732 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 2, APRIL 2014

Transcription

1 732 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 2, APRIL 2014 A General Self-Organized Tree-Based Energy-Balance Routing Protocol for Wireless Sensor Network Zhao Han, Jie Wu, Member, IEEE, Jie Zhang, Liefeng Liu, and Kaiyun Tian Abstract Wireless sensor network (WSN) is a system composed of a large number of low-cost micro-sensors. This network is used to collect and send various kinds of messages to a base station (BS). WSN consists of low-cost nodes with limited battery power, and the battery replacement is not easy for WSN with thousands of physically embedded nodes, which means energy efficient routing protocol should be employed to offer a long-life work time. To achieve the aim, we need not only to minimize total energy consumption but also to balance WSN load. Researchers have proposed many protocols such as LEACH, HEED, PEGASIS, TBC and PEDAP. In this paper, we propose a General Self-Organized Tree-Based Energy-Balance routing protocol (GSTEB) which builds a routing tree using a process where, for each round, BS assigns a root node and broadcasts this selection to all sensor nodes. Subsequently, each node selects its parent by considering only itself and its neighbors information, thus making GSTEB a dynamic protocol. Simulation results show that GSTEB has a better performance than other protocols in balancing energy consumption, thus prolonging the lifetime of WSN. Index Terms Energy-balance, network lifetime, routing protocol, self-organized, wireless sensor network. I. INTRODUCTION WITH the advances in Micro-Electro-Mechanical Systems (MEMS)-based sensor technology, low-power digital electronics and low-power wireless communication [1], [2], [3], it is now possible to produce wireless sensor nodes in quantity at low cost. Although these sensor nodes are not as powerful or accurate as their expensive macro-sensor counterparts, we are able to build a high quality, fault-tolerant sensor network by making thousands of sensor nodes work together. Through the cooperation of wireless sensor nodes, WSN collects large amounts of information and sends them to the Base Station (BS). WSN has a wide range of potential applications [10], including military surveillance, disaster prediction, environment monitoring, etc. Thus it has become one Manuscript received June 30, 2012; revised February 09, 2013; accepted February 25, Date of publication April 02, 2014; date of current version April 10, This work was supported in part by Major National Science and Technology Special Program of China under Grant 2011ZX The authors are with Department of Modern Physics and State Key Laboratory of Particle Detection and Electronics, University of Science and Technology of China, Hefei , Anhui, China ( anzhao7@mail.ustc.edu.cn; wujie@ustc.edu.cn; zhjjhz@mail.ustc.edu.cn; llf@mail.ustc.edu.cn; kytian@mail.ustc.edu.cn). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TNS of the most important research fields and has aroused extensive research interest. Generally, wireless sensor nodes are deployed randomly and densely in a target region, especially where the physical environment is so harsh that the macro-sensor counterparts cannot be deployed. After deployment, the network cannot work properly unless there is sufficient battery power. In general, WSN may produce quite a substantial amount of data, so if data fusion could be used, the throughput could be reduced [4]. Because sensor nodes are deployed densely, WSN might generate redundant data from multiple nodes, and the redundant data can be combined to reduce transmission. Many well-known protocols implement data fusion, but almost all of them assume that the length of the message transmitted by each relay node should be constant, i.e., each node transmits the same volume of data no matter how much data it receives from its child nodes [12]. PEGASIS [7], PEDAP [8] and TBC [17] are typical protocols based on this assumption and perform far better than LEACH [4], [5] and HEED [6] in this case. However, there are quite a few applications in which the length of the message transmitted by a parent node depends not only on the length of its own, but also on the lengths of the messages received from its child nodes. In an extreme case, the relay node should transmit the length of the message which is the sum of its own sensed data and the received data from its children [12]. Energy consumption of a node is due to either useful or wasteful operations. The useful operations include transmitting or receiving data messages, and processing requests. On the other hand, the wasteful consumption is due to the operation of constructing routing tree, overhearing, retransmitting because of harsh environment, dealing with redundant broadcast overhead messages, and idle listening to the media. In this paper, we propose a General Self-Organized Treebased Energy Balance routing protocol (GSTEB). We consider a situation in which the network collects information periodically from a terrain where each node continually senses the environment and sends the data back to BS [11]. Normally there are two definitions for network lifetime: a) The time from the start of the network operation to the death of the first node in the network [13]. b) The time from the start of the network operation to the death of the last node in the network. In this paper, we adopt the first definition. Moreover, we consider two extreme cases in data fusion: Case (1) The data between any sensor nodes can be totally fused. Each node transmits the same volume of IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.

2 HAN et al.: GENERAL SELF-ORGANIZED TREE-BASED ENERGY-BALANCE 733 data no matter how much data it receives from its children. Case (2) The data can t be fused. The length of message transmitted by each relay node is the sum of its own sensed data and received data from its children. The remainder of the paper is organized as follows: Section II reviews related works. The network and radio models of our proposal are discussed in Section III. Section IV describes the architectures and details of GSTEB. In Section V we present our simulations in contrast to the simulations of other known protocols. Finally, Section VI concludes the paper. II. RELATED WORKS A main task of WSN is to periodically collect information of the interested area and transmit the information to BS. A simple approach to fulfilling this task is that each sensor node transmits data directly to BS. However, when BS is located far away from the target area, the sensor nodes will die quickly due to much energy consumption. On the other hand, since the distances between each node and BS are different, direct transmission leads to unbalanced energy consumption. To solve these problems, many protocols have been proposed. Of the protocols proposed, hierarchical protocols such as LEACH, HEED, PEGASIS, TBC and PEDAP can achieve satisfactory solutions. In LEACH [4], [5], for the entire network, nodes selected according to a fraction p from all sensor nodes are chosen to serve as cluster heads (CHs), where p is a design parameter. The operations of LEACH are divided into several rounds. Each round includes a setup phase and a steady-state phase. During the setup phase, each node will decide whether to become a CH or not accordingtoapredefined criterion. After CHs are chosen, each of other nodes will select its own CH and join the cluster according to the power of many received broadcast messages. Each node will choose the nearest CH. During the steady-state phase, CHs fuse the data received from their cluster members and send the fuseddatatobsbysingle-hopcommunication.leachuses randomization to rotate CHs for each round in order to evenly distribute the energy consumption. So LEACH can reduce the amount of data directly transmitted to BS and balance WSN load, thus achieving a factor of 8 times improvement compared with direct transmission. In [6], the authors proposed a hybrid, energy-efficient, distributed clustering algorithm (HEED). HEED is an improvement of LEACH on the manner of CH choosing. In each round, HEED selects CHs according to the residual energy of each node and a secondary parameter such as nodes proximity to their neighbors or nodes degrees. By iterations and competition, HEED ensures only one CH within a certain range, so uniform CHs distribution is achieved across the network. Compared with LEACH, HEED effectively prolongs network lifetime and is suitable for situations such as where each node has different initial energy. For Case1, LEACH and HEED greatly reduce total energy consumption. However, LEACH and HEED consume energy heavily in the head nodes, which makes the head nodes die quickly. S. Lindsey et al. proposed an algorithm related to LEACH, and it is called PEGASIS [7]. PEGASIS is a nearly optimal power efficient protocol which uses GREEDY algorithm to make all the sensor nodes in the network form a chain. In PEGASIS, the (i mod N)th node is chosen to be a leader and the leader is the only one which needs to communicate with BS in round i. N is the total amount of nodes. Data is collected by starting from both endpoints of the chain, and transmitted along the chain, and fused each time it transmits from one node to the next until it reaches the leader. So PEGASIS sharply reduces the total amount of data for long-distance transmission and achieves a better performance than LEACH by 100% to 300% in terms of network lifetime. Tree-Based Clustering (TBC) [17] is also an improved protocolofleach.itformsseveralclustersinthesamewayas LEACH, and each cluster has a cluster-head (CH). The nodes within a cluster construct a routing tree where the cluster-head is the root of it. For tree configuration, the cluster-head uses the distance information between the member nodes and itself. Each node is location-aware, it can estimate the distance between the root and itself. Every cluster is divided into some levels. The distance of a node to the root is the basis for determining its level in the cluster. The cluster-head is at level-0(root) and a node in level will choose the node in and nearest to itself as its parent node. Data transfer simultaneously happens between the nodes in two neighboring levels, and each node fuses the received data and transmits it to its parent. TBC is an excellent protocol in which each node records the information of its neighbors and builds topography through computing, which is similar to GSTEB. But some cluster-heads in the network consume more energy than other nodes when BS is located far away. PEDAP [8] is a tree-based routing protocol that makes all the nodes form a minimum spanning tree, which costs minimum energy for data transmitting. It also has another version called PEDAP-PA which slightly increases energy for data transmitting but balances energy consumption per node. PEDAP has the same network assumptions as PEGASIS and uses data fusion. However, both PEDAP and PEDAP-PA are protocols that need BS to build the topography which will cause a large amount of energy waste. This is because if the network needs BS to build the topography, BS should send a lot of information to the sensor nodes, including what time is the Time Division Multiple Access (TDMA) slot, who are their child nodes and who are their parent nodes. This kind of information exchanging will cause a lot of energy to be wasted or will cause a long delay. III. NETWORK AND RADIO MODEL In our work, we assume that the system model has the following properties: sensor nodes are randomly distributed in the square field and there is only one BS deployed far away from the area. Sensor nodes are stationary and energy constrained. Once deployed, they will keep operating until their energy is exhausted. BS is stationary, but BS is not energy constrained. All sensor nodes have power control capabilities, each node can change the power level and communicate with BS directly.

3 734 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 2, APRIL 2014 Sensor nodes are location-aware. A sensor node can get its location information through other mechanisms such as GPS or position algorithms. Each node has its unique identifier (ID). In order to compare the performance of GSTEB with the performance of other protocols, we use different radio models for Case1 and Case2. Since Case1 is the typical characteristic of PEGASIS [7] and PEDAP [8], we use the same radio model as in PEGASIS analysis [4], [5], [7], [8], which makes it easier to verify our simulation results and compare the performance of GSTEB with that of PEGASIS. In this model, the energy dissipation of the radio caused by running the transmitter or receiver circuitry equals nj/bit and the energy dissipation of the radio caused by running the transmit amplifier equals pj/bit/m. It is also assumed that a path loss due to free-space propagation model is used. The energy consumption of transmitting a k-bit packet to a distance d and receiving that packet is: Transmitting Receiving For Case2, we use the same model as in HEED [6], which makes it easier to verify the simulation results and compare the performance of GSTEB with that of HEED. This model uses both the free-space propagation model and two-ray ground propagation model to approximate the path loss due to wireless channel transmission. When, the free-space propagation model is employed and uses pj/bit/m for the transmitter amplifier. When, the two-ray ground propagation model which leads to a path loss is employed and uses pj/bit/m for the transmitter amplifier. is a threshold transmission distance which can be computed by: square root. For both cases, the medium is assumed to be symmetric so that the energy required for transmitting a message from node A to node B or from node B to node A is the same. IV. GENERAL SELF-ORGANIZED TREE-BASED ENERGY-BALANCE ROUTING PROTOCOL The main aim of GSTEB is to achieve a longer network lifetime for different applications. In each round, BS assigns a root node and broadcasts its ID and its coordinates to all sensor nodes. Then the network computes the path either by transmitting the path information from BS to sensor nodes or by having the same tree structure being dynamically and individually built by each node. For both cases, GSTEB can change the root and reconstruct the routing tree with short delay and low energy consumption. Therefore a better balanced load is achieved compared with the protocols mentioned in Section II. The operation of GSTEB is divided into Initial Phase, Tree Constructing Phase, Self-Organized Data Collecting and Transmitting Phase, and Information Exchanging Phase. A. Initial Phase In Initial Phase, the network parameters are initialized. Initial Phase is divided into three steps. Step 1: When Initial Phase begins, BS broadcasts a packet to all the nodes to inform them of beginning time, the length of time slot and the number of nodes N. When all the nodes receive the packet, they will compute their own energy-level (EL) using function: EL is a parameter for load balance, and it is an estimated energy value rather than a true one and only used in Case2, i is the ID of each node, and is a constant which reflects the minimum energy unit and can be changed depending on our demands. Step 2: Each node sends its packet in a circle with a certain radius during its own time slot after Step 1. For example, in the time slot, the node whose ID is i will send out its packet. This packet contains a preamble and the information such as coordinates and EL of node i. All the other nodes during this time slot will monitor the channel, and if some of them are the neighbors of node i, they can receive this packet and record the information of node i in memory. The nodes which are not in the range of can t monitor the preamble in this time slot, so they can know they are not the neighbors of node i and will turn off their radios, then switch to sleep mode to save energy. After all nodes send their information, each node records a table in their memory which contains the information of all its neighbors. Step 3: Each node sends a packet which contains all its neighbors information during its own time slot when Step 2 is over. Then its neighbors can receive this packet and record the information in memory. The length of time slots in Steps 2 and 3 is predefined, thus when time is up, each node has sent its information before Initial Phase ended. After Initial Phase, each node records two tables in memory which contain the information of all its neighbors and its neighbors neighbors. These two tables are defined as Table I and Table II. Each node works according to them in the following phases. Initial Phase is a significant preparation for the next phases. After Initial Phase, GSTEB operates in rounds. For GSTEB and all other protocols mentioned, the round has the same meaning. In a round, the routing tree may need to be rebuilt and each sensor node generates a DATA_PAK that needs to be sent to BS. When BS receives the data of all sensor nodes, a round ended. Round is not a real time measurement unit, but it reflects the ability for transmitting the collected data for sensors, so round is a suitable time measurement unit for WSN lifetime. Each round contains three phases, including Tree Constructing Phase, Self-Organized Data Collecting and Transmitting Phase, and Information Exchanging Phase. B. Tree Constructing Phase Within each round, GSTEB performs the following steps to build a routing tree. Between Case1 and Case2 there are some differences in the steps of routing tree constructing: Step 1: BS assigns a node as root and broadcasts root ID and root coordinates to all sensor nodes.

4 HAN et al.: GENERAL SELF-ORGANIZED TREE-BASED ENERGY-BALANCE 735 TABLE I NETWORK LIFETIMES OF DIFFERENT SCHEMES Step For Case1, because data fusion technique is implemented, only one node which communicates directly with BS can transmit all the data with the same length as its own, which results in much less energy consumption. In order to balance the network load for Case1, in each round, a node with the largest residual energy is chosen as root. The root collects the data of all sensors and transmits the fused data to BS over long distance. For Case2, because data can t be fused, it will not save the energy for data transmitting by making fewer nodes communicate directly with BS. When one of the sensor nodes collects all the data and sends it to BS, it would deplete its energy quickly. In this case BS always assigns itself as root. 2: Each node tries to select a parent in its neighbors using EL and coordinates which are recorded in Table I. The selection criteria are: 1) For both Case1 and Case2, for a sensor node, the distance between its parent node and the root should be shorter than that between itself and the root. 2) For Case1, each node chooses a neighbor that satisfies criterion 1 and is the nearest to itself as its parent. And if the node can t find a neighbor which satisfies criterion 1, it selects the root as its parent. 3) For Case2, the process of Tree Constructing Phase can be regarded as an iterative algorithm. Besides criterion 1, for a sensor node, only the nodes with the largest EL of all its neighbors and itself can act as relay nodes. If the sensor node itself has the largest EL, it can also be considered to be an imaginary relay node. Choosing the parent node from all the relay nodes is based on energy consumptions. Any of these consumptions is the sum of consumption from the sensor node to a relay node and that from the relay node to BS. The relay node which causes minimum consumption will be chosen as the parent node. It is true that this relay node should choose its parent node in the same way. So a path with minimum consumption is found by iterations. And by using EL, GSTEB chooses the nodes with more residual energy to transmit Fig. 1. Topography generated if each node chooses the nearest as parent. data for long distance. If the sensor node cannot find a suitable parent node, it will transmit its data directly to BS. Step 3: Because every node chooses the parent from its neighbors and every node records its neighbors neighbors information in Table II, each node can know all its neighbors parent nodes by computing, and it can also know all its child nodes. If a node has no child node, it defines itself as a leaf node, from which the data transmitting begins. As discussed above, for Case1, because each packet sent to the parent nodes will be fused, the minimum energy consumption can be achieved if each node chooses the node nearest to it. But if all nodes choose their nearest neighbors, the network may not be able to build a tree. Fig. 1 shows a network of 100 nodes in this situation. We can find that some clusters are formed, but they cannot connect with others. Thus in GSTEB, we use criterion 1 in Case1 to limit the search direction. By this approach, a routing tree is constructed and some nodes still have the possibility of connecting to their nearest neighbors. For Case2, criterion 1 should also be obeyed and this criterion helps to save the energy for data transmitting to a certain extent. To build a routing tree, for Case1, each node follows the steps. But for Case2, we use BS to compute the topography. Even though we can fulfill this work without the control of BS, a large amount of energy is wasted in the next phase. C. Self-Organized Data Collecting and Transmitting Phase After the routing tree is constructed, each sensor node collects information to generate a DATA_PKT which needs to be transmitted to BS. For Case1, TDMA and Frequency Hopping Spread Spectrum (FHSS) are both applied. This phase is divided into several TDMA time slots. In a time slot, only the leaf nodes try to send their DATA_PKTs. After a node receives all the data from its child nodes, this node itself serves as a leaf node and tries to send the fused data in the next time slot.

5 736 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 2, APRIL 2014 Fig. 2. Process of a time slot in Self-Organized Data Collecting and Transmitting Phase for Case1. stands for leaf nodes, which need to transmit data in the time slot. stands for other nodes, they are parent nodes. All the leaf nodes try to send their DATA_PKTs in a time slot which divide all other nodes into three situations (see the following paragraphs). This figure shows what all the nodes do for the three situations. Each node knows the ID of its parent node. In each time slot, in order to reduce communication interference, we apply FHSS in which each child node communicates with its parent node using the frequency hopping sequence determined by its parent node ID. Each TDMA time slot is divided into three segments as follows (see Fig. 2). Segment1:Thefirst segment is used to check if there is communication interference for a parent node. In this segment, each leaf node sends a beacon which contains its ID to its parent node at the same time. Three situations may occur and they divide all the parent nodes into three kinds. For the first situation, if no leaf node needs to transmit data to the parent node in this time slot, it receives nothing. For the second situation, if more than one leaf node needs to transmit data to the parent node, it receives an incorrect beacon. For the third situation, if only one leaf node needs to transmit data to the parent node, it receives a correct beacon. The operation of the second segment depends on the three situations.

6 HAN et al.: GENERAL SELF-ORGANIZED TREE-BASED ENERGY-BALANCE 737 Segment 2: During the second segment, the leaf nodes which can transmit their data are confirmed. For the first situation, the parent node turns to sleep mode until next time slot starts. For the second situation, the parent node sends a control packet to all its child nodes. This control packet chooses one of its child nodes to transmit data in the next segment. For the third situation, the parent node sends a control packet to this leaf node. This control packet tells this leaf node to transmit data in the next segment. Segment 3: The permitted leaf nodes send their data to their parent nodes, while other leaf nodes turn to sleep mode. The process in one time slot is shown in Fig. 2. For Case2, each node chooses its parent by considering not the distance but the total energy consumption. In our simulation results, we will show that there may be many leaf nodes sharing one parent node in one time slot. If all the leaf nodes try to transmit their data at the same time, the data messages sent to the same parent node may interfere with each other. By applying Frequency Division Multiple Access (FDMA) or Code Division Multiple Access (CDMA), the schedule generated under competition is able to avoid collisions. However, the accompanying massive control packets will cause a large amount of energy to be wasted. By using the control of BS, the energy waste can be reduced and thus the process may be much simpler. At the beginning of each round, the operation is also divided into several time slots. In the time slot, the node whose ID is i turns on its radio and receives the message from BS. BS uses the same approach to construct the routing tree in each round, and then BS tells sensor nodes when to send or receive the data. In each TDMA time slot, the nodes work in turns defined by BS. When BS receives all the data, the network will start the next phase. D. Information Exchanging Phase For Case1, since each node needs to generate and transmit a DATA_PKT in each round, it may exhaust its energy and die. The dying of any sensor node can influence the topography. So the nodes that are going to die need to inform others. The process is also divided into time slots. In each time slot, the nodes whose energy is going to be exhausted will compute a random delay which makes only one node broadcast in this time slot. When the delay is ended, these nodes are trying to broadcast a packet to the whole network. While all other nodes are monitoring the channel, they will receive this packet and perform an ID check. Then they modify their tables. If no such packet is received in the time slot, the network will start the next round. For Case2, BS can collect the initial EL and coordinates information of all the sensor nodes in Initial Phase. For each round, BS builds the routing tree and the schedule of the network by using the EL and coordinates information. Once the routing tree is built, the energy consumption of each sensor node in this round can be calculated by BS, thus the information needed for calculating the topology for the next round can be known in advance. However, because WSN may be deployed in an unfriendly environment, the actual EL of each sensor node may be different from the EL calculated by BS. To cope with this problem, each sensor node calculates its EL and detects its actual residual energy in each round. We define the calculated EL as EL1 and the actual EL as EL2. When the two Fig. 3. Routing tree generated by GSTEB for 100 nodes randomly deployed in asquareforcase1. ELs of a sensor node are different, the sensor node generates an error flag and packs the information of actual residual energy into DATA_PKT, which needs to be sent to BS. When this DATA_PKT is received, BS will get the actual residual energy of this sensor node and use it to calculate the topology in the next round. V. COMPARATIVE ANALYSIS AND SIMULATION RESULTS A MATLAB simulation of GSTEB is done for both Case1 and Case2 to evaluate the performance. For Case1, we first compare GSTEB with PEGASIS and use the same network model as PEGASIS. We generate a randomly distributed 100 to 400 nodes network of square area m m with BS located at (50 m, 175 m) and use DATA_PKT length of 2000 bits and CTRL_PKT length of 100 bits. We let each node have 0.25 J initial energy. Fig. 3 and Fig. 4 show the routing tree generated by GSTEB and PEGASIS for exactly the same 100 node topology. In Fig. 3 the triangle is root node and in Fig. 4 the triangle is head node and the rectangle is tail node. As seen, the routing tree generated by GSTEB is better. Since PEGASIS uses GREEDY algorithm to form a chain, long links may exist between parent nodes and child nodes, which will cause an unbalanced load. As for GSTEB, each node tends to choose the nearest neighbor. Even though it may not be able to select the global optimal solution, long links are avoided. Fig. 5 shows that the time when the first node dies changes within a range from 100 nodes to 400 nodes in the network. Because network works in rounds, the time measurement unit is round but it is not a real time measurement unit. We can find that GSTEB performs much better than PEGASIS and prolongs network lifetime by about 100% to 300% in Fig. 5. To compare GSTEB with TBC, we use the same parameters as TBC as shown in [17]. BS is located at (50 m, 120 m) and the length of a DATA_PAK is 4000 bits. We compare the performance of GSTEB with the existing simulation results of TBC.

7 738 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 2, APRIL 2014 Fig. 4. Routing tree generated by PEGASIS for 100 nodes randomly deployed in a square for Case1. Fig. 5. For Case1, we compare the time when first node dies for GSTEB and PEGASIS for the number of nodes from 100 to 400 in the square area, the network working in round makes Round be the time measurement unit. Table I lists the lifetime of the sensor network in terms of the round when a node begins to die and in terms of the round when all the nodes die. In Table I, the simulation results are taken from [17] except GSTEB. We can find that GSTEB performs better than all other protocols. Both GSTEB and TBC are protocols which require that each node records the information of its neighbors and they use similar approach to transmit data. For GSTEB, each node needs to record the information of its neighbors neighbors, so it needs more memory. But GSTEB can achieve a better performance in energy saving, because each node has more opportunities to choose the nearest neighbor as the parent. Moreover, there are several cluster-heads that need to communicate with BS in TBC. The simulation results of TBC show that it performs well since BS is located near the sensor field. If BS is located far away, the energy consumption of TBC increases quickly because of the cluster-heads. And GSTEB will outperform it. When the network lifetime is defined as the time from the start of the network operation to the death of the last node in the network, PEDAP [8] achieves an optimal solution. It is because PEDAP builds a minimum spanning tree, and it consumes the least energy for data transmitting in each round. In this situation, we make some slight changes to GSTEB. We need not consider the load balance, but in each round the total energy consumed by the network should be the minimum. GSTEB chooses the node nearest to BS as the root like PEDAP, and then GSTEB builds a routing tree and the topology remains unchanged until the root dies. Once the root dies, the network reconstructs a new routing tree in the same way. We compare the improved GSTEB with PEDAP. Fig. 6 shows the routing tree generated by PEDAP of the same topology as Fig. 3 and Fig. 4. Simulation results show that by using the same model as PEGASIS, in each round, the data transmitting in the routing tree generated by the improved GSTEB consumes less than 0.5% extra energy compared with the routing tree generated by PEDAP. PEDAP is a protocol that requires BS to build the topography. To achieve that, BS should send a broadcast packet which contains all the topology information and the schedule information to all sensor nodes or BS informs the sensor nodes one by one when the topography needs to be rebuilt. If BS sends a packet which contains all the information, each node has to receive the whole packet to get its own information. This will cause a large amount of energy to be wasted. On the other hand, with BS informing the nodes one by one, the network can save energy, but causes a long delay. As compared, in the improved GSTEB, when the root node is chosen, all other nodes compute and find their own parents by themselves in parallel without any information exchange, so the energy consumption can be neglected. Once a node dies, other nodes remove its information from their tables and rebuild the routing tree. As a result, the improved GSTEB only consumes a little more energy for data transmitting than PEDAP, but the energy for building the routing tree is greatly reduced, so the improved GSTEB works almost as well as an optimal solution. For Case2, we change the initial energy of each node to 2 J and J, we also generate a randomly distributed 100 to 400 nodes network of square area m mwithbs located at (50 m, 175 m) and use DATA_PKT length of 2000 bits and CTRL_PKT length of 100 bits. We compare GSTEB with HEED in this case. Fig. 7 shows that the time when the first node dies changes within a range from 100 nodes to 400 nodes in the network. In Fig. 7 the time measurement unit is also round, but it is not a real time measurement unit. Fig. 8 shows the routing tree generated by GSTEB for 100 nodes deployed in a square area in this case. Clearly, GSTEB performs far better than HEED and prolongs the network lifetime by more than 100%. In Fig. 9 and Fig. 10, we can find that almost all nodes have the same residual energy for GSTEB when the first node dies. For HEED, the nodes nearer to BS have much more residual energy than the nodes farther from BS. This is a typical load imbalance. HEED and GSTEB both aim to balance the network load for Case2. HEED selects CHs by considering the residual energy,

8 HAN et al.: GENERAL SELF-ORGANIZED TREE-BASED ENERGY-BALANCE 739 Fig. 6. Routing tree generated by PEDAP for 100 nodes randomly deployed in asquareforcase1. Fig. 8. Routing tree generated by GSTEB for 100 nodes randomly deployed in asquareforcase2. Fig. 9. Residual energy of the network when first node dies for GSTEB. Fig. 7. For Case1, we compare the time when first node dies for GSTEB and HEED for the number of nodes from 100 to 400 in the square area, the network working in round makes Round be the time measurement unit. but only the local energy balance is attained. It is because CHs cannot dynamically decide to communicate with BS by single-hop or multi-hops. If they choose the single-hop, CHs farther from BS consume more energy. On the other hand, even if they choose the multi-hops, CHs nearer to BS have to transmit more data and consume more energy. For GSTEB, each node dynamically decides to communicate with BS directly or through others. All the nodes try to find neighbors with higher EL as parent nodes. In each round, the nodes with higher EL consume more energy, which leads to a whole energy balance as shown in Fig. 9. Both in Case1 and Case2, GSTEB is suitable for different initial energy circumstances. For Case1, nodes with the most Fig. 10. Residual energy of the network when first node dies for HEED. energy will be chosen as the root and communicate with BS far away. In Case2, nodes with more energy will be selected as

9 740 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 2, APRIL 2014 parent more often. Both cases will cause the nodes with more initial energy to transmit or receive more data and consume more energy. We calculate the waste of energy and the delay of GSTEB for both cases. For Case1, since the topography is built by self-organizing, each node is able to choose its parent simultaneously, it consumes little energy and this only causes a short delay in Tree Constructing Phase. In Self-Organized Data Collecting and Transmitting Phase, each leaf node needs to send a beacon which is received by its parent node. The beacon is a short frame which contains only the ID of the leaf node. We assume that the length of the beacon is 20 bits. Compared with the energy consumption caused by DATA_PKT (2000 bits), the energy waste caused by the beacon is only 1% of it. On the other hand, because all the leaf nodes transmit data at the same time, the delay of data transmitting is j-based TDMA time slots where j is the maximal level of the tree. The delay is shorter than that of PEGASIS and PEDAP. For Case2, BS should send a packet which contains the topology information and the schedule information to the leaf nodes one by one, which will lead to extra energy consumption. The CTRL_PKT is longer than the beacon in Case1 since it contains more information. We assume that the length of CTRL_PKT sent by BS to each node is 200 bits. Each sensor node needs to receive a control packet in a round. Because the data can t be fused, it has to be transmitted for a long distance, which causes significantly large energy consumption. Moreover, the relay nodes need to transmit more data, which causes a long delay. As a result, even though the process of the information exchanging between BS and sensor nodes leads to a waste of energy and a delay, the ratio between wasteful consumption and useful consumption is acceptable, and this method makes the process much simpler, so it is suitable for Case2. VI. CONCLUSIONS In this work, we introduce GSTEB. Two definitions of network lifetime and two extreme cases of data fusion are proposed. The simulations show that when the data collected by sensors is strongly correlative, GSTEB outperforms LEACH, PE- GASIS, TREEPSI [9] and TBC. Because GSTEB is a self-organized protocol, it only consumes a small amount of energy in each round to change the topography for the purpose of balancing the energy consumption. All the leaf nodes can transmit data in the same TDMA time slot so that the transmitting delay is short. When lifetime is defined as the time from the start of the network operation to the death of the first node in the network, GSTEB prolongs the lifetime by 100% to 300% compared with PEGASIS. In some cases, we are more interested in the lifetime of the last node in the network. Some slight changes are made to make the performance of GSTEB similar to that of PEDAP. So GSTEB is nearly the optimal solution in Case1. When the data collected by sensors cannot be fused, GSTEB offers another simple approach to balancing the network load. In fact, it is difficult to distribute the load evenly on all nodes in such a case. Even though GSTEB needs BS to compute the topography, which leads to an increase in energy waste and a longer delay, this kind of energy waste and longer delay are acceptable when compared with the energy consumption and the time delay for data transmitting. Simulation results show that when lifetime is defined as the time from the start of the network operation to the death of the first node in the network, GSTEB prolongs the lifetime of the network by more than 100% compared with HEED. REFERENCES [1] K. Akkaya and M. Younis, A survey of routing protocols in wireless sensor networks, Elsevier Ad Hoc Network J., vol. 3/3, pp , [2] I. F. Akyildiz et al., Wireless sensor networks: A survey, Computer Netw., vol. 38, pp , Mar [3] Sohrabi et al., Protocols for self-organization of a wireless sensor network, IEEE Personal Commun., vol. 7, no. 5, pp , Oct [4] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, Energyefficient communication protocols for wireless microsensor networks, in Proc. 33rd Hawaii Int. Conf. System Sci., Jan. 2000, pp [5] W. B. Heinzelman, A. Chandrakasan, and H. Balakrishanan, An application-specific protocol architechture for wireless microsensor networks, IEEE Trans. Wireless Commun, vol. 1, no. 4, pp , Oct [6] O. Younis and S. Fahmy, HEED: A hybrid, energy-efficient, distributed clustering approach for ad hoc sensor networks, IEEE Trans. Mobile Computing, vol. 3, no. 4, pp , [7] S. Lindsey and C. Raghavendra, Pegasis: Power-efficient gathering in sensor information systems, in Proc. IEEE Aerospace Conf., 2002, vol. 3, pp [8] H.O.TanandI.Korpeoglu, Powerefficient data gathering and aggregation in wireless sensor networks, SIGMOD Rec., vol.32,no.4,pp , [9] S. S. Satapathy and N. Sarma, TREEPSI: Tree based energy efficient protocol for sensor information, in Proc. IFIP Int. Conf., Apr. 2006, pp [10] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, Wireless sensor networks: A survey, Computer Netw.s, vol. 38, no. 4, pp , [11] R. Szewczyk, J. Polastre, A. Mainwaring, and D. Culler, Lessons from sensor network expedition, in Proc. 1st European Workshop on Wireless Sensor Networks EWSN 04, Germany, Jan , [12] W. Liang and Y. Liu, Online data gathering for maximizing network lifetime in sensor networks, IEEE Trans Mobile Computing, vol.6, no. 1, pp. 2 11, [13] J. H. Chang and L. Tassiulas, Energy conserving routing in wireless ad hoc networks, in Proc. IEEE INFOCOM, 2000, vol. 1, pp [14] G. Mankar and S. T. Bodkhe, Traffic aware energy efficient routing protocol, in Proc. 3rd ICECT, 2011, vol. 6, pp ,. [15] N. Tabassum, Q. E. K. Mamun, and Y. Urano, COSEN: A chain oriented sensor network for efficient data collection, in Proc. IEEE ITCC, Apr. 2006, pp [16] M. Liu, J. Cao, G. Chen, and X. Wang, An energy-aware routing protocol in wireless sensor networks, Sensors, vol. 9, pp , [17] K. T. Kim and H. Y. Youn, Tree-Based Clustering(TBC) for energy efficient wireless sensor networks, in Proc. AINA 2010, 2010, pp