CONSUMED-ENERGY-TYPE-AWARE ROUTING FOR WIRELESS SENSOR NETWORKS A Thesis submitted to the Graduate School University of Arkansas at Little Rock in partial fulfillment of requirements for the degree of MASTER OF SCIENCE in Computer Science in the Department of Computer Science of the Donaghey College of Information Science and System Engineering May, 2007 Shinya Ito BS, University of Aizu, Fukushima, Japan, 2005 Ⓒ Copyright 2007 Shinya Ito All Rights Reserved UNIVERSITY OF ARKANSAS AT LITTLE ROCK DONAGHEY COLLEGE OF IMFORMATION SCIENCE SYSTEM ENGINEERING DEPARTMENT OF COMPUTER SCIENCE This thesis, “Consumed-Energy-Type-Aware Routing for Wireless Sensor Networks,” by Shinya Ito, is approved by: Thesis Adviser ________________________________________________ Kenji Yoshigoe Assistant Professor of Computer Science Thesis Committee ________________________________________________ Coskun Bayrak Professor of Computer Science ________________________________________________ Xian Liu Associate Professor of System Engineering Program Coordinator ________________________________________________ Coskun Bayrak Professor of Computer Science Graduate Deans ________________________________________________ Ibrahim Nisanci Dean of the Graduate School Fair Use This thesis is protected by the Copyright Laws of the United States (Public Law 94-533, revised in 1976). Consistent with fair use as defined in the Copyright Laws, brief quotations from this material are allowed with proper acknowledgement. Use of this material for financial gain without the author’s express written permission is not allowed. Duplication I authorized the Head of Interlibrary Loan or the Head of Archives at the Ottenheimer Library at the University of Arkansas at Little Rock to arrange for duplication of this thesis for educational or scholarly purposes when so requested by a library user. The duplication shall be at the user’s expense. Signature ____________________________________________________ Acknowledgement I would like to express my gratitude to those who gave me the possibility to complete this thesis. First of all, I would like to thank my thesis advisor Dr. Yoshigoe for his generous support and stimulating suggestions. His encouragement helped me throughout this research. I would like to thank my thesis committee members Dr. Coskun Bayrak and Dr. Xian Liu for all their help and valuable advice. I also would like to thank the University of Arkansas at Little Rock for giving me a chance to learn computer science in the United States and offering me a scholarship to continue the coursework. Finally, I would like to give my special thanks to my parents. Without their support, it would have been impossible to have studied in the United States. i Table of Contents PAGE List of Tables ..................................................................................................................... iii List of Figures .................................................................................................................... iv Glossary of Acronyms ...................................................................................................... vii Abstract ............................................................................................................................ viii Chapter 1 : Introduction ...................................................................................................... 1 1.1 Wireless Sensor Networks ................................................................................................... 1 1.2 Motivation ............................................................................................................................. 5 1.3 Contribution ......................................................................................................................... 7 1.4 Organization ......................................................................................................................... 7 Chapter 2 : Background and Related Works....................................................................... 8 2.1 Routing Algorithms .............................................................................................................. 8 2.1.1 Traditional Routing Algorithms .................................................................................. 8 2.1.2 Routing Algorithms in Mobile Ad-Hoc Networks (MANETs) .................................. 10 2.1.3 Routing Algorithms in WSNs ..................................................................................... 13 2.1.4 Energy Aware Routings in WSNs .............................................................................. 18 2.1.5 Hierarchical Technique in WSNs .............................................................................. 19 2.2 Other Energy Aware Protocols .......................................................................................... 22 2.3 Localization Algorithms ..................................................................................................... 27 Chapter 3 : Methodology .................................................................................................. 29 3.1 Consumed-Energy-Type-Aware Routing (CETAR) .......................................................... 29 3.2 CETAR with Biased Consumed Energy (BCE) ................................................................ 33 ii 3.3 CETAR with Aggressively-and-Adaptively Biased Consumed Energy (AABCE) .......... 35 3.4 Effect of Weighting Function on the BCE and AABCE ................................................... 38 3.5 CETAR for GEAR ............................................................................................................... 40 3.6 The Application of CETAR for the Other Type of WSN Protocol .................................... 46 Chapter 4 : Evaluation of CETAR .................................................................................... 49 4.1 Simulation Models ............................................................................................................. 49 4.1.1 Assumption for the Simualtion .................................................................................. 49 4.1.2 Common Experimental Settings ................................................................................ 51 4.1.3 Type of Traffics for CETAR ........................................................................................ 52 4.2 Evaluation for CETAR with BCE...................................................................................... 55 4.2.1 Experimental Settings for BCE ................................................................................. 55 4.2.2 Evaluation for BCE ..................................................................................................... 56 4.2.3 Effectiveness of β for BCE .......................................................................................... 60 4.3 Evaluation for CETAR with AABCE ................................................................................ 63 4.3.1 Experimental Setting for AABCE .............................................................................. 63 4.3.2 Evaluation for AABCE ............................................................................................... 64 4.3.3 Effectivity of Weighting Functions for AABCE ........................................................ 68 4.3.4 Effectiveness of AABCE over BCE in Dynamically Non-Uniform Traffic .............. 72 Chapter 5 : Conclusion and Future Works ........................................................................ 76 5.1 Contributions of This Research..................................................................................... 77 5.2 Discussions for Future Works ....................................................................................... 77 REFERENCES ................................................................................................................. 79 iii List of Tables PAGE Table 1 - The average percentage of packet sent over GEAR for CETAR with BCE ......59 Table 2 - The percentage of packet sent over GEAR for CETAR with AABCE ..............68 Table 3 - Improvement of CETAR with AABCE over CETAR with BCE (1) .................68 Table 4 - The average of experimental result in Figure 4.18 and Figure 4.19...................75 Table 5 - Improvement of CETAR with AABCE over CETAR with BCE (2).................75 iv List of Figures PAGE 1.1 - Sensor nodes scattered in a sensor field ......................................................................1 1.2 - The components of sensor node ..................................................................................3 1.3 - A State Diagram for WSNs .........................................................................................4 1.4 - Power consumption of node subsystems (mW) ..........................................................5 2.1 - AODV algorithm .......................................................................................................12 2.2 - Greedy forwarding. y is x’s closest neighbor y to the destination node D ................15 2.3 - Node x faced routing hole for destination D..............................................................16 2.4 - Greedy Other Adaptive Face Routing (GOAFR) ......................................................17 2.5 - Packet forwarding with and without clustering and aggregation ..............................20 2.6 - S-MAC scheduling ....................................................................................................24 2.7 - Routing by COMPOW in a typical non-homogeneous networks .............................26 3.1 - The situation at the deployment of nodes in WSNs ..................................................30 3.2 - Intensive traffic from the fixed sets of nodes ............................................................30 3.3 - Intensive traffic from the semi-fixed sets of nodes ...................................................31 3.4 - An example of CETAR .............................................................................................33 3.5 - An example of CETAR with BCE ............................................................................35 3.6 - Graph of β and actually used square root β adapted by CEs(Ni) and CEr(Ni)............37 3.7 - Utility functions changes with increase in β ..............................................................39 3.8 - Greedy packet forwarding .........................................................................................41 3.9 - Load balancing and the shortest path routing ............................................................42 3.10 - An Example of GEAR incorporating CETAR ........................................................44 v 3.11 - Routing on the hierarchical network structure ........................................................47 4.1 - Relation of transmission range between fixed and adjustable power ........................50 4.2 - 10 given source and target region pairs .....................................................................52 4.3 - 10 sources out of 30 source candidates .....................................................................53 4.4 - 10 sources out of 10 source candidates .....................................................................54 4.5 - The network partitioning ...........................................................................................55 4.6 - CETAR with BCE (uniform traffic) ..........................................................................57 4.7 - CETAR with BCE (non-uniform traffic: A cluster of 10 out of 30 closest senders) 58 4.8 - CETAR with BCE (non-uniform traffic: A cluster of 10 closest senders) ................59 4.9 - CETAR with BCE for different β (uniform traffic) ..................................................61 4.10 - CETAR with BCE for different β (non-uniform traffic: A cluster of 10 out of 30 closest senders) ..........................................................................................................62 4.11 - CETAR with BCE for different β (non-uniform traffic: A cluster of 10 closest senders) ......................................................................................................................63 4.12 - CETAR with AABCE (uniform traffic) ..................................................................65 4.13 - CETAR with AABCE (non-uniform traffic: A cluster of 10 out of 30 closest senders) ......................................................................................................................66 4.14 - CETAR with AABCE (non-uniform traffic: A cluster of 10 closest senders) ........67 4.15 - CETAR with AABCE with utility functions (uniform traffic)................................69 4.16 - CETAR with BCE with utility functions (non-uniform traffic: A cluster of 10 out of 30 closest senders) .................................................................................................70 4.17 - CETAR with BCE with utility functions (non-uniform traffic: A cluster of 10 closest senders) ..........................................................................................................71 vi 4.18 - CETAR with BCE for 2 stage sender selection (uniform traffic) ...........................73 4.19 - CETAR with BCE for 2 stage sender selection (non-uniform traffic: A cluster of 10 out of 30 closest senders) ......................................................................................74 vii Glossary of Acronyms and Terms ACK AODV CE CETAR COMPOW CS CTS DATP DSR EAR GEAR GG GOAFR GPS HEED LEACH MAC MANET MPR OLSR PAMAS PEGASIS RDG RNG RREP RREQ RTS S-MAC SYNC TDMA TDRPF WSN Acknowledgement Ad-hoc on Demand Distance Vector Routing Consumed Energy Consumed-Energy-Type-Aware Routing Common Power Career Sensing Clear To Send Dynamically Adaptive Transmission Power Dynamic Source Routing Energy Aware Routing Geographical and Energy Aware Routing Gabriel Graph Greedy Other Adaptive Face Routing Global Positioning System Hybrid Energy-Efficient Distributed Clustering Low Energy Adaptive Clustering Hierarchy Media Access Control Mobile Ad-Hoc Network Multi-Point Reply Optimized Link State Routing Power-Aware Multi-Access Protocol with Signaling Power-Efficient Gathering in Sensor Information Systems Restricted Delaunay Graph Relative Neighborhood Graph Route Reply Route Request Request To Send Sensor-Media Access Control Synchronization Time Division Multiple Access Topology Dissemination Based on Reverse-Path Forwarding Wireless Sensor Network viii ABSTRACT Wireless sensor networks (WSNs) typically operate with sensor nodes with limited energy which is difficult or even impossible to be replaced due to the remote environment in which they are deployed. Thus, power management of sensor nodes is one of the important topics in WSNs. In this thesis, a new metric for energy aware routing, consumed-energy-type-aware routing (CETAR), is proposed to extend the lifetime of the WSNs. CETAR uses statistics of the energy consumed for each type of node activities including sensing, data processing, data transmission as a source node, and data reception/transmission as a routing node for routing decision. In particular, CETAR selects a node with high residual energy which seldom plays a role of source node as a routing node. Idea is to maintain the energy of active source nodes to prolong the functionality of the WSNs. Biased Consumed Energy (BCE), proposed in this thesis, derives a bias factor for a node that frequently plays a role of routing node and a node that frequently plays a role of source node. Simulation study demonstrates that the lifetime of the geographical and energy aware routing (GEAR) can significantly extend with CETAR based on BCE. Aggressively-and-Adaptively BCE (AABCE) further extends BCE by dynamically adapting the extent of the bias factor among consumed-energy types as the amount of consumed energy at each node changes. Simulation study demonstrates that the lifetime of GEAR with CETAR based on AABCE can further extended that of GEAR with CETAR based on BCE. ix This thesis also demonstrates the CETAR can be widely applied to the other type of network architecture to extend the lifetime of WSNs. With its adaptability to hierarchical network, the significance of CETAR to extend the lifetime of WSNs is clear. 1 Chapter 1 Introduction 1.1 1Wireless Sensor Networks Wireless Sensor Networks (WSNs) have been receiving a great amount of attention recently due to their substantial applicability to improve our lives [1]-[6]. They aid us by extending our ability to accurately monitor, study, and control objects and environments of various scales and conditions such as human bodies, geological surveys, habitats, and security surveillance. A WSN is composed of a large number of networked sensor nodes that are densely deployed either inside the phenomenon or to its proximity as shown in Figure 1.1 [7]. Figure 1.2 - Sensor nodes scattered in a sensor field [7] 2 This figure shows sensed data is delivered to the user. Suppose data is sensed by the sensor node A inside the sensor field. Since the transmission range of radio for each sensor is short, A, at first, passes sensed data to the neighbor node B. In this example, this data may be routed by the path A-B-C-D-E-Sink. Since sink is already connected to the Internet, it can deliver sensed data to the user directly from sink. Sensor nodes in WSNs can also autonomously process and cooperatively analyze sensed data inside networks so that they can prune the redundant data observed inside a network and deliver only necessary data to the user through sinks. Furthermore, WSNs can dynamically adapt its topology. After sensor nodes are deployed in a sensor field, they autonomously find the neighbor nodes and start communicating with each other in various ways, normally using multi-hop communications. This allows random deployment of sensors in inaccessible terrains or disaster relief operations. Figure 1.2 shows components of a sensor node including sensor, processor, storage and transceiver. Once the sensing unit captures an event, it converts the analog signal in to digital signal and passes it to the processor for possible processing. The data can be stored in the storage and later transmitted to a downstream sensor node. In this figure, extra components are enclosed in dotted lines. For instance, a location finding system such as GPS is not always necessary, and centroid localization calculated from the location of sinks is used instead [8]. 3 Figure 1.3 - The components of sensor node [7] An operational sensor node can be in one of the three states; active, idle, or sleep state. Figure 1.3 shows a node state transition diagram. As soon as node is powered on, its state transits to the idle state. In the idle condition, node senses career and moves to the active state if some signal is received (JOB_READY). On the other hand, if node does not sense any signal and timer is expired, it goes to the sleep state (TIMEOUT). In the sleep condition, node turns off most of its process and keeps remaining inactive. Even when node is inactive, it still senses wake up signal to turn itself on when WAKE_UP signal is received. In the active state, they perform activities such as processing, transmission, and reception. Immediately after its job is done (JOB_DONE), it moves to the idle state and wait a signal. 4 Figure 1.4 - A State Diagram in WSNs Energy consumed by radio signal transmission and reception in sensor node is dominant as shown in Figure 1.4. The amount of energy consumed for sensing and processing are significantly smaller than that consumed for the radio activities. Among the energy consumed by the node, energy used by the idle condition (listening) is also dominant. This amount increases as the operation time of node increases. Significant amount of literatures exist that efficiently schedule the sleep mode and listening mode [9]-[13]. Thus, that is not the scope of this thesis. With the advancement of technologies that has enabled a sensor device to have higher cost performance and better capabilities like high resolution sensing, observations in vast fields that require a number of sensor devices has become possible. Networking of these intelligent yet low-cost sensor devices is expected to revolutionize information gathering and processing in many situations. In contrast to various applicability of WSNs, the capacity of batteries inside their nodes is highly limited, and it is unrealistic to replenish battery of the nodes in many cases as described in Section 1.2. Therefore, the preservation of such vital energy at each sensor node is significantly important in WSNs 5 1.2 Motivation The study to reduce energy consumption rate in WSNs is significantly important since energy of battery consumed by every single activity such as sensing, processing, and transmission of data cannot be easily recharged due to geographical, environmental or sometimes financial reasons. For instance, sensor nodes could be deployed around volcanic areas that human beings cannot easily access. They could also be deployed in very large area where the redeployment can be costly. Even if their energy of nodes can be replenished in other cases, it is unrealistic for the user to manage energy information for each sensor node individually due to the size of WSNs. Thus, such management tasks are accomplished via collaboration among sensor nodes which requires significant energy overhead. As the energy of sensor nodes is depleted, the topology of the sensor network can be changed frequently. This would require reorganization of the network that expends additional energy consumption by the sensor nodes. Figure 1.5 - Power consumption of node subsystems (mW) [14] 6 In these environments, nodes should autonomously construct their connection after they are deployed, and the deployment of sensor nodes is one-time-only. This means the lifetime of sensor nodes will directly determine the lifetime of sensor networks. This study is motivated by the following observations regarding energy efficiency in sensor nodes: 1) In many applications, the frequency of sensing activities among the deployed sensor nodes in the network is not uniformly distributed. This is because, in many cases, we cannot specifically identify a set of definite observation points at the time of deployment phase of the network. 2) Energy consumed by radio signal transmission and reception in sensor node is dominant as shown in Figure 1.4. Transmission and reception consume almost 70 percent of total energy used for all node activities [14]. Therefore, reducing energy for transmission and reception activities has significant impact for extending the lifetime of sensor nodes. 3) The set of actively sensing nodes, as sources of data origination, consume extensive energy. Thus, their residual energy should be considered more precious than the residual energy of the node which does not perform sensing activities; however, no means are investigated to preserve actively sensing node in the literature. 4) Energy-Aware Routing (EAR) algorithms attempt to minimize energy requirements at each node or an overall network to transfer individual packets and to maximize the operation time of a given network. It normally calculates the least cost path based on several metrics including residual energy, 7 transmission power, and node location. Out of these metrics, residual energy plays the primary role in the routing decision. Though various EAR algorithms have been proposed and studied in literatures, none of the EAR algorithms take in consideration the amount of energy each type of activities consume. 1.3 Contribution This thesis proposes an approach to preserve the energy of actively sensing nodes in the WSNs and investigates its effectiveness. The major contributions of this thesis are as follows: 1) It proposes and investigates the method for preserving actively sensing source node in the WSNs. This method, Consumed-Energy-Type-Aware Routing (CETAR), improves the lifetime of the actively sensing nodes and the WSNs in many cases. 2) It investigates the adaptability of CETAR in well-known energy aware routing. 1.4 Organization The organization of the proposed work has the following structure layout. Chapter 2 provides the overviews of earlier studies of routing algorithms, localization algorithms, and methods that improve lifetime of WSNs. Chapter 3 describes CETAR, and demonstrates its adaptability to geographical and energy aware routing (GEAR) [8]. Chapter 4 evaluates the performance of CETAR on GEAR simulator. Chapter 5 proposes future work. 8 Chapter 2 Background and Related Works 2.1 Routing Algorithms Routing is the process used by data communication networks to deliver a packets from a source device to a destination device. For instance, the Internet dominantly uses a hop-by-hop routing model where each host or router that handles a packet at first checks the destination address in the IP header, computes the next hop that will bring the packet one step closer to its destination, and delivers the packet to the next hops where the process is continuously repeated. The rest of this section reviews a set of hop-by-hop routing algorithms as they are currently dominant in the Internet. 2.1.1 Routing Algorithms on the Internet Protocol There are mainly two types of routing algorithms in the Internet: distance vector algorithm and link state algorithm. Distance vector algorithms use the Bellman-Ford algorithm. The algorithm was developed by Richard Bellman and Lester R. Ford (Jr.) independently in [15] and [16]. This approach assigns a number and the cost to each of the links between each node in the network. Nodes will send information from point A to point B via the path that results in the lowest total cost (i.e. the sum of the costs of the links between the nodes used) [17]. At a initial point, each node only knows of its immediate neighbors, and the direct cost involved in reaching them. The information; that is, the list of destination, the total cost to each, and the next hop node for each destination 9 that consist of routing table, are passed. Each node send to each neighbor its own current total cost to get to all the destinations it knows. The neighboring nodes examine this information, and compare it to what they already know; that is, something which represents an improvement on what they already have, they insert in their own routing tables. Finally, all the nodes in the network will discover the best next hop for all destinations and the lowest total cost. When applying link-state algorithms, each node uses entire cost information of the network by using graph. To produce this graph, each node floods the entire network with information about what other nodes it can connect to, and each node then independently assembles this information into a graph. Using this graph, each router then independently determines the best route from itself to every other node. The algorithm used to derive the best route in link-state algorithm is Dijkstra's algorithm, proposed by Dijkstra in [18], Dijkstra's algorithm builds a tree data structure with the current node itself as the root node, and establishing a path to every other node in the network. It starts forming a tree containing only itself. Then, from the set of nodes which it has not yet added to the tree, it adds the node which has the lowest cost to reach an adjacent node which already appears in the tree. This continues until every node appears in the tree. This tree then serves to construct the routing table which gives the best next hop which has the least cost. Distance-vector routing protocols are simple and efficient in small networks, and require little management, if any. However, they do not scale well, and have poor convergence properties. On the contrary, advantage of link-state routing is that it reacts more quickly and in a bounded amount of time, to connectivity changes. Also, the 10 link-state packets that are sent over the network are smaller than the packets used in distance-vector routing. However, it requires more storage and more computing to run than distance-vector routing does. In the wireless sensor network, both type of traditional routing algorithm does not fit because of processing ability, memory, and severe bandwidth and power constraint. For instance, a large number of nodes might be deployed in the sensor field. To exchange routing table for those nodes consumes so much energy, bandwidth, and memory. Furthermore, it needs so much processing ability to fuse or aggregate the information inside the network. It is unrealistic situation for the energy-limited sensors in WSNs. 2.1.2 Routing Algorithms in Mobile Ad-Hoc Networks (MANETs) Both Mobile Ad-Hoc Networks (MANETs) and WSNs consist of wireless mobile nodes, and their nodes communicate with the other nodes in them by using wireless signal and multi-hop data relay. However, WSNs have several different characteristics. First, a battery of node in WSNs is not easily rechargeable because nodes are likely to be deployed in such a hostile environment and/or with high volume that access to these nodes can be risky and/or costly. This is not the case with MANETs where the battery of a laptop computer, cell phone, and PDA, etc can be easily recharged, Second, the high volume of WSN nodes require a significant amount of overhead during packet transmission. Likewise, the employment of global addressing scheme is also difficult in terms of energy efficiency and bandwidth utilization since the overhead of ID maintenance is high. Thus traditional IP-based protocol, applicable to MANET, may not be applicable for WSNs. Third, mobility of each node in WSNs are normally static after 11 deployment while that of MANET can be frequent. Thus, WSNs basically assume static node configuration after node deployment while MANET assumes dynamic node configuration. As known from their differences above, WSNs pursue energy efficiency, on the other hand, MANETs pursue the efficiency of throughput and decreasing communication delay. Regardless, the pursuit of MANET is also desirable and can be used in WSN application since the determination of routing technique in WSN is application specific and to balance energy-efficiency and communication efficiency is important. Routing protocol in MANET can be active or proactive. Reactive protocol requires to retrieve route information before initiating communication. Dynamic Source Routing (DSR) [19] and Ad Hoc on Demand Distance Vector Routing (AODV) [20] are the examples of reactive protocols. Proactive protocol keeps route information by exchanging periodic messages. Optimized Link State Routing Protocol (OLSR) [21] and Topology Dissemination Based on Reverse-Path Forwarding (TBRPF) [22] are the example of proactive protocols. In DSR [19], proposed by Johnson et al., sender node contains all route information to the destination in each packet header, called source routing. If sender initiates communication with destination-unknown node, at first sender node flood Route Request (RREQ) message inside the network. Each packet contains ID and each node broadcasts a received packet only one time. Each node relays packet by attaching its address information. Packet arriving at the destination has complete and ordered information of address of nodes of the traced path. Destination node uses this route and sends Route Reply (RREP) message to the sender. 12 AODV [19] proposed by Perkins et al., has routing table to determine which node needs to send their data for which node. Specifically, each node only knows the next relay node to route packet and it is not concerned about how the packet is delivered to the destination. In Figure 2.1 (a), for example, sender node S which does not know the routing path to a destination, at first floods a RREQ message inside the network. Relay node records the information of a node that relayed the message as the next hop node to the sender. By RREQ, all nodes inside the network come to have the route to send packet for sender node as described by dotted line. By using this information, destination node can send RREP message to the source node by unicast as shown in Figure 2.1 (b). At the same time, relay nodes know the information to relay the node to destination of that sender. By completing the processes mentioned above, bidirectional link is established between sender and destination. (a) (b) Figure 2.1 - AODV algorithm. OLSR [20], proposed by Clausen et al. optimizes the flooding by using Multi-Point Relay (MPR) set. MPR set consists of the previous relay nodes of packets to 13 a node. In the traditional flooding, each node needs to relay a packet at least one time; In OLSR, on the contrary, uses MPR set so that only the minimum amount of nodes relays are needed. Each node knows which node's MPR itself is for. If the topology of the network is changed by locomotion of nodes, the routing table in proactive type routing is affected. TBRPF [22], proposed by Ogier et al., uses the stable link information and difference information to avoid this effect. The objective of TBRPF is to create source tree as the shortest path for all nodes in the network. By possessing the same tree created by the same topology, all nodes can communicate each other. Since use of flooding technique is not efficient and creation of tree each time consumes substantial amount of resource, it maintains this tree by exchanging periodic reported sub tree information as difference information. 2.1.3 Routing Algorithms in WSNs A plethora of routing protocols has been investigated in WSNs [23]-[26], [32]. The earliest work known as geographic routing was introduced by Finn [23]. Geographic routing refers to a family of techniques to route data packets in a communication network. The main idea is that packets should be aware of their destination and messages will be routed hop-by-hop to nodes closer to the destination until the message reaches its final destination, which could be a point or a region in the case of geocasting. This implies that the hosts participating in the routing process should be aware of their geographic positions. Face Routing, introduced by Kranakis et al. in [24], walks along faces of planar graphs and proceeds along the line connecting to the source and the destination. Besides guaranteeing to reach the destination, it does so with O(n) messages, 14 where n is the number of network nodes. However, this is unsatisfactory since even a simple flooding algorithm can reach the destination with O(n) messages. Cost of routing can be determined by the distance between the source and the destination. Based on this observation, Intanagonwiwat et al. proposed directed diffusion [25], a data-centric protocol for sensor network applications. It achieves energy savings by cashing and processing data in network and by selecting empirically good paths. However, without geographic routing support, there is initial and periodic interest that results in low rate data flooding throughout the network. Since packet flooding consumes substantial amounts of energy in network, Karp et al. proposed Greedy Perimeter Stateless Routing (GPSR) [26], which elegantly avoided packet-flooding problems by deriving a planar graph1 out of the original network graph. Though GPSR is one of the most popular geographical routing protocols, it is originally designed for ad-hoc networks. From the characteristics of the geographic routing, GPSR tries to take the shortest path to forward packets to the next hop nodes as in Figure 2.2 1 A planar graph is a graph that can be drawn so that no edges intersect [27] 15 Figure 2.2 - Greedy forwarding. y is the x’s closest neighbor y to the destination node D [26] In this figure, to sends packet from node x to D, x at first sends its packet to the nearest neighbor node y, then y routes that packet to its nearest neighbor. GPSR assumes that each node knows its neighbor’s location information by using the GPS or localization method as described in [28]. The availability information of neighbors is exchanged by using some periodic local communication protocol. It continues until the packet arrives at destination node D. At the same time, it also follows the perimeter of the planer graph (i.e., path x-w-v-D or x-y-z-D as described in Figure 2.4) to circumvent the routing holes2. Greedy forwarding does not always work well. Figure 2.4 shows the situation which node x faced routing hole to send packet to the destination D. In this figure, x has the packet which destination is D. There are two ways to send this packet: the path w-v or y-z. However, for node x, neighbor node w and y are farther position than itself to the destination. So greedy forwarding will stop packet transmission at this time. To avoid this 2 A node in the routing hole has no neighbors closer to the destination than themselves as in Figure 2.4. 16 problem, some protocols have been proposed. One of them is right hand rule incorporated in GPSR. In right hand rule, after the node x received data, it sends this data to the first existing node which direction is counterclockwise from the node from which x received data. This approach requires a heuristic, the no-crossing heuristic, to force the right-hand rule to find perimeters that enclose voids in regions where edges of the graph cross [26]. Though GPSR is using Relative Neighborhood Graph (RNG) [29] and Gabriel Graph (GG) [30], it is recommended to use Restricted Delaunay Graph (RDG) [31] for gaining theoretically or practically high performance. GPSR takes short path; however, traffic will concentrate on the planar graph. Thus, only the nodes in that path will drain batteries quickly while other nodes will remain intact. Furthermore, GPSR assumes nodes to operate in promiscuous listening mode and is consequently always consuming energy. Figure 2.3 Figure 2.4 - Node x faced routing hole for destination D [26] 17 Kuhn et al. proposed Greedy Other Adaptive Face Routing (GOAFR) [32], a geographical routing algorithm that is both asymptomatically worst-case optimal and average-case efficient. Similar to GPSR, GOAFR combines greedy routing and face routing. As in Figure 2.6, starting at s, GOAFR proceeds in greedy mode until reaching the n1 facing routing hole F. The algorithm switches to face routing mode and explores the boundary of F to n2, the node closest to t on F's boundary. GOAFR falls back to greedy mode again and finally reaches t. In the next section, a set of routing algorithms especially suited for conserving energy of sensor nodes in WSNs are described. Figure 2.5 Figure 2.6 - Greedy Other Adaptive Face Routing (GOAFR) In [33], Gupta et al. investigated the performance of a new clustering algorithm for sensor networks. The proposed approach balances the load among clusters and simultaneously tries to cluster the sensor nodes close to high-energy cluster-heads, called 18 gateways. Simulation results have demonstrated the high efficiency of load balanced clustering for sensor networks applying different routing methods. This clustering algorithm is applied to the other WSN routing approaches and improves most of their performances in terms of the lifetime of the sensor networks. 2.1.4 Energy Aware Routings in WSNs Energy-Aware Routing (EAR) algorithms try to minimize energy requirements at each node or an overall network to transfer each packet and to maximize the operational time of a given network. It normally calculates least cost path based on several metrics such as residual energy, transmission power, and node location. Out of these metrics, residual energy plays the primary role in routing decision, and many protocols has been proposed in [8], [25], and [34]. Woo et al. [25] first proposed energy aware routing concept such as “maximize time to partition” and “minimize maximum node cost.” Deployment of these concepts results in excellent solutions; however, it is difficult to directly implement them in a local algorithm when even the centralized version of the same problem is NP-complete3. Chang et al [34] obtained the result that in order to maximize the lifetime of WSNs, the traffic should be routed such that the energy consumption is balanced among nodes in proportion to their remaining energy. Stojmenovic et al. [35] proposed a power-cost routing, which considers both the distance to the destination and the consumed energy of a node in the route selection. The particular power metric they used assumes arbitrary dense network and arbitrary adjusted transmission power. However, the specific format of the cost function will not have 3 Any of a class of computational problems for which no efficient solution algorithm has been found. Many significant computer-science problems belong to this class—e.g., the traveling salesman problem, satisfiability problems, and graph-covering problems [36]. 19 major impact on the performance mentioned in [8]. In addition, [35] did not address routing with the presence of a communication hole. Yu et al. [8], elegantly avoided the problem of GPSR (explained in the previous section) by proposing Geographical and Energy Aware Routing (GEAR). GEAR will not deterministically follow a particular set of nodes while routing around holes. GEAR uses two metrics for routing decision. One is the distance between packet sending node and its neighbor nodes, and the other is the remaining energy level of neighbor nodes. This simple heuristic in GEAR tends to avoid nodes that have been frequently visited before to achieve load balancing. 2.1.5 Hierarchical Technique in WSNs Many topologies have been proposed in WSNs. Topology types are mainly classified into four parts in terms of a way for each node to deliver data to the sink. 20 Figure 2.7 - Packet forwarding with and without clustering and aggregation [37] In a network which uses single hop transmission without clustering like (a) in Figure 2.5 utilizes the nodes which use single hop transmission to the sink after sensing information. This way is easy to implement because there is no special routing technique to transfer data to the sink. However, if the size of network becomes large, it consumes so much energy to transmit a signal to the sink. In Figure 2.5 (b) which uses multi hop transmission without clustering, on the other hand, each sender uses its neighbor nodes as a relay node for packet delivery. Packet is transmitted by multi-hop route and delivered to the sink or target region. This approach is a more efficient way to save the energy 21 consumption and used in [8], [26], and [32] compared to (a) in Figure 2.5 since the energy used for sending activity is only for sending their neighbors. However, traffic concentrates normally on the nodes near by sink [38]. Some applications solve this problem by allocating cluster head among the nodes [39]-[40]. The network which uses single hop transmission with clustering like (c) in Figure 2.5 (c) splits the network into sections so that each node inside each section can communicate each other by one hop. One cluster head is chosen from each section and directly communicates with each node inside the section where a cluster head is selected. A cluster head gathers and fuses information for utilizing energy and bandwidth, and delivers it to the sink directly. This topology is used in [39], [40], and [41]. However, the larger the network size, the greater the consumed energy. Therefore, the network which uses single hop transmission with clustering like (c) in Figure 8 is not suitable for the large size network. The network which uses multi hop transmission with clustering like (d) in Figure 2.5 has a similar approach with the network (c). The difference is that both nodes inside clusters and cluster heads use multi-hop data transmission among nodes or among cluster heads to relay packets. This approach can decrease the required energy to deliver packet from the nodes to the cluster heads. Then, cluster heads use multi-hop data transmissions among them to the sink. It can efficiently expand the size of cluster, and network more efficiently compared to the approach like (a)-(c) because it does not need to use long-range transmission to a sink. This approach is proposed by Younis et al. as Hybrid Energy-Efficient Distributed Clustering (HEED) in [37]. 22 2.2 Other Energy Aware Protocols Heinzelman et al. [39], proposed Low Energy Adaptive Clustering Hierarchy (LEACH), a clustering based protocol that uses randomized rotation of local cluster heads to evenly distribute the energy load among the sensors in the network. LEACH assumes that all nodes can transmit packets with enough power to reach the sink if needed, and that each node has computational power to support different Media Access Control (MAC) Protocol for scheduling. Therefore, it is not applicable to networks deployed in large area [42]. In [40], Power-Efficient Gathering in Sensor Information Systems (PEGASIS) [40] was proposed as an enhancement over the LEACH protocol. PEGASIS is a near optimal chain-based protocol, and in order to extend network lifetime, nodes need only communicate with their closest neighbors, and they take turns in communicating with a sink. When the round of all nodes communicating with the sink ends, a new round starts, and so on. This reduces the power required to transmit data per round as the power draining is spread uniformly over all nodes. PEGASIS has two main objectives. First, it increases the lifetime of each node by using collaborative techniques. Second, it allows only local coordination between nodes that are close together so that the bandwidth consumed in communication is reduced. Unlike LEACH, PEGASIS avoids cluster formation and uses only one node, instead of multiple nodes, in a chain to transmit to the sink. When a sensor node is idle, it is wasteful to keep sensors on and listening the signal from other nodes. Motivated by this observation, Singh et al. [43] proposed an Power-Aware Multi-Access Protocol with Signaling (PAMAS) for ad-hoc radio networks by conserving battery power at nodes by intelligently powering off nodes that are not actively transmitting or receiving packets. 23 On the contrary to the scheduling type MAC protocol used in LEACH, S-MAC [44] proposed by Ye et al. combined benefits of TDMA and contention protocols. Since idle listening consumes significant energy, it periodically turns on/off sensor's radio when sleeping, and differs each node's active state in such a way that each node, except node’s neighbors, has a different timing of active state. It reduces overhearing and message passing. Together with overhearing avoidance and message passing, S-MAC obtains significant energy savings over always-listening MAC protocols. At the same time, a node wakes up for a short period of time at the end of each transmission (adaptive listening). Adaptive listening largely reduces the latency due to periodic sleeping and increases throughput. Figure 2.8 shows timing relationship between a receiver and different senders. CS stands for carrier sense4. Synchronization (SYNC) packet is sent to the other nodes at first to notify the existence of the sender for starting communication. SYNC has two types of information; that is, an address and waiting time of a sender. If the sender does not receive the reply for SYNC message within a defined waiting time, it will randomly choose sleeping time and go to sleep state. For example, sender 1 at first sends SYNC packet and does not receive any reply, thus it will sleep. The sender 2 at first sends request-to-send (RTS) packet to win the medium, and the receiver will reply with a clear-to-send (CTS) packet. If sender 2 receives CTS, it can start sending data. The sender 3 sends SYSC packet since it is new node. While CS, it receive the reply for 4 Carrier Sense describes the fact that a transmitter listens for a carrier wave before trying to send. That is, it tries to detect the presence of an encoded signal from another station before attempting to transmit. If a carrier is sensed, the node waits for the transmission in progress to finish before initiating its own transmission [45] 24 SYNC packet and send RTS packet. After receiving CTS packet from receiver, it can send data to the receiver. Figure 2.8 - S-MAC scheduling [44] Position-based algorithm called SPAN [46] proposed by Cen et al. selects some nodes as coordinators based on their positions. The coordinators form a network backbone used to forward messages. A node should become a coordinator if two neighbors of a non-coordinator node cannot reach each other directly or via one or two coordinators. In [47], Hull et al. investigated the end-to-end performance of various congestion avoidance techniques in a sensor network. They proposed a method called Fusion that combines three congestion control techniques that operate at different layers of the traditional protocol stack. These techniques include a hop-by-hop flow control, a source 25 rate limiting scheme similar to the adaptive rate control mechanism that meters traffic being admitted into the network, and a prioritized MAC layer that gives a backlogged node priority over non-backlogged nodes for access to the shared medium. Rodoplu et al. [48] proposed a general model in which the power consumption between two nodes at distance d is u(d)=dα+c for some constants α and c, and described several properties of power transmission that are used to find neighbors for which a direct transmission is the best choice in terms of power consumption. Banerjee et al. [49] investigated the transmission cost to reliably send packets from source to destination node including re-transmission activity caused by link errors. Kubisch et al. defined a strategy to select a minimum transmission power for each node in a given distributed sensor network in [50]. Network operators define a target number of reachable neighbors for all nodes. A node adjusts its transmit quality according to its current neighbor count. If a node has more neighbors than necessary, it scales down its transmit power to the target range. If too few neighbors exist, the node increases its power. Narayanaswamy et al. [51] presents a power control protocol named Common Power (COMPOW). Their goal is to choose the smallest common power level by each node that preserves connectivity, maximizes traffic carrying capacity, reduces contention in the MAC layer; and it requires low power to route packets. In their approach, several routing daemons run in parallel at each node for each power level. Each routing daemon exchanges control messages with its counterparts at the neighboring nodes to maintain its own routing table. CLUSTERPOW [52] was designed to overcome some of the short comings of COMPOW by accounting for non-uniform distributions of nodes. It introduces a hierarchy, where by closely located nodes are allowed to form a cluster and choose a 26 small common power to interact with each other as in Figure 2.9. In this figure, a sender, S, tries to send a packet to a destination node, D, via inter-cluster relay nodes, N1, N2, and N3. Different clusters communicate among themselves at a different (higher) power levels. Intentionally, most of the intra-cluster communication is done at a lower power level, and the (possibly rare) inter-cluster communication is carried out with a higher power. As before, each node runs multiple daemons, which constantly exchange reachability information with neighbors. This incurs a significant message overhead. Figure 2.9 - Routing by COMPOW in a typical non-homogeneous networks. S is the sender, N1 and N2 is the inter-cluster relay nodes, and the node D is the destination [51] The PARO proposed by Gomez et al. in [53] maximizes the number of nodes used between the source and destination within a signal transmission range between them so as to minimize the energy consumption of the overall network. A scheme proposed by the same authors in [54] shows the impact of an individual variable-range transmission 27 power control on the physical and network connectivity, network capacity, and power saving of wireless multihop networks such as ad hoc and sensor networks. 2.3 Localization Algorithms Typically, the aforementioned routing algorithms in WSNs mentioned above assume that they already have some system to localize the location of nodes. Global Positioning System (GPS) is the most popular way to get location information of nodes; however, because of its size and cost, it is still not easy to deploy for a number of sensors. The localization method proposed in [55] uses an idealized radio model and proposes a simple connectivity-based localization method for such devices in unconstrained outdoor environments. Rao et al. [56] focused mainly on the sensor network that no nodes have location information. Their approach involved assigning virtual co-ordinates to each node so that each node has virtual connection with its neighbor nodes by local connectivity, and then it applies a standard geographic routing over these coordinates. Since nodes always know their neighbors and keep connectivity with them, this technique can be applied in most WSN settings. Their approach shows that greedy routing performs better using virtual coordinates than using true geographic coordinates. The cricket compass system proposed in [57] reports position and orientation indoors for a handheld device and informs an application running on the device of the position and orientation in a local coordinate system established by the fixed set of beacons. Though it is hard to achieve the high accuracy of localization in indoors by interference of radio, it achieved an accuracy with a few degrees of true value indoors. In [58], the angle of arrival technique is applied in the ad-hoc network without reference points that transmit high power signals. 28 The nodes collect the angle information from neighbors and derive coordinates by using angle-based triangulation. 29 Chapter 3 Methodology In this section, a new energy-aware-routing metric called Consumed-Energy-Type-Aware Routing (CETAR) is proposed. Then, the CETAR is employed to Geographic and Energy Aware Routing (GEAR) to demonstrate its adaptability to a general energy-aware-routing algorithm. 3.1 Consumed-Energy-Type-Aware Routing (CETAR) Some WSN applications constantly monitor and report environmental changes in an entire region. For such WSN applications, the frequency of any sensor node being the origin of data sending activity is more or less uniformly distributed throughout the WSN, and survivability of each node is more or less equally important. In many other WSN applications, however, some sensors actively capture and disseminate more information than the rest of the sensors do. An example is that a collection of sample data is sent from a target region whose exact location is unknown at the deployment of sensor nodes as described in Figure 3.1. This figure describes the situation as some phenomena exits in fixed regions inside the sensor field. Since an observer does not know the exact location of phenomenon at the deployment of sensor nodes, he has to deploy sensor nodes to a wider range of field than actual regions of his interest. 30 Figure 3.1 - The situation at the deployment of nodes in WSNs Suppose nodes are uniformly deployed in the sensor field as described in Figure 3.2 and Figure 3.3. Figure 3.2 describes conceivable case in which the intensive traffics are generated from the fixed sets of nodes. In this case, active sensors sense and transmit more information than the rest of sensors do, and the location of data originator is easily predicted since their location remains the same. Traffic in this figure is not uniformly distributed throughout the sensor field, so we define it as a non-uniform traffic. Figure 3.2 - Intensive traffic from the fixed sets of nodes 31 On the other hand, Figure 3.3 describes conceivable case in which the intensive traffics are generated from the semi-fixed sets of nodes. In this case, active sensors also sense and transmit more information than the rest of sensors do; however, the location of data originator is not easily predicted since their location changes frequently time after time as described specific time-span t1, t2, and t3 in the Figure 3.3. As seen in the figure, the nodes inside the phenomenon A happened in the time-span t1 are painted gradated gray. The other phenomena, happening during t2 and t3, are also painted gray and black. Since traffic in this figure is not uniformly distributed throughout the sensor field and the location of intensive traffic changes, we define it as a dynamic non-uniform traffic. Figure 3.3 - Intensive traffic from the semi-fixed sets of nodes As an actual example, let us suppose several braces of birds will migrate in some sensor-deployed field in a forest during habitat monitoring observation. Then WNS will start sensing some information about those birds. In this situation, either the observer 32 originates queries for the sensors which retrieve some data of environmental change or the movement of birds triggers transmission of sensed information to the observer, bursts of sensed information originated from a target area especially around birds happen. After their young bird leave their nest, next bird will start migrating in a different place. We can imagine that the most highly activated nodes are the data originating sensors. When those sensors are used for routing activity of the other sensors, their energy can be depleted rapidly. Replacement of those highly active sensor nodes so as to extend the duration of monitoring activity may not be efficient or even not be practical in certain cases such as a deployment of WSN on a wide and distant location like in a deep forest. For this type of WSN application, special attention needs to be given to minimize the energy consumption of the sensor nodes that actively disseminate the data. Lifetime of active sensor nodes may be extended if the information regarding what type of activities each sensor node consumes energy for, is available to its neighbor nodes so as to accomplish intelligent packet forwarding decisions. CETAR is the focus of this research. CETAR’s routing decision is based on the frequency of the different activities (i.e., transmission, reception, sensing, and processing) in which each node engages. The investigation of CETAR comprised of data sending activity and data routing activity at each node because their energy consumptions are dominated by transmission and reception operations which in turn dominate the total energy consumed for node activities [14]. We define the data sending activity at a node as it transmits data as a source node to its direct neighbor node. The data routing activity at a node is defined as it receives and transmits data as an intermediate router. In the sensor network activities, the importance of every node's role is sometimes not equal. There are 33 possibilities that only particular nodes could continue to originate collected data as shown in Figure 3.4. In that case, keeping such sensor nodes active as much as possible is the rational way to prolong the lifetime of sensor network rather than keeping them engaged in both data sending and routing activities. CETAR is particularly suited for the applications with this type of characteristics. The CETAR is a general solution to preserve historically active source nodes and has a potential to improve existing energy-aware routing algorithms in general to prolong the lifetime of the WSNs. Figure 3.4 - An example of CETAR 3.2 CETAR with Biased Consumed Energy (BCE) CETAR employs simple statistics for separate types of energy consuming activities. For instance, each node keeps statistics of the energy consumed for data transmissions as a source node and data transmission and reception as an intermediate 34 router. Since the transmission and reception operations dominate energy consumption of sensor nodes in WSNs, such statistics are useful for identifying which nodes are primarily active as a source node and which nodes are primarily active as a routing node. In particular, the statistics can be used to establish routing paths so that CETAR selects a sensor node with high residual energy as well as one that which has rarely consumed energy as a source node. We can define a Biased Consumed Energy (BCE) at node i, Ni as, BCE ( N i ) (CEs ( N i )) (1 )(CEr ( N i )) (1) where CEs(Ni) and CEr(Ni) are consumed energy of Ni used for data sending and that used for routing activities, respectively. These statistics can be easily recorded by each node. β is a tunable weight from 0 to 1. If β is 0.5, BCE(Ni) becomes equivalent to the total consumed energy of Ni without bias. β value is shared in the whole network and each node calculates BCE based on β. A simple example in Figure 3.5 illustrates how CETAR can preserve energy of the node which frequently plays a source node for potentially improving the lifetime of WSNs. Suppose Ns tries to send a packet to Nd via one of the neighbor nodes, Nx and Ny. Assuming the transmission cost among the direct neighbor nodes are the same, the least cost path can be derived based on the total energy consumption at each node i, CEi. Since CEx=0.55 and CEy=0.6, the least cost path from Ns to Nd is Ns-Nx-Nd. Thus, the node which frequently plays a source node, Nx, will consume its energy which could be used in future transmission of data originated from its sensor. When equation (1) with β=0.9, cost will be computed as BCE(Nx)=0.9(0.5)+0.1(0.05)≈0.46 35 and BCE(Ny)=0.9(0.2)+0.1(0.4)≈0.21 respectively. Consequently, the least cost path from Ns to Nd is Ns-Ny-Nd. That is, Ni will choose Ny as the next hop so as to preserve the energy of the active node, Nx. Figure 3.5 - An example of CETAR with BCE 3.3 CETAR with Adaptively-and-Aggressively Biased Consumed Energy (AABCE) Use of fixed β could abuse energy resources in the WSN since not enough statistics are available to predict which nodes can be identified as sending nodes immediately after the deployment of the network. For example, one node can be initially performing sending activities only for a very small period of time. Even if the node has no more future data to send it will be overprotected compared to its neighbor nodes that could potentially become active at a later time. In this situation, BCE in equation (1) with relatively high β is likely to increase the total energy consumption in the network without 36 a benefit. Based on this observation, instead of using predefined weight β, adaptively BCE using β as an adaptive weight that can be adjusted dynamically based on the statistics of consumed energy for sending, CEs(Ni) , and routing, CEr(Ni), in equation (2). This statistical information of energy is used to calculate β. By using those values, β is calculated as follows. CEs ( N i ) CEs ( N i ) CEr ( N i ) (2) where CEs(Ni)+ CEr(Ni) ≤ 1.0. When CEs(Ni) and CEr(Ni) consume none of their energy, value of β is set 0 as a default value. In the equation (1), either side of the co-efficient, β or 1-β, will be linearly increased/decreased if the other side of the co-efficient is linearly decreased/increased. That means, the rate of change in bias for data sending activity part (CEs ( Ni )) and that for routing activity part (1 )(CEr ( Ni )) are moderate throughout an entire range of β. Since we prefer to aggressively and rationally preserve active nodes which are originating packets, we define aggressively-and-adaptively biased consumed energy AABCE for data origination or routing at node i, Ni calculated as follows. AABCE ( N i ) sqrt ( )(CEs ( N i )) (1 sqrt ( ))(CEr ( N i )) (3) where β is the tunable weight from 0 to 1. Linear and square root value for β and 1-β are described in Figure 3.6. As seen in Figure 3.6, β changes linearly from 0 to 1. On the other hand, square root β value sharply increases when β is close to 0 while its rate of increase becomes moderate as β approaches 1. 37 Figure 3.6 - Graph of β and actually used square root β adapted by CEs(Ni) and CEr(Ni) BCE sometimes excessively preserves nodes which perform data originating activity compared to AABCE. AABCE can prevent this situation. For example, suppose the case that Nx already consumed 10 percent of the energy for sending and 89 percent of the energy for routing, and Ny consumed 20 percent of the energy for sending and 10 percent of the energy for routing so far in the Figure 3.5. In this case, CEs(Nx)=0.1, CEr(Nx)=0.89, CEs(Ny)=0.2, and CEr(Ny)=0.2. It calculates BCE(Nx) = 0.9(0.1) + 0.1(0.89)≈0.18 and BCE(Ny) = 0.9(0.2) + 0.1(0.2)=0.21 where β is 0.9. In this example, even if the energy of node Nx is substantially depleted, CETAR with BCE will choose the least cost path from Ns to Nd as Ns-Nx-Nd. CETAR 38 with BCE underestimates energy used for routing activity lets the node’s battery depleted quickly, and it shortens the lifetime of the network. On the other hand, when AABCE is used in the same scenario, β for Nx is 0.1/(0.1+0.89)=0.1 and β for Ny is 0.2/(0.2+0.2) = 0.5 and, AABCE(Nx) = sqrt(0.1)(0.1)+(1-sqrt(0.1))(0.89)≈0.64 and AABCE(Ny) = sqrt(0.5)(0.2)+(1-sqrt(0.5))(0.2)≈0.23 In this case, CETAR with AABCE will choose the least cost path from Ns to Nd as Ns-Ny-Nd. Thus, CETAR with AABCE successfully avoids aforementioned situation by adaptively using proper weight for each weight. In the section 3.5, GEAR is used as an example to show how CETAR can be incorporated into existing energy-aware routing protocols in general to gain the lifetime of the WSNs. 3.4 Effect of Weighting Function on the BCE and AABCE Figure 3.6 shows the relationship between β and square root β on how the value of squat root β changes with increase in β. Equation (3) used square root β as a coefficient of each type of consumed energy. As seen in Figure 3.6, β changes linearly from 0 to 1. On the other hand, square root β value sharply increases when β is close to 0 while its rate of increase becomes moderate as β approaches 1. This type of function is called the utility function5 for β. To use this function as a coefficient of equation (3), we need to set the minimum/maximum value of utility function in the range of β as 0/1. Many utility 5 A mathematical expression that assigns a value to all possible choices 39 functions other than square root β can be used. Utility functions which are the arctan( ) and 2 1 (1 ) 2 where 0 ≤ β ≤ 1 is described in the Figure 3.7. In this figure, as β increases from 0 to 1, arctan( ) and 2 1 (1 ) 2 sharply increases as square root β does. However, square root β sharply increases even when β is low enough. This characteristic will contribute for putting more weight for data originating activity from the early stage of WSN activity. Effect of utility function for AABCE compared to linearly increased β is evaluated in Chapter 4. We will consider which value of β in equation (1) results in extending the lifetime of active sensor nodes for CETAR with BCE in the next chapter. The effect of weighting function for using a coefficient in equation (3) for CETAR with AABCE will also be considered. Figure 3.7 - Utility Functions changed with increase in β 40 3.5 CETAR for GEAR GEAR considers both the residual energy and the distance to the destination when selecting a routing node. The idea of CETAR is incorporated into GEAR to evaluate a relative performance improvement of CETAR. GEAR uses the following assumptions [8]. 1) It is assumed that locations of node in this simulator are determined by using one of those existing localization algorithms as described in the section 2.3. Knowledge of node location is necessary for the routing decision of GEAR. 2) There exists some periodic packet exchange protocol done in the underlying layer to exchange the message containing information: location of a node and its remaining energy. This information is shared among nodes and its neighbors. It might be possible to be done in MAC layer, for example, piggybacking the neighbor information in the MAC layer control packets, such as RTS, CTS, and ACK. 3) Each radio-link is bi-directional. This is not an unreasonable choice as unidirectional links can be selectively black-listed at the MAC layer [59]-[60]. 4) Sensor node consumes unnegligible amount of energy even when it is in a listening or idle state as shown in Figure 1.4. Thus, TDMA-style MAC protocol proposed in [61] or SMAC [44] that enables a sensor node switch to sleep mode can be used when it is not transmitting or waiting packet. GEAR controls the number of disseminated packets by only considering a certain target region instead of flooding entire networks with the packets by using 41 energy-aware and geographically-informed neighbor-selection heuristics to route a packet toward the destination region. Specifically, greedy forwarding as described in the Figure 3.8 and energy balancing technique as described in the Figure 3.9 is used. Figure 3.9 explains the energy aware portion of GEAR. In this figure, there is source node, S, and destination node, T, and the shortest path is S-A-B-C-D-T. If the shortest path is used every time, the nodes A, B, C, D on the path S-A-B-C-D-T will be depleted quickly since intermediate routing nodes consume energy for routing including data transmission and reception in contrast to S and T, which concentrate on data transmission/reception. If S and T on the path S-A-B-C-D-T are also depleted, there is no way to send packets from P to Q, called a network partition. Therefore, instead of using the path S-A-B-C-D-T, the alternate paths, in which the load balancing and the shortest path routing can spread the load of energy consumption, is shown in Figure 3.9. This not only for extends the lifetime of node S and T longer but also delays intermediate nodes from quick depletion of their energy. R C N = Node = Path = Transmission range Figure 3.8 - Greedy packet forwarding 42 Considering a node N trying to forward a packet whose destination is centroid C in target region R, the node N routes the packet progressively toward the target region as in Figure 3.8. At the same time, it tries to balance the energy consumption across all its neighbors. The next hop determined by the smallest learned cost across all neighbors is: h( N , R) c( N , N min ) h( N min , R) (4) Learned cost is the combination of distance from sender to its neighbor node Ni, residual energy of node N, and the learned cost of its neighbor Ni to the target region R, h(Ni,R). If a node does not have h(Ni,R) for a neighbor Ni, it computes the estimated cost c(Ni, R) of Ni as a default value for h(Ni, R) as follows: αd(Ni, R) + (1-α)e(Ni) (5) where d(Ni,R) is the normalized distance from Ni to the centroid C of the region R and expressed as d ( N i , R) Distance ( N i , R ) Max N j Nei ( N j ) ( Distance ( N j , R )) (6), Figure 3.9 - Load balancing and the shortest path routing Error! Reference source not found. 43 and e(Ni) is the normalized consumed energy at node Ni, and expressed as e( N i ) CE( N i ) Max N j Nei ( N j ) (CE( N j )) (7), where CE(Ni) is the consumed energy at Ni, Nei(Nj) is a set of neighbors of Nj, and α is a tunable weight from 0 to 1. If α is 1, learned cost is purely determined by the distance from the neighbors to the target region R. Since h(Ni,R) cannot be calculated for non-adjacent neighbors Ni and R at the beginning, the estimated cost c(Ni,R) as a initial learned cost of neighbors is used instead. From the result of learned cost calculation, the next hop satisfying high residual energy which is also closer to the destination will be selected. If all neighbors are further than N to the destination C, this means N is in a routing hole. That is, no nearer neighbors exist between N and C. In this case, N first updates its learned cost by combining its neighbor’s learned cost and the cost of forwarding a packet to the neighbor. Consequently, N will send the updated learned cost to its neighbor to prevent subsequent packets from falling into the hole. GEAR calculates a learned cost with fixed weighting factor, α, for the distance between two nodes and a residual energy of node. We adopt CETAR scheme to GEAR so that the residual energy component can be further categorized to different types of energy consumption. Thus, when the consumed energy of an active source node is still high, the node will receive greater weight so as to prevent it from being selected as a part of the routing path. In particular, energy portion of the equation (5) is modified to e( Ni ) BCE ( Ni ) MaxN j Nei ( N ) ( BCE ( N j )) (8) 44 A simple example in Figure 3.10 illustrates how aggressive preservation of active source nodes can potentially improve the lifetime of WSNs. (a) (b) Figure 3.10 - An Example of GEAR incorporating CETAR Suppose Ni is trying to send a packet to R via one of the neighbor nodes, Nx and Ny as an intermediate routing node. Based on GEAR, h(Nx,R)=c(Nx,Nz)+h(Nz,R) and h(Ny,R)=c(Ny,Nz)+h(Nz,R). For α=0.5, CEs(Nx)=0.9, CEr(Nx)=0.01, CEs(Ny)=0.1, and CEr(Ny)=0.2 are used in Figure 3.10 (a), the GEAR will compute h(Nx,R)=0.5(1/3)+0.5(0.91/0.91)+10≈10.66 and h(Ny,R)=0.5(3/3)+0.5(0.3/0.6)+10=10.75, respectively. Thus, the highly active source node, Nx, will continue to be a part of routing path between Ns and R. On the other hand, the GEAR with CETAR with β=0.9 will compute 45 h(Nx,R) = 0.5(1/3)+0.5((0.9(0.9)+0.1(0.01))/0.811)+10≈10.66 and h(Ny,R) = 0.5(3/3)+0.5((0.9(0.1)+0.1(0.2))/0.38)+10≈10.64. Consequently, node Ni will choose Ny as the next hop BCE sometimes excessively preserves nodes which perform data originating activity. As an example, we show the particular case that AABCE is effective compared to BCE. We assume the same situation with the section 3.3 such as α=0.5, β=0.9 for BCE, and CEs(Nx)=0.1, CEr(Nx)=0.89, CEs(Ny)=0.2, and CEr(Ny)=0.2 in the Figure 3.10 (b). We can know BCE(Nx)≈0.18, BCE(Ny)=0.21, AABCE(Nx)≈0.64, and AABCE(Ny) ≈0.23 from the section 3.3. By using aforementioned BCE and AABCE, h(Nx,R) for GEAR using CETAR with BCE is calculated as, h(Nx,R)=0.5(2/3)+0.5(0.18/0.38)+10≈10.57 and h(Ny,R)=0.5(3/3)+0.5(0.21/0.38)+10≈10.77 where β is 0.9. Similarly, learned cost for GEAR using CETAR with AABCE is calculated as, h(Nx,R)=0.5(2/3)+0.5(0.64/0.64)+10≈10.83 and h(Ny,R)=0.5(3/3)+0.5(0.23/0.36)+10≈10.81 While CETAR with BCE could not avoid using energy depleted node Nx as the next hop, CETAR with AABCE successfully choose Ny as the next hop. Those examples demonstrate that CETAR with AABCE can adaptively and rationally prevents active source nodes from being selected as a part of routing path in 46 GEAR. Thus, the lifetime of the heavily-used individual sensor nodes can be extended, and that of entire WSNs can be extended accordingly. In the next chapter, CETAR is incorporated into GEAR simulator and evaluated the performance in terms of the lifetime of network. 3.6 The Application of CETAR for the Other WSN Topology In this section, the possibility of CETAR for the other type of WSN topology is discussed. The type of network protocol is classified according to their network structure such as flat, hierarchical, or location-based [42]. In the flat network, each node has the same role, on the contrary to the hierarchical network which has the cluster head and normal node, and they serve different roles each other. Location-based network utilize the location information of nodes acquired from a localization technique and use them in the routing decision rather than flooding packets in the whole network. GEAR [8] proposed in the previous section has the two attributes that are flat and location-based. Thus we will seek the applicability of CETAR for the hierarchical network. Hierarchical network is mainly classified into two types according to the type of cluster’s role of whether they use the single hop routing or multi-hop routing for the data delivery to the observer (sink) as described in the Figure 2.5 (c) and (d). We focus on the Figure 2.5 (d) since CETAR needs multi-hop routing to save the energy of active sender nodes. This idea cannot be applicable for the single hop routing as described in the Figure 2.5 (c). Figure 3.11 shows that three clusters are sending their packet to the sink by using inter-cluster head multi-hop routing. Also nodes in each cluster send packet to their 47 cluster head by using multi-hop data transmission. There are several assumptions for this network as following: 1) Each node knows its absolute locations by some localization algorithm. 2) Network is separated into segments and nodes inside each segment will form a cluster. Each cluster has one cluster head. 3) Nodes inside a cluster can only communicate with nodes inside the same cluster or their cluster head by using multi-hop routing. Clusters heads can communicate each other by using multi-hop routing. 4) Even if the cluster heads have a plenty of battery, their energy is not unlimited. Also, the replenishment of battery is costly since nodes in WSNs are deployed in distant or dangerous area where human can not easily enter. 5) All underlying layer used below or above network layer is abstracted. Figure 3.11 – Routing on the hierarchical network structure 48 In this figure, we assumes that energy used inside each cluster is already optimized by assuming existing multi-hop routing protocols like GEAR. In this case, each node in a cluster will deliver their packet for their cluster head at first. After cluster heads receives packets from their cluster, they deliver packets to the observer by using multi-hop routing. Therefore, we can substitute routing in hierarchical network in Figure 3.11 as the overlay network among cluster heads. In this case, CETAR can be incorporated to a routing algorithm used among the clusterheads. Consequently, the same logic of applying CETAR for the routing scheme as described in the section 3.5 can be used in this example. Locally-improved routing decisions among nodes inside the clusters and cluster heads can create globally-improved routing decisions for the entire network. 49 Chapter 4 Evaluation of CETAR 4.1 Simulation Models 4.1.1 Assumptions for the simulation Performance of power management in WSNs could be measured by total energy consumed in a system or the number of packets being sent and/or received before the network partitions. Latter metric is chosen over the former one in order to measure the duration of operable time for the uniform and non-uniform traffic described in the section 4.1.3. Many of routing algorithms proposed in recent literatures use dynamic adaptive transmission power (DATP) described in Figure 4.1. Thus, GEAR is implemented with DATP to save unnecessary transmission cost even though the GEAR did not provide it in [8]. It is well known that the transmission power of Pt=PrDα is required to transmit signal to the receiver where Pr is the receiving power and D is the distance from sender to receiver. Pr needs to be higher than threshold Pthr. If the received signal is lower than Pthr, receiver can not receive the signal properly. Since the transmission power of GEAR is fixed, we can know at most Dα-Diα transmission power will be wasted at most from the Figure 4.1. The value of α depends on the transmission media and antenna characteristics. This value is typically around 2 for the short distance links which is less than 100 meters and omni-directional antennae, and around 4 for longer distance in the 2.4 GHz transmission band [49]. For the sake of research considered in this manuscript, α=2 is 50 assumed because wireless sensor nodes do not typically use in 2.4 GHz frequency band. A margin of Dj (0 ≤ Dj ≤ D−Di), in addition to distance Di, can be applied to set a transmission power level to (Di+Dj)α in order to reduce error rate in data link layer. Same DATP is used for CETAR. We assume link error-free environment since the relationship between signal attenuation and error rate is complex. Thus, all signals are assured 100 % to be delivered to the next hop. Figure 4.1 - Relation of transmission range between fixed and adjustable power 51 4.1.2 Common Experimental Settings The number of nodes in the network ranges from 400 to 4800 nodes while its density is kept constant. For instance, the geometric area of a 600 node network is 1200 units by 1200 units square. This means one node exists per 2400 units2 of area. This node density will be used for all evaluations of network configurations. To keep the density of the number of nodes constant, the size of the sensor network area is doubled when the number of nodes is doubled. After the number of nodes to be deployed in the sensor field of simulator was entered, the length of the edge of the geometric area in the simulator is automatically calculated as 20 6 p where p is the number of nodes in the network. If p is 600, for instance, the area will be 1200 unit square. The energy level of each node is initialized to 1 joule. For the GEAR with a fixed transmission power, 0.001 joule is consumed for either transmission or reception of a packet. For the GEAR/CETAR with DATP, the energy required for transmitting a packet can range from 0 to 0.001 joule while the receiving energy is always fixed as 0.001 joule. Node’s transmission range is fixed at a 100 unit distance across all simulations for the GEAR. For GEAR/CETAR with DATP, transmission range was changed based on the distance to the next node located within the maximum transmission range D. In this experiment, an interference-free environment is assumed. That is, only the transmission power to send for distance Di instead of Di+Dj is required. Value of α=0.5 is used in equation (5) as used in [8]. Link error rate was ignored in this simulation experiment since the experiments in [8] did not use link error metrics. Our intention was to measure the optimal performance gain of the proposed schemes over GEAR. 52 4.1.3 Type of Traffics for CETAR The following types of traffic to generate packets are used in our simulator. 1) Uniform traffic Pairs of source nodes and target regions are uniformly distributed throughout the entire network. Ten source and target region pairs are randomly selected and paired with each other. This experiment measures the performance of the network with applications requiring relatively uniformly distributed communication patterns. Figure 4.2 - 10 given source and target region pairs 2) Non-uniform traffic (a cluster of 10 moderately close nodes) Source nodes are clustered so as to concentrate part of the traffic. An initial source node is selected randomly out of all nodes in the network. Then the rest of source nodes are randomly selected out of 29 nodes which are the closest to the initially selected source node. Then, 10 target regions are randomly selected and paired with the source nodes. This experiment measures the 53 performance of the network where active source nodes are moderately close to each other. Such conditions exit in many circumstances in the real WSNs. Figure 4.3 - 10 sources out of 30 source candidates 3) Non-uniform traffic a cluster of 10 closest nodes) Source nodes are clustered so as to concentrate part of the traffic. An initial source node is selected randomly out of all nodes in the network. Then a set of nodes which are the closest to the initial source node is selected to form a cluster of 10 source nodes. Ten target regions are randomly selected and paired with the source nodes. This experiment measures the performance of the network where active source nodes are adjacent to each other. 54 Figure 4.4 - 10 sources out of 10 source candidates In the following sections, we conduct the high level simulations of GEAR, GEAR with DATP, CETAR, and CETAR with DATP. We are only interested in how packet can be routed to a target region, and the packet dissemination portion of GEAR is not considered in our experiment. Experiments are conducted to measure the number of packets successfully delivered to the target regions before the network partitioning in both uniform and non-uniform traffic environments. The network is partitioned if all the given sources are partitioned from their respective destinations as described in the Figure 4.5. This figure shows that the source node has no way to deliver its packet to the target region because of the energy-depleted nodes (shown in gray). 55 Figure 4.5 - The network partitioning 4.2 Evaluation for CETAR with BCE 4.2.1 Experimental Setting for BCE In this experiment, BCE of equation (1) described in the section 3.2 is incorporated into the residual energy information of GEAR. The value of β is set to 0.9 in section 4.2.2 intuitively since this value is good enough to avoid the data sending node from routing activities. For example, when β is set to 0.9, the weight of data sending activities is nine times larger than that of routing activities. The effect of this value is further discussed in the section 4.2.3. Experimental assumptions and settings are the same with the section 4.1.1., and traffic generators used in this experiment are the same with 4.1.2. We conducted the simulation experiment for 50 times on the network consisting of 400, 600, and 800 nodes; 30 times on the network consisting of 1200 and 2400 nodes; and 20 times on the network consisting of 4800 nodes. Because of time constraint, the number of measurements conducted in the evaluations for each node density is different. 56 For example, to conduct one simulation on 4800-node network, it takes 20 minutes on the author’s machine (Pentium M 1.6 GHz, 1GB RAM). The experimental result which has outside the range of 95% confidence interval from the average of experimental result is eliminated. 4.2.2 Evaluation for BCE Figure 4.7 shows the experimental result of the number of packets delivered successfully before the network partitioning for the uniform traffic experiment. This figure shows the relation between the node density and the total number of packets sent from 10 randomly chosen sources to the target regions. Theoretically, the maximum number of packet transmissions for the GEAR without transmission power control will be 10000 since each node has 1 joule (10000-unit) energy and they consume one unit of energy per transmission or reception of a packet. Therefore, the results of GEAR and CETAR with BCE have almost the similar results and do not exceed 10000 packets delivery. GEAR with DATP can send 48.2% more packets on average than the GEAR without transmission power control. The result indicates that significant energy saving occurs at each node by dynamically adjusting transmission power. Furthermore, the CETAR with BCE and DATP can send 70.5% more packets on average than the GEAR. 57 Figure 4.6 Figure 4.7 - CETAR with BCE (uniform traffic) Figure 4.8 shows the results of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using a cluster of 10 out of 30 closest senders. In this experiment, CETAR with BCE can send 10.9% more packets than GEAR. GEAR with DATP and CETAR with BCE and DATP can send on average of 45.6% and 62.3% more packets, respectively, than GEAR. 58 Figure 4.8 - CETAR with BCE (non-uniform traffic: A cluster of 10 out of 30 closest senders) Figure 4.9 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using a cluster of 10 closest senders. In this experiment, CETAR with BCE can send 12.1% over GEAR. GEAR with DATP and CETAR with BCE and DATP can send on average of 43% and 61.6% more packets, respectively, than GEAR. 59 FigureFigure 4.9 - CETAR with BCE (non-uniform traffic: A cluster of 10 closest senders) Our experiments demonstrate that aggressively preserving active source nodes is effective for extending the lifetime of WSNs regardless of network size. The result of the experiment evaluating the performance of CETAR with BCE is summarized in the Table 1. Table 1 - The average percentage of packet sent over GEAR for CETAR with BCE Uniform Nonuniform (10/30) Nonuniform GEAR CETAR with BCE GEAR with DATP 0% (9579) 0% (5121) 0% 0% (9529) 10.9 % (5681) 12.1 % 48 % (14206) 45.6 % (7458) 43 % CETAR with BCE and DATP 70 % (16338) 68.9 % (8652) 61.6 % 60 (10/10) (4815) (5399) (6890) (7785) 4.2.3 Effectiveness of β for BCE In this section, we investigate which value is most effective for CETAR with BCE. Value of β (0.6, 0.7, 0.8, 0.9, and 0.95) is evaluated. If β become less than 0.5, the weight of routing activity becomes higher than that of sending activity. We do not use β which is less than 0.6 because our objective is to investigate the effect of biased-consumed-energy type in data originating activities. This experiment is similar to the experiments conducted in the section 4.2.1. The only difference is that we repeat different value of β for each experiment. Figure 4.10 shows the relation between the node density and the total number of packet to be sent from 10 randomly chosen sources and target region pairs by using BCE with several types of β. The simulation result of GEAR and GEAR with DATP is the same with the section 4.2.1 in the Figure 4.10. Results of CETAR with BCE and DATP for all β experimented exceed the result of GEAR and GEAR with DATP. CETAR with BCE and DATP when β=0.6 and 0.95 have the worst result among measured β. CETAR with BCE and DATP when β=0.7 and 0.9 have the second or nearly best results, and the CETAR with BCE and DATP when β=0.8 has the best result. 61 Figure 4.10 - CETAR with BCE for different β (uniform traffic) Figure 4.11 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using BCE with a cluster of 10 out of 30 closest senders. The results show the same trend as those of the previous experiment as shown in Figure 4.10 Results of CETAR with BCE and DATP for all β experimented exceed the result of GEAR/GEAR with DATP. CETAR with BCE when β=0.6 and 0.95 have the worst result among measured β. CETAR with BCE and DATP when β=0.7 and 0.9 have the second or nearly best results ,and the CETAR with BCE and DATP when β=0.8 has the best result. 62 Figure 4.11 - CETAR with BCE for different β (non-uniform traffic: A cluster of 10 out of 30 closest senders) Figure 4.12 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using BCE with a cluster of 10 closest senders. Most of tendencies are the same with previous experiment. However, β=0.9 instead of β=0.8 has the best results. This result indicates that β=0.8 may not be large enough for a sender node in a tight cluster to avoid participation to routing activity since one or more sender nodes are adjacent to the sender. 63 Figure 4.12 - CETAR with BCE for different β (non-uniform traffic: A cluster of 10 closest senders) From the above experiments, we successfully investigated how the value of β effects the simulation of CETAR with BCE. The result is sensitive based on the used traffic model. So far we mentioned CETAR with BCE. We are going to evaluate CETAR with AABCE in the next section. 4.3 Evaluation for CETAR with AABCE 4.3.1 Experimental Setting for AABCE In this section, we evaluate CETAR with AABCE. BCE used in the previous section is replaced with AABCE of equation (3) in the section 3.2. Experimental 64 assumptions and setting is the same with the section 4.1.1, and the traffic generators used in this experiment are also the same with 4.1.2 except for the section 4.3.4. 4.3.2 Evaluation for AABCE Figure 4.13 shows the relation between the node density and the total number of packet sent from 10 randomly chosen sources to the target region pairs by using AABCE. Theoretically, the maximum number of packet transmissions for the GEAR without transmission power control will be 10000 as described in 4.2.2. Thus, GEAR and CETAR with AABCE has almost the same result. However, GEAR with DATP and CETAR with AABCE and DATP made clear difference. GEAR with DATP can send over 48% more packets on average than the GEAR without transmission power control throughout all sizes of networks measured. Furthermore, the CETAR with AABCE and DATP can send over 80% more packets on average than the GEAR throughout all sizes of networks measured. 65 Figure 4.13 - CETAR with AABCE (uniform traffic) Figure 4.14 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using AABCE with a cluster of 10 out of 30 closest senders. In this experiment, CETAR with AABCE can send 22.1% over GEAR. On the other hand, GEAR with DATP can send 45% more packets on average than the GEAR without transmission power control throughout all sizes of networks measured. Furthermore, the CETAR with AABCE and DATP can send over 83.2% more packets on average than the GEAR throughout all size of networks measured. 66 Figure 4.14 - CETAR with AABCE (non-uniform traffic: A cluster of 10 out of 30 closest senders) Figure 4.15 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using AABCE with a cluster of 10 closest senders. In this experiment, CETAR with AABCE can send 23.3% over GEAR. On the other hand, GEAR with DATP can send 43% more packets on average than the GEAR without transmission power control throughout all sizes of networks measured. Furthermore, the CETAR with AABCE and DATP can send over 79.4% more packets on average than the GEAR throughout all size of networks measured. 67 Figure 4.15 - CETAR with AABCE (non-uniform traffic: A cluster of 10 closest senders) As a conclusion, all experiment about AABCE is improved compared to BCE as described in the Table 2 and Table 3. For uniform traffic, CETAR with AABCE and DATP can send 5.8% more packet than CETAR with BCE and DATP on average. For the non-uniform traffic with 10 sources out of 30 sender candidate, CETAR with AABCE can send 10% more packet than CETAR with BCE on average and CETAR with AABCE and DATP can send 9.2% more packet than CETAR with BCE and DATP on average. For the non-uniform traffic with a 10 closest nodes, CETAR with AABCE and DAPT can send 9.9% more packet than CETAR with BCE and DATP on average and CETAR with AABCE and DAPT can send 11.1% more packet than CETAR with BCE and DATP on average. At this time, we used equation (3). β in equation (3) is weighted by the square root. For further investigation, we evaluate the effect of weighting functions as a 68 coefficient of β in the section 4.3.3 and evaluate the adaptability of AABCE over BCE by using different types of traffic generator named two-stage source node selection in the section 4.3.4. Table 2 - The percentage of packet sent over GEAR for CETAR with AABCE Uniform Nonuniform (10/30) Nonuniform (10/10) GEAR CETAR with AABCE GEAR with DATP 0% (9579) 0% (5121) 0% (4815) 0% (9613) 22.1 % (6257) 23.3 % (5939) 48 % (14206) 45.6 % (7458) 43 % (6890) CETAR with AABCE and DATP 70 % (17300) 83.2 % (9386) 79.4 % (8625) Table 3 - Improvement of CETAR with AABCE over CETAR with BCE (1) Uniform Nonuniform (10/30) Nonuniform (10/10) CETAR with AABCE over CETAR with BCE 0% CETAR with AABCE and DATP over CETAR with BCE and DATP 5.8 % 10 % 8.3 % 9.9 % 11.1 % 4.3.3 Effectivity of Weighting Functions for AABCE In this section, the effectiveness of utility functions as a coefficient of β in equation (3) is evaluated. In equation (3), square root is incorporated with the weighting factor β. Additional utility functions are employed for CETAR with equation (3) and compared to each other. Figure 4.16 shows the relation between the node density and the total number of packet sent from 10 randomly chosen sources to the target region pairs by using AABCE with eight kinds of weighting functions. In this figure, AABCE with performs the 69 best in many cases. Occasionally, AABCE with AABCE with 1 (1 ) 2 performs better than AABCE without weighting function performs the worst. AABCE with arctan( ) performs the second worst. 2 Figure 4.16 - CETAR with AABCE with utility functions (uniform traffic) Figure 4.17 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using AABCE with a cluster of 10 out of 30 closest senders for eight weighting functions. The trend of the results are identical to that of previous experiment shown in the Figure 4.16. 70 Figure 4.17 - CETAR with AABCE with utility functions (non-uniform traffic: A cluster of 10 out of 30 closest senders) Figure 4.18 shows the result of the number of packets successfully delivered before network partitioning for the non-uniform traffic experiment using CETAR with AABCE with a cluster of 10 closest senders for eight weighting functions. In this experiment, AABCE with reason why AABCE with 1 (1 ) 2 performed the best. It is conceivable that the 1 (1 ) 2 performed the best is due to its network configuration. In this experiment, source nodes are formed as a cluster as each node adjoins each other. Therefore it is possible that sources cannot avoid using their neighbor node from routing activities frequently. However, since AABCE with 1 (1 ) 2 highly preserve source node from using routing even when β is small as shown in Figure 71 4.18, it prevents adjacent source nodes from frequently using routing activities from the initial stage. Figure 4.18 - CETAR with AABCE with utility functions (non-uniform traffic: A cluster of 10 closest senders) As a conclusion, AABCE with performs the best. AABCE with arctan( ) which has less steep weighting function moderately increased their 2 experimental result compared with AABCE without square root. On the other hand, the result of AABCE with 1 (1 ) 2 which has steep weight curve got better than AABCE with when highly non-uniform (10 sources out of 10 source candidates) traffic is used. From the following result, we can conclude that AABCE with works 72 the best. On the other hand, too steep or moderate (linear) weighting function works worse except for the particular situation. 4.3.4 Effectiveness of AABCE over BCE in Dynamically Non-Uniform Traffic In this section, we verify if the AABCE is effective or not over BCE for dynamically non-uniform traffic described in section 3.1. Two cases whose lifetime of sending node would be susceptible to the effect of the value β are created. 1) Case 1 Nodes and target regions are uniformly distributed throughout the entire network. From those nodes, (the number of nodes in the system)/10 of nodes are chosen as the sender candidates. For each packet transmission in the simulation, one sender node is chosen from the source node candidates and paired with one target regions. After all sender candidates consumed 20% of their energy for sending activities, fixed set of 10 senders is chosen (those senders can send their packet for any target regions) and the simulation is continued. 2) Case 2 A cluster of sender candidates which consists of 30 adjacent nodes is formed at first and 10 sender nodes are chosen from the cluster. Those 10 sender nodes are matched with any target regions for every packet transmissions. After all senders consumed 20% of their energy for sending, we select a disjoint set of 10 sender node from the same cluster of 30 adjacent nodes. and simulation is continued. Even 20% of energy is consumed for sending, the value of 0.2 is relatively high (=0.447). By using this biased β, original senders are prevented from routing activities well. For the CETAR with 73 BCE in this experiment, we used β=0.9 for the consistency to compare the experimental result acquired in section 4.2.2. Figure 4.19 shows the result for case 1. CETAR with AABCE performed better than CETAR with BCE and CETAR with AABCE and DATP performed better than CETAR with BCE and DATP. The numerical result of comparison among those experiments is shown in the Table 4 as “Uniform”. CETAR with AABCE can send 16% more packet than BCE. Moreover, AABCE with DATP can send 6.9% more packet than BCE with DATP compared to the 5.8% improvement of section 4.3.2. Figure 4.19 - CETAR with AABCE for 2 stage sender selection (uniform traffic) Figure 4.20 shows the result of case 2. CETAR with AABCE performed better than CETAR with BCE, and CETAR with AABCE and DATP performed better than CETAR with BCE and DATP. The numerical result is shown in the Table 4 as 74 “Nonuniform (10/30)”. CETAR with AABCE can send 23.5% more packet than CETAR with BCE compared to the 10% improvement of section 4.3.2. Moreover, CETAR with AABCE and DATP can send 21.1% more packet than CETAR with BCE and DATP compared to the 8.7% improvement of section 4.3.2. Figure 4.20 - CETAR with AABCE 2 stage sender selection (non-uniform traffic: A cluster of 10 out of 30 closest senders) The improvement of result from CETAR with BCE to CETAR with AABCE is slightly better than the section 4.3.2. for the uniform traffic as described in the Table 5 because the traffic is well distributed and sender nodes are easily avoided from routing activities even when there are many highly weighted senders in the system. On the other hand, non-uniform traffic of CETAR with AABCE improved significantly over CETAR 75 with BCE that is 10% or more. This is because, the clustered senders have more chance to directly avoid using active data originating node from routing activities because they are adjacent to each other. The total number of packet sent is decreased in those experiments since the number of sender nodes is fixed in the middle of the experiments. Table 4 - The average of experimental result in Figure 4.19 and Figure 4.20 Uniform Nonuniform (10/30) CETAR with BCE 9482 CETAR with AABCE 11009 CETAR with BCE and DATP 15852 CETAR with AABCE and DATP 16948 4212 5206 6706 8137 Table 5 - Improvement of CETAR with AABCE over CETAR with BCE (2) Uniform Nonuniform (10/30) CETAR with AABCE over CETAR with BCE 16 % CETAR with AABCE and DATP over CETAR with BCE and DATP 6.9% 23.5 % 21.1 % 76 Chapter 5 Conclusion and Future Works Consumed-energy-type-aware routing (CETAR), proposed in this thesis preserves the energy of active source nodes by discouraging them to participate for routing tasks. CETAR uses statistics of the energy consumed for each type of node activities including sensing, data processing, data transmission as a source node, and data receiving/transmission as a routing node for routing decision. In particular, CETAR selects a node with high residual energy which seldom plays a role of source node as a routing node. Idea is to maintain the energy of active source nodes to prolong the functionality of the WSNs. Two types of CETAR are proposed. Biased consumed energy (BCE) is derived based on a bias factor, β, of pre-fixed value that can be applied to discourage an active source node from joining a routing path. Aggressively-and-Adaptively BCE (AABCE) is derived based on β whose value can be dynamically updated based on the amount of energy consumed by sending and routing activities. Simulation model is developed to evaluate the relative performance improvement of the CETAR over geographic and energy aware routing (GEAR). Simulation results demonstrated that the lifetime of the CETAR based on BCE improves that of the GEAR for all experiments conducted. Furthermore, that of the CETAR based on AABCE further improves that of the CETAR based on BCE. Our simulation evaluations demonstrate that CETAR can significantly extend the lifetime of WSNs especially for non-uniform traffic in which sender nodes are 77 closely clustered and a active source is highly likely to have to route packets of its neighbor if CETAR is not available. 5.1 Contributions of this research This research addressed the new energy-aware routing scheme to extend the lifetime of WSNs. The main contributions of my research presented in this thesis are as follows: 1) A new metric used for energy aware routing to prolong the lifetime of WSNs is proposed and investigated. 2) The adaptability of CETAR is demonstrated for a wide range of network topology which use residual energy information for deciding routing path. 3) Dynamic power control is incorporated on GEAR simulator proposed by Yu et al. in [8] and its performance gain is evaluated. 4) The effect of the new metric is evaluated using GEAR simulator. 5.2 Discussions for Future Works Further evaluation is needed to obtain more accurate performance gain of the proposed mechanisms by increasing the number of experiments. We intend to further investigate the performance of the proposed energy-aware metric in a more dynamic environment including interference and the energy consumptions incurred by the exchange of information among neighboring nodes. Furthermore, we intend to incorporate CETAR to different EAR simulators other than GEAR to enhance the validity of its adaptability demonstrated in section 3.6. 78 Though CETAR with AABCE outperformed that with BCE, AABCE does not always put appropriate biased cost for a sending activity in many cases. Suppose a node consumes a great amount of energy for routing activities at first, and it suddenly becomes active as a sender node. In this case, this node cannot immediately be prevented from routing activity because the value of β would be significantly low due to the energy consumed for routing in the past based on equation (2). One possible solution is to consider not only the total statistics of consumed-energy type but also recent statistics of consumed-energy type to control the bias factor. 79 References [1] K. Martinez, J. Hart, and R. Ong, "Environmental Sensor Networks," IEEE Computer 37(8) pp. 50-56, 2004 [2] Workshop: "Sensors for Environmental Observatories," Final report available at http://wtec.org/seo/final/Sensors_for_Environmental_Observatories.pdf, University of Washington, Seattle, November 30 – December 2, 2004 [3] T. T. Hsieh, "Using sensor networks for highway and traffic applications," Potentials, IEEE, pp. 13-16, vol 23, issue 2, April-May, 2004 [4] K. Martinez, J. K. Hart, R. Ong, "Environmental Sensor Networks," Computer, v.37 n.8, p.50-56, August 2004 [5] A. Mainwaring, J. Polastre, Robert Szewczyk, David Culler, and John Anderson, "Wireless Sensor Networks for Habitat Monitoring," WSNA'02, Atlanta, Georgia, September 2002 [6] M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, "A Survey of Application Distribution in Wireless sensor networks," EURASIP J. Wirel. Commun. Netw. 5, 5 (Oct. 2005), 774-788 [7] I. F. Akyildiz, S. Weilian, Y. Sankarasubramaniam, and E. Cayirci, “A survey on sensor networks,” IEEE Commun.Mag., vol. 40, pp. 102-114, Aug. 2002 [8] Y. Yu, R. Govindan, and D. Estrin, “Geographical and Energy Aware Routing: A Recursive Data Dissemination Protocol for Wireless Sensor Networks,” UCLA Computer Science Department Technical Report UCLA/CSD-TR-01-0023, May 2001 80 [9] G. Pei and C. Chien., "Low power TDMA in Large Wireless Sensor Networks," Military Communications Conference (MILCOM'01), Communications for Network-Centric Operations: Creat, October 2001 [10] W. Ye, J. Heidemann, and D. Estrin., "An Energy-Efficient MAC Protocol for Wireless Sensor Networks.," In Proceedings of the 21st International Annual Joint Conference of the IEEE Computer and Communica, June 2002 [11] V. Rajendran, K. Obraczka, and J. J. Garcia-Luna-Aceves., "Energy-Efficient, Collision-Free Medium Access Control for Wireless Sensor Networks.," The First ACM Conference on Embedded Networked Sensor Systems (Sensys'03), November 2003 [12] T. V. Dam and K. Langendoen, "An Adaptive Energy-Efficient MAC Protocol for Wireless Sensor Networks.," The First ACM Conference on Embedded Networked Sensor Systems (Sensys'03), November 2003 [13] J. Polastre, J. Hill, and D. Culler, "Versatile Low Power Media Access for Wireless Sensor Networks," ACM Sensys, November 2004 [14] C. Schurgers, V. Tsiatis, S. Ganeriwal and M. Srivastava, "Optimizing sensor networks in the energy-latency-density design space," IEEE Trans. Mob. Comp., vol.1, pp.70-80, January-March 2002 [15] R. Bellman, "On a Routing Problem," In Quarterly of Applied Mathematics, 16(1), pp.87-90, 1958 [16] L. R. Ford (Jr.), "Network Flow Theory," the RAND Corporation, Paper P-923, August 1956 [17] Wikipedia, "Geographic Routing," http://en.wikipedia.org/wiki/Routing, April 2007 81 [18] E. W. Dijkstra, "A note on two problems in connexion with graphs," In: Numerische Mathematik, PP. 269–271, 1959 [19] D. B. Johnson, D. A. Maltz, and Y-C. Hu, "The Dynamic Source Routing Protocol for Mobile Ad Hoc Networks (DSR)," RFC 4728. July 2004 [20] C. Perkins, E. Belding-Royer, and S. Das. "Ad hoc On-Demand Distance Vector (AODV) Routing,", RFC 3561. July. 2003 [21] T. Clausen and P. Jacquet, "Optimized Link State Routing Protocol (OLSR)," RFC 3626, October 2003 [22] R. Ogier, F. Templin and M. Lewis, "Topology Dissemination Based on Reverse-Path Forwarding (TBRPF)," RFC 3684, February 2004 [23] G. G. Finn, "Routing and Addressing Problems in Large Metropolitan-scale Internetworks," ISI Report, ISI/RR-87-180, Mar. 1987 [24] E. Kranakis, H. Singh and J. Urrutia, "Compass routing on geometric networks," Proc. 11 th Canadian Conference on Computational Geometry, Vancouver, August, 1999 [25] C. Intanagonwiwat, R. Govindan and D. Estrin, "Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks,” submitted to ACM MobiCom 2000 [26] B. Karp and H. T. Kung, “Gpsr: Greedy perimeter stateless routing for wireless networks,” In Proc. ACM Mobicom, Boston, MA, 2000 [27] Wikipedia, “Planar Graph," http://en.wikipedia.org/wiki/Planar_graph, March 2007 [28] P. Bahl, V. N. Padmanabhan, "RADAR: An In-Building RF-Based User Location and Tracking System," IEEE Infocom 2000, pp:775--784, 2000 82 [29] J. W. Jaromczyk and G. T. Toussaint. Relative neighborhood graphs and their relatives. Proc. IEEE, 80(9):1502--1517, 1992 [30] D. W. Matula and R. R. Sokal, "Properties of Gabriel Graphs Relevant to Geographic Variation Research and Clustering of Points in the Plane," Geogr. Anal. 12: 205–222, 1980 [31] J. Gao, L. Guibas, J. Hershberger, L. Zhang, A. Zhu, “Geometric Spanner for Routing in Mobile Networks,” Proc. ACM MobiHoc01, pp. 45-55, 2001 [32] F. Kuhn, R. Wattenhofer, and A. Zollinger, "Worst-Case Optimal and Average-Case Efficient Geometric Ad-Hoc Routing," ACM MOBIHOC 2003 [33] G. Gupta and M. Younis, "Performance Evaluation of Load-Balanced Clustering of Wireless Sensor Networks," 10th International Conference on Telecommunications (ICT'03)., March 2003 [34] J. H. Chang, L. Tassiulas, "Energy Conserving Routing in Wireless Ad-hoc Networks," INFOCOM '2000, March 2000 [35] I. Stojmenovic and X. Lin, "Power aware localized routing in wireless networks," IEEE Transactions on Parallel and Distributed Systems, Vol. 12, No. 11, pp. 1122-1133, November 2001 [36] Encyclopedia Britannica, “NP-complete problem,” http://www.britannica.com/eb/article-9056426/NP-complete-problem, April 2007 [37] O. Younis and S. Fahmy, "Distributed Clustering in Ad-hoc Sensor Networks: A Hybrid, Energy-Efficient Approach," INFOCOM'04, March 2004 83 [38] M. Haenggi, "Energy-balancing strategies for wireless sensor networks," in Proc. IEEE International Symposium on Circuits and Systems (ISCAS '03), vol. 4, pp. IV-828-IV-831, Bangkok, Thailand, May 2003 [39] W. Rabiner Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy-Efficient Communication Protocol for Wireless Microsensor,” Networks. In Proceedings of the 33rd International Conference on System Sciences (HICSS ’00), 2000 [40] S. Lindsey and C.S. Raghavendra, PEGASIS: Power-Efficient GAthering in Sensor Information Systems, Aerospace Conference Proceedings, March 2002 [41] F. Ye, H. Luo, J. Cheng, S. Lu, and L. Zhang, "A Two-Tier Data Dissemination Model for Large-scale Wireless Sensor Networks," MOBICOM'02, August 2002 [42] J. N. Al-Karaki, and A. E. Kamal, "Routing Techniques in Wireless Sensor Networks: A Survey," IEEE Wireless Communications, vol.11, no.6, December 2004 [43] S. Singh and C. S. Raghavendra, "PAMAS: Power Aware Multi-Access protocol with Signalling for Ad Hoc Networks," (to appear) ACM Computer Communications Review, 1999 [44] W. Ye, J. Heidemann, and D. Estrin, "An Energy-Efficient MAC Protocol for Wireless Sensor Networks," INFOCOM02. 2002 [45] Wikipedia, "Carrier Sense," http://en.wikipedia.org/wiki/Carrier_sense [46] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, "Span: an Energy-Efficient Coordination Algorithm for Topology Maintenance in Ad Hoc Wireless Networks," ACM Wireless Networks Journal, Volume 8, Number 5, September 2002 84 [47] B. Hull, K. Jamieson, and H. Balakrishnan, "Mitigating Congestion in Wireless Sensor Networks," ACM SenSys04. 2004 [48] V. Rodoplu and T. H. Meng, “Minimum energy mobile wireless networks.” IEEE J. Selected Areas in Communications, 17(8):1333-1344, August 1999 [49] S. Banerjee and A. Misra, “Adapting Transmission Power for Optimal Energy Reliable Multi-hop Wireless Communication,” Wireless Optimization Workshop, Sophia-Antipholis, March 2003 [50] M. Kubisch, H. Karl, A. Wolisz, L. Zhong, and J. Rabaey, "Distributed Algorithms for Transmission Power Control in Wireless Sensor Networks," Proceeding of IEEE Wireless communications and networking Conference, pp. 558-563, March 2003 [51] S. Narayanaswamy, V. Kawadia, R. S. Sreenivas, and P.R.Kumar, "Power Control in Ad-Hoc Networks: Theory, Architecture, Algorithm and Implementation of the COMPOW protocol." In Proceedings of the European Wireless Conference Next Generation Wireless Networks: Technologies, Protocols, Services and Applications, pages 156-162, Florence, Italy, February 2002 [52] V. Kawadia and P. Kumar, "Power Control Clustering in Ad Hoc Networks," Proceedings of IEEE INFOCOM, pp. 459-469, March 2003 [53] J. Gomez, A.T. Campbell, M. Naghshineh, and C. Bisdikian, "PARO: Supporting Dynamic Power Controlled Routing in Wireless Ad Hoc Networks," in ACM/Kluwer Journal on Wireless Networks (WINET), Volume 9, Issue 5, pp. 443-460, 2003 [54] J. Gomez and A. T. Campbell, “A Case for Variable-Range Transmission Power Control in Wireless Multihop Networks,” IEEE INFOCOM, 2004 85 [55] N. Bulusu, J. Heidemann, and D. Estrin, "GPS-less Low-Cost Outdoor Localization for Very Small Devices," IEEE Personal Communication, October 2000 [56] A. Rao, C. Papadimitriou, S. Ratnasamy, S. Shenker, and I. Stoica, "Geographic routing without location information," In Proc. ACM Mobicom, San Deigo, CA, 2003 [57] N. B. Priyantha, A. Chakraborty, H. Balakrishnan, "The ,Cricket Location-Support system," Proc. 6th ACM MOBICOM, Boston, MA, August 2000 [58] D. Niculescu and B. Nath, "Ad Hoc Positioning System (APS) Using AOA," IEEE GLOBECOM 2001, San Antonio, November 2001 [59] M. K. Marina and S. R. Das, "Routing performance in the presence of unidirectional links in multihop wireless networks," In Mobihoc. ACM, 2002. [60] V. Ramasubramanian and D. Mosse. "Srl: Providing a bi-directional abstraction for unidirectional ad hoc networks," Submitted to Transactions on Networking, 2002 [61] Sensoria Corporation, http://www.sensoria.com/ [62] M. Woo, S. Singh, and C. S. Raghavendra, “Power-Aware Routing in Mobile Ad Hoc Networks,” ACM/IEEE MOBICOM, 1998 [63] Wikipedia, “Geographic Routing,” http://en.wikipedia.org/wiki/Geographic_routing, March 2007 [64] V. Kawadia and P. Kumar, "Principles and Protocols for Power Control in Wireless Ad Hoc Networks," IEEE Journal on Selected Areas in Communications, vol. 23, No. 1, pp.76-88, January 2005 [65] V. Kawadia and P. Kumar, "Power Control and Clustering in Ad Hoc Networks," Proceedings of IEEE INFOCOM, pp. 459-469, March 2003 86 [66] J. Hightower and G. Borriello, "Location systems for ubiquitous computing," IEEE Computer, vol. 34, no. 8, pp. 57–66, August 2001 [67] S Bajaj, L. Breslau, D. Estrin, K. Fall, S. Floyd, P. Haldar, M. Handley, A. Helmy, J. Heidemann, P. Huang, S. Kumar, S. Mc- Canne, R. Rejaie, P. Sharma, K. Varadhan, Y. Xu, H. Yu, and D. Zappala, "Improving simulation for network research," Technical Report 99-702b, University of Southern California, March 1999. revised September 1999, to appear in IEEE Computer [68] P. Bahl, V. N. Padmanabhan, "RADAR: An In-Building RF-Based User Location and Tracking System," IEEE Infocom 2000, pp:775--784, 2000 [69] J. F. Kurose, K. W. Ross, "A Top-Down Approach Featuring the Internet - 4th edition, " Addison Wesley, ISBN: 978-0321227355, 2004 [70] S. Ando, Y. Tamura, Y. Tobe, M. Minami, "Sensor Network Technologies," Tokno-Denki Publishing, ISBN: 978-4-501-32470-4, May 2005 [71] S. Ito and K. Yoshigoe, "Consumed-Energy-Type-Aware Routing for Wireless Sensor Networks," to appear in IEEE Wireless Telecommunications Symposium, April 26-28, 2007 [72] S. Ito and K. Yoshigoe, "Exponentially Aggressive Preservation of Nearly Depleted Nodes for Wireless Sensor Networks," to appear in the International Conference on Information Technology: New Generations, April 2-4, 2007