中華大學碩士論文

(1)

中華大學碩士論文

題目：

在無線感測網路中探討基於虛擬網格之路由演算法

系所別：資訊工程學系碩士班學號姓名：M09502026 李志軒指導教授：梁秋國博士

中華民國九十九年二月

(2)

A Study of Grid-Based Routing Protocols in Wireless Sensor Networks

Student：Chih-Shiuan Li

Advisor：Dr. Chiu-Kuo Liang

Department of Computer Science and Information Engineering Chung Hua University, Hsin Chu, Taiwan

February, 2010

(3)

i

Abstract

A wireless sensor network is composed of a large number of tiny sensor nodes which can monitor environment, communicate wirelessly, do computations and report the occurrences of interesting events (e.g. forest fire, chemical spills, etc.). In wireless sensor networks, Grid structure was proposed for energy efficient data routing.

Instead of propagating query messages from sink to all sensors to set up data forwarding information, the grid structure uses only sensors located at the grid points needed to acquire the forwarding information. To periodically collect data is an important task from an area of interest for time-sensitive applications. The sensed data must be gathered and transmitted to a base station for further processing to meet the end-user queries. Since the network consists of low-cost nodes with limited battery power, it is a challenging task to design an efficient routing scheme that can collect massive data and offer good performance in energy efficiency, and long network lifetimes.

In this thesis, we consider the problem of grid-bases routing in wireless sensor networks (WSNs). We consider two different models for the problem and propose two routing models for energy-balanced and energy-efficient routing, respectively.

Our schemes don't use global flooding to transmit data and thus are suitable for large of WSNs. Simulation results show that our schemes achieves a better solution in load balancing and energy-efficient in WSNs.

(4)

ii

List of Figures

Figure 1.1. A wireless sensor network schematic illustration.. ... 2

Figure 1.2. Sensor nodes communicate with the sink directly ... 3

Figure 2.1(A). Interests propagation ... 10

Figure 2.1(B). Initial gradients set up ... 10

Figure 2.1(C). Data delivery along reinforced path ... 10

Figure 2.2. An example of greedy forwarding.. ... 11

Figure 2.3. A sensor network example. ... 12

Figure 2.4. One event source node B and one sink S ... 13

Figure 2.5. Two-tier query and data forwarding between source B and Sink S. ... 14

Figure 2.6. Grid construction and roles in GGR networks... 16

Figure 2.7. Other Dissemination Nodes are chosen by varying the cell size when rebuilding the grid ... 17

Figure 2.8. Grid condition before using LBDD hand-off method.. ... 19

Figure 2.9. Grid condition after using LBDD hand-off method.. ... 19

Figure 3.1. The energy consumption on the forwarding path.. ... 24

Figure 3.2. Different Dissemination Nodes are chosen by moving the DP to the right direction when rebuilding the grid.. ... 25

Figure 3.3. The spiral shifting path ... 26

Figure 3.4. A Steiner Point P in a △ABC ... 27

Figure 3.5. A Steiner tree with three points A, B and C ... 28

Figure 3.6. Two Steiner Points X and Y in a square... 28

Figure 3.7. A Steiner tree with four points A, B, C and D ... 28

Figure 3.8. The network construction in STGR ... 30

Figure 3.9. Routing paths in STGR ... 31

Figure 4.1. Standard deviation of remaining energy. ... 35

Figure 4.2. Node death rate vs. number of rounds ... 36

Figure 4.3. Node death rate vs. number of grid shifting ... 37

Figure 4.4. A routing path using GGR protocol ... 38

Figure 4.5. A routing path using STGR protocol ... 38

Figure 4.6. A routing path with a hole using GGR protocol ... 39

Figure 4.7. A routing path with a hole using STGR protocol... 40

Figure 4.8. Transmission hops in each rounds ... 41

Figure 4.9. Total energy consumption in each rounds ... 41

(6)

iv

List of Tables

Table 4.1. The parameters of experiment ... 33

(7)

1

Chapter 1 Introduction

The desire to create smart environments has existed for decades, and recent advances in such areas as pervasive computing, precision machinery, and wireless and sensor networking now allow this dream to become a reality. Wireless sensor networks (WSN) have recently come into prominence because they hold the potential to revolutionize many segments of our economy and life, from environmental observing and preservation, to manufacturing and business asset management, to automation in transportation and health-care industries. These sensor networks consist of peculiar nodes that are able to interact with their environment by sensing or controlling physical parameters; these nodes have to collaborate to fulfill their tasks as usually, a single node is incapable of doing so; and they use wireless communication to enable this collaboration. In essence, the sensor nodes without such a network contain at least some computation, wireless communication, and sensing or control functionalities. Unlike a centralized system, a sensor network is subject to a unique set of resource constraints such as finite on-board battery power and restricted communication ability [1, 2, 3]. As an advance in wireless communication and electronics technology has enabled the development of low-cost sensor nodes, wireless sensor networks have been pervasive in various applications.

A wireless sensor network is composed of a large number of sensor nodes that are densely deployed [4, 5]. A sensor node typically has embedded processing capabilities and onboard storage; the node can have one or more sensors operating in the acoustic, seismic, radio, infrared, optical, magnetic, and chemical or biological

(8)

2

domains. The node has communication interfaces, typically wireless links, to neighboring domains. The sensor node also often has location and positioning knowledge that is acquired through a global positioning system (GPS) or local positioning algorithm [6]. Sensor nodes are scattered in a special domain called a sensor field. Each of the distributed sensor nodes typically has the capability to collect data, analyze them, and route them to a sink point.

Among the various scopes one of the major applications of sensor network is to collect information periodically from a remote terrain where each node continually senses the environment and sends back this data to the sink for further analysis. For the purposes of our works, a wireless sensor network include perhaps several hundred even thousands sensor nodes deployed at random, and potential multiple data sink.

One role of the data sink is to disseminate data collection instructions to sensor nodes within the network [7]. Figure 1.1 depicts a typical WSN arrangement.

Sensor field User

Internet

Sink

Sensors

Wireless Links

Figure 1.1. A wireless sensor network schematic illustration.

Since large numbers of sensor nodes are densely deployed, neighbor nodes may be very close to each other. Hence, multi-hop communication in wireless sensor networks is expected to consume less power than the traditional single hop

(9)

3

communication. As shown in Figure1.2, it might cause high energy consumption, when sensor nodes send message directly to the sink.

Sink

Sensors

Wireless Links

Figure 1.2. Sensor nodes communicate with the sink directly.

Furthermore, the transmission power levels can be kept low, which is highly desired in covert operations. Multi-hop communication can also effectively overcome some of the signal propagation effects experienced in long-distance wireless communication.

One of the most important constraints on sensor nodes is low power consumption requirement. Sensor nodes carry limited, generally irreplaceable, power source.

Therefore, while traditional networks aim to achieve high quality of service (QoS) provisions, sensor network protocols must focus primarily on power conservation.

They must have inbuilt trade-off mechanisms that give the end user the option of prolonging network lifetime at cost of lower throughput or higher transmission delay[6]. So, there are two important issues in design routing protocol for wireless sensor networks. One is that how to prolong the operating time of networks, the other is that how to collecting data quickly.

(10)

4

This paper studies the problem of scalable and efficient data dissemination in a large-scale sensor network from potentially multiple sources to potentially multiple sinks. In this paper a source is defined as a sensor node that detects a target or an event of interest, and generates data to report the target or event. A sink is defined as a user that collects these data reports from the sensor network.

The communication of the query instructions from sinks to sensors and the retrieval of data corresponding to those queries from the sensors continue to be an important research area. The communication must be efficient, particularly in the face of failure, and must maintain a high level of coverage in the sensor network.

Satisfying these goals of energy efficiency and load balancing creates a challenging problem.

Ours works presented in this paper builds on previous attempts to satisfy the routing problem, and in particular uses the same grid structures in Two-Tier Data Dissemination (TTDD) protocol [8] and Geographic Grid Routing (GGR) protocol [9, 10]. We attempt to improve on the energy efficiency and load balancing by proposing a data dissemination routing protocol, the Spiral Grid Routing (SGR) protocol with low energy consumption and load balance features. And we presented a routing protocol named Steiner Trees Grid Routing (STGR) protocol, to improve on the deliverer and energy efficient.

The rest of the paper is organized as follows. In Section 2, we give an overview of the related routing protocols. Details of the Spiral Grid Routing protocol and the Steiner Trees Grid Routing protocol are given in Section 3. Section 4 shows a comparative analysis and simulation results. Finally, Section 5 presents a concluding remark.

(11)

5

Chapter 2 Related Works

Although the advance of technological progress causes the research of wireless sensor network development very rapid. How to retard energy consumption of sensor nodes is still a major issue. Therefore, there are many of protocols that have been developed for wireless sensor networks. In the following sections we will be divided into two parts to describe: non grid-based routing protocols and grid-based routing protocols. Several non grid-based routing schemes, Several schemes, such as Directed Diffusion, Sensor Protocol for Information via Negotiation (SPIN) [15], and Low Energy Adaptive Clustering Hierarchy (LEACH) [16], are suggested and investigated for the purpose of using less energy by a sensor node in collecting information and receiving/transmitting data. If a specific sensor node consumes power more rapidly in transferring data, the network lifetime will be decreased more rapidly. The lifetime of a network is defined as the duration of time prior to the failure of the first node due to battery exhaustion, and, in the worst case, it may cause network partitioning. Much research has been conducted related to routing schemes that use the limited resources available in WSNs more efficiently. These schemes typically focus on finding the minimum energy path for optimizing energy usage. Because using the lowest energy paths may not always be optimal from the point of view of the network lifetime and long-term connectivity, an Energy Aware Routing (EAR) scheme is proposed. The EAR utilizes sub-optimal paths occasionally, and therefore is more efficient.

Additionally, simulation results are also presented that demonstrate an increase in network lifetimes of up to 40% compared to comparable directed diffusion routing

(12)

6

[17]. Therefore, the basic concept of EAR has been applied to grid-based routing schemes for energy efficiency.

2.1 Non Grid-Based Routing protocols

In this section, we will introduce two routing protocols without using grid structure to help routing.

2.1.1 Directed Diffusion(DD)

The traditional way of data transmission is using flooding to deliver data. The data sink sends the query to sensor nodes which are within the radio range of sink, the sensor which received the query from the sink would check the implementation of the directive and continue to send the query to its neighboring sensor nodes. After that next sensor which received the same query would do the same actions again, the query would flow each sensor nodes to the entire network.

After the sensor node collected information, sensor node would send collecting data back to sink by flooding. According to [7], it can be seen that flooding has several deficiencies such as overlap and resource blindness. The same query will be retransmission many times, it cause the huge energy consumption and network jam.

In view of this, an important feature of directed diffusion (DD) [7] is that interest and data propagation and aggregation are determined by localized interactions (message exchanges between neighbors or nodes within some vicinity). DD is a data-centric protocol which means that the routing path wouldn’t be create until the sink sent the query out. Directed diffusion consists of several elements: interests, data messages, gradients, and reinforcements. An interest message is a query or an interrogation which specifies what a user wants. Typically, data in sensor networks is the collected or processed information of a physical phenomenon. Such data can be an event which is a short description of the sensed phenomenon. A sensing task is disseminated throughout the sensor network as an interest for named data. This

(13)

7

dissemination sets up gradients within the network designed to draw events.

Specifically, a gradient is direction state created in each node that receives an interest.

Events start flowing towards the originators of interests along multiple gradient paths.

The sensor network reinforces one, or a small number of these paths. Directed diffusion divided routing protocol into three phases: interest propagation, data propagation and reinforcement.

 Interest propagation

The sink uses flooding to send interest (query) to its neighboring sensor nodes, after these sensor nodes which are next to the sink received interest, sensor nodes will send back a gradient to the sink. Shown in the figure 2.1(A), sink sent out the interest to its neighboring nodes, and there are three sensor nodes received the interest. Shown in Figure 2.1(B), when the sensor node A received this interest, node A would sent a gradients back to the sink. The purpose of doing this is to create a back path from the sensor node to the sink. Next, node A would continue to send this interest to its neighboring nodes. Sensor nodes would check the duplicate receiving interest message, once received the same interest, the sensor node would check the gradients between transmitter and receiver itself. Shown in Firure2.1, after sensor node A broadcasted interest to neighbors, sensor node D would received this interest. And node D would broadcast interest and sent a gradient back to node A, after that node D will also received the same interest from sensor node C. In this moment, sensor node D found no gradient to sensor node C, node D would send a gradient to node C in

order to create a back path, but node D wouldn't broadcast this interest again. This process would stop when the interest reached the data source node, and go into the second part of protocol.

 Data Propagation

Shown in Figure 2.1, once sensor node B received the interest and confirmed the

(14)

8

source of event. Sensor node B would start to begin data propagation; it broadcasted data to neighboring nodes. Sensor nodes would check gradients as soon as received the data from data source node B. If the sensor node has gradient between node B and itself, the sensor would continue to broadcast the data to its neighbors. Finally, according to gradient generated in interests propagation phase, the sink would get the data.

 Reinforcement

Shown in figure 2.1(B), it can be found that there are several routing paths from source node B to the sink. It is wasteful if every times node B transfers data to the sink by using all of these paths. So, when the first time node B transmitted data to the sink, a relay node decided to broadcast data basis on gradients. The relay node would add a count for the transmitted node. Finally, the sink would know sensor node A has higher count than the other two sensor nodes. So, the sink would choice a high frequency path to transport data, and the sink can choice the other path to continue transmission once the high frequency path failed.

To compare with flooding protocol, directed diffusion does better in energy consumption, because directed diffusion doesn’t need to use flooding every time. But, directed diffusion has a weakness when the sink sent a query to the wireless sensor networks. In the large sensor field, the sink broadcasted query would waste a lot of energy by using flooding data forwarding.

Source Event

Sink

C

A B

D Interest

Figure 2.1(A). Interests propagation

(15)

9

Source Event

Sink

C

A B

D Gradients

Figure 2.1(B). Initial gradients set up

Source Event

Sink

C

A B

D

Figure 2.1(C). Data delivery along reinforced path

2.1.2 Geographic Routing

In wireless sensor networks, the typical routing protocol which sensor node has GPS device is greedy perimeter stateless routing (GPSR) [11]. Because of sensor nodes have GPS device; each node knows the location information itself. So, as a node want to transmit data to the sink, it used greedy forwarding protocol [13] to transmit data. Shown in figure 2.2, sensor node x needs to send data to destination node D. Every node in sensor field knows the location information of destination node and its location information. Sensor node x would communication with its neighboring nodes in order to choose a node which is closest to destination node D.

After doing this process, sensor node x would choose sensor node y to been the next hop relay node. Sensor node y would do the same procedure to find next hop relay node, finally, data could reach the destination node D. In some researches [9, 10]

mention that greedy forwarding has a big defect, when a node doesn’t have any neighboring nodes which is closer destination node than itself, this node couldn’t

(16)

10

continue to transmit data. So, such as GPSR [11] proposed a detour protocol to solve dead end problem. But such a protocol can't avoid the redundant energy consumption on the routing.

x

y

D

Figure 2.2. An example of greedy forwarding. The node x chose next hop node y which is closest to the destination D.

2.2 Grid-Based Routing protocols

Grid structure was proposed for energy efficient data routing. Instead of propagating query messages from sink to all sensors to set up data forwarding information, the grid structure uses only sensors located at the grid points needed to acquire the forwarding information. There are three related papers talking about this topic.

2.2.1

Two

Tier Data Dissemination(TTDD) Model

In order to reduce energy consumption cause by flooding in large scale sensor field, two tier data dissemination (TTDD) [8] routing model was been proposed.

TTDD assumed that there is a battlefield which can set up wireless sensor networks just like Figure 2.3 shown. In this case, TTDD postulated soldiers as sinks and postulated tanks as events. Soldiers could use wireless sensor networks to detect tank.

(17)

11

In the military scenario, soldiers need to track tank frequently, if every time soldiers send out queries by flooding to entire sensor field, it takes a big cost.

Source A

Source B

Sink 1 Sink 2

Sink 3

Figure 2.3. A sensor network example. Soldiers use the sensor network to detect tank In TTDD, when a sensor detected an event, this sensor would set itself as an event source. The source node divided the sensor field plane into a grid of cells. Each cell is a α x α square, the size of α determined by the user. The source node propagates the announcement of event to each cross point of grid, called dissemination points (DPs); the source node located at the first dissemination point.

Shown in Figure 2.4, sensor node B set itself as a source and it would calculate the location of adjacent four neighboring dissemination points by given its location (x, y) and cell size α. And source node B promoted itself as the first dissemination node (DN) and sent a data-announcement message to adjacent dissemination points location used simple greedy geographical forwarding. A dissemination node stores a few pieces of information for the grid structure, including the data announcement message, the dissemination point it is serving and the upstream dissemination node's location.

When a node received announcement message, it could calculate the distance between the location of dissemination point and itself. If the distance less than half of transmission ranges of sensor node, the sensor node would communication with its neighboring nodes in order to find a sensor node which is closest to the dissemination

(18)

12

point location. The closest sensor node would promote itself to become a dissemination node, and did the same process to choose DNs. The data announcement message is recursively propagated through the whole sensor field so that each dissemination point on the grid is served by a dissemination node. Duplicate announcement messages from different neighboring dissemination points are identified by the sequence number carried in the announcement and simply dropped.

α

B S

Figure 2.4. One event source node B and one sink S.

When the source node sent data-announcement message to construct grid, it also broadcasted the announcement of event, so every dissemination node knows the location of event source. As the sink send out a task, the dissemination node which is closest to the sink would check the task exist or not. The sink starts with flooding its query with its location of primary agent, to its immediate dissemination node. The dissemination node records primary agent's location and forwards the query to its upstream dissemination node until the query reaches the source node. The data are returned to the dissemination node along the way that the query traverses. The dissemination node forwards the data to primary agent, and finally to the sink. If sensor field didn't have information which the sink wanted, the dissemination node could give the sink a feedback quickly. Shown in Figure 2.5, sink S send out the query

(19)

13

and the closest DN knew source node B which had the information. So, the DN relayed task to source node B by using simple greedy geographic forwarding, and source B transported data to the sink by using the same way.

Task Data

B S

Figure 2.5. Two-tier query and data forwarding between source B and Sink S.

According to [9, 10], it found that every time sensor nodes detected a new event or the original event moved, would lead sensor nodes to create a new grid. When the wireless sensor networks operated in a long time, it may because sensor nodes maintain routing path hardy, and redundant energy consumption. The grid structure used by TTDD achieves reliable multipath message transport, but is far more efficient than a blind network flood. The use of geographic routing also reduces the number of nodes that must maintain routing state information to those devices belonging to the set of Dissemination Nodes. If a single link fails, the affected downstream DNs often have one or two spare upstream links that can be used instead. In addition to its use in increasing communication reliability, the ACK message can be of use in dealing with path failures. In the event that no more upstream links are available, an ACK message informs further downstream nodes that the path has failed. In contrast to geographic grid routing [9, 10], the TTDD routing protocol uses only a single path for

(20)

14

transmission of data messages back to the sink. The path could easily be disrupted by a failed sensor and force the entire Advertisement-Request phase to be repeated – a process which requires significant energy resources.

2.2.2 Geographic Grid Routing(GGR)

In order to solve these problems of TTDD routing protocol, geographic grid routing (GGR) [9, 10] has been proposed. GGR used the same grid structure as TTDD, but the difference is that GGR used the sink to be the source of the grid. Although TTDD uses a similar grid structure to GGR, it is used in a less efficient and reliable manner. TTDD uses a two-way Advertisement-Request exchange that is less efficient than the sink tasking scheme used by GGR. TTDD also only uses multiple paths for the Advertisement phase so the data request could be lost on the single chosen path.

The operation stage of GGR has been divided into three parts:

 Grid Construction and Task Dissemination

The dissemination of data request messages has a dual purpose in a GGR network; establishment of forwarding gradients and task assignment. GGR builds upon the idea of a forwarding grid introduced by TTDD for routing messages back to the sink. The major difference is that the GGR grid is based at the data sink rather than the data source, like the TTDD grid. The components of the grid are shown in Figure 2.6. The cross-points of the grid are called dissemination points (DPs). The grid is based at the location of the assigning sink, which serves as the first DP. Shown in Figure 2.6, the sink sends the task message to each of the four adjacent DPs using simple geographic forwarding through intermediate nodes. And sensor nodes would compare with each neighboring nodes in order to choose a DN which is closest to DP.

Eventually the task message reaches DNs all over the sensor field. Using information included in the task message, dissemination nodes rank and stores the most energy efficient upstream links toward the sink.

(21)

15 Cell size

Sink

Dissemination

Nodes Intermediate

Nodes

TASK Message Dissemination Dissemination

Points

Figure 2.6. Grid construction and roles in GGR networks.

 Data Forwarding

Like dissemination nodes, sensing nodes also create routing table entries when a task message is received. Observation data is generated according to the task message description, formatted into a data message, and sent to the lowest cost upstream link in the routing table. Again intermediate nodes provide message transport between DNs via simple geographic forwarding. The data message is moved over the grid using the most energy efficient route available to the sink.

 Grid Maintenance

As a sink gathers data from its forwarding grid, those nodes most involved in the forwarding activities will have their energy stores reduced more quickly than other nodes. The sink eventually rebuilds the grid using a different cell size. As shown in Figure 2.7, even small changes in the cell size can completely change the overlay grid structure and involve a different set of nodes in the forwarding process to prolong time to failure.

The grid maintenance process is very important for sensor replacement. Newly deployed sensors can have a chance to become DNs and play an important role in the

(22)

16

network when the grid is reconstructed. In contrast, TTDD does not include a mechanism for grid reconstruction, and GGR provided a grid maintenance process by varying the grid cell size. To vary the grid cell size can’t ensure every sensor node to become the dissemination node, so the maintenance process of GGR is hard to improve the nodes in the forwarding process to prolong time to failure.

Dissemination Nodes

Sink Dissemination

Nodes Sink

(a) (b)

Figure 2.7. Other Dissemination Nodes are chosen by varying the cell size when rebuilding the grid.

2.2.3 Load Balance Data Dissemination(LBDD)

TTDD doesn't have a network maintenance process and GGR proposed varying the cell size, but it still can't get a good result in prolong the network’s life. So, a low energy consumption and load balance data dissemination (LBDD) [13] model was been proposed.

At the grid construct phase, LBDD is different from TTDD and GGR routing protocols. In the LBDD model, sensor nodes have been set up the grid information including the cell size and the location of initial DP before spilled on the sensor field.

Once sensor nodes spilled on the sensor field, every node calculate its location and find the DP which is closest to itself according to grid information. After doing the processes, DNs can been chosen by comparing with each neighbor nodes. Finally, the sink announces its location to every DN by simply greedy forwarding along the gird.

(23)

17

When the source node needed to transmit data to the sink, it would transport data to closest DN by greedy forwarding. As the DN received data from the source node, the DN would calculate location relationship between the sink and itself, and the DN relayed data to the sink along the routing path which is most energy efficiently. The grid structures are the same both in LBDD and GGR. However, in LBDD, the authors proposed a hand-off mechanism, which is different from the mechanism in GGR, to change the grid structure in order to have the better performance in load balance. The main idea of the hand-off mechanism is to hand-off all the Dissemination Nodes to the geographically upper-right region. Figure 2.8 and 2.9 shows the hand-off process in LBDD, after doing hand-off process, a new group of DNs were chosen, and then data transmission would use the new DNs and nodes along the new grid.

Although the work in LBDD outperforms in load balance than TTDD, it is still not good enough. Therefore, in our work, we propose a different hand-off scheme to get the better performance in load balance.

DN

DP

Sink

Figure 2.8. Grid condition before using LBDD hand-off method.

(24)

18

Sink New DP DN

DP

Figure 2.9. Grid condition after using LBDD hand-off method.

(25)

19

Chapter 3 Our Proposed Schemes

3.1 Preliminary

An energy-balanced mechanism for deployment and topology control primarily consists of two phases— topology construction and the network maintenance phase [18]. Spiral Grid Routing scheme is able to provide an energy-balanced maintenance of networks in wireless sensor networks, and Steiner Trees Grid Routing scheme is able to provide an energy-efficient routing protocol in wireless sensor networks.

Our proposed Spiral Grid Routing (SGR) scheme is a hierarchical protocol for dissemination queries in a sensor network and retrieving the corresponding data. The scheme consists of three stages; grid establishment, data forwarding, and grid maintenance. For the purpose of this work, a sensor network is assumed to have the following characteristics. The sensor field is made up of hundreds or perhaps thousands of small, cheap sensing devices that are randomly deployed throughout a two-dimensional area of interest. The power supply is restricted due to the size of sensor nodes. Short-range radios with static transmission power are used due to the energy constraint. Therefore, multi-hop forwarding schemes are used to achieve long-range communication. The sensing devices are assumed to have a fixed and known location within the sensor field. An immobile data sink is deployed with the area of interest, and has location knowledge and an infinite power source.

Our proposed Steiner Trees Grid Routing (STGR) scheme is also a hierarchical protocol. For the purpose of this work, a sensor network is assumed to have the same characteristics as SGR.

(26)

20

3.2. Spiral Grid Routing Protocol

In this section, we describe the SGR protocol in detail. We make the following assumptions in SGR. Each sensor node is aware of its location after deployment (e.g., using some localization techniques). All sensor nodes, sinks and targets are stationary.

3.2.1 Network Construction

Once the sensor nodes are deployed in the sensor field, the sink node starts to construct the grid structure. The sink divides the plane into a grid of cells.

Cross-points of the grid are the Dissemination Points (DPs). The size of the cells is determined by the sink such that DPs are not within direct transmission range. The sink’s position is the first DP. Knowing its own position and the size of each cell, the sink is able to send a data request (in the form of a data-announcement message) to each adjacent Dissemination Point in the grid. Simple geographic forwarding is used to reach these locations. Upon receiving the data-announcement message, the closest known sensor node to each DP promotes itself to become a Dissemination Node (DN).

Then, the DN forwards the data-announcement message to each of its adjacent DPs, except the point from which the data-announcement message was received. These actions are repeated as the data-announcement message propagates and Dissemination Nodes are chosen throughout the sensor field. The data-announcement messages contain the additive hop count and used energy of DNs along the upstream path.

Using this information, Dissemination Nodes store the most energy efficient upstream links toward the sink. An upstream is dropped if a lower or equal cost link can be offered to the neighboring Dissemination Node. The forwarding grid provides multiple paths toward the sink while avoiding routing loops.

It should be noticed that the grid in our protocol is the same as the one in GGR protocol.

(27)

21

3.2.2 Routing Phase

After the grid structure is constructed, the sink is able to send the queries to the interested area for collecting information. Since the sink is aware of the location of the interested area, simple greedy geographic routing scheme is used to send a data request (in the form of QUERY message) to the next Dissemination Point in the grid.

Then, the corresponding Dissemination Node which is closest to the Dissemination Point will receive the QUERY message. Next, the Dissemination Node will forward the QUERY message to next Dissemination Point which is closer to the region of interest. The forwarding procedure at a Dissemination Node continues till the QUERY message stops at a Dissemination Node which is within in the region of interest.

Upon reaching the region of interest, the QUERY message must be disseminated to all sensors in the defined area. Restricted flooding is the simplest and most effective way to distribute the QUERY message to all nodes within the target region.

A region of interest is specified as a rectangular area and is divided into sub-regions by the virtual grid structure. A Dissemination Node is only concerned with disseminating the QUERY message to sub-regions that are contained in surrounding cells. During the restricted flooding, a message is only broadcast once by each node that is inside the defined region of interest. Sensors within those sub-regions may then use the DN as the first hop when sending observation data to the sink. These flooding actions are in addition to the regular grid construction actions described above for Dissemination Nodes.

Any node that is within the target region of a received QUERY message stores the appropriate routing information and starts to send the sensed data (in the form of DATA message) to the sink. The routing information contains the appropriate

(28)

22

upstream DNs through which DATA messages will be forwarded. Having established paths to the sinks through Dissemination Nodes, data forwarding can occur. Data forwarding involves reception and forwarding of DATA messages, acknowledgment of those messages and reception of acknowledgments for messages sent. The acknowledgment messages sent between DNs are used to acknowledge DATA messages and for flow control.

Upon receiving a DATA packet at a Dissemination Node, the local routing table is checked for an upstream link that may be used to reach the sink. The message is then buffered in case retransmission is necessary. At this moment an acknowledge message is sent back to the sender indicating acceptance or rejection, and providing flow control and energy consumption information. The forwarding procedure at a Dissemination Node continues till the DATA message stops at the sink node.

3.2.3 Network Maintenance

As a sink gathers data from its forwarding grid, those nodes most involved in the forwarding activities will have their energy stored reduced more quickly than other nodes. As shown in Figure 3.1, the intermediate nodes will cost more energy than other nodes and the upstream path will be no longer available. Therefore, the sink eventually has to rebuild the grid in order to balance the energy consumption. In GGR [9, 10], the authors rebuild the grid structure by using a different cell size. Although they rebuild the grid, but the location of first DP, which is the location of sink node, still remains the same in the new grid structure. This means that some nodes involved in the forwarding path in the old grid will be still as the nodes involved in the new grid. Thus, the approach in GGR cannot achieve a good load balance in energy consumption of sensor nodes. In this paper, we propose a mechanism for rebuilding

(29)

23

the grid structure to obtain a better performance in load balance. Our proposed mechanism is called the Spiral Grid Shifting.

Event Sink Relay nodes

Figure 3.1. The energy consumption on the forwarding path.

The main idea of Spiral Grid Shifting is as follows. As we know that the location of sink node is also the location of first DP. If we change the location of first DP, then the locations of all DPs are also changed according to our procedure of grid construction. This will result in obtaining a new grid. Figure 3.2 shows the situation.

Figure 3.2(a) shows the grid structure before grid shifting operation. Figure 3.2(b) shows the new grid structure after the grid shifting operation. Note that the sink node knows the location of the new first DP; therefore it can send the data requests on the new constructed grid structure by sending the data requests to the new DP. Since the grid is changed, the nodes involved in the forwarding activities are also changed, which results in the balance of energy consumption during the data collection process.

(30)

24 DN

DP

Sink Sink

New DP DN

DP

(a) (b)

Figure 3.2. Different Dissemination Nodes are chosen by moving the DP to the right direction when rebuilding the grid. The same network is shown in both (a) and (b).

As shown in Figure 3.2, the shifting is achieved by moving DP located at the sink node to the right direction. However, this is not efficient in term of energy consumption since after a few shifting processes, the grid will be the same as the one in the first place. Therefore, in order to achieve more load balance in energy consumption, we proposed a spiral shifting path. As shown in Figure 3.3, the shifting of DP follows the spiral path. After moving DP into the center of the spiral path, the shifting process will be started from the sink node again. Note that the number of shifting following the spiral path and the shifting distance are determined by the sink node.

Sink New DP

Grid Shifting

Sink New DN

Grid Shifting

(a) (b)

Figure 3.3. The spiral shifting path.

(31)

25

3.3 Steiner Trees Grid Routing Protocol

Before going further, let us first explain the network model in our work. As we have known that the main task of sink node is to disseminate the user queries into a sensor network and retrieve the corresponding data from the sensor network. For this purpose, a sensor network is assumed to have the following characteristics. The sensor field is made up of hundreds or perhaps thousands of small, cheap sensing devices that are randomly deployed throughout a two-dimensional area of interest.

The power supply is restricted due to the size of sensor nodes. Short-range radios with static transmission power are used due to the energy constraint. Therefore, multi-hop forwarding schemes are used to achieve long-range communication. The sensing devices are assumed to have a fixed and known location within the sensor field. An immobile data sink is deployed with the area of interest, and has location knowledge and an infinite power source.

Now, we are going to introduce our proposed routing scheme, which is called the Steiner Trees Grid Routing (STGR) protocol. In order to reduce the total energy consumption for data transmission between the source node and the sink node, we construct a different virtual grid structure instead of virtual grid in GGR. Our idea is to construct the virtual grid structure based on the square Steiner trees. In the following, we will first introduce the concept of Steiner trees. Then, our virtual grid structure will be explained.

The Steiner tree problem, created by Jakob Steiner, is a problem in combinatorial optimization. The Steiner tree problem is similar to the minimum spanning tree problem, but the difference between the Steiner problem and the minimum spanning tree is that, in the Steiner tree problem, extra intermediate vertices may be added to

(32)

26

the graph in order to reduce the length of the spanning tree. These new vertices introduced to decrease the total length of connection are known as Steiner points [13].

The basic idea of our approach is using the grid routing which is based upon a square Steiner tree. In that way we need know how to add extra vertices in the grid.

Basis on [13], the Steiner point in a triangle is also a Fermat point. There are some ways to find Steiner point in a triangle form, shows in Figure 3.4, we can use any one side of a triangle to draw a regular triangle and circumscribe this regular triangle, just like two dashed lines in Figure 3.4. Two dashed lines will cross a point in original triangle, which is the Steiner point. In the Figure 3.5, points A, B and C connect with Steiner point P, become to a Steiner tree with three vertices. In view of Steiner tree with three points, we can use similar way to find a Steiner tree with four points.

Figure 3.6 shows that there is a square have two Steiner points X and Y by connected two vertices of circumscribed circle. So shows in Figure 3.7, points A, B, C and D connect with Steiner points which closest to itself, become a Steiner Tree.

A

B

C P

Figure 3.4. A Steiner Point P in a △ABC.

A

B

P

C

Figure 3.5. A Steiner tree with three points A, B and C.

(33)

27

A

B C

D

X Y

Figure 3.6. Two Steiner Points X and Y in a square.

A

B C

D

X Y

Figure 3.7. A Steiner tree with four points A, B, C and D.

3.3.1 Network Construction

Once the sensor nodes are deployed in the sensor field, the sink node starts to construct the grid structure. The sink divides the plane into a grid of cells.

Cross-points of the grid are the Dissemination Points (DPs). The size of the cells, denoted as , is determined by the sink such that DPs are not within direct transmission range. The sink is the first DP. Knowing its own position and the size of each cell, the sink is able to send a data request (in the form of a data-announcement message) to each adjacent Dissemination Point in the grid. For a sink at location Ls=(x, y), dissemination points are located at Lp = (xi, yj) such that:

{𝑥_𝑖 = 𝑥 + 𝑖 ∙ 𝛼, 𝑦_𝑖 = 𝑦 + 𝑗 ∙ 𝛼; 𝑖, 𝑗 = ± 0, ± 1, ± 2 … }.

According to Figure 3.6, the positions of X and Y depends on the length of square, so Steiner Points (SP) are located at Lsp = (xi, yj) such that:

(34)

28

{𝑥_𝑖 = 𝑥 ± 𝑖 ∙ 3

6 𝛼, 𝑦_𝑖 = 𝑦 ± 𝑗 ∙𝛼

2; 𝑖, 𝑗 = ± 0, ± 1, ± 2 … }

Simple geographic forwarding is used to reach these locations. Upon receiving the data-announcement message, the closest known sensor node to each DP and SP promotes itself to become a Dissemination Node (DN), and DN will record belong to which DP or SP. Then, the DN forwards the data-announcement message to each of its adjacent DPs and SPs, except the point from which the data-announcement message was received. These actions are repeated as the data-announcement message propagates and Dissemination Nodes are chosen throughout the sensor field. There is an example shown in Figure 3.8.

DNs for DPs Sink DNs for SPs

Figure 3.8. The network construction in STGR.

The data-announcement messages contain the additive hop count and used energy of DNs along the upstream path. Using this information, Dissemination Nodes store the most energy efficient upstream links toward the sink. An upstream is dropped if a lower or equal cost link can be offered to the neighboring Dissemination

(35)

29

Node. The forwarding grid provides multiple paths toward the sink while avoiding routing loops.

3.3.2 Routing Phase

Any node that is within the target region of a received QUERY message stores the appropriate routing information and starts to send the sensed data (in the form of DATA message) to the sink. The routing information contains the appropriate upstream DNs through which DATA messages will be forwarded. DN choice which way to transmit DATA message depending on DN belong to DP or SP. If DN belongs to SP, DN will choose a upstream along the hexagon. There are two conditions if DN belongs to DP. While the location of DP, Lp = (xi, yj) wasn't parallel or vertical as the location of sink, L_s=(x, y). DN would choice the upstream along the hexagon.

Otherwise, DN would choice the upstream along the grid. As shown in Figure 3.9, DN transported data about the event A along the grid, and DN transported data about the event B along the hexagon.

DNs for DPs Sink DNs for SPs

Event A Event B

Figure 3.9. Routing paths in STGR.

Having established routing paths to the sinks through Dissemination Nodes, data forwarding process can be started. Data forwarding involves reception and forwarding

(36)

30

of DATA messages, acknowledgment of those messages and reception of acknowledgments for messages sent. The acknowledgment messages sent between DNs are used to acknowledge DATA messages and for flow control. If a single link fails, the affected downstream DNs often have one or two spare upstream links that can be used instead. In addition to its use in increasing communication reliability, the acknowledgments for messages can be used in dealing with path failures. In the event that no more upstream links are available, an acknowledgment for messages informs further downstream nodes that the path has failed.

Upon receiving a DATA packet at a Dissemination Node, the local routing table is checked for an upstream link that may be used to reach the sink. The message is then buffered in case retransmission is necessary. At this moment an acknowledge message is sent back to the sender indicating acceptance or rejection, and providing flow control and energy consumption information. The forwarding procedure at a Dissemination Node continues till the DATA message stops at the sink node.

(37)

31

Chapter 4 Simulation Results

In this chapter, we evaluate the performance of SGR and STGR. In section 4.1 we describe the simulation environment. The performance of SGR and STGR are introduced in section 4.2 and 4.3, respectively.

4.1 Performance Metrics and Simulation Environment

Here we simple describe about simulation setup and experiment model, the simulation environment is described as follows. The simulation program is written in C# on .NET platform. The sensor network contains 2000 nodes in the squared region of size 500  500 meter². Sensors are uniformly distributed in the area with density d

= (80 nodes / 100 meter²). The grid cell size is 100  100 meter². For simplicity, we assume that all sensors are identical. The communication range of each sensor is 40 meters. Each sensor has the initial power level with 6000 Joules. Each data transmission and receiving will take 66 Joules and 39 Joules, respectively. Each measurement of our simulation results represents an average over 20 executions. We list these parameters on the Table 4.1.

Table 4.1. The parameters of experiment.

Parameter Meanings Value

As Area size 500*500 meter²

N Number of nodes 2000 nodes

Cs Cell size 100*100 meter²

send the energy of transmistting 66 Jules Recive the energy of reception 39 Jules

H_D Handoff duration 10 times

SR the radio range of sensor node

40 meter Sε Sensor node's energy 6000 Jule

(38)

32

4.2. Simulation Results of Spiral Grid Routing Protocol

In this section, we evaluate the performance of our proposed scheme. Our experimental results are based on a full implementation of the previously described GGR and LBDD protocols. We focus on the comparison of load balance feature among our proposed SGR, GGR, and LBDD protocols.

We focus on the load balance feature of energy consumption. The performance metrics include the remaining energy after executing several queries and the nodes death rate after collecting several rounds of data.

For evaluating the remaining energy, we use the standard deviation among nodes to evaluate the performance of load balance. That is, the smaller the standard deviation of remaining energy is, the better performance in load balance has achieved.

The standard deviation Std is calculated by using the following equation:

𝑆𝑡𝑑 = (𝑥_𝑖 − 𝑥 )² 𝑁 − 1

Where x_iis the remaining energy for node i, 𝑥 is the average energy remaining of all sensors and N is the number of sensors in the field.

In this simulation, we assume that the sink node has 50 queries to be sent to the region of interest and each query will collect data from the region of interest for 20 times. The grid shifting will be launch after collecting 10 times of data. Figure 4.1 shows the simulation result. As we can see in Figure 4.1, our proposed SGR protocol outperforms than GGR and LBDD protocols in the balance of remaining energy. This means that our SGR protocol can distribute the work of energy consumption more evenly to other nodes in the grid cell. Especially, as the number of queries been

(39)

33

executed grows, the load balance feature in energy consumption shows that our SGR protocol performs much better than the other protocols.

Figure 4.1. Standard deviation of remaining energy.

For evaluating the node death rate, we are interesting in knowing the percentage of nodes which are out of energy after executing queries. If the percentage is low, then the lifetime of the network can be prolonged. In this simulation, we evaluate the node death rate by executing a query sent from sink. Once a sensor within the region of interest received the query message, it will send back the observation data to the sink periodically. For convenience, we call it a round after the sensor node has sent back 20 times of data. Note that we only compare our proposed SGR protocol with LBDD protocol. The reason is that, as the result from Figure 4.1, SGR and LBDD protocols are more load balance than GGR. Therefore, we focus on the comparison of these two protocols to understand which one can make the network’s lifetime longer.

Figure 8 shows the simulation result. In Figure 4.2, the node death rate of our SGR protocol is much lower than LBDD when the number of rounds is less than 100. Our SGR protocol will obtain the same node death rate as LBDD after 160 rounds. This means that our SGR protocol can effectively prolong the lifetime of the network.

0 200 400 600 800 1000 1200 1400 1600

10 20 30 40 50

Standard deviation of remaining energy (Jules)

Number of events been excuted

SGR LBDD GGR

(40)

34

Figure 4.2. Node death rate vs. number of rounds.

For evaluating the effectiveness of our grid shifting, we are interested in knowing the node death rate after performing several grid shifting processes. In this simulation we compare our SGR protocol only with the GGR protocol. This is because that our SGR and LBDD protocols are taking the similar mechanism to change the grid structure in which the cell size is fixed. These are different from the mechanism taken by GGR in which the cell size is changed in order to change the grid structure. Figure 9 shows the simulation result. In this simulation, we compare the node death rate of our SGR with three different cell sizes of GGR protocol, which are fixed size, large size and small size. The fixed size of GGR in the simulation means that the cell size is fixed during the simulation. The large size means that the cell sizes are incremented from initial cell size, say , to 1.1, 1.2, …, and 1.9, and then return to  and increment the cell size as previous again. The small size means that the cell sizes are decremented from initial cell size to 0.9, 0.8, …, and 0.1, and then return to  and decrements the cell size as previous again. In Figure 4.3, the node death rate of our SGR protocol is lower than all three cell sizes of GGR protocol.

Especially, the node death rate of the small size approach of GGR is much larger than

0 1 2 3 4 5 6 7 8 9 10

20 40 60 80 100 120 140 160

Node Death Rate(%)

Number of Rounds

SGR LBDD

(41)

35

the others after executing more than 60 grid shifting or changing processes. This is because that more nodes are involved in the data forwarding process when the grid size is small. Therefore, more nodes will be out of energy in this situation. Finally, our SGR protocol will obtain the smaller node death rate than GGRs, indicating that our grid shifting mechanism performs more effective than the grid changing mechanism in GGR protocol.

Figure 4.3. Node death rate vs. number of grid shifting.

0 10 20 30 40 50 60 70 80

10 20 30 40 50 60 70 80

Node Death Rate(%)

Number of Grid Shifting

SGR GGR(fixed) GGR(big) GGR(small)

(42)

36

4.3. Simulation Results of Steiner Trees Grid Routing Protocol

In this section, we evaluate the performance of our proposed scheme. Our experimental results are based on a full implementation of the previously described GGR protocols. We focus on the comparison of energy-efficient feature among our proposed STGR and GGR protocols.

Simulation Results

Figure 4.4. A routing path using GGR protocol.

Figure 4.5. A routing path using STGR protocol.

(43)

37

In this experiment, we are going to show that our proposed routing scheme is as robust as GGR protocol. Therefore, we have done two different situations: normal and abnormal (holes within the sensor field) cases. For the normal environment case, Figure 4.4 shows the routing path from a source node to a sink node in GGR protocol.

Figure 4.5 shows the routing path of our proposed STGR protocol with the same situation in Figure 4.4. As we can see in Figure 4.4, the source node in GGR protocol will forward data along either horizontal or vertical directions to the sink node while it will forward data along the hexagonal path first in our STGR protocol and then follow the horizontal or vertical directions to the sink node.

For the abnormal case, we assume that there are some holes within the sensor field. Figure 4.6 shows the routing path from a source node to a sink node in GGR protocol where there exists a hole between source node and sink node. Figure 4.7 shows the routing path of our proposed STGR protocol with the same situation in Figure 4.6. As we can see in Figure 4.7, our proposed STGR protocol can forward data the source node in GGR protocol will forward data along either horizontal or vertical directions to the sink node while it will forward data along the hexagonal path first in our SPGR protocol and then follow the horizontal or vertical directions to the sink node.

(44)

38

Figure 4.6. A routing path with a hole using GGR protocol.

Figure 4.7. A routing path with a hole using STGR protocol.

In this simulation, we are going to show that our proposed STGR scheme will achieve better performance in term of total energy consumption. Therefore, we have done two experiments based on two different criteria: the transmission hops in each round and the total energy consumption in each round. For these purposes, we assume

(45)

39

that the sink node has 20 queries to be sent to the region of interest and each query will collect data from the region of interest for 20 times.

For the transmission hops in each round, Figure 4.8 shows the simulation result. As we can see in Figure 4.8, the transmission hops in our STGR protocol is less than that in GGR protocol. This means that our STGR protocol can use fewer sensors to complete transmission task. In addition, according to the result of transmission hops, STGR can reach the data more quickly than GGR.

Figure 4.8. Transmission hops in each round.

For the total energy consumption in each round, Figure 4.9 shows the simulation result. As we can see in Figure 4.9, the total energy consumption in our STGR protocol is less than that in GGR protocol. This means that our STGR protocol can achieve a longer life time of the sensor network than that in GGR protocol.

0 5 10 15 20 25 30 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

T o ta l tra ns m is si o n ho ps

Rounds

SPGR GGR

(46)

40

Figure 4.9. Total energy consumption in each round.

0 10000 20000 30000 40000 50000 60000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Energy total consumption (J)

Rounds

SPGR GGR

(47)

41

Chapter 5 Conclusions

In this paper, we present a spiral grid routing protocol to achieve better performance in load balance in term of energy consumption and a Steiner Trees Grid routing protocol to achieve better performance in data deliverer and energy efficient.

Our approach of SGR tries to have every node in the grid cell to become the dissemination node once and STGR tries to let the transmission hops fewer. This will make the energy consumption more evenly among the sensors in the cell and reduced the total energy consumption. Simulation results show that our proposed scheme performs better than previous other schemes, such as GGR [9, 10] and LBDD [12], in terms of the remaining energy distribution of sensors and the node death rate after executing queries.

(48)

42

References

[1] Diane Cook, Sajal Das," Smart Environments: Technology, Protocols and Applications", 2004, ISBN: 978-0-471-54448-7

[2] Feng Zhao,Leonidas J. Guibas," Wireless sensor networks: an information processing approach", 2004, ISBN: 1558609148

[3] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, ―A Survey on Sensor Networks,‖ IEEE Communications Magazine, vol. 40, pp. 102–116, Aug.

2002. (From RideSharing Fault Tolerant Aggregation in Sensor Networks Using Corrective Actions)

[4] I. F. Akyildiz, W. SuCorresponding Author Contact Information, Y.

Sankarasubramaniam and E. Cayirci, " Wireless sensor networks: a survey", Computer Networks Volume 38, Issue 4, 15 March 2002, Pages 393-422.

[5] Chalermek Intanagonwiwat, Deborah Estrin, Ramesh Govindan and John Heidemann, " Impact of Network Density on Data Aggregation in Wireless Sensor Networks", 22nd IEEE International Conference on Distributed Computing Systems (ICDCS'02).

[6] Al-Karaki, J.N. , Kamal, A.E., " Routing techniques in wireless sensor networks:

a survey", Wireless Communications, IEEE, Dec. 2004,Volume: 11, Issue:

6,page(s): 6- 28

[7] C. Intanagonwiwat, R. Govindan, and D. Estrin, ―Directed diffusion: a scalable and robust communication paradigm for sensor networks,‖ Proceedings of the sixth annual international conference on Mobile computing and networking, pp.

56–67, 2000.

中 華 大 學 碩 士 論 文