KEYWORDS underwater acoustic sensor networks; three-dimensional environment; collaborative localization; multi-anchor nodes; upgrade anchor node

Transcription

1 WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2016; 16: Published online 26 November 2014 in Wiley Online Library (wileyonlinelibrary.com) RESEARCH ARTICLE MANCL: a multi-anchor nodes collaborative localization algorithm for underwater acoustic sensor networks Guangjie Han 1,2 *, Chenyu Zhang 1, Tongqing Liu 1 and Lei Shu 3 1 Department of Information & Communication Engineering, Hohai University, Changzhou, Jiangsu, China 2 Changzhou Key Laboratory of Sensor Networks and Environmental Sensing, Changzhou, China 3 Guangdong Petrochemical Equipment Fault Diagnosis Key Laboratory, Guangdong University of Petrochemical Technology, Maoming, China ABSTRACT Localization is an essential and major issue for underwater acoustic sensor networks (UASNs). Almost all the applications in UASNs are closely related to the locations of sensors. In this paper, we propose a multi-anchor nodes collaborative localization (MANCL) algorithm, a three-dimensional (3D) localization scheme using anchor nodes and upgrade anchor nodes within two hops for UASNs. The MANCL algorithm divides the whole localization process into four sub-processes: unknown node localization process, iterative location estimation process, improved 3D Euclidean distance estimation process, and 3D DV-hop distance estimation process based on two-hop anchor nodes. In the third sub-process, we propose a communication mechanism and a vote mechanism to determine the temporary coordinates of unknown nodes. In the fourth sub-process, we use two-hop anchor nodes to help localize unknown nodes. We also evaluate and compare the proposed algorithm with a large-scale localization algorithm through simulations. Results show that the proposed MANCL algorithm can perform better with regard to localization ratio, average localization error, and energy consumption in UASNs. Copyright 2014 John Wiley & Sons, Ltd. KEYWORDS underwater acoustic sensor networks; three-dimensional environment; collaborative localization; multi-anchor nodes; upgrade anchor node *Correspondence Guangjie Han, Department of Information & Communication Engineering, Hohai University, Changzhou, Jiangsu, China. hanguangjie@gmail.com Present Address: Suzhou Keda Technology Co., Ltd, Suzhou, China 1. INTRODUCTION Underwater acoustic sensor networks (UASNs) have been drawing much attention because of their potential applications, ranging from environmental research, resource exploration, and emergency events for military purposes [1,2]. For these applications, localization is a fundamental issue. Localization in UASNs is challenging due to several reasons: (I) unavailability of a global positioning system (GPS) [3]; (II) high and variable propagation delays because the communication links are based on acoustic wireless technology [4]; (III) limited bandwidth capacity and energy; and (IV) difficult to deploy sensor nodes at precise locations in underwater environments [5]. Considering these challenges, it is necessary to develop a localization algorithm to reduce distance measurement error and improve localization accuracy and localization ratio, even in sparse underwater sensor networks. Although localization is widely studied for wireless sensor networks (WSNs) [6 ], the existing algorithms for WSNs cannot be directly applied to UASNs. The architecture of UASNs may vary depending on the specific applications [9]. Typically, there are two types of architectures: stationary UASNs and mobile UASNs. In stationary UASNs, sensor nodes are stationary after deployment, irrespective of water current. In mobile UASNs, underwater nodes are equipped with floating buoys which can be inflated by pumps to adjust their depths to cover the entire monitored space according to predetermined schemes or with the help of automatic mobile sensor nodes such as autonomous underwater vehicles (AUVs), unmanned underwater vehicles (UUVs), and low-power gliders. A 62 Copyright 2014 John Wiley & Sons, Ltd.

2 G. Han et al. Multi-anchor nodes collaborative localization algorithm satellite onshore sink bottom node acoustic link underwater node AUV AUV trajectory surface buoy ocean surface ocean bottom Figure 1. A three-dimensional mobile underwater acoustic sensor networks architecture. AUV, autonomous underwater vehicles. three-dimensional (3D) mobile UASN architecture is shown in Figure 1. According to the application requirements, different kinds of sensor nodes can be deployed in UASNs, that is, onshore sinks, surface buoys, underwater sensor nodes, bottom nodes, and automatic mobile sensor nodes. They are deployed to perform collaborative monitoring tasks over a given monitoring region [10]. In order to maximize localization ratio and improve localization accuracy while saving more energy to prolong network lifetime, we propose a multi-anchor nodes collaborative localization (MANCL) algorithm. The MANCL algorithm divides the whole localization process into four sub-processes: unknown node localization process, iterative location estimation process, 3D Euclidean distance estimation process, and 3D DV-hop distance estimation process based on two-hop anchor nodes. We use upgrade anchor nodes within two hops to help residual unknown node with localization during the second to the fourth rounds of localization processes. The improved 3D Euclidean distance estimation process consists of two mechanisms (a communication mechanism and a vote mechanism) to effectively determine the temporary coordinates of an unknown node. In communication mechanism, non-localized unknown nodes use localized sensor nodes within communication range to estimate their temporary coordinates. In vote mechanism, neighboring anchor nodes and upgrade anchor nodes vote to determine the temporary coordinates of non-localized sensor nodes. In the 3D DV-hop distance estimation process, we use average twohop anchor node distance to calculate the coordinates of unknown nodes [11]. In summary, the improved 3D Euclidean distance estimation process is based on the 3D Euclidean distance estimation proposed in large-scale localization (LSL) algorithm [12], which will be introduced detailedly in the following sections. The contributions of this paper are depicted as follows. A multi-anchor nodes collaborative localization algorithm was proposed for UASNs. Detailed descriptions of how multi-anchor nodes collaboratively help unknown nodes with localization are given in the following sections. Selection criteria are specially designed to select appropriate upgrade anchor nodes and two-hop anchor nodes to help unknown nodes with localization. Two mechanisms, a communication mechanism and a vote mechanism, are proposed to effectively determine the temporary coordinates of an unknown node, that is, select the temporary coordinates of the unknown node from two possible positions in the improved 3D Euclidean distance estimation process. The influencing factors of localization error and energy consumption in each process of the MANCL algorithm were also been analyzed. We evaluated the MANCL algorithm with extensive simulations, and results show that the MANCL algorithm can achieve a high localization ratio and localization accuracy with low energy consumption in 3D UASNs. The rest of this paper is organized as follows. Section (2) reviews some related works. Section (3) describes the proposed MANCL localization algorithm in detail. Section (4) gives a detailed analysis of simulation results. And finally, Section (5) makes a conclusion and discusses future research issues of localization algorithms in UASNs. 2. RELATED WORK We previously classified localization algorithms into three categories [13]: (I) stationary localization algorithms; (II) mobile localization algorithms; and (III) hybrid localization algorithms. In this section, we summarize and analyze several typical localization algorithms. A range-free and anchor-free localization scheme, localization with directional beacons for UASNs was proposed by Luo et al. [14]. An AUV patrols over the monitoring region slowly under a predefined trajectory broadcasting beacon messages. By listening to two or more beacons sent from the AUV, unknown nodes can calculate their positions. It is energy-efficient by eliminating internode communication and does not suffer from distance Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 63

3 Multi-anchor nodes collaborative localization algorithm G. Han et al. measurement errors, but it takes a long time to locate all the unknown nodes in a UASN. A hierarchical approach LSL algorithm was presented by Zhou et al. [12]. It is a distributed range-based localization algorithm. The algorithm divides the whole localization process into two parts: anchor node localization and unknown node localization. The algorithm is based on a 3D Euclidean distance estimation method and a recursive location estimation method to estimate distances between unknown nodes and more non-neighboring reference nodes and then calculates the coordinates of the unknown nodes. During the 3D Euclidean distance estimation process, the choice among four possibility locations is made locally by voting by its neighboring nodes. But the authors did not explain why these neighboring nodes are not directly used to help with localization. We will discuss this problem further in Section 3. Bian et al. [15] proposed a hyperbola-based localization scheme (HLS). Instead of using the circle-based least squares location estimation method, the proposed scheme uses the hyperbola-based approach for event locations detection and a normal distribution for estimation error modeling and calibration. Compared with the circle-based approach, HLS is more robust against distance measurement error and has a higher localization ratio. The authors listed three distance measurement techniques adopted in WSNs and analyzed their potential use in UASNs. In general, HLS is suitable for accurate localization in UASNs. However, sensor nodes are required to send long-range signals (around 1000 meters) in HLS, which consumes more energy. In mobile UASNs, locations of sensor nodes change frequently with water current [14,16]. A multi-frequency active localization method based on time difference of arrival was presented in [17]. Surface buoys locate themselves using a GPS after initial deployment and periodically broadcast localization information with lowfrequency acoustic signals. Relay nodes communicate with each other to divide the network into multiple localization domains and calculate max hop for each domain. Unknown nodes remain dormant till detecting an event. During awake time, they receive coordinates from surface buoys and locate themselves. After localization, unknown nodes start high-frequency signal-sending devices to broadcast a message report package (MRP), which contains the detected events and their coordinates. The package broadcasting stops when the value of max hop is zero or the MRP is received by any relay node. Relay nodes send the received MRP to surface buoys for further dispensation. As mentioned earlier, the majority of localization algorithms in UASNs use acoustic range measurements to estimate distance. Callmer et al. [1] utilized triaxial magnetometers and a vessel with a known magnetic dipole to silently localize sensor nodes. Unknown nodes are localized by listening to the messages of the dipole. The ferromagnetic field created by the dipole is measured by the magnetometers. Each unknown node is equipped with a pressure sensor and an accelerometer used for depth estimation and sensor orientation estimation, respectively. The trajectory of the vessel and the positions of unknown nodes are estimated simultaneously by using an extended Kalman filter. Thus, common problems in UASNs such as time synchronization, limited bandwidth capacity and high propagation delays can be avoided. Cheng et al. [19] and Teymorian et al. [20] studied the localization problem in sparse 3D underwater sensor networks. The depth information is used to transform the 3D underwater positioning problem into its two-dimensional counterpart via a projection technique. They proved that a non-degenerative projection preserves network localizability, and they then designed a distributed localization framework for 3D UASNs which they called underwater sensor positioning (USP). Traditional 3D localization techniques require the existence of at least four non-coplanar anchor nodes within communication range of the to-belocalized node. However, in the USP scheme, the requirement is eliminated by using sensor depth information and a location projection technique that maps the positions of neighboring anchor nodes from one plane to another. The framework can be applied with any ranging method proposed for two-dimensional terrestrial sensor networks. Through simulations, they concluded that the USP algorithm could improve localization capabilities with low computation and communication requirements. The depth information is computed with a pressure sensor. Hydraulic pressure may be different, even at the same depth, under the impact of temperature and sensitivity, especially in the ocean environment. Thus, how to balance hydraulic pressure must first be resolved. Besides, the pressure sensor has its own measurement range. A collaborative localization scheme (CLS) for mobile UASNs was proposed [21], where sensor nodes collaborate to determine their coordinates autonomously without using long-range transponders. Sensor nodes are first deployed at the ocean surface; then they use buoyancy control to descend deeper into the ocean. Once the desired depth is reached, the sensor node travels back to the surface. Although sensor nodes know their depth from pressure sensors, their positions in the other two dimensions change continuously due to the motion caused by currents. In order to track the descending nodes, sensor nodes are classified into two categories: profilers and followers. A profiler travels to a depth first. Then, followers track the trajectory of the profiler. All the sensor nodes descend with the same speed. The location of a profiler is a prediction of the followers future locations. CLS is an anchor-free and energy-effective self-localization strategy that does not require prior node planning. But for a sparse or nonhomogenous network, the performance of CLS can be affected significantly. Time synchronization is required in order to achieve high localization accuracy. Moreover, the followers must chase closely after the profiler to avoid communication breakdown. Because an absolutely stationary network does not exist in real applications, the most probable future research will take sensor mobility into account in localization 64 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

4 G. Han et al. Multi-anchor nodes collaborative localization algorithm algorithms. Besides, it is necessary to develop efficient collaborative localization algorithms in which multiple anchor nodes dynamically collaborate to achieve accurate and energy-efficient localization. In the next section, we will present our proposed localization scheme MANCL in detail. 3. MULTI-ANCHOR NODES COLLABORATIVE LOCALIZATION ALGORITHM In this section, we first introduce the network model and relative definitions, then describe collaborative localization mechanisms and present the MANCL algorithm, which includes four sub-processes: (I) unknown node localization; (II) iterative location estimation; (III) improved 3D Euclidean distance estimation; and (IV) 3D DV-hop distance estimation based on two-hop anchor nodes Network model and relative definitions To deal with large-scale UASN localization, we propose a collaborative localization algorithm. A 3D UASN partition of the MANCL model is shown in Figure 2. There are three types of sensor nodes in the network: surface buoys, anchor nodes, and unknown nodes. Surface buoys are floating on the water surface. Anchor nodes and unknown nodes are deployed randomly in the water. Unknown nodes anchor node surface buoy ordinary node can communicate with neighboring anchor nodes, but they cannot directly exchange messages with surface buoys. The monitoring area is divided into several small equal cube-shaped regions, which is accomplished by the surface buoys. Each anchor node belongs to one cube region according to its coordinates. Surface buoys record the number of anchor nodes in each cube region and send these messages to each anchor node. The collaborative localization algorithm carries out both inside one cube unit and among multiple cube units. Definition 3.1. upgrade anchor node. If the trust value of a localized unknown node is no smaller than the trust value threshold, the localized node becomes an upgrade anchor node and helps in other unknown nodes localization. As depicted in Figure 3(a), unknown node O 1 uses the coordinates of A 1, A 2, A 3,andA 4 to calculate its own location and trust value. The trust value of O 1 is larger than the trust value threshold, so O 1 marks itself as an upgrade anchor node. In Figure 3(b), upgrade anchor node O 1 participates in the localization of an unknown node O 2. Definition 3.2. half-localized unknown node. If the trust value of a localized unknown node is smaller than the trust value threshold, the localized unknown node becomes a half-localized unknown node. Definition 3.3. one-hop anchor node. Anchor nodes within communication range of an unknown node are called one-hop anchor nodes, namely, neighboring anchor nodes. (a) monitoring region (b) one cube unit Definition 3.4. one-hop upgrade anchor node. Upgrade anchor nodes within communication range of an unknown node are called one-hop upgrade anchor nodes. Definition 3.5. two-hop anchor node. The neighboring anchor nodes of an unknown node s one-hop anchor nodes or an unknown node s upgrade anchor nodes are defined as two-hop anchor nodes. Figure 2. A large-scale three-dimensional underwater acoustic sensor networks partition. We use an example to illustrate the earlier definitions. Referring to Figure 4, O 1 and O 3 are unknown nodes. O 2 A 3 A 3 A 1 A 1 A 5 A 5 O1 O2 O1 O2 A 2 A 4 A 2 A 4 ordinary node (a) anchor node upgrade anchor node (b) Figure 3. Upgrade anchor nodes help with localization. Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 65

5 Multi-anchor nodes collaborative localization algorithm G. Han et al. and O 4 are upgrade anchor nodes. A 2 and A 4 are one-hop anchor nodes of O 1. O 2 is the upgrade anchor node of O 1. A 1, A 3, A 5,andA 6 are two-hop anchor nodes of O 1. A 7 is not the two-hop anchor node of O 1, because the joint of A 7 and O 1 is an unknown node. According to Definition 3.5, although the intermediary node between upgrade anchor node A 4 and unknown node O 1 is an anchor node O 4,we still think that O 4 is not a two-hop anchor node of O 1,in order to avoid accumulative errors Collaborative localization mechanism There already existed some collaborative localization algorithms in WSNs and UASNs [21 24]. In the collaborative localization algorithms, sensor nodes cooperate with each other to determine the coordinates of all unknown nodes in a network; usually iterative schemes, mobile elements, or distance estimation methods are used to address the limitations of a static sensor network during the localization process. The collaborative localization algorithms aim to maximize the localization ratio and localization accuracy to improve network performance. In the MANCL algorithm, we use one-hop anchor nodes, one-hop upgrade anchor nodes, and two-hop anchor nodes to help with unknown nodes localization. The collaborative localization algorithm carries out both inside one cube unit and among multiple cube units. We give a simple example to illustrate three kinds of anchor nodes used in the proposed collaborative localization algorithm. In Figure 5(a), the sum of one-hop anchor nodes and upgrade anchor nodes of an unknown node is three; the unknown node needs one two-hop anchor node to help with localization. In Figure 5(b), the sum of one-hop anchor nodes and upgrade anchor nodes of an unknown node is A 1 A O4 5 A2 O1 A 3 O 2 A 4 O 3 A 6 A 7 anchor node ordinary node upgrade anchor node Figure 4. An example of a part of network topology. two; it cannot be localized without the help of at least two two-hop anchor nodes. In Figure 5(c), the unknown node has only a one-hop anchor node or upgrade anchor node, and it still needs coordinates messages of at least three two-hop anchor nodes Algorithm description During the localization process, anchor nodes broadcast their coordinates periodically. All the unknown nodes that receive these coordinates messages can estimate their distances to the corresponding anchor nodes. If an unknown node receives four (or more than four) non-coplanar coordinates from different anchor nodes, the unknown node can calculate its position by using a multilateral localization method. However, in large-scale UASNs, not all the unknown nodes can receive coordinates messages from four or more anchor nodes, so it is necessary to use other distance estimation methods to help with localization. In the MANCL algorithm, we use a received signal strength indicator (RSSI) technique to calculate distances between anchor nodes and unknown nodes during the unknown node localization process and the iterative location estimation process. If some unknown nodes still cannot locate themselves after the first and the second localization phases, the algorithm will turn to the next process. In the improved 3D Euclidean distance estimation process, we develop two mechanisms to determine the temporary position of an unknown node. In the fourth process, a two-hop distance estimation scheme is used to calculate average two-hop anchor node distance. Thus, non-localized unknown nodes can use the information to calculate their positions. The complete localization process of the MANCL algorithm is illustrated in the chart titled MANCL Algorithm. Table I lists the parameters used in the MANCL algorithm Unknown node localization process. At the beginning of the localization phase, there are only three types of sensor nodes in the UASN: surface buoys, anchor nodes, and unknown nodes. All the anchor nodes set their trust value to 1, and they broadcast localization data packages periodically to help with unknown nodes localization. The format of the localization data package is depicted in Figure 6. The meaning of each field is explained as follows. ordinary node anchor node anchor node or upgrade anchor node (a) (b) (c) Figure 5. Different kinds of anchor nodes used in the multi-anchor nodes collaborative localization algorithm. 66 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

6 G. Han et al. Multi-anchor nodes collaborative localization algorithm MANCL Algorithm 1: Input: Location of anchor nodes.l a / and trust value threshold 2: Output: Location of unknown nodes.l o / and trust value 3: Anchor nodes broadcast coordinate messages 4: if N a C N ua 4 then 5: if N a 4 then 6: Calculate L o and 7: else : Select 4 N a upgrade anchor nodes 9: Calculate L o and 10: if then 11: return L o and 12: else 13: return Fail_L o 14: else if N a 1 or N ua 1 then 15: flag true 16: if N a C N ua D 3 then 17: Send localization request message R 1 1: if N k > 0 and N ta 1 19: Decide P o or P 0 o then 20: Calculate distance to two-hop anchor node 21: flag true 22: if flag = false then 23: Send localization request message R 2. 24: if N ta > 4.N a C N ua / then 25: Calculate average two-hop distance 26: Calculate L o and 27: else 2: return Fail_L o 29: if then 30: return L o and 31: else 32: return Fail_L o 33: return Fail_L o msg_type: message type; set type of a localization data package to 1 s_time: time when this package is sent n_type: node type; set type of an anchor node to 1 ID: unique identification number of a sending node coord: coordinates of a sending node tr_value: trust value of a sending node r_energy: residual energy of a sending node max_hop: the maximum value among minimum hop counts Each unknown node keeps a counter, N a, to register the number of one-hop anchor nodes. When an unknown node receives a data package from a different ID, N a increases by one. There are two cases: (a) N a 4. The unknown node first calculates its coordinates by using a multilateral localization method and then calculates its trust value. If the trust value is no smaller than a predefined trust value threshold, this unknown node will mark itself as an upgrade anchor node. Then it broadcasts a localization-assisting data package to help with localization. If the trust value is smaller than the threshold, this unknown node marks itself as a half-localized unknown node and plays a part in the communication mechanism. (b) N a < 4. Go to the iterative location estimation process Iterative location estimation process. With the progress of the localization algorithm, more and more unknown nodes are localized and become upgrade anchor nodes. Upgrade anchor nodes can help improve the localization ratio of the network. If N a CN ua 4andN a 4, the unknown node will use the coordinates information of anchor nodes to calculate its position. If N a < 4, the unknown node will select 4 N a upgrade anchor nodes, which have the maximum C 1 among N ua Table I. Parameters used in multi-anchor nodes collaborative localization algorithm. Parameters R N a N ua N ta N k N l R 1 R 2 T _number Maxhop Aver_hop Max_anchor T h I h Meaning Communication range. Trust value. Trust value threshold. The number of one-hop anchor nodes of an unknown node. The number of one-hop upgrade anchor nodes of an unknown node. The number of two-hop anchor nodes of an unknown node. The number of one-hop half-localized unknown nodes of a non-localized unknown node. The number of marking nodes. Localization request message in the improved three-dimensional Euclidean distance estimation process. Localization request message in the three-dimensional DV-hop distance estimation process. The total number of one-hop anchor nodes and upgrade anchor nodes of an unknown node. The maximum value among minimum hop counts from an anchor node to other anchor nodes. The average of the maximum values among minimum hop counts from an anchor node to all the other anchor nodes. The anchor node that is Maxhop hop counts away from another anchor node. Hop count threshold in three-dimensional DV-hop distance estimation process. Interval hop count in three-dimensional DV-hop distance estimation process. Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 67

7 Multi-anchor nodes collaborative localization algorithm G. Han et al. msg_type s_time n_type ID coord tr_value r_energy max_hop Figure 6. Format of a localization data package broadcast by anchor node. msg_type s_time n_type ID coord tr_value r_energy Figure 7. Format of a localization-assisting data package broadcast by upgrade anchor node. upgrade anchor nodes to help with localization. C 1 is defined as Equation (1), E residual 1 C 1 D 1 C ˇ1 C 1 (1) E initial D estimation where E initial and E residual are initial and residual energy of an upgrade anchor node, respectively. is the trust value of an upgrade anchor node. D estimation is estimated distance between an unknown node and an upgrade anchor node. 1, ˇ1, and 1 are weighted values that satisfy 1 C ˇ1 C 1 D 1. Synthesizing trust value calculation formula in LSL [12] and SLMP [25] algorithms, we define trust value as Equation (2), < 1 (anchor node) D ı : 1 P n p (others) id1.u xi / 2 C.v y i / 2 C.w z i / 2 (2) where.u, v, w/ are the estimated coordinates of an unknown node,.x i, y i, z i / are the coordinates of an anchor node i or an upgrade anchor node i.whenn a 4, n D N a. When N a < 4, n D 4. ı is a trust value operator defined as Equation (3), ı D nx id1 q j.u x i / 2 C.v y i / 2 C.w z i / 2 l 2 i j (3) in which l i is the measured distance between an unknown node and reference node i. From Equation (2) and Equation (3), we can observe that is essentially a normalized ı. When, the unknown node is localized successfully and becomes an upgrade anchor node. Otherwise, it becomes a half-localized unknown node to broadcast localization-assisting data packages in the communication mechanism phase and will continue to locate itself. The format of a localization-assisting data package is shown in Figure 7. The meaning of each field is the same as that of a localization data package, but the msg_type of a localization-assisting data package is 2 and n_type is Improved three-dimensional Euclidean distance estimation process. Zhou et al. [12] proposed a Euclidean distance estimation scheme for 3D sensor networks. They aimed to estimate the distance between two non-neighboring nodes from known one-hop distance measurements. The unknown node must know three one-hop neighboring anchor nodes or upgrade anchor nodes to perform the Euclidean distance estimation scheme. We extend this algorithm by adding two key mechanisms: a communication mechanism and a vote mechanism. In Figure, an unknown node E broadcasts a localization request message; its three one-hop anchor nodes or upgrade anchor nodes (A, C, andd) receive this message and forward it to the two-hop anchor nodes. After comparison, unknown node E selects two-hop anchor node B to help with localization. E wants to estimate its distance to B, which is two hops away. Note that there should be at least one anchor node among A, C, andd. Assume that C is an anchor node. Three one-hop neighbors can calculate two possible positions E c and Ec 0 for the unknown node E. The choice among possibilities is made locally by voting when E has more immediate neighbors with estimates to B. The distance estimate to B is not available until E obtains more information from its neighbors [12]. However, this method has some shortcomings; the locations of intermediate nodes that are used to vote have two circumstances: (a) If the location of an intermediate node is known, why not use the intermediate node and anchor nodes A, C, and D to calculate the position of E? (b) If the location of an intermediate node is unknown, it is likely to vote for an incorrect position, which will result in larger localization errors. In order to improve localization accuracy, we propose two mechanisms, a communication mechanism and a vote mechanism, to determine the temporary coordinates of an unknown node E, that is, select the temporary coordinates X Z O D ( ) E E c C ' E c ordinary node anchor node anchor node or upgrade anchor node Figure. Three-dimensional Euclidean estimation. A B Y 6 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

8 G. Han et al. Multi-anchor nodes collaborative localization algorithm msg_type s_time n_type ID r_energy Figure 9. Format of a localization request message R 1. of the unknown node E from two possible positions E c and Ec 0. We describe the two mechanisms which have the same functionality in detail as follows. Definition 3.6. localization request message R 1.Ifthe sum of one-hop anchor nodes and upgrade anchor nodes of an unknown node is smaller than four, the unknown node will broadcast a localization request message R 1.R 1 can be forwarded to two-hop anchor nodes at most. The format of the message R 1 is shown in Figure 9. The meaning of each field is explained as follows: msg_type: message type; set type of R 1 to 3 s_time: time when the package is sent n_type: type of a sending node ID: unique identification number of a sending node r_energy: residual energy of a sending node Each unknown node keeps two counters, N k and N ta,to register the number of neighboring half-localized unknown nodes and two-hop anchor nodes, respectively. Communication mechanism. This mechanism is based on a half-localized neighboring unknown node of the non-localized unknown node, which uses the communication relationship between the half-localized neighboring unknown node and the non-localized unknown node. If there are several half-localized unknown nodes, select the half-localized unknown nodes that has the largest C 2. C 2 is defined as Equation (4), E residual 1 C 2 D 2 C ˇ2 C 2 (4) E initial D unknown localized where E initial, E residual, and are initial energy, residual energy and trust value of a half-localized unknown node, respectively. D unknown localized is the estimated distance between a to-be-localized unknown node and a half-localized unknown node. 2, ˇ2, and 2 are weighted values that satisfy 2 C ˇ2 C 2 D 1. Suppose that a non-localized unknown node unknown_1 has a half-localized neighboring unknown node half _unknown_2, which has the maximum C 2 among half-localized neighboring unknown nodes. If the distance between half _unknown_2 and Ec 0 is larger than R and the distance between half _unknown_2 and E c is smaller than R or equal to R, unknown_1 will choose E c as its temporary position. Otherwise, if the distance between half _unknown_2 and E c is larger than R, and the distance between half _unknown_2 and Ec 0 is smaller than R or equal to R, unknown_1 will choose Ec 0 as its temporary position. Or else, the unknown node cannot be localized in improved 3D Euclidean distance estimation process. Vote mechanism. In the improved 3D Euclidean distance estimation process, T_number of an unknown node is three. But for other possible coordinates that should be excluded, T_number is likely to be larger than three. Thus, if T_number of E c is three, and T_number of Ec 0 is larger than three, the unknown node will choose E c as its temporary position. If T_number of Ec 0 is three, and T_number of E c is larger than three, the unknown node will choose Ec 0 as its temporary position. Otherwise, the unknown node cannot be localized in the improved 3D Euclidean distance estimation process. From the earlier discussion, the unknown node knows three one-hop anchor nodes or upgrade anchor nodes and its temporary coordinates. The unknown node still needs a two-hop anchor node to help in localization. The selection criteria of the two-hop anchor node are depicted as Equation (5), C 3 D D unknown one C D one two (5) where D unknown one is the distance between a to-belocalized unknown node and a neighboring anchor node or the distance between a to-be-localized unknown node and a neighboring upgrade anchor node, and D one two is the distance between a neighboring anchor node and a two-hop anchor node or the distance between a neighboring upgrade anchor node and a two-hop anchor node. The unknown node will select a two-hop anchor node that has the smallest C 3 among all the two-hop anchor nodes. Then, it will calculate the Euclidean distance between the temporary coordinates and the two-hop anchor node. Thus, the unknown node can calculate its coordinates and trust value by using three one-hop anchor nodes or upgrade anchor nodes and a two-hop anchor node. When, the unknown node is localized successfully and becomes an upgrade anchor node Three-dimensional DV-hop distance estimation process based on two-hop anchor nodes. In a large-scale sparse deployed UASN, not all the unknown nodes can be localized successfully during the earlier three sub-processes. If non-localized unknown nodes do not have any neighboring half-localized unknown node or cannot meet the localization requirements, they will turn to the fourth localization process. There are four cases for the residual non-localized unknown nodes: (a) The sum of one-hop anchor nodes and upgrade anchor nodes of the non-localized unknown node is three. (b) The sum of one-hop anchor nodes and upgrade anchor nodes of the non-localized unknown node is two. (c) The non-localized unknown node has only one onehop anchor node or only one upgrade anchor node. Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 69

9 Multi-anchor nodes collaborative localization algorithm G. Han et al. Figure 11. Format of a localization request message R 2. Na Nua 3 Na Nua 2 Na 1or Nua 1 Figure 10. Cases (a), (b), and (c) in three-dimensional DV-hop distance estimation process. (d) The non-localized unknown node has no one-hop anchor node or upgrade anchor node. In the earlier four cases, non-localized unknown nodes in situation (a), (b), and (c) have at least one neighboring anchor node or upgrade anchor node, as illustrated in Figure 10. The non-localized unknown node in situation (a), (b), and (c) needs one two-hop anchor node, two two-hop anchor nodes and three two-hop anchor nodes to help with localization, respectively. The unknown nodes in the fourth case cannot be localized, because they do not have any neighboring one-hop anchor node or upgrade anchor node. We use a 3D DV-hop distance estimation scheme based on two-hop anchor nodes to estimate the distance between an unknown node and its two-hop anchor nodes. The cases in Figure 10 all satisfy N a C N ua < 4, N a 1orN ua 1; thus, we need 4.N a CN ua / two-hop anchor nodes to help with localization. At the beginning of the localization algorithm, when anchor nodes broadcast their coordinates, IDs, and hop counts, each anchor node has already recorded every minimum hop count to other anchor nodes and forwarded the coordinates messages to its neighboring anchor nodes. Then, each anchor node registers the maximum value among the minimum hop counts from itself to other anchor nodes. The average of the maximum values among minimum hop counts from an anchor node to other anchor nodes is calculated with Equation (6), Aver_hop D P Ta id1 Maxhop i T a (6) in which T a is the total number of anchor nodes in a UASN, and Maxhop i is the maximum value among minimum hop counts from anchor node i to all the other anchor nodes. The non-localized unknown node will choose 4.N a C N ua / two-hop anchor nodes that have the maximum C 4. C 4 is calculated with Equation (7), 1 C 4 D C ' Maxhop D unknown one C D one two Aver_hop where D unknown one is the distance between a to-belocalized unknown node and a neighboring anchor node or the distance between a to-be-localized unknown node and a neighboring upgrade anchor node, and D one two is the (7) distance between a neighboring anchor node and a twohop anchor node or the distance between a neighboring upgrade anchor node and a two-hop anchor node. and ' are weighted values that satisfy C ' D 1. Definition 3.7. localization request message R 2. After the third localization sub-process, if an unknown node s total number of anchor nodes and upgrade anchor nodes is still smaller than four, the unknown node will broadcast a localization request message R 2. The localization request message R 2 will be forwarded to its two-hop anchor nodes. The format of a localization request message R 2 is shown in Figure 11. The meaning of the fields, which are different from the former, are explained as follows: msg_type: message type; set type of R 2 to onehop_id: ID of a one-hop anchor node or an upgrade anchor node twohop_id: ID of a two-hop anchor node t_hop: hop count threshold; the same as T h i_hop: interval hop count; the same as I h r_energy: residual energy of a sending node Definition 3.. forwarding primary path of R 2. Path from the non-localized unknown node to the anchor node that is t_hop away to forward the localization request message R 2. Definition 3.9. marking node. We mark anchor nodes of every i_hop on forwarding primary path of R 2 as marking nodes, which are used to calculate the average two-hop distance of anchor nodes. After receiving a localization request message R 2,the receiver will select a next-hop anchor node that has the maximum residual energy, so as to reach the anchor node that is t_hop away. Every anchor node on the forwarding primary path of R 2 must obey this criterion until localization request message R 2 is forwarded to the destination anchor node. Figure 12 gives an example to explain identifying nodes and the forwarding primary path of R 2. We mark anchor nodes on every three hop counts as marking nodes in this instance. At the beginning, unknown node O broadcasts a localization request message R 2 ; the first-hop anchor node marks itself as a marking node, then forwards R 2 to the next-hop anchor node; anchor nodes on every three hop counts mark themselves as marking nodes constantly; the anchor node on the last hop, t_hop away from the unknown node, also marks itself as an identifying node. Thus, 690 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

10 G. Han et al. Multi-anchor nodes collaborative localization algorithm Figure 12. An explanation of marking nodes and the forwarding primary path of R 2. A 1, A 5,andA 9 are marking nodes as shown in Figure 12. fo, A 1, A 2, A 4, A 5, A 6, A, A 9 g forms the forwarding primary path of R 2. fo, A 1, A 2, A 4, A 5, A 7 g and fo, A 1, A 3 g are not forwarding primary paths because the total number of hop counts is smaller than T h. We can determine the number of marking nodes N l according to the relationship between Max_anchor and T h 1. If Maxhop T h 1, there are T h hop counts between the one-hop anchor node and Max_anchor. N l is defined as Equation (), < T h 1 I N l D h C 1 : Th 1 I h j Th 1 I h k C 2 (others) is integer If Maxhop < T h 1, there are Maxhop C 1 hop counts between the one-hop anchor node and Max_anchor. N l is defined as Equation (9), < Maxhop I N l D h C 1 : Maxhop I h j Maxhop I h k C 2 (others) is integer Average one-hop anchor node distance can be calculated with Equation (10), P q j i.x i x j / 2 C.y i y j / 2 C.z i z j / 2 hop i D P j i h, ij i, j 2f1, 2, :::, N l g (10) where.x i, y i, z i / and.x j, y j, z j / are coordinates of marking nodes i and j, respectively, and h i,j is the hop counts between anchor node i and j.i j/. The unknown node which sends the localization request message R 2 will receive N l average one-hop distance feedback messages. Then, the unknown node will calculate Aver_hop and d unknown two by using Equation (11) and Equation (12). Aver_hop D () (9) P Nl id1 hop i N l (11) D unknown two D 2 Aver_hop (12) O A A A A A 6 A 4 A A A A9 11 A 10 Figure 13. Forwarding process of a localization request message R 2. Thus, the unknown node can calculate its position by using 4.N a C N ua / two-hop anchor nodes and N a C N ua one-hop anchor nodes or upgrade anchor nodes and its trust value to decide whether to become an upgrade anchor node. During this process, each unknown node only sends the localization request message R 2 once in order to reduce energy consumption. We give an example to illustrate the forwarding process of a localization request message R 2 as shown in Figure 13. In this instance we set T h D, I h D 3; Maxhop of A 1 is. The number on the first line under anchor nodes represents the maximum hop counts that the anchor node can reach. The number on the second line under the anchor nodes is the residual energy. In Figure 13, R 2 is forwarded to A 2 by A 1 according to its onehop_id and twohop_id. Except for A 1, A 2 has another three neighboring anchor nodes: A 3, A 4,andA 5. Because A 3 can only forward R 2 for one more hop, and the residual energy of A 5 is smaller than that of A 4, A 2 selects A 4 as the next forwarding hop. According to this standard, an optimum forwarding path fa 1, A 2, A 4, A 7, A 9, A 10, A 11, A 12, A 13 g is formed. A 1, A 7, A 11,andA 13 mark themselves as marking nodes automatically to calculate average one-hop distances and pass back the calculation results to unknown node O. 4. LOCALIZATION ERROR AND ENERGY CONSUMPTION In this section, the influencing factors of localization error and energy consumption in each sub-process of the MANCL algorithm are analyzed theoretically. A 12 A 13 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 691

11 Multi-anchor nodes collaborative localization algorithm G. Han et al Localization error analysis In the MANCL algorithm, unknown nodes estimate their distances to anchor nodes by using the RSSI technique or estimation methods such as the improved 3D Euclidean distance estimation algorithm and 3D DV-hop distance estimation algorithm. Inevitably, there exist some measurement errors. Because we use upgrade anchor nodes to help with localization, accumulation error will affect localization accuracy as well. Thus, the main influencing factors of localization error include propagation delay and accumulation error Localization error caused by propagation delay. Due to severe signal delay and degradation caused by multi-path propagation and high temporal and spatial variability of the channel conditions, communication lag between sending and receiving nodes is larger in practice. Thus, the measured distance is larger than the real distance between two sensor nodes. Assume that.x 1, y 1, z 1 /,.x 2, y 2, z 2 /,.x 3, y 3, z 3 /,and.x 4, y 4, z 4 / are coordinates of four anchor nodes, and.x, y, z/ are the coordinates of an unknown node. The distances between an unknown node and four anchor nodes are d 1, d 2, d 3,andd 4 respectively. We can obtain:.x 1 x/ ˆ< 2 C.y 1 y/ 2 C.z 1 z/ 2 D d1 2.x 2 x/ 2 C.y 2 y/ 2 C.z 2 z/ 2 D d 2.x ˆ: 3 x/ 2 C.y 3 y/ 2 C.z 3 z/ 2 D d3 2 (13).x 4 x/ 2 C.y 4 y/ 2 C.z 4 z/ 2 D d4 2 By subtracting the fourth equation from the first three equations, we can obtain Equation (14). Equation (14) can be simplified as follow, Through simple computation, we obtain Equation (16), x D a 3.b 2 c 3 b 3 c 2 /s 1 Ca 3.b 3 c 1 b 1 c 3 /s 2 Ca 3.b 1 c 2 b 2 c 1 /s 3.a 3 b 1 a 1 b 3 /.a 3 c 2 a 2 c 3 /.a 3 b 2 a 2 b 3 /.a 3 c 1 a 1 c 3 / ˆ< y D a 3.a 3 c 2 a 2 c 3 /s 1 Ca 3.a 1 c 3 a 3 c 1 /s 2 Ca 3.a 2 c 1 a 1 c 2 /s 3.a 3 b 1 a 1 b 3 /.a 3 c 2 a 2 c 3 /.a 3 b 2 a 2 b 3 /.a 3 c 1 a 1 c 3 / ˆ: z D a 3.a 2 b 3 a 3 b 2 /s 1 Ca 3.a 3 b 1 a 1 b 3 /s 2 Ca 3.a 1 b 2 a 2 b 1 /s 3.a 3 b 1 a 1 b 3 /.a 3 c 2 a 2 c 3 /.a 3 b 2 a 2 b 3 /.a 3 c 1 a 1 c 3 / (16) Equation (16) can be simplified to Equation (17), x D xs 1 Cˇxs 2 C x s 3 m ˆ< y D ys 1 Cˇys 2 C y s 3 m ˆ: z D zs 1 Cˇzs 2 C z s 3 m of which x D a 3.b 2 c 3 b 3 c 2 /, ˇx D a 3.b 3 c 1 b 1 c 3 /, x D a 3.b 1 c 2 b 2 c 1 /, y D a 3.a 3 c 2 a 2 c 3 /, ˇy D a 3.a 1 c 3 a 3 c 1 /, y D a 3.a 2 c 1 a 1 c 2 /, z D a 3.a 2 b 3 a 3 b 2 /, ˇz D a 3.a 3 b 1 a 1 b 3 /, z D a 3.a 1 b 2 a 2 b 1 /, m D.a 3 b 1 a 1 b 3 /.a 3 c 2 a 2 c 3 /.a 3 b 2 a 2 b 3 /.a 3 c 1 a 1 c 3 / (17) The measurement distances d 1, d 2, d 3,andd 4 have error increments Md 1, Md 2, Md 3 and Md 4, respectively. Distances between the unknown node and its four neighboring ˆ< 2.x 1 x 4 /x C 2.y 1 y 4 /y C 2.z 1 z 4 /z D x1 2 x2 4 C y2 1 y2 4 C z2 1 z2 4 C d2 4 d2 1 2.x 2 x 4 /x C 2.y 2 y 4 /y C 2.z 2 z 4 /z D x2 2 ˆ: x2 4 C y2 2 y2 4 C z2 2 z2 4 C d2 4 d2 2, 2.x 3 x 4 /x C 2.y 3 y 4 /y C 2.z 3 z 4 /z D x3 2 x2 4 C y2 3 y2 4 C z2 3 z2 4 C d2 4 d2 3 (14) < a 1 x C b 1 y C c 1 z D s 1 a : 2 x C b 2 y C c 2 z D s 2 (15) a 3 x C b 3 y C c 3 z D s 3 of which a 1 D 2.x 1 x 4 /, b 1 D 2.y 1 y 4 /, c 1 D 2.z 1 z 4 /, a 2 D 2.x 2 x 4 /, b 2 D 2.y 2 y 4 /, c 2 D 2.z 2 z 4 /, a 3 D 2.x 3 x 4 /, b 3 D 2.y 3 y 4 /, c 3 D 2.z 3 z 4 /, s 1 D x1 2 x2 4 C y2 1 y2 4 C z2 1 z2 4 C d2 4 d2 1, s 2 D x2 2 x2 4 C y2 2 y2 4 C z2 2 z2 4 C d2 4 d2 2, s 3 D x3 2 x2 4 C y2 3 y2 4 C z2 3 z2 4 C d2 4 d2 3 anchor nodes should be d 1 C Md 1, d 2 C Md 2, d 3 C Md 3, and d 4 C Md 4 actually. The error increments will affect s 1, s 2,ands 3 in Equation (17). After revising, we obtain Equation (1), ˆ< ˆ: of which x D x.s 1 CMs 1 /Cˇx.s 2 CMs 2 /C x.s 3 CMs 3 / m y D y.s 1 CMs 1 /Cˇy.s 2 CMs 2 /C y.s 3 CMs 3 / m z D z.s 1 CMs 1 /Cˇz.s 2 CMs 2 /C z.s 3 CMs 3 / m Ms 1 D Md4 2 2Md 4d 4 Md1 2 C 2Md 1d 1, Ms 2 D Md4 2 2Md 4d 4 Md2 2 C 2Md 2d 2, Ms 3 D Md4 2 2Md 4d 4 Md3 2 C 2Md 3d 3 (1) 692 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

12 G. Han et al. Multi-anchor nodes collaborative localization algorithm According to Equations (17) and (1), we can calculate.mx, My, Mz/ between the estimated coordinates, and the real coordinates as depicted in Equation (19), ˆ< ˆ: Mx D xms 1 CˇxMs 2 C x Ms 3 m My D yms 1 CˇyMs 2 C y Ms 3 m Mz D zms 1 CˇzMs 2 C z Ms 3 m ˆ< ˆ: (19) 4.2. Energy consumption Energy-efficient communication in sensor networks is of paramount importance because the devices have limited battery life. The unbalanced power consumption among sensor nodes may cause network partition [26]. Thus, it is necessary to optimize energy consumption to prolong the lifetime of UASNs. Different methods for reducing power consumption in WSNs and UASNs have been explored [27 29]. During the localization procedure, energy consumption primarily consists of three aspects: sending messages, x D x.s 1 CMs 1 /C.ˇxC2a 3 b 3 Mz 1 2a 3 c 3 My 1 /s 2 C. x C2a 3 c 2 My 1 2a 3 b 2 Mz 1 /s 3 mc2.a 3 My 1 b 3 Mx 1 /.a 3 c 2 a 2 c 3 / 2.a 3 Mz 1 c 3 Mx 1 /.a 3 b 2 a 2 b 3 / y D y.s 1 CMs 1 /C.ˇyC2a 3 c 3 Mx 1 2a 2 3 Mz 1/s 2 C. y C2a 2 a 3 Mz 1 2a 3 c 2 Mx 1 /s 3 mc2.a 3 My 1 b 3 Mx 1 /.a 3 c 2 a 2 c 3 / 2.a 3 Mz 1 c 3 Mx 1 /.a 3 b 2 a 2 b 3 / z D z.s 1 CMs 1 /C.ˇzC2a 2 3 My 1 2a 3 b 3 Mx 1 /s 2 C. z C2a 3 b 2 Mx 1 2a 2 a 3 My 1 /s 3 mc2.a 3 My 1 b 3 Mx 1 /.a 3 c 2 a 2 c 3 / 2.a 3 Mz 1 c 3 Mx 1 /.a 3 b 2 a 2 b 3 / (20) Therefore, we need to adjust the estimated coordinates shown in Equation (17) referring to the distribution model of.mx, My, Mz/. The theoretical analysis offers a mathematical tool to study the localization error caused by propagation delay, our ultimate objective is to be able to infer the distribution model of the.mx, My, Mz/ to reduce localization errors in real applications based on actual tests Localization error caused by accumulation error. During the iterative location estimation process, successfully localized unknown nodes become upgrade anchor nodes if their trust values are no smaller than the trust value threshold. When residual to-be-localized unknown nodes use these upgrade anchor nodes to help calculate their coordinates, accumulative localization errors are generated. Suppose that an unknown node already has three neighboring anchor nodes; it needs to select one upgrade anchor node, of which the calculated coordinates are.x 1 C Mx 1, y 1 C My 1, z 1 C Mz 1 /. Assume that there is no distance measurement error. Thus, we can obtain the coordinates of an unknown node using Equation (20), of which Ms 1 D Mx 2 1 C My2 1 C Mz2 1 2x 1Mx 1 2y 1 My 1 2z 1 Mz 1. We can calculate the difference between the real coordinates and the estimated coordinates using Equation (17) and Equation (20). In the first sub-process of the MANCL algorithm, localization error mainly results from propagation delay. In the other three sub-processes, localization error is impacted by both propagation delay and accumulation error. Because the algorithm uses upgrade anchor nodes to help with localization, localization error caused by propagation delay accumulate during the localization process. It means that localization error caused by propagation delay is a major factor in influencing localization accuracy in UASNs. receiving messages, and calculations. We will analyze energy consumption in each sub-process in detail. Assume that there are M anchor nodes and N unknown nodes in a UASN. Energy consumption of a sending message is E s, energy consumption of a receiving message is E r and energy consumption of a calculation is E c Energy consumption in the first sub-process. In this process, both anchor nodes and unknown nodes consume energy. Each anchor node needs to broadcast localization data package once during the first process. So energy consumption of each anchor node is E a D E s (21) If an unknown node has N a one-hop anchor nodes, it has to calculate N a distances and decide whether to calculate its coordinates and trust value. (a) If N a < 4, the unknown node receives N a localization data packages and calculates N a distances. The energy consumption E o is E o D N a.e r C E c / (22) (b) If N a 4, the unknown node receives N a localization data packages and calculates N a distances. Besides, it needs to calculate its coordinates and trust value. Thus, the energy consumption E o is E o D N a.e r C E c / C 2E c (23) Energy consumption in the second sub-process. Because anchor nodes have already broadcast localization data packages during the first process, in the second Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 693

13 Multi-anchor nodes collaborative localization algorithm G. Han et al. process, only upgrade anchor nodes will broadcast localization-assisting data packages. Unknown nodes that have fewer than four neighboring anchor nodes will receive these packages and calculate corresponding measurement distances. Energy consumption of an upgrade anchor node is E 0 ua D E s (24) (a) If N ua < 4 N a, the unknown node receives N ua localization-assisting data packages and calculates distances and C 1 for N ua times. Energy consumption E 0 o is E 0 o D N ua.e r C 2E c / (25) (b) If N ua 4 N a, the unknown nodes receives N ua localization-assisting data packages and calculates distances and C 1 for N ua times; in addition, it needs to calculate its coordinates and trust value, so energy consumption E 0 o is E 0 o D N ua.e r C 2E c / C 2E c (26) Energy consumption in the third sub-process. If an unknown node has N a anchor nodes, N ua onehop upgrade anchor nodes, N ta two-hop anchor nodes, and N k neighboring half-localized unknown nodes, the energy consumption in the communication mechanism and the vote mechanism can be classified as follows. (a) Communication mechanism Each one-hop anchor node receives R 1 at least once. If this one-hop anchor node is an intermediary node between an unknown node and N ta two-hop anchor nodes, it needs to forward R 1 and feedback messages from two-hop anchor nodes. Thus, the energy consumption of a onehop anchor node is Ea 00 D Er (others).1 C N ta /.E s C E r /.forward R 1 / (27) Each one-hop upgrade anchor node receives R 1 at least once. If this one-hop upgrade anchor node is an intermediary node between an unknown node and Nta 0 two-hop anchor nodes, it will forward R 1 and feedback messages from two-hop anchor nodes. The energy consumption of a one-hop upgrade anchor node is Eua 00 D Er (others) 1 C N 0 ta.es C E r /.forward R 1 / (2) Each neighboring half-localized unknown node receives R 1 once. Then, it feeds back its coordinates in return. Thus, the energy consumption of a half-localized unknown node is E 00 k D E r C E s (29) Each two-hop anchor node receives R 1 from one-hop anchor nodes or upgrade anchor nodes. Then, it feeds back its coordinates in return. The energy consumption of a two-hop anchor node is Eta 00 D E r C E s (30) The unknown node sends R 1 once and receives N k feedback messages from half-localized unknown nodes and N ta feedback messages from two-hop anchor nodes. It also needs to calculate C 2 for N k times, C 3 for N ta times, and its coordinates and trust value once each. Thus, the energy consumption of an unknown node is E 00 o D E s C.N k C N ta /E r C.N k C N ta C 2/E c (31) (b) Vote mechanism Each one-hop anchor node receives and forwards R 1 at least once. If this one-hop anchor node is an intermediary node between an unknown node and N ta two-hop anchor nodes, it needs to forward R 1 and feedback messages from two-hop anchor nodes. Thus, the energy consumption of a one-hop anchor node is Ea 000 D Er (others).1 C N ta /.E s C E r /.forward R 1 / (32) Suppose that the one-hop upgrade anchor node is an intermediary node between an unknown node and Nta 0 two-hop anchor nodes. The energy consumption of a one-hop upgrade anchor node is the same as that of a one-hop anchor node, Eua 000 D Er (others) 1 C N 0 ta.es C E r /.forward R 1 / (33) Each two-hop anchor node receives R 1 from its one-hop anchor nodes or upgrade anchor nodes. Then, it feeds back its coordinates in return. The energy consumption of a two-hop anchor node is Eta 000 D E r C E s (34) The unknown node sends R 1 once and receives N ta feedback messages from two-hop anchor nodes. It also needs to calculate C 3 for N ta 694 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

14 G. Han et al. Multi-anchor nodes collaborative localization algorithm times, its coordinates and trust value once each. Thus, the energy consumption of an unknown node is E 000 o D E s C N ta E r C.N ta C 2/E c (35) Energy consumption in the four sub-process. Unknown nodes have already known the information of its two-hop anchor nodes during the improved 3D Euclidean distance estimation process. Therefore, in the 3D DV-hop distance estimation process, an unknown node will choose 4.N a C N ua / two-hop anchor nodes to help with localization. Thus, there are four kinds of energy consumption in this process. The first and last marking nodes on the forwarding primary path of R 2 send or receive data packages once. Other marking nodes send and receive data packages twice. Each marking node needs to calculate Aver_hop once. Thus, the energy consumption of a marking node is < E s C E r C E c E i D : (the first and the last marking nodes) 2.E s C E r / C E c (other marking nodes) (36) Other anchor nodes on the forwarding primary path receive and send data package R 2 twice. The energy consumption of the anchor node is E af D 2.E s C E r / (37) Although anchor nodes that are not on the forwarding primary path may also receive the localization request message R 2, they do not forward R 2. Thus, the energy consumption of the anchor node is E naf D E r (3) The unknown node sends R 2 once and receives N l feedback messages from marking nodes. It also needs to calculate C 4 for N ta times, and its coordinates and trust value once each. Thus, the energy consumption of an unknown node is 5. SIMULATIONS E 0000 o D E s C E r C 3E c (39) In this section, we evaluate the performance of the proposed localization algorithm. The algorithm was implemented using MATLAB. We named the MANCL algorithm that uses the communication mechanism MANCL_C, and the MANCL algorithm that uses the vote mechanism MANCL_V. Due to the high localization ratios of the MANCL_C, MANCL_V, and LSL algorithms, the performance differences are not obvious unless the trust value threshold is large enough. Thus, we compare the MANCL_C, MANCL_V, and LSL algorithms in terms of different number of anchor nodes, unknown nodes, and communication ranges under a large trust value threshold. Then, we examine the performance of the MANCL_C, MANCL_V, and LSL algorithms under different trust value thresholds. Lastly, we analyze the MANCL_V algorithm in terms of different hop count thresholds and interval hop counts. To ensure reliability of the evaluation results, 50 simulation runs were performed for each set of simulation conditions Simulation settings In our experiments, sensor nodes are randomly deployed in a 3D space. The simulation region is divided into 125 small cube-shaped regions. Table II lists the parameters used in simulations Performance metrics The performance of the proposed algorithm is investigated in terms of localization ratio, average localization error, and average energy consumption Localization ratio. Localization ratio is the percentage of localized unknown nodes among the unknown nodes. The higher the localization ratio, the more unknown nodes can be used to collaboratively accomplish localization. We define localization ratio as Equation (40), Loc_ratio D N s N o (40) where N s is the number of localized unknown nodes, and N o is the sum of unknown nodes deployed initially in a UASN. Table II. Parameters used in simulations. Simulation region size 500 m 500 m 500 m Divided cube region size 100 m 100 m 100 m Number of unknown nodes Communication range m Initial energy J Sending energy consumption 1.44 J Receiving energy consumption 1.15 J Calculation energy consumption 0.1 J Trust value Interval hop count 2 4 Hop count threshold 5 9 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 695

15 Multi-anchor nodes collaborative localization algorithm G. Han et al Average localization error. Localization error is the average distance between the estimated coordinates and the real coordinates of an unknown node as defined in Equation (41). Average localization error is the sum of localization errors divided by the number of localized unknown nodes, as depicted in Equation (42), q Loc_error D.u x/ 2 C.v y/ 2 C.w z/ 2, (41) P Ns id1 Aver_locerror D Loc_error i (42) N s where.u, v, w/ are real coordinates of an unknown node,.x, y, z/ are estimated coordinates of an unknown node, Loc_error i is the localization error of a localized unknown node i,andn s is the number of localized unknown nodes Average energy consumption. Average energy consumption is defined as the total energy consumption of the network divided by the total number of sensor nodes, which includes anchor nodes and unknown nodes. Aver_consumption D P Aamount id1 EA i C P O amount jd1 EO j (43) A amount C O amount A amount and O amount are the number of anchor nodes and unknown nodes, respectively. EA i is the energy consumption of anchor node i during the localization process, and EO j is the energy consumption of unknown node j during localization process Impact of anchor node ratio In this set of simulations, we examine the performance of the three localization algorithms, MANCL_C, MANCL_V, and LSL, in terms of localization ratio, average localization error, and average energy consumption. The anchor node density varies from 10% to 35% with increments of 5%, as depicted in Figure 14. The localization ratios of the three algorithms increase with the increase of the number of anchor nodes, as shown in Figure 14(a). The more anchor nodes, the higher the localization ratio. For instance, when the anchor node percentage of the MANCL_V algorithm is 15%, the localization ratio is nearly 90%; when the anchor node percentage is 30%, the localization ratio can reach nearly 96%. Thus, we can increase the number of anchor nodes to achieve a higher localization ratio in sparse UASNs. The MANCL_C and MANCL_V algorithms outperform the LSL algorithm in general. The performance results also reveal that MANCL_V can improve the localization ratio more efficiently than the other two algorithms, especially in sparse deployment. Figure 14(b) represents the relationship between the anchor node ratio and average localization error. When the anchor node ratio is relatively small, the average localization error of LSL algorithm is smaller than that of the other two algorithms. With the increase of anchor node ratio, the average localization errors of the MANCL_C algorithm and the MANCL_V algorithm decrease rapidly and become smaller than that of LSL algorithm. This is owing to that at the beginning, when the anchor node ratio is relatively small, MANCL_C and MANCL_V use a 3D DV-hop distance estimation scheme to help with localization, which is not accurate. With the increase of anchor node ratio, more unknown nodes can calculate their coordinates during the first three localization processes. Besides, we take residual energy, trust value, and distance estimation into account when selecting appropriate upgrade anchor nodes to help with localization. However, we cannot expect to reduce average localization error by simply increasing the anchor node ratio, because the improvement is limited when the number of anchor nodes increases to a certain point. Figure 14(c) depicts the relationship between average energy consumption and anchor node ratio. For LSL algorithm, when the network is sparse, although unknown nodes broadcast localization request messages continuously, unknown nodes cannot receive enough coordinates information from anchor nodes. Therefore, the average energy consumption of the LSL algorithm is high. With the increase of anchor node ratio, the average energy (a) (b) (c) Figure 14. Impact of anchor node ratio: (a) localization ratio, (b) average localization error, and (c) average energy consumption. LSL, large-scale localization; MANCL, multi-anchor nodes collaborative localization; V, vote mechanism; C, communication mechanism. 696 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

16 G. Han et al. Multi-anchor nodes collaborative localization algorithm consumption of LSL algorithm becomes lower and lower. Compared with the LSL algorithm, the MANCL_C and MANCL_V algorithms consume less energy. This is because the unknown nodes in MANCL_C algorithm and MANCL_V algorithm send localization request messages once, when needed. When the anchor node ratio is relatively small, most unknown nodes have to broadcast localization request messages during the third and fourth localization sub-processes; thus, the average energy consumption increases with the anchor node ratio at the beginning. However, after the anchor node ratio reaches 25%, the average energy consumption decreases as the anchor node ratio increases because most of the unknown nodes can be localized during the first two localization subprocesses, so they do not need to send localization request messages any more Impact of the number of unknown nodes In this subsection, the number of unknown nodes varies from 100 to 600 with increments of 100. Figure 15 compares the three algorithms in terms of localization ratio, average localization error, and average energy consumption. Figure 15(a) shows that for the three algorithms the more unknown nodes there are, the higher the localization ratios are. When the number of unknown nodes varies from 100 to 300, localization ratios of the three algorithms increase rapidly. After that, the localization ratios change more slowly. The localization ratios of the MANCL_V and MANCL_C algorithms are larger than that of LSL algorithm, in general. The results also illustrate that the vote mechanism performs better than the communication mechanism in improving network localization ratio when unknown node density changes. We observe from Figure 15(b) that with an increase in unknown node density, average localization errors increase at the beginning. However, as the unknown node density continues to increase, the average localization errors of the three algorithms decrease slowly. This can be explained as follows. When the number of unknown nodes varies from 100 to 300, most unknown nodes can localize themselves, and some of them become upgrade anchor nodes to help other residual unknown nodes with localization, which results in accumulative errors. If we continually increase the unknown node density, more localized unknown nodes will become upgrade anchor nodes; thus, the non-localized unknown nodes have more choices to select appropriate upgrade anchor nodes according to the selection criteria. Thus, average localization errors decrease. However, the LSL algorithm selects upgrade anchor nodes only on the basis of trust value. Thus, the average localization error of the LSL algorithm is larger than that of MANCL_C and MANCL_V algorithms. With the increase of unknown node density, average energy consumption becomes higher and higher owing to the increment of forwarding data packets, as shown in Figure 15(c). Compared with LSL algorithm, MANCL_C and MANCL_V algorithms save more energy Impact of communication range In this subsection, we study the impact of communication range on the LSL, MANCL_C and MANCL_V algorithms. We change the communication range from 60m to 140m. The results are plotted in Figure 16. From Figure 16(a), we can see that when the communication range is 60m, the localization ratios of the three schemes are all equal to zero. With the increase of communication range, the localization ratios increase correspondingly. After the communication range increases to 120m, the localization ratios show no apparent change. The results also indicate that the three algorithms perform similarly in improving localization ratio when the communication range of the sensor nodes changes. Figure 16(b) compares the performance of the three algorithms in average localization error. Note that the average localization errors of the three algorithms increase with the increase of communication range before the communication range reaches 90m, due to the accumulative errors caused by upgrade anchor nodes. When we continue to increase the communication range, unknown nodes can communicate with more anchor nodes and have a greater (a) (b) (c) Figure 15. Impact of the number of unknown nodes: (a) localization ratio, (b) average localization error, and (c) average energy consumption. LSL, large-scale localization; MANCL, multi-anchor nodes collaborative localization; V, vote mechanism; C, communication mechanism. Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 697

17 Multi-anchor nodes collaborative localization algorithm G. Han et al. (a) (b) (c) Figure 16. Impact of the communication range: (a) localization ratio, (b) average localization error, and (c) average energy consumption. LSL, large-scale localization; MANCL, multi-anchor nodes collaborative localization; V, vote mechanism; C, communication mechanism. (a) (b) (c) Figure 17. Impact of the trust value threshold: (a) localization ratio, (b) average localization error, and (c) average energy consumption. LSL, large-scale localization; MANCL, multi-anchor nodes collaborative localization; V, vote mechanism; C, communication mechanism. probability of selecting appropriate anchor nodes to help with localization during the first localization sub-process. Figure 16(c) examines average energy consumption of the three algorithms under different communication ranges. When the communication range varies from 70m to 100m, average energy consumption of the three algorithms grows rapidly because more upgrade anchor nodes broadcast coordinates messages to help with localization. When the number of upgrade anchor nodes increases to a certain point, the majority of unknown nodes have been localized, and fewer unknown nodes need to broadcast localization request messages; thus, the average energy consumption grows more slowly Impact of trust value threshold For UASNs that require highly precise location information, the trust value threshold should be set to a relatively high value. However, if the trust value threshold is too large, the available upgrade anchor nodes will be greatly reduced, which results in a dramatic decrease in localization ratio. Figure 17(a) compares the localization ratios of the three algorithms under different trust value thresholds. At the beginning, average localization ratios decrease slowly with the increase of the trust value threshold, after the trust value threshold reaches 0.95, the localization ratios of the three algorithms decrease abruptly. Because an unknown nodes is accounted to be localized successfully unless its calculated trust value is larger than the trust value threshold; thus, when the trust value threshold is set to 1, the localization ratios of the three algorithms drop to zero. With the increase of the trust value threshold, the available upgrade anchor nodes decrease significantly. Thus, more non-localized unknown nodes have to rely on the fourth localization sub-process to estimate their coordinates, which are not accurate. As a result, the average localization errors of MANCL_C and MANCL_V algorithms decrease slowly as shown in Figure 17(b). For the LSL algorithm, the average localization error also decreases with the increase of the trust value threshold, because the increase of the trust value threshold can reduce the accumulation error caused by upgrade anchor nodes. Besides, the LSL algorithm does not have a 3D DV-hop distance estimation process. However, when the trust value threshold is set to 1, few unknown nodes can be localized successfully, which results in zero localization ratios; thus, at this time it is meaningless to discuss the average localization errors of the three algorithms. Figure 17(c) suggests that before the trust value threshold reaches 0.95, the average energy consumption of the three algorithms changes little. Because the trust value threshold has a little influence on average energy 69 Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

18 G. Han et al. Multi-anchor nodes collaborative localization algorithm (a) (b) (c) Figure 1. Impact of the hop count threshold and the interval hop count: (a) localization ratio, (b) average localization error, and (c) average energy consumption. consumption as long as there are enough anchor nodes to maintain the localization process. However, when the trust value threshold increases to 0.95, average energy consumption of the LSL, MANCL_C, and MANCL_V algorithms increases rapidly because more unknown nodes cannot be localized successfully; they have to send more localization request messages, which results in the increase of the average energy consumption Impact of hop count threshold and interval hop count The earlier simulations illustrate that the performance of MANCL_V algorithm is better than that of MANCL_C algorithm, in general. Thus, in this set of simulations, we use the MANCL_V algorithm to study the impact of hop count threshold and interval hop count. We change the hop count threshold from 5 to 9 and the interval hop count from 2 to 4 to evaluate the performance of the MANCL_V algorithm in terms of localization ratio, average localization error, and average energy consumption. The results are depicted in Figure 1. From Figure 1(a), we can conclude that for a fixed hop count threshold, the localization ratio of MANCL_V algorithm decreases with the increase of interval hop count. The results also indicate that the localization ratio of MANCL_V algorithm increases with the increase of hop count threshold. This is because the average one-hop anchor node distance calculated by marking nodes is more accurate, thus, more localized unknown nodes can become upgrade anchor nodes to help with localization. Figure 1(b) presents the average localization error of MANCL_V algorithm under different interval hop counts and hop count thresholds. With the increase of hop count threshold, the average localization error decreases correspondingly. The smaller the interval hop count is, the more marking nodes on the forwarding primary path of R 2 there are. Thus, the estimated coordinates of unknown nodes are more accurate, which leads to smaller localization error. Although more marking nodes on the forwarding primary path of R 2 can reduce the average localization error, it also results in higher energy consumption due to more forwarding data packages and more calculations, as depicted in Figure 1(c). 6. CONCLUSIONS AND FUTURE RESEARCH ISSUES In this paper, we present a collaborative localization algorithm for large-scale sparse UASNs. The MANCL algorithm divides the whole localization process into four sub-processes: unknown node localization process, iterative location estimation process, improved 3D Euclidean distance estimation process, and 3D DV-hop distance estimation process based on two-hop anchor nodes. The improved Euclidean distance estimation process consists of two mechanisms (a communication mechanism and a vote mechanism) to effectively determine temporary coordinates of unknown nodes during the Euclidean distance estimation process. Simulations illustrate that the performance of MANCL_V algorithm is better than that of MANCL_C algorithm, in general. During the 3D DV-hop distance estimation process, we use the average two-hop anchor node distance to help with unknown node localization. The results indicate that the MANCL algorithm can achieve a high localization ratio with a relatively small average localization error and low average energy consumption in sparse UASNs. We also investigate the impact of anchor node ratio, unknown node density, communication range, trust value threshold, hop count threshold and interval hop count on localization ratio, average localization error and average energy consumption. In real applications, we can select different network parameters according to the simulation results to meet applications requirements. The future research issues of localization algorithms possibly are: Because underwater sensor nodes inevitably move with water current, designing a mobility model to consider and simulate real water conditions is an important issue. The theoretical analysis in this paper only offers a mathematical tool to study the localization error Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 699

19 Multi-anchor nodes collaborative localization algorithm G. Han et al. caused by propagation delay, the distribution model should be analyzed deeply based on actual tests to reduce localization errors in real applications. Because underwater sensor nodes are expensive and the costs quickly rise for deep water, mobile anchor nodes assisted localization algorithms should be specifically designed. ACKNOWLEDGEMENTS The work is supported by Qing Lan Project, Natural Science Foundation of JiangSu Province of China (no. BK ), and The Applied Basic Research Program of Nantong Science and Technology Bureau (no. BK ). REFERENCES 1. Erol-Kantarci M, Mouftah HT, Oktug S. A survey of architectures and localization techniques for underwater acoustic sensor. Communications Surveys and Tutorials COMSUR 2011; 13(3): Tan H, Diamant R, Seah WKG, Waldmeyer M. A survey of techniques and challenges in underwater localization. Ocean Engineering 2011; 3(14 15): Guo Z, Luo H, Hong F, Yang M, Lionel ML. Current progress and research issues in underwater sensor networks. Computer Research and Development 2010; 47 (3): Chen B, Pompili D. Minimizing position uncertainty for under-ice autonomous underwater vehicles. Computer Networks 2013; 57(1): Teymorian AY, Cheng W, Ma L, Cheng X, Lu X, Lu Z. 3D underwater sensor network localization. IEEE Transactions on Mobile Computing 2009; (12): Jiang J, Han G, Zhu C, Dong Y, Zhang N. Secure localization in wireless sensor networks: a survey. Journal of Communications 2011; 6(6): Gopakumar A, Jacob L. Power-aware range-free wireless sensor network localization using neighbor distance distribution. Wireless Communications and Mobile Computing 2013; 13(5): Bao H, Zhang B, Li C, Yao Z. Mobile anchor assisted particle swarm optimization (PSO) based localization algorithms for wireless sensor networks. Wireless Communications and Mobile Computing 2012; 12(15): Guo Y, Liu Y. Localization for anchor-free underwater sensor networks. Computer & Electrical Engineering 2013; 39(6): Waldmeyer M, Tan H, Seah WKG. Multi-stage AUV-aided localization for underwater wireless sensor networks. In 2011 IEEE Workshops of International Conference on Advanced Information Networking and Applications, Biopolis, 2011; Zhang C, Han G, Jiang J, Shu L, Liu G, Rodrigues JJPC. A Collaborative localization algorithm for underwater acoustic sensor networks. In 2014 IEEE International Conference on Computing, Management and Telecommunications, ComManTel 2014, Da Nang, Vietnam, 2014; Zhou Z, Cui J, Zhou S. Efficient localization for largescale underwater sensor networks. Ad hoc Networks 2010; (3): Han G, Jiang J, Shu L., Xu Y, Wang F. Localization algorithms of underwater wireless sensor networks: a survey. Sensors 2012; 12(2): Luo H, Guo Z, Dong W, Hong F, Zhao Y. LDB: localization with directional beacons for sparse 3D underwater acoustic sensor networks. Journal of Networks 2010; 5(1): Bian T, Venkatesan R, Li C. Design and evaluation of a new localization scheme for underwater acoustic sensor networks. In Global Telecommunications Conference, Honolulu, USA, 2009; Erol-Kantarci M, Oktug S, Vieira L, Gerla M. Performance evaluation of distributed localization techniques for mobile underwater acoustic sensor networks. Ad Hoc Networks 2011; 9(1): Yang K, Guo Y, Wei D, Jin Y. MFALM: an active localization method for dynamic underwater wireless sensor networks. Computer Science 2010; 37(1): Callmer J, Skoqlund M, Gustafsson F. Silent localization of underwater sensors using magnetometers. Advances in Signal Processing 2010; 2010(1): Cheng W, Teymorian AY, Ma L, Cheng X, Lu X, Lu Z. Underwater localization in sparse 3D acoustic sensor networks. In The 27th Conference on Computer Communications, Phoenix, AZ, USA, 200; Teymorian AY, Cheng W, Ma L, Cheng X, Lu X, Lu Z. 3D underwater sensor network localization. IEEE Transactions on Mobile Computing 2009; (12): Benslimane A, Saad C, Konig J, Boulmalf M. Cooperative localization techniques for wireless sensor networks: free, signal and angle based techniques. Wireless Communications and Mobile Computing 2014; 14(17): Liang Q, Leung VCM, Meng W, Adachi F. Cooperative communications and sensing. Wireless Communications and Mobile Computing 2014; 13(13): Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd.

20 G. Han et al. Multi-anchor nodes collaborative localization algorithm 23. Zhu S, Ding Z. Distributed cooperative localization of wireless sensor networks with convex hull constraint. IEEE Transactions on Wireless Communications 2011; 10(7): Tan X, Li J. Cooperative positioning in underwater sensor networks. IEEE Transaction on Signal Processing 2010; 5(11): Zhou Z, Peng Z, Cui J, Shi Z, Bagtzoglou AC. Scalable localization with mobility prediction for underwater sensor networks. IEEE Transactions on Mobile Computing 2011; 10(3): Ibrahim S, Liu J, Al-Bzoor M, Cui J, Ammar R. Towards efficient dynamic surface gateway deployment for underwater network. Ad Hoc Networks 2013; 11(): Santhakumar M, Asokan T. Power efficient dynamic station keeping control of a flat-fish type autonomous underwater vehicle through design modifications of thruster configuration. Ocean Engineering 2013; 5: Dhurandher SK, Obaidat MS, Gupta M. Energized geocasting model for underwater wireless sensor networks. Simulation Modelling Practice and Theory 2013; 27: Wang X, Shang J, Luo Z, Tang L, Zhang X, Li J. Reviews of power systems and environmental energy conversion for unmanned underwater vehicles. Renewable and Sustainable Energy Reviews 2012; 16(4): AUTHORS BIOGRAPHIES Guangjie Han received the Ph.D. degree from Northeastern University, Shenyang, China, in From 2004 to 2006, he was a Product Manager for the ZTE Company. In February 200, he finished his work as a Postdoctoral Researcher with the Department of Computer Science, Chonnam National University, Gwangju, Korea. From October 2010 to 2011, he was a Visiting Research Scholar with Osaka University, Suita, Japan. He is currently a Professor with the Department of Information and Communication System, Hohai University, Nanjing, China. He is the author of over 130 papers published in related international conference proceedings and journals, and is the holder of 55 patents. His current research interests include sensor networks, computer communications, mobile cloud computing, and multimedia communication and security. Dr. Han has served as a Cochair for more than 20 international conferences/workshops and as a Technical Program Committee member of more than 70 conferences. He has served on the Editorial Boards of up to 16 international journals, including the International Journal of Ad Hoc and Ubiquitous Computing, Journal of Internet Technology and KSII Transactions on Internet and Information Systems. He has served as a Reviewer of more than 50 journals. He received the 2014 Second International Conference on Computing, Management, Computing, Communications and IT Applications Conference and Telecommunications and International Conference on Communications and Networking in China Best Paper Awards. He is a member of the Association for Computing Machinery. Chenyu Zhang is currently pursuing her Master s degree in Department of Communication and Information System at Hohai University, China. She received B.S. degree in Communication Engineering from Hohai University, China, in Her current research interests are localization and path planning for Wireless Sensor Networks. Tongqing Liu is in the Suzhou Keda Technology Co., Ltd, China. He recevied his Master s degree from Department of Communication and Information System from Hohai University, China, in His current research interests is localization for Wireless Sensor Networks. Lei Shu received the Ph.D. degree from the National University of Ireland, Galway, Ireland, in Until March 2012, he was a Specially Assigned Researcher with the Department of Multimedia Engineering, Graduate School of Information Science and Technology, Osaka University, Japan. Since October 2012, he has been with Guangdong University of Petrochemical Technology, Maoming, China as a Full Professor. Since 2013, he has been a Ph.D. Supervisor with Dalian University of Technology, Dalian, China, and a Master Supervisor with Beijing University of Posts and Telecommunications, Beijing, China. He has also been the Vice Director of the Guangdong Provincial Key Laboratory of Petrochemical Equipment Fault Diagnosis, Guangdong University of Petrochemical Technology. He is the Founder of the Industrial Security and Wireless Sensor Networks Laboratory. He is the author of over 200 papers published in related conference proceedings, journals, and books. His current H-index is 1. His research interests include wireless sensor networks, multimedia communication, middleware, security, and fault diagnosis. Dr. Shu served as a Cochair for more than 50 various international conferences/workshops, e.g., the Wirel. Commun. Mob. Comput. 2016; 16: John Wiley & Sons, Ltd. 701