Recent advances in wireless technology, Micro-Electro-Mechanical System (MEMS) technology and low power microprocessors have enabled a new generation of large-scale Wireless Sensor Networks (WSNs). WSNs promise advantage in wide range of applications such as planetary explosion, ocean monitoring, wildlife habitat monitoring, detecting forest fires and floods, patient monitoring and vehicular or pedestrian traffic monitoring. In WSNs, generally, a large number of sensor nodes are deployed in the remote terrain. These nodes coordinate to establish a communication network, monitor specified tasks, and report sensed data periodically or spontaneously to the sink node. In massive-scale WSNs, a wireless communication range of the sensor node is not long enough to transfer the sensor data directly to the sink node, hence traditional approach for data collection in WSNs involves multi-hop wireless communication among the sensor nodes. However the multi-hop wireless communication may not always be possible due to network partitioning. The network partitioning may be easily occurred when the sensor nodes are deployed sparsely in the target area or when the existing sensor nodes are out of order by running out their butteries or some other software/hardware failures. One solution to tackle this issue is a use of redundant sensor nodes. However it is required massive number of sensor nodes for monitoring a large area such as planet, forest and ocean. Another solution is a use of mobility. If the sensor nodes have availability of movement, they can carry the sensor data to the sink node as well as construct a wireless communication path. We call such the sensor network, which is designed as disruption tolerant, Disruption Tolerant Sensor Networks (DTSNs).

Traditional works of utilizing mobility in DTSNs is mainly focused on mobility of sink node, where the sink node moves predetermined route to collect the sensor data from the static or mobile sensor node. In such the works, the network performance heavily depends on the behavior of mobile sink. For example, if the mobile sink is out of order, all the sensor data whose destination is this mobile sink are dropped. On the other hand, the conventional works focusing on mobility of sensor node is mainly exploiting there random or predictable mobility, i.e., sensors attached to animals and humans. Utilizing random mobility and predictable mobility for sensor nodes is able to make the network scalable compared to utilizing a controlled mobility for sink node. However, on the other hand, random and predictable mobility sometimes generate huge delay, which causes data loss due to buffer overflow. Therefore, in this dissertation, we study and exploit the controlled mobility for sensor nodes to improve network performance in terms of data reachability, scalability and energy efficiency. Availability of controlled mobility for sensor node is currently growing; especially the emerging technology of multiple cooperated robots and embedded sensing devices brings us the use of mobile sensor robot for exploration in the large area such as the Moon, Mars and oceans. Hence the basic study of exploiting controlled mobility for sensor node is feasible to bring an emerging paradigm to the sensor network in the not-too-distant future.

Utilizing the controlled mobility, mobile node can move toward a wireless communication area that is covered by the destination node of data, instruct nearby node to move to a specific location or collaborate with other nodes to transfer data. In DTSNs, how to move and collaborate to minimize energy consumption for physical movement is a key problem, since the mobile node generally consumes much more energy than wireless communications in a unit time, and in sparse networks, mobile node moves for a long duration to carry the sensor data compared to duration of using wireless communication. To reduce the moving distance of mobile node, we can utilize store-carry-and-forward model, where the existing nodes relay the data from source to sink in one or more hops. Utilizing store-carry-and-forward model, the mobile node reduces the moving distance for wireless communication distance whose maximum value is the wireless communication range of wireless interface. This model is also beneficial in perspective of scalability because decisions of relaying can be made locally. However with decreasing the number of mobile node, the number of relaying from source node to sink per one bundle of sensor data is also decreasing. To overcome the issue, in this dissertation, we introduce an idea of concentrating the mobile nodes on a pre-determined path, in which sets of mobile nodes, called relay nodes, facilitate the connectivity to the sink node. Another crucial and challenging problem in DTSNs is the requirements of long-term communication, such as transmissions of continuous video from a specific hot spot. This is because that the mobile node has the difficulties on both storage and energy constrains. To overcome the above issues, in this dissertation, we introduce two significant approaches: sink-side approach and peer-side approach.

In the sink-side approach, we construct a single or multiple path(s) for relay nodes from the sink node, where the data is relayed hop by hop to the sink, so that the mobile node which is other than the relay node can reduce the distance of movement by forwarding the sensor data to the relay node which is on the constructed path. In the sink-side approach, we study the impact of single and multiple paths for the relay nodes. In our study, we observed that increasing the number of relaying nodes enlarges the affected area of reducing the moving distance for the mobile nodes which are other than relay node. However, if there is a constraint on the available number of relay nodes, increasing the number of relay nodes per path decreases the total number of path, and hence, the affected area of reducing the moving distance declines by a certain degree. Therefore, in this dissertation, we propose an algorithm to determine the optimal length and the number of paths to minimize the average moving distance of all the mobile nodes in the network to reduce the energy consumption. We also propose an algorithm to determine an optimal route for mobile node to the nearby constructed path. Our simulation result shows that optimizing the total number of paths reduces significant amount of energy for movement, when the sensor data is small size compared to the storage size of mobile node and generated intermittently.

For the continuous and massive-sized data from the hot spot, we need to take care about the storage overflow at the source node as well as the energy efficiency. Therefore, in the peer-side approach, we construct a direct path from the mobile node which is a source of the continuous and massive-sized sensor data to the sink node, on which the relay nodes may do round trip to transfer the sensor data. We also propose the algorithm for determining an optimal scheduling of relay nodes to meet the deadline of storage overflow at the source node. In the peer-side approach, we also propose an algorithm of local and temporal collaboration between nearby paths for relay nodes, where we partially combine and share the paths, so that the average moving distance per bundle can be reduced. In the algorithm, we construct an optimal tree within the nearby paths, which is base on the known algorithm for constructing the Euclidean Steiner Minimum Tree. Since the each path is temporal and has own duration, the tree-shaped combined path should be tolerant to dynamic change of its topology. Hence, we also propose the algorithm to handle the dynamic removal/addition of path(s) from/to the combined tree. The algorithm does not need to re-construct a whole tree configure the small part of the tree for the remaining/new path(s) so that we can reduce the overhead of re-constructing the whole tree. The simulation result shows that using tree-shaped combined paths we can reduce the energy consumption for movement compared to the single path based approach.

In this dissertation, to study the impact and evaluate our two approaches; sink-side approach and peer-side approach, we conduct the simulations and discuss the result. From the result, we can observe that, for non-continuous sensor data, the sink-side approach is better than the peer-side approach, and for continuous sensor data, the peer side approach is better than the sink-side approach in terms of energy efficiency and reachability of sensor data.

本論文は「A Study on Controlled Mobility in Disruption Tolerant Sensor Networks(分断耐性センサネットワークにおけるモビリティ制御に関する研究)」と題し、環境情報の取得・収集を目的としたセンサノード群による無線ネットワークの構築が困難な場合に、センサノードを自律的に移動可能とすることで、センサノードの移動そのものをデータ配送手段として捉えてネットワークを構築する際のモビリティ制御手法について検討を行ったものであり、全七章から構成されている。

第一章は「Introduction(序論)」であり、従来の環境固定型無線センサノード群による無線センサネットワークにおける課題を整理し、本論文において主題となる分断耐性センサネットワークの必要性を明らかにしている。また、分断耐性センサネットワークにおける、モビリティ制御の必要性と、センサノードの移動そのものをデータ配送として利用する場合の要求条件を整理し、その要求条件を満たすデータ収集手法であるBS-Pipe (BS-Pipe: Path Selection from a Base Station)及び PN-Pipe (PN-Pipe: Path Selection from Peer Nodes)の研究の経緯について述べている。

第二章は「Disruption Tolerant Sensor Networks(分断耐性センサネットワーク)」と題し、分断耐性センサネットワークの概要と既存のセンサネットワーク、モバイルアドホックネットワークとの違いを、システムモデル、アプリケーションモデル、ネットワークモデル、モビリティモデルに関して既存の手法と対比させながら整理し、データ配送のためのセンサノードのモビリティ制御の必要性とその特徴について述べている。

第三章は「Basic Concept and Problem Framework(基本概念と問題の枠組み)」と題し、分断耐性センサネットワークにおける課題と、トラフィックモデルやセンサノードの種類、またアプリケーションモデル、ネットワークモデルを含む問題設定を行い、解決手法であるBS-PipeとPN-Pipeについての概要と、その基本概念である "pipeline" のモデル化について述べている。 "pipeline" では移動によるデータ配送を行うセンサノードをいくつかの経路に集中させ、経路内でのそれらセンサノード間の協調によって、移動距離の短縮を図っている。

第四章は「BS-Pipe: Path Selection from a Base Station(BS-Pipe: ベースステーションからの経路選択手法)」と題し、センサデータの配送先であるベースステーションから、センシング領域への "pipeline" の構築を行うための経路選択手法(BS-Pipe)について述べている。BS-Pipeでは、ベースステーションが構築した "pipeline" へ、周囲に存在するセンシングの役割を担っているセンサノードからのデータを転送することで、センサデータの配送に要する送信元から宛先までの移動距離の短縮を図っている。そのため、BS-Pipeでは、 "pipeline"内のセンサノードの移動制御手法だけでなく、周囲に存在するセンサノードの "pipeline" への移動経路についての最適化も行っている。

第五章は「PN-Pipe: Path Selection from Peer Nodes(PN-Pipe: ピアノードからの経路選択手法)」と題し、各センサノード(ピアノード)からの経路構築要求に対する、センサノードとベースステーション間をピア・トゥ・ピアで結ぶ "pipeline" の構築手法と、各センサノードの要求条件を満たすための "pipeline"の最適化を行っている(S-PN-Pipe)。さらに、第五章ではセンサノード・ベースステーション間で構築している "pipeline" を近隣の "pipeline" と統合させることで、ピア・トゥ・ピアに "pipeline" を構築する場合に比べた移動距離の短縮を目的とした、M-PN-Pipeについて述べている。M-PN-Pipeでは、 複数のセンサノード(経路構築要求元)・ベースステーション間において、ベースステーションを根とした最適な木構造の経路を決定し、また、構築した木(経路)の各枝に対し、最適なノード数を配置することにより、データ配送による移動距離の短縮し、エネルギー消費量の削減を図っている。

第六章は「Analysis and Evaluation (解析とシミュレーション)」と題し、BS-Pipe、及びPN-Pipeのそれぞれの性能評価をシミュレーションにより解明している。さらに、シミュレーションにより、BS-PipeとPN-Pipeについての性能比較及び議論を行い、システム要求であるセンシングを行うセンサノードの数に対するデータの配送を行っているセンサノードの数の割合が高い場合は、PN-Pipeの方が最適であり、逆に、その割合が低い場合は、BS-Pipeの方がより多くのエネルギー消費量を削減できることを解明している。

第七章は「Conclusion(結論)」であり、論文の成果と今後の展開をまとめている。

以上これを要するに、本論文は、センサノード群を自律的に移動可能とし、無線で接続されることによって環境情報を取得するための無線センサネットワークではデータ収集が困難な分断化された状況においてもデータ収集を可能とする、センサノードの効率的なモビリティ制御手法の提案と、その性能解析を行ったものであって、電子情報学に貢献するところが少なくない。 よって本論文は博士(情報理工学)の学位論文として合格と認められる。