中華大學碩士論文

(1)

中華大學碩士論文

題目：

在無線感測網路中具低延遲之省電路由演算法

系所別：資訊工程學系碩士班學號姓名：M09502020 黃禹傑指導教授：梁秋國博士

中華民國九十九年二月

(2)

Energy Efficient Routing Schemes for Achieving Low Latency in Wireless Sensor Networks

Student：Yu-Jie Huang

Advisor：Dr. Chiu-Kuo Liang

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

February, 2010

(3)

Abstract

In wireless sensor networks, it is an important task to periodically collect data from an area of interest for time-sensitive applications. The sensed data must be gathered and transmitted to a base station (BS) 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 minimize delay and offer good performance in energy efficiency, and long network lifetimes.

In this thesis, we propose two routing protocols, called Degree-Constrained Minimum Spanning Tree (DCMST) and Cluster-based Minimum Spanning Tree with Degree-Constrained (CMST-DC), to collect information efficiently. These two routing protocol are efficient in the ways that they ensure maximal utilization of network energy; they make the lifetime of the network longer, as well as they take much less time to complete a round. The experimental results show the better performance in terms of network latency, which is defined as the time duration for delivering data packets from all the nodes to the BS. It also shows that our protocols can perform better than several routing protocols in total transmission distance, which implies the less total energy consumption. Therefore, our proposed routing scheme gives a good compromise between energy efficiency and latency.

(4)

List of Figures

Figure 1.1 An example of a wireless sensor network. ... 2

Figure 1.2 Sensor nodes send data directly to the base station ... 3

Figure 1.3 Sensor nodes relay data to the base station ... 4

Figure 2.1 Sensor nodes are divided into 5 clusters using LEACH approach ... 8

Figure 2.2 Non-cluster head node sends data to its cluster head ... 9

Figure 2.3 Cluster heads send data directly to the base station ... 10

Figure 2.4 The furthest node from BS connects with its nearest neighbor ... 11

Figure 2.5 A chain formed by PEGASIS approach ... 11

Figure 2.6 Sensor nodes relay data to the leader node. ... 12

Figure 2.7 Chain formation phase of COSEN... 14

Figure 2.8 Chain formation phase of COSEN (5 chains are formed) ... 14

Figure 2.9 Cluster heads are formed a higher level chain ... 15

Figure 2.10 An example after clustering phase of GSEN ... 17

Figure 2.11 Cluster heads are formed a higher level chain ... 18

Figure 4.1 The construction process of the degree-constrained MST under our proposed scheme ... 26

Figure 4.2 The final tree and chain structures constructed by degree-constrained MST (a) and PEGASIS (b)... 26

Figure 4.3 The construction process of the degree-constrained MST while the degree of node d has gotten up to the degree limitation ... 27

Figure 4.4 The construction process of DCMST ... 30

Figure 4.5 Token passing approach ... 33

Figure 4.6 Transmission direction and sequence ... 34

Figure 4.7 The process of CMST-DC: A lower level trees is formed among each cluster ... 36

Figure 5.1 Delay time comparison for CMST-DC and COSEN ... 39

Figure 5.2 Total transmission distance comparison for DCMST and COSEN ... 40

Figure 5.3 Energy times delay with varying network size ... 40

Figure 5.4 Delay time comparison for CMST-DC and GSEN ... 42

Figure 5.5 Total transmission distance comparison for DCMST and GSEN ... 43

Figure 5.6 Energy times delay with varying network size ... 43

Figure 5.7 The routing paths constructed by GSEN ... 44

Figure 5.8 The routing paths constructed by CMST-DC ... 44

(6)

Chapter 1 Introduction

Wireless sensor networks (WSNs) are one of the most important technologies that will change the world [1] in that such networks can provide us with fine-granular observations about the physical world where we are living. A wireless sensor network is composed of a large number of tiny and inexpensive sensor nodes which can collect the information for further applications. Potential applications of wireless sensor networks span a wide range of areas; including disaster rescue, energy management, medical monitoring, logistics and inventory management, remote monitoring of seismic activities, environmental factors (e.g., air, water, soil, wind, chemicals), precision agriculture, factory instrumentation, and military reconnaissance, etc [2, 3].

With their capabilities for monitoring and control, the sensors are expected to be widely deployed. Such a network can provide a fine global picture through the collaboration of many sensors with each observing a coarse local view [4, 5].

In a wireless sensor networks, each sensor node is normally battery operated and equipped with the following modules [2, 6]:

 A sensing module which is capable of sensing some information (e.g., an acoustic, a seismic, a temperature, or a brightness sensor) or monitoring some entity in environment [7, 8].

 A digital unit for processing the signals from the sensors and can make some simple mathematical computation such as summation, average, or data sorting of collected information. This unit is also performing network protocol functions [6].

(7)

 A radio module for wireless communication which can send collected information to users wirelessly. With this communication module, sensor nodes can communicate with other nodes, and the users can send some queries to the sensor nodes to tell them what data the users need.

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 base station (BS), which is usually located at considerably far from the target field [9], for further analysis as shown in Figure 1.1.

User Base station

Sensors

Target region

Figure 1.1An example of a wireless sensor network

Periodical information collection from unreachable remote terrain and then transmit information to a remote base station is one of the targeted applications of sensor networks. But the energy restriction of battery operated sensor nodes certainly makes this task difficult and complicated because once they are deployed, the network can keep operating only until the battery power is sufficient. But it is almost impossible to replace the battery of the sensor nodes which are location-unknown once deployed over an inaccessible terrain. Therefore, it is desirable that the network protocols should take care of issues which are related to energy-efficiently, self-configuration, fault-tolerance, and delay etc [6, 10]. Especially, energy efficiency

(8)

In order to save energy, it is useful to fuse the sensed data into more meaningful information before transmitting to the base station. This is because that, as sensors are deployed densely, they might generate huge redundant and similar data from multiple nodes can be combined together so that the required number of transmission to the BS can be reduced. We also have to reduce the amount of long-distance data transmissions, since the energy consumption of data transmission scales with the transmission distance [12]. As shown in Figure 1.2, if all the sensor nodes send their data directly to the BS which is located far from the target region, it might cause a high energy consumption of each sensor node. The lifetime of this network will be shorter because each node runs out of its energy quickly.

Base station

Figure 1.2 Sensor nodes send data directly to the base station.

Comparatively, if a sensor node collects all the data relayed from other nodes, there is only one sensor transmitting its data to BS directly. As shown in Figure 1.3, this scheme which is called multi-hop routing reduces the energy consumption caused by the long transmission distance and prolongs the lifetime of the network effectively.

(9)

Base station

Figure 1.3 Sensor nodes relay data to the base station.

Another important issue in design consideration of a sensor network is data delivery time since in most cases data from sensor network are time critical as in the case of battle field or medical or security monitoring system. Such applications are required to receive the data from sensor nodes with minimum delay [6, 13].

In this thesis, we propose two hierarchical tree based protocols called

Degree-Constrained Minimum Spanning Tree (DCMST) and Cluster-based Minimum Spanning Tree with Degree-Constrained (CMST-DC) in which sensors are grouped into several clusters. In every cluster, a routing tree is constructed for data transmission. One sensor node is elected as a cluster head in every cluster based on the residual energy and this node remains as a cluster head for an optimal number of rounds. Among all cluster heads, a routing tree is also constructed. One cluster head is selected to be the leader of all cluster heads based on some measures at every round.

All nodes in a cluster send messages to the cluster head. Besides, all cluster heads send the information to the leader of cluster heads. The leader is the node that transmits the information to the BS. After several rounds, new group of cluster heads are selected. Due to the hierarchical tree structure, our protocol requires much lower time and energy as compared to other protocols of the wireless sensor networks for data collection.

(10)

The rest of the thesis is organized as follows. In Chapter 2, we give an overview of the related routing protocols. The network and communication models of our proposed protocol are discussed in Chapter 3. A detail description of our approach is presented in Chapter 4. Chapter 5 shows a comparative analysis and some simulation results. Finally, Chapter 6 presents a concluding remark.

(11)

Chapter 2 Related Works

Among various proposed routing protocols the hierarchical protocols LEACH [12], PEGASIS [8], COSEN [14] and GSEN [15] provide elegant solutions to minimize energy consumption and to lengthen network lifetime. We will describe these protocols in detail in the following sections.

2.1 Low-Energy Adaptive Clustering Hierarchy (LEACH)

Because BS is usually located considerably far from the target field [9], the amount of data transmitted from sensor nodes to BS should be reduced. On the other hand, sensor networks contain too much redundant and duplicate data for an end-user to process, therefore, automated methods of combining or aggregating the data into a small set of meaningful information is required [16, 17]. Based on these thought, the authors in [12] proposed a hierarchical cluster-based protocol called LEACH (Low-Energy Adaptive Clustering Hierarchy), in which the nodes organize themselves into local clusters with one node acting as the local base station or cluster-head (CH). Once the cluster-head has all the data from the nodes in its cluster, the cluster-head node aggregates the data and then transmits the compressed data to the base station. Since the base station is far away in the scenario we are examining, this is a high energy transmission. However, since there are only a few cluster-heads, this only affects a small number of nodes.

(12)

LEACH uses rotation of the cluster head in order to evenly distribute the energy consumption. The operation of LEACH is organized into rounds. Each round begins with a set-up phase followed by a steady-state phase.

During the setup phase, initially, when clusters are being created, each node decides whether or not to become a cluster-head for the current round. This decision is based on the suggested percentage of cluster heads for the network and the number of times the node has been a cluster-head so far. This decision is made by the node n choosing a random number between 0 and 1. If this random number is less than the threshold T(n), the node is a cluster head. This threshold T(n) is set as:

𝑇 𝑛 =

𝑃

1 − 𝑃 ∗ (𝑟 mod 1 𝑃)

, if 𝑛 ∈ 𝐺 0 , otherwise

(1)

where P is desired percentage to become a cluster-head (e.g., P = 0.05), r is the current round, and G is the set of nodes that have not been cluster-heads in the last

P

1 rounds. Using this threshold, each node will be a cluster-head at some point

within P

1 rounds. During round 0 (r = 0), each node has a probability P of becoming

a cluster-head. The nodes that are cluster-heads in round 0 cannot be cluster-heads for the next

P

1 rounds. Thus the probability that the remaining nodes are cluster-heads

must be increased, since there are fewer nodes that are eligible to become cluster-heads. After 1 1

P rounds, T = 1 for any nodes that have not yet been

cluster-heads, and after P

1 rounds, all nodes are once again eligible to become

cluster-heads.

Each node that has elected itself a cluster-head for the current round broadcasts

(13)

an advertisement message to the rest of the nodes. All cluster-heads transmit their advertisement using the same transmit energy and the non-cluster-head nodes must keep their receivers on during this phase of set-up to hear the advertisements of all the cluster-head nodes. After this, each non-cluster-head node decides the cluster to which it will belong for this round. This decision is based on the received signal strength (RSS) of the advertisement. The cluster-head advertisement heard with the largest signal strength is the cluster-head to whom the minimum amount of transmitted energy is needed for communication. After each node has decided to which cluster it belongs, it must inform the cluster-head node that it will be a member of the cluster.

Each node transmits this information back to the cluster-head.

As an example, Figure 2.1 shows that 5 cluster-head are chosen as a cluster-head, and non-cluster-head nodes have chosen its cluster-head.

Base station

Sensor (Non cluster head) Sensor (Cluster head)

Figure 2.1 Sensor nodes are divided into 5 clusters using LEACH approach.

The cluster-head node receives all the messages for nodes that would like to be included in the cluster. Based on the number of nodes in the cluster, the cluster-head node creates a TDMA schedule telling each node when it can transmit. This schedule is broadcast back to the nodes in the cluster.

(14)

Once the clusters are created and the TDMA schedule is fixed, data transmission can begin. Assuming nodes always have data to send, they send it during their allocated transmission time to the cluster head. This transmission uses a minimal amount of energy (Chosen based on the received strength of the cluster-head advertisement). Data aggregation is used after the cluster-head collects all the data from the nodes in its cluster. Once the previous work is done, the data is sent to BS by each cluster-head as shown in Figure 2.2 and Figure 2.3.

Initial data

Base station

Figure 2.2 Non-cluster head node sends data to its cluster head.

The authors of LEACH consider the scene in which BS is located far from target field, and the redundant data is high. A novel hierarchical clustering protocol is proposed, but it has some disadvantage. First, the cluster setup overload that needs to be carried by the network at every round. Secondly, there are still many long distance transmissions in the network.

(15)

Aggregated data

Base station

Figure 2.3 Cluster heads send data directly to the base station.

2.2 Power-Efficient Gathering in Sensor Information Systems (PEGASIS)

On the other hand, PEGASIS [8] forms a chain including all nodes in the network. A chain is formed by using a greedy algorithm so that each node can only communicate with its closest neighbor. In each round, a randomly selected node in the chain takes turn to transmit the aggregated information to the BS. PEGASIS saves energy by selecting only one leader node to transmit to the BS while other nodes transmit only to its local neighbor. However, it will cause excessive delay introduced by the distant node in a single chain.

To construct the chain, PEGASIS assume that all nodes have global knowledge of the network. To ensure that all nodes have close neighbors is difficult as this problem is similar to the traveling salesman problem, so the authors of PEGASIS use a greedy algorithm to create a routing path.This greedy approach is done before the first round of data transmission. The greedy algorithm starts with the furthest node from the BS. It begins with this node in order to make sure that nodes farther from the

(16)

BS have closer neighbors, as in the greedy algorithm the neighbor distances will increase gradually since nodes already on the chain cannot be revisited. Starting with the furthest node from BS, each node finds the closest neighbor node that has not joined in the chain in order. Until PEGASIS forms a chain including all nodes in the network. The furthest node from BS shown in Figure 2.4 finds its closest neighbor node, then that node continues to find next closest neighbor. The final chain forms by PEGASIS is shown in Figure 2.5.

Base station Sensor node

The furthest node from BS

Figure 2.4 The furthest node from BS connects with its nearest neighbor.

Sensor node

Base station

Figure 2.5 A chain formed by PEGASIS approach.

(17)

For gathering data in each round, each node receives data from one neighbor, fuses with its own data, and transmits to the other neighbor on the chain. Nodes take turns transmitting to the BS. It will choose one node as a leader to transmit to the BS in round. In a given round, the data transmission starts from the ends of the chain.

Then the leader node collects all data and sends it to the BS. It performs data fusion at every node except the end nodes in the chain to reduce the amount of data transmitted between sensors and BS. Each node will fuse its neighbor’s data with its own to generate a single packet of the same length and then transmit that to its other neighbor.

Figure 2.6 shows that data is relayed from the ends of the chain to the cluster-head.

Once the cluster-head receive the data from its neighbors, it sends this aggregated data which includes all the information of the entire network to BS.

Sensor node

The leader in current round Direction of data transmitting

Base station

Figure 2.6 Sensor nodes relay data to the leader node.

PEGASIS saves energy by selecting only one leader node to transmit to the BS while other nodes node transmit to only local neighbor. It enables a considerable amount of energy savings in the network, because PEGASIS doesn’t reconstruct a chain every round. However it will cause excessive delay introduced by the single chain for the distant node. When the network size becomes comparatively large, the

(18)

2.3 Chain Oriented Sensor Networks (COSEN)

In [14], Tabassum et al. proposed a hierarchical chain based protocol called COSEN which considers both energy consumption and delay. The main idea is to group sensors into two different level chains which are several lower level chains and one higher level chain to reduce the high network delay introduced by a long chain for the distant node of PEGASIS. In each lower chain which are formed to include all the sensor nodes, one sensor is elected as a chain-leader based on the residual energy, and one higher level leader is selected among all lower level leaders at every round. All nodes in a lower level chain relay its data to the lower level leader, and all lower level leaders send the data to the higher level leader. Then the higher level leader will transmit the information including the data from all other sensors to the BS. After several rounds, new lower level leaders are selected.

The procedure of COSEN operates in two phases - chain formation phase followed by data transmission phase. In the chain formation phase, several lower level chains are formed to include all sensor nodes. Each chain is of fixed length which is called CL (chain length). Simulation results of COSEN show that a chain containing around 15-20% of the sensors gives the optimal results. Chain formation starts from the node at the furthest position from BS using a greedy algorithm. A node in a chain selects the nearest live node that is not already inserted into any other chain and adds it to the chain. If the chain length exceeds the predefined length, CL, new chain formation starts. This way of chain formation process continues until all the live nodes are grouped into chains. Positions of the nodes are obtained by methods based on triangulation [10, 18, 19], where nodes approximate their positions using signal strengths from a few known points. To save this extra energy consumption by the

(19)

extra negotiation, this process just takes place in COSEN once at the beginning of network setup. Figure 2.7 and Figure 2.8 shows the process of chain formation phase which starts from the furthest node from BS. Since the length of the first chain exceeds CL (We set CL as 5 for this example.), the next node is added to the new chain as shown in Figure 2.7. The chain formation process stops after all the live nodes are grouped into the chain (Figure 2.8).

Sensor node

The furthest node from BS The link between two nodes

Base station

Figure 2.7 Chain formation phase of COSEN

Sensor node

The link between two nodes

Base station

Figure 2.8Chain formation phase of COSEN (5 chains are formed)

(20)

Unlike PEGASIS, where leaders are chosen randomly in every round, COSEN selects leaders for every chain based on the remaining energy in each sensor of the chain after a predefined number of rounds. Once the leaders are selected, a higher level leader is selected among the leaders using a greedy algorithm. The higher level leader is the only node that sends information to the BS. The higher level leader selection is based on 3 criteria: 1. distance from BS, 2. energy remains in the node, and 3. not selected as higher level leader for the last N/CL (or N/CL + 1) rounds.

COSEN tries to let the nodes closer to BS take turn to be higher level leader more frequently than the nodes those are far from the BS.

Figure 2.9 shows a higher level chain is constructed among cluster-head.

Sensor

The link between two nodes Cluster-head

The link between two CH

Base station

Figure 2.9 Cluster heads are formed a higher level chain.

After the formation of the chain and selection of leaders, sensors start data collection operation. This should be noted that chain formation phase does not precede data collection phase always. It precedes data collection phase whenever it is necessary to reconstruct new chains. The data transmission scheme in COSEN is alike the scheme in PEGASIS. Each end node in a chain starts by transmitting to the next

(21)

node. The node in the next position receives the data and fuses this data with its own and transmits it to the next node. This is how data propagate from the furthest node in the chain to the chain-leader.

2.4 Group-based Sensor Networks (GSEN)

In [15], Tabassum et al. proposed another hierarchical chain based protocol called GSEN which also considers both energy consumption and delay. The main idea of GSEN is to partition sensors into several chains to reduce the high network delay introduced by a long chain for the distant node of PEGASIS, too. The major difference between the approach of GSEN and COSEN is that GSEN groups clusters before forming chains and COSEN groups clusters during chain formation. GSEN also operates in two phases - chain formation phase followed by data transmission phase. In the chain formation phase, several chains are formed with one leader in each cluster. Tabassum et al. borrow the clustering method from LEACH and limit the number of groups limited to five in a 100-node network. Each sensor chooses a number between 0 and 1. If the number is less than a threshold T(n) shown in equation (1), the node broadcast itself as the leader. In GSEN, they assumes the group formation takes place only once at the beginning of the network set-up phase. So equation (1) reduces to T = p, when r = 0, in this case. Non-leader nodes receiving the broadcast from leaders will decide to which leader it will join depending on the received signal strength by itself. Nodes then inform the leader by sending an acknowledgement signal. After collecting all the signals, each leader node creates a chain using the same greedy algorithm in COSEN to connect all the nodes in the cluster.

(22)

This node selects the nearest live node that is in the same cluster and adds it to the chain. This way of chain formation process continues until all the live nodes within the same cluster are grouped into chains. Unlike LEACH where cluster set-up takes place at every round, GSEN prefers to reconstruct groups after every R number of rounds. Group leader keeps on changing at every round in a random order in every group. All the leaders will also connected as a higher level chain and one leader is selected as the higher level leader just like the process in COSEN. As shown in Figure 2.10, there are five clusters being grouped. Then five lower level chains and 1 higher level chain constructed among the lower level leaders are formed in Figure 2.11 in the set-up phase.

Base station

Figure 2.10 An example after clustering phase of GSEN

(23)

Base station

Figure 2.11 Cluster heads are formed a higher level chain.

In data transmission phase, the authors use the same algorithm in COSEN to collect the data sensed by sensors. Each sensor node sends the sensed information to its lower level leader. Then, each lower level leader sends the information towards higher level leader. At the end of a round, higher level leader sends the information to BS. Although COSEN and GSEN outperforms in energy consumption and network delay than both LEACH and PEGASIS, it is still not good enough. In our work, we propose a hierarchical tree based protocol which will get better performance than them both in energy consumption and network delay. Our proposal is completely self-organized which means that the cluster setup and routing path calculation are all carried out by the sensor nodes themselves without involving BS and energy efficient with very limited delay.

(24)

Chapter 3 Network and Communication Models

For the sake of uniformity, the models we used are very similar as used in LEACH, PEGASIS, COSEN, and GSEN. In this chapter, we are going to explain the network and communication models. The assumptions of proposed network model we consider will be discussed in section 3.1, and the communication model will be explained in detail, too. We also make some analysis about the energy consumption.

3.1 Network Model

In our proposed protocol, we consider the following network model assumptions:

 Data are transmitted periodically from the sensor network to the remote BS and delay critical, especially in the cases of battle field, medical or security monitoring system. In such applications, collecting data from all sensor nodes is urgent, and it is necessary to receive the data from sensor nodes as soon as possible [6, 13].

 The BS is located far from the sensor network and fixed. It is important as in some cases such as battle field and habitat monitoring, since such applications need to run unattended, diagnostic and monitoring tools are essential [20].

 All sensor nodes are homogeneous, energy constrained and immobile.

 All sensor nodes have the ability to aggregate the data received from other sensors with its own data. In addition to helping avoid information overload, data aggregation, also known as data fusion, can combine several unreliable

(25)

data measurements to produce a more accurate signal by enhancing the common signal and reducing the uncorrelated noise. The classification performed on the aggregated data might be performed by a human operator or automatically.

 All sensors nodes have the ability to transmit data to any other sensor node or directly to the BS [21, 22].

 The radio of sensors has power control and can expend the minimum required energy to reach the intended recipients. The radios can be turned off to avoid receiving unintended transmissions.

3.2 Communication Model

We use the same radio model as used in LEACH, PEGASIS, COSEN and GSEN.

In this model, a radio dissipates E_elec=50nJ/bit to run the transmitter or receiver circuitry and Eamp=1pJ/bit/m² for the transmitter amplifier. Therefore the transmission cost to transmit a k-bit message to a distance d is given by the equation (2):

𝐸_𝑇𝑥 𝑘, 𝑑 = 𝑘𝐸_{𝑒𝑙𝑒𝑐} + 𝑘𝐸_𝑓𝑠𝑑² , 𝑑 < 𝑑₀

𝑘𝐸_{𝑒𝑙𝑒𝑐} + 𝑘𝐸_𝑚𝑝𝑑⁴, 𝑑 ≥ 𝑑₀ 2 , where Efs=10pJ/bit/m² and Emp=0.0013pJ/bit/m⁴ are required amplifier energy for free space and multipath model respectively to send data at an acceptable signal to noise ratio (SNR). The threshold distance d can be given by equation (3): ₀

𝑑₀ = 𝐸𝑓𝑠 𝐸_𝑚𝑝 (3) The receiving cost is given by equation (4):

𝐸_𝑟𝑥 = 𝑘𝐸_{𝑒𝑙𝑒𝑐} (4) It is assumed that the radio channel is symmetric so that to transmit a message in both directions between two nodes requires the same energy at a given SNR. Since receiving and transmitting are both high cost operation, the number of receiving and

(26)

transmitting should be minimal. Also, to achieve the energy saving, we should let only one node transmit to BS and all nodes transmit only to local neighbor. On the other hand, there is also a cost of 5nJ/bit/message for data fusion [6, 12].

(27)

Chapter 4 Our Proposed Schemes

In this chapter, the operation of our proposed Degree-Constrained Minimum Spanning Tree (DCMST) and Cluster based Minimal Spanning Tree with Degree-Constrained (CMST-DC) routing protocol will be discussed in detail. It can be divided into two phases: cluster formation phase followed by data transmission phase.

The basic idea of our approach is to construct a minimum spanning tree to replace the chain in each cluster in order to reduce the energy consumption from data transmission. We also consider about the data collection delay so that it doesn’t scale fast even in a large network.

4.1 Preliminary

Since all nodes in a cluster need to send data to the cluster head, we use the idea of minimum spanning tree (MST for short) to shorten the total transmission distance to reduce the energy consumption from data transmission. This means that we construct a minimum spanning tree of the nodes in a cluster. However, it is possible that a node in the computed MST will be connected with many other nodes. In such case, this node needs to receive and fuse more data collected from its neighbors than other nodes. Since receiving and fusing data consume the energy, this may cause the node to die earlier than other nodes.

In order to avoid the situation that a node will be connected with many other

(28)

nodes, we introduce the degree constraint to each tree node. For a connected, edge-weighted, and undirected graph, the degree of a node is the number of edges in which it participates. Given a positive integer d, the Degree-Constrained MST is a spanning tree with the smallest weight among all possible spanning trees which contain no nodes of degree greater than d. This problem is NP-hard, because the Hamiltonian Path problem (Problem ND1 in Garey and Johnson [23]), which is

NP-complete, is a special case of d-MST with d = 2 and all edge weights identical.

For the cases that the vertices are points in the plane, and edge weights are the Euclidean distances between these points, Monma and Suri [24] showed that there always exists a MST with degree no more than five. Paradimitriou and Vazirani [25]

proved that finding a d-MST in the Euclidean plane is NP-hard when d = 3, and conjectured that it remains NP-hard when d = 4. It should be noticed that the chain in PEGASIS, COSEN or GSEN is a tree (not necessary a MST) with no nodes of degree greater than 2.

In our proposal, sensor nodes are deployed randomly in the target field. Several clusters are formed with one leader in each cluster. All nodes within a cluster form a tree among them by using a greedy algorithm. Within a cluster, one node is selected as cluster head by some criteria. After that a higher level tree is formed including all lower level cluster head nodes. Among these cluster heads only one chosen head node sends information to BS.

In the next section, we present the detailed implementation of the distributed protocol for constructing the degree constrained minimum spanning trees.

(29)

4.2 Construction of a Degree-constrained Minimum Spanning Tree in Wireless Sensor Networks

Sensor nodes are deployed randomly in the target field. The tree formation algorithm starts with a starting node from BS. This first starting node will be explained in the next section. The procedure of construction of a MST in a wireless sensor network is described as the following steps:

Step 1: A node which is a starting node first set itself as a tree node (A node which is already in a tree) and will broadcast a Find-Nearest-Neighbor (FNN) message with the largest transmission range to find the nearest live node.

Step 2: Once a node receives the FNN message, it sets a backoff timer of t1 seconds, where t1 is uniformly distributed in some range and depends on the received signal strength (RSS) of the received message. The more signal strength of the received message, the less t1 will be.

Step 3: When the timer expires, the node sends back an acknowledgement (ACK) message with its node identification using minimum required the energy to reach the starting node. If a node hears other ACK messages before its t1 timer expires, it cancels its timer.

Step 4: Each tree node received the ACK message will set a backoff timer of t2 seconds. Again, t2 is relative to the received signal strength of the received ACK message.

Step 5: When t2 expires, the node sends a confirmation (CFM) message with node ID to inform the node sent the ACK message to be the next starting node and the link between them can be constructed.

(30)

The above process will be repeated for finding next nearest live neighbor node until no live neighbors exist.

As an example, Figure 4.1 shows the process of constructing a minimum spanning tree with degree-constrained for a network of five live nodes in a cluster. At the beginning, node a is the starting node and it broadcasts a FNN message. Node b sends back an ACK message to node a since it is the nearest neighbor of node a (Figure 4.1(b)). After node a reply a CFM message to node b, the link between nodes a and b is established and node b will be the next starting node. After node b sends a

FNN message, node c will reply the ACK message and the link between nodes b and c will be established (Figure 4.1(c)). After node c sends a FNN message, only nodes d and e can reply since nodes a and b are already in the tree. In this case, node d will reply an ACK message earlier than node e since it is closer to node c. At this time, nodes a, b and c can hear the ACK message from node d and they will set a backoff timer for themselves of a period of time according to the received signal strength from the ACK message sent by node d, respectively. In this case, the timer of node b will expire first and it sends a CFM message to node d. Therefore, node d will connect to node b, not to node c (Figure 4.1(d)). The procedure will continue until all live neighbor nodes are found. Figure 4.2(a) shows the final tree structure constructed by our procedure for the example in Figure 4.1, compared with the chain structure constructed as we showed in Figure 4.2(b).

(31)

e a

c b d

e a

c b d

e a

c b d

e a

c b d

(a) (b)

(c) (d)

Figure 4.1 The construction process of the degree-constrained MST under our proposed scheme.

e a

c b d

e a

c b d

(a) (b)

Figure 4.2 The final tree and chain structures constructed by degree-constrained MST (a) and PEGASIS (b).

Our procedure for finding the degree-constrained MST is simple. First, we let each node has a counter to indicate the current degree of itself. When a nearest live neighbor node replies with an ACK message to the starting node, some tree nodes will receive this message. Before a tree node can set a backoff timer, it has to first check if

(32)

its current degree counter is less than the degree limitation, say d. If its current degree counter equals to d, then it does not need to set a backoff timer to response the ACK message. When a link is established, the current degree counter will be incremented by one for both nodes connected by the link. When the above procedure stops, a degree constrained spanning tree with no nodes of degree greater than d is obtained for a set of sensor nodes.

As shown in Figure 4.3(a), we let the degree limitation be 3. Node d is current starting node and it broadcast a FNN message. Node e will reply the ACK message, and some tree nodes will received this ACK message. Since the degree of node b equals to the degree limitation, it doesn’t set its t backoff timer. In this case, the ₂ timer of node d will expire first and it sends a CFM message to node e. Therefore, node e will connect to node d, not to node b which is the nearest tree node to node e.

After this link is established, the current degree counter of node d and node e will be incremented by one.

f

a

c b d

(a) e

f

a

c b d

(b) e

1 1

3 3

1

1 1 2

1 0

0 0

.

Figure 4.3 The construction process of the degree-constrained MST while the degree of node d has gotten up to the degree limitation.

(33)

4.3 DCMST: Degree-Constrained Minimum Spanning Tree

The operation of our proposed Degree-Constrained Minimal Spanning Tree (DCMST) routing protocol can be divided by two phases: tree formation phase followed by data transmission phase. In the following sub-sections we discuss each of them in details.

4.3.1 Cluster Formation Phase

The network establishment begins with the formation of trees. Several lower level trees are formed to include all the sensor nodes. All nodes connected by a tree are treated as within a cluster and the tree among them is constructed by using the algorithm described above. Within a cluster, one node is selected as cluster head based on the remaining energy in each sensor of the tree after a predefined number of rounds.

After that a higher level tree is formed including all lower level cluster head nodes.

Among these cluster heads only one chosen head node sends information to BS.

The tree formation algorithm is the same as that we described in section 4.2, and each sensor has a cluster size counter to calculate how many nodes are in the cluster.

The process starts with the furthest node from BS. This furthest node is treated as a tree node and a starting node of the first tree. We can simply let sink broadcast a message to find the furthest node. Once a node receives this message, it will set a backoff timer. The more signal strength of the received message, the shorter timer will be. When the timer expires, the node broadcasts a FNN message to start the construction of degree-constrained minimum spanning tree. If the other nodes receive the FNN messages before its timer expires, it cancels its timer. Then the process of construction of a degree-constrained MST will be repeated for finding next nearest

(34)

cluster size (CS). This CS is fixed, and we set it as 20% of the number of sensor nodes which is picked by LEACH, COSEN and GSEN in our simulations. The last starting node will broadcast a FNC (Find-Next-Cluster) message to inform the nearest live neighbor node to be the starting node of the next cluster and the process proceeds for finding the nodes in next cluster. Note that the messages broadcasted by a node are including the cluster identification to ensure that two nodes in different clusters won’t be linked.

As an example, Figure 4.4 shows the process of construction of DCMST for a network. To simplify this example, we just deploy six sensors and set the Cs as 3.

Figure 4.4(a) shows that node c is current starting node and the CS of its cluster will exceed 3 if the next nearest neighbor joins its cluster. Therefore, node c broadcast a FNC message to inform the nearest live neighbor node to be the starting node and cluster head of the next cluster. In this case, node d will become the next starting node to continue the process (Figure 4.4(b)), and node e is the next nearest neighbor of node d. Note that, even if the distance between node e and node b is shorter than the distance between node e and node d, node e doesn’t connect with node b since they are not in the same cluster (Figure 4.4(c)). The final result of this example is shown in Figure 4.4(d)

(35)

f

a

c b d

(a) e

f

a

c b d

(b) e

f

a

c b d

(c) e

f

a

c b d

(d) e

Figure 4.4 The construction process of DCMST

For an N-node network where each tree contains CS nodes, the number of trees is N/CS (if N mod Cs = 0) or N/CS+1 (if N mod Cs ≠ 0). After all the nodes are included in a tree, next target is to choose a leader node in a tree. We select leaders for each tree based on the remaining energy in each sensor of the tree and change the leaders after several rounds. We change the leader after k rounds where k  N/CS, since the simulation results of COSEN perform better while k N/CS. Once the lower level leaders are selected, a higher level tree is constructed among the lower level leaders using the same algorithm. A higher level leader will be selected in every round and it is the only node to send data to BS. For the higher level leader selection the criteria we consider are the distance from nodes to BS and energy remained in the node. We first normalize a notation dw (distance-weight) from 0 to 1 depends on the distance from nodes to BS. The longer distance from nodes to BS, the less dw will be.

Again, we normalize the other notation ew (energy-weight) from 0 to 1 depends on

(36)

energy remained in the node. The more energy remained in the node, the bigger ew will be. Then we use a weight w to be the criteria to choose one higher level leader from the nodes in the higher level tree. At the beginning of a round, all nodes within the higher level tree will broadcast a message containing the weight w, and one node that has the highest weight will be the higher level cluster head in this round. The weight w can be given by the equation (5):

𝑤 =𝑑𝑤 + 𝑒𝑤

2 (5) We consider that the tree formation takes place whenever every 10% nodes of the initial deployed sensors die. This is due to the optimal total transmission distance of tree and for efficient load balanced.

4.3.2 Data Transmission Phase

After the formation of tree and selection of cluster heads, sensors start data collection operation. At the beginning of the data collection and transmission phase each cluster head accumulates data from the member nodes within its cluster.

However, there may have a long delay during the data collection and transmission phase if the cluster size is large. Data collection delay is defined as the time duration (in time slots) for delivering data packets from all the nodes to the BS. It is especially important for many time-critical applications, such as battlefield surveillance and fire detection. Therefore, it is important to minimize the delay for data collection. In order to achieve the minimum delay for data collection, we adapt the similar time schedule mechanism as in SHORT [26]. The idea for minimizing the delay in SHORT is to generate as many communication pairs (parallel packet transmissions) as possible in each time slot. For a network of n sensors and one cluster head, at most



^{n 2}^/ ⁱ



(37)

communication pairs can be generated in the i slot. So the minimum delay for _th completing packet transmission during the data collection phase is (



^log2n



¹) slots.

In the following, we describe our approach for data collection within a cluster. If a node is elected as a cluster head it needs to take the responsibility to calculate the time schedule for data collection. Thus, the first step for a cluster head is to collect the information from all nodes in order to know the network topology within its cluster.

To collect the information, a cluster head sends a token packet toward each end node of the tree to travel around all nodes in the tree. Each node in the tree marks its node ID onto the token and sends the marked token to its other children nodes which have not received the token yet. The intermediate node continues to send a token to its children nodes till the token is traveling around all its children nodes and returning back to itself. After that, it sends the token back to its parent node.

As shown in Figure 4.5, if node c is elected as a head node, it sends a token toward node a which then sends the token back to node c. Since node c has other neighbors, it sends the token to node b which then sends the token back to node c.

Finally, node c sends the token toward node d to collect the information. However, since node d has other neighbors, it sends the token toward node f which then sends the token toward node g. After node g receives the token, it sends the token back to its parent node f and the token will be sent toward node d. Node d sends the token toward node e which then sends back to node d. After that since node d has no other unsent children nodes, it sends the token to its parent node, c. The process continues till the token reaches to the head node. After the token is returned to the head node, all node IDs and their topology information are collected. Then, the head node can execute the scheduling algorithm proposed in SHORT to generate all communication pairs.

(38)

a

f

c b

d

e

g

Direction of Tokens

Figure 4.5 Token passing approach.

In Figure 4.6, we will generate three time slots to collect data, say S1 = {(g, f), (e, d), (b, c)}, S2 = {(f, d), (a, c)} and S3 = {(d, c)}, since only the leaf nodes and the

nodes which already collect all the information of children nodes can send its data to the parent node. The transmission direction for each communication pair is also identified. This is how data propagate from the end nodes to the cluster head in a tree.

Every cluster head then transmits the information to the next cluster head in the higher level tree using the same fashion. Whenever the higher level cluster head gets all the information, it transmits the information to BS after data fusion. It should be noticed that once the time schedule is obtained, it is no need to re-compute the schedule unless a new cluster head is elected. Therefore, the computing cost is small.

In case any sensor node’s remaining energy is lower than 5% of the initial energy, it broadcast a message including a new schedule for its children nodes and parent node. The children nodes of this senile node will simply bypass the senile node and continue operation.

(39)

a

f

c b

d

(a)

e

g

a

f

c b

d

(b)

e

g

a

f

c b

d

(c)

e

g

Figure 4.6 Transmission direction and sequence

4.4 CMST-DC: Cluster-based Minimal Spanning Tree with Degree-Constrained

The operation of the other of our proposed routing protocols, Cluster-based Minimum Spanning Tree with Degree-Constrained (CMST-DC), can also be divided by tree formation phase and data transmission phase. In the following sub-sections we discuss each of them in details.

4.4.1 Cluster Formation Phase

Unlike the algorithm we described in section 4.3.1, the network establishment begins with the formation of clusters. Several clusters are formed with one leader in each cluster. All nodes within a cluster then form a tree among them by using our algorithm. Within a cluster, one node is selected as cluster head by some criteria after

(40)

k rounds. After that a higher level tree is formed including all cluster head nodes.

Among these cluster heads only one chosen head node sends information to BS.

At the beginning of cluster formation only, we adapt the same algorithm as LEACH for the selection of cluster head node. We prefer the idea of LEACH where each sensor chooses a number between 0 and 1. If the number is less than a threshold

 

n

T shown in equation (1), the node broadcasts itself as the leader. Non-leader nodes receiving the broadcast decide by themselves to which leader it will join depending on the signal strength and inform the corresponding leader by sending an acknowledgement. After collecting all the acknowledgement signals, each leader node initiates tree formation starting from itself connecting all the nodes in the cluster.

Once the tree is constructed by using the algorithm we described in section 4.2, one higher level leader is selected among lower level leaders by using the algorithm we discussed in section 4.3.1 where one node that has the highest weight will be the higher level cluster head. The main difference between our approach and LEACH is that, unlike LEACH where cluster set-up takes place at every round, our approach prefers to re-build clusters after a certain number of rounds. Thus, once the clusters are formed they remain fixed until next cluster formation phase is needed. Figure 4.7(a) shows the network topology after we use the clustering algorithm of LEACH.

Figure 4.7(b) shows the final tree structure constructed by our procedure

(41)

(a) (b)

Figure 4.7 The process of CMST-DC: A lower level trees is formed among each cluster.

Unlike the algorithm we discussed in section 4.3.1, where trees are reconstructed every 10% of initial sensors’ death, we now reconstruct CMST-DC every 20 rounds.

The higher level leader is still reselected every round.

4.4.2 Data Transmission Phase

The same as the process of DCMST protocol which is discussed in section 4.3.2, sensors start data collection operation after the formation of degree-constrained minimum spanning tree and selection of cluster heads in CMST-DC. We use the same algorithm we described in section 4.3.2 for data collection to minimize the delay time.

The approach is completely suited to both of our proposed protocols, DCMST and CMST-DC.

(42)

Chapter 5 Performance Evaluation

In this chapter, we evaluate the performance of our proposed routing protocol. In section 5.1 we first describe our simulation environment and experiment models.

Then we explain the performance metrics we used to compare with other protocols in section 5.2. The simulation results of our proposal compared with COSEN and GSEN are shown in section 5.3 and 5.4 respectively.

5.1 Simulation Environment

The proposed DCMST and CMST-DC schemes were evaluated by extensive computer simulations and compared with other protocols, using the C# programming language on .net platform. In our simulation, we consider different network density from 100 nodes to 500 nodes with 5% of the nodes being cluster heads. Each sensor’s location is represented by Cartesian coordinates and placed in a non-overlapping random position. All the nodes have the same initial energy of 0.25 Joules and have the ability of adjusting their transmission power to minimize interference and ensure error-free packet transmissions. A node is considered dead when its energy becomes zero and excluded for the consecutive rounds. The length of each data packet is 4000 bits. The communication model we used is the same as that we described in section 3.2. The simulation results are discussed in section 5.3 and 5.4, each representing an averaged summary over 100 runs.

(43)

5.2 Performance Metrics

To evaluate the performance of a routing protocol, we select three general metrics which are extensively used in many previous papers.

 Total transmission distance: This is defined as the sum of the transmission distance of all sensor nodes in a round. Since the energy consumption of data transmission scales with the transmission distance [5], the shorter total transmission distance, the less energy will be consumed from data transmission.

Then we can prolong the network lifetime.

 Delay Time: This is defined as the total unit time we need to collect all data produced by all sensor nodes and send it to the BS in a round. Since in most applications, data from sensor network are time critical as in the case of battle field or medical or security monitoring system, this is a very important of design consideration of a sensor network.

 Product of the energy consumed and the average end to end delay: In some routing applications, the total transmission distance and the average delay time metrics may not produce clear conclusions. Therefore, the product of the energy consumed and the average end-to-end delay metric can be selected as a general metric. An optimal routing protocol should have the minimum value for the product of the energy consumed and the average end to end delay [28].

5.3 Simulation Results of DCMST

In this section, the performance of our proposed routing protocol is evaluated.

We compared DCMST with COSEN in our simulation. All nodes are placed randomly in a place of 50 meter * 50 meter. The base station is located at (0, 100). The

(44)

simulation results are shown as the following figures.

Figure 5.1 shows the simulation result of average delay time we need to collect all data produced by all sensor nodes and send it to the BS. Our DCMST approach performs better than COSEN in delay time about 9% while we put 100 sensor nodes and 21% while we put 500 sensor nodes in the network. It shows that our protocol always has less delay to deliver information to the BS from distant nodes as compared to COSEN. The higher network density, the better simulation result of delay performs.

This simulation result shows that our DCMST approach can be used well in many time critical applications in different network density.

Figure 5.1 Delay time comparison for DCMST and COSEN.

Figure 5.2 shows the simulation result of total transmission distance. As shown in Figure 5.2, we found DCMST performs better than COSEN. As the network density increased, the total transmission distance of both scheme increased, too. But DCMST always performs about 15% better than COSEN. The improvement is achieved by using the tree-based structure instead of chain-based structure. Since the

Delay Time

0 20 40 60 80 100

100 200 300 400 500

Number of Nodes

Delay (Unit Time)

DCMST COSEN

(45)

energy consumption of data transmission scales with the transmission distance, we can also say that total energy consumption of DCMST is less than COSEN.

Figure 5.2 Total transmission distance comparison for DCMST and COSEN.

Total Transmission Distance

0 200 400 600 800 1000 1200

100 200 300 400 500

Number of Nodes

Total Transmission Distance(m)

DCMST COSEN

Energy * Delay

0 2 4 6 8 10 12

100 200 300 400 500

Number of Nodes Energy * Delay (Joule*Unit Time) DCMST COSEN

(46)

In many routing applications, the average end-to-end delay metric may not produce clear conclusions although it is clear from Figure 5.2 that the end to end delay is lower for DCMST. Therefore, the product of the energy consumed and the average end-to-end delay metric can be selected as a general metric. The experiment result regarding of the product of energy consumed and time delay is shown in Figure 5.3. It can be clearly seen from Figure 5.3 that DCMST performs much better than COSEN.

5.4 Simulation Results of CMST-DC

In this section, we compared CMST-DC which is clustering before MST construction with GSEN in our simulation. All nodes are placed randomly in a place of 100 meter * 100 meter. The base station is located at (50, 200). The simulation results are shown as the following figures.

The simulation result of average end-to-end delay for both routing scheme is compared in Figure 5.4. Our CMST-DC approach performs better than GSEN in delay time about 45%. Notice that, the delay time of GSEN is higher than COSEN, because the cluster size of each cluster is different in GSEN. The delay time will get higher if one of the clusters of GSEN has more sensor nodes. It shows that our protocol always has less delay to deliver information to the BS from distant nodes as compared to COSEN. The higher network density, the better simulation result of delay performs.

(47)

Figure 5.4 Delay time comparison for CMST-DC and GSEN.

Figure 5.5 shows the simulation result of total transmission distance. In addition, we also add LEACH in this simulation to compare it with GSEN and CMST-DC. As shown in Figure 5.5, we found CMST-DC performs much better than GSEN and LEACH. Since each sensor nodes in LEACH send sensed information directly to its cluster head, the total transmission distance of LEACH is the longest. The total transmission distance of CMST-DC is the shortest because of the tree-based structure.

Since the energy consumption of data transmission scales with the transmission distance, we can also say that total energy consumption of CMST-DC is less than GSEN and LEACH.

Delay Time

0 10 20 30 40 50 60 70 80

100 200 300 400 500

Number of Nodes

Delay (Unit Time)

CMST-DC GSEN

(48)

Figure 5.5 Total transmission distance comparison for DCMST and GSEN.

Figure 5.6 Energy times delay with varying network size.

Total Transmission Distance

0 1000 2000 3000 4000 5000 6000 7000

100 200 300 400 500

Number of Nodes

Total Transmission Distance(m)

CMST-DC GSEN LEACH

Energy * Delay

0 2 4 6 8 10 12 14 16 18

100 200 300 400 500

Number of Nodes Energy * Delay (Joule*Unit Time) CMST-DC GSEN

(49)

The simulation result of the product of the energy consumed and the average end-to-end delay is shown in Figure 5.6. It can be clearly seen from Figure 5.6 that CMST-DC performs much better than GSEN.

Cluster head Sensor node Delay time: 30

Total transmission distance: 1178.42

Figure 5.7 The routing paths constructed by GSEN

Cluster head Sensor node Delay time: 15

Total transmission distance: 943.87

Figure 5.8 The routing paths constructed by CMST-DC

(50)

Figure 5.7 and Figure 5.8 shows the results of routing paths by using GSEN and our CMST-DC approach on a sensor nodes distribution of a run from the experiment with 100 nodes in the network respectively. Five clusters are formed and we colored each cluster different colors. As shown in Figure 5.7, there are 5 clusters formed as chains using the scheme of GSEN, and a higher level chain is formed through the cluster heads. It can be seen that our protocol performs better both in the delay time and total transmission distance than GSEN.

(51)

Chapter 6 Conclusion

In this thesis, we propose two new hierarchical tree-based routing protocols for efficiently collecting data in a sensor network. For designing the protocols, we consider how to reduce the delay of data collection and how to shorten the total transmission distance in order to reduce the energy consumption. We also consider the product of the energy consumed and the delay time. This is a general metric in sensor networks. Our protocols show better performance than COSEN [11] and GSEN [6] in terms of network delay, total transmission distance and the product of energy consumed and the delay time.

(52)

Reference

[1] 10 emerging technologies that will change the world. Technology Review, vol. 106, no.1, pp. 33-49, Feb 2003.

[2] M. Khan, G. Pandurangan, and B. Bhargava, Energy-Efficient Routing Schemes for Wireless Sensor Networks, Technical Report CSD TR 03-013, Dept. of Computer Science, Purdue University, 2003.

[3] N. Bulusu, D. Estrin, L. Girod, and J. Heidemann, “Scalable Coordination for Wireless Sensor Networks: Self-configuring Localization Systems,” in Proceedings of the Sixth International Symposium on Communication Theory and Applications (ISCTA),July 2001.

[4] D. Li, K.D. Wong, Y.H. Hu, and A.M. Sayeed, Detection, Classification and Tracking of Targets, IEEE Signal Processing Magazine, Vol. 19, pp. 17-29, March 2002.

[5] S. Madden, M.J. Franklin, J.M. Hellerstein and W. Hong, TAG: a tiny aggregation service for ad-hoc sensor networks, OSDI, December 2002.

[6] W. R. Heinzelman, A. Chandrakasan, and H.Balakrishnan, “An Application-Specific Protocol Architecture for Wireless Microsensor Networks”, IEEE Trans. Wireless Commun., vol. 1, no. 4, Oct. 2002, pp. 660-670.

[7] Kuei-Ping Shih; Sheng-Shih Wang; Pao-Hwa Yang; Chau-Chieh Chang,

"CollECT: Collaborative Event deteCtion and Tracking in Wireless Heterogeneous Sensor Networks," Computers and Communications, 2006. ISCC '06. Proceedings. 11th IEEE Symposium on , vol., no., pp. 935-940, 26-29 June 2006

[8] Lindsey, S.; Raghavendra, C.S., "PEGASIS: Power-efficient gathering in sensor

中 華 大 學 碩 士 論 文