Graduate Programs | 25
First
Last
Date
Advisor
Title
Reza
Moazzez Estanjini
5/22/11
Paschalidis Vehicle Scheduling and Routing for Data Transport in Wireless Sensor Networks
Abstract: The traditional approach for data relaying in Wireless Sensor NETworks (WSNETs) involves multi-hop communications from data sources to destinations. In addition to problems that multi-hop routing may cause when networks become partitioned, relaying data over a large number of hops reduces the lifetime of sensor nodes. It is known that the use of some mobile elements, which in this dissertation are referred to as Message Ferries (MFs), to transport data from node to node can overcome the mentioned difficulties. In this dissertation, a WSNET in which MFs are used is termed Mobile Sensor NETworks (MSNETs). In MSNETs, the data transfer efficiency depends heavily on the path followed by the MFs, hence the main challenge is the scheduling of the MFs. This dissertation addresses this issue within two settings: a static perspective and a dynamic one. The static setting is suitable for MSNETs whose underlying dynamics are relatively slow. For such scenarios, the proposed solutions in this dissertation have performance guarantees, some present novel results on scalability of MSNETs, and some even improve the best-known solutions to some classical problems in combinatorial optimization. On the other hand, for MSNETs whose underlying dynamics are faster and less predictable, a dynamic perspective is more suitable. For such scenarios, this dissertation addresses the problems in a Markov Decision Process (MDP) framework. A powerful novel approximate Dynamic Programming (DP) algorithm for MDP problems is presented. The algorithm is of the Actor-Critic (AC) type and uses a Least Squares Temporal Difference (LSTD) learning method. The use of the algorithm is not restricted to MSNET problems; it can also be used for many MDP problems as well as Partially Observable Markov Decision Process (POMDP) problems arising in other application areas. The forklift dispatching problem illustrated in this dissertation, which can be seen as an instance of the MF scheduling problems in MSNETs, is in fact an example of a real world problem in warehouse management where the use of the proposed algorithm is shown to be effective. Thomas
Vitolo
1/25/11
Casta単on
Practical Algorithms to Discover Degree Constrained Spanning Trees in Sparsley Connected Drafts
Abstract: Choosing a network topology to connect communication nodes subject to specific objectives, constraints, and properties is a broadly studied problem, with approaches varying widely depending on the hardware capabilities and limitations, the required performance criteria, and the available budget. This thesis is motivated by a topology problem in which naval, ground, air, and space vehicles require secure, high bandwidth data communications across long distances in an unstable environment. Each wireless connection in the network requires a pair of directional antennae; the connection is point-to-point line-of-sight, not broadcast over a wider region. While some nodes may be stationary, stable and secure, the majority of the hundreds of nodes in the network are moving. As a result, a line-of-sight connection between a pair of nodes is subject to predictable and unpredictable interruption. Because the network is constantly evolving, new topologies must be generated very quickly so that network connectivity is maintained. This thesis defines the connectivity problem as finding a modified degree constrained spanning tree. It develops optimal algorithms which can generate the topology backbone for large point-to-point wireless networks quickly, even in networks with hundreds of nodes. The algorithms presented are compared to other network generating algorithms. This thesis extends the algorithms to new variations of this topology problem including scenarios where the antennae in the network consist of multiple incompatible technologies, and uses discrete time stages to maximize full connectivity over a time horizon. The results of the thesis provide algorithms for the design of real-time topology maintenance algorithms, as well as bounds for comparison with the performance of faster heuristic approaches for topology maintenance.
Boston University College of Engineering | Division of Systems Engineering