Research Article
A mechanism of topology optimization for underwater acoustic sensor networks based on autonomous underwater vehicles
International Journal of Distributed Sensor Networks 2017, Vol. 13(1) Ó The Author(s) 2017 DOI: 10.1177/1550147716686979 journals.sagepub.com/home/ijdsn
Ming He1,2,3, Fangxin Liu1,2, Zhuang Miao1, Huan Zhou4 and Qiuli Chen1,2,3
Abstract Limited energy, high mobility, and unsteady acoustic communication links render the underwater acoustic sensor networks subject to performance reduction. To solve this problem, a topology optimization method, also called TO-A algorithm, is developed based on topology reconfiguration. First, the proposed method optimizes the coverage rate of underwater acoustic sensor networks by adjusting the location of sensor nodes through simulating fish behaviors. Second, to optimize the connectivity of underwater acoustic sensor networks, the proposed method, using edge nodes, repairs disconnected positions and eliminates key nodes in underwater acoustic sensor networks. Third, a topology optimization strategy, also called TO-DA algorithm, is developed for Double-autonomous underwater vehicles to improve the robustness and adaptability of the network topology. When the inherent law of underwater acoustic sensor networks topology formation is further found, the method for optimizing network topology is proposed, which based on the triangle principle eliminates those key nodes and can help with the survivability of network regeneration. The method proves reasonable and valid by stimulation and contrast examinations. The comparison of TO-A algorithm and TO-DA algorithm shows that under low energy consumption, TO-A algorithm can keep the network coverage at about 97% for a long time, connectivity rate above 89%, while the TO-DA algorithm can improve the survivability of the network by above 50% at the expense of 8.5% of the network coverage. Keywords Underwater acoustic sensor networks, Double-autonomous underwater vehicles, topology optimization
Date received: 14 September 2016; accepted: 8 December 2016 Academic Editor: Hongyi Wu
Introduction Underwater acoustic sensor networks (UASNs) are capable of underwater environment in real time, accurate and effective monitoring, and can be widely applied to various fields such as underwater military target surveillance,24 marine data collection, monitoring water quality, seabed mineral resources exploration, and assisted navigation.1,25–30 The quality of communication among the sensor nodes and the reliability of topology of UASNs have been seriously affected due to
1
College of Command Information Systems, PLA University of Science and Technology, Nanjing, China 2 Nanjing University of Information Science and Technology, Nanjing, China 3 The 61th Research Institute of PLA, Beijing, China 4 Tongji University, Shanghai, China Corresponding author: Fangxin Liu, College of Command Information Systems, PLA University of Science and Technology, No. 1 Haifu Lane, Guanghua Road, Qinhuai District, Nanjing 210007, China. Email:
[email protected]
Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License (http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (http://www.uk.sagepub.com/aboutus/ openaccess.htm).
2
International Journal of Distributed Sensor Networks
the factors of complex marine environment, such as the high latency, low bandwidth, and easy attenuation of signal and multipath effects, seriously affecting.15–23 Currently, optimization for terrestrial sensor networks has some achievements. Fu and Han2 solve the sensor node–distributed optimization problems of the terrestrial sensor network by quantum genetic algorithm; Aziz et al.3 consider comprehensively the battery power of sensor and other factors, then employing the distributed topology control technology to optimize the energy consumption of wireless sensor networks; Wang et al.4 present a parallel particle swarm optimization (PSO) strategies based on the mobility of sensor node, so that distribution of the node in sensor networks reaches local optimum; Mi and Zhou5 proposed a terrestrial wireless sensor network routing algorithm multi-objective optimization. Considering the particularity of UASNs in complex environments and underwater acoustic communication mode, Pioneers’ achievements in the optimization strategy are not fully applicable to UASNs. As adopted in Jafri et al.,6 the connectivity of UASNs has been optimized from the aspect of protocol design, but there are key nodes in the UASNs, making the invulnerability of network topology unable to be improved (the key node means all single-hop neighbor nodes of this node belong to subsets which do not communicate with each other with the number of subsets at k(k 2) and these subsets are connected by this node only). As adopted in Xia et al.,7 the fish-inspired underwater acoustic sensor node deployment algorithm optimized the coverage of topology of the UASNs. As adopted in Manjula and Sunilkumar,8 the particle swarm algorithm achieved optimization for the layout of the underwater acoustic sensor node. Both algorithms make the distribution density of the sensor nodes and the distribution density of the target event to attain a nice match so that the approximate optimal for the coverage of UASNs is realized. However, there are also some limitations as follows: (1) (2)
(3) (4)
It cannot ensure the communication between the all sensor nodes; Throughout the life of UASNs always exist the key nodes, and this problem has not been effectively solved; The interruption phenomenon happens during data collection of sensor nodes; On the condition of being difficult to replace the equipment (e.g. the sensor) under water, it cannot maintain effective monitoring during the long-time mission.
This article takes into consideration both relevant features of UASNs and topological reconstruction of the underwater non-uniform coverage of UASNs. Moreover, we use the autonomous underwater vehicles
(AUVs) to present two different methods for the topological optimization of two kinds of UASNs based on, respectively, topology reconfiguration (TO-A algorithm) and Double-autonomous underwater vehicle (Double-AUV); a new version of AUVs (TO-DA algorithm) is proposed in this article, which can overcome those limitations mentioned above. TO-A algorithm is divided into two stages: in the first stage, the artificial fish swarm algorithm (AFSA) is adopted to readjust the position of the sensor nodes that are inspired by node deployment algorithm7 and then the coverage of the UASNs is optimized; in the second stage, the entire connectivity of the UASNs is restored using edge node in the UASNs to repair the disconnected position and eliminate key nodes to improve the invulnerability of network, so as to made up for the shortage of AFSA in the first stage. Shadow zones cause high bit error rates, loss of connectivity, and dramatical impact,9,31–33 so this article presents a new version of AUVs, called Double-AUVs (see Figure 1). By this way, (1) the connectivity among UASNs inside the shadow zone is established by Double-AUVs so as to avoid an information collection vacuum period generated by shadow zone while its carrying out mission; (2) a topology optimization mechanism based on Double-AUVs is adopted which is proposed in this article and an UASN’s key node is eliminated by triangle dilute methods to extend the life cycle of UASNs by sacrificing part of the coverage of the UASNs. The remainder of this article is organized as follows. In section ‘‘Network model and problem description,’’ we present our system model and raise the problem that needs solving. In section ‘‘The TO-A algorithm,’’ our new algorithm (TO-A algorithm) is described in detail. In section ‘‘The TO-DA algorithm,’’ we present our mechanism of topology optimization. An empirical evaluation of TO-A algorithms and TO-DA algorithm is provided in section ‘‘Experimental study.’’ The article concludes with recommendations for future work in section ‘‘Conclusion.’’
Network model and problem description Network model m events, to be monitored, are unevenly distributed in the target area D. Monitoring event set is defined as T = ft1 , t2 , t3 , . . . , tm g, (ti 2 D, i = 1, 2, . . . , m); the UASNs consist of underwater acoustic sensors implemented for coverage of the event. The set of sensors is defined as S = fs1 , s2 , s3 , . . . , sn g; they are deployed with k AUVs (AUVs denote normal AUVs in TO-A algorithm, and Double-AUVs in TO-DA algorithm). AUVs are defined as A = fa1 , a2 , a3 , . . . , ak g in TO-A algorithm, at the same time Double-AUVs are defined as A = fa1 , a2 , a3 , . . . , ak g and B = fb1 , b2 , b3 , . . . , bk g.
He et al.
3
B
A
Double -AUV
B
Double -AUV
Figure 1. The reason for Double-AUVs.
B stands for the sensor set installed on Double-AUVs. These two types of AUVs collectively stand for AUV nodes in the model. Sensors and AUVs are selected using an omni-directional Boolean sensing model. rs and rc represent maximum sensing radius and communication radius of sensor nodes, respectively. ras refers to maximum sensing radius, while rac denotes the communication radius of AUV nodes. Definition 1. Coverage is represented as the ratio of the total event monitored by sensor nodes arranged in the monitoring area and the total number of events actually exist in the monitoring area.34,35 Definition 2. Neighborhood topology is represented as local network topology that is constituted of neighbor nodes directly connected to sensor node Si . It is defined as F(si ) = G(L(si ), E(si )), where L(si ) denotes set of nodes in the neighborhood of si and sj denotes set of links in the neighborhood of si . In order to match with the actual situation, the energy consumption of sensor nodes in coordinate axis (include x, y, and z axes) moving per unit distance is, ! ! ! ! ! respectively, defined as Cx , Cx , Cy , Cy , Cz , and ! Cz . The energy consumption of sensor node move per ! unit distance in any direction (defined as Ci ) can be calculated by the vector-calculation. Definition 3. The total energy consumption of movement of sensor nodes in UASNs is measured as follows C=
n X X ~ ~ C i 3 d i
ð1Þ
i = 1 8d
! where j di j represents moving distance of node Si in a certain direction during the process of optimization, and disi represents the total distance of node Si moves. In other word, disi is the total value of movingP distance ! of node Si in all directions, measured as disi = 8d j di j.
In this article, sensor nodes are divided into three categories: key nodes, edge nodes and ordinary nodes. Definition 4. If the sensor nodes are removed to increase the number of connectivity of UASNs branch, such kinds of nodes are called key nodes. Definition 5. Peripheral node (or the importance of node is low), in the other word, removing the sensor nodes causes neither increasing the number of connectivity of UASNs branch nor generating a key node, is called edge node. Definition 6. In addition to the edge nodes and the key nodes, other nodes are known as the ordinary nodes. Hypothesis 1. Underwater acoustic sensors have function about auxiliary locating. The node can exchange information between adjacent nodes to get the set L(si ). Through the set L(si ), we can know the location of their own and other nodes. Hypothesis 2. AUVs node can communicate directly with the base station on the water. Hypothesis 3. In the process of mission, the energy of AUVs node approximates infinitely large (represent + ‘).36–38 Hypothesis 4. There will be no new failure while topology optimization is performed by AUVs. Hypothesis 5. There is no data loss when communication occurs among nodes. Hypothesis 6. rc \rac . Meanwhile, AUVs can adjust the radius of perception and the radius of communication to the maximum range according to actual requirements. Given the high cost of AUVs, the number of AUVs should be much smaller than the number of sensors, namely K\\N.
4
International Journal of Distributed Sensor Networks
(a)
Figure 3. Neighborhood topology of s7 and the neighborhood node reachability matrix: (a) the neighborhood topology of node S7 and (b) the reachable matrix in the neighborhood of node S7 .
Figure 2. Local network topology: (a) the local topology.
Relevant judgment Calculation of neighborhood topology of nodes. According to the Hypothesis 1, the sensor node si knows its own location and location of the sensor in its neighborhood, so we can compute the corresponding neighborhood topology as follows F(si ) = G(L(si ), E(si ))
(b)
ð2Þ
There is an example (see Figure 2): node S7 knows location of node S6 , node S8 , and node S9 , and node S6 knows location of node S2 , node S5 , and node S7 ; so, we know the length of link e7, 8 (s7 , s8 ), the length of link e7, 9 (s7 , s9 ) and the angle between link and link, set as; then, we can calculate the angle between link e7, 8 (s7 , s8 ) and link e7, 9 (s7 , s9 ) and set as a according to cosine theorem and sine theorem; finally, we can know clearly the angle between link e7, 9 (s7 , s9 ) and link e8, 9 (s8 , s9 ), set as g. Meanwhile, we can judge whether a communications link exists between node s8 and node s9 by equation (3). We can obtain their neighborhood topology F(s7 ) according to the state in equation (3) 8 dis(s 8 , s9 ) = > qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi > < len(e7, 8 )2 + len(e7, 9 )2 2len(e7, 8 )len(e7, 9 )cosa c > > : dis(s8 , s9 ) rc , state : connected dis(s8 , s9 ).r , state : disconnected ð3Þ
Judgment for the key node and the edge node. Uneven characteristics of actual network are reflected in the network model, so there are key nodes and non-key nodes. In this article, we have a more detailed classification of the non-key nodes, the ordinary nodes, and the edge nodes. In order to prevent the network topology from breaking off after the node has been removed, the neighborhood topology of the node is supposed to be connected. That neighborhood topology of the node is connected is the sufficient condition to ensure that the network topology is still connected after the node was removed. We can obtain neighborhood topology F(si ) of node si
by the above method and then we can get the reachability matrix according to neighborhood topology F(si ). As shown in Figure 3, (a) represents the neighborhood topology of node s7 ; (b) represents the reachable matrix in the neighborhood of node s7 . The reachable matrix formalism describes the reachable relation among the nodes in the neighborhood topology of certain nodes. The set of neighbor of the node Si is represented as L(si ) = fsj jd(sj , si ) rc , j = 1, 2, . . . , ng and a judgment set is established as P(si ) = ;, according to reachable matrix of nodes in neighborhood of node si , and the elements of the reachable matrix of the node si are represented by Aij : (1) (2)
(3) (4)
Node sj , that matrix elements satisfy Aij = 1 is added to the set P(si ); Check the values of Akj in the set P(si ) where the node sk exists in the reachable matrix F(si ). If Akj = 1, add node sj to the set P(si ); Repeat above procedure until all nodes in the set L(si ) are visited; Compare the set P(si ) and the set L(si ). If Pðsi Þ = L(si ), it indicates that the neighborhood topology of node si still remains connected after the node si is removed and the node si is a non-key node. Then continue to judge whether the node si is the edge node; if P(si ) 6 = L(si ), it proves that the node si is a key node, marked the node as si .
For example (see Figure 2), the set of neighbor of the node s7 is L(s7 ) = fs6 , s8 , s9 g, yet P(s7 ) = fs8 , s9 g, so it is obviously P(si ) 6¼ L(si ). We know node s8 and node s9 belong to the same connected branch, but node s6 is an isolated node. As a result, the node s7 is judged to be the key node and the failure of node s7 leads to large probability of network topology segmentation. After determining the node si is not the key node, make judgment about edge node. Assuming node sj as the neighbor node of node si , its neighborhood topology is F(sj ), delete si from F(sj ), and then determine whether the node sj is the key node by the method of
He et al.
5
judging the key nodes. If the node sj is the key node, it is shown that the removal of si will generate a new key node. So, si is not an edge node; otherwise, si is the edge node, which is marked. A method of localization of underwater sensor network is proposed by Heidemann et al.,10 and its effectiveness is verified by experiments on the sea. In this article, we use AUVs to patrol the mission area, gathering information about the marine environment, and the location of the nodes. And the location of the disconnected position is located by the method mentioned above.10
used as food, the process of sensor nodes tending to events is equivalent to the process of artificial fish looking for food, and the sensor node simulates fish foraging, rear-end collision, and clustering behavior in order to complete the adjustment of positions. The perception of artificial fish is greater than its communication range (because the perception of sensor nodes is greater than its communication range), set to visual (the sphere range that artificial fish itself is regarded as the center of the sphere and rs as the radius), the step ! size of maximum movement is lstep , Li represents the current position of the artificial fish si , and the number of iterations is MaxIter.
Goal description
Definition 7. Fitness is the number of targets covered by a sphere with a radius rs for a sensor in the target monitoring area. The fitness can be expressed by equation (4)
The UASNs topology changes frequently, resulting in the decrease of network performance due to the complex ocean environment. In order to improve the performance of UASNs topology and prolong the network lifetime, the network topology optimization mechanism based on AUVs is proposed (for both TO-A and TODA algorithms). Two optimization processes of the algorithm are shown in the following figure. Figure 4 is a schematic diagram of the optimization process of two algorithms.
Nsensor (si ) = ftj jd(si , tj ) rs , j = 1, 2, . . . , mg
ð4Þ
Definition 8. The number of sensors that covers target event tj is called covering multiplicity of tj , covering multiplicity can be expressed by equation (5) Kej =
n X
J (d(si , tj ) rs )
ð5Þ
i=1
The TO-A algorithm Coverage optimization for UASNs inspired by fish swarm
where J () represents the indicator function. When the condition in brackets is satisfied, its value was 1, otherwise, its value was 0.
In the process of UASNs coverage optimization inspired by fish swarm, the sensor nodes are treated as artificial fish, events in the target monitoring area are
Definition 9. The allowable degree of crowding around a location in a target monitoring area is called congestion factor.
Figure 4. The flowchart of topology optimization.
6
International Journal of Distributed Sensor Networks
The allowable degree of crowding is known as the congestion factor, around a location in the target monitoring area. The congestion factor of the sensor A is expressed as d(si ) = K deg 3 Nsensor (si )
ð6Þ
where the constant K deg is the expected covering multiplicity of a single event and Nsensor (si ) represents the number of events in the coverage of sensor si . Nneigh (si ) represents the number of sensor nodes in the neighborhood of the sensor si and Nnear (si ) represents the number of sensor nodes in the sensing range of sensor si . The process of coverage optimization of UASNs inspired by fish swarm is as follows: Step 1 Nsensor (si ) = 0 can be divided into three cases. Step 1.1 Nneigh (si ) = 0, then perform foraging operations: randomly chosen to move in any direction when the step size does not exceed lstep in the perception range visual of the artificial fish and the new position is expressed as ! ! ! L0 i Li ! L0 i = Li + rand(lstep ) ! ! 0 L i Li
ð7Þ
where rand(lstep ) represents a random number from 0 to the maximum step size lstep . If the artificial ! fish is moved, which means that the sensor node to L0 i makes the Nsensor (si ) increase, the foraging operation was successful. Otherwise, the foraging operation failed. Step 1.2 Nneigh (si ).0, then perform rear-end collision operation: find the optimal neighbor nodes sb , that is to say, sb = argmaxsj 2L(si ) fNsensor (sj )g, if Nsensor (sb ). Nsensor (si ) and Nnear (sb )\d(sb ), it is proved that the number of covering events of sensor node sb is more than the number of covering events of sensor node si , and the location of the sensor node sb is not crowded, then move sensor node si to the location of sensor node sb at a step, the following operation is performed ! ! ! Lb Li ! L0 i = Li + rand(lstep ) ! ! Lb Li
ð8Þ
! If sensor node si moves to L0 i , making the Nsensor (si ) increased, the rear-end collision operation was successful; otherwise, the rear-end collision operation failed. Step 1.3 Nneigh (si ).1, then perform clustering operation: determine the location of the neighbor node center sc by equation (9) P ! Lj ! Lc =
sj 2L(si )
Nneigh (si )
ð9Þ
Respectively, the values of Nsensor (sc ) and Nnear (sc ) of the center position sc are calculated by equations (10) and (11) P Nsensor (sj ) Nsensor (sc ) =
Nnear (sc ) =
sj 2L(si )
Nneigh (si ) P Nnear (sj ) sj 2L(si )
Nneigh (si )
ð10Þ
ð11Þ
If Nsensor (sc ).Nsensor (si ) and Nnear (sc )\d(sc ), it proves the number of covering events of sensor node sc is more than the number of covering events of sensor node si , and the location of the sensor node sc is not crowded. Then, move sensor node si to the location of sensor node sc at a step. The following operation is performed ! ! ! Lc Li ! L0 i = Li + rand(lstep ) ! ! Lc Li
ð12Þ
! If the sensor node si moves to L0 i , making the Nsensor (si ) increased, the clustering operation was successful; otherwise, the clustering operation failed. Step 2 Nsensor (si ) = 0, the sensor node si moves to the center position of the event covered. Then, perform Steps 1. Each sensor node in the target area performs the steps mentioned above. After a total of 100 iterations, the process of UASNs coverage optimization is over. In this process, each node can control itself, in accordance with the rules of the algorithm to get the global optimal value. It requires the step size of each time node to randomly move a size not exceeding its own perceptual range, so as to ensure that any event during the process of foraging operation is not skipped, and when rear-end collision or clustering occurs, there will not be too crowded among sensor nodes.
Connectivity optimization for UASNs based on edge nodes After coverage optimization for UASNs inspired by fish swarm is completed, though the network coverage has been optimized, there are disconnected positions and key nodes existing in the UASNs. Repair the disconnected position and eliminate the key nodes by edge nodes, so that the connectivity of UASNs can be optimized. Selection of the repair node. The repair node is selected from edge nodes, which is used to repair the disconnected position and remove the key nodes in the UASNs. The process of the selection of repair node is shown in Figure 5.
He et al.
7
Figure 5. Schematic diagram of the process of repair node selection.
1. Initiating the request of repair operation, that is, Request messages Broadcast the Request(si , Li , hops) message by the key node, including key node’s ID, location, hops of search, and other information, where hops is the counter for counting the number of hops and the initial value of hops is set by si , the Request(si , Li , hops) message is sent each time, and the hops minus 1, until the hops becomes 0, and the Request(si , Li , hops) message is no longer being delivered. 2. The edge node processes the Request message Edge node sj within the number of hops receives the Request message, the distance between si and sj is calculated according to the position information Li of si and its own position information Lj , and then calculate ! ! the C 0 = k d(si , sj ) jCi j, where jCi j represents the energy consumption moving per unit distance along the ! direction of sj to si , d(si , sj )jCi j represents estimation of energy consumption of moving, and k represents the adjustment coefficient, used to control the remaining energy while edge node sj is moving to si . sj is compared with C 0 regarding the current value of their own energy. If C 0 is greater than the current own energy, it is indicated that sj does not have enough energy to move to si , or even if it can move to the C, it cannot continue to work properly, then do not save the Request message
and forward it; if C 0 is less than the current own energy, then check the list of messages. If the list has received the same Request(si , Li , hops) message before, then forward it. Otherwise, it shall be added to the list and forwarded, until hops of Request(si , Li , hops) message was reduced to 0 when the message would be discarded. Edge nodes need to establish the queue of Request messages and set timer of Request message. The queue of the Request messages is constructed with each key node further away from sj . The shorter the distance that sj moves to each key node, the lesser the time that edge node spends to arrive at the key node, and then the less the time spent on local topology restoration. We use the following equation (13) to calculate the time that is required to complete the local topology restoration by the edge node sj D P
Trepair = Tcomput D +
i=1
v
li ð13Þ
where D represents the total number of steps that the edge node sj moves; Tcomput represents the time for the edge node sj to calculate the cost of moving direction each time; v represents the moving speed of the edge node P sj ; li represents the ith step size of the edge node sj ; D i = 1 li represents the total movement distance of the edge node sj . While the timer of the Request message expires, the edge node sj selects a Request message from the queue
8
International Journal of Distributed Sensor Networks
of Request messages and replies the Response(si , sj , Lj ) message to its sponsors si , including the key node ID, the edge node ID, and location information of the edge node. 3. The key node processes the Response message The key node takes into account energy, movement distance, and utilization of coverage of the edge node. In this way, it can construct the queue of Response messages and selects a number of edge nodes from the queue to send the Take message to itself. The key node sets the timer of the Response message after the Request message is sent. If the key node has not received any Response message that is returned by the edge node during a certain period of time measured by the timer, then the key node expands the search range, increases the initial value of hops, and broadcasts the Request message again. 4. The edge node processes the Take message The timer of the Take message is set at the same time as the edge node sends the Response message. If the edge node has received the Take message during a certain period measured by the timer, the edge node is selected as the node for repairing the key node that has sent the Take message and started the process of moving; when the timer expires and the edge node does not receive the Take message, the next key node from the queue of the Request messages will be selected to send the Response message and the timer of the Take message will be started. The process of the selection of the repair node for repairing the disconnected position is similar to the above process, but the request of repair (namely the Request message) is initiated by the AUVs that discover the disconnected position and information interaction between the AUVs and edge nodes. The movement of the repair node oriented by virtual force. In the following description, we define the disconnected position of UASNs and the position of the key node as the target position. When the repair node is selected, the repair node begins to move to the target position. We assume that the nodes will be affected by four kinds of virtual force in the process of moving. They are the gravitation of the target position, repulsion among neighboring nodes, repulsion of obstacles, and random disturbance force. The target position will generate gravitation based on Hooke’s law for the repair node to attract the repair node to move it; the repulsion will be generated among single-hop neighboring nodes based on Coulomb’s law; in order to avoid a collision during the movement, the
obstacle on the movement route of the repair node will generate repulsion based on Coulomb’s law; random disturbance force is used to adjust the operator for avoiding the repair node to fall into motion trap, and it is in the condition of long-term stop or reciprocating motion. Assuming that o represents the target position, sj denotes the repair node, and each virtual force is expressed as ! ! Fg (o, sj ) = kg d(o, sj ) 8 kr mi mj c ! < !2 , 0\d(si , sj )\r Fr (si , sj ) = d(si , sj ) : 0, d(si , sj ) rc 8 ko mi mj c ! < !2 , 0\d(si , sj ) r Fo (si , sj ) = d(si , sj ) : 0, d(si , sj ).rc ! Fw (sj ) = rand(0, Qrand(0, 360
ð14Þ ð15Þ
ð16Þ ð17Þ
where kg , kr , and ko represent the coefficient of proportional; mi and mj represent the quality of the sensor ! node and mi = mj ; d(o, sj ) represents Euclidean distance ! between the repair node and the target position, d(si , sj ) represents Euclidean distance between the two sensor nodes; Q represents disturbance amplitude of random disturbance force. The resultant force acted by the repair node is expressed by equation (18). When the resultant force is 0 or in the opposite direction of the last resultant force, the node does not move. If it does not move kw times in a row, then a random disturbance occurs, and the step size and direction of the movement are generated randomly 8 P P ! ! ! > < Fr (si , sj ) + Fo (si , sj ) + Fa (o, sj ) ~ sj 2D(si ) ð18Þ F = sj 2L(si ) > ! :F w (sj ), kw = 3 The movement of the repair node is represented as follows ~ F P(t + Dt) = l + P(t) ~ F
ð19Þ
where P(t) represents the position coordinate of the repair node at time t, P(t + Dt) denotes the position coordinate of the repair node after Dt, ~ F=j~ Fj represents the unit vector of the moving direction of the repair node, and l represents step size of movement. The maximum unit step size of the movement oriented by virtual force is set to lmax , then absolutely l lmax . The repair node is in constant motion oriented by virtual force until it gets into the communication range of the target position and the mission of local topology restoration is completed.
He et al.
9
The selection of several important parameters such as kg , kr , ko , and lmax is described as follows: kg represents the gravitational coefficient of the target position. In order to ensure that the repair node, in a relatively short time, could move into the communication range of the target location, kg should be set to the maximum of the proportional coefficient (i.e. kg , kr , and ko ); kr and ko , respectively, indicate the repulsion among the single-hop neighboring nodes and the repulsion generated by the obstacle. The obstacle is still the sensor node. In order to make the repair node successfully avoiding obstacles, it is set as ko .kr . The setting of the maximum step size of the repair node is similar to that of the maximum moving step size of the artificial fish. If the setting is too small, it will increase the number of mobile, which leads to increased energy consumption; if the setting is too large, it will cause the node to move back and forth in the process of moving; because it is the movement of the sensor nodes and in order to facilitate unified throughout the optimization, we set lmax = lstep . The method for supplying the coverage. Although there is no impact on the connectivity of the network topology, it may lead to individual events unable to be covered and affect the coverage of UASNs. In order to fill up the gaps that may be left after the repair node is removed and ensure the network topology unchanged, the position of original neighbor nodes of the repair node can be fine adjusted by virtual force. In this process, the original neighbor nodes of the repair node are affected by two kinds of virtual force: (1) the gravitation of the original location of the repair node based on Hooke’s law; (2) the force among the neighbor nodes based on Coulomb’s law (whether it is the gravitation or the repulsion depends on the distance among the nodes). Two kinds of virtual force are expressed as equations (20) and (21). If the conditions of dopt = 1:78rs and rc .1:78rs are satisfied, the distance among the sensor nodes is dopt , which can ensure the connectivity and achieve the optimal coverage11 ! ! Fb (o0 , si ) = kb d(o0 , si ) 8 k mm c i j > > !2 , 0\d(si , sj ) dopt ! < d(si , sj ) Fc (si , sj ) = kc mi mj > > : !2 , dopt \d(si , sj ) d(si , sj )
ð20Þ
ð21Þ
where kb and kc are the coefficient of proportionality. In order to move the original neighborhood node of the repair node to the position of the original repair node, we should set kb .kc . Because the range of the movement of the node in the process is smaller, kb and kc are set to smaller order of magnitude.
The resultant force acted by the original neighbor nodes of the repair node is expressed as follows X ! ! 0! Fi = Fb (o , si ) + Fc (si , sj )
ð22Þ
sj 2L(si )
The position of nodes in the original neighborhood of the repair node is adjusted under the action of the resultant force, which fills up the gaps caused by removing the repair node to prevent the coverage of UASNs from being lowered.
The TO-DA algorithm The process of repair The reasons why we present the TO-DA algorithm for optimizing the topology of the UASNs are as follows: (1)
(2)
(3)
The underwater acoustic sensor has limited energy and the frequent topology evolution for maintaining the connectivity of the network topology leads to the energy of sensor nodes greatly reduced in complex ocean environment; The shadow zone still exists in the process of mission execution, and the communication signal attenuation among the underwater acoustic sensors or the signal-to-noise ratio (SNR) is large, which cannot maintain stable/effective communication. This is because the acoustic wave cannot penetrate the shadow zone with much signal distortion. The number of AUVs is limited because of its expensive cost contrast to the sensor node. Replacing the key nodes with AUVs leads to the weakening of the ability about global monitoring, data acquisition, and so on. Sometimes the implementation of topology reconfiguration is caused by the energy deficiency of individual underwater acoustic sensors or the poor communication caused by suspended matter in sea water.
In the view of the above, we present the TO-DA algorithm. It is the improvement of the TO-A algorithm. Each AUV carries a underwater acoustic sensor. They exchange the information and fix position by a wire. The wire can penetrate shadow zone without much signal distortion. The AUV can clean or charge the sensor that is connected to the AUV for avoiding the frequent topology evolution. When the sensor nodes in the monitoring area are monitored by DoubleAUVs, the information about the failure node is also recorded for statistics. If the disconnected node is found, then bi that the sensor attached to the
10
International Journal of Distributed Sensor Networks
Double-AUVs moves to the L(Si ) for repairing. Then, set the state of bi as busy. On this basis, if Double-AUVs find the disconnected position of the UASNs, proceed as follows: Case 1: The disconnected node does not belong to set B. Step 1 According to Hypothesis 1.The node bi knows its own location information and the location information of the nodes in the neighborhood. Then, the whole network is divided into smaller network topology, and the failure node si is judged to belong to the ith part of topology of the network. Step 2 In the ith part of topology of the network, according to the method mentioned above, the method of classification is used to determine the classification of the node bi (the key node or the edge node). Then, Double-AUVs find the disconnected position by the method is mentioned in section ‘‘Network model and problem description.’’ Step 2.1 If bi is the edge node Double-AUV Ai initiates the request of requirement which means that Ai sends the Request message with information about disconnected position to bi . After bi received the Request message, bi returns the Response message that is used to answer the request of repair. The time of repair is calculated by equation (23) D P
Trepair = Tcomput D +
i=1
v
li ð23Þ
Figure 6. The process of selecting node for reconstruction.
where D represents the needed moving steps of the node bi , Tcomput represents the time for the node bi to calculate the cost of moving direction each time, v represents the moving speed of the node bi , li represents the ith step D P li represents the total movesize of the node bi , and i=1
ment distance of the node bi . Step 2.2 If bi is the key node In the ith part of topology of the network, local topology reconstruction is carried out relatively to the whole topology network. After bi turns into the edge node, jump to Step 2.1. The process of the local topology reconstruction is as follows or shown in Figure 6: The key node bi sends Find message In the ith part of network topology of the key node bi , bi sends the Find(bi , Li , hops, Links) message to all nodes in the topology network i in addition to itself. The Find(bi , Li , hops, Links) message includes key node’s ID, location, hops of search, and other information, where hops is the counter for counting the number of hops, and the initial value of hops is set by bi . Every time the Find message passes by, the hops minus 1. The Find message is no longer being passed after hops becomes 0. The edge node sj processes the Find message After the edge node sj in the i part of topology of the network received the Find message sj , calculate the distance d(bi , sj ) between the bi and the sj according to the position information Li about bi and the position information Lj about itself and then calculate the total energy C 0 = k d(bi , sj ) jCi j. The edge node sj is compared with C 0 regarding the remaining energy of itself.
He et al. If C 0 is more than the remaining energy of sj , it is indicated the edge node sj does not have enough energy for moving to the position of bi or even if the edge node sj can move to the position of bi , it cannot do its work, so sj does not save the Find message and forward it. If C 0 is less than the remaining energy of sj , then check the list of messages. If the list has received the same Find message, then forward it. Otherwise, the edge node sj sends the Answer(sj , Lj , sLink) message to bi , including ID number of sj , the position of sj , the nodes number connected to the edge node sj , and change node’s own information by the information in Links. The key node bi processes the Answer message After the key node bi received the Answer message that returned by sj , then send the Restructure (bi , new key, hops). message to all nodes in the topology network i in addition to itself, the message includes the ID number of the bi , the new key node number, and the hops that are used for traversing. The value of hops is set by bi . Every time the Restructure message passes by, the hops minus 1. If hops = 0, the Restructure message is no longer being passed. If the node in the topology network i has received the Restructure message, then the bi from the Restructure message shall be compared with the order number of self-connected nodes. If they are consistent, the node that has been compared changes the order number of nodes connected to itself by the information of new key (i.e. the node changes the self-connected node). Meanwhile, reply the Run(num) message to bi , including its own number that represents the node, which has changed the order number of self-connected nodes. If they are not consistent, it indicates the node did not connect with the key node. Then, forward the Restructure message. The key node bi processes the Run message When the key node bi has received Run message, the key node bi compares the num with the order number of self-connected nodes. If all nodes connected with the key node bi replied the Run message, the key node bi changes the order number of self-connected nodes by sLink in the Answer message (i.e. the key node changes the self-connected node) and sends the Move message to sj . Meanwhile, the key node bi moves to the position of sj according to Lj in the Answer message. The edge node sj processes the Move message After the edge node sj received the Move message, sj moves to the position of the key node bi according to Li in the Find message that the edge node sj has received. Meanwhile, bi is moving to the position of the node si that needs to be repaired. Step 3 Use the method mentioned in section ‘‘The method for supplying the coverage’’ to avoid the coverage loss. Case 2: The disconnected node belongs to set B. The Double-AUVs can directly repair the connectivity of UASNs.
11 Step 1 The Double-AUVs just send the information to sensor node bi attached to the Double-AUVs and ask the sensor node for returning. Then, clean the sensor node bi or charge the sensor node bi , because the location and path of the sensor have been recorded in the Double-AUVs when the sensor node bi has been dispatched. Step 2 When the sensor node bi has been cleaned or charged, the Double-AUVs send bi back to the original location according to the information that has been recorded in the AUV node Ai . Step 3 When the sensor node bi returns to the original position and communicates with the neighbor nodes, bi sends a confirmation message to the Ai through the wire.
The process of eliminating key nodes Frequent repairs will decrease the lifetime of the entire network, and lead to gaps in information collection, and so on. In order to improve the performance of UASNs, it is required to first eliminate key nodes after the repair is finished. By Definition 5, after the edge of the node is removed, neither the communication branch is increased, nor new node key nodes is formed. After the optimal node is selected, the information about other sensor nodes in the connected branch can be known by Hypothesis 1. Then, select the nearest edge node from the key node in random and reconstruct the neighborhood topology of the key node through the triangular mechanism. Eliminating the key node requires three nodes at least due to the stability of the triangular structure and the limited energy of the underwater acoustic sensor. Then, we introduce the concept of triangle mechanism. At least select the nodes in the neighborhood of the key node that can be connected with the key node for establishing a connection (see e5, 10 (s5 , s10 ) and e7, 10 (s7 , s10 ) in Figure 7(b)) with the key node according to the cosine theorem and the sine theorem. After the key node was eliminated, the neighborhood topology can be reestablished by the method for determining theneighborhood topology mentioned above (see the dotted line e6, 10 (s6 , s10 ) and e9, 10 (s9 , s10 ) in Figure 7(b)). The above process is shown in Figure 7. After the process mentioned above is finished, the nodes s6 , s7 , and s10 are not the key nodes. Although removing the edge node will cause the loss of coverage, the optimization of invulnerability of the network topology performance is greater than the loss of coverage.
Experimental study In this section, we conduct simulation experiments repeatedly to demonstrate the effectiveness and
12
International Journal of Distributed Sensor Networks
Table 1. The experimental parameters. Symbol
Description
Value
rs rc Vt v e Q lstep lmax Interval Interval
Sensing radius of the sensor Communication radius of the sensor Propagation velocity of sound underwater Speed of sensor The initial energy of sensor Disturbance amplitude The largest moving step size of artificial fish The largest moving step size of repair node The timer of optimization period The timer of optimization period
40 m 60 m 1500 m/s 3 m/s 30 J 18 m 15 m 15 m 60 min 60 min
Figure 7. The process of eliminating the key nodes: (a) the local topology and (b) the local topology after the key node was eliminated.
efficiency of the proposed algorithm by the Monte Carlo method. The experimental environment is as follows: there are 50 events required to be monitored, which are unevenly distributed in the 400 m 3 400 m 3 400 m underwater monitoring area. The UASNs composed of 20 sensors in use for monitoring the events and there are five AUVs added to the UASNs. The locations of sensors and target events influenced by the ocean current12 model have changed. Table 1 shows the experimental parameters. The coefficient of proportionality kg , kr , ko , kb , and kc are set as 106 , 102 , 104 , 104 , and 102 , respectively, and unit energy ! ! ! ! ! consumption in each direction Cx , Cx , Cv , Cv , Cz , ! ! ! ! ! and Cz are set as Cx = Cy = 5 3 102 J, Cx = Cy ! ! = 2 3 102 J, Cz = 8 3 102 J, andCz = 1 3 102 J, respectively.
The TO-A algorithm analysis Compared with the PSO algorithm mentioned by Bhavya and Nithya13, we carry on three comparison experiments. Experiment 1. A comparison experiment about coverage optimization. Figure 8 shows the change curve of the degree of coverage with time gradually after the optimization algorithm is carried out. Solid line represents the
Figure 8. The change of coverage in the optimization process.
effect of TO-A algorithm. Dashed line represents the effect of PSO algorithm. It can be seen that the final value of the coverage after PSO algorithm is less than that of the TO-A algorithm in this article, because the TO-A algorithm inspired by fish swarm in this article is capable of global search so that the global extremum can be realized. But both of them can achieve nearly optimal in the coverage and guarantee the coverage over 95%. We also find PSO algorithm can achieve the optimal in the coverage earlier, because PSO algorithm obtains the value of local optimal quickly. So, the curve can be earlier convergent.
Experiment 2. Compare the number of the key node after coverage optimization. The experimental scenario is set as follows: the number of events increased from 50 to 70 and the number of sensors distributed in the UASNs is 20, 25, and 30, respectively. The TO-A algorithm and PSO algorithm are carried out in these three scenarios. We have statistics on experimental times of key nodes exist in the 100th round experiment (see Figure 9). It can be seen that with the increase in the number of events, there is a significant reduction in the
He et al.
Figure 9. The number of experiments that exists the key nodes.
13
Figure 10. The curve of recovering connectivity that is randomly selected from 100 experiments.
number of experiments of the existence of key nodes, because during the execution of the two algorithms, sensors always move toward the place where the density of events is high. The events distributed are more even so that sensors deployed are more even after the events in the sparse area are increased. But the PSO algorithm is easy to fall into local optimum so that the overall result is worse than the TO-A algorithm (inflection point as shown in Figure 9). In addition, as shown in Figure 9, when the sensor node is set to 25, the overall effect after the implementation of the algorithm is optimal. The number of sensor nodes is less difficult to uniformly cover the events when the number of events is constant. The number of sensor nodes is increased too much and events are distributed unevenly, leading to sensor nodes deployed unevenly and the key nodes are easily generated. Experiment 3. A comparison experiment about recovery of the connectivity. First, we verify the optimization of the connectivity by executing the TO-A algorithm proposed in this article. We set the number of events to 50 and the number of sensor nodes to 20. The connectivity of network is repaired by the edge nodes (this method is mentioned in section ‘‘Connectivity optimization for UASNs based on edge nodes’’). The connectivity of UASNs was recovered in 100 repeated experiments. Figure 10 shows the connectivity recovery curves of 4 experiments were randomly selected from the 100 repeated experiments. Figure 11 shows the change curve of the experimental results through performing two kinds of algorithms during 500 s. As said by Bhavya and Nithya,13 eight self-moving nodes (the AUVs) were deployed in the above
Figure 11. The change of connectivity through the algorithm: (a) the TO-A algorithm and (b) the PSO algorithm.
experimental background. When the network is not connected, the self-moving node is used to repair. Figure 11(a) shows clearly, in addition to initializing repair operation that takes nearly 40 s, and others only take 10 s. It is indicated that the PSO algorithm achieves the connectivity of UASNs quickly. But it is easy to generate the disconnected network because in the work by Bhavya and Nithya,13 the key nodes were not eliminated, which leads to weaken the invulnerability of UASNs. Even the deviation of the location of AUVs nodes that is influenced by ocean currents will occur and leads to the disconnected UASNs and the
14
Figure 12. The change of connectivity when disconnected location is restored by the algorithm.
UASNs need to be repaired again (as it has been shown time in Figure 8(a) around 240 s). Although the TO-A algorithm proposed in this article spends a longer time to initialize the repair operation (the average cost is about 55 s in 100 repeated experiments), the TO-A algorithm will enhance the invulnerability of UASNs through eliminating the key nodes and it is not easy to disconnect the network under the influence of ocean currents and all kinds of biology. Figure 11(b) shows the connectivity of UASNs which was repaired by the TO-A algorithm within 500 s that the experiment starts. In the process of the experiment in 500 s, after the PSO algorithm mentioned by Bhavya and Nithya13 is carried out, we can calculate the connectivity rate of UASNs is 372/500=74% and the connectivity rate of UASNs after the TO-A algorithm proposed in this article is carried out is 445/500=89%. We can also find a phenomenon that the longer experiment time is, the advantages of the TO-A algorithm more obvious is, so the TO-A algorithm has better scalability.
The TO-DA algorithm analysis Compare with the PSO algorithm mentioned by Bhavya and Nithya13 and the fish swarm optimization algorithm mentioned by Pandey and Pompili14 (called Fish Swarm optimization algorithm [FSO]). Experiment 1: a comparative experiment about repairing the connectivity. When the failure node occurs in the topology of the network, the number of connected nodes decreases sharply. Figure 12 shows the change of the number of connected nodes with the time after the implementation of three algorithms, which are TO-DA algorithm, the PSO algorithm, and the FSP algorithm, respectively. It can be seen the curve of PSO algorithm
International Journal of Distributed Sensor Networks
Figure 13. The comparison of coverage and survivability.
is relatively stable, because the PSO algorithm is the local optimum; the curve of FSP algorithm is fluctuated, because large number of sensors were selected to move when the FSP algorithm is executed. After performing the above two algorithms, the number of connected sensor nodes is less than the result from the TO-DA algorithm proposed in section ’’ The TO-DA algorithm.’’ We can also see that the start points of three curves in Figure 12 are different, because the method of deployed node is different in three algorithms. During the execution process of TO-DA algorithm, there is a larger changein the number of connected nodes, because the process of repairing node happens always in the implementation of the operation of eliminating the key nodes. In the process of eliminating the key nodes, on one hand, it will speed up the covering of all nodes in UASNs; on the other hand, there will be the inevitable decline in performance of the connectivity of UASNs when the edge node in the mobile, but the time of declining in performance is short. So, we just need a very short period of time to achieve the overall optimization of the repair. Experiment 2: the analysis of coverage and performance of survivability. Figure 13 is the change of the survivability and the coverage after the algorithm is executed. The left ordinate, solid blue line (representing PSO algorithm (1) in Figure 13) and blue circles line (representing TO-DA algorithm (1) in Figure 13) represent a change of the coverage in PSO optimization algorithm and the TO-DA algorithm over time, respectively; the right ordinate, solid green line (representing PSO algorithm (2) in Figure 13) and green circles line (representing TO-DA algorithm (2) in Figure 13) represent a change of the survivability in PSO optimization
He et al.
Figure 14. The comparison of connectivity and energy consumption.
algorithm and the TO-DA algorithm over time, respectively. It can be seen in Figure 13 that though the coverage of the UASNs after the TO-DA algorithm is carried out is less than the effect by implementation of the PSO algorithm, the survivability after the TO-DA algorithm is executed is far beyond the effect after the implementation of the PSO algorithm. Through statistical analysis, although at the expense of coverage of approximately 8.5% in the optimization process, survivability performance is improved by about 50%. Because the process of TO-DA algorithm execution is always accompanied with the operation of eliminating the key nodes, which makes the topology change, the coverage is reduced, but the key nodes are eliminated, and the network lifetime is prolonged. This method is more suitable for long-time underwater missions. Experiment 3: the performance analysis of the network after the end of the process of optimization. Figure 14 shows the connectivity and the overall energy consumption of the UASNs during the process of the optimization. In the optimization phase, the line slope of the energy consumption is larger because there are lots of moving nodes. In the whole UASNs network, the energy consumption of the nodes gradually tends to be stable and the growth is stopped at 110 s. This situation indicates the process of the implementation of FSP algorithm is finished. As we can see from Figure 14, the connectivity curve shows an upward trend with time, the curve reaches the peak near the 60 s. There is a clear advantage compared with the FSP algorithm. It reaches the peak at about 43 s, then the curve about TO-DA
15 algorithm drops to 0.9 in a short period of time, then backs to the peak near 85 s and maintain stably, because the implementation of the TO-DA algorithm is finished near the 60 s, and the connectivity of UASNs has been optimized, but during the following processes of eliminating the key nodes, there is inevitable decline in local connectivity. So, the overall connectivity of the network decreased slightly due to the edge nodes leaving its original location; however, the connectivity of UASNs is restored immediately with the completion of eliminating the key nodes. At the same time, as it is seen from Figure 14, when the time is at 120 s and curve of the energy consumption of the topology of UASNs does not leveled off, there is still a small slope. It indicates the energy consumption continues to increase, and the repair process of connectivity of UASNs is still in progress. The above three experiments about TO-DA algorithm are not difficult to be found in the UASNs. As the time passed, the automatic repairing and restructuring of the sensor nodes makes an increase in the density of the sensor nodes. This enables the connectivity among the nodes turn better and the survivability of local topology of network has been improved, but the coverage occurs redundant phenomenon in a certain extent.
Conclusion In this article, we propose a topology optimization scheme based on reorganization, namely, TO-A algorithm that maintains connectivity and coverage, and a topology optimization mechanism based on DoubleAUVs, namely, TO-DA algorithm that prolongs the life cycle and connectivity of UASNs. To be more specific, the TO-A algorithm’s purpose is to address the problem about topology affected by ocean currents and other environmental factors that leads to the dynamic evolution of network and generates gradual reduction in the performance of UASNs; the TO-A algorithm not only adjusts the location of sensor nodes by simulating the fish swarm behavior so as to optimize the coverage, but also uses the edge nodes to repair the disconnected position in the UASNs so as to eliminate the key nodes and optimize the connectivity. Although the TO-A algorithm presented above, compared with other algorithms can effectively improve the connectivity of the network and optimize the coverage with lower energy consumption, the TO-DA algorithm is a further research for executing the underwater long-term missions. We redefine the Double-AUVs and propose the TO-DA algorithm. The TO-DA algorithm not only repairs the failed node by Double-AUVs, but also takes into account the frequent topology repair, which leads to a reduction in the lifetime of the network. So, after
16 the repair is completed, the algorithm begins to carry out the operation that eliminates the key nodes, which will improve regeneration and survivability of UASNs in uncertain ocean environment. The simulation results, respectively, demonstrate the performance of UASNs after the algorithms are executed; all the algorithms can restore the connectivity of UASNs, but the TO-A algorithm is more efficient in coverage, and the TO-DA algorithm is more efficient in survivability. The TO-A algorithm can guarantee the network coverage is maintained at about 97%, connectivity rate above 89% with lower energy consumption and the TO-DA algorithm can improve survivability of the network effectively above 50% with lose of 8.5% of the network coverage. Research results have certain contributions to the research of topology of UASNs and have extensive application value. In the near future, on one hand, we plan to make reasonable assumptions about the AUVs own energy, control threshold value of AUVs’ energy, and further ensure the efficiency of the network, because the energy of AUVs is infinite in the hypothesis of this article; on the other hand, the uncertainty factors that affect the performance of UASNs also include the dynamic nature of the boundary of different temperature water layers, the mesoscale vortex, and other factors such as human factor; so, we will consider these factors in the follow-up study. Declaration of conflicting interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Natural Science Foundation of China under Grant Nos 61301159 and 61303267, Natural Science Foundation of Jiangsu Province under Grant Nos BK20150721 and BK20161469, China Postdoctoral Science Foundation under Grant Nos 2015M582786 and 2016T91017, Engineering Research Center of Jiangsu Province under Grant No. BM2014391, and Primary Research & Development Plan of Jiangsu Province under Grant No. BE2015728.
References 1. Erol-Kantarci M, Mouftah HT and Oktug S. A survey of architectures and localization techniques for underwater acoustic sensor networks. IEEE Commun Surv Tutor 2011; 13(3): 487–502. 2. Fu H and Han S. Optimal sensor node distribution based on the new quantum genetic algorithm. J Sens Actuator 2008; 21(7): 1259–1263.
International Journal of Distributed Sensor Networks 3. Aziz AA, Sekercioglu YA, Fitzpatrick P, et al. A survey on distributed topology control techniques for extending the lifetime of battery powered wireless sensor networks. IEEE Commun Surv Tutor 2013; 15(1): 121–144. 4. Wang X, Wang S and Ma J. Parallel particle swarm optimization based mobile sensor node deployment in wireless sensor networks. J Comput 2007; 30(4): 563–568. 5. Mi Z and Zhou J. A multi-object optimization routing algorithm with constraint for wireless sensor networks. J Appl Sci-Electron Inf Eng 2008; 26(3): 239–243. 6. Jafri MR, Ahmed S, Javaid N, et al. AMCTD: adaptive mobility of courier nodes in threshold-optimized DBR protocol for underwater wireless sensor networks. In: Proceedings of the 2013 eighth international conference on broadband, wireless computing, communication and applications (BWCCA), Compiegne, 28–30 October 2013, pp.93–99. IEEE. 7. Xia N, Wang C, Zheng R, et al. Fish swarm inspired underwater sensor deployment. Acta Autom Sin 2012; 38(2): 295–302. 8. Manjula RB and Sunilkumar SM. Coverage optimization based sensor deployment by using PSO for antisubmarine detection in UWASNs. Ocean Electron (SYMPOL) 2013; 2013: 15–22. 9. Nguyen ST, Cayirci E, Yang L, et al. A shadow zone aware routing protocol for acoustic underwater sensor networks. IEEE Commun Lett 2009; 13: 366–368. 10. Heidemann J, Stojanovic M and Zorzi M. Underwater sensor networks: applications, advances and challenges. Philos T R Soc A 2012; 370(1958): 158–175. 11. Luo H, Guo Z, Dong W, et al. LDB: localization with directional beacons for sparse 3D underwater acoustic sensor network. J Netw 2010; 5(1): 28–38. 12. Caruso A, Paparella F, Vieira LFM, et al. The meandering current mobility model and its impact on underwater mobile sensor networks. In: Proceedings of INFOCOM 2008, Phoenix, AZ, 13–18 April 2008, pp.221–229. Piscataway, NJ: IEEE. 13. Bhavya R and Nithya N. Efficient hierarchical route allocation for underwater sensor network. Int J Emerg Sci Eng (IJESE) 2014; 2(4): 2319. 14. Pandey P and Pompili D. Region of feasibility of interference alignment in underwater sensor networks. IEEE J Oceanic Eng 2014; 39(1): 189–202. 15. Masiero R, Azad S, Favaro F, et al. DESERT underwater: an NS-Miracle-based framework to Design, Simulate, Emulate and Realize Test-beds for underwater network protocols. In: Proceedings of OCEANS, Yeosu, South Korea, 21–24 May 2012, pp.1–10. New York: IEEE. 16. Najeeb F, Khan ZA, Qasim U, et al. A new advanced energy efficient routing protocol for UWSNs. J Basic Appl Sci Res 2014; 4(2): 43–53. 17. Peng Z, Wang D, Shi Y, et al. Containment control of networked autonomous underwater vehicles with model uncertainty and ocean disturbances guided by multiple leaders. Inform Sciences 2015; 316: 163–179. 18. Xia N, Wang S, Zheng R, et al. Study on localizability judgment in underwater sensor networks based on skeleton extraction and rigidity theory. Chin J Comput 2014; 37(26): 1–14.
He et al. 19. Shen J, Tan H, Wang J, et al. A novel routing protocol providing good transmission reliability in underwater sensor networks. J Internet Technol 2015; 16(1): 171–178. 20. Coutinho RWL, Boukerche A, Vieira LFM, et al. A novel void node recovery paradigm for long-term underwater sensor networks. Ad Hoc Netw 2015; 34: 144–156. 21. Alam SN and Haas Z. Coverage and connectivity in three-dimensional networks. In: Proceedings of the 12th annual ACM international conference on mobile computing and networking (MobiCom’06), Los Angeles, CA, 24–29 September 2006. 22. Caiti A, Grythe K, Hovem JM, et al. Linking acoustic communications and network performance: integration and experimentation of an underwater acoustic network. IEEE J Oceanic Eng 2013; 38(4): 758–771. 23. Liu Y and Zhou G. Key technologies and applications of internet of things. In: Proceedings of the fifth international conference on intelligent computation technology and automation (ICICTA), Zhangjiajie, China, 12–14 January 2012, pp.197–200. New York: IEEE. 24. Yoon S, Azad AK, Oh H, et al. AURP: an AUV-aided underwater routing protocol for underwater acoustic sensor networks. Sensors 2012, 12: 1827–1845. 25. He M, Chen Q, Chen X, et al. Fish swarm inspired Ad hoc networks node random mobility optimization model in 3D environment. Chin J Sci Instrum 2014; 35(12): 2826–2834. 26. Sanchez A, Blanc S, Yuste P, et al. Advanced acoustic wake-up system for underwater sensor networks. Commun Inf Sci Manage Eng 2012; 2(2): 1–10. 27. Mitsche D, Resta G and Santi P. The random waypoint mobility model with uniform node spatial distribution. Wirel Netw 2014; 20(5): 1053–1066.
17 28. Xie S and Wang Y. Construction of tree network with limited delivery latency in homogeneous wireless sensor networks. Wirel Pers Commun 2014; 78(1): 231–246. 29. Li J, Li X, Yang B, et al. Segmentation-based image copy-move forgery detection scheme. IEEE T Inf Foren Sec 2015; 10(3): 507–518. 30. Gu B, Sheng VS, Tay KY, et al. Incremental support vector learning for ordinal regression. IEEE Trans Neural Netw Learn Syst 2015; 26: 1403–1416. 31. Xia Z, Wang X, Sun X, et al. A secure and dynamic multi-keyword ranked search scheme over encrypted cloud data. IEEE T Parall Distr 2015; 27: 340–352. 32. Pan Z, Zhang Y and Kwong S. Efficient motion and disparity estimation optimization for low complexity multiview video coding. IEEE T Broadcast 2015; 61: 166–176. 33. Xia Z, Wang X, Sun X, et al. Steganalysis of LSB matching using differences between nonadjacent pixels. Multimed Tools Appl 2014; 75: 1947–1962. 34. Xia Z, Wang X, Sun X, et al. Steganalysis of least significant bit matching using multi-order differences. Secur Commun Netw 2014; 7(8): 1283–1291. 35. Fu Z, Sun X, Liu Q, et al. Achieving efficient cloud search services: multi-keyword ranked search over encrypted cloud data supporting parallel computing. IEICE T Commun 2015; E98-B(1): 190–200. 36. Chen B, Shu H, Coatrieux G, et al. Color image analysis by quaternion-type moments. J Math Imaging Vis 2015; 51(1): 124–144. 37. Wen X, Shao L, Xue Y, et al. A rapid learning algorithm for vehicle classification. Inf Sci 2015; 295(1): 395–406. 38. Zheng Y, Jeon B, Xu D, et al. Image segmentation by generalized hierarchical fuzzy C-means algorithm. J Intell Fuzzy Syst 2015; 28(2): 961–973.