity in the interval ∆t relatively to the velocity of the target;

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY • Xv : is a random variable of the Gaussian distribution; • Xθ : is a random variable of the Gaussian distribution. Then, the new positions are given by the following equations 3.4 and 3.5. Xi (t0 + ∆t ) = Xi (t0 ) + di (t0 + ∆t ) ∗ cos(θi (t0 + ∆t ))

(3.4)

Yi (t0 + ∆t ) = Yi (t0 ) + di (t0 + ∆t ) ∗ sin(θi (t0 + ∆t ))

(3.5)

di (t0 + ∆t ) = vi (t0 + ∆t ) ∗ ∆t

(3.6)

where: The Gauss Markov model gives a precise position of the target based on the previous averaged velocities and directions. But in reality this is not the practical case, because the target can be in an area concentrated around the position given by the Gauss Markov. In fact, the real values of velocity and direction can be superior or inferior to the averaged values deduced from the previous detections. We have then a set of possible positions and not only a precise position. Based on those facts we will modify the Gauss Markov prediction model to take into account these assumptions. In the proposed prediction method, the predicted position of a target is not a specific deterministic position but is a set of all the possible target positions (corresponding to the possible velocities and directions). In addition, these positions do not have the same probabilities due to the fact that some velocities and directions are more probable than others. Then, we affect a probability to each set of possible positions. To this end, the steps of the scheduling scheme are detailed in the following subsections. 3.5.2.1. The first Step In a first step we will consider all the previous values of the velocity and direction of the targets moving. Those values corresponds to the previous detections. Figure 3.7 and Figure 3.8 represents respectively an example of a registration of the velocity and direction variations.

Figure 3.7.: The variation of the velocity

3.5.2.2. The second Step In the previous step, we collected all the values of the velocity which varies in the interval [Vmin , Vmax ] where Vmin and Vmax are respectively the minimal and maximal recorded values of velocity. We will then divide this interval to smaller intervals ∆V which are respectively [Vmin , Vmin + ∆V ] , [Vmin + ∆V, Vmin + 2∆V ] ,...., [Vmax − ∆V, Vmax ]. Then, we determine

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY

Figure 3.8.: The variation of the direction the probability that the next velocity is in one of those intervals. The result of this step is a set of velocity intervals and a probability of each one of those intervals. The probability is based on the previous detections. We will determine the number of previously observed velocities for each velocity subinterval. Having the number of all the previous detections, we will determine the probability of being in this interval.

Figure 3.9.: The decomposition of the velocity into intervals Let us consider for example the case depicted by the Figure 3.9. The interval [V min, V max] is divided into three subintervals. For the first interval [Vmin , Vmin +∆V ] we have 5 previously recorded detections for which the velocity is in this interval. Given that the total number of detections is 10, the probability that the velocity will be in this interval is 0.5. The same calculations are done for the other two intervals. The table 3.3 represents the probability of each sub interval. The velocity subinterval [Vmin , Vmin + ∆V ] [Vmin + ∆V, Vmin + 2∆V ] [Vmin + 2∆V, Vmax ]

The number of previous observations 5 3 2

The probability of the interval 0,5 0,3 0,2

Table 3.3.: The probabilities of the Velocity intervals The same analysis is done on the previous detected directions of the targets, and we will have a set of possible intervals of directions. Each interval will have a probability of occurrence. The Figure 3.10 and table 3.4 illustrates the calculation of the directions probabilities.

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY

Figure 3.10.: The decomposition of the direction into intervals The direction subinterval [θmin , θmin + ∆θ] [θmin + ∆θ, θmin + 2∆θ] [θmin + 2∆θ, θmax ]

The number of previous observations 3 3 4

The probability of the interval 0,3 0,3 0,4

Table 3.4.: The probabilities of the direction intervals 3.5.2.3. The third Step In the previous step, we determined the sets of possible directions and velocities and assigned probabilities to each subinterval. The motion of a sensor is described using both velocity and direction. Then we will match every value of the velocity to the values of the directions to have all the couple sets matching the velocity and the direction. For each one of the sets we determine a probability which is equal to the multiplication of velocity and direction subintervals probabilities. The table 3.5 illustrates the probabilities of the couples of velocity and direction. [Vmin , Vmin + ∆V ] [θmin , θmin + ∆θ] [θmin + ∆θ, θmin + 2∆θ] [θmin + 2∆θ, θmax ]

p=0.15 p=0.15

[Vmin + ∆V, Vmin + 2∆V ] p=0.09 p=0.09

[Vmin + 2 ∗ ∆V, Vmax ] p=0.06 p=0.06

p=0.2

p=0.12

p=0.08

Table 3.5.: The areas probability

3.5.2.4. The fourth Step For the proposed scheduling scheme, we will divide the time into slot times denoted by ST . The slot times represent the periodicity of target positions prediction and consequently the decisions moments of the sensors status. In this step, for each subarea (couple of velocity and direction) determined in the previous step, the core node will determine which sensors ensures k-coverage of it. Now we have a set of subareas and for each one we correspond the list of the sensors that will cover it at a particular time t. Then, we will determine the averaged probability of target presence in the subareas. Based on the resulting probabilities, we will classify the subareas into three kinds.

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY • The subareas for which the probability is superior to the averaged probability. The core node indicates to the sensors that cover those areas to be active in the interval [t, t + ST ] of duration ST ; • The subareas for which the probability is less than the averaged probability. The apparition of targets in these subareas is low probable and then we propose that the sensors that covers this kind of subareas will not be active for the entire interval [t, t + ST ]. They will be active only in the interval [t, t + S2T ]. Operating at this manner, we will economize the power of those sensors because they will be active for a small interval. In the case where the sensors detect a target, they will be in active sensing mode; • Until now, we have considered only a little part of the area to be covered by the Wireless Sensor Network. We considered the areas that are probable considering the previous movements of the targets. But, in the monitored area, many new targets can appear at any time. Those targets were not considered in the estimation of the next movements because they were not previously detected. Those areas have to be considered to avoid the presence of black areas of coverage. So, for the areas that does not belong to one of the two previous subareas, we propose that those areas be 1-covered. If the sensor charged of monitoring that area has detected a target, the core node will in that case activate other sensors to ensure the required k-coverage. Managing the sensors using this scheduling scheme ensures that the monitored area will be covered non uniformly depending on the probability of the targets apparition in the subareas. The areas that are high probable will be always k-covered. For the areas less probable we will have an alternation in time between partial k-coverage and 1-coverage. And finally the areas that are Zero probable will be only 1-covered. In case of events apparition in a subarea, more sensors that cover this area will be ordered to shift to active state to ensure k-coverage.

3.5.3. Selecting the model parameters The scheduling solution presented is based on a set of parameters which are: • the slot time ST duration. As presented below, we will estimate the predicted positions of the targets at the beginning of each slot time. Also, we evaluate the previous values of direction and velocity depending on the behavior of the targets during a specific interval equal to a slot time. • the historic depth. As mentioned above we will predict the next movements of a target based on the previous detections (characterized by the direction and velocity). The historical depth means how many previous detections will be considered in the current prediction; • The ∆V is the size of the sub intervals of the [Vmin , Vmax ] interval; • The ∆Θ is the size of the sub intervals of the [Θ − min, Θmax ] interval; Having optimized results of the proposed solution is based on a good choice of the values of these parameters. In the followings we will present the importance of choosing a good value of the parameters. We will also indicate how the choice of these parameters values will be done.

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY 3.5.3.1. The choice of Slot Time The proposed solution is based on the previously detected movements of the targets. Those movements are evaluated at the end of each slot time. If the slot time is very small, the displacements of the target will not be significant when compared to the sensing ranges of the sensors. In this case, when considering the displacement from one slot to another, it is not certain that the target has moved from a sensor coverage area to another and then the scheduling has not a contribution. Considering a large value of the slot time will give a lack in the sensing quality and density. When considering a large value of the slot time, the target can cross more than one sensor coverage area in one slot time (an area larger than the area covered by the sensor). In this case, many sensors will not be considered and we will affect the quality of the coverage and tracking of the targets. Then we can deduce that the good value of the Slot Time to be chosen is related to the averaged velocity and the average distance realizing by a target when crossing a cell covered by a sensor. The slot time is equal to the averaged time required for a target to cross a sensor coverage area. Then the formula that gives the Slot time value is given by equation 3.7. ¯ D SlotT ime = ¯ , V

(3.7)

where: ¯ The averaged distance realized by a target when crossing a cell; • D: • V¯ : is an expected averaged velocity of a target. 3.5.3.2. The choice of the historic depth For the parameter historic depth, it is important to choose the adequate events used in the prediction model. In fact, every target has a certain behavior in its moving. If we consider all the movements of a target, we can use values of the direction or the velocity that are not homogenous with the current behavior of the target. In this case, those values can give a false estimation of the future position of the target and then a false decision about the sensors that will cover this target in the future slot times. • For the velocity, when looking in the previous recorded velocities, we stop when we find a velocity equal to Zero. In fact, a null velocity means that the target was not moving, then it began a new displacement. All the previous detections velocities will not be considered in the estimation of the target position; • For the direction, every target has a line of sight. When looking in the previous events, we will stop considering the detected events when the direction is not in the same line of sight of the other directions. The possible directions are varying in the interval [0,2*π]. Given a direction Θ, the directions in the same Line Of Sight of this direction are in the quarter centralized around Θ. The decision of the historical depth is presented in Algorithm 3 where CST is the current slot time.

3.5. CHAPTER AN OPTIMIZED 3. SCHEDULING SCHEDULING SCHEMES SCHEME FOR BASED WIRELESS ON TARGET SENSOR NETWORKS MOBILITY Algorithm 3 The historical depth for the direction i=CST Repeat i=i-1 ¯ = Average(Θ[CST − 1]..Θ[i]) Θ ¯ Until (i=Last Previous Event) OR (Θ(i) is not in the same LOS(Θ) ¯ If Θ[i] is not in the same LOS(Θ) Then Depth=[CST-1..i+1] else Depth=[CST-1..i] 3.5.3.3. The choice of ∆V and ∆Θ The values of ∆V and ∆Θ are of big importance in the proposed scheduling scheme. As represented in the algorithm specifications, the possible area of the target presence in the future will be divided into several small areas which are determined by those two values. In this section we will discuss the choice of the size of those areas. Lets denote this area A(∆V, ∆Θ). If the value of A(∆V, ∆Θ) is very small, we will have a lot of sub possible areas. When having a lot of possible areas, the probabilities of the sub areas will be the same and then all the nodes covering those sub areas will be classified as nodes of higher probabilities. Also, when having small areas, a sensor Si may cover two or more neighboring areas and then it will be affected in the same time to high and low probable areas, which is not the goal of the proposed scheduling method. An other case is a choose of a large value for the A(∆V, ∆Θ) parameter. In that case, we will have a small number of sub areas because the sub areas will be large in surface. This case is not also optimized because the sub areas will be large and it is possible that it contains partly an area which is very less probable compared to the global area. In this case low probable areas will be treated as high probable areas which is not the main goal of the proposed solution. Then we have to make a good choice of this parameter to ensure a good decomposition into sub areas that reflects different distinct behaviors of the targets in relation to the velocity and direction. Based on this analysis, the optimal value of the area A(∆V ,∆Θ) is equal to the area covered by a sensor. In that case, the sensor will not probably appear in more than one area and also we will not have in that area distinct zones with different probabilities. To determine the values of ∆V ,∆Θ we follow those steps: • The possible area of presence of the target is denoted AP and the area covered by a sensor is denoted AS . At a first step we determine the number of the sub areas which A is equal to N = ASp ; • To have N sub √ areas, the possible interval of the target velocity [Vmin ..Vmax ] will be divided into N sub intervals. Also, √ the possible interval of the target moving directions [Θmin ..Θmax ] will be divided into N sub intervals; • The value of ∆V is given by ∆V =

Vmax √−Vmin ; N

• The value of ∆Θ is given by ∆Θ =

Θmax √−Θmin . N

3.6. PERFORMANCE EVALUATION OF THE TARGETS MOBILITY BASED SCHEDULING CHAPTER 3. SCHEDULING SCHEMES FOR WIRELESS SENSOR NETWORKS

3.6. Performance evaluation of the targets mobility based scheduling In this section, we will first present the simulation model used to evaluate the performances of the proposed scheduling protocol. We will then represent the results of some conducted simulations. The aim of those simulations is to compare the proposed protocol to the TDMA Round Robin scheme used in many scheduling solutions. The two schemes will be compared in terms of the averaged number of activated sensors and the lifetime of the network.

3.6.1. The Simulation Model The characteristics of the simulation model are listed in the followings. • The sensors are deployed in area of 1000*1000 m2 . The required coverage quality is 3coverage. We considered the deployment solution proposed in [33]. We deployed initially three sets of sensors. Each set will operate in a slot time and ensures 3 coverage of the monitored area. • We considered sensors for which the sensing Range RS = 40meters. For the targets we considered targets for which the range is equal to 5 meters; • The lifetime of each sensor is equal to 5 days. The value of the slot time chosen is 15 minutes; • The targets to be detected by the sensors are deployed as events moving across the monitored area. We considered different numbers of targets to evaluate the performances of the proposed scheme in many operating cases.

3.6.2. The activated sensors In this first part of the simulations, we will evaluate the averaged number of the activated sensors at a slot time. We evaluated this metric for many values of the targets number. Figure 3.11 represents the results of the simulation. The number of activated sensors in a slot time is less when considering the proposed protocol. In fact, when considering the TDMA scheduling scheme, in a given slot time all the area to be monitored is fully 3 covered. But, for our solution the density of the coverage is variable within the whole area and depends at first on the movements of the targets. The other deduction, is that for our solution the averaged number of the activated sensors increases in relation with the number of the targets. In fact, as presented in the specifications of the solution, the number of the sub areas fully covered depends on the occurrence and presence of the targets in those areas. Then, as well as the targets are present as well as more sensors are needed to ensure additional fully covered areas. But, for the TDMA scheduling solution, the sensors are activated in a static predetermined manner to ensure a fully covered area independently of the number and position of the appeared targets. For those reasons, the averaged number of the activated sensors for the TDMA Round Robin scheme is constant. For the proposed scheme, the biggest number of the averaged activated sensors will be reached when considering a large number of the targets. In this worst case, we will have all the area

3.6. PERFORMANCE EVALUATION OF THE TARGETS MOBILITY BASED SCHEDULING CHAPTER 3. SCHEDULING SCHEMES FOR WIRELESS SENSOR NETWORKS

Figure 3.11.: The Averaged Number of Sensors fully covered because the targets are present every where. Then, in the worst case we will have the same performances than the TDMA scheduling scheme.

3.6.3. The lifetime of the network The second conducted simulation compares the proposed protocol and the TDMA Round Robin scheme. The comparison metric is the lifetime of the network. We considered different values of the targets number and for each value we evaluated the lifetime of the network. The results of this simulation are represented in Figure 3.12. The results of this simulation show that the proposed scheduling solution is better than the Round Robin scheme when considering the lifetime of the network. For all the values of the number of the targets, the lifetime of the network is greater for the proposed protocol. We remark that for the proposed protocol, the lifetime decreases when the number of the targets increases. In fact as well as more targets are present in the monitored area as well as the number of the activated sensors is greater. As a consequence, when the sensors are more used the lifetime of the network decreases. But for the TDMA scheduling method, the same number of sensors is activated at each slot and then the lifetime is constant independently of the targets and their apparitions. So, our solution is adapted to the behavior of the targets and permits a more optimized use of the sensors resources. We deduce also from this simulation that in the most charged use of the sensing nodes (apparition of many targets), the proposed protocol will give the same or better performances than the TDMA scheduling scheme. In fact, when the area is the most possible charged with targets, the amount of the needed activated sensors have to ensure a fully 3-covered area, which is the default coverage given by the TDMA scheme. Then in the worst cases, the proposed scheme will have the same performances given by the TDMA scheduling solution.

3.7. CHAPTER CONCLUSION 3. SCHEDULING OF THE CHAPTER SCHEMES FOR WIRELESS SENSOR NETWORKS

Figure 3.12.: The Lifetime of the Network

3.7. Conclusion of the chapter In this chapter, we presented two scheduling schemes to extend the lifetime of the deployed network. In the first scheme, we proposed at first a deployment scheme based on the variation of the sensing range of the sensors in relation with the sensors’ energy. We coupled the energy based deployment scheme with a scheduling scheme based on the energy consumption of the sensors. The number of activated sensors is updated when the energy of the sensors changes to ensure that only the needed sensors are activated. In the second part of this chapter, we proposed an other scheduling scheme based on the recorded targets movements. The scheduling scheme is based on an approximation of the next possible locations of the targets. Based on the deduced ulterior positions, the sensors deployed will be either in active or sleep status. The proposed scheduling scheme provides a non uniform repartition of the sensors in the monitored area.

DynTunKey: A Dynamic Distributed Group Key Tunneling Management Protocol

4

4.1. Introduction to the chapter Some applications in which are used the WSNs exchange sensitive and top secret data as for example the military applications. The scarcity of the computational, memory and energy resources, as well as the native vulnerabilities of the radio transmission protocols, increase the need for security services that protect the WSN-based applications. One of the most crucial requirements regarding the security of WSNs is authentication. This stems from the fact that radio links are, by nature, open and vulnerable to various identity spoofing attacks. Moreover, the confidentiality of the gathered data is often an important concern. Secure tunnels have been widely used in traditional wired and wireless networks to guarantee confidentiality and authentication. Unfortunately, the use of traditional tunneling protocols, mainly IPSec and SSL, is not suitable with the specific features of WSNs. Most of the research published in the literature has focused on the development of light cryptographic algorithms that comply with the sensor nodes capabilities. However, tunnel management protocols and the underlying key handling schemes have not been addressed. The objective is to present a solution that permits: • the authentication of the sensing nodes; • the formation of adhoc secure channels for every cluster; • the security of the exchanges between the nodes. To ensure those objectives we propose in this chapter a distributed and dynamic tunnel and group key management protocol for WSNs called DynTunKey. We introduce a tunneling approach that takes into account the characteristics of the cryptographic algorithms that are typically used for WSNs. The most important contributions of the solution presented are the listed in the following: • The proposed approach adapts to heterogeneous WSNs. In other terms, instead of using 1-to-1 tunnels, we rather build many-to-many tunnels. We will have a unique tunnel

55

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.1. INTRODUCTION TO THE CHAPTER TUNNELING MANAGEMENT PROTOCOL for many sensors communicating together. That is advantageous because we will ensure several communications between multiple nodes through the same security tunnel; • In our work, we introduced a new concept which is the CSA (Cluster Security Association). The CSA is an abstraction of the established many-to-many tunnels and represents shared security attributes between many sensor nodes. • In the proposed approach, the tunnel key used for the encryption of the communicated data will not have an infinite validity time and will be changed periodically. This characteristic is advantageous for the robustness of the security protocol and protects against nodes compromise; • In the proposed approach the key is not pre-deployed at the sensors, but is dynamically generated. The communicating sensors contribute in a secure manner in this generation. Then we avoid a full centralized management scheme; • The proposed protocol provides for a dynamic integration of new sensor at any time to the global architecture without compromising the security needed. A new appearing sensor is automatically inserted into the communicating nodes without the need of updates in the other nodes. The proposed protocol permits to all the sensors to exchange directly secured data. This is useful in many applications that necessitate the exchange of data between all the sensors. In particular: • Military target tracking application needs such communication. When using a group key each sensor can report its gathered data using the group key to the other sensors. This is advantageous and permits collaboration between the sensors without the need of central node; • Firefighting applications. In this case the sensors are deployed with the firefighter. If all the sensors have the same shared key, any one of the sensors can directly send data to other sensors. And then the collaboration between the members of the team will be easier by having an efficient dialog between the sensors that represent the firefighter. The rest of the chapter is organized as follows. In section 4.2, we will present some of the most important security solutions and key distribution schemes used for Wireless Sensor Networks. Section 4.3 describes the global security architecture that will be used to protect heterogeneous WSNs. The tunnel initialization phase is addressed in Section 4.4. A protocol for distributed negotiation and management of the Cluster Security Associations required to establish the many-to-many secure tunnels is introduced in Section 4.5. In Section 4.6, we will analyze the robustness of the proposed tunneling protocol. Section 4.7 presents the simulation model including the deployment of the sensors and the mobility models for the targets. In the same section, we present the results of some simulations to compare our protocol to recent key management solutions for Wireless Sensor Networks. The same section assesses the efficiency of the proposed protocol and compares it to the classical tunneling IPSec approach in terms of communication and processing overhead. Finally, Section 4.8 concludes the paper.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.2. RELATED WORKS TUNNELING MANAGEMENT PROTOCOL

4.2. Related works In this section, we present the most cited security solutions used for the Wireless Sensor Networks. We especially focus on the key management side of each protocol. For each solution, we will present the advantages and contributions of the proposed protocol when compared to this solution. The first kind of solutions proposes a straight pairwise private key sharing scheme between every pair of nodes [34–36]. At a deterministic manner, every node shares a key with every other sensor and the communications are done over 1-to-1 secured tunnels. When compared to the needed security requirements in the architecture considered, those solutions present the following lacks. • The first problem with that solution is that n-1 keys have to be stored at each sensor node, where n is the number of the nodes in the network. So a large memory space is used to store all the keys. For our solution, only one key is used between a set of sensors that belongs to the same group; • A compromise of a node, will compromise its communication for all the network, because this sensor node stores in its memory all the network keys. This is not the case in our solution because the key is always renewed; • The other major lack of this solution is that it does not permit a simple group communication. In fact, if a sensor needs to send data to many sensors, it have to do this in many messages. Many copies of the message have to be sent directly to the sensors using the pairwise shared key with each one of the sensors. This is impractical when the number of the sensors belonging to the group become larger. For our solution we have only one key shared with all the sensors. According to this, a node sends the data only one time using the shared group key; • An other problem with this solution is the add of a new sensor to the list of the communicating node. In fact, in those solutions all the keys are predeployed in the sensors. Then to add a new sensor we have to update the keys stored in all the sensors to add the key that will be used with the new sensor. For our solution, we will show that a new sensor is dynamically integrated in a secured communicating group. Other probabilistic solutions were presented and used for WSNs. Those solutions are qualified as Random Key Distribution Solutions. In the basic random key scheme [37] Eschenauer and Gligor introduced a probabilistic key predistribution scheme for sensor networks. This solution is based on three steps which are respectively: the key predistribution, the determination of the shared key phase and the path key establishment. In the key predistribution phase,initially a large pool of P keys are chosen and each sensor will be equipped with a key ring stored in its memory. The key ring is consisted of randomly chosen keys from the set P. Then the neighboring sensors will find which is the common key in their rings. This key will be used to secure the data sent between the two sensors. This is done in the shared key establishment phase. The last phase is the path-key establishment phase. In fact, this solution is probabilistic because it is not guaranteed that all the combination of the pair nodes shares a common key in their randomly chosen rings. Then a path-key have to be assigned for those sensor nodes through two or more links established at the end of the second phase. This solution when compared with the previous solutions necessitates less amount of memory in

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.2. RELATED WORKS TUNNELING MANAGEMENT PROTOCOL the sensors because each sensor does not store a key with all the other sensors. Inspired by this work, additional random key predistribution schemes have been proposed in [38–43]. The main addition of those works is to increase the resilience of the network against node capture and ensure a smaller need for communication intermediate paths. Those solutions also optimizes the required operation time and the number of the stored keys. But, despite all the added techniques, it is always a probabilistic solution. The major lacks of those solutions for our context are presented in the following. • such as the previous kind of solutions, those ones stores a lot amount of keys in the memory of each sensor; • this solution does not permit direct group communication between sensors, because the links established are 1-to-1 links. Also those links are not directly established because it is impossible to find a shared key between all the pairs of nodes. Then if a sensor needs to send data to a group of nodes, it has to do it in a separate manner for each sensor. An other category of the keys generation solutions are the centralized key management schemes. In those schemes a central node called the KDC (Key Distribution Center) controls and generates the keys used by the sensors. One of the protocol functioning at this manner is the LKHW protocol proposed in [44]. In this scheme, the Core Node is treated as a KDC and all keys are logically distributed in a tree rooted at the base station. • In this solution the sensors do not contribute in the elaboration of the keys. Then a compromise of the central node compromises all the network chain; • An other lack in this solution is that the keys are distributed in a tree manner. Then to have a communication between a set of nodes, we do not certainly have a direct secure link between them. Then a group communication is difficult to be proceeded in this schemes because it will be done as many separate secured connections in a tree communication manner; • Having keys distributed in a tree manner does not facilitate the regeneration of the keys and the integration of a new sensor node in the communicating trusted group of sensors. In PIKE [45], Chan and Perrig propose a solution that is not a fully centralized solution. The basic idea in PIKE is to use sensor nodes as trusted intermediaries to establish shared keys between nodes. In this solution, they proposed that the key will be established between two sensors through a common trusted third node somewhere within the sensor network. For this solution initial keys are distributed such as for any two nodes A and B, there is a node C that shares a key with both A and B. Therefore, the key establishment protocol between A and B can be securely routed through C. In this solution, the establishment of the key is secured and the number of initially deployed keys at the sensor is less than the previous solutions. But, it is not suitable for group communication, because its lower probable that all the nodes of the same group have a common trusted node. The same lacks of the previous categories of solutions are present with this kind of solution. The LEAP protocol described by Zhu et al. [46] takes an approach that utilizes multiple keying mechanisms. In this scheme four kinds of messages are established between the different types of sensors.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.2. RELATED WORKS TUNNELING MANAGEMENT PROTOCOL • An individual key shared with the core node (predistributed); • A group key that is shared by all the nodes in the network (predistributed); • Pairwise keys shared with immediate neighboring nodes; • A cluster key shared with multiple neighboring nodes. LEAP protocol permits several kinds of communications depending on the needed communicating nodes. This solution provides a many-to-many tunnel protocol like our proposed solution. But when compared with our solution, it has some lacks: • The number of the deployed keys at the sensors is large since every pair of sensor nodes needs a key; • The keys used for several kinds of communication are predeployed into the sensors. This solution uses a static key and does not propose a dynamic generation of the key. For the solution we proposed, the key is renewed after a validity interval; • In this solution the keys used for cluster communication are predefined. Then this solution does not permit a dynamic belonging to the groups. When a sensor needs to change from a group, the key stored in its memory have to be updated and replaced by the cluster key of the new group. That solution is not practical because it needs direct static intervention with each group change. For the proposed solution, the group key is regenerated automatically at periodic times. Then if a sensor changes from a group it is automatically integrated in the secured new group when it contributes to the elaboration of the group key. The protocol NSKM presented in [47], is a protocol that manages different kinds of keys such as the LEAP protocol [46]. The difference is that the cluster keys are calculated by every node within particular cluster. Despite this change, the main lacks of the LEAP protocol are the lacks of this protocol, in particular the absence of a rekeying solution. In [48], the authors propose a solution called RDKM(Real-Time Dynamic key Management). The main feature of this solution is that it establishes a real-time rekey mechanism based on the search-triggered splay tree architecture. It designs and realizes the rekey mechanism based on the splay tree, which can provide random function to generate new keys and make the dynamic key management feasible. In this solution, the cluster heads organize the keys of their members into a splay tree-architecture key pool. The cluster head shares with each one of the member nodes belonging to its cluster a pairwise key. Those keys are established through messages shared between the sensors and the corresponding cluster heads. This solution presents an efficient storage and rekeying solution, but it does not ensure direct group communication between several sensors because the sensors do not share a unique cluster key. The Table 4.1 illustrates the comparison between the proposed solution and the previously cited solutions.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.2. RELATED WORKS TUNNELING MANAGEMENT PROTOCOL Hierarchy Keys per Sensor Probabilistic Pairwise Keys solutions [34–36] random Secured key predis- paths through tribution trusted schemes sensors [37–43] LKHW Tree based key [44] management

Distribution Number of mechanism of keys

Group

static predeployed

one key with every sensor

Many tunnels per group

No

static predeployed

One ring pool of keys (selected randomly)

Many tunnels per group

No

static predeployed

Many tunnels per group

No

PIKE [45]

Secured paths through trusted sensors

static predeployed

Many tunnels per group

No

KM [48]

Tree based key management

computed through exchanges

Many tunnels per group

On demand (one sensor departure or attack on a secure channel)

LEAP [46] and NSKM [47]

many kinds of keys (pairwise and group)

static or computed on predeployed

One tunnel per group

No

The proposed protocol DynTunKey

One group key

computed on exchanges

One key with the superior node in the tree One ring pool of keys (equal to the number of needed trusted nodes) One key for initial exchanges, pairwise keys and one cluster key in each sensor One key for group communication + One key with every sensor One key for initial exchanges and one group key in each sensor

One tunnel per group

Periodic

Protocol

Regeneration

Communication of the keys

Table 4.1.: Comparison of Key Management Solutions

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.3. ARCHITECTURAL ISSUES TUNNELING MANAGEMENT PROTOCOL

4.3. Architectural issues In this section, we discuss the architectural aspects related to the implementation of encrypted tunnels on a WSN infrastructure. We first emphasize the need for using protected tunnels in the particular context of heterogeneous WSNs. Then, we provide a global overview on the proposed distributed security approach. Finally, we present the communication exchanges.

4.3.1. Need for encrypted tunnels in WSNs The proposed security protocol DynTunKey is adapted to an heterogeneous wireless sensor network. The architecture considered is two layered composed of the core nodes and the sensing nodes. The architecture of the network considered is similar to the architecture presented in Section 3.5.1 and represented by Figure 3.5. • For the proposed security protocol DynTunKey, in addition to its general functionalities, the core node will be responsible of the management and the establishment of the group key and tunnels; • In addition to their sensing and reporting rules, the sensing nodes will contribute in the construction of the tunnels and the group key. In hostile scenarios, relaying of critical data must be secure. Since data would be relayed through many nodes, care must be taken to ensure that the data aggregated at intermediate node is not corrupted. When receiving an alert message, the core node should also accurately verify the identity of the originating sensor node. In fact, the adversary can deploy sensor nodes that can deliver false information to the analysis center. Moreover, the legitimate sensor nodes are prone to be corrupted, because of weak physical protection, so as to be under the control of the enemy. Therefore, it appears that authentication and confidentiality are among the most crucial security properties that should be fulfilled when implementing an hierarchical infrastructure. Encrypted tunnels constitute a promising alternative to address these needs since they have been widely used in many contexts in traditional networks.

4.3.2. Proposed security architecture At the foundation of our approach is the idea of assembling the verification operations of the alert messages originating from multiple sensor nodes to a unique verification step. Based on this reasoning, we define a new requirement called k-security [49]. A WSN is called k-secure if, and only if, the following properties hold: 1. Every sensor node si possesses a private key denoted by κi . 2. A unique public key π and a subsequent algorithm can be used to verify whether k signatures of the same message generated by distinct sensors are valid or not. The key characteristics of our work are listed in the following: 1. The proposed protocol DynTunKey share a unique group key between a set of sensors that belong to the same geographic zone. Then all the sensors can exchange data between them in a secure and simple manner. Every sensor can send secured data either to the Core Node or to the other sensors, because the group key is known by all the types of sensors;

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.3. ARCHITECTURAL ISSUES TUNNELING MANAGEMENT PROTOCOL 2. The proposed key generation method includes multiple phases that can be organized according to the context in which the WSN is implemented. To implement this functionality, we consider the group Diffie-Hellman key exchange protocol introduced in [50]; 3. We will no longer use asymmetric encryption because the exchanged tunnel group key is a symmetric key. This is advantageous because the use of the symmetric encryption needs less resources than asymmetric encryption. It is then well adapted to the context of the Wireless Sensor Networks; The basic steps of the proposed security scheme are given below: 1. The core node periodically sends messages to the sensor nodes asking for new information. The sensors that have gathered new information will send reply messages; 2. The core node builds sensor clusters based on the location of the detected events, in the sense that a cluster will include the sensor nodes that have detected the same event; 3. The CSA is set up for every cluster. The tunnel establishment process is authenticated using threshold public key cryptography. The nodes of the cluster share the same symmetric group key using group Diffie-Hellman key exchange.

4.3.3. Communication exchanges For our work, many messages will be exchanged between the Core Node and the sensors. We have two major kinds of messages. The first ones are the messages required to establish the tunnels. The second kind of messages is used to report the events detected by the sensors. The two kinds of messages are sent periodically. We divided the time into slots. At the end of each slot, the sensors send the gathered data to the Core node. This reported information is encrypted using the symmetric key shared and established between the core node and all the sensors. The group symmetric key used has a validity interval which is equal to N time slots. In this manner we will not generate a group key in each reporting slot but we will generate a key used in many time slots. This aims to decrease the time required for the establishment of the keys. This is illustrated by Figure 4.1.

Figure 4.1.: The communication exchanges In the following sections, we describe the protocols that we have designed to support the implementation of the aforementioned security process.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.4. INITIAL AND AUTHENTICATION TUNNELING EXCHANGES MANAGEMENT PROTOCOL

4.4. Initial and authentication exchanges The communication begins with initial exchanges. This part is composed of five steps and three types of exchanged messages. At the end of those steps a symmetric key for the group is calculated. For each tunnel, all the nodes contribute to the calculation of this Group Key. This key is denoted GKDH (Group Key Diffie Hellman). When all those messages have been exchanged, the sensors and the Core Node are authenticated using their private keys. The method used to establish this key is derived from the Diffie Hellman key agreement protocol [51] and the work of D. Augot et al. [50] which present a method to generate a key for a group. Initially, all the nodes agree on a cyclic group G and on two numbers p and g, where p is a prime number and g is a primitive root modulo p. The nodes have also a pre deployed value S which is secret and known only by the trusted sensors. • Step 1: At this step, the first message INIT is sent by the Core Node to all the nodes in its coverage area to announce the beginning of an initial exchange phase. This message is sent periodically. By the nature of the Wireless sensor Network, the message is broadcasted to all the senor nodes. {Type of message , INIT} Moreover, the Core Node picks a random natural number Rc. This number is the contribution of the Core Node in the group key generated. • Step 2: The sensing nodes will participate to the construction of the CSA. When receiving the INIT message, every sensor node Si picks a random integer Ri. Then, the sensor node calculates g S∗Ri , joins its Identifier to this value and sends the whole message to the Core Node. In this message, the sensor also sends its digital signature. The field AUTH is a digest of the message and is signed by the sensor using its private key KSi. This ensures the integrity and the authentication of the sensor node. {Type of message,[Identifier,gs∗Ri ]},{AUTH}KSi • Step 3: At this step, the Core Node has received the contribution of all the sensors. For each message, it verifies the integrity and the authenticity of the originating sensor node using the common public key π. When verifying the identity of the sensor nodes, only the trusted sensors will contribute to the elaboration of the Group Key. If an intruder tries to send a contribution, it will not be authenticated and then its contribution will not be considered and will be rejected from the group. Based on the identifiers of the sensors and relatively to their deployment locations, the Core Node will organize the sensors into groups. Each group is the set of the sensors that detect the same event in the same zone. For each group, a common tunnel and a group key will be calculated. • Step 4: At this step, the core node has received the contribution of all the sensors and classified them into groups. If the Core Node classified the sensors into N groups,

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.4. INITIAL AND AUTHENTICATION TUNNELING EXCHANGES MANAGEMENT PROTOCOL then the core node will perform the following tasks for each group in a separated manner. For each value received g S∗Ri , the Core Node computes the resulting value (g S∗Ri )Rc = g S∗Ri∗Rc . The core node has picked the number Rc in the first step. Then, the Core Node sends to all the sensors belonging to the same group those values with the identifiers of the sensors. The message sent is represented below. {Type of message,[Identifier of S1, gS∗R1∗Rc ],..., [Identifier of Si, gS∗Ri∗Rc ],..., [Identifier of Sn, gS∗Rn∗Rc ]} , {AUTH}KCN The payload AUTH is the digest of the message. This digest is encrypted using the private key of the Core Node KCN. The payload AUTH and the signature ensure both integrity of the message and authentication of the core node. • Step 5: At this step of the exchanges, each sensor can calculate the Group Key for the cluster to which it belongs. When receiving the previous presented message, each sensor verifies the authentication to check if the message is sent by the Core Node or by an intruder, this is done using the public key of the core node. The sensor checks also the integrity of the message to be sure that it has not been modified. After those checks, every sensor node looks for its identifier and extracts the calculated value corresponding to it. For example, the sensor Si will consider the value g S∗Ri∗Rc and removes its secret value Ri from it to obtain g S∗Rc . Then it calculates the Group Key which is done by the following equation. GKDH = g S∗Rc ∗ (

Q

i∈M 6=(CN ) g

S∗Ri∗Rc )

= g S∗Rc(1+R1+..+Ri+..+Rn)

This operation is performed by every sensor and then all the sensors have the same shared key. The Core Node also calculates the group key GKDH which will be the session key for that Group of the sensors. Now, the initial exchange has finished. At the end of those exchanges, we have two major results. First, those messages authenticate the Core Node and the sensors and then only the trust nodes will participate in the establishment of the secure channels. Secondly, after those exchanges the nodes have shared a secret Group key that will be used in further communications and sessions. Those initial exchanges are illustrated by Figure 4.2. In this figure, the sensors S1, S2 and S3 have detected the same events but the sensor S4 does not detect any event. When receiving the Init message, the sensor S4 does not send a response but the other sensors exchange all the messages and complete all the steps from step 1 to step 5 to establish the Group Key.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.5. CLUSTER SA NEGOTIATION EXCHANGES TUNNELING MANAGEMENT PROTOCOL

Figure 4.2.: The Group Key establishment

4.5. Cluster SA negotiation exchanges 4.5.1. Sensor clustering Now that all the sensors have established the group key, a second phase is performed by the Core Node to identify the sensors that will participate to the CSA and their cryptographic preferences. This part is decomposed into two messages exchanged between the Core Node and the sensors. Those steps are performed for each group of sensors. • Step 1: At a first step, the Core Node sends a message to all the sensors. In this message, the Core Node demands to the sensors to choose their preferences for the CSA. It sends the cryptographic suites supported by it. The message sent is in the following format. {Type of message,{Mid, SAi}GKDH },{digest}GKDH Type of message indicates the type of the sent message. For this case, the node who receives the messages detects that it is request to choose a cryptographic suite. Mid is the message identifier. SAi states the cryptographic algorithms supported by the Core Node for encryption and signature. – The previous two payloads are encrypted using the previously negotiated Key GKDH. This ensures the confidentiality of the transmitted data. digest is a digest of the global message and is encrypted using the negotiated key GKDH. – This payload ensures the integrity and authentication of the sent data. • Step 2: At this time, all the sensors have received the supported algorithms sent by the Core Node. Each sensor decrypts the received message using the symmetric Key GKDH,

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.5. CLUSTER SA NEGOTIATION EXCHANGES TUNNELING MANAGEMENT PROTOCOL and then verifies the integrity of the message using the same key. Every sensor chooses its preferences of cryptographic suites and responds by sending a message in this format. {Type of message,{Mid,ID Sensor,SAr}GKDH }, {digest}GKDH Mid is the message identifier sent by the core node in the previous message. ID sensor is the identifier of the sensor which sends the response. SAr states the cryptographic suite chosen from the offered choices sent in the payload SAi. – These three payloads are encrypted using the Group Key GKDH. digest is a digest calculated and encrypted using the key GKDH. • Step 3: At this step, the Core Node has received all the responses from the sensors and performs some tasks: – it decrypts the message using the symmetric group key GKDH; – it calculates the digest of the sent payload; – it decrypts the payload {digest}GKDH using the key GKDH;

– it compares the calculated digest and the received one to verify the integrity of the message. If those tests are positive, the sensor is then authenticated and can participate to the Cluster Security Association CSA.

4.5.2. Establishment of the CSA In the previous steps, the core node has received the identifiers of all the sensors and the selected cryptographic suites. The core node can establish the CSA corresponding to the responses. The CSA of each group of sensors contains the following information: • the list of the sensors; • the key of the session GKDH already established in the initial exchanges; • the cryptographic suite (for signature and encryption). The characteristics of the CSA (despite the key) are then sent to all the sensors and are encrypted using the previous Group Key GKDH. The payload {digest}GKDH ensures the integrity and authenticity of the Core Node. The message sent is represented below. {Type of message , {list of sensors , cryptographic suite}GKDH }, {digest}GKDH At this final step of exchanges, each sensor node has the cryptographic suite, the list of trusted sensors and the session key. An illustration of the CSA establishment protocol is represented in Figure 4.3.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.6. ROBUSTNESS OF THE ALGORITHM TUNNELING MANAGEMENT PROTOCOL

Figure 4.3.: The CSA establishment steps

4.6. Robustness of the algorithm In this part, we will discuss and prove the robustness of the proposed security protocol either for the exchanged messages or the validity of the key. We will also discuss the impact of a node compromise on the establishment process of the group tunnel and the CSA.

4.6.1. Messages Robustness In this part, we will analyze the robustness of each one of the messages exchanged in the establishment of the group key. For each message we will analyze its ability to support the authenticity, the confidentiality and the integrity of the sent data. • The message: {Type of message, [Identifier,gs∗Ri ]}, {AUTH}KSi This message is sent by each sensor to the core node to present its contribution in the

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.6. ROBUSTNESS OF THE ALGORITHM TUNNELING MANAGEMENT PROTOCOL group key. – The authentication is verified because the field AUTH is encrypted using the private key of the sensor KSi; – The integrity is done by the field AUTH which is a digest of the original message. This part cannot be modified by an intruder because it necessitates the use of the private key; – Either the original data is sent in clear mode but the confidentiality is ensured. An intruder cannot use this information. The important information sent is the contribution of the sensor which is gS∗Ri . But the values of S and g are secret values and are deployed in trusted nodes and are unknown for the intruder. Then it cannot separate the value gS∗Ri into separate correct values of S and Ri. • The message: {[Identifier of S1, gS∗R1∗Rc ],..., [Identifier of Si, gS∗Ri∗Rc ],..., [Identifier of Sn, gS∗Rn∗Rc ]} , {AUTH}KCN This message is the message in which the Core Node sends all the contributions of the sensors to all the nodes. Those values will be used to calculate the group key. – The authentication is verified because the field AUTH is encrypted using the private key of the Core Node KCN. Then an intruder cannot sign the message using this key; – The integrity is done by the field AUTH which is a digest of the original message. This part cannot be modified by an intruder because it necessitates the use of the private key of the Core Node; – Either the original data is sent in clear mode but the confidentiality is ensured. An intruder cannot use this information. The important information sent is the contributions of the sensors which are the sets of gS∗Ri∗Rc . But the values of S and g are secret values and are deployed in trusted nodes and are unknown for the intruder. Then the latter cannot separate the value gS∗Ri∗Rc into separate correct values of S, Ri and Rc. • The messages:

– {Mid, SAi}GKDH {digest}GKDH – {list of sensors , cryptographic suite}GKDH , {digest}GKDH – {Mid,ID Sensor,SAr}GKDH }, {digest}GKDH

In those messages, the Core Node and the sensors establish the preferences for the CSA. Those three messages use the same techniques to ensure the security requirements. – The authentication is verified because the field digest is encrypted using the group key. This key is only known by the trusted sensors; – The integrity is ensured by the field digest which is a digest of the original message. This part cannot be modified by an intruder because it necessitates the use of the group key;

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.6. ROBUSTNESS OF THE ALGORITHM TUNNELING MANAGEMENT PROTOCOL – The message sent is encrypted using the group key. An intruder cannot decrypt the data because the group key has been distributed in a secure manner.

4.6.2. Key Validity In addition to the security performed in the key negotiation steps, we have an other propriety of the developed protocol DynTunKey. This propriety gives a more robust security scheme. In fact, the established tunnels has not an infinite validity. As presented previously in the specifications of the proposed protocol, the group key is periodically established with the contribution of the different communicating nodes that belong to the same group. From an other point of view, a sensor that belongs to one group and has moved to another group will be automatically deleted from its initial group and affected to the new group. This propriety enhances the global security of DynTunKey, because the sensor S1 that changes from one group G1 to another group G2 will not be able to decrypt the messages of its first group G1. In fact, in the regeneration of the keys, the corresponding node S1 will share the key with the sensors that belong to its newest group G2. Then the old key of the group G1 (known by S1) will no longer be a valid key and it cannot use it to decrypt the messages of its initial group. This is a protection against unauthorized access. In the case where the key does not change periodically, the sensor that has moved from one group to another can maintain a copy of its initial Group Key. Then, it can hear and decrypt the messages of the first group without belonging to this group. But for DynTunKey, this problem is automatically resolved because the keys are not of infinite validity and the sensor will have only a valid key used in the group to which it belongs in the current instant. Given that the group key has a finite and short validity duration, an attack based on cryptanalysis fails. In fact, such an attack needs processing time. Therefore, the periodic renewal of the key will be a protection mechanism because the attacker will not have enough time to deduce the group key and use it to decrypt the exchanged data.

4.6.3. The effect of a node compromise Till now we have demonstrated the robustness of the CSA establishment process relative to the messages robustness and key validity. The network implemented operates in the general cases in a hostile environment and the sensor nodes can be compromised by two means. For the first compromise method the attacker will destroy the sensor to be in a denial of service. For the process of the group key establishment, if a node is not operating the group key and the CSA will be correctly established. In fact, the list of the sensors that should collaborate to establish the group key is not predetermined and any trusted nodes can participate in the establishment of the key in a secured manner. Then, if a node is destroyed the proposed key establishment process is not affected. The second type of a node compromise is a trial of the sensors content reading. The aim of this compromise is to use the node ID and the keys stored in the sensor node. If this is done, the intruder will be able to participate in the key establishment process. To prevent this kind of compromise, we propose two preventive mechanisms: • For the first protection solution, the memory contents stored in the sensors should be

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL encrypted. In that case if an intruder tries to access the content it should decrypt it. If we use a strong cryptographic scheme, the attacker will not be able to read in clear the ciphered content; • For the second protection solution, we propose a physical protection of the nodes by the use of self destructive sensors. This kind of sensors is automatically destructed at any attempt of an external sensor manipulation or attempt of content reading. In that case, if any intruder tries to read the content the node will be automatically destructed and then the memory that contains the cryptographic keys and security information is flushed. By the mean of these two protection mechanisms, the intruder will not be able to use the content of the sensor either for sending data or participating in the group establishment process because the information stored in the sensor will not be available.

4.7. Performance evaluation In this section, we will present in a first part the simulation model used in the evaluation of DynTunKey. This model contents respectively the deployment model and the generation of the events used in the simulation. We will then represent the results of some conducted simulations. In the first simulations, we will compare the proposed protocol DynTunKey to some recent works that propose a key management solution for Wireless Sensor Networks. The second simulations aim to compare DynTunKey to the classic IPSec protocol [52, 53].

4.7.1. Simulation Model 4.7.1.1. Deployment of the sensors The sensors are deployed in the area to ensure a k-covered area. The deployment strategy is described in [10, 33]. We divided the area into zones. As presented in the specifications of DynTunKey, the sensors that belong to the same zone establish a shared tunnel with the Core Node. We divided the lifetime of the sensors into slot times. As presented in the communication model, each sensor will report its collected data to the core node at the end of the slot time. The Group Key validity is a set of slot times. 4.7.1.2. Deployment of the targets In conventional sensor networks, static targets are used and considered in the research works. Our study is interested by securing the detected events. In our case we will consider many targets in the monitored region and those targets are moving. Initially we deployed random targets in the covered area of the Wireless Sensor Network. Those targets are mobile. We considered respectively two models of mobility which are the Random Walk and the Gauss Markov mobility models [12]. For each of those models we evaluated some metrics to compare DynTunKey and other key management solutions. The simulations and results will be presented in the next sections. In this subsection we will introduce those two mobility models. • The Random Walk Model

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL Starting from the initial distribution of the targets at time t0 , we assume that each target performs a 2-D random walk movement. With this mobility model, each target Ti travels from its current location to a new location by randomly choosing a direction θ ∈ [0..2∗π] and a distance di ∈ [dmin ..dmax ] in a prefixed time interval ∆t . For our case, each target will have its own distance and direction in each time interval. Considering the initial positions Xi (t0 ) and Yi (t0 ) of a particular target at time t0 , the formulas 4.1 and 4.2 give the new position of a target Ti after the random movement. Xi (t0 + ∆t ) = Xi (t0 ) + di (t0 + ∆t ) ∗ cos(θi (t0 ) + ∆t ),

(4.1)

Yi (t0 + ∆t ) = Yi (t0 ) + di (t0 + ∆t ) ∗ sin(θi (t0 + ∆t )),

(4.2)

where di (t0 + ∆t ) and θi (t0 + ∆t ) are random variables and represent respectively the distance and direction of displacement that occurs in the period between t0 and t0 + ∆t . This mobility is done in every time interval ∆t and for every target Ti . • The Gauss Markov Model The Random Walk model is a pure random model. In fact, for this model there is not a logic moving of the targets. The Gauss Markov model is considered to be a more realistic mobility model. In fact, it relates the current displacement distance di (t0 + ∆t ) and the current direction θi (t0 + ∆t ) with its previous displacement di (t0 ) and direction θi (t0 ). So this model takes into account the previous movements of each target and then we will have a more realistic mobility path either in direction or speed. The formula 4.3 gives the new displacement di (t0 + ∆t ) relatively to di (t0 ) di (t0 + ∆t ) = a ∗ di (t0 ) + (1 − a) ∗ d¯ + where:

p

1 − a 2 ∗ Xd ,

(4.3)

– a: is the tuning parameter ∈ [0..1]; ¯ is the average displacement distance in the interval ∆t relatively to the speed – d: of the target; – Xd : is a random variable of the Gaussian distribution. The formula 4.4 gives the new direction θi (t0 + ∆t ) relatively to the previous direction θi (t0 ) θi (t0 + ∆t ) = θi (t0 ) + where: – b: is the tuning parameter ∈ [0..1];

p

1 − b2 ∗ Xθ ,

– Xθ : is a random variable of the Gaussian distribution.

(4.4)

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL Then, the new positions are given using equations 4.5 and 4.6. Xi (t0 + ∆t ) = Xi (t0 ) + di (t0 + ∆t ) ∗ cos(θi (t0 + ∆t )).

(4.5)

Yi (t0 + ∆t ) = Yi (t0 ) + di (t0 + ∆t ) ∗ sin(θi (t0 + ∆t )).

(4.6)

4.7.2. Comparison of DynTunKey to MAKM, NSKM and RDKM In this first part of the simulations, we will compare the proposed protocol DynTunKey to recent key management protocols for Wireless Sensor Networks. In particular, we will consider the protocols MAKM [54], NSKM [47] and RDKM [48]. To evaluate these security protocols, we will compare them relatively to the storage space required for the cryptographic keys used in all the network. In the conducted simulations: • we considered many densities in the deployment of the sensors to ensure k-coverage; • we deployed targets in the covered area. The targets are periodically moving using the random walk and the gauss markov mobility protocols‘; • As used in the simulation of the previous works, the size of each key is 20 bytes. For each one of the security protocols, we evaluated the storage space required for the network keys used. We have considered two scenarios corresponding to the mobility models. Figure 4.4 represents the evaluation of the security protocols when the targets moves using the Random Walk Model. This figure illustrates an evaluation of the key storage space for DynTunKey and the protocols MAKM [54], NSKM [47] and RDKM [48].

Figure 4.4.: Key Storage Space for Random Walk Figure 4.5 represents the evaluation of the key storage space when the deployed targets moves using the Gauss Markov Model. Such as the case for the random walk simulations,we evaluated the key storage space for all the keys needed for DynTunKey and the protocols MAKM [54], NSKM [47] and RDKM [48].

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL

Figure 4.5.: Key Storage Space for Gauss Markov - The first deduction in those simulations, is that the proposed protocol DynTunKey outperforms all the other solutions regarding to the Key Storage Space metric. The proposed protocol gives good performances either considering the Gauss Markov mobility or the Random Walk mobility. This result is logic, because for DynTunKey the number of the keys needed and stored in the nodes of the network does not depend only of the number of the sensors but depends directly of the behavior and occurrence of the targets in the monitored area. In fact, all the mentioned solutions establish keys for all the network nodes despite the reporting necessity meaning that the key storage space is always maximal. Then, as a consequence the performances of DynTunKey will be better or at least will give the same performances of the other protocols when all the sensors establish keys with the Core Node. - An other remark is that as well as the sensors density is greater as well as the shift between the Storage space for our solution and the other ones become greater. For example, when considering the 1-coverage and 2-coverage densities, the key storage space for DynTunKey is closer to the values of the other protocols. But, when the value of k becomes greater, the key storage space is larger. In fact, as said previously the number of created and stored keys in DynTunKey is dynamic and depends not only of the number of the sensors but also is established only for the sensors that have to report their detections. And then, the number of the keys is adapted to the number of the events that occurs in the monitored area. But, for the other solutions, the number of the stored keys depends only of the number of the sensors and then the relation between the number of the sensors and the number of the keys is direct. As a general conclusion, DynTunKey gives good performances when compared to the other protocols in all the sensors density cases and targets behavior.

4.7.3. Comparison of DynTunKey and IPSec In the previous section, we compared the proposed protocol DynTunKey to recent proposed key management protocols. Those protocols are used in Wireless Sensor Networks but does not use the concept of tunneling. To the best of our knowledge, our solution is the first one that uses the concept of tunnels to ensure the security for the Wireless Sensor Networks. Based on those facts, we will compare DynTunKey to an other protocol based on the tunneling

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL management. We conducted simulations to compare the proposed protocol to the classic IPSec protocol [52, 53]. The metrics that will be used in the comparison are respectively the number of the tunnels, the number of messages and the security latency. 4.7.3.1. Number of tunnels For this first part of the simulation, we will compare DynTunKey and IPSec regarding the number of created tunnels. We considered different values of the Key Validity period going from 1 slot time to 14 slot times. For each of those values, we calculated the number of created tunnels for IPSec and DynTunKey protocol. The results of the simulations are shown in the following figures.

Figure 4.6.: The Number of tunnels for the Random Walk Model Figure 4.6 represents the variation of the created tunnels for each one of the protocols considering the Random Walk mobility model.

Figure 4.7.: The Number of tunnels for the Gauss Markov Model Figure 4.7 represents the variation of the created tunnels for each one of the protocols considering the Gauss Markov mobility model. The two Figures 4.6 and 4.7 shows that the number of the created tunnels for DynTunKey are less than those created for IPSec.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL In fact, when considering the IPSec protocol, each event will be reported in a separate manner. And then, the number of tunnels is equal to the number of the detected and reported events. But for our protocol, the events are gathered and reported at the end of each reporting period. Also, all the sensors that belong to the same zone establish a unique tunnel. And then, for DynTunKey, a set of sensors (belonging to the same zone) and a set of events (that occur in the same period of key validity) share the same tunnel. - Considering those facts, it is logical that the number of tunnels for IPSec is very greater than those for the proposed protocol DynTunKey. As a conclusion, this part of simulations shows that despite the model of mobility and behavior of targets, the proposed protocol gives a better performance than IPSec when considering the number of created tunnels. - It comes also from those simulations that the number of tunnels decreases according to the period. In fact, we considered the same events with the same mobility, and when the value of period is greater a more number of events are reported using the same Shared Group Key. So as well as the value of the period is greater as well as we have less number of tunnels. But, we have to remember that a great value of key validity period means that the Group Key will be used for a longer time which is not suitable for the security protocol robustness. So we have to choose a median value of the period to ensure both a small number of tunnels and an accepted value of Group Key periodicity regeneration.

The RW Model The GM Model

Period ∈ [1..6] 1860 1167

Period ∈ [7..14] 160 106

Table 4.2.: The standard deviation of the number of tunnels The table 4.2 represents the standard deviation of the number of tunnels. We measured the standard deviation of the tunnels number for the small values of the period and the big ones. In this table we can observe that for the two models of mobility, the variation of the tunnels number is great when considering the small values of the period (ranging from 1 to 6 slots). But for the big values of the period, the difference between two successive periods is not big when compared with variation between the small values of the period. In fact, when considering the smallest values of the period between 1 and 6 slots times we will have a standard deviation equal to 1860 for Random Walk and equal to 1167 for the Gauss Markov model. But for the biggest values of the period in the interval from 7 to 14 slot times, the standard deviation is equal to 160 for the random walk model and 106 for the gauss markov model. So from the experiments and results, we can determine the best value of the period. For the simulated case, we can select the value of the period in the first interval [1..6] that minimizes the value of the number of tunnels when compared with the previous period. Also the most adequate period is the one from which we will not have big variance with the tunnels number of the next periods. When considering this value, we can ensure two major proposed advantages of DynTunKey which are respectively the smallest value of the key period validity (to decrease the risk of an attack) and almost the minimal value of number of tunnels that can be reached which is one of the principal goals of DynTunKey.

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL 4.7.3.2. Number of messages For this simulation, the metric evaluated is the number of exchanged messages between the sensors and the core node. We considered different values of the period going from 1 slot time to 14 slot times. For each one of those values, we calculated the number of exchanged messages for IPSec and the proposed protocol. As well as the previous simulation, we considered two models of mobility which are the random walk and the gauss markov mobility. The variation of the number of messages in relation with the period when the targets are moving in a Random Walk manner is represented by the following figure.

Figure 4.8.: The Number of messages for the Random Walk Model The variation of the number of exchanged messages when considering the Gauss Markov mobility model is represented by the following Figure.

Figure 4.9.: The Number of messages for the Gauss Markov Model It comes from Figures 4.8 and 4.9 that the number of messages decreases according to the period. For the two mobility models, when comparing DynTunKey protocol and IPSec, the number of messages exchanged to establish all the tunnels for DynTunKey is less than those

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.7. PERFORMANCE EVALUATION TUNNELING MANAGEMENT PROTOCOL exchanged for IPSec. This is a logical result because for IPSec every event is reported at a separated manner and then messages are exchanged for every event detected to establish a dedicated tunnel. But for DynTunKey, many events detected by a sensor node are reported on a same tunnel. So, there is only one exchange of tunnel establishment messages for the proposed protocol. This establishment is performed at the beginning of the key validity interval. Hence, we can deduce that the proposed protocol outperforms IPSec when considering the communication overhead resulting from the establishment of the encrypted tunnels.

The RW Model The GM Model

Period ∈ [1..6] 42745 20466

Period ∈ [7..14] 4948 2191

Table 4.3.: The standard deviation of the number of messages As well as for the number of established tunnels, we represent in table 4.3 the standard deviation of the number of messages. We represent the standard deviation for small and big period values, when considering the two mobility models. The same conclusions can be deduced. For the small values of period we have a big variance between the successive values, but for the big values of the periods the deviation is less than the one for the small values. Then, the chosen value have to be in the small values of period. This value will verify the two criterions. At a first point it is smaller enough to reduce the key validity interval and it gives also a remarkable variance of the number of messages when compared with the next values. 4.7.3.3. Security latency In the third part of the simulation, we compare the proposed protocol DynTunKey and IPSec regarding the time required to establish all the tunnels.

Figure 4.10.: The Latency time for the Random Walk Model We developed two programs, the first one implements the exchange of messages as described in the specifications of DynTunKey. The second program implements the exchange of the messages to establish an IPSec tunnel. The same list of events is used for both protocols

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.8. CONCLUSION OF THE CHAPTER TUNNELING MANAGEMENT PROTOCOL and we considered such as the two previous simulations the Randow Walk and Gauss Markov mobility models. The two programs were executed on the same laptop and we measured the total time of execution of each program. The results of this simulation are represented in Figures 4.10 and Figure 4.11.

Figure 4.11.: The Latency time for the Gauss Markov Model Despite the mobility model, the time required to establish the tunnels for DynTunKey is less than the time required for IPSec. This is a logical result because when considering IPSec, the tunnel are created in separate manner because each sensor exchanges its specific messages with the core node. But for the proposed protocol DynTunKey, only at the end of each period many sensors exchange simultaneously messages with the core node to establish a common tunnel or CSA. Also, the number of tunnels in DynTunKey protocol are less than those for IPSec and then the time of creation of tunnels will be certainly lower. Table 4.4 represents the standard deviation of the Latency time. And for the same reasons, the best chosen period have to be in the small values. Then, a median value of the regeneration period will guarantee a small key validity interval, an optimal value of the tunnels number, an optimal number of messages and reduced establishment latency.

The RW Model The GM Model

Period ∈ [1..6] 17.61 14.6

Period ∈ [7..14] 1.46 1.45

Table 4.4.: The standard deviation of the security latency

4.8. Conclusion of the chapter In this chapter, we proposed DynTunKey which is a secure group key and tunneling management protocol for Wireless Sensor Networks. This protocol aims to establish dynamic secure tunnels between the nodes. To optimize the creation of the tunnels, the protocol creates a shared Cluster Security Association common to the sensor nodes that detect the same event and belong to the same geographic zone. The proposed protocol has many advantages. Regarding the security, it permits a dynamic generation of a periodic Group key. This Shared Group key is a symmetric key and then the encryption of the data uses less computational

CHAPTER 4. DYNTUNKEY: A DYNAMIC DISTRIBUTED GROUP KEY 4.8. CONCLUSION OF THE CHAPTER TUNNELING MANAGEMENT PROTOCOL resources than an asymmetric solution. In addition to the inputs in terms of security, the proposed protocol is useful in many applications that necessitate a secure communication between all the nodes such as military target tracking or firefighting collaborating team. We conducted some simulations to evaluate the performance of the proposed protocol and we have shown that the proposed approach considerably reduces the key storage space, the processing and communication overhead. In those simulations, we considered two mobility models of the targets to show that the DynTunKey gives good results in all the cases.

Border Surveillance using sensor based thick-lines

5

5.1. Introduction to the chapter In this chapter we will propose a border surveillance system based on WSNs. The WSNs are used in many applications covering both military and civilian domains. The first goal of the use of a WSN-based solution is to monitor a particular event in a sensed area. In particular, a surveillance application monitors either an area or a borderline. One of the most recent surveillance applications of WSNs is the border surveillance application. This kind of applications is becoming critical due to the increase of the risks of intrusion on borders. Due to the global risks near their borders, governments are frightened of the appearance of intruders on their borders, either for unauthorized importation of goods or for terrorism actions. Then all governments have demonstrated a need of a good and efficient surveillance system to control suspect activities on borders. Typically, WSNs within these applications are based on small devices connected through radio links and are able to detect the presence of intruders in the monitored area. However, a good monitoring necessitates certain coverage requirement to avoid shadowing areas and missing measurements. Based on these facts, the sensors have to be efficiently deployed to provide a good quality of coverage. An efficient deployment of sensors has to satisfy many requirements including the degree of coverage, network connectivity, and network maintainability. For these reasons one of the most important tasks in implementing a WSN for Border Surveillance is the deployment phase. In this chapter, we present an implementation framework of Border surveillance solutions using Wireless Sensor Networks. In particular, we provide the following contributions: • a global network architecture for border surveillance based on three types of components deployed on a thick line architecture set up along the border; • a deployment strategy to comply with many constraints on the network operation, including coverage maintenance, connectivity preservation, and routing quality of service; • a mechanism of routing management to ensure an optimal and efficient process of data relaying and reporting between the different nodes. We also present the mechanism of elaboration of the routing tables at the different levels of the network.

80

5.2. RELATED CHAPTER 5. BORDER WORKS SURVEILLANCE USING SENSOR BASED THICK-LINES The rest of the chapter is organized as follows. In Section 5.2 will be presented works that treats the border surveillance applications and the routing protocols adapted for adhoc networks. The architecture of the network and the different node features are introduced in Section 5.3. Section 5.4 describes the detailed steps of the deployment strategy proposed. In Section 5.5, we present a routing scheme to be used in our framework to permit efficient communication between the several deployed nodes. Section 5.6 assesses the efficiency of the proposed deployment and routing protocol through some conducted simulations. Finally, Section 5.7 concludes the chapter.

5.2. Related Works In this section, we present in a first part, a summary study of the WSN based border surveillance applications. In a second part, we compare the proposed routing protocol and the existing routing solutions adapted to adhoc networks.

5.2.1. Border Surveillance applications Many recent works have addressed border surveillance applications based on WSNs. Many solutions using WSNs have organized the network nodes as a line-sensor [55–59], where every movement going over a barrier of sensors is detected, in this case, sensor nodes deployment should guarantee barrier coverage. Compared to full coverage, a barrier coverage based on a perfect linear deployment requires fewer sensor nodes and may experience radio disconnection due to sensor failure and depletion. In addition, this kind of linear architecture does not permit the tracking of intruders, since the intruders are detected only when crossing the line-sensor and their future movements cannot be tracked. Furthermore, in the barrier coverage model, the sensor nodes are not able to locally determine whether the barrier coverage is ensured [60]. This limitation impedes the development of localized algorithms. Barrier and k-barrier coverage require, for instance, that all crossing paths in a monitored region are covered by at least one sensor node or k distinct sensor nodes, respectively. In [61], Sun et al. proposed a 3-layered hybrid network architecture for border patrol. In Bordersense [61], advanced sensor networks technologies have been deployed, including multimedia sensor nodes, mobile sensor nodes, and scalar sensors, which could be deployed underground or above the ground. The authors analyzed the k-barrier coverage requirements in terms of sensor density when the sensor nodes are deployed randomly in a belt in front of the border according to the Poisson point process with spatial density. The high cost of the deployment of multimedia sensors (cameras) makes the deployment of Bordersense for larger borders difficult. Furthermore, a uniform distribution of sensor nodes cannot be achieved in such environment. Liu et al. [62] investigated the construction of sensor barrier on long strip area of irregular shape when sensors are distributed according to Poisson point process. To ensure that trespassers cannot cross the border undetect, multiple disjoint sensor barriers will be created in distributed manner covering large scale boundaries [62]. Then, a segmentation technique has been proposed to achieve continuous barrier coverage of the whole area. In [63], Dudek et al. proposed a demonstrator, consisting of 10 sensor nodes, for WSNs in the frame of border surveillance application and with which the basic functionality of the network along with trespasser detection have been tested. The case of border surveillance when sensor are randomly but non-uniformly deployed has been investigated in [55, 56]. Uniform sensor

5.2. RELATED CHAPTER 5. BORDER WORKS SURVEILLANCE USING SENSOR BASED THICK-LINES deployment is useful in theoretical analysis but remains unrealistic in the real deployment where a non-uniform placement is obtained do to several environmental factors. Saipulla et al. [56] assessed that barrier coverage could be obtained under line based deployment and outperforms that of Poisson model. K-barrier coverage has been the focus of [60], S. Kumar et al. proposed an algorithm to determine whether a belt region is k-barrier coverage or not and introduced the notions of weak and strong barrier coverage. Weak barrier coverage detects intrusion attempts when trespassers go over a barrier of stealthy sensors, however, strong barrier coverage detects intrusion when even sensors are not stealthy. Most of the proposed frameworks for border surveillance assumed a line-based network where barrier or k-barrier coverage is ensured. Under this assumption, we drastically reduce the number of needed sensors to monitor a given border, this amelioration is obtained at the expense of the performance of the network. In other words, line-based deployment at best ensures the detection of all intrusion attempts, but will not allow to take an internal decision about tracking the intruders. Furthermore, the sensor nodes placement which follows a uniform distribution [61, 62] is not realistic due to environmental factors and sensor nodes deployment process (air-dropping, throwing from aircraft). Another limitation of the proposed models is the silence about the connectivity issue which is closely related to coverage issue. To enhance the performances of border surveillance WSN, it is recommended to deal with connected-coverage. In the presented deployment solution, we will deal with connected coverage in border surveillance WSN under the full coverage model and we will also treat the routing process which has not been addressed in the presented works.

5.2.2. The adhoc routing protocols In this subsection, we briefly compare the proposed routing protocol to two well known routing protocols for adhoc networks. In literature, the routing protocols are classified in two main categories. The reactive protocols: this kind of protocols establishes and updates the routes of the network on demand (when needed). The most known reactive protocols are presented in the followings. • AODV (Ad-hoc On-demand Distance Vector)[64]: When a source node requires a route, it creates routes on the fly and maintained as long as the source is needed. For multicast groups AODV builds a tree. The wireless nodes store an entry route to each node of the network. When the network is very dense these tables will be very voluminous. AODV requires little energy consumption and does not require high computing power, so it is suitable for use with WSNs. • DSR (Dynamic Source Routing)[65]: DSR is similar to AODV given that the routes are established on a demand when a node have data to be transmitted. However, it uses source routing instead of relying on the routing table of each intermediate router, meaning that the source node should have an indication on all the nodes that will be crossed to reach a destination. The DSR protocol is not very efficient in large networks because each source must have the vision of all the network. The cost of maintenance of routing tables is important. But it is suitable for small and medium networks.

5.2. RELATED CHAPTER 5. BORDER WORKS SURVEILLANCE USING SENSOR BASED THICK-LINES The proactive protocols: this routing protocol establishes and maintains continuously routes in advance without the wait for an event occurrence. The most known proactive protocol is DSDV (Destination-Sequenced Distance Vector routing)[66]. The DSDV protocol guarantees loop-free routes. At every node are stored route entries to all the nodes of the network. The path to a destination is determined using the distance vector shortest path routing protocol. DSDV introduces a heavy load on the network due to the periodic update messages. The protocol is not suitable for large scale networks because the bandwidth is used in the updating messages. In the particular case of the WSNs, the DSDV is not suitable because it consumes a lot of energy for routing tasks. The characteristics of the routing solution we proposed are listed in the followings: • the method presented is an hybrid method: at the initiation of the network, the proposed method establishes all the routes of all the network like the proactive protocols. After that, no periodic update is done. The routes are only updated if a node detects that the routes are no longer valid like the reactive protocols. Therefore, the proposed protocol permits an efficient routing and requires a little energy consumption for routing tasks; • The routes established in the proposed method are the lightest ones that can be done because they are composed of only one entry at each node. Each sensing node can send its gathered data to the routing node through other nodes. We also gave a solution that permits unicast communications between any two nodes using the same reduced routing table.

Table 5.1.: A comparison of our routing protocol to other routing protocols Protocol Category of the protocol The routing tables AODV Reactive protocol: when a the number of entries (the sensor will send data to next hop) in a node routing another one, the table is equal to the number corresponding route is of nodes communicating with established it. DSR Reactive protocol: when a source routing: the route sensor will send data to entry indicates all the another one, the intermediate nodes to reach corresponding route is the destination. established DSDV Proactive protocol: the every node contains an routes are periodically indication of a route to all updated the other nodes of the network. The proposed Hybrid protocol: proactive at the routing table of each routing the initiation of the network node contains a unique entry. protocol and reactive in the rest of the network lifetime

5.3. ARCHITECTURAL CHAPTER 5. BORDER SURVEILLANCE ISSUES USING SENSOR BASED THICK-LINES

5.3. Architectural issues In this section, we will present the global architectural issues of the network that will be used in the conceived border surveillance system. We will first give a summarized presentation of the linear networks. Then, we present the network architecture, the hierarchy of the nodes used and the functionalities assured by each type of node. Finally, we will present the network architecture and the network topology.

5.3.1. Presentation of the linear networks Many of WSN applications involve placing the sensors in a linear form, making a special class of these networks defined as a Linear Sensor Network(LSN). The LSNs are a topology of repartition of the nodes constituting the WSN network. One can find three categories of LSNs according to the repartition of the different nodes: thin, thick and very thick [67]. In addition, based on the hierarchical organization of the deployed nodes, the LSNs can be classified into several categories: one-level and multi-level LSNs. • The Thin LSNs: The most basic LSN is the one where the nodes deployed are organized in a unique dimensional linear form. We can distinguish one-level thin LSNs for which the network is composed on a unique type of nodes having the mission of sensing events. The thin LSNs can be multi-level in the case where more than one kind of nodes are deployed. In the latter case, some nodes are used for sensing and the others for communication and routing tasks. Regardless to their hierarchical level, the nodes are deployed on the same line. • The Thick LSNs: In this topology the linearity does not exist at all the nodes levels. Only the nodes constituting the upper levels are organized in linear infrastructure. The sensing nodes are scattered in an area delimited by two lines. They monitor the presence of intruders and send the data collected to the nodes deployed on the central line. As the thin LSNs, a thick LSN can be one-level or multi-level depending on the kinds of the sensors deployed. • The Very thick LSNs: For this architecture all the nodes are placed in an area delimited by two parallel lines. The very thick LSNs may be one-level or multi-level.

5.3.2. Node Hierarchy Sensor networks can be classified into two categories: flat sensor networks (one-level) and heterogeneous sensor networks (multi-level). In a flat WSN, all the sensor nodes have the same sensing, communication and processing characteristics. A heterogeneous WSN integrates various sensor types with different capabilities and functionalities. The presence of heterogeneous nodes (i.e., nodes with an enhanced energy capacity or communication capability) in a WSN has the advantage of increasing network reliability and lifetime. Typically, a large number of inexpensive nodes perform simple sensing tasks, while a few expensive nodes (that may be embedded on mobile platforms) provide network control, data filtering, fusion and transport. This segregation of roles promotes a cost-effective design of the network as well as a more efficient implementation of the overall sensing application. We consider the particular case of a heterogeneous wireless sensor network for border surveillance. To ensure efficient border surveillance, the network is a three layer hierarchical network,

5.3. ARCHITECTURAL CHAPTER 5. BORDER SURVEILLANCE ISSUES USING SENSOR BASED THICK-LINES where layers are built on three types of nodes: - The Basic Sensing Nodes (BSN) constitute the first hierarchical level. They are the most elementary nodes in the considered architecture and are low powered when compared to the nodes of the other layers. The first role of a BSN is to detect the occurrence of an event (e.g. the presence of an intruder). The BSNs collaborate to relay the information gathered to the next layer in an optimized manner. A multi-hop communication may take place between the nodes of the first and second layer. - The Data Relay Nodes (DRN) form the second hierarchical layer of the network. The DRNs have no sensing capabilities. Their main task is to collect the data gathered by the BSNs and collaborate to relay it to the next layer. They implement intelligent functions to optimize the relay process and reduce the resources needed for the relay. Thus, each DRN has hierarchically a set of BSNs under its control and is responsible of routing their reports to the nodes belonging to the third layer. The DRNs are more powerful and have more capabilities than the BSNs. - The Data Dissemination Nodes (DDN) compose the third layer of the network. Their function is to send the collected data to the Network Control Center (NCC) for analysis. The DDNs analyze, preprocess and fusion the related events reported by the DRNs. The global nodes hierarchy is represented in Figure 5.1 .

Figure 5.1.: The nodes hierarchy

5.3.3. The Network topology In this subsection, we will present the topology of the network or, in other terms, the physical repartition of the nodes. We consider a thick linear network because it enlarges the area covered by the network and permits a tracking of the intruders. In fact, if we consider a thin linear network, the target will be detected only one time when it crosses the line covered. But, if we consider a thick linear network, a whole strip is covered and then many BSNs will detect the target along this strip and the sensing task is more realistic, more efficient, and reflects the behavior and displacements of the targets. Thus, in addition to the detection process, intrusion tracking and location estimation can be performed. The DRNs will be deployed on the central line. The BSNs and the DDNs are scattered in a strip around the DRNs line. The topology of the network is represented by Figure 5.2.

5.4. THE DEPLOYMENT CHAPTER 5. BORDER SURVEILLANCE STRATEGY USING SENSOR BASED THICK-LINES

Figure 5.2.: The network topology

5.4. The deployment strategy In this section, we will present the deployment strategy developed. The aim of the deployment is to ensure an optimal and efficient quality of coverage and connectivity.

5.4.1. The deployment of the DRNs As presented in the previous section, we will consider a thick linear topology. For the DRNs, we will deploy them in a linear manner and this line will be the main artery of the network. When deploying the DRNs on this virtual line, we have to ensure that all the DRNs are mutually connected. This connectivity can be ensured through one or multiple hops. Such linear repartition is connected when any DRN have at least two other DRNs in its communication range. The two neighboring nodes should be placed exclusively at the DRN right and left side meaning that the DRN have to be placed between the two nodes. In fact, if a DRN of the line is only connected to sensors on its right (or left) side, the line of DRNs will be partitioned into two isolated DRN sub-lines and the connectivity of the network will be broken. We denote by RCDRN the communication range of the DRNs. An example of a DRN broken line is represented by Figure 5.3.

Figure 5.3.: A DRN broken line An efficient DRNs deployment should ensure the linear connectivity. As represented in Figure and deploy a BSN 5.4 , when considering the segment lines S1 , S2 and S3 of width RCDRN 2 in each segment, we will have a full connected DRN line. In fact, the maximum distance between any couple of points P1 in S1 and P2 in S2 will be less than RCDRN . The distance between any couple of points P2 in S2 and P3 in S3 will be less than RCDRN . Then, having will ensure that each DRN is connected to a left and a DRN in each segment of width RCDRN 2 right neighbor. Generalizing this result to all the segments of width RCDRN will ensure a full 2

5.4. THE DEPLOYMENT CHAPTER 5. BORDER SURVEILLANCE STRATEGY USING SENSOR BASED THICK-LINES connectivity of the DRN line. When having two connected neighbors in the two directions, each DRN will have a path (through multiple hops) to all the other DRNs of the network. Given that the border is generally vast, we cannot perform a deterministic deployment and in the general cases the sensors deployed are thrown from air. Then, we cannot determine an exact position of the DRN but only have an approximative control on the portion of area in which will a sensor land. Thus, a good deployment should ensure the presence of at least one DRN in each segment of width RCDRN . 2

Figure 5.4.: The required density for DRN deployment Based on these constraints, we can determine the number of the needed DRNs. Let us denote by LLine the length of the line on which the DRNs will be deployed. The number NDRN of needed DRNs to have a linear connectivity is given by the following equation. NDRN =

2 LLine LLine = RCDRN /2 RCDRN

(5.1)

The deployment of the NDRN nodes will not be done in a deterministic manner but as shown previously, the random deployment performed should ensure that each segment of width RCDRN will contain one DRN. 2

5.4.2. The deployment of the BSNs After deploying the DRNs on the main arterial line of the network, we can perform the deployment of the BSNs. For the BSNs deployment process, we have two major constraints to be satisfied. • The first constraint states that any deployed BSN should be reachable in one hop by at least one DRN. As stated in the architectural issues, the DRNs will be responsible of communicating management tasks to the BSNs. Our choice is that this communication should be done in one hop to make the communication easier and avoid a possible lateness of transmission. • The first constraint is related to the connectivity needs but the major task of the BSNs is to ensure an efficient sensing of the monitored area. The second constraint to be satisfied states that the number and the BSNs repartition should ensure an efficient monitoring. The proposed deployment scheme that satisfies these two constraints is detailed in the following subsections.

5.4. THE DEPLOYMENT CHAPTER 5. BORDER SURVEILLANCE STRATEGY USING SENSOR BASED THICK-LINES 5.4.2.1. The BSNs connectivity based deployment To ensure the first constraint, the deployment process should guarantee that the farthest BSN have to be in the communication area of at least one DRN. Knowing that all the DRNs are deployed on the same line, the width of the strip chosen (centered on the DRN line) should guarantee that any BSN position in the strip is distanced by less than RCDRN to at least one DRN. In the deployment process for the DRNs, we have proven that the farthest DRNs are distanced by RCDRN . As illustrated by Figure 5.5, the farthest possible BSN position is the intersection of the DRN communication disc and bisector of the segment of width RCDRN .

Figure 5.5.: The determination of the width of the strip To satisfy the first deployment constraint, we can easily deduce that the width of the strip should verify the following relation. WStrip =

√

3 RCDRN

(5.2)

Remark. If the designer needs a larger √ surveillance area, he can choose a width of the strip larger than the recommended value 3 RCDRN . All the details of the √ proposed solution are valid regardless to the value chosen. If the width of the strip is > 3 RCDRN , a BSN located on the border of the strip may be far from all the deployed DRNs by a distance > RCDRN . Therefore, it is not connected directly to any DRN. We notice that when a DRN sends data to a BSN (or a broadcast message to all the BSNs), it can be sent in multi hop through intermediate BSNs. It is evident that √ a multi hop relay process necessitates additional treatments than one hop relay. The choice of 3 RCDRN is not rigorous but it is the optimal value that provides the largest width of the strip and the simplest communication process from DRNs to BSNs done in all the cases in one hop. 5.4.2.2. The BSNs sensing based deployment Till now, we determined the dimensions of the strip to be covered. In the second step, we will present the deployment solution that allows an efficient sensing of the strip. To determine the number of the needed BSNs, we will use the method presented in Section 2.2.3. The details are presented in the indicated section, we will highlight only the main results of the used method. The surface Sc sensed by a BSN is given by the following formula. Z Rs =Rsmax Z θ=2Π∗pθ (Rs ) Sc = (Rt + Rs )dRs dθ (5.3) Rs =0

θ=0

5.5. THE ROUTING CHAPTER 5. BORDER TECHNIQUE SURVEILLANCE USING SENSOR BASED THICK-LINES To have k-coverage of the strip, it has been demonstrated that the density of the BSNs repartition should be k ρS = . (5.4) Sc The total number of needed BSNs is given by the following formula. NBSN = ρS × AStrip

(5.5)

where AStrip = WStrip × LStrip is the measure of the area of the strip to be monitored. The BSNs will be scattered in the monitored area using a uniform and random deployment. Remark. The designer of the network can use any other method to determine the number of required BSNs to cover the strip. For example, the deployment method based on energy consumption [68], the method based on geographical patterns [69] or any other deployment method presented in the literature can be used.

5.4.3. The deployment of the DDNs Let us first notice that the DDNs deployment has not many requirements to comply with. In fact, we do not have neither coverage density needed nor required connectivity between the deployed DDNs. A DDN needs only to receive data from the DRNs, either in one-hop connectivity or multiple hop DRNs connectivity. Thus, we need only to define the number of the DRNs using the same superior hierarchical DDN. Having the total number of the DRNs deployed, we can easily determine the number of needed DDNs in the whole network. This number is given by equation 5.6. NDDN =

NDRN NDRN perDDN

(5.6)

To be sure that the DDNs are √in the communication range of the DRNs, they will be deployed in a strip of width equal to 3 RCDRN and centered around the DRN line. The DDNs will be randomly and uniformly deployed in the strip. We do not need really rigorous positions of the DDNs because in all the cases these nodes will receive data routed through many DRNs.

5.5. The routing technique In this section, we present the routing procedure used in the proposed architecture. The process of detection begins at the BSNs level which detect and prepare the data related to the detected intrusions in their area of coverage. Then, the BSN has to deliver this data to the nearest DRN directly or through other intermediate BSNs. After receiving the reports, each DRN will deliver the gathered data to a DDN on a multi hop routing through other DRNs. The routing process is illustrated by Figure 5.6.

5.5.1. The routing from BSN to DRN √ Knowing that the BSNs are deployed in a strip of width 3 RCDRN and that the BSNs communication range RCBSN is smaller than RCDRN one can easily notice that many BSNs cannot directly send data in one hop to the DRN. The data will then be relayed from the originating BSN to the DRN through intermediate BSNs.

5.5. THE ROUTING CHAPTER 5. BORDER TECHNIQUE SURVEILLANCE USING SENSOR BASED THICK-LINES

Figure 5.6.: The routing process of sensed data 5.5.1.1. The problem of broadcast storm Given that the physical transmission channel for WSNs is by nature a diffusion channel, that the BSNs are deployed as a random graph (without predefined hierarchical dependency infrastructure), and that the BSNs transmit immediately the packets received, the packet will be at each hop duplicated. We will then have a flooding of multiple copies of the same packet in the network (a broadcast storm). The packet will finally reach the DRN concerned but in multiple copies which necessitates additional treatments to detect the redundancy and consider only one copy of the packet. In the case of heavy load on the network, the sensor nodes will be highly uselessly used and will consume rapidly their energy which will affect the global lifetime of the network. The problem of a broadcast storm is depicted on Figure 5.7.

Figure 5.7.: An illustration of a broadcast storm

5.5. THE ROUTING CHAPTER 5. BORDER TECHNIQUE SURVEILLANCE USING SENSOR BASED THICK-LINES To avoid the flooding, a packet sent by a BSN should indicate the next hop that will be in charge of relaying it to another hop in direction to the requested DRN node. In that case, when a BSN sends a packet, it is received by all its neighbors due to the constraints of the physical transmission channel. But when receiving the packet, the neighbors will read the field that indicates the next hop and only the BSN indicated as the next hop will relay the packet. The non concerned BSNs will drop it. We will always have a unique copy of each packet until it reaches the DRN destination. Then, we are sure that no broadcast storm occurs and by consequence a fewer number of nodes will be used in the relay process and the energy consumption for many BSNs will be economized. The DRNs will receive a unique copy of the packet and there is no longer a need of duplicated packets detection. In practice, this is done by the mean of routing tables. The routing tables will logically arrange the BSNs and DRNs in a spanning tree to eliminate the physical loops. Each BSN is assumed to have a routing table that indicates the route to the corresponding DRN. Each line of the routing table contains the following information: • The Device Identifier (the Mac address) of the next hop; • The Device Identifier of the DRN; and • The number of hops to reach the DRN through this route. 5.5.1.2. The construction of the routing tables Due to the adhoc nature of the deployed WSN, the network physical infrastructure can change at any moment without any anterior knowledge. For this reason, the use of static routing tables is not practical. The routing tables of the BSNs should be constructed dynamically to be self organized and adapted rapidly to any change in the network physical infrastructure. The establishment of the routing tables is done in a 2-step process through exchanges between the BSNs and DRNs. First Step: Periodically, each DRN sends a beacon message that indicates its presence. This message is a multicast message because it can be used by any BSN. At the end of this step, the BSNs that can send directly data (in one hop) to a DRN will detect the presence of the DRN. The transmission range used when sending the beacon message has to be RCBSN instead of RCDRN . In fact, if the beacon message is sent using RCDRN communication range it will reach BSNs that are distanced from the DRN of more than RCBSN . Thus, they will suppose having a DRN directly reachable but, in fact, it is not in their communication range and the routing information will be false. At the end of the first step, a BSN that can directly reach a DRN will add a direct route to a DRN in their routing table. The device identifier of the DRN is sent in the beacon message having the following format. {DRN Presence, DRN Device Identifier}

Second step: Each BSN will send its routing table to its neighbors. When receiving the routes declarations from a neighbor, each BSN will update its routing table if necessary. The update is done if the received route indicates fewer hops to reach a DRN than the available

5.5. THE ROUTING CHAPTER 5. BORDER TECHNIQUE SURVEILLANCE USING SENSOR BASED THICK-LINES route. If the routes of a BSN changes, it sends immediately its local routing table to inform the neighbors of the new established route. This message has the following format. {Route Update, Source Device Identifier, DRN Device Identifier, number of hops} where: • Source Device Identifier indicates the device identifier (or the Mac address) of the BSN that have sent the route update. This field is necessary because the BSNs that will register the routing entry in their routing tables should indicate this device identifier as the next hop; • DRN Device Identifier indicates the DRN corresponding to the routing entry. This information is necessary in the data packets, because the BSN should indicate the DRN to which the data should be delivered; • number of hops indicates the number of hops that will be crossed to reach the DRN. When a BSN receives a route update, it increments the number of hops received and compares it to the number of hops indicated in its routing table. If the received route indicates a fewer number of hops, it updates the routing entry using the received one, else the routing entry does not change. After some exchange rounds between the BSNs, all the routing tables will be filled in and every BSN will have an indication of a route (e.g. the next hop) in direction of the nearest DRN. The number of the rounds needed to have a converged network is relative √ to RCBSN and to the width of the strip 3 RCDRN . The routing protocol should also ensure that the routing entries reflects the infrastructure of the network and is up to date. With the proposed establishment solution of routes, the routing tables will be updated automatically if the network changes because the BSNs send immediately their routing tables in case of change. A network change consists at a modification of the DRNs number or positions, which will cause the change of the routing entries of the closest BSNs. This modification will be diffused to all the other BSNs. The proposed routes establishment process provides an efficient update of the routes only if all the BSNs are in operation, meaning that it does not give solutions in case of a BSN failure or a deployment of a new BSN. In case of a node failure, all the routes are still valid even if the routes are interrupted. For example, let consider a BSN B1 having its neighbor B2 as a next hop. Despite the reason, the node B2 can be out of service (breakdown or energy expiration) and then the indicated route will no longer be valid. In the case of non change of the network infrastructure (e.g. the line of DRNs is not modified), no BSN is going to send route updates. Then, the BSN B1 will not receive any route update and the route remains registered while being not valid. In case of a new node deployment, if the architecture of the network does not change, no route update will be sent and then the new node will not receive any route announcement. In the following we propose respectively methods to update the routing entries in the case of a BSN failure or the add of a new BSN. The solution to the problem of a BSN failure: In the case of node failure, some routes will not be valid. As a solution, many routing protocols classified as proactive protocols propose

5.5. THE ROUTING CHAPTER 5. BORDER TECHNIQUE SURVEILLANCE USING SENSOR BASED THICK-LINES that the wireless nodes send periodically their routing tables (despite an event or modification of the routing table) to ensure that a periodic update of the routing tables of all the BSNs is done. This solution efficiently resolves the problem but the nodes will uselessly and rapidly consume their energy by sending the routes update. As a reaction to a node failure, we propose that the update is done on demand to avoid the consumption of energy in tasks that may be unnecessary. When a BSN has gathered data, it will send this data to the DRN through the next hop indicated in its routing entry either of being still valid or invalid. When receiving the data from a BSN, the DRN is asked to send an acknowledgment unicast message to the originating BSN as a response. If a DRN response is received, the BSN can implicitly be sure that the data has been delivered and that the route is valid and no action is needed. But if no DRN response has been received, the BSN deduces that the route through the next hop is no longer valid. Upon the occurrence of this event, the BSN will send a message {Route Failure Notification} to indicate that the route is no longer valid. This message will be received by all the neighbors of the BSN, which will also diffuse it until it reaches the DRN. When the DRN receives a message {Route Failure Notification}, it deduces that there is a problem of routing entries validity. In the next step, the DRN will send a message {Flush Routing entries} in diffusion to order all the BSNs in its communication range to delete all their routing tables. Now, when all the tables of BSNs in the area of this DRN are flushed, the previously detailed steps 1 and 2 will be done and all the tables of the BSNs will be newly filled in and by consequence valid. The presented method gives a solution to the nodes failures avoiding the classical periodic updates solutions which consumes a lot of energy. Figure 5.8 illustrates an example of this process.

Figure 5.8.: The solution to a node failure

The solution to a new deployed BSN: When a new BSN is deployed it is not sure that the existing BSNs are sending route updates. Then, the new BSN will not receive route updates. The method we propose to integrate and set up a valid routing table of the deployed BSN is

5.5. THE ROUTING CHAPTER 5. BORDER TECHNIQUE SURVEILLANCE USING SENSOR BASED THICK-LINES initiated by the latter. When deployed, the BSN sends a message {Request of Route} to all its neighbors. When receiving this message, the neighbors of the new BSN will respond in a unicast message. This message is a route update message containing the available route at the neighbor. The new deployed BSN will receive many routes, it will select the best one (the fewer number of hops) and stores it as a valid route entry. A recapitulation of the messages needed in all the steps of routes establishment and alternative scenarios is illustrated by Table 5.2. Table 5.2.: The used messages for routes establishment Message Signification DRN Presence A DRN sends this message in multicast to notify its presence to the nearest BSNs. Route Update This unicast message is sent by a BSN to announce to the neighbors a routing entry. Route Failure This multicast message is sent by a BSN to announce to its Notification neighbors a routing failure. Flush Routing This multicast message is sent by a DRN to order to the entries BSNs to flash their routing entries. Request of This multicast message is sent by a new deployed BSN to Route its neighbors looking for a route to the DRN.

5.5.2. The routing from DRN to DDN Once an optimal route is built, the data is relayed through the BSNs and reaches the corresponding DRN. This data is sent through intermediate DRNs until it reaches a DDN. As presented in the deployment strategy section, each DRN will have two adjacent DRN neighbors and then at least two possible paths to a DDN. The routing table of each DRN contains then two entries describing the possible routes. The entries have the following attributes. • The device identifier of the DRN neighbor (the next hop); • The device identifier of the DDN; • The number of intermediate DRNs on this route. When reporting data, the DRN selects the best route corresponding to the minimal intermediate hops between the DRN and the DDN. The establishment of the routes at the DRNs is done through route updates between the DRN neighbors. The mechanism of routing entries establishment is the same than the case of routing between BSNs and DRNs. The DDNs will notify their presence using periodic beacon messages. The DRNs that will receive a DDN beacon will register a direct routing entry to a DDN and the others will receive and register the routing entries from the neighboring DRNs.

5.5.3. The routing to the BSNs In this subsection, we present the routing mechanism in destination to the BSNs. A BSN node can receive data either from a DRN or a BSN. To send data to a BSN, the DRN constructs

5.6. PERFORMANCE CHAPTER 5. BORDER EVALUATION SURVEILLANCE USING SENSOR BASED THICK-LINES a unicast message to the BSN and sends this message. Given that all the BSNs are in the communication range of the DRN, the message will certainly reach the BSN. Sending data from DRNs to BSNs does not require a routing management process. For the case where a BSN will send data to another BSN, we propose that the BSN constructs a data payload to which it sticks the BSN destination address. It indicates on the payload that the message will be relayed to another BSN. The all is sent to a DRN using the classical routing from a BSN to a DRN. The message sent by the BSN has the following format. {Message to Relay, Data Payload, BSN Destination Address} When receiving a message marked to be relayed to another BSN, the DRN will send the received message as a unicast message and indicates the address joint by the source BSN as the destination address. Given, that all the BSNs are in the communication range of the DRN, the message will reach the BSN. Operating at this manner, we will not have necessity of maintaining routing management between the BSNs. Most of the existing routing protocols for adhoc networks establish dynamic routes between all the wireless nodes. This treatment necessitates continuous updates and a large amount of exchanged data in the routing tables that becomes enormous when the number of deployed nodes is greater. The main contribution of our solution is to avoid the complex process of the routes establishment and the use of complex routing tables.

5.6. Performance evaluation In this section, we first present the simulation model used to evaluate the performances of the proposed network architecture. Then, we discuss the results of some conducted simulations. The aim of these simulations is to analyze both the deployment process and its impact on the routing process proposed.

5.6.1. The simulation model The characteristics of the simulation model are listed in the followings:

• The communication range of the DRNs is equal to 80 meters;

√ • The monitored area considered in the simulation is 1000m * 138 meters ( 3 ∗ RCDRN ); • We will consider in the simulation many values for the k-coverage quality varying form k=1 to k=9; • We also considered many sensing and communication ranges for the BSNs to study their impact on the quality of the deployment. Considering this simulation model, we conducted two simulations. • In the first simulation, we study the impact of the k-coverage parameter on the routing technique and the number of the hops needed to reach the DRN line.

5.6. PERFORMANCE CHAPTER 5. BORDER EVALUATION SURVEILLANCE USING SENSOR BASED THICK-LINES • In the second simulation, we varied the BSNs communication range and studied the impact of this parameter on the number of non-connected BSNs. A non-connected BSNs is a node that does not have a path to the DRN line.

5.6.2. The variation of the number of hops In this simulation, we varied the coverage factor k and measured for each value the mean number of hops needed for a BSN to reach the DRN line. We also measured the maximum number of hops for each value of k. We fixed the communication range of the BSNs to RCBSN = 20m. The results of this simulation are represented by Figure 5.9.

Figure 5.9.: The number of hops We can observe in this simulation, that the worst mean value for the number of hops is equal to 3 intermediate nodes, which is an acceptable value. Also, when considering the worst case of all the conducted simulations, the maximum number of hops detected for all the BSNs is equal to 9 intermediate sensors. Then, we can deduce that the proposed deployment and routing techniques, gives good performances when evaluating the number of hops needed to reach the DRN line.

5.6.3. The number of non-connected BSNs In the second simulation, we have observed the number of the non-connected BSNs that have not a route to a DRN. This metric gives an indication on the quality of the deployment and the relative connectivity percentage. In the first part of this simulation, we fixed the parameter k to 1-coverage and varied the BSNs communication range RCBSN between 6m and 24m. The table 5.3 represents for different values of the communication range, the total number of the deployed BSNs and the number of the BSNs that do not have a route to a DRN. Figure 5.10 represents the percentage of non-connected sensors for 1-coverage. We remark that the deployment and routing solution gives a small number of sensors non-connected. In the second part of this simulation, we conducted the same simulation as the previous case but for 2-coverage deployment. The table 5.4 represents for different values of the Communication range, the total number of deployed nodes and the number of the nodes that have not a route to a DRN node. Figure 5.11 represents the percentage of non-connected sensors for

5.6. PERFORMANCE CHAPTER 5. BORDER EVALUATION SURVEILLANCE USING SENSOR BASED THICK-LINES

Table 5.3.: The non-connected BSNs for RCBSN 6m 8m Number total deployed nodes 5328 2975 Non-connected nodes 374 181 RCBSN 16m 18 m Number total deployed nodes 746 589 Non-connected nodes 42 28

1-coverage 10 m 12 m 1905 1333 101 93 20 m 22 m 477 395 25 19

14 m 973 55 24 m 331 12

Figure 5.10.: Percentage of non-connected BSNs for 1-coverage 2-coverage. We notice on table 5.4 that the number of the non-covered sensors in 2-coverage is less than those for the case of 1-coverage. In fact, for 2-coverage, we have more sensors deployed than in the case of 1-coverage; and thus we have more possibilities of finding routes. Therefore, when having a larger value for k (which is the general case in most deployments), we may have a little percentage of not connected nodes. For example, for this simulation, the mean value of the percentage of non connectivity for 2-coverage is 0.36% and for 1-coverage it is 5.5%. In the general cases despite of the coverage degree needed, we remark that the percentage of non connected BSNs has a maximal value equal to 7% for one coverage and 0.6% for 2coverage which is an acceptable value. This lack of connectivity is inevitable because we are in the case of a random deployment strategy. But, we notice that the percentage of the non connected nodes is not enormous when compared with the total number of deployed nodes. If

Table 5.4.: The non-connected BSNs for RCBSN 6m 8m Number total deployed nodes 10655 5949 Non-connected nodes 17 11 RCBSN 16m 18 m Number total deployed nodes 1491 1177 Non-connected nodes 6 6

2-coverage 10 m 12 m 3810 2665 7 7 20 m 22 m 954 790 4 4

14 m 1946 7 24 m 662 4

5.7. CONCLUSION CHAPTER 5. BORDER OF THE SURVEILLANCE CHAPTER USING SENSOR BASED THICK-LINES

Figure 5.11.: Percentage of non-connected BSNs for 2-coverage we want a total covered network (like deterministic solutions), this lack can be overcome by two possible means. The first one is using mobility where the non connected BSNs perform small distance movements to reach a connectivity area of another BSN. Another solution is to deploy additional BSNs in an incremental process to cover the small black areas and therefore provide a 100% connectivity of the deployed network.

5.7. Conclusion of the chapter In this chapter we presented the details of a solution for the border surveillance application using a hierarchical Wireless Sensor Network. We presented the general architectural aspects of the network used. We also proposed deployment and routing techniques to be used in this network. The deployment has to satisfy several constraints in particular the coverage of the monitored area and the connectivity of the nodes to ensure a good quality of coverage and an efficient exchange of the gathered data. We presented an efficient solution for routes establishments adapted to the adhoc particularities. We also addressed the problem of routes updates when a BSN fails or when new BSNs are deployed.

DWBS: A Distributed Wireless Border Surveillance System

6

6.1. Introduction The WSNs are used in many fields to build surveillance systems to detect and report intrusion related events. One of the most recent domains in which the WSNs have been used is Border Surveillance. To have an efficient tracking of malicious related events, the WSNs are widely used for border surveillance. Many research works have focused on providing useful solutions, requiring an important interest to hard problems including sensor deployment, energy consumption control, and security. One among the most investigated issues is the deployment technique to set up in the design of the network. In fact, an efficient tracking and reporting strategy needs an optimal choice of the sensors’ number and locations. In this work, we consider the context of border surveillance using a hierarchical WSN and propose a network and deployment model through multiple lines of protection. The solution conceived is titled DWBS as Distributed Wireless Border Surveillance System. We also propose mathematical models to evaluate the network performances and support planning tasks. The main contributions of the work presented in this chapter are the following: • We built a WSN based surveillance system called DWBS that is able to provide a controllable surveillance of infiltration within a large area neighboring the border. Techniques are developed to plan and dimension the deployed network. They use two types of sensors, DRNs and BSNs, responsible respectively for building a communication network and sensing coverage within the monitored area; • The second contribution is the setup of a deployment model taking in consideration real deployment conditions. The deployment method consists at paving the monitored area with paving patterns of predetermined shapes translating the environment conditions; • Using the paving deployment technique, we proposed two deployment strategies for DWBS. The first one is a deterministic deployment method and the second is more generalized and provides a controlled random deployment method. In the present chapter, we introduce the DWBS architecture and define the paving deployment technique. We consider in this chapter the deterministic deployment case for DWBS. The rest of the chapter is organized as follows. In Section 6.2 will be presented works that treats the

99

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.2. RELATED WORKS SYSTEM problem of deployment of sensors. In Section 6.3, we conducted mathematical calculations to analyze and determine the path and the sensor landing position when thrown. In Section 6.4, we present a description of the paving technique used in our work. We present in the same section, the architectural issues of the deployed network. Section 6.5 describes the deterministic deployment model proposed for the DRNs and the BSNs. In Section 6.6, we present some conducted simulations to evaluate the performances of the deployed network. Finally, Section 6.7 concludes the paper.

6.2. Related Works We can classify the deployment strategies for WSNs into three categories, they are: the Static Nodes placement with controlled deployment, the Static Nodes placement with random deployment and the dynamic nodes placement with random deployment [70]. For the static nodes placement with controlled deployment scheme [55, 57, 60, 62, 63, 71– 75], the sensors are placed in positions chosen to ensure full coverage of the monitored area. These sensors are static and never change their positions. This solution gives an optimal and guaranteed quality of coverage; but, positioning the static sensor nodes at specific places necessitates easy, direct and full control of the monitored area. Unfortunately, this is not always guaranteed on frontiers and large monitored areas because of geographical irregularities nature. For this reason, this deployment strategy is not suitable for large scale applications and border surveillance. In the static nodes placement with random deployment scheme [76–79], the sensors are static, but are not placed in deterministic positions. In fact, this deployment scheme is used when placing the sensors at deterministic positions is risky or infeasible. In that case, the sensors are dropped, from air, which leads to randomly spreading them in the monitored area and will not ensure the total coverage or connectivity of the monitored area because the sensors’ distribution may be non uniform. Based on this, many research works have proposed a random deployment of large populations of nodes to overcome uncertainty. In that case, having a large number of sensors can give a greater density and may permit a good coverage of the monitored area. This deployment technique can be used in all kinds of monitored areas and many types of applications. However, it shows a major drawback related to the high cost it generates. For the dynamic nodes placement with random deployment [19, 56, 61, 80–87], the sensors are able to move within the monitored area. In a first step of the scheme, the sensors are randomly spread in the sensed area. Similarly to the previous class, the sensors will not ensure good coverage of the area of interest. To deal with this lack of coverage, a second step assumes that the sensor nodes will be able to move after deployment by changing their positions to ensure the required quality of coverage. This method can be used in large scale applications because initially the sensors are placed randomly in the monitored area. Also, given that the sensors can move to ensure the required coverage, an excess of nodes redundancy is not needed. This solution has two drawbacks: first, the motion may cost a lot of energy and second the sensor motion cannot be handled properly because of nature irregularities.

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM The main advantages of the DWBS network architecture and deployment scheme is that we used hierarchical networks to ensure both coverage and connectivity. For the deployment process, we used controlled random deployment avoiding the cost of the deterministic deployment. Having a controlled random deployment, the positions of the sensors are controlled using a mathematical model that gives the expected positions of the sensors. For the size of the network, the solution we proposed is scalable and can be used either for large network, small networks or scalable networks. Table 6.1 illustrates a comparison of those deployment techniques and the method we proposed in this paper within some metrics and factors.

6.3. The characterization of the paving patterns In this work, we have an area to be covered using wireless sensor networks. The deployment technique that will be used is to pave the monitored area with predetermined shapes of paving patterns. We mean by paving the monitored area, the operation of choosing the placement of the paving patterns to ensure a given deployment goal such as the connectivity or coverage. As presented previously, an airplane or a helicopter will follow a linear movement and at specified positions the sensors will be dropped. A paving pattern is the area in which can a sensor land when thrown from air. In our work, the shape of the paving patterns is not random but has predetermined shapes translating the environment conditions (Figure 6.1). The first set of considered environment characteristics are related to the airplane in charge of the deployment. It includes altitude and velocity when the sensor is thrown. The second set of environment characteristics include the characteristics of the wind experienced when deploying such as velocity and direction. Each paving pattern can contain one or more sensors. In this section, we give a mathematical characterization of the shape and size of paving patterns. The deployment parameters that will be considered are respectively wind parameters, the altitude and velocity of the airplane. In a second part, we decompose the paving patterns into sub zones based on the values that the wind parameters may likely have. These sub regions will be characterized by the probability of sensor landing.

6.3.1. The shape of the paving pattern In this section, we present the mathematical calculations performed to characterize the paving patterns. To represent the positions of the landing sensor, we consider an orthonormal basis (Ox,Oy,Oz). For the sake of model clarity, we consider the following parameters: • FWx : is the force of the wind applied on the x axis; • FWy : is the force of the wind applied on the y axis; • FWz : is the force of the wind applied on the z axis; • m: is the mass of the thrown sensor node; • V0 : is the real number representing the velocity of the airplane on the Ox axis when the sensor is thrown on the x axis; • h: is the altitude of the airplane when it throws the sensor;

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM Deployment method

Type of deployment

Motion of nodes

[71, 72]

Deterministic

Static

[73, 74]

Deterministic

Static

[75]

Deterministic

Static

[60]

Random

Static

[63]

Random

Static

[55]

Random

Static

[57]

Random

Static

[62]

Random

Dynamic

[76–81]

Random

Dynamic

[82]

Deterministic and Random

Dynamic

[83]

Deterministic and Random Random

Dynamic

Random

Dynamic

[61]

Random

Static

[56]

Random

Dynamic

DWBS

Controlled Random Deployment

Static

[19, 84– 86] [87]

Static

Type and Scale of the network Uniform, Small Scale Networks Uniform, Large Scale Networks Uniform, Large Scale Networks Heterogeneous, Large Scale Networks Uniform, Large Scale Networks Uniform, Large Scale Networks Uniform, Large Scale Networks Heterogeneous, Scalable Networks Uniform, Large Scale Networks Heterogeneous, Large Scale Networks Uniform, Large Scale Networks Uniform, Belt Region

Optimization objective

Heterogeneous, Belt Region Uniform, Scalable Networks Uniform, Belt Region Heterogeneous, Scalable Networks

Coverage Control using mobility Probabilistic Coverage

Predefined Coverage Control Predefined Coverage and Connectivity Control Predefined Connectivity Control Missing Coverage Control (dense networks) Probabilistic Coverage Control and data fidelity Probabilistic Coverage and Connectivity Control Data Fidelity Coverage Control using mobility Coverage Control using mobility Coverage Control using mobility Coverage Control using mobility Probabilistic Coverage

Coverage Control using mobility Mathematical Based connectivity and coverage control

Table 6.1.: Comparison of the deployment methods • tL : is the moment at which the sensor lands. Using the above forces, we determine the landing position of the sensor. To have this, we should evaluate the position coordinates of landing on the Ox and Oy axes. The forces applied to a sensor when thrown on the three axes, are given by the following

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM equation. F~(t) = (FWx (t), FWy (t), FWz (t) + mg)

(6.1)

For simplification, we suppose in our case that the wind force on the z axis is equal to Zero. FWz (t) has only an effect on the duration of the landing time. Having the relation F~ = m ~γ , the motion is given by the following equation. ~ = ( FWx (t) , FWy (t) , g) (6.2) γ(t) m m To get the velocity, we should integrate ~γ assuming that the airplane velocity vector is equal ~ is given by the following equation. to (V0 , 0, 0) at the moment of sensor drop. The value of V V ~(t) = (

Zt

FWx (u) du + V0 , m

0

Zt

FWy (u) du, gt) m

(6.3)

0

The landing positions LP on the three axes are given by equation (6.4) knowing that at t=0 the sensor is located on the airplane (origin). The landing motion and the forces applied on the sensors when landing are represented by Figure 6.1. LP~(t) = (

Zt Zu 0

0

FWx (v) dv du + V0 t, m

Zt Zu 0

FWy (v) 1 dv du, gt2 ) m 2

(6.4)

0

Figure 6.1.: The landing motion and paving pattern If we assume that during sensor fall down, the wind direction and intensity are constant; then FWx and FWy are constant, and equation (6.4) can be written as FWy 2 1 2 FW LP~(t) = ( x t2 + V0 t, t , gt ). (6.5) 2m 2m 2 We notice that if the airplane altitude is q denoted by h at the start of the sensor fall down, 2h then the sensor will reach land at tL = g . One can easily determine in that case that

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM F

h

Wx the coordinates of the landing point are given by ( mg + V0 positions are given by the following system of equations 6.6.  q FWx h 2h   + V LP = 0 x  mg g

   LP = y

q

2h FWy h g , mg , h).

The Landing

(6.6)

FW y h mg

The previously presented values correspond to a precise value of FWx , FWy and V0 . To characterize the form of the paving pattern resulting of these coordinates, we should consider the different possible values of FWx , FWy and V0 and study the form of the area drawn by LPx and LPy given by the system of equations 6.6. Given that FWx , FWy are the forces of wind respectively on Ox and Oy axes, the relation between them is given in accordance to a wind model that describes the behavior of the wind in direction and velocity. In the followings, we consider several wind models and determine the resulting possible landing position of the sensor when considering each model. One can distinguish three generic deployment scenarios depending on the characteristics of the wind parameters. 6.3.1.1. The elliptic wind model The forces of the wind verifies FWx ≤ FWx,max and FWy ≤ FWy,max . In this model, there is two strictly positive real numbers a and b defining the relation between the maximal values of the wind forces FWx,max and FWy,max as: 2 FW x,max

+

2 FW y,max

=1 (6.7) a2 b2 Let us now characterize the landing area (or paving pattern) of a sensor for the case of the elliptic wind model. Based on Equation (6.7), the relation between FWx and FWy is illustrated by the following relation. 2 2 FW FW x + 2y ≤ 1 a2 b This inequation can also be written in the form. t4L 4m2 t4 a2 4mL2

2 FW x

+

t4L 4m2 t4 b2 4mL2

2 FW y

(6.8)

≤1

(6.9)

From Equations (6.6) and (6.9) we deduce the relation between LPx and LPy by the following inequation. (LPx − V0 tL )2 (

t2L

a 2 2m )

LPy2

+ (

b t2L 2 2m )

≤1

(6.10)

This inequation gives an elliptic relation between LPx and LPy . The resulting elliptic disc ES representing the set of the positions in which a sensor S can land is centered on CE =

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM (V0 tL , 0, 0), the major semi axis of ES is equal to b t2L 2m .

a t2L 2m

and the minor semi axis is equal to

The paving pattern corresponding to the elliptic wind model is depicted by Figure 6.2.

Figure 6.2.: The elliptic wind model

6.3.1.2. The circular wind model For the circular model, the wind forces verifies the following characteristics: • The forces of the wind on the Ox and Oy axes are dependent;







~ ] and all • The norm of the wind force denoted by F varies in the interval [0, F~ max the directions of the wind are possible;

• The value of the airplane velocity when the sensor is thrown is controllable and equal to a precise and fixed value.

In this case, the values of the wind forces FWx and FWy on the Ox and Oy axes are given by the following equations.



~  F =

F cos(α)  Wx     



~ (6.11) F =

F sin(α) W y        , with α in [0, 2π].

Then relatively to the circular wind model, the landing positions are given by the set of equations 6.12.  q kF~ kcos(α) h   LPx = + V0 2h  mg g (6.12)   ~ kF ksin(α) h  LPy = mg

We denote by

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM

s

~



F h 2h

~ β( F ) = and w = V0 mg g

The relation between LPx and LPy is then given by Equation 6.13. 

LPx − w β(F )

2

+



LPy β(F )

2

=1

(6.13)

Based

for a given value of the wind force

on the deduced relation, the expected landing area

~

~

F is a circle C of center CC (w, 0, 0) and radius β( F ). The whole possible landing area of a sensor for all the

is then the union of all the

circles

~

~ possible values of the wind forces norm F varying in the interval [0, F ]. Then the kF~ kmax

max h

landing area (e.g. the paving pattern) will be a disk of radius R = and centered mg q 2h on CDisc = (V0 g , 0, 0). The paving pattern corresponding to the circular wind model is represented by Figure 6.3.

Figure 6.3.: The circular wind model

6.3.1.3. The free wind model (or rectangular wind model) For the free wind model, the wind forces verify the following characteristics: • The forces of the wind on Ox and Oy axes are independent; • The maximal value of the wind force on Ox axis FWx is equal to FWx,max . Given that the wind force can be in the same direction or the opposite direction of the airplane, then FWx varies in the interval [−FWx,max , FWx,max ]; • The maximal value of the lateral wind force on the Oy axis FWy is equal to FWy,max . Given that the wind force can be in the left or the right side of the airplane, then FWy varies in the interval [−FWy,max , FWy,max ];

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM • The value of the airplane velocity when the sensor is thrown, is controllable and is equal to a precise value.

Based on these characteristics and using the equation of LPx and LPx given by Equation 6.6, the landing positions for the free wind model are described in the followings: • The position of landing LPx on the Ox q q axis can be in the interval FWx,max h 2h FWx,max h [− m g + V0 g , + V0 2h m g g ]. LPx then varies on a segment centered on q 2h F W x,max ; Cx = V0 2h g and of width = m g

FWy,max h FWy,max h , ]. m g m g FWy,max 2h = ; m g

• The position of landing LPy on the Oy axis can be in the interval [− LPy then varies on a segment centered on Cy = 0 and of width

Considering the possible landing values on the Ox and Oy axes and given that the LPx and LPy are independent variables, then all the combinations of LPx positions and LPy positions are possible. Based on q this, the paving pattern is simply a rectangle centered on the center FWx,max 2h 2h and a length equal to of fall down CR = (V0 g , 0, 0) having a width equal to m g FWy,max 2h . m g

The paving pattern for the free wind model is illustrated by Figure 6.4.

Figure 6.4.: The rectangular wind model Remark. In the next parts of this work, we will consider the elliptic model in our calculations and analysis since the circular case is a particular case of it (achieved when a=b) and that the rectangular model is very easy to analyze.

6.3.2. The customization of the paving patterns Since the statistics of the wind variations affects the dimension of the paving patterns, we will customize the paving patterns to the wind characteristics observed through time. For this, we assume that a sensor is thrown from a point (0,0,h) and that the statistics on the wind forces shows n scenarios. For the ith scenario, the wind forces FWx and FWy vary respectively in [0, FWx,maxi ] and [0, FWy,maxi ]. The wind force components cannot exceed the values FWx,maxn

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.3. THE CHARACTERIZATION OF THE PAVING PATTERNS SYSTEM and FWy,maxn . We denote by P wi , where i=1...n, the probability that FWx is in [0, FWx,maxi ] and FWy in [0, FWy,maxi ]. We suppose that the statistics of the wind forces verifies the following system of inequalities. P w1 < P w2 < .. < P wi < .. < P wn = 1

(6.14)

We suppose now that there are 2n strictly positive real numbers ai and bi , i=1..n, verifying the following system of inequalities.   a1 < a2 < ...... < ai < ai+1 < ....... < an (6.15)  b1 < b2 < ...... < bi < bi+1 < ....... < bn We suppose that the relation between FWx,maxi and FWy,maxi is led by the following formula. 2 FW x,max

a2i

i

+

2 FW y,max

b2i

i

=1

(6.16)

Using Equation (6.10), the landing area of a sensor for the ith scenario is an elliptic disc a t2

b t2

i L and a minor semiaxis equal to i2mL . En denoted by Ei of a major semiaxis equal to 2m is the maximal elliptic disc in which the sensor can land and corresponds to the nth wind scenario. Knowing the inequalities depicted by the relation (6.15), we deduce that the relation between the different constructed elliptic discs is given by the following expression.

E1 ⊂ E2 ⊂ .... ⊂ Ei ⊂ Ei+1 .... ⊂ En

(6.17)

Now, we partition the set of possible landing positions En to disjoint sub areas that does not intersect. We denote by Eni (where i=1,...,n), the disjoint sub areas to which will be divided the ellipse En . Those sub areas are represented by the following system of equations.  1  En = E1 (6.18)  i En = Ei \Ei−1 , i = 2..n

This decomposition is depicted by Figure 6.5.

Figure 6.5.: The decomposition of the ellipse En We denote by PEni the probability that the sensor lands in the subarea Eni . The wind forces and probabilities depicted by the system of equations (6.14) are useful to generate the probability that the sensor lands in the sub-elliptic areas. It is easy to show that these probabilities are given by:

6.4. CHAPTER DWBS: PAVING 6. DWBS: BASED A DISTRIBUTED DEPLOYMENT WIRELESS TECHNIQUE BORDER ANDSURVEILLANCE ARCHITECTURE SYSTEM   PEn1 = P w1 

(6.19)

PEni = P wi − P wi−1 , i = 2, ..., n

Given that the intensity of the wind force components cannot exceed the values FWx,maxn and FWy,maxn , then the probabilities of the sub elliptic areas Eni verifies the following equation. n X

PEni = P wn = 1

(6.20)

i=1

6.4. DWBS: paving based deployment technique and architecture In this part, based on the sensors’ landing positions characterization, we will present the general concepts of the paving technique used in the deployment of the several sensors types.

6.4.1. The paving of a monitored area In this subsection, we present the paving technique used to cover the monitored area meaning that the technique determines the placement of the paving patterns (i.e. the center) in the monitored area. This placement should ensure an efficient coverage of the operation field, in the sense that the number of pavement elements is minimum and the paving patterns cover the whole area (or at least a maximum part of it). One special case can be distinguished when the monitored area is a strip. Two solutions can be distinguished as depicted by Figure 6.6 and Figure 6.7. In the first solution, the paving patterns are overlapped and arranged as presented by Figure 6.6. Such a paving will not cover totally the strip but will ensure the coverage of a maximal part of it. The width of the strip WStrip and the distance δ between two successive paving patterns are depicted by the following equations in relation to the minor semi axis M inAxis and major semi axis M ajAxis.  WStrip = 2M inAxis (6.21) 2M ajAxis − δ > 0

To minimize the Area A non covered by the paving patterns, we should bring closer the centers of the paving patterns (e.g. Area A closer to 0 and δ closer to 0).

Figure 6.6.: A maximal part paving of a strip In the second solution, the paving patterns are overlapping in such a way that their intersection belong to the frontier of the strip. In that case, we have a full coverage of the strip as depicted by Figure 6.7.

6.4. CHAPTER DWBS: PAVING 6. DWBS: BASED A DISTRIBUTED DEPLOYMENT WIRELESS TECHNIQUE BORDER ANDSURVEILLANCE ARCHITECTURE SYSTEM

Figure 6.7.: A total paving of a strip The calculations shows that the optimal value of the distance between two successive paving patterns δ is given by the following equation. v u  2 ! u W Strip δ = 2t 1 − (M ajAxis)2 (6.22) 2M inAxis

Proof. To have the optimal distance δ between two paving patterns, we should determine the intersection between the ellipse representing the paving pattern and the horizontal border of the strip. The horizontal border corresponds to the coordinate on the Oy axes equal to W strip/2. The equation of the paving pattern is given by the following relation. 

x M ajAxis

2

+



2 y =1 M inAxis

. Substituting the variable y with the value Wstrip/2, we deduce that the value of the coordinate x is equal to. v u  2 ! u W Strip x = ±t 1 − (M ajAxis)2 2M inAxis . By consequence, the best value that can be chosen for δ is given by Equation 6.22. Knowing that the length of the j t h strip LStripj is finite, the number of needed paving patterns NP Ej to cover it is given by the following equation.   LStripj (6.23) NP E j = δ The determined number of paving patterns is minimum; Since, if the number of paving patterns of a minimum coverage is less than NP Ej , the strip will not be totally covered since the centers of two consecutive paving patterns will have a distance higher than δ. It can be easily seen that to cover the strip totally using the second method needs more paving patterns than covering it partially using the first method. A solution to get a paving of a general monitored area by paving patterns is to partition the area into strips and to cover every strip using one of the above mentioned methods. A total coverage using the method depicted by Figure 6.7 is depicted by Figure 6.8.

6.4.2. Architectural issues In this subsection, we define the architecture of a Distributed Wireless Border Surveillance system called DWBS capable of monitoring the physical presence of intruders in a monitored

6.4. CHAPTER DWBS: PAVING 6. DWBS: BASED A DISTRIBUTED DEPLOYMENT WIRELESS TECHNIQUE BORDER ANDSURVEILLANCE ARCHITECTURE SYSTEM

Figure 6.8.: The paving of a large area area. To ensure efficient border surveillance, the network DWBS is a three layered hierarchical network, where layers are built on three types of nodes: BSN, DRN and DDN. The details of the functionalities of the several nodes are given in Section 5.3. For the network topology, we will consider a multi linear network called DWBS. The topology of the considered network is composed of many lines. Each line is organized as a thin linear subnetwork. On the lines are deployed BSNs and DRNs. The topology of the network considered is represented by Figure 6.9.

Figure 6.9.: The topology of DWBS We propose an aerial deployment of DWBS in such a way that the radio connectivity and efficient coverage are ensured. An airplane will be moving and at regular intervals throws one

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.5. DWBS: THE DETERMINISTIC DEPLOYMENT CASE SYSTEM or more sensors (DRNs or BSNs). These sensors are supposed to land on the linear form to provide a thin linear network. The aerial deployment of the sensors is depicted by Figure 6.10.

Figure 6.10.: The aerial deployment of the sensors The path followed by the sensor when landing and the shape of the area (paving pattern) in which a sensor can land are detailed in Section 6.3.

6.5. DWBS: the deterministic deployment case In this section, we present the deterministic deployment method for the sensors. We assume in the followings that the sensors’ landing possible position has an elliptic form. We consider the deterministic case where the wind forces gives that the sensors lands in the center of the paving pattern.

6.5.1. The deployment model for DRNs For the sake of model clarity we denote by the ellipse ED the set of the positions in which a DRN can land when thrown from the airplane for the maximal value of the wind force. To build a solution let us first partition the monitored area into strips of width WStrip separated by ∆. Two successive paving patterns ED are separated by δD . We suppose that the communication range Rc verifies the following relation stating that any DRN can communicate with its neighbors and any position in the strip can be reached by a DRN. s   2 WStrip 2 δD Rc ≥ + (6.24) 2 2 An example of the area paving for the DRNs is represented by Figure 6.11. We consider the simplest case in which all the DRNs land in the center of the paving patterns which are carriers of the communication discs of the DRN. That case occurs when the lateral wind forces FWy on y-axis is equal to 0 and the frontal wind force FWx on x axis is fixed. To ensure connectivity, the deployment of DRNs should be made in such a way that the paving patterns positions satisfy three constraints: 1. The first constraint states that each point in the monitored area should be at least reachable by a DRN. To ensure this constraint, all the monitored area should be covered by the DRNs communication discs to ensure that any position is reachable by at least

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.5. DWBS: THE DETERMINISTIC DEPLOYMENT CASE SYSTEM

Figure 6.11.: An area paving for the DRNs one DRN. This deployment goal is resolved if the paving patterns positions verify the following formula.

M onitoredArea ⊆

N[ PE

Disc(P Ei , Rc )

(6.25)

i=1

where: - NP E is the number of paving patterns (the number of deployed DRNs); - Disc(P Ei , Rc ) is the DRN communication disc centered on the center of the ith paving pattern P Ei and of radius Rc . Each strip will be covered by the paving patterns using the method presented by Figure 6.7. Figure 6.12 shows the resulting coverage by the radio communication discs carried by the paving patterns.

Figure 6.12.: The strip paving for the DRNs To ensure the coverage expressed by Formula 6.25, we divide the whole area into strips separated by ∆ = 2d, with d is given by the following formula. r WStrip δD d = Rc2 − ( )2 − >0 (6.26) 2 2

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.5. DWBS: THE DETERMINISTIC DEPLOYMENT CASE SYSTEM The value of d is >0 because the communication range Rc is greater than the distance that separates the center of the paving pattern and the intersection of the paving pattern and the strip. As illustrated by Figure 6.12, to avoid having a non covered area, the communication discs should intersect. The resulting covered area around a strip is of width WStrip + 2d which explains the value chosen of ∆ = 2d expressed by Equation (6.26). Then, the minimum number NP E of DRNs needed to have a total cover of the monitored area is given by the following expression. NStrips

NP E =

X

NP E j

(6.27)

j=1

where: - NStrips is the number of the strips; - NP E j is the number of paving patterns in the j th strip given by Equation 6.23; We remark that in the general cases, the communication range Rc verifies the Inequation , the strips should intersect because (6.24). We notice that in the case where Rc < W Strip 2 the discs of range Rc cannot cover the whole strip. 2. The second constraint is that any two neighboring nodes belonging to the same strip should be connected. Given that the DRNs are supposed to land in the center of ED , then the distance between two successive paving patterns δD should verify the following inequation. δD ≤ Rc (6.28) 3. The third constraint is that the strips should be connected to avoid having an isolated part of the network. At this end, the distance between two strips ∆ have to be well chosen to permit inter strips communication. To ensure that the strips remain connected, the value of ∆ should be controlled and verifies the following inequation. ∆ + WStrip ≤ Rc

(6.29)

The previously cited constraints and the chosen values of δD and ∆ ensure the constraints but are only valid for the deterministic case deployment. For the deterministic case, we supposed that the DRN lands in the center of the paving pattern. The general case where the DRN lands randomly in any position in the paving pattern will be deeply analyzed through mathematical models in Chapter 7.

6.5.2. The deployment model for BSNs In this section, we present the adapted deployment model for the BSNs within a paving pattern that we assume with elliptic form. In the DRNs deployment, we divided the monitored area into strips. When considering the BSNs, the deployment goal is to ensure a linear coverage of the lines centered on the strips.

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.5. DWBS: THE DETERMINISTIC DEPLOYMENT CASE SYSTEM

Figure 6.13.: The choice of deployment positions of BSN Figure 6.13 illustrates an example of BSNs deployment that ensures full coverage of the line. On the same figure is shown the repartition of the paving patterns EB that carry the sensing disks of the BSNs. To have a full coverage of the line, the distance δB between two paving patterns EB should be ≤ 2Rs . Given that the BSNs are supposed to land in the center of the paving pattern EB , if the distance δB is > 2Rs then the centers of the coverage discs are distanced by more than 2Rs . By consequence there is a line of segment between the two successive BSNs that is not covered. Knowing that the length of the j t h strip LStripj is finite and the sensing range Rs , the number of BSNs (or paving patterns) needed to cover it is given by the following equation.   LStripj (6.30) NBSNj = 2Rs Then, the minimum number of BSNs needed to have a linear coverage of NStrips belonging to the monitored area is given by the following expression. X

NBSN j

(6.31)

j≤NStrips

In the presented deployment scenarios, the distance δB between two BSN paving patterns is 2Rs permits a coverage of a line. As well as the distance δB becomes less than 2Rs as well as the area monitored by the BSNs becomes larger because the sensing disks will intersect. Then, if the designer of the network needs that the area sensed is larger that a line, it should reduce the distance δB to have the required coverage quality. An example of BSNs deployment that ensures thick linear surveillance is depicted by Figure 6.14.

Figure 6.14.: BSNs thick linear surveillance The optimal distance between two paving patterns δB that permits a total coverage of a strip of width WStrip Senseded is given in the following formula. s   WStrip Senseded 2 δB = 2 Rs2 − (6.32) 2

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.6. PERFORMANCE EVALUATION SYSTEM

6.6. Performance evaluation In this section, we present the results of conducted simulations to evaluate the performances of the deterministic deployment of DWBS. We will respectively evaluate the deployment of the DRNs and the BSNs.

6.6.1. Evaluation of the DRNs deployment In the first simulation, we consider an area A to be monitored of size 1000 ∗ 1000 m2 . We varied the communication range Rc in [10,50] meters and measured the number of deployed sensors and the number of points that are reached by at least one DRN. We considered three deployment strategies: • a deterministic deployment of DWBS. The number of deployed nodes and repartition is done as presented in Section 6.5.1; • a uniform random deployment. The sensors are deployed uniformly in a random manner. The number of deployed sensors is given by the following formula N=

2 ∗ mes(A); πRc 2

• a uniform random deployment. The number of sensors is equal to the number of sensors deployed for DWBS method. The aim of this simulation is to compare the DWBS to the random uniform deployment strategy considering different densities of sensors. We mention that the number of deployed sensors for DWBS is less than the number of sensors deployed for the random deployment case. The results of the conducted simulations are illustrated in Figure 6.15.

Figure 6.15.: The number of connected points When comparing the DWBS deployment strategy and the random uniform deployment, the number of points connected and reached by at least one DRN is higher for the DWBS deployment. In addition regardless to the communication range of the DRNs, DWBS gives a percentage of connected points very close to 100%.

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.6. PERFORMANCE EVALUATION SYSTEM

6.6.2. Evaluation of the BSNs deployment In this part of the simulations, we present the results of conducted simulations to evaluate the performances of the BSNs deployment. We fixed the sensing range and varied the width of the strip in the interval [20,32]. For each value of the strip width, we evaluated the number of deployed nodes and the percentage of covered points of three deployment strategies: • The BSNs deterministic deployment: We deployed the BSNs as presented in Section 6.5.2; • The uniform random deployment for 1-coverage: The number of BSNs is determined using the random deployment to provide 1-coverage; • The uniform random deployment for 2-coverage: We deployed a dense network to provide 2-coverage. The 2-coverage is only used to have a more dense network, but in the required coverage quality each point should be covered by one BSN. Figure 6.16 depicts the number of sensors returned by each one of the aforementioned deployment strategies.

Figure 6.16.: The number of deployed BSNs Figure 6.17 represents the percentage of sensed points for each one of the deployment strategies. The first remark that can be deduced is that regardless to the width of the strip, the deterministic case of DWBS gives a coverage percentage close to 100%. We can remark that for the 1-coverage random deployment, we have less deployed BSNs than the DWBS case. But when observing the quality of coverage, the percentage of covered points is at max equal to 50% which is a bad coverage quality. In addition, even if a more dense network is deployed (the random case for 2-coverage) the quality of coverage does not reach the quality given by DWBS. These simulations demonstrates the efficiency of the deterministic case of DWBS either for coverage or connectivity.

CHAPTER 6. DWBS: A DISTRIBUTED WIRELESS BORDER SURVEILLANCE 6.7. CONCLUSION OF THE CHAPTER SYSTEM

Figure 6.17.: The percentage of covered points

6.7. Conclusion of the chapter In this chapter we developed a Distributed Wireless Border Surveillance system (DWBS) for an efficient border surveillance. The main contributions aim at presenting a deployment model that takes into account realistic parameters such as the wind, the altitude and velocity of the airplane from which the sensors are thrown. We conducted mathematical calculations to determine the landing positions and the shape of the landing area of the sensors. This landing area was referred to as a paving pattern. The novelty of the DWBS deployment method is that it is based on a paving of the monitored area using paving patterns. Based on these parameters, we presented the general architectural aspects of the network used. We also proposed a deterministic deployment based on a paving technique to ensure both connectivity and coverage of a monitored area. In the next chapter, we will consider more general cases and present the random deployment based on the paving technique.

DWBS: The controlled random deployment case

7

7.1. Introduction In Chapter 6, we introduced the DWBS network based on an aerial deployment of sensors. The main contribution presented in the previous chapter is the paving based deployment technique. This deployment is based on an analysis of the paving pattern in which can a sensor land when thrown from the airplane in relation with the several parameters in particular the wind forces and the velocity of the airplane. We presented in the previous chapter, a deterministic deployment method based on the paving technique supposing that the wind forces are constant and the sensors land in the center of the paving patterns. In this chapter, we extend the work presented in Chapter 6 by considering the controlled random deployment case for DWBS. The main contributions added in this chapter are the following: • A mathematical model is built to provide a tight control on the quality of coverage for communication of the network built on the sensors. In particular, it allows to compute the probability of connectivity by controlling a set of parameters including dropping point locations, number of dropped DRNs, and dropping altitude. • A mathematical model that estimates the sensing coverage quality of the network is built. The aim of this model is to present a solution that can be used to evaluate the coverage performance of the network using light computation resources. In the present chapter, we present the controlled random deployment case of DWBS. The rest of the chapter is organized as follows. In Section 6.2 will be presented the controlled random deployment for the DRNs. Based on this deployment model, we present in the same section through mathematical propositions the model impact on DRN connectivity. In Section 6.3, we present the model impact on inter strips connectivity. Section 6.4 is devoted to the controlled random deployment model for BSNs and its impact on the quality of coverage. In Section 6.5, we compare the DWBS controlled random deployment to the random deployment method. Section 6.6 assesses the efficiency of the connectivity and coverage proposed models in relation to the model parameters through some conducted simulations. In Section 6.7, we analyze the global performances of the DWBS network. Finally, Section 6.8 concludes the chapter.

119

7.2. CHAPTER THE DRNS 7.CONTROLLED DWBS: THE CONTROLLED RANDOM DEPLOYMENT RANDOM DEPLOYMENT CASE CASE

7.2. The DRNs controlled random deployment case In this section, we consider the random case for the DRNs deployment, meaning that we are not sure that the DRN lands in the center of the paving pattern ED but it can be in any position of this expected area. We analyze the impact of the presented model on the network performances related to the connectivity of DWBS. We consider in this section many cases of the DRNs deployment parameters and evaluate for each one the probability of the DWBS network connectivity. The simplest case corresponds to the following values of the model parameters.   Rc > 2WED where:



(7.1)

δD = aj+1 − aj ≤ WED

• WED is the width of the ellipse ED . • aj is the throwing position of DRNj . • δD = aj+1 − aj is the distance between two successive throwing positions. The values of δD is constant. Considering the most distant possible DRN positions consists at considering the maximal value of δD which is WED . Then, the maximal possible distance between two neighboring DRNs is equal to 2WED . Given the relation Rc > 2WED then the most distanced DRNs are separated by less than Rc and by consequence they can surely communicate together. Based on this case analysis, we will have a fully connected DRN line with a probability equal to 1. We consider in the followings, the most generic possible cases for the values of the deployment parameters and determine in these cases the probability to have a full DRN connected line. This case corresponds to the values of Rc and δD corresponding to equation (7.2) or equation (7.3).

δD ≤ WED ≤ Rc ≤ 2WED

(7.2)

δ D ≤ Rc < W E D

(7.3)

For the values considered in Equation (7.2) and Equation (7.3), we are sure that for any position of DRNj , we will have at least one possible position of DRNj+1 that permits connectivity because the distance between the two ellipses δD is less than Rc . Based on the same equations, the furthest possible distance between DRNj and DRNj+1 is δD + WED . The possible communication range Rc may be less than δD + WED which means that the DRNs may be not connected. We can deduce that in this case the probability that two neighboring DRNs are connected is non equal to 0 but is not equal to 1. Figure 7.1 illustrates an example of this case. If the DRNj is located in an initial position, the neighboring DRNj+1 will be connected to DRNj only if it lands in the hashed area Sj+1 .

7.2. CHAPTER THE DRNS 7.CONTROLLED DWBS: THE CONTROLLED RANDOM DEPLOYMENT RANDOM DEPLOYMENT CASE CASE

Figure 7.1.: The DRN neighbors connectivity This surface is relative to the position at which is located the sensor DRNj . Then the probability that two neighboring DRNs are connected is not equal to 1. The DRNj can be located in any position in the ellipse ED of width WED . We consider that the possible positions in the ellipse ED are strips of very small width ∆W . We denoted by Sj,k,∆W the domain of the k th strip of width ∆W in which can be located DRNj . When DRNj is in the k th strip Sj,k,∆W , we correspond a domain Sj+1 (k) in which should be DRNj+1 to permit connectivity between the two DRNs. Figure 7.1 depicts an example of decomposition into strips and the resulting constraints for two neighboring DRN connectivity. Let PSj,k,∆W be the probability that DRNj is in the strip Sj,k,∆W and PSj+1 (k) be the probability that the DRNj+1 is in the surface Sj+1 (k). Theorem 2.1 The probability PLC of having a DRN line composed of ND DRNs connected, is given by the following relation.

PLC =

NY D −1

P Cj,j+1 ,

(7.4)

j=1

where P Cj,j+1 is the probability that the DRNj and DRNj+1 are connected, and it is given by the following equation. ⌈WED /∆W ⌉

P Cj,j+1 =

X

PSj,k,∆W PSj+1 (k) .

(7.5)

k=1

Proof

Every line of DRNs in a strip is composed of ND DRNs. As shown in Figure 7.2, to have the linear connectivity, every pair of neighboring nodes should be connected. Then, the probability of line connectivity PLC is the product of the connectivity probability P Cj,j+1 for each two neighboring nodes. The product loop on counter j is used to consider all the DRNs belonging to the strip. The Probability of Line Connectivity PLC is then given by Equation 7.4.

7.2. CHAPTER THE DRNS 7.CONTROLLED DWBS: THE CONTROLLED RANDOM DEPLOYMENT RANDOM DEPLOYMENT CASE CASE

Figure 7.2.: The DRN positions for Line Connectivity The probability of connectivity between the two neighboring DRNs P Cj,j+1 is the product of the probability PSj,k,∆W that DRNj is in the strip Sj,k,∆W and the probability PSj+1 (k) that the DRNj+1 is in the related area Sj+1 (k). In Equation (7.5), the loop on the counter k is used to consider all the possible strips Sj,k,∆W in which can be DRNj . Then, the probability that DRNj and DRNj+1 are connected is given by Equation (7.5). The value of ∆W has not an impact on the resulting value of P Cj,j+1 , because for any value of ∆W we will consider all the area in the ellipse ED and then consider all the possible positions. In the case of a uniform probability of a DRN landing in ED , the expression of the probabilities PSj+1 (k) and PSj,k,∆W are given in the following proposition. Proposition 2.1 The probabilities PSj+1 (k) and Pj,k,∆W are given by the following formulas. PSj+1 (k) =

mes(Sj+1 (k)) mes(ED )

(7.6)

PSj,k,∆W =

mes(Sj,k,∆W ) mes(ED )

(7.7)

The probability of being in a subarea of an ellipse is equal to the division of the area of the sub ellipse and the area of the whole ellipse. The calculation of the measure of a subellipse surface is detailed in Appendix A. Remarks. 1. In Equations 7.6 and 7.7, we considered the case in which the probability of the DRN landing in ED is uniform. In Section 6.3.2, we have shown that in the general case the landing position of a DRN in the ellipse ED is not uniform. In that case, ED is divided into n i , i=1,2,...,n relatively to the probabilities of wind velocities and directions. We sub areas ED i . denote by PE i the probability that the sensor lands in ED D

Figure 7.3.: The decomposition of the ellipse ED An illustration of that case is depicted by Figure 7.3. The probabilities PSj+1 (k) and PSj,k,∆W for the non uniform case are given by the following formulas.

7.3. CHAPTER THE DRNS 7.INTER DWBS: STRIPS THE CONTROLLED CONNECTIVITY RANDOM DEPLOYMENT CASE

PSj+1 (k) =

n i (k)) X mes(Sj+1 i=1

PSj,k,∆W =

i ) mes(ED

n i X mes(Sj,k,∆W ) i=1

i ) mes(ED

PE i

(7.8)

PE i

(7.9)

D

D

2. In the particular case where Rc < δD , the probability of linear connectivity is given by Theorem 2.1. Considering this case, many possible positions of DRNj in its paving pattern cannot have a set of DRNs distanced of less than Rc and then cannot be connected to the neighboring DRNj+1 .

Figure 7.4.: A case where connectivity is impossible As depicted by Figure 7.4, if DRNj node is located in the surface delimited by the positions in the interval [aj , aj+1 − Rc[, it will have no probability to be connected to its neighbor. Then DRNj have to be located in the surface delimited by [aj+1 − Rc , WED ]. We notice then that for some values of the counter k, the term PSj,k,∆W PSj+1 (k) in Equation (7.5) is equal to Zero. The probability P Cj,j+1 that DRNj and DRNj+1 are connected in that case is given by the following formula. ⌈WED /∆W ⌉

P Cj,j+1 =

X

PSj,k,∆W PSj+1 (k)

(7.10)

a −Rc ⌉ k=⌈ j+1 ∆W

7.3. The DRNs inter strips connectivity In this section, we study the impact of the proposed deployment model on the inter strips connectivity. In the previous section, we have determined the probability of linear connectivity in a linear strip. But as presented, we deploy the DRNs in many strips that have to be connected. In this section, we study the impact of the deployment model on the inter strips connectivity. Every neighboring strips have to be connected in the sense that two neighboring strips have DRNs that can communicate together. In fact, for a given DRN is Stripi , we assume that there is a DRN in the next Stripi+1 that will ensure the connectivity. Then, we propose an

7.3. CHAPTER THE DRNS 7.INTER DWBS: STRIPS THE CONTROLLED CONNECTIVITY RANDOM DEPLOYMENT CASE expression that evaluates the probability that the DRN in Stripi is connected to a DRN in Stripi+1 . In a first step, we determine the probability that a DRN (x, y) located in the position (x, y) is not connected to any DRN located in the adjacent strip.

Figure 7.5.: The needed intersection area for inter strips connectivity As presented by Figure 7.5, if the DRN is located in (x,y), a DRN belonging to the next strip have to be in the hatched area to have inter strips connectivity. Proposition 3.1 The probability PN ISCD (x, y) that the DRN located in (x, y) is not connected to any DRN of the next strip is given by the following formula.

PN ISCD (x, y) =

n=N (x,y) 

Y

n=1

where:

mes(IP En ,DRN (x,y) ) 1− mes(P En )



(7.11)

• N (x, y) is the number of paving patterns that intersects with the connectivity area of DRN (x, y); • IP En ,DRN (x,y) is the intersection surface between the nth paving pattern and the connectivity area of the DRN when located in the position (x, y). Proof As presented by Figure 7.6, the connectivity area of the DRN intersects with one or more paving patterns. Each paving patterns surely contains one DRN. To be not connected to the next strip, each one of the other DRNs (belonging to the next strip) must not be in this intersection area. The number N (x, y) of paving patterns that intersects with the connectivity area of DRN (x, y), can be easily computed in relation to Rc , the distance inter strips ∆ and the positions of the DRN(x,y). In other words, N (x, y) is the number of the DRNs that can be connected to the DRN when being in (x,y). The counter n is used to consider all the N (x, y) paving patterns. Given that IP En ,DRN (x,y) is the intersection surface between the nth paving pattern and the connectivity area of the

7.3. CHAPTER THE DRNS 7.INTER DWBS: STRIPS THE CONTROLLED CONNECTIVITY RANDOM DEPLOYMENT CASE

Figure 7.6.: The intersection with the paving patterns DRN, the probability PN ISCD (x, y) is then given by Equation (7.11). In the following theorem we determine the probability PISC that all the network is connected meaning that all the strips are connected. Theorem 3.1 1. The probability PISC that all the NStrips are connected is given by the following equation. PISC =

i=NStrips −1

Y

PISCi,i+1 ,

(7.12)

i=1

where PISCi,i+1 is the probability of connectivity between two successive line strips i and i + 1, and it is given by the following equation. PISCi,i+1 = 1 − PN ISCi,i+1

(7.13)

2. The probability PN ISCi,i+1 of non connectivity between two successive line strips is given by the following equation. k=NDRNi

PN ISCi,i+1 =

Y

PN ISCD

(7.14)

k=1

where NDRNi is the number of DRNs of the it h strip. 3. The probability PN ISCD that a DRN located in any position in a paving pattern is not connected to any DRN in the next strip is given by the following equation.

PN ISCD

1 = WP E .LP E

x=W Z P E y=L Z PE x=0

PN ISCD (x, y)dxdy

y=0

where: • WP E is the width of the rectangle corresponding to a paving pattern; • LP E is the length of the rectangle corresponding to a paving pattern;

(7.15)

7.4. CHAPTER THE BSNS 7. CONTROLLED DWBS: THE RANDOM CONTROLLED DEPLOYMENT RANDOM DEPLOYMENT CASE CASE • PN ISCD (x, y) is given in Proposition 3.1. Proof 1. To have a network connected, the NStrips belonging to the deployed network should be connected meaning that any two neighboring strips i and i + 1 are connected. Then the probability PISC is the product of the probabilities PISCi,i+1 for any two adjacent strips. 2. We have many DRNs at each strip line. The strips i and i + 1 are not connected only if any DRN belonging to the strip i is not connected to the next strip i + 1. PN ISCi,i+1 is then given by the formula 7.14. 3. We do not have an exact position of the DRN landing position and it can be in any position of the paving pattern area. Then for a DRN, to have the probability that it is not connected to the adjacent strip, we should not consider only a precise position (x, y) of the DRN but all the possible positions of the DRN in the paving pattern. The integrals on x and y are used to consider all the possible positions in a paving pattern. Then, PN ISCD is given by the formula 7.15.

7.4. The BSNs controlled random deployment case For the deterministic case, we supposed that the BSN lands in the center of the paving pattern. In subsection 6.5.2, we have shown that the required distance δB between two successive paving patterns EB of width WEB should be equal to 2Rs to ensure the linear coverage. But, the linear connectivity is ensured only if the BSN lands in the center of the paving pattern. In this section, we consider the random case for the BSNs deployment meaning that we are not sure that the BSN lands in the center of the paving pattern EB but can be in any position of this expected area. We suppose in this section that δB ≥ 2 WEB meaning that two adjacent paving patterns EB does not intersect. The goal of the BSN deployment is the coverage of a line in the monitored area. We analyze the impact of the presented model on the DWBS network performances by evaluating the probability PM LC that the monitored line is covered. Proposition 4.1 The probability PT N S (x) that a target T located in a position x on a segment of line of width L is not detected by any BSN is given by the following relation.

PT N S (x) =

mes(EB \ IP Ej ,Disc(x,Rs ) ) mes(EB \ IP Ej+1 ,Disc(x,Rs ) ) × mes(EB ) mes(EB )

(7.16)

where: • IP En ,Disc(x,Rs ) is the intersection area between the nth paving pattern and Disc(x, Rs ); • Disc(x, Rs ) is the disc of center x and radius Rs . Proof When a target T is located in a position x it can be detected by a BSN only if it is located in the disc Disc(x, Rs ) of range Rs and centered on x. As presented in Figure 7.7, when the target T is located on the line of width L between the centers of two successive paving patterns P Ej and P Ej+1 , the disc Disc(x, Rs ) intersects with those two paving patterns. We

7.4. CHAPTER THE BSNS 7. CONTROLLED DWBS: THE RANDOM CONTROLLED DEPLOYMENT RANDOM DEPLOYMENT CASE CASE

Figure 7.7.: The intersection between Discx,Rs and paving patterns denote by IP Ej ,Disc(x,Rs ) and IP Ej+1 ,Disc(x,Rs ) respectively the intersection areas between the disc Disc(x, Rs ) and the paving patterns P Ej and P Ej+1 . The target T is sensed by the BSN belonging to the j th paving pattern P Ej only if it is located in the area IP Ej ,Disc(x,Rs ) . The target T is not covered by the BSN belonging to P Ej and P Ej+1 if they are located respectively in (EB \IP Ej ,Disc(x,Rs ) ) and (EB \ IP Ej+1 ,Disc(x,Rs ) ). Then, we can deduce that the probability PT N S (x) is given by the Formula 7.16. Theorem 4.1 The probability PM LC that all the line to be monitored of length Llength is covered is given by the following formula.   Llength (1 − PL,N cov ) , (7.17) PM LC = L where PL,N Cov is the probability that a line of length L is not fully covered, is given by the following formula.

PL,N cov

1 = L

x=L Z

PT N S (x) dx.

(7.18)

x=0

Proof To have the probability PL,N cov that the line is not covered, we should consider all the possible positions x of a target on the line and for each position compute the probability PT N S (x). A line of width L is not covered if any point belonging to this line is not covered. In Equation (7.18), we have considered all the possible positions x on a segment of width L and all the probabilities PT N S (x). Having a full line of width LLength to be covered, we should partition the monitored line to several segments of width L. The probability that the line of width LLength is fully covered is the probability that all the segment lines of width L are covered. Then the probability PM LC that all the monitored line is covered is given by Equation (7.17).

7.5. THE COMPARISON OF THE PROPOSED MODEL TO THE RANDOM DEPLOYMENT CHAPTER 7. DWBS: THE CONTROLLED RANDOM DEPLOYMENT CASE

7.5. The comparison of the proposed model to the random deployment In this section, we present the results of the first conducted simulation. In this simulation, we evaluate the probability of DRNs line connectivity PLC and compare it to other deployment solutions. The aim of this comparison is to show that the values returned by the mathematical model proposed are close to the values returned by the classical simulations such as the random deployment. We consider that the area to be monitored is of size 1000*1000 meters. This area will be divided into strips centered on the lines that will held the DRNs and the BSNs deployed. As presented, in the previous sections, this value depends on three parameters which are respectively: (a) the communication range Rc (b) the distance between two successive deployment positions and paving elements δD and (c) the width of the ellipse WED . We conducted three scenarios of simulations to evaluate the behavior of the proposed model in relation with each one of the parameters.

7.5.1. The variation of PLC in relation with Rc We fixed the values of the distance between two paving patterns δD to 50 meters and WED to 100 meters and varied the Rc value. For each value of Rc , we computed the probability of linear connectivity. For this simulation, we evaluated at first two deployment strategies: • The first one represents the value of the DRN line connectivity PLC returned by the proposed model illustrated by Theorem 2.1; • For the second one, we performed random deployment of the DRNs. We deployed many sets of DRNs in random positions in the ellipses ED and measured the number of deployments that gives full DRN line connectivity. This case represents PLC in real deployment cases and classical simulations. The aim of those two simulations is to compare the results returned by the mathematical model using the proposed arithmetical calculations and those returned by a real deployment case. When interpreting the results represented by Figure 7.8, we remarked that despite having the same curve progression but there is a gap between the model value and the random deployment real strategy. After a thorough comparison between the proposed model operating principle and the random deployment strategy, we concluded that the main gap is due to the fact that in our model we considered all the possible combinations of deployment positions and measured the mean probability of connectivity. But, in the random deployment, despite the facts that we performed many random iterations but we are still in random positions and we will not certainly cover all the possible DRN positions in the ellipse ED . For that, we added another kind of deployment scenario. In this metric, we will not consider random positions of the DRNs but we fixed from the beginning some possible positions of the DRNs spaced by a predetermined Step and measured the connectivity probability for all the possible combinations of those positions. The value calculated is then a mean value between all the possible predetermined positions. The descriptor parameter of this metric is the Step between the fixed positions. We considered two cases, which corresponds respectively to Step=20 meters and Step=10 meters.

7.5. THE COMPARISON OF THE PROPOSED MODEL TO THE RANDOM DEPLOYMENT CHAPTER 7. DWBS: THE CONTROLLED RANDOM DEPLOYMENT CASE

Figure 7.8.: The variation of PLC in relation with Rc Then, we have four scenarios and corresponding metric values computed which corresponds respectively to the model value, the random deployment, random deployment with respectively Step=20 meters and Step=10 meters. The results of the simulations for the four metrics are represented by Figure 7.8.

7.5.2. The variation of PLC in relation with δD

Figure 7.9.: The variation of PLC in relation with δD In this simulation, we fixed the communication range Rc to 110 meters and the ellipse width WED to 100 meters. We varied the value of δD to study the impact of this parameter on PLC . Such as the previous simulation, we varied the considered parameter and measured respectively the value of PLC corresponding to the proposed model, the random deployment, random deployment with respectively Step=20 meters and Step=10 meters. The results of this simulation are represented by Figure 7.9.

7.5. THE COMPARISON OF THE PROPOSED MODEL TO THE RANDOM DEPLOYMENT CHAPTER 7. DWBS: THE CONTROLLED RANDOM DEPLOYMENT CASE

7.5.3. The variation of PLC in relation with WED In this simulation, we fixed the communication range Rc to 110 meters and the distance between two paving element δD to 50 meters. We varied the value of WED to study the impact of this parameter on PLC . Such as the previous simulation, we varied the considered parameter and measured respectively the value of PLC corresponding to the proposed model, the random deployment, random deployment with respectively Step=20 meters and Step=10 meters. The results of this simulation are represented by Figure 7.10.

Figure 7.10.: The variation of PLC in relation with WED

7.5.4. Interpretation of the simulation results In this subsection, we will comment and analyze the results of the simulations represented by Figures 7.8, 7.9 and 7.10. - The first deduction concerns the global behavior of the curves. The progression of the curve relative to the proposed model is the same than the curve for the random deployment and random deployment for prefixed DRN positions. We can deduce that the connectivity probability returned by the proposed mathematical model returns the same results of the realistic and classical conducted simulations. Given, that our model is based on light mathematical formulations, we can then use it to substitute the classical simulations based on random deployments which require higher computation resources and longer processing times; - In the simulation results, for the random deployment with prefixed DRN positions, it is clear that as well as the Step is less as well as the values of PLC are closer to the value returned by our model. Also, as well as we use more precise discrete positions as well as we are close to the model given values. We can deduce that the results of our model are closer to the classical realistic simulation scenarios and can be used to evaluate the connectivity of the deployed network;

7.6. CHAPTER THE EVALUATION 7. DWBS: OFTHE DWBS CONTROLLED CONNECTIVITY RANDOM AND DEPLOYMENT COVERAGE CASE - The third deduction is relative to the behavior of PLC in relation with each one of the aforementioned parameters. • We remark on Figure 7.8 that the PLC values grows up when the values of the communication range Rc becomes greater. In fact, when Rc is greater, two neighboring DRNs have more probabilities to be connected and by consequence all the DRN line connectivity is greater; • On Figure 7.9, the value of PLC decreases when the value of δD becomes greater. It is evident that as well as the distance between the throwing positions of the DRNs is greater as well as the DRNs are more distanced and we have less cases the provides connected DRNs; • Finally, as observed on Figure 7.10, the probability of DRN line connectivity PLC decreases when the value of WED becomes greater. In fact when the width of the ellipse ED grows up the possible distance between neighboring DRNs becomes greater. By consequence, the probability of connectivity will decrease as well as the WED values increases.

7.6. The evaluation of DWBS connectivity and coverage In this section we evaluate the DWBS performances and study the behavior of the connectivity and coverage using the proposed mathematical models.

7.6.1. The evaluation of DWBS connectivity

Figure 7.11.: The variation of PLC in relation to WED and Rc In the previous simulations, we have just observed the global behavior of the curve of PLC and verified its homogeneity with the results returned by random deployment simulations. In these simulations, we represent more generalized variations of PLC in relation with many possible values of the different parameters. We also analyze the variations of PLC to both two parameters at the same time to have more representative variations of PLC . Those simulations can also be used in dimensioning the network and help in the choice of the different parameters

7.6. CHAPTER THE EVALUATION 7. DWBS: OFTHE DWBS CONTROLLED CONNECTIVITY RANDOM AND DEPLOYMENT COVERAGE CASE values to have a better probability of network connectivity. In the first simulation scenario, we calculated the variation of PLC in relation to both Rc and WED . We fixed the parameter δD to 50 meters. We varied the value of WED between 70 and 240 meters and Rc between 50 and 155 meters. We measured the value of PLC returned by our model for all the combinations of WED and Rc values. The results of this simulation are represented by Figure 7.11. In the second simulation scenario, we calculated the variation of PLC in relation to both δD and WED . We fixed the parameter Rc to 110 meters. We varied the value of WED between 30 and 240 meters and δD between 5 and 120 meters. We measured the value of PLC returned by our model for all the combinations of WED and δD values. The results of this simulation are represented by Figure 7.12.

Figure 7.12.: The variation of PLC in relation to WED and δD In the third simulation scenario, we calculated the variation of PLC in relation to δD and Rc . We fixed the parameter WED to 100 meters. We varied the value of Rc between 50 and 155 meters and δD between 5 and 120 meters. We measured the value of PLC for all the combinations of Rc and δD values. The results of this simulation are represented by Figure 7.13.

Figure 7.13.: The variation of PLC in relation to δD and Rc The deductions that can be resulted from this simulation confirms the observations done on the previous simulations. These interpretations are represented by the followings.

7.6. CHAPTER THE EVALUATION 7. DWBS: OFTHE DWBS CONTROLLED CONNECTIVITY RANDOM AND DEPLOYMENT COVERAGE CASE • As observed on Figures 7.11 and 7.12, the probability of DRN line connectivity PLC is better for the smallest values of WED . We observe in the two Figures, that for fixed values of Rc and δD , the curve of PLC decreases as well as the value of WED becomes greater. In fact when the width WED of the ellipse ED becomes larger, the distance between neighboring DRNs is greater and by consequence the probability of connectivity will decrease; • We remark on Figures 7.11 and 7.13 that the PLC value is better for the largest values of the communication range Rc . The result is logical because when having a greater value of Rc , we have more probabilities that two neighboring nodes are connected; • On Figures 7.12 and 7.13, the best values of PLC are observed for the smallest values of δD . In real cases, when the paving patterns are closer, two neighboring DRNs will be closer and we will have more possibilities for connectivity; • Despite, the behavior of PLC in relation with the different parameters, we remark that for some combinations of the parameters values, we never reach an acceptable value of connectivity probability. For example, on figure 7.11 when WED is >130 and for all the possible Rc values we never reach the value PLC = 1. The same deduction can be realized on Figures 7.12 and 7.13. These deductions mean that the values of the model parameters are of great importance and have to be well chosen. Then a dimensioning is essential because for some values we will have a low probability of connectivity and the deployed network will not satisfy one of the important goals of the network deployment which is the connectivity. In particular, one can notice that despite the values taken by Rc , δD and WED , the probability of linear connectivity reaches high values of PLC provided that Rc ≥ δD + WED . The simulation results can be used either to evaluate the network performances or in dimensioning the network. The presented figures helps the designer of the network in the process of the several parameters choice to guarantee the connectivity performances needed by the implemented network.

7.6.2. The evaluation of DWBS coverage In this part of the simulations, we evaluate the probability of linear coverage PM LC given by Theorem 4.1. We conducted simulations to compare the proposed coverage control model to the random deployment. Such as the case of the connectivity control model, the simulations have shown that the results returned by our mathematical model have the same behavior than the random deployment simulations. We varied the width WEB of the ellipse EB between 15 and 100 meters and the coverage range Rs from 5 to 70 meters while the distance between two successive paving patterns δB is fixed to 50 meters. We computed using the proposed coverage control mathematical model, the probability that the monitored line is fully sensed. The results of this simulation are represented by Figure 7.14.

7.7. CHAPTER THE DWBS7.GLOBAL DWBS: PERFORMANCE THE CONTROLLED PROBABILITIES RANDOM DEPLOYMENT CASE

Figure 7.14.: The Probability of Linear Coverage Based on the results of this simulation, we can evaluate the behavior of the Linear coverage in relation to the different parameters. - The results show that for a fixed value of WEB , as well as the range Rs becomes greater as well as the linear coverage is more probable. In real cases when the sensing range Rs becomes greater we have a higher probability of coverage. - This simulation also shows that as well as the width of the paving pattern WEB is smaller as well as the linear coverage probability is better. This is logical because when the size of the ellipse EB is smaller we have more precision on the positions in which can the BSN land and then we have higher probabilities that the proposed deployment model gives the required quality of coverage. - In addition to the study of the linear connectivity in relation to the several parameters, the conducted simulations can be used in dimensioning the DWBS network. The results of the simulation show that the best values of PM LC are reached only for some combinations of Rs and WEB values. In fact, when WEB is fixed to 30 meters the value of Rs should be ≥ 40 meters to have a PM LC = 1. And when WEB is fixed to 70 meters the value of Rs should be ≥ 60 meters to have a PM LC = 1. In particular, one can notice that despite the values taken by Rs , δB and WEB the probability of linear coverage PM LC reaches high values provided that Rs ≥

δ B + WE B . 2

The results of this simulations are useful in the choice of the Rs and WEB values that ensures the required probability of coverage.

7.7. The DWBS global performance probabilities In the previous simulations, we evaluated using the proposed mathematical models separately the linear connectivity and coverage in the same strip. But, to have good global performance, the inter strips connectivity and the combined linear coverage and connectivity should be verified. Then to evaluate the global probability of the network performance, we should evaluate two other metrics which are respectively the probability of inter strips connectivity PISC and the probability of linear connectivity and sensing coverage denoted by PLSC .

7.7. CHAPTER THE DWBS7.GLOBAL DWBS: PERFORMANCE THE CONTROLLED PROBABILITIES RANDOM DEPLOYMENT CASE

7.7.1. The inter strips connectivity In addition to the intra strips connectivity, the deployed network should ensure the inter strips connectivity. In this simulation, we evaluate the probability of inter strips connectivity PISC given in Theorem 3.1. This probability depends on two parameters which are the communication range Rc and the distance between two adjacent strips ∆. We fixed the width of the strips W Strip to 20 meters and varied the communication range Rc between 30 meters and 90 meters and ∆ between 0 and 100 meters. For each combination of Rc and ∆ values, we evaluated PISC . The results of this simulation are given by Figure 7.15.

Figure 7.15.: The Probability of Inter Strips Connectivity PISC Based on this simulation, we can evaluate the behavior of the inter strips connectivity in relation to the different parameters. • We remark on Figure 7.15, that when the distance between two strips ∆ becomes greater, we have less opportunities that two adjacent strips are connected if Rc has a fixed value; • On the same figure, we remark that for a fixed value of ∆, the area in a strip covered by a DRN in the other strip gets larger when Rc increases and by consequence the value of PISC becomes greater; • The results of the simulation show that the best values of PISC are reached for the lowest values of ∆ and the largest values of Rc . In particular, one can notice that despite the values of Rc and ∆, the probability of inter strips connectivity becomes larger when the values of Rc and ∆ verifies Rc > ∆. As well as the value of Rc is larger than ∆ as well as we are closer to PISC = 1. The value of PISC is very close to 1 when the value of Rc verifies Rc ≥ ∆ + W Strip. The proposed model can be used in the dimensioning of the network while helping in the choice of the parameters’ values to ensure higher values of inter strips connectivity probability.

7.7.2. The linear connectivity and coverage To have good global performances, the deployment of BSN and DRN on a line should ensure both connectivity and coverage. In this simulations, we vary respectively the parameters Rc

7.8. CHAPTER CONCLUSION 7. DWBS: OF THE THE CHAPTER CONTROLLED RANDOM DEPLOYMENT CASE and Rs and determine for each combination the probability PLSC that the line is connected and sensed. We fixed the value of the paving patterns WED and WEB to respectively 100 meters and 30 meters. The distance between two DRNs or BSNs paving patterns is fixed to 50 meters. The results of this simulation are represented by the following figure.

Figure 7.16.: The Probability of Linear coverage and connectivity PLSC The results of the simulation show that the best values of PLSC are reached for the largest values of Rs and Rc . In fact, when Rs ≥ 50 and Rc ≥ 105, PLSC is > 0.8. In particular, one can notice that despite the values taken by the model parameters the probaδ +W bility of coverage and connectivity can reach high values provided that Rs is closer to B 2 EB and Rc is closer to δD + WED . To reach a high probability PLSC equal to 1, the values of the parameters should verify δ B + WE B and Rc ≥ δD + WED . 2 Then the simulations conducted can give an indication about the required values of the several model parameters to ensure that PLSC is greater than a given value. Rs ≥

7.8. Conclusion of the chapter In this chapter we presented the controlled random deployment case for the DWBS network. We proposed mathematical control models to evaluate the performances of the deployed network. The proposed models permits the evaluation of the linear connectivity, the inter strips connectivity for the whole network connectivity and the coverage capabilities of DWBS. We analyzed through simulations, the performances of the proposed models. The results of the simulations can be used as a dimensioning model and helps the designer of the network in the choice of the deployment parameters.

Conclusions and perspectives

8

The main outcome of this thesis is 3-fold: a) the development of a set of deployment techniques suitable for WSNs. We considered many constraints to be satisfied. We considered two main applications which are the target surveillance and border surveillance; b) the development of a security strategy based on Group key establishment and tunneling; and c) the establishment of scheduling strategies to extend the network lifetime. We proposed two scheduling strategies having two different metrics of decision. This chapter summarizes the achieved work in this thesis, showing the major contributions in every axis. After that, it presents some of the main open problems that remain unsolved and can be addressed by future works.

8.1. Deployment strategies In this thesis we set up several deployment methods suitable for WSNs. The deployment techniques are presented in the followings. Deployment based on radio irregularities: The first deployment strategy is based on the radio propagation effects such as path loss and shadowing. In this method, we studied the impact of the radio irregularities on the sensing range of a sensor. Based on this study, we proposed a deployment technique that offers a good quality of sensing under the impact of radio irregularities. An energy based deployment technique: The transmission and sensing ranges of a sensor depend on the energy. When the energy of the nodes decays, the coverage range of the sensors becomes less and then the area covered by the node is smaller. We studied the impact of the energy decay on the sensing performance. Therefore, we proposed a deployment technique that takes into account this constraint and provides a good quality of sensing during all the network lifetime regardless the decay of the energy. A deployment based on geographical patterns: We proposed another deployment technique that provides a non uniform and dynamic sensing of the monitored area. This uniformity is function of the geographical patterns of the monitored area. In the proposed method, we divides the monitored area into subareas classified into different categories (river, mountain...). To each category we affect a quality of sensing in function of an initial probability of targets apparition. We proposed in this method an iterative and dynamic deployment. The historical

137

8.2. SCHEDULING SCHEMES CHAPTER 8. CONCLUSIONS AND PERSPECTIVES tracking of the targets in the monitored area will be used to update the probability of targets presence in the subareas, and by consequence the quality of sensing of the subareas. This update can be done either with a new nodes deployment or a move of existing nodes to other subareas. A thick linear network for Border surveillance: We setup a framework for border surveillance using a hierarchical Wireless Sensor Networks. We used a thick linear topology to permit the tracking of the targets near the border line. We proposed a deployment technique of the several nodes of the network to ensure sensing of the events and communication between all nodes. We proposed a soft routing technique suitable for the proposed architecture to ensure communication between all the nodes and a non costly update of the routing entries. DWBS, A Distributed deployment scheme for Wireless Border Surveillance: We built a WSN to ensure the surveillance of infiltration within a large area near the border line. In the proposed deployment model, the sensors are supposed to be thrown from an airplane. The main contribution is the setup of a new deployment method, which consists at paving the monitored area with paving patterns of predetermined shapes. The paving patterns are not randomly chosen but have predetermined shapes translating the environment conditions. Through the analysis of the forces applied on the sensors when landing, we characterized precisely the shapes of the paving patterns. The paving based deployment strategy ensures both coverage and connectivity. We proposed mathematical models that permit the compute of sensing and coverage probabilities in relation to the environment characteristics. The mathematical models are built to provide a tight control on the quality of sensing and communication and evaluate the network sensing efficiency. In addition, the mathematical models are developed to plan and dimension the deployed network.

8.2. Scheduling schemes We proposed in this thesis two scheduling scheme to ensure a better energy consumption control. An energy based scheduling scheme: We proposed in this thesis a deployment method that provides an efficient sensing quality when the sensors’ energy decays. This method is based on a deployment of redundant sensors to ensure the coverage in the worst cases. Therefore, we proposed a scheduling scheme that takes benefits of this redundancy. During the lifetime of the network, depending on the sensors’ residual energy, only the needed sensors are activated and the others are in sleep mode. Operating at this manner, the sensing quality is preserved and the lifetime of the network is extended because the sensors are used only when needed. A targets movement based scheduling scheme: We set up another scheduling scheme. The factor of choice of the sensors status depends on an anticipation of the targets’ possible locations. This method is based on a thorough analysis of the targets’ recorded movements and anticipate their future possible positions. Depending on this, the monitored area is non uniformly covered: the most likely areas to contain targets are k-covered and the others are

8.3. SECURITY PROTOCOL CHAPTER 8. CONCLUSIONS AND PERSPECTIVES 1-covered. This method preserves the sensing of the area and extends the network lifetime because the sensors dynamically alternate between active and sleep status.

8.3. Security Protocol We proposed a dynamic tunneling and key management security protocol called DynTunKey. This security protocol ensures authentication of the nodes and provides mechanisms for data integrity and confidentiality. The main contribution of the proposed protocol is the use of the tunnels in the WSN context. To the best of our knowledge, we are the first to propose the use of tunnels in WSNs. The tunnels created use a group key to permit communication between many nodes. The group key is established in a distributed and secured process. We also introduced a new concept called the CSA (Cluster Security Association) which is an abstraction of the established many-to-many tunnels and represents shared security attributes between many sensor nodes. The proposed security protocols ensures self organization.

8.4. Future Works The final section of this thesis opens new fields of research showing problems that remain still unsolved and which can be addressed in future works. • In general cases, the irregularity models define the sensing domain as default values outside the sensors area of deployment. These models are only approximate after deployment and generate non negligible errors. Then, the deployment model should be adapted to the deployment context. In a future work, we will propose a novel radio irregularity scheme to cope with the varying climate and vegetation characteristics of the environment, where the monitoring platform is deployed. A technique based on the neighboring sensors collaboration may enhance the irregularity perception near a sensor. • To extend the network lifetime, we will propose in a future work a solution based on sensors able to change their sensing domains. The sensors reduce their sensing range to consume less energy and extend their lifetime. The network will be composed of nodes having different sensing ranges. We should then adapt the deployment models taking into account the non uniformity of the sensors’ coverage domains. • Development of node failure management system. The system developed should guarantee two complementary tasks. The first one is an algorithm of detecting, locating and predicting power-deficient sensors and relay nodes. The second is an algorithm of replacing the power-deficient sensors and relay nodes before failure. An adaptive version of the DWBS will be developed to deploy the additional sensors to substitute the failed nodes. This is an important feature for reliability of sensor networks by replacing power-deficient sensors and relay nodes before failure. • In a future work, we will propose tracking techniques to follow the movement of the intruders in the monitored area. The sensor network used is a multi linear network like DWBS. The first and final sensor lines of the network are always activated ensuring,in some sense, a total coverage of the input and output edges. To extend the network

8.4. FUTURE WORKS

CHAPTER 8. CONCLUSIONS AND PERSPECTIVES

lifetime, a sensor belonging to the other lines is activated if a target will cross its sensing domain. The first contribution of the tracking technique is the continuous tracking using less energy consumption because the sensors are activated when needed. The second contribution is the prediction of the target’s exit time and the output segment on the edge.

Bibliography [1] D.C. Cox, R. Murray, and A. Norris. 800mhz attenuation measured in and around suburban houses. AT&T Bell Laboratory Technical Journal, 673(6), 1984. [2] R.C. Bernhardt. Macroscopic diversity in frequency reuse systems. IEEE Journal on Selected Areas in Communications, SAC5:862–878, 1987. [3] C. Gui and P. Mohapatra. Power conservation and quality of surveillance in target tracking sensor networks. In Proc. ACM MobiCom’04, pages 129–143, September 2004. [4] J. Ma, Q. Chen, D. Zhang, and L.M. Ni. Empirical study of radio signal strength in sensor networks. In Technical Report, March 2006. [5] G. Zhou, T. He, S. Krishnamurthy, and J. Stankovic. Impact of radio irregularity on wireless sensor networks. In The Second International Conference on Mobile Systems, Applications, and Services (MobiSys’04), Boston, MA, June 2004. [6] M. Woehrle, D. Brockhoff, and T. Hohm. A new model for deployment coverage and connectivity of wireless sensor networks. In Technical Report, September 2007. [7] J.M. Molina-Garcia-Pardo, A. Martinez-Sala, M.V. Bueno-Delgado, E. Egea-Lopezand L. Juan-Llacer, and J. Garca-Har. [8] J. Zhang, H. Zhang, J. Lv, and L.Yin. Research on impact of radio irregularity on self-localization of wsn. Key Engineering Materials, 474-476:2161–2166, 2011. [9] T. Rappaport. Wireless Communications: Principles and Practice. Prentice Hall PTR, Upper Saddle River, NJ, USA, 2nd edition, 2001. ISBN 0130422320. [10] M. Hamdi, N. Boudriga, and M. S. Obaidat. Whomoves: An optimized broadband sensor network for military vehicle tracking. International Journal of Communication Systems, 21:277–300, 2008. ISSN 1074-5351. [11] X. He and X. Gui. The localized area coverage algorithm based on delayed start scheme for wsn. Journal OF Software, 4(3), May 2009. [12] T. Camp, J. Boleng, and V. Davies. A survey of mobility models for ad hoc network research. In Wireless communication and mobile computing (WCMC): Special issue on mobile ad hoc networking research, trends and applications, September 2002. [13] S. Kumar, T. H. Lai, and J. Balogh. On k-coverage in a mostly sleeping sensor network. In Proceedings of the 10th Annual International Conference on Mobile Computing and Networking (Mobicom), pages 144–158, 2004.

141

Bibliography

Bibliography

[14] M. Perillo and W. Heinzelman. Optimal sensor management under energy and reliability constraints. In Proceedings of the IEEE Conference on Wireless Communications and Networking, March 2003. [15] M. Cardei and D. Du. Improving wireless sensor network lifetime through power-aware organization. In ACM Wireless Networks, May 2005. [16] J. Lu, L. Bao, and T. Suda. Coverage-aware sensor engagement in dense sensor networks. In Proceedings of the International Confernece on Embedded and Ubiquitous Computing (EUC 2005), Dec 2005. [17] J. Salzmann, S. Kubisch, F. Reichenbach, and D. Timmermann. Energy and coverage aware routing algorithm in self organized sensor networks. In Proceedings of Networked Sensing Systems (INSS’07), June 2007. [18] R. Akl and U. Sawant. Grid-based coordinated routing in wireless sensor networks. In Proceedings of 4th IEEE Conference on Computer Communications and Networking, January 2007. [19] G. Wang, G. Cao, T. La Porta, and W. Zhang. Sensor relocation in mobile sensor networks. In Proceedings of the 24th International Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM’05), Miami, FL, March 2005. [20] R.P. Liu, G. Rogers, S. Zhou, and J. Zic. Topology control with hexagonal tessellation. In International Journal of Sensor Networks, pages 91–98, 2007. [21] R.P. Liu, G. Rogers, and S. Zhou. Honeycomb architecture for energy conservation in wireless sensor networks. In Proceedings of IEEE Globecom, San Francisco, November 2006. [22] Z. Jiang, J. Wu, A. Agah, and B. Lu. Topology control for secured coverage in wireless sensor networks. In Proceedings of the 3rd IEEE International Workshop on Wireless and Sensor Networks Security (WSNS’07), October 2007. [23] Y. Wu, S. Fahmy, and N.B. Shroff. Optimal qos-aware sleep/wake scheduling for timesynchronized sensor networks. In Invited Sessions on Optimization of Communication Networks 40th Annual Conference on Information Sciences and Systems (CISS), Princeton, NJ, 2006. [24] M. Bagheri and M. Hefeeda. Randomized k-coverage algorithms for dense sensor networks. In Proceedings of the IEEE INFOCOM 2007 Minisymposium, Anchorage, AK, pages 2376–2380, May 2007. [25] B. Wang, K.C. Chua, V. Srinivasan, and W. Wang. Scheduling sensor activity for point information coverage in wireless sensor networks. In Proceedings of the WiOpt, 2006. [26] Y. Cai, M. Li, W. Shu, and M-Y. Wu. Acos: an area-based collaborative sleeping protocol for wireless sensor networks. In International Journal of Ad Hoc and Sensor Wireless Networks, 2006.

Bibliography

Bibliography

[27] B. Liang, J. Frolik, and X.S. Wang. A predictive qos control strategy for wireless sensor networks. In The 1st Workshop on Resource Provisioning and Management in Sensor Networks (PRMSN’05) in conjunction with the 2nd IEEE MASS, Washington DC, November 2005. [28] H. Zhang and J.C. Hou. Maintaining sensing coverage and connectivity in large sensor networks. pages 453–457. [29] A. Giusti, A.L. Murphy, and G.P. Picco. Decentralized scattering of wakeup times in wireless sensor networks. In Proceedings of the 4th European Conference on Wireless Sensor Networks (EWSN), Delft, The Netherlands, Lecture Notes on Computer Science, volume 4373, pages 245–260. Springer, January 2007. [30] B. Wang, K.C. Chua, V. Srinivasan, and W. Wang. Sensor density for complete information coverage in wireless sensor networks. In Proceedings of the EWSN 2006, Zurich, Switzerland, February 2006. [31] Zh. Dingxing, X. Ming, Ch. Yingwen, and W. Shulin. Probabilistic coverage configuration for wireless sensor networks. In Second International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM 2006), Wuhan, September 2006. [32] J.W. Branch, G.G. Chen, and B.K. Szymanski. Escort: energy-efficient sensor network communal routing topology using signal quality metrics. In Proceedings of the 4th IEEE International Conference on Networking (IEEE ICN’05), Reunion Island, France, April 2005. [33] R. Bellazreg, M. Hamdi, and N. Boudriga. Coverage control and irregular radio propagation in wireless sensor networks. In Proc. WCNC 2009, September 2009. [34] H. Chan, A. Perrig, and D. Song. Random key predistribution schemes for sensor networks. In Proceedings IEEE Symposium on Security and Privacy, May 2003. [35] W. Du, J. Deng, Y. Han, and P. Varshney. A pairwise key pre-distribution scheme for wireless sensor networks. In Proceedings of the Tenth ACM Conference on Computer and Communications Security (CCS 2003), pages 42–51, October 2003. [36] S. A. Cametepe and B. Yener. Combinatorial design of key distribution mechanisms for wireless sensor networks. In Proc. 9th European Symp. Research Computer Security, 2004. [37] L. Eschenauer and V. D. Gligor. A key-management scheme for distributed sensor networks. In CCS ’02: Proc. 9th ACM Conf. Comp. and Commun. Security, New York: ACM Press, pages 41–47, 2002. [38] R. D. Pietro, L. V. Mancini, and A. Mei. Random key-assignment for secure wireless sensor networks. In SASN ’03: Proc. 1st ACM Workshop on Security of Ad Hoc and Sensor Networks, New York: ACM Press, pages 62–71, 2003. [39] D. Liu and P. Ning. Establishing pairwise keysdistributed sensor networks. In CCS ’03: Proc. 10th ACM Conf. Comp. and Commun. Security, New York: ACM Press, pages 52–61, 2003.

Bibliography

Bibliography

[40] W. Du and al. A key management scheme for wireless sensor networks using deployment knowledge. In Proc. IEEE INFOCOM, Hong Kong, pages 586–97, 2004. [41] W. Bechkit. New key management schemes for resource constrained wireless sensor networks. In IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks, WoWMoM, Lucca, 2011. [42] T. D. Subash and C. Divya. Novel key pre-distribution scheme in wireless sensor network. In International Conference on Emerging Trends in Electrical and Computer Technology, ICETECT, pages 959–963, 2011. [43] A. Rasheed and R. Mahapatra. Key predistribution schemes for establishing pairwise keys with a mobile sink in sensor networks. In IEEE Transactions on Parallel and Distributed Systems, pages 176–184, 2011. [44] R. D. Pietro and al. Lkhw: A directed diffusion-based secure multicast scheme for wireless sensor networks. In ICPPW ’03: 32nd International Conference on Parallel Processing Workshops., IEEE Computer Society Press, pages 397–406, 2003. [45] H. Chan and A. Perrig. Pike: Peer intermediaries for key establishment in sensor networks. In IEEE Infocom, 2005. [46] S. Zhu, S. Setia, and S. Jajodia. Leap: Efficient security mechanisms for large-scale distributed sensor networks. In CCS’03: 10th ACM Conference on Computer and Communications Security, New York, ACM Press, pages 62–72, 2003. [47] I. S. Gawdan, C. O. Chow, T. Zia, and Q. I. Sarhan. A novel secure key management module for hierarchical clustering wireless sensor networks. In Proc. IEEE International Conference on Computational Intelligence Modelling and Simulation, CIMSim 2011, art. no. 6076377, pages 312–316, 2011. [48] Y. Zhang, X. Li, J. Yang, Y. Liu, N. Xiong, and A. V. Vasilakos. A real-time dynamic key management for hierarchical wireless multimedia sensor network. In Multimedia Tools and Applications, pages 1–21, 2012. [49] M. Sliti, M. Hamdi, N. Boudriga, and A. Helmy. An elliptic threshold signature framework for k-security in wireless sensor networks. In The 15th IEEE International Conference on Electronics, Circuits, and Systems, 2008. [50] D. Augot, R. Bhaskar, V. Issarny, and D. Sacchetti. An efficient group key agreement protocol for ad-hoc networks. In IEEE Workshop on Trust, Security and Privacy in Ubiquitous Computing, WoWMoM 2005, Taormina, Italy. [51] W. Diffie and M. E. Hellman. New directions in cryptography. IEEE Transactions on Information Theory, 22:644–654, Nov 1976. [52] S. Kent and K. Seo. Security architecture for the internet protocol, rfc 4301. IEEE International Conference on Electronics, Circuits, and Systems, December 2005. [53] C. Kaufman and Ed Microsoft. The internet key exchange (ikev2) protocol, rfc 4306. December 2005.

Bibliography

Bibliography

[54] D. Du, H. Xiong, and H. Wang. An efficient key management scheme for wireless sensor networks. In International Journal of Distributed Sensor Networks, 2012. [55] A. Saipulla, B. Liu, and J.Wang. Barrier coverage with airdropped wireless sensors. In Proceedings of IEEE MilCom, 2008. [56] A. Saipulla, C. Westphal, B. Liu, and J. Wang. Barrier coverage of line-based deployed wireless sensor networks. In INFOCOM, 2009. [57] K.F. Ssu, W.T. Wang, F.K. Wu, and T.T. Wu. K-barrier coverage with a directional sensing model. In International Journal on Smart Sensing and Intelligent Systems, volume 2, March 2009. [58] M. Chitnis, Y. Liang, J. Y. Zheng, P. Pagano, and G. Lipari. Wireless line sensor network for distributed visual surveillance. In Proceedings of the 6th ACM symposium on Performance evaluation of wireless ad hoc, sensor, and ubiquitous networks,Tenerife, Canary Islands, Spain, 2009. [59] G. Yang and D. Qiao. Barrier information coverage with wireless sensors. In INFOCOM, 2009. [60] S. Kumar, T. H. Lai, and A. Arora. Barrier coverage with wireless sensors. In In Proceedings of the ACM MobiCom, 2005. [61] Z. Sun, P. Wanga, M. C. Vuran, M. A. Al-Rodhaan, A. M. Al-Dhelaan, and I. F. Akyildiz. Bordersense: Border patrol through advanced wireless sensor networks. In Ad Hoc Networks, 2011. [62] B. Liu, O. Dousse, J. Wang, and A. Saipulla. Strong barrier coverage of wireless sensor networks. In Proceedings of the 9th ACM international symposium on Mobile ad hoc networking and computing ,Hong Kong SAR, China, May 2008. [63] D. Dudek, C. Haas, A. Kuntz, M. Zitterbart, D. Krger, P. Rothenpieler, D.Pfisterer, and S. Fischer. A wireless sensor network for border surveillance. In ACM Conference on Embedded Networked Sensor Systems (SenSys’09),Berkeley, CA, USA, November 2009. [64] C. E. Perkins and E. M. Belding-Royer. Ad-hoc on-demand distance vector routing. In WMCSA, pages 90–100, 1999. [65] D. B. Johnson and D. A. Maltz. Dynamic source routing in ad hoc wireless networks. In Mobile Computing, pages 153–181. Kluwer Academic Publishers, 1996. [66] C. E. Perkins and P. Bhagwat. Highly dynamic destination-sequenced distance-vector routing (dsdv) for mobile computers. In SIGCOMM, pages 234–244, 1994. [67] I. Jawhar, B.N. Mohamed, and D.P.Agrawal. Linear wireless sensor networks: Classification and applications. 34(5):1671–1682, 2011. [68] R. Bellazreg, M. Hamdi, and N. Boudriga. On the impact of irregular radio propagation on coverage control and sleep scheduling in wireless sensor networks. In Proceedings of the 8th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA 2010), Hammamet, Tunisia, May 2010.

Bibliography

Bibliography

[69] R. Bellazreg, N. Boudriga, S. An, and I. Nakouri. An heterogeneous wireless sensor networks deployment strategy for border surveillance. In Proceedings of the 10th IEEE Vehicular Technology Society Asia Pacific Wireless Communications Symposium (APWCS 2013), Seoul, Korea, August 2013. [70] M. Younis and K. Akkaya. Strategies and techniques for node placement in wireless sensor networks: A survey. Elsevier Ad-Hoc Network Journal, 6(4):621–655, 2008. [71] A. Brooks, A. Makarenko, T. Kaupp, S. Williams, and H. Durrant-Whyte. Implementation of an indoor active sensor network. In Proceedings of the 9th International Symposium on Experimental Robotics, Singapore, June 2004. [72] V.A. Petrushin, G. Wei, O. Shakil, D. Roqueiro, and V. Gershman. Multiplesensor indoor surveillance system. In Proceedings of the 3rd Canadian Conference on Computer and Robot Vision (CRV’06), Quebec city, June 2006. [73] E.S. Biagioni and G. Sasaki. Wireless sensor placement for reliable and efficient data collection. In Proceedings of the 36th Annual Hawaii international Conference on System Sciences (HICSS’03), Big Island, Hawaii, volume 5, January 2003. [74] K. Kar and S. Banerjee. Node placement for connected coverage in sensor networks. In Proceedings of the Workshop on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt’03), Sophia Antipolis, France, 2003. [75] J. Bredin, E. Demaine, M. Taghi Hajiaghayi, and D. Rus. Deploying sensor networks with guaranteed capacity and fault tolerance. In Proceedings of the 6th ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHOC’05), Urbana-Champaign, Illinois, May 2005. [76] G.A. Gupta S. Toumpis. Optimal placement of nodes in large sensor networks under a general physical layer model. In Proceedings of 2nd IEEE Conference on Sensor and AdHoc Communications and Networks (SECON’05), Santa Clara, CA, September 2005. [77] T. Clouqueur, V. Phipatanasuphorn, P. Ramanathan, and K.K. Saluja. Sensor deployment strategy for target detection. In Proceedings of the 1st ACM international Workshop on Wireless Sensor Networks and Applications (WSNA’02), Atlanta, Georgia, September 2002. [78] M. Ishizuka and M. Aida. Performance study of node placement in sensor networks. In Proceedings of the 24th International Conference on Distributed Computing Systems Workshops (Icdcsw’04),Hachioji, Tokyo, Japan, volume 7, March 2004. [79] X. Zhang and S.B. Wicker. How to distribute sensors in a random field? In Proceedings of the 3rd International Symposium on Information Processing in Sensor Networks (IPSN’04), Berkeley, CA, April 2004. [80] M. Younis K. Akkaya. Coverage and latency aware actor placement for wireless sensor and actor networks. International Journal of Sensor Networks, 3:152–164, May 2008. [81] A. Kansal and al. Controlled mobility for sustainable wireless sensor networks. In Proceedings of IEEE Sensor and AdHoc Communications and Networks (SECON’04), Santa Clara, CA, October 2004.

Bibliography

Bibliography

[82] G. Wang, G. Cao, and T. La Porta. Movement-assisted sensor deployment. In Proceedings of the 23rd International Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM’04), Hong Kong, March 2004. [83] J. Wu and S. Yang. Smart: A scan-based movement assisted sensor deployment method in wireless sensor networks. In Proceedings of the 24th International Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM’05), Miami, FL, March 2005. [84] D. Lavanya and S.K. Udgata. Swarm intelligence based localization in wireless sensor networks. Lecture Notes Computer Sciences, pages 317–328, 2011. [85] C. Ozturk, D. Karaboga, and B. Gorkemli. Probabilistic dynamic deployment of wireless sensor networks by artificial bee colony algorithm. Sensors, 11(6):6056–6065, 2011. [86] N. Ahmed, S.S. Kanhere, and S. Jha. A pragmatic approach to area coverage in hybrid wireless sensor networks. Wireless Communications and Mobile Computing, 11:23–45, 2011. [87] G. Wang, L. Guo, H. Duan, L. Liu, and H. Wang. Dynamic deployment of wireless sensor networks by biogeography based optimization algorithm. Journal of Sensor and Actuator Networks, 1:86–96, 2012. [88] R. Bellazreg, N. Boudriga, and M. Hamdi. A dynamic distributed key tunneling protocol for heterogeneous wireless sensor networks. In IEEE International Conference on Trust Security and Privacy in Computing and Communications TrustCom, 2012. [89] C. Karlof, N. Sastry, and D. Wagner. Tinysec: A link-layer security architecture for wireless sensor networks. In SenSys ’04: Proc. 2nd International Conference on Embedded Networked Sensor Systems, New York: ACM Press, pages 162–75, 2004. [90] A. Perrig, R. Szewczyk, V. Wen, D. Culler, and J. D. Tygar. Spins: Security protocols for sensor networks. In The Seventh Annual International Conference on Mobile Computing and Networking (MobiCom 2001), 2001. [91] Mayank Saraogi. Security in wireless sensor networks. In ACM SenSys, 2004.

Computation of the measure of a sub ellipse

A

In this appendix, we will present the demonstration of the measure of the surface of a sub ellipse.

Figure A.1.: The surface delimited by f(x) The surface S delimited by f(x) and depicted by Figure A.2 is given by the following formula S=

Zy1

f (y)dy.

y0

To measure the area of the sub ellipse, we consider the following figure. We present in the following the needed calculations and the final formula of the subellipse area SAA”A′ . The equation of an ellipse is given by the following relation x2 y 2 + 2 = 1. a2 b  then we have x2 = a2 1 − q 2 hence x = ±a 1 − yb2

y2 b2



148

APPENDIX A. COMPUTATION OF THE MEASURE OF A SUB ELLIPSE

Figure A.2.: The measure of a sub ellipse ⌢ We consider the surface delimited by the arc AB and the segments AB and A’B’. In that case q

f (y) = a

1−

y2 . b2

We have : y0 = b

q

1−

x20 a2

p = ab a2 − x20

having: 0 < y0 ≤ b then 0 < and SABB ′ A′ = a

Ry0 q

1−

y2 b2

Ry0 q

1−

0

and SABB ′′ A′′ A′ = 2a

y0 b

0

≤1

dy y2 b2

dy

and SAA0 A′ = SABB ′ A′ − SABB ′ A0 or SABB ′ A0 = y0 xx0 then we have : SAA′′ A′ = 2 SAA0 A′ Lets calculate the measure of

ABB ′ A′ .

We have

SABB ′ A′

Zy0 r y2 =a 1 − 2 dy. 0

We denote by

y b

= sin(t) ⇒ dy = bcos(t)dt arcsin(

R

then SABB ′ A′ = ab

0

knowing that 0 < we deduce that

y0 ) b

p

1 − sin2 (t)cos(t)dt

≤ 1, then 0 < arcsin( yb0 ) ≤

y0 b

y

y

arcsin( b 0 )

SABB ′ A′ = ab

R 0

0

π 2

arcsin( b 0)

cos2 tdt = ab

R 0

0

1+cos(2t) dt 2

APPENDIX A. COMPUTATION OF THE MEASURE OF A SUB ELLIPSE

= = or

ab 2 ab 2

y

iarcsin( 0 b) sin(2t) 2 0   arcsin( yb0 ) + 21 sin(2arcsin( yb0 )) h

t+

sin(2arcsin( yb0 )) = 2sin(arcsin( yb0 )cos(arcsin(y0 b)) = 2 yb0 cos(arcsin( yb0 )) We also have, cos2 (arcsin( yb0 )) = 1 − sin2 (arcsin( yb0 )) = 1 −

y02 b2

then, cos(arcsin( yb0 )) =

q

1−

y02 b2

and then, sin(2arcsin( yb0 ))

=

2 yb0

q

1−

we can deduce that  ab SABB ′ A′ = 2 arcsin( yb0 ) +

y02 b2

y0 b

q

1−

y02 b2



and SAA0 A′ =

ab 2



arcsin( yb0 )

+

then we deduce that  ′ SAA”A = ab arcsin( yb0 ) +

y0 b

q

y0 b

q

1−

1−

Replacing x0 with its value x0 = a following formula. SAA”A′

y02 b2



− y0 xx0

y02 b2



− 2y0 xx0

q

1− "

y02 , b2

the area of the sup ellipse will be given by the

y0 y0 = ab arcsin( ) − b b

r

y2 1 − 20 b

#