Location-Aware Protocols for Energy-Efficient Information Processing in Wireless Sensor Networks

Transcription

1 Location-Aware Protocols for Energy-Efficient Information Processing in Wireless Sensor Networks by Harshavardhan Sabbineni Department of Electrical and Computer Engineering Duke University Date: Approved: Prof. Krishnendu Chakrabarty, Advisor Prof. John Board Prof. Loren Nolte Prof. Kishor Trivedi Dr. John Zachary Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Electrical and Computer Engineering in the Graduate School of Duke University 2009

2 Abstract (Electronics and Electrical Engineering) Location-Aware Protocols for Energy-Efficient Information Processing in Wireless Sensor Networks by Harshavardhan Sabbineni Department of Electrical and Computer Engineering Duke University Date: Approved: Prof. Krishnendu Chakrabarty, Advisor Prof. John Board Prof. Loren Nolte Prof. Kishor Trivedi Dr. John Zachary An abstract of a dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Electrical and Computer Engineering in the Graduate School of Duke University 2009

3 Copyright c 2009 by Harshavardhan Sabbineni All rights reserved.

4 Abstract Advances in the miniaturization of microelectromechanical components have led to battery powered and inexpensive sensor nodes, which can be networked in an ad hoc manner to perform distributed sensing and information processing. While sensor networks can be deployed in inhospitable terrain to provide continuous monitoring and processing capabilities for a wide range of applications, sensor nodes are severely resource-constrained; they typically run on batteries and have a small amount of memory. Therefore, energy-efficient and lightweight protocols are necessary for distributed information processing in these networks. The data provided by a sensor node is often useful only in the context of the location of the data source. Thus, sensor networks rely on localization schemes to provide location information to sensor nodes. The premise of this thesis is that location-aware protocols, which are based on the assumption that sensor nodes can estimate their location, improve the efficiency of data gathering and resource utilization of wireless sensor networks. Location-awareness improves the energy-efficiency of the protocols needed for routing, transport, data dissemination and self-organization of sensor networks. Existing sensor network protocols typically do not use location information effectively, hence they are not energy-efficient. In this thesis, we show how location information can be leveraged in novel ways in sensor network protocols to achieve energy efficiency. The contributions of this thesis are in four important areas related to network protocol design for wireless sensor networks: 1) self-organization; 2) data iv

5 dissemination or node reprogramming; 3) service differentiation; and 4) data collection. Work on self-organization (SCARE) and data dissemination (LAF) was carried out from 2002 to 2004 and the work on service differentiation (SensiQoS) and data collection (HTDC) was carried out from 2004 to This thesis first presents a new approach for self-configuration of ad hoc sensor networks. The self-configuration of a large number of sensor nodes requires a distributed solution. We propose a scalable self-configuration and adaptive reconfiguration (SCARE) algorithm that exploits the redundancy in sensor networks to extend the lifetime of the network. SCARE distributes the set of nodes in the sensor network into subsets of coordinator nodes and non-coordinator nodes. While coordinator nodes stay awake, provide coverage, and perform multi-hop routing in the network, non-coordinator nodes go to sleep. When nodes fail, SCARE adaptively re-configures the network by selecting appropriate non-coordinator nodes to become coordinators and take over the role of failed coordinators. This scheme only needs local topology information and uses simple data structures in its implementation. SCARE organizes nodes into coordinator and non-coordinator nodes. A recent work, termed Ripples [106] has improved upon the selforganization and reconfiguration mechanism proposed in SCARE. It uses a lightweight clustering algorithm to elect cluster heads instead of coordinator nodes based on location information as proposed by SCARE. Ripples selects fewer cluster-head nodes compared to the number of coordinator nodes elected by SCARE by varying the cluster radius and consequently realizes more energy savings while providing comparable sensing coverage. This thesis next presents an energy-efficient protocol for data dissemination in sensor networks. Sensor networks also enable distributed collection and processing of sensed data. These networks are usually connected to the outside world with base stations or access points through which a user can retrieve the sensed data for v

6 further inference and action. Dissemination of information is a challenging problem in sensor networks because of resource constraints. Conventional methods use classical flooding for disseminating data in a sensor network. However, classical flooding suffers from disadvantages such as the broadcast storm problem. We have proposed an energy-efficient scheme that uses the concept of virtual grids to partition (selfconfigure) the set of nodes into groups of gateway nodes and internal nodes. While gateway nodes forward the packets across virtual grids, internal nodes forward the packets within a virtual grid. The proposed location-aided flooding protocol (LAF) reduces the number of redundant transmissions and receptions by storing a small amount of state information in a packet and inferring the information about nodes that already have the packet from the modified packet header. More recent work [55] has extended the virtual grid concept proposed by LAF to non-uniform sensor network deployments. In [55], non-uniform virtual grids are used to improve upon the energy savings provided by LAF and achieve higher energy savings for non-uniform sensor network topologies. This thesis also addressees the challenging problem of timely data delivery in sensor networks. We propose SensiQos, which leverages the inherent properties of the data generated by events in a sensor network such as spatial and temporal correlation, and realizes energy savings through application-specific in-network aggregation of the data. This data delivery scheme is based on distributed packet scheduling, where nodes make localized decisions on when to schedule a packet for transmission to save energy and to which neighbor they should forward the packet to meet its end-to-end real-time deadline. Finally, this thesis presents an energy-efficient data collection protocol for sensor networks. It is based on a combination of geographic hash table and mobile sinks that leverage mobile sinks to achieve energy-efficiency in event-driven sensor networks. Next, an analysis of the energy savings realized by the proposed protocol is presented. vi

7 Simulation results demonstrate significant gains in energy savings for data collection with change in various parameter values. In summary, this thesis represents an important step towards the design of location-aware energy-efficient protocols for self-configuration, data dissemination, data delivery, and data collection in wireless sensor networks. It is expected to lead to even more efficient protocols for data dissemination, routing, and transport-layer protocols for energy-constrained and failure-prone sensor networks. vii

8 To my parents viii

9 Contents Abstract List of Tables List of Figures Acknowledgements iv xv xvi xx 1 Introduction Design Challenges in Wireless Sensor Networks Sensor Network Architectures Data-Dissemination Protocols Self-Configuration Protocols Data Delivery of Delay Sensitive Traffic Data Collection with Mobile Sinks Relevance of Thesis Research Thesis Outline Self-Configuration and Adaptive Reconfiguration Introduction Relevant Prior Work Outline of SCARE Basic Scheme Network Partitioning Problem ix

10 2.4 Details of SCARE Time Relationships Ensuring Network Connectivity Message Complexity Space Complexity Performance Evaluation Simulation Methodology Simulation Results Number of Coordinators Selected Control Message Overhead Mobility Network Lifetime Effect of Node Failures Effect of Location Estimation Error Summary Location-Aided Flooding for Data Dissemination Introduction Related Prior Work Location-Aided Flooding Modified Flooding Location Information Virtual Grids Packet Header Format LAF Node Types Information Dissemination Using LAF x

11 3.3.7 Resource Management in LAF Completeness of the Data Dissemination Procedure Analysis Space Complexity Time Complexity Errors in Location Estimates Performance Evaluation Simulation Model Data Acquired in the System with Time Energy Dissipated in the System with Time Impact of Number of Grids Impact of Packet Size on Energy Savings Impact of Node Degree on Energy Savings Impact of Network Size on LAF Impact of Error in Location Estimate Summary An Energy-Efficient Data-Delivery Scheme for Delay-Sensitive Traffic Introduction Related Work SensiQoS Design Location Information Packet Header Format SPEED Protocol SensiQoS Packet Scheduler Data Aggregation xi

12 4.3.6 MAC-layer QoS Support Feedback-based Congestion Control Analysis Network-wide Speed Expected Number of Transmissions Impact of Localization Error Space Complexity Time Complexity Performance Evaluation Energy Model Simulation Environment Service Differentiation Energy Savings Packet Deadline Miss Ratio Node Density Impact of Aggregation Factor Impact of Event Occurrence Summary Data Collection in Event-Driven Networks with Mobile Sinks Introduction Related Work HTDC Design System Model Virtual Grid Construction Geographic Hash Table xii

13 5.3.4 Hash Functions Local Event Announcer Node Data Collection HTDC Messages Load Sharing Routing of Messages to the Mobile Sink Analysis Model and Notation Energy Consumption Time Complexity Other Data Collection Algorithms Reactive Data Collection (RDC) Continuous Data Collection (CDC) Ideal Data Collection (IDC) Performance Evaluation Energy Model Simulation Environment Effect of Mobile Sinks Effect of the Number of Events Effect of Cell Size Impact of Error in Localization Summary Conclusions and Future Work Thesis Contributions Future Work xiii

14 6.2.1 Energy-Efficient Reliability for Correlated Data Real-Time Data Collection Protocol: RT-HTDC Bibliography 156 Biography 169 xiv

15 List of Tables 1.1 Comparison of different data dissemination protocols Comparison of different self-organization protocols Radio characteristics [48] Simulation Environment Parameters Simulation Environment Parameters xv

16 List of Figures 1.1 Inventory tracking using sensor networks Battlefield monitoring using sensor networks A typical sensor node An example of ad hoc sensor network deployment A hierarchical sensor network Information dissemination: (a) SPIN protocol; (b) Recursive geographic forwarding; (c) Multi-path forwarding; d) Directed diffusion Self-organization in sensor networks: Connected sensor cover Self-organization in sensor networks: ASCENT Procedure Initialize Procedure recvdatapacket Sensing and transmission radii of a node Network partitions in the basic scheme Illustration of the relationships between the time intervals Procedure TimerExpire The state diagram for SCARE Result of self-configuration using SCARE Illustration of how network partitioning is prevented in SCARE Approximation of coverage region of a sensor node by a square Coverage due to SCARE versus all nodes xvi

17 2.12 Number of coordinators selected with an increase in nodes Coverage versus number of nodes for SCARE and Span Fraction of nodes selected as coordinators in SCARE and Span Coordinators selected in SCARE versus an ideal number of coordinators selected based on square tiling Number of control messages sent for self-configuration Packet loss rate as a function of pause time Fraction of nodes remaining with time for Span and SCARE Effect of node failure on SCARE and a RDC method Effect of error in distance estimation on SCARE Number of Coordinators selected versus s/r Coverage obtained versus s/r Example illustrating modified flooding Energy savings in modified flooding over classical flooding Example of a virtual grid Packet header format in LAF Illustration of gateway nodes and internal nodes in a virtual grid Procedure GWNodePacketForward Procedure InternalNodePacketForward Linear network with N nodes Energy consumption for LAF (analytical result) Test network used in the simulations Data disseminated in the system with time A zoom-in view of Fig Energy savings due to LAF Effect of number of virtual grids on energy consumption xvii

18 3.15 Effect of number of packet size on energy savings Effect of average degree of a node on energy consumption Effect of network size on LAF In a sensor network, performing data aggregation may increase the latency of the data packets. In this connected topology, data generated by nodes B, C, D, E is correlated. If node C waits for packets from node E to perform data aggregation, this may increase the delay for its packets to reach the sink node A Packet Header for SensiQoS Illustration of SPEED protocol Procedure recvdatapacket Packet Organization in SensiQoS Service differentiation Normalized histogram of the packet arrival times for the high priority event Energy consumption Packet deadline miss ratio with increasing number of flows Energy consumed with increasing node density Average delay with increasing node density Energy consumed versus aggregation factor Energy consumed versus event frequency An event has occurred in the sensor network with four mobile sinks Local Event Announcer Node. D is the hashed location of the LEAN node. A is the LEAN node for this virtual cell. B and C are the perimeter nodes for the LEAN node Source nodes in each cell send an event announcement message to LEAN LEAN nodes send a DCR packet to the mobile sink to request data collection xviii

19 5.5 Procedure AnnounceEvent Procedure SendDCRPacket Procedure DataCollection Total number of transmissions versus number of mobile sinks Data-collection delay versus the number of mobile sinks Total number of transmissions versus number of events Data-collection delay versus number of events Total number of transmissions versus cell size Data collection request xix

20 Acknowledgements The work presented in this dissertation would not have been possible without the help of many individuals, and I wish to thank everyone that has helped make it possible. I am indebted to my advisor, Prof. Krishnendu Chakrabarty, for his constant support over the years. He has been an excellent mentor and guide, not only guiding me in transforming preliminary ideas into effective solutions but also enabling me to move forward when I got stuck. His support, energy and discipline have been an inspiration for me which I shall remember in the future for my own professional life. I am grateful to the members of my thesis committee, Prof. John A. Board, Prof. Kishor Trivedi, Prof. Jeffrey Krolik, Prof. Loren Nolte and Dr. John Zachary for their encouragements. I would like to address my special thanks to Dr. John Zachary for hosting me during the summer of 2003 and for the many stimulating discussions on sensor networks. I have been fortunate to have made several good friends at Duke, including Vamsee Pamula, Narayan Kovvali and Vijay Srinivasan. Finally, I express my deepest gratitude and appreciation to my family, who have always being there with me to share my happiness and grief. I dedicate this thesis to them for their love and sacrifice. Last, but not least, I would like to thank my grandmother, Jhansi Lakshmi Bai Talasila, for being a tremendous source of inspiration and teaching me the value of hardwork. xx

21 1 Introduction Advances in the miniaturization of microelectromechanical structures have led to battery-powered and cheap sensor nodes that have sensing, communication and processing capabilities. These sensor nodes can be networked in an ad hoc manner to perform distributed sensing and information processing. Such ad hoc sensor networks provide greater fault tolerance and sensing accuracy and are typically less expensive compared to the alternative of using only a few expensive and isolated sensors. These networks can also be deployed in inhospitable terrain or in hostile environments to provide continuous monitoring and processing capabilities for a wide range of applications [71], [77], [89], [122], [35], [64], [19]. A typical sensor network application is inventory tracking in factory warehouses. As illustrated in Fig. 1.1, a single sensor node can be attached to each item in the warehouse. These sensor nodes can then be used for tracking the location of the items as they are moved within the warehouse. They can also provide information on the location of nearby items as well as the history of movement of various items. Once deployed, the sensor network needs little human intervention and can function autonomously. 1

22 Factory Warehouse Internet User Computer Sensor nodes Figure 1.1: Inventory tracking using sensor networks. Another typical application of sensor networks lies in military situations. Sensor nodes can be air-dropped behind enemy lines or in inhospitable terrain. These nodes can self-organize themselves and provide unattended monitoring of the deployed area by gathering information about enemy defenses and equipment, movement of troops, and areas of troop concentration. They can then relay this information back to a friendly base station for further processing and decision making. This is illustrated in Fig. 1.2, where the presence of an enemy tank in the monitored area is relayed to the command center. Sensor nodes are typically characterized by small form-factor, limited battery power, and a small amount of memory. For example, the Spec Motes sensor nodes from Berkeley have a 900MHz radio with a radio range of 40 ft and a total stored energy on the order of 1J [40]. Due to their limited resources, many of the methods developed for the Internet and mobile ad hoc networks cannot be directly applied to sensor networks. New approaches and network protocols are required to solve the problems of localization, routing, naming, self-organization and data dissemina- 2

23 Command center Enemy tank Ad hoc sensor network Monitored area Figure 1.2: Battlefield monitoring using sensor networks. tion in sensor networks. The premise of this thesis is that location-aware protocols, which are based on the assumption that sensor nodes can estimate their location, improve the efficiency of data gathering and resource utilization of wireless sensor networks. Location-awareness improves the energy-efficiency of the protocols needed for routing, addressing, in-network data storage and security. A number of protocols proposed in the literature do not use location information [49], [17], and are therefore not energy-efficient. Sensor nodes can typically estimate their location either through the use of GPS [42] or using less expensive localization mechanisms [33, 1]. The use of location information for improving the efficiency of network routing protocols in wireless ad hoc networks has also been investigated [129], [109], [73]. This thesis is targeted towards wireless sensor networks, which are typically more resource-constrained than wireless ad hoc networks. We propose a number of location-aware and energy-efficient network protocols for self-organization, data dissemination, service differentiation and data collection. The remainder of this chapter is organized as follows. The challenges involved 3

24 in designing protocols for wireless sensor networks are discussed in Section 1.1. In Section 1.2, we discuss some typical sensor network architectures. In Section 1.3, various data dissemination protocols for wireless sensor networks are discussed. Section 1.4 describes the need for self-configuration in sensor networks and presents an overview of self-configuration algorithms presented in the literature. In Section 1.5, we outline our data delivery scheme for delay-sensitive traffic that leverages location information to deliver the data to the destination in a timely manner. Section 1.6 describes our data collection mechanism for event-driven sensor networks that uses a geographic hash table to locate the mobile sinks as well as to route the packets to the mobile sink to collect the data in an energy-efficient way. Finally, we present an overview of the thesis in Section Design Challenges in Wireless Sensor Networks In this section, we review some issues involved in the design of self-organization and data dissemination protocols in wireless sensor networks. Due to their resource constraints and unique application requirements, sensor networks pose a number of challenges. These are summarized below: Small Memory: Sensor nodes usually have a small amount of memory. Hence, sensor network protocols should not require the storage of a large amount of information at the sensor node. For example, the Berkeley Spec Motes have 3Kb of memory [40]. Limited Battery Power: Sensor nodes typically have a small form factor with a limited amount of battery power [76]. Furthermore, radio communication typically costs more in terms of energy compared to computation costs in a sensor node. Therefore, protocols designed for sensor networks should utilize only a few control messages. 4

25 Fault Tolerance: Sensor nodes are prone to failure. This may be due to a variety of reasons. Loss of battery power may lead to failure of the sensor nodes. Similarly, when sensor nodes are deployed in hostile or harsh environments as in the case of military or industrial applications, sensor nodes might be easily damaged. Thus, protocols designers should build fault tolerance into their algorithms for improving the utility of sensor networks. Self-Organization: Sensor nodes are often air-dropped in hostile or harmful environments. It is not possible for humans to reach these sensor nodes. Besides, it is not possible for humans to repair each sensor node, as often the number of sensor nodes is quite large. Hence, self-organization of sensor nodes to form a connected network is an essential requirement. Scalability: The number of sensor nodes in a sensor network can be in the order of hundreds or even thousands. Hence, protocols designed for sensor networks should be highly scalable. 1.2 Sensor Network Architectures This section describes a typical sensor node and discusses different sensor network architectures. A typical sensor node shown in Fig. 1.3 consists of four basic components: a power unit that may be battery-powered, a sensing unit that may consist of one or more sensors, a processing unit that consists of a CPU to provide a basic processing capabilities, a DSP chip to provide limited signal processing functions, and a transceiver to provide untethered communications. Sensor network applications such as inventory tracking, perimeter defense, and environmental monitoring require careful planning in the design of the protocols, including the choice of the sensor network architecture and the amount of redundancy to be present in the network. There are several sensor network architectures that protocol designers might 5

26 Figure 1.3: A typical sensor node. consider for their applications: Homogeneous versus heterogeneous: A sensor network may consist of homogeneous or heterogeneous nodes. In a homogeneous sensor network, all the sensor nodes have similar sensing and processing abilities. As a typical sensor network can have up to thousands of nodes, homogeneous sensor networks are economical due to reasons of scale. A heterogeous sensor network may consist of sensor nodes with different sensor types, power capacities and processing abilities. An example of a heterogeneous sensor network is a habitat monitoring network where sensor nodes with cameras perform the video sensing while sensor nodes with recorders perform audio sensing, both with different power requirements and processing abilities. Thus, different protocols are needed for homogeneous and heterogeneous sensor networks. Random versus deterministic deployment: Sensor nodes can be deployed by air-dropping them (Fig. 1.4) or throwing them randomly in a target area or they can be placed at pre-determined locations using a deterministic scheme. Protocols for self-configuration of a randomly-deployed network may not be 6

27 Figure 1.4: An example of ad hoc sensor network deployment. well suited for a deterministically-deployed sensor network. Similarly, data dissemination algorithms designed for deterministic sensor networks may not perform well when used in randomly-deployed sensor networks. Hierarchical versus flat topology: Designers can select either a flat topology or a hierarchical cluster-based topology depending on the application for their protocol. Hierarchical topologies are generally more suited for sensor networks as they allow data fusion and other common functions within a cluster, thus minimizing communication outside a cluster. A 3-level hierarchical sensor network is shown in Fig Static versus mobile: Sensor networks can consist of either static or mobile nodes, or a mixture of both static and mobile nodes. Depending on the composition of the particular sensor network, they may require very different algorithms for self-organization and data dissemination protocols. 1.3 Data-Dissemination Protocols Once sensor nodes are deployed, efficient protocols are needed to disseminate the data sensed by the sensor nodes. Data dissemination involves the sending of the sensed data to the nodes that requested the data from the area where the event has 7

28 Figure 1.5: A hierarchical sensor network. occurred. As many sensors might be present in the area of event occurrence, data might be duplicated and nodes may receive multiple copies, or data may be lost due to a lossy communication channel. Several schemes have been proposed for data dissemination in sensor networks. In this section, we describe the state-of-the-art data dissemination protocols for wireless sensor networks. Traditionally, flooding is used in networks to disseminate information [52]. It is also used in several routing algorithms in sensor networks [49]. In flooding, the source node broadcasts the packet to all its neighbors. Each node that receives the packet stores a copy of the packet and broadcasts the packet to all its neighbors. Flooding terminates when a maximum number of hops are reached or the destination of the packet is the node itself. Flooding is robust to node failures and delivers the packet to all the nodes in the network provided the network is lossless. However, the following problems might exist if flooding is done indiscriminately. As each node may be in the transmission 8

29 range of many other nodes, each node might receive multiple copies of the same packet, thereby resulting in wastage of energy. Similarly, as sensor networks are typically very dense, heavy contention might result because many nodes are trying to acquire the channel at the same time. These problems are collectively referred to as the broadcast storm problem [78]. In addition to nodes receiving redundant packets as discussed above, another problem might occur in wireless sensor networks. If the packet received by a node already has some or all of the data contained in the packet, wastage of energy occurs. This is known as the overlap problem [63]. Gossiping [85] is another data dissemination protocol traditionally used in ad hoc networks. In gossiping, the source node sends the packet to a randomly-selected neighbor. Each node that receives the packet randomly selects a neighbor and sends the packet to it. This process is repeated by all the nodes that have received the packet. Thus, data is disseminated throughout the network. In flooding, when a node with high degree receives a packet, it broadcasts the packet to all its neighbors. In turn, all the neighbors broadcast new copies of the packet. Thus, the network is flooded with multiple copies of the same packet. Gossiping avoids the implosion problem by sending the packet to only one neighbor. This also results in energy savings. However, the information is also disseminated at a slower rate compared to flooding. Moreover, gossiping does not solve the overlap problem. SPIN protocols are a set of resource-adaptive information dissemination protocols for wireless sensor networks [63]. In SPIN, nodes use meta-data to describe the data they possess. Nodes negotiate through a set of protocols to request the data they do not possess. When a node obtains new data, it broadcasts a ADV message to all of its neighbors with the meta-data describing the new data. Nodes that have received the ADV message checks the meta-data to see if it already has the data. Otherwise, it sends a REQ message to the sender of the ADV message requesting the data. The sender responds with a DATA message containing the requested data 9

30 (a) (b) (c) (d) Figure 1.6: Information dissemination: (a) SPIN protocol; (b) Recursive geographic forwarding; (c) Multi-path forwarding; d) Directed diffusion. and the protocol terminates. SPIN achieves energy savings by eliminating requests for redundant transmissions of data. Upon receipt of a ADV message, a node need not send a REQ message if it already has the data. Similarly, a node can aggregate its data with the newly received data and send an ADV message for the aggregated data. Nodes are also resource-adaptive in SPIN. Nodes poll their system resources for the amount of remaining energy and make informed decisions about disseminating information. This 10

31 is shown in the Fig. 1.6(a), where node A advertises its data by broadcasting an ADV packet. Nodes B and C respond by requesting the data using a REQ packet. GEAR [125] is a recursive data dissemination protocol for wireless sensor networks. A target region is specified in each query packet. Initially, the query packet is forwarded towards the target region. GEAR uses a set of geographically informed heuristics to route packets to the target region. Once the packet reaches the target region, it then uses a recursive geographic forwarding scheme to disseminate the packet within the target region. Each node within the target region splits its region into subregions and sends a copy of the query packet to each of the subregions. This recursive splitting and forwarding is terminated when a node is the only one in the subregion. The splitting of a region into subregions in recursive geographic forwarding is illustrated in Fig. 1.6(b). Information dissemination in sensor networks is made information-aware in [24]. Every data packet is assigned a priority level based on its information content and criticalness. Packets carry a small amount of state to help make forwarding decisions at individual nodes. The ReInForM technique in [24] uses local knowledge of channel error rates and neighborhood at each node. It provides the desired amount of reliability in data delivery by sending multiple copies of the same packet through multiple paths from the source to the sink. Multi-path forwarding from a source to a sink is illustrated in Fig. 1.6(c). A total of five paths exist between the source and the sink. The number of packets sent to the sink through each path assuming the communication links are lossless is shown in the figure. Directed diffusion [49] is a data-centric paradigm for disseminating information. In Directed diffusion, data is named using attribute-value pairs. Query for a sensing task is distributed throughout the network as an interest for named data. This dissemination sets up gradients within the network to draw events matching the interest. The sensor network re-inforces a small number of these paths. Fig. 1.6(d) 11

32 Table 1.1: Comparison of different data dissemination protocols. Dissemination Avoids Avoids Resource Fault Protocol overlap? implosion? adaptive? tolerance Flooding No No No High Gossiping No No No Low SPIN Yes Yes Yes High GEAR No Yes Yes Medium Directed Diffusion Yes No Yes High ReInForM No No Can be incorporated High shows the three steps involved in directed diffusion, including interest propagation from the sink to the source, the setup of gradients from source to the sink, and the re-inforcement of better paths from the source to the sink. Directed diffusion positively re-inforces certain paths and negatively others to repair the paths that have failed nodes in them. It is a reactive routing technique and enables in-network aggregation of data by application-specific filters at each node in the network. A qualitative comparison of different data dissemination protocols is shown in Table 1.1. Improving the efficiency of broadcasting has been studied extensively in the context of wireless ad hoc networks. However, the resource constraints of wireless sensor networks present new challenges that have not been studied in the prior work. In this thesis, we present a location-aided information dissemination protocol that uses the concept of virtual grids to reduce the redundant receptions of the flooded packet and saves energy. 1.4 Self-Configuration Protocols Several sensor network applications require unattended autonomous operation for extended periods of time. Hence, sensor nodes should self-organize themselves and perform data gathering and processing in spite of node failures, loss of temporary com- 12

33 Figure 1.7: Self-organization in sensor networks: Connected sensor cover. Figure 1.8: Self-organization in sensor networks: ASCENT. munication links and node movement. In this section, a number of self-organization protocols in ad hoc sensor networks are discussed. Span [17] attempts to save energy by switching off redundant nodes without affecting the connectivity of the network. In Span, a limited set of nodes self-organize themselves to form a multi-hop forwarding backbone while other nodes go to sleep. Nodes make decisions based on their local topology information. A TDMA-based self-organization scheme for sensor networks is presented in [100]. Each node uses a superframe, similar to a TDMA frame, to schedule different time slots for different neighbors. In each slot, a node can only talk to that neighbor for which the slot is reserved. Either code division multiple access (CDMA) or fre- 13

34 quency division multiple access (FDMA) is used to prevent collision of packets among potentially interfering links. However, this scheme does not take advantage of the redundancy inherent in wireless sensor networks to power off some nodes. The work in [36] introduced the concept of connected sensor cover for selforganizing the sensor network to achieve energy savings. A connected sensor cover for a query is the minimum set of sensors such that the sensing region of the selected sensors covers the entire geographical region of the query and the selected set of sensors form a connected communication graph. At each stage, the algorithm selects a path of sensors that connects an already connected sensor to a partially connected sensor. The selected path is then added to the already selected sensors at that stage. The algorithm terminates when the selected set of sensors completely cover the query region. This is illustrated in Fig. 1.7 where nodes A, B, C, D, E, F, G, H, and I completely cover the sensor field and form a connected sensor cover. This algorithm results in self-organization of the sensor nodes for a specific query to improve the energy efficiency of the sensor network. The techniques mentioned in [17, 100, 36] consider a flat topology for the sensor network. Alternatively, self-organization can also be achieved by grouping sensor nodes into clusters. A self-organizing algorithm termed Rapid, for message-efficient clustering based on the concept of budget allocation is presented in [62]. Rapid uses very few messages to produce clusters of bounded size. A node that wants to build the cluster initiates the process by allocating a certain budget to itself and broadcasts the remaining budget among its neighbors by sending each neighbor a message. Nodes that receive the budget account for themselves and distribute the remaining budget among themselves. Each node that receives a message sends an acknowledgment to its parent when either the budget is exhausted or it has received acknowledgments from all its children. The algorithm terminates when the node that initiated the algorithm receives acknowledgments from all the neighbors it sent 14

35 a message to. The Rapid algorithm always produces clusters less than the desired limit. However, sometimes the produced cluster may be quite small compared to the desired bound. This is improved in Persisent [62], where a node does not immediately send an acknowledgment to its parent on receiving acknowledgments from all its children. It checks to see if the budget allocated to each of its neighbors has been exhausted. If the budget is not exhausted, it distributes the remaining budget among the neighbors it has not previously explored. Thus a node returns an acknowledgment only when its allocated budget has been met or if further growth is not possible. These algorithms produce single clusters of bounded size using few messages. ASCENT [15] is a self-organizing scheme that provides topology control for sensor networks. In ASCENT, each node assesses its connectivity and adapts its participation in the multi-hop network. ASCENT has several phases. Upon initialization, each node enters into a listening-only phase called the neighbor discovery phase, where each node obtains an estimate of the number of neighbors actively transmitting based on local measurements. At the end of this phase, nodes enter into the join decision phase, where they decide whether to join the multi-hop sensor network. During this phase, a node may temporarily join the network for a certain period of time to check if contributes to improved connectivity. If a node decides to join the network for a longer the network for a longer time, it enters into an active phase and participates in the routing protocols of the network. If a node decides not to join the network, it enters into an adaptive phase where it turns itself off for a period of time or reduces its transmission range. Fig. 1.8 illustrates the various phases involved in ASCENT. The sink node A sends help messages in the neighbor discovery phase that results in neighborhood announcement by other nodes in the join decision phase. Nodes that have joined the network are in the active phase and participate in forwarding data from the source to the sink. ASCENT improves the energy-efficiency of the sensor 15

36 network without a significant improvement in message loss. It is also adaptive to the traffic in the network and is stable under various network traffic conditions. A qualitative summary of the characteristics of different data dissemination protocols is presented in Table 1.2. While the problem of self-organization of sensor networks has been studied in considerable detail, the use of location information to improve energy-efficiency has largely been ignored. This thesis presents a scheme that exploits the redundancy inherent in wireless sensor networks to improve the energy-efficiency of the selforganization process. Table 1.2: Comparison of different self-organization protocols. Self-Configuration Query Fault Energy Topology protocol specific? tolerance efficiency Span Flat No High Medium TDMA Flat No Low Medium Connected Sensor Cover Flat Yes Medium High Rapid Hierarchical No Low High Persistent Hierarchical No Low High ASCENT Flat Yes High High 1.5 Data Delivery of Delay Sensitive Traffic The sensed information in a sensor network typically has the following characteristics. 1. The information in a sensor network is closely tied to the location of the sensor nodes where data is generated. For example, tracking applications only care where a target is located, not the ID of the reporting node. 2. The generated data can have different real-time deadlines. For example, in a forest fire detection application, the data that suggests that a forest fire is imminent may have a shorter real-time deadline to reach the sink node than the data that consists of periodic temperature readings. 16

37 3. The data generated by sensor nodes has both spatial and temporal correlation. Such correlation of the data can be leveraged for performing in-network data aggregation, thereby potentially reducing the number of packet transmissions and hence, extending the lifetime of the network. Our focus here is on the timely delivery of sensor-generated data in an energyefficient way. We are interested in event-driven sensor networks, where multiple detections of the same event can occur within a short time period in a relatively close geographic region. Many applications of sensor networks such as habitat monitoring and target tracking fit this model, as events in these sensor networks often occur in specific geographic regions. For example, in a habitat monitoring sensor network application, a query might look like this: Notify me within 2 minutes whenever the number of animals in a geographic region [x, y] increases above 50? Traditional service differentiation protocols such as IntServ and DiffServ [75] for wired networks support real-time traffic with latency constraints through end-toend signaling and resource reservation. However, such protocols are not suitable for wireless sensor networks due to several reasons, e.g., dynamic topology changes due to node addition, failure, and node mobility. The design of an energy-efficient data-delivery scheme with latency constraints is a major challenge for wireless sensor networks. There is a trade-off between performing in-network data aggregation and achieving timeliness. Methods that optimize data aggregation by enabling maximum path sharing increase the queueing delay at relay nodes due to increased inbound traffic and waiting time for the arrival of the data packets to be aggregated [61]. In real-time applications, such increased queuing delays typically results in longer packet delivery latencies and can make the packets miss their timeliness deadlines and thus can overshadow the energy savings of the in-network aggregation. 17

38 In this thesis, we propose a novel data delivery scheme for delay-sensitive traffic called SensiQoS that significantly reduces the energy consumption in wireless sensor networks without reducing the number of packets that meet the end-to-end real-time deadline. SensiQoS leverages the inherent properties of the data generated by events in a sensor network such as spatial and temporal correlation, and realizes energy savings through application-specific in-network aggregation of the data. SensiQoS maximizes energy savings by adaptively waiting for packets from upstream nodes to perform in-network processing without missing the real-time deadline for the data packets. SensiQoS is a distributed packet scheduling scheme where nodes make localized decisions on when to schedule a packet for transmission to save energy and to which neighbor they should forward the packet to meet its end-to-end realtime deadline. We present a randomized algorithm where nodes adapt locally to the network-traffic to maximize energy savings in the network. Simulation results show the energy-efficiency of the proposed approach without the reducing the number of packets that miss the real-time deadline. 1.6 Data Collection with Mobile Sinks There is a new focus in the research community on data collection with the help of mobile sinks in sensor networks [114], [58], [74], [16], [110], [116], [127], [27], [108], [34], [38], [92]. Typical sensor network applications generate large amounts of data and send that data to the base station using multi-hop routing. However, transporting large quantities of data to the base station can quickly drain the limited energy resources of the sensor nodes and reduce the lifetime of the sensor network. One way to significantly reduce communication cost in sensor networks is to perform in-network aggregation (e.g., AVG and MIN) [19, 11] or in-network processing (e.g., beamforming [79], [5])). However, due to the inherent loss of detail, these techniques do not provide the fine data granularity desired by several sensor network applica- 18

39 tions. Often times, all of the data generated by the sensor network is necessary for the purposes of accuracy and modeling in these applications. For instance, it has been noted that in a networked structural health monitoring application [20], more than 500 samples per second are required to efficiently detect damages. Another example is the the Sonoma Redwoods project [103] where the biologists researching the project require as much detailed data from the sensor network as possible to evaluate various physical models and test various hypotheses over the data. Besides, such multi-hop routing with static sink nodes results in the early death of the one-hop neighbors of the sinks and renders the sensor network unusable. This is expected because most of the data traffic is relayed to the sink by these nodes, thus greatly increasing their energy consumption, resulting in their untimely death and partition of the network topology. Consequently, nodes located remotely will be unable to report to the sink, which greatly reduces the lifetime of the WSN. Besides, in many situations, a static sink may be infeasible because of deployment or security constraints. Similarly, in hostile or inhospitable environments, it may not be feasible to replenish the batteries of the sensor nodes. Recently, mobile data sinks have been proposed as a solution for data collection to balance the energy consumption among the sensor nodes throughout the network geographically [58], [74], [16], [110], [116], [127], [27]. Mobile sinks, either proactively or reactively, move in the sensor network and collect the data from the sensor nodes. This not only solves the early death problem for the one-hop neighbors of the sink but also extends network lifetime by distributing the responsibility of relaying data to the sink among many nodes in the sensor network. Furthermore, mobile sinks are not only a solution to prolong network lifetime, but also a requirement for many applications. The sink mobility assumption may be useful for applications such as target tracking, emergency preparedness, and habitat monitoring. It is also advantageous to use the strategy of mobile sinks in other environments, where vehicles 19