Mobile Wireless Networks
We propose in this section to study various mobile ad-hoc networking architectures. There are three main categories and each category is dened by the mobility of the nodes and by the network dynamicity. The three categories are:
• Mobile ad-hoc networks (MANETs).
• Delay or disruption tolerant networks (DTNs).
• Opportunistic networks.
Mobile Ad-Hoc Networks
Wireless networks can be classied into two categories: infrastructure-based net- works and ad-hoc networks (Sensor networks, MANETs, DTNs). When the network is deployed with an infrastructure, the nodes communicate with one or many base stations. The set of all base stations (such as access points) are connected to a backbone. On the other hand, the ad-hoc networks have no infrastructure and are totally decentralized. Indeed, the networks consists of autonomous mobile nodes that are connected through wireless links and which may play the role of routers to ensure communication between any pair of source-destination .
In ad-hoc networks and especially in MANETs, each pair of nodes can be con- nected by a wireless link or by multi-hop paths. Then, source and destination nodes can be out of range of each other and even though communicate. The network topol- ogy can evolve over time as nodes are expected to move. Nonetheless, the MANET routing protocols tend to maintain end-to-end paths between all the nodes. This property highlights the dierence with opportunistic networks where it is assumed that no end-to-end paths exist and forwarding a packet from a source to a desti- nation relies only on the mobility of the nodes and the contacts occurred between them.
Delay Tolerant Networks
The Delay or Disruption Tolerant Network (DTN) architecture is a paradigm that has been advanced in order to provide services for challenging networks  (ensuring connectivity in rural areas [37, 96], military networks , underwater networks , . . . ). The DTNs are deployed in environments where the connectivity is intermittent or scheduled and may undergo high error rates, long delays and low data rates. These networks cannot maintain end-to-end paths between nodes due to the mentioned constraints. Nevertheless, forwarding a packet from a source to a destination in these networks relies on the store-and-forward multi-hop routing based on the occurrence of contacts between intermediate nodes.
A Key Parameter for Mobile Wireless Networks: Human Mobility
With the growth of these mobile networks and the sophistication of portable com- putation and communication devices, the human mobility becomes a paramount phenomenon that impacts the structure of the network as shown in [29, 59]. Hence, it is important to understand the human mobility to ensure accurate evaluations and analyses of the performance of communication protocols and applications. In the following section, we give a more deep information about human mobility and list its major properties detailed in the literature.
Towards Understanding Human Mobility
To reproduce human mobility, two rst individual models have been proposed: the Random Walk or Brownian motion  and the Random Waypoint model .
Nodes that use Random Walk move, at each step, with a speed chosen from a distribution towards a random position on a moving area. When a node reaches its destination, it xes a new speed and picks new random position to which it have to move and the process is repeated. When the Random Waypoint model is considered, the same process as Random Walk is reproduced with adding pause times between each step. Many other mobility models that are similar to two cited approaches are detailed in [26, 79]. There are also group mobility models proposed in the literature, such as , where the movement of nodes are correlated to the one of the clusterhead.
Such models have been widely used in simulations to attest the eciency of communication protocols in the context of intermittent connections. Despite the popularity of such synthetic models, it has been shown that they are unable to really reproduce the complexity of human behavior [29, 79]. From these perspective, many researchers have focused on collecting data and especially encounters between nodes from real life scenarios (conferences, student campuses, meetings,. . . ). Relaying on real traces, it is possible to identify salient properties of human mobility and perform
more accurate simulations and more reliable evaluations.
Social Network Analysis and Link Prediction
with other behaviors. In other words, using such a model generate users that have not preferred contacts (strong friends) and consequently no consistent social inter- actions. To emphasize these ndings, Thakur et al. have proposed to compute a similarity metric. It measures the degree of similarity of the behaviors of two mobile nodes and the behavior of each node is expressed by an association matrix. The columns of the matrix represent the possible locations that a node can visit and the rows express time granularity (hours, days, weeks, etc.). The dominant behavioral patterns are tracked using the Singular Value Decomposition (SVD) . For more details about the similarity metric computation, we refer the reader to .
From this perspective, the nodes in such networks are strongly interdependent and the interactions between them govern the structure of the network. This fact has motivated researchers to apply the Social Network Analysis (SNA)  to extract intrinsic properties of the network and to exploit them to design more ecient communication protocols.
Table of contents :
1.1 Constraints and Challenges in Mobile Multi-Hop Networks
1.1.1 Major Problems Encountered
1.1.2 The Human Mobility: Understanding it to Harness its Prop- erties
1.2 Motivations and Contributions
1.3 Dissertation Organization I Proposition of a Link Prediction Framework and Metrics for Human-Centered Mobile Wireless Networks
2 Human-Centered Mobile Wireless Networks: Human Mobility in the Service of Link Prediction
2.2 Mobile Wireless Networks
2.2.1 Mobile Ad-Hoc Networks
2.2.2 Delay Tolerant Networks
2.2.3 Opportunistic Networks
2.2.4 A Key Parameter for Mobile Wireless Networks: Human Mo- bility
2.3 Human Mobility
2.3.1 Towards Understanding Human Mobility
2.3.2 Encounter Traces
2.3.3 Human Mobility Properties
2.4 Social Network Analysis and Link Prediction
2.4.1 Overview on the Social Network Analysis
2.4.2 Link Prediction in Social Networks
3 Tensor-Based Link Prediction Framework for Mobile Wireless Net-works
3.2 Related Work
3.3 Description of the Tensor-Based Link Prediction Method
3.3.2 Matrix of Scores Computation
3.3.3 Matrix of Scores Interpretation
3.4 Performance Evaluation and Simulation Results
3.4.1 Simulation Traces
3.4.2 Performance Analysis
4 Improving Link Prediction in Mobile Wireless Networks by Con- sidering Link and Proximity Stabilities
4.2 Related Work
4.3 Tensor-Based Link Prediction Framework: A Reminder
4.3.2 Overview on Tensor-Based Link Prediction Technique
4.3.3 Matrix of Scores Computation
4.3.4 Matrix of Scores Interpretation
4.4 How To Quantify Link and Proximity Stabilities and How To Use Them?
4.4.1 How Can Link and Proximity Stabilities Improve Link Pre- diction?
4.4.2 The Entropy Measure
4.4.3 Quantifying Link and Proximity Stabilities by Means of Time Series Entropy Estimation
4.4.4 Joining the Entropy Estimations to the Tensor-Based Link Prediction Framework
4.5 Simulations Scenarios and Performance Evaluation
4.5.1 Simulation Traces
4.5.2 Simulation Results and Performance Analysis
5 A Joint Model for IEEE 802.15.4 Physical and Medium Access Control Layers
5.2 Related Work
5.3 Developed IEEE 802.15.4 Model for Smart Grid
5.3.1 IEEE 802.15.4 PHY Model Description
5.3.2 Operation Details for the IEEE 802.15.4 MAC Model and the Interactions with the PHY Model
5.4 Simulation and Results
5.4.1 Cheking the Validity of the Joint Model
5.4.2 Comparison between Combined PHY and MAC Layers and Simple MAC Layer Models
5.4.3 Evolution of Node Performance with Growing Densities
6 Cooperation Enforcement for Packet Forwarding Optimization in Multi-hop Mobile Ad-hoc Networks
6.2 Related Work
6.3 System Model And Problem Formulation
6.3.1 « The Weakest Link » TV Game Principle
6.3.2 The Proposed Model: Analogy with « The Weakest Link » TV Game
6.4 Self-Learning Repeated Game Framework and Punishment Mechanism113
6.4.1 The Punishment Mechanism
6.4.2 The Self-Learning Repeated Game Framework Description
6.5 Performance Evaluation and Simulations Results
7.1 Contributions of the Thesis
7.1.1 Proposing a Tensor-Based Link Prediction Framework
7.1.2 Improving Link Prediction Eciency by Considering Link and Proximity Stabilities
7.1.3 Other Works Towards the Improvement of the Evaluation and the Design of Communication Protocols in Human-Centered Wireless Networks
7.2 Future Research Directions
7.2.1 Implementing the Tensor-based Link Prediction Framework on Real Testbeds
7.2.2 Investigating the Contributions of Other Expressions for the Stability Measure
7.2.3 Exploiting the Joint Model for IEEE 802.15.4 PHY and MAC Layers for Proposing New Performance Metrics
7.2.4 Enhancing the Self-Learning Repeated Game
B Résumé en Français
B.2 Proposition de Méthode et de Métriques pour la Prédiction des liens dans les Réseaux Mobiles Sans-Fil Centrés sur l’Aspect Humain
B.2.1 Mobilité Humaine, Analyse des Réseaux Sociaux et Prédiction des Liens
B.2.2 Méthode de Prédiction des Liens Basée sur les Tenseurs pour les Réseaux Sans-Fil Mobiles
B.2.3 Amélioration de la Méthode de Prédiction des Liens en Con- sidérant les Stabilités de Lien et de Proximité
B.3 Autres Contributions Avancées dans la Thèse