Node activity scheduling and routing in Wireless Sensor Networks

Participants : Cédric Adjih, Ichrak Amdouni, Pascale Minet.

The need to maximize network lifetime in wireless ad hoc networks and especially in wireless sensor networks requires the use of energy efficient algorithms and protocols. Motivated by the fact that a node consumes the least energy when its radio is in sleep state, we achieve energy efficiency by scheduling nodes activity. Nodes are assigned time slots during which they can transmit and they can turn off their radio when they are neither transmitting nor receiving. Compared to classical TDMA-based medium access scheme, spatial bandwidth use is optimized: non interfering nodes are able to share the same time slots, collisions are avoided and overhearing and interferences are reduced. In our work about time slots assignment, two cases are studied. First, when nodes require equal channel access, we use node coloring. Second, when nodes have heterogeneous traffic demands, we designed the traffic aware time slot assignment algorithm TRASA. Unlike the majority of previous works, we generalize the definition of node coloring and slot allocation problems. Indeed, we set the maximum distance between two interfering nodes as a parameter of these problems. We prove that they are NP-complete, making heuristic approaches inevitable in practice. A central directive of this thesis is to design self-adaptive solutions. This adaptivity concerns many aspects such as the mission given by the application, the heterogeneity of node traffic demands, the network density, the regularity of network topology, and the failure of wireless links.

In the GETRF project, we target the energy efficiency in wireless sensor networks. We proposed node activity scheduling approaches that determine active and inactive slots for sensor nodes as to enable them to turn off their radio and save energy in the inactive slots.

1. First, we proposed a scheduling algorithm based on node coloring of grid sensor networks called VCM. This proposal was strengthen with mathematical analysis of the optimal number of colors needed to color an infinite grid. VCM produced an optimal number of colors when the transmission range tends to infinity. Also, this algorithm does not require message exchange between sensors to determine colors.

2. Second, this work was extended to adapt it to general graphs: the graph is divided into cells and the color of the cell is the color of the node on the left bottom of the cell. Nodes inside the cell are scheduled successively.

In addition to the energy efficiency, we targeted the delay optimization for data collection applications in grid wireless sensor networks. We profit from the previous work VCM and integrate it with a new hierarchical routing method to minimize data collection delays.

Time slot and channel assignment in multichannel Wireless Sensor Networks

Participants : Pascale Minet, Ridha Soua, Erwan Livolant.

Applying WSNs in industrial environment requires fast and reliable data gathering (or data convergecast). If packets are forwarded individually to the sink, it is called raw data convergecast. We resort to the multichannel paradigm to enhance the data gathering delay, the robustness against interferences and the throughput. Since some appli- cations require deterministic and bounded convergecast delays, we target conflict free joint time slot and channel assignment solutions that minimize the schedule length. Such solutions allow nodes to save energy by sleeping in any slot where they are not involved in transmissions. We extend existing multichannel results to take into account a sink equipped with multiple radio interfaces and heterogeneous traffic demands. Indeed, we compute the theoretical bounds, that is the minimum number of time slots needed to complete convergecast, in various topologies with different traffic demands. These bounds are provided for different acknowledgment policies. For each of them, we provide a graph-based interference model. We also give optimal schedules that achieve these optimal bounds. We formalize the problem of multichannel slot assignment using integer linear programming and solve with GLPK tool for small configurations.

We propose MODESA, a centralized joint time slot and channel assignment algorithm. We prove the optimality of MODESA in lines, multilines and balanced trees topologies. By simulations, we show that MODESA outperforms TMCP, a well known subtree-based scheduling. We improve MODESA with different channel allocation strategies depending on the channel selection criteria (channels load balancing or preference of channels with the best qualities). Moreover, we show that resorting to multipath routing minimizes the convergecast delay. This work is extended in MUSIKA to take into account multi-sinks WSNs and traffic differentiation: the problem is formalized using integer linear programming and solved with GLPK. Simulations results show that the schedule length is minimized and the buffer size is reduced. We then address the adaptivity challenge. The slot assignment should be more flexible and able to adapt to application and environment variability (e.g., alarms, temporary additional demands). Theoretical bounds on the number of additional slots intro- duced to cope with traffic changes, are given. AMSA, an incremental solution, is proposed. Its performances are evaluated in two cases: retransmissions or temporary changes in appli- cation needs.

WSN Redeployment

Participants : Pascale Minet, Saoucene Ridene, Ines Khoufi.

This is a joint work with Telecom SudParis: Anis Laouiti.

In many applications (e.g military, environment monitoring), wireless sensors are randomly deployed in a given area. Unfortunately, this deployment is not efficient enough to ensure full area coverage and total network connectivity. Hence, all the considered area must be covered by sensors ensuring that any event is detected in the sensing range of at least one sensor. In addition, the sensor network must be connected in terms of radio communication in order to forward the detected event to the sink(s). Thus, a redeployment algorithm has to be applied in order to achieve these two goals.

In this context, we have proposed redeployment algorithms based on virtual forces. DVFA, is our Distributed Virtual Forces Algorithm. Each node in the network executes DVFA and computes its new position based on information collected from its neighbors. Performance evaluation shows that DVFA gives very good coverage rate (between 98% and 100%) and ensures the connectivity between sensors.

Moreover, in a real environment, obstacles such as trees, walls and buildings may exist and they may impact the deployment of wireless sensors. Obstacles can prohibit the network connectivity between nodes and create some uncovered holes or some accumulation of sensors in the same region. Consequently, an efficient wireless sensors deployment algorithm is required to ensure both coverage and network connectivity in the presence of obstacles. We have focused on this problem and enhanced our Distributed Virtual Force Algorithm (DVFA) to cope with obstacles. Simulation results show that DVFA gives very good performances even in the presence of obstacles.

Opportunistic routing cross-layer schemes for low duty-cycle wireless sensor networks

Participants : Mohamed Zayani, Paul Muhlethaler.

This is a joint work with Nadjib Aitsaadi from University of Paris 12.

The opportunistic aspect of routing is suitable with such networks where the topology is dynamic and protocols based on topological information become inefficient. Previous work initiated by Paul Muhlethaler and Nadjib Aitsaadi consisted in a geographical receiver-oriented scheme based on RI-MAC protocol (Receiver-Initiated MAC). This scheme is revised and a new contribution proposes to address the same problem with a sender-oriented approach. After scrutinising different protocols belonging to this classification, the B-MAC protocol is chosen to build a new opportunistic cross-layer scheme. Our choice is motivated by the ability of this protocol to provide to a sender the closest neighbor to the destination (typically a sink). In other words, such a scheme enables us to obtain shorter paths in terms of hops which would increase the efficiency of information delivery. In counterparts, as it relies on long preambles (property of B-MAC) to solicit all the neighborhood, it needs larger delays and energy consumption (1% of active time). Nevertheless, this proposal remains interesting as the studied networks are dedicated to infrequent event detection and are not real time-oriented.

Starting from a simulator coded by Nadjib Aitsaadi for the receiver-oriented scheme, the new scheme has been coded under many variants. On top of ideal techniques, a realistic variant has been considered and modelled. Its particularity can be summarized in the election process of the next hop. Indeed, it is based on sending bursts by the potential candidates to receive a packet from a sender. These bursts express the closeness of each candidate to the destination and correspond to the binary complement of the distance to this destination.

The opportunistic cross-layer scheme, when designed with RI-MAC, has shown solid performances in carrying the information about a rare event detection to a sink. This is verified for an event detected by several nodes. Nevertheless, the efficiency of such a design becomes less obvious when the detection is performed by a very small number of nodes. the opportunistic routing using RI-MAC relies on a minor set of potential candidates to forward a packet. In other words, a sender can only select an awake neighbor (typically closer to the sink) as the next hop. To overcome this limitation, we initially proposed to limit the number of hops to reach the sink.The principle of B-MAC perfectly matches with this idea. It is also important to highlight the ability of an opportunistic cross-layer built over B-MAC to avoid collisions. B-MAC- and RI-MAC-based proposals are suitable to convey emergency packets in dense and large WSNs when the event is reported by a significant set of nodes. When this set is limited, the sake of efficiency rather suggests a scheme based on B-MAC. It should be remembered that the proposed schemes extremely limit the energy consumption compared to classical networks.

Data dissemination in Urban Environment

Participants : Belhaoua Asma, Nadjib Achir, Paul Muhlethaler.

Over the last decade, wireless sensor networks have brought valid solutions to real-world monitoring problems. Sensors are now incorporated in all our modern life facilities, such as mobile phones, vehicles, buses, bus stations, bikes, etc. For example, mobile phones, with their increasing capabilities are used as voice communication device but also as a sensing device able to collect data such as image, audio, GPS position, speed, etc. All these sensors could play an important role in the provisioning of a multitude of dynamic information about their environmental trends. Considering that, WSN could be considered as a valid solution to urban monitoring problems by bringing new services for the city or for the citizens. According to the last requirement, the main question that we need to answer is how the data could be collected and/or transmitted? Several algorithms were developed recently for sensor data gathering in WSN. However, the majority of existing works on WSN has focused only on specific areas applications, such as environmental monitoring, military target tracking, weather forecast, home automation, intrusion detection, etc. In this training we studied the existing strategies of dissemination in Delay/ Disruption Tolerant Networks (DTN). The main objective is to identify those that can be applied to urban environments. We implemented and tested several strategies in the WSNet network simulator on a dense network.