Journal of Communications and Information Networks, 2018, 3(1): 67-83 doi: 10.1007/s41650-018-0008-3

Research papers

Adaptive Routing Protocol for Lifetime Maximization in Multi-Constraint Wireless Sensor Networks

Fouad El Hajji,, Cherkaoui Leghris,, Khadija Douzi,

Corresponding authors: Fouad El Hajji,fouad.elhajjietu@univh2m.ma

作者简介 About authors

Fouad El Hajji was born in 1988. He received his B. S. and M. S. degrees in electronic and automatic engineering from Hassan II University, Casablanca, Morocco, in 2009 and 2011, respectively. He is now a Ph. D. student in network and communications engineering at Faculty of sciences and technologies, Hassan II University of Casablanca. His main research interests include communications and wireless sensor network applications, network security, offensive and defensive techniques. E-mail:fouad.elhajjietu@univh2m.ma.

Cherkaoui Leghris is a professor at the Hassan II University in Casablanca, Morocco. He provides training for engineers on networking technologies, which is now focused on IPv6, wireless networks, IoT, broadband, network security. He also leads several scientific research efforts in ICT technologies with his project 4-Any. He has many publications in many conferences and scientific journals. E-mail:cleghris@yahoo.fr.

Khadija Douzi is a professor researcher at Faculty of Sciences and Technologies, Hassan II University of Casablanca, Morocco. She is a member of the LIM laboratory in addition to being the director of the A2S research team. Her research interests include E-orientation system, dynamic learning content model for adaptive learning environment, multiple mean linear regression model and WSNs. E-mail:kdouzi@yahoo.fr.

Abstract

Routing in wireless sensor networks plays a crucial role in deploying and managing an efficient and adaptive network. Ensuring efficient routing entails an ever-increasing necessity for optimized energy consumption and reliable resource management of both the sensor nodes and the overall sensor network. An efficient routing solution is characterized by its ability to increase network lifetime, enhance efficiency, and ensure the appropriate quality of service. Therefore, the routing protocols need to be designed with an ultimate objective by considering and efficiently managing many characteristics and requirements, such as fault tolerance, scalability, production costs, and others. Unfortunately, many of the existing solutions lead to higher power consumption and communication control overheads, which not only increase network congestion but also decrease network lifetime. In addition, most of these protocols consider a limited number of criteria, in contrast to the complexity and the evolution of WSNs. This paper presents a new adaptive and dynamic multi-criteria routing protocol. Our protocol operates in multi-constraint environments, where most of the current solutions fail to monitor successive and continuous changes in network state and user preferences. This approach provides a routing mechanism, which creates a routing tree based on the evaluation of many criteria. These criteria can cover the topological metrics of neighboring nodes (the role of the nodes in intracommunications, connections between different parts of the network, etc. ), the estimated power consumption to reach each direct neighbor, the path length (number of hops to the sink), the remaining energy of individual sensor nodes, and others. These criteria are controlled and supervised dynamically through a detection scheme. In addition, a dynamic selection mechanism, based on multi-attribute decision-making methods, is implemented to build and update the routing tree. In response to changes in the network state, user preferences, and application-defined goals, the election mechanism provides the best routing neighbor between each node and the sink.

PDF (4879KB) Metadata Metrics Related articles Export EndNote| Ris| Bibtex  Favorite

Cite this article

Fouad El Hajji. Adaptive Routing Protocol for Lifetime Maximization in Multi-Constraint Wireless Sensor Networks. [J], 2018, 3(1): 67-83 doi:10.1007/s41650-018-0008-3

Ⅰ. INTRODUCTION

In the world of computer networks and telecommunicationsystems, wireless sensors networks(WSNs)present an active and dynamic research area that grows exponentially. A WSN is composed of hundreds or thousands of tiny embedded devices called motes that have inbuilt features for sensing, processing, and communicating information over wireless channels, deployed either randomly or uniformly, inside a phenomenon or very close to it. Once deployed, the sensor nodes must be able to auto-organize themselves into a WSN. The limitations of these networks include constrained power, susceptibility to node failures, node mobility, a large-scale deployment, a dynamic network topology, etc. The application requirements, like the quality of service (QoS), expose the network administrator to multiple challenges and increase the need for the efficient management of WSN features[1].

In WSNs, the nodes consume energy in different ways. However, the primary energy cost is dissipated in data transmission. In addition, a WSN has many characteristics and limitations that must be evaluated and taken into consideration[2]. Firstly, due to the density of sensor nodes in WSNs and a large number of nodes transmitting simultaneously, medium access control for these networks and scalability issues is a challenge. Secondly, due to the small size of sensor nodes, they have limited computing, bandwidth, and communicating capabilities. Finally, sensor nodes broadcast their data, therefore there may be a high probability of data redundancy. These limitations pose many challenges to the design and maintenance of sensor networks at all levels of the networking protocol stack[1,2].

In WSNs, due to an aleatory deployment, certain sensor nodes are connected to a greater number of motes as part of these network paths. Managing WSN challenges is impracticable without considering node weightiness. To address these issues, certain measures, such as centrality and connectivity, have been proposed[3]. Centrality is related to the position of motes(nodes)in a network and leads to the determination of their relative impact on it. Therefore, we applied centrality measures to define the importance and the role of a node within a network. This also assists in calculating the importance and relationship of each node with its neighboring nodes. In addition, it may help in resolving various inherent problems and developing efficient algorithms for WSNs[4]. One of these problems is routing data inside the network.

In network processing and data transmission, communication between motes is structured as a routing tree with a base station as their root. Always using the shortest path routing may not be optimal for the lifetime of a network, sustainable connectivity, network capacity, and dealing with failures. Conversely, when the robustness of a topology and dealing with failures gains popularity and applications increase in complexity, it becomes imperative that the creation of a routing tree be based on the objective of ensuring robustness in addition to existing standard criteria(such as the network lifetime). This encourages the development of adaptive routing protocols that consider multiple criteria, as opposed to the single-criteria approaches in use today[5]. However, there has been less research in the field of multiple constraint-based routing protocols, which consider many performance goals. Such criteria should consider latency requirements, the robustness of the topology, as well as node characteristics such as the energy remaining at the sensor nodes and the power consumption model of the nodes[6,7].

This paper proposes how centrality measures and multicriteria algorithms can be combined to increase the performance of WSNs through the introduction of a dynamic multicriteria routing approach. Our objective is to reduce the use of shortest paths between each node and the sink, so that nodes with a high centrality will not be overloaded, leading to less power consumption. This strategy can balance the energy consumption throughout the entire network by limiting the transmitted packet rate for each node, increasing the lifetime by decreasing the risk of failure for crucial nodes (higher centrality), avoiding congestion and reducing both the packet retransmission rate and redundancy. In addition, a core concept based on multi-attribute decision-making(MADM)methods is implemented to select the best alternatives(paths between each node and sink, as well as routing nodes)to create a routing tree according to user preferences, application requirements, and network state changes.

The rest of this paper is organized as follows. Firstly, we present a brief overview of previous studies related to the WSN routing problem. In section Ⅲ, we present the dynamic multi-criteria routing approach and the use of centrality measures for power attribution and detection of critical nodes. Section IV describes the different mechanisms and algorithms used in our model. Then, we discuss the experimental setup and numerical results in section V. The final section presents the main conclusion and future directions of this work.

Ⅱ. RELATED WORKS

Routing in WSNs is very complicated and challenging, which differs from standard routing. This is due to several characteristics that differentiate WSNs from conventional communication systems like wireless ad-hoc networks. First, it is impracticable to create a global addressing strategy for deploying all the sensor nodes. Second, sensor nodes are limited in terms of transmission power, energy power(batteries), treatment capacity and data storage, which require an efficient resource management[2]. Owing to such constraints, many approaches and protocols have been presented and applied to solve the routing data conundrum in WSNs.

Data-centric protocols are based on query specifications and depend on the designation of desired and specific data, which help exclude multiple transmissions of similar data and suppress many redundant packets[2]. SPIN is a basic datacentric protocol developed based on channel capacity control and data negotiation between nodes in order to remove redundant data and reduce energy consumption. Later, directed diffusion was developed and it revolutionized the datacentric routing concept. Subsequently, many other protocols have been proposed either based on directed diffusion or following a similar concept like rumor routing, COUGAR, and Information-Directed Routing[5].

Comparable to standard communication networks, scalability remains one of the principal design specification of WSNs. Hierarchical protocols organize networks in sets of nodes called clusters. An elected node, named a cluster head (CH), manages each cluster, realizes data aggregation, and reduces transmissions in order to save energy[8]. LEACH is one of the first hierarchical routing protocols. The concept presented in LEACH has been a key inspiration for other hierarchical routing protocols like TEEN, PEGASIS and APTEEN, although many protocols have been developed, like energy-aware routing for cluster-based sensor networks and self-organizing protocols. In Ref. [9], Hussain and Matin proposed a hierarchical cluster-based routing (HCR) protocol where nodes self-organize into clusters and each cluster is managed by a set of associates called a headset. They improved the HCR protocol by using a genetic algorithm to determine the number of clusters, the cluster heads, the cluster members, and the transmission schedules. Inspired by this work, a large family of clustering protocols has been developed.

Location-based protocols are realized by using position information to transmit the data between nodes and sink. Generally, location information is used to determine the distance separating two distinct nodes, so that the energy consumption can be estimated, thereby reducing the overall dissipated energy and increasing the network lifetime[2]. The most famous of these protocols is the geographic adaptive fidelity (GAF) protocol, which is a location-based protocol proposed for the first time for mobile adhoc networks(MANETs)[5]. The table 1 presents a summary of the most popular routing protocols used in the WSN field.

As a result, the main challenges for the design of routing protocols for WSNs are still the energy efficiency and reliable data routing due to the limited resources of nodes, such as energy and radio transmitter capacity. The ultimate aim of the routing protocol design is to preserve node operation for as long as possible, thus expanding the network lifetime. However, with the increase in application and user requirements, many other criteria must be taken into consideration during the design of a protocol. This has led to a change in the protocol design strategy from being based on a single criterion and step to a new approach considering a set of criteria and many routing objectives.

In this context, many new approaches have been developed. Sert et al. [10] presented a new approach to solve the hotspots problem in clustering-based routing named a Multi-Objective Fuzzy Clustering Algorithm. This algorithm employs the remaining energy, distance to the sink and density of the nodes, nearly all parameters considered thus far, together with fuzzy logic in estimating the competition radius. This radius can be changed in each round. Other solutions proposing the adjustment of the election cluster-head mechanism have increased the efficiency. Khan et al. [11]proposed a Fuzzy-TOPSIS technique, based on multi-criteria decision-making, to select the CH efficiently and effectively to maximize the WSN lifetime. They considered several criteria including the residual energy, the node energy consumption rate, the number of neighbor nodes, the average distance between neighboring nodes and the distance from the sink. A threshold based intra-cluster and inter-cluster multi-hop communication mechanism is used to decrease energy consumption.

In summary, we can note some facts.

• Existing protocols differ according to topological characteristics, the mechanisms used and the adaptation to technology evolution. Furthermore, many proposed protocols use complicated algorithms which present serious implementation difficulties.

• The principal aims of these solutions are to extend the network lifetime, reduce the power consumption, and guaran tee a minimum quality of service and data transmission. Despite their importance and significance, these objectives are still ineffective due to the actual complexity of WSNs. Therefore, more objectives and criteria should be integrated into future work.

• Most of the protocols do not take into account user preferences and changes that can occur during the exploitation phase due to user and environment adjustments.

Table 1   Some routing protocols for WSNs

categoryrepresentative protocols
location-based protocolsMECN,SMECN,GAF,GEAR,Span,TBF,BVGF,GeRaF
SPIN,directed diffusion,rumor routing,
data-centric protocolsCOUGAR,ACQUIRE,EAD,informationdirected routing,gradient-based routing,energy-aware routing,information-directed routing,quorum-based information dissemination,home agent based information dissemination
hierarchical protocolsLEACH,PEGASIS,HEED,TEEN,APTEEN
QoS-based protocolsSAR,SPEED,energy-aware routing

New window| CSV


In the next section, we present our new multi-criteria routing approach, which responds to these critical challenges based on the analysis of the topological and energetic characteristics of the network, the integration of user preferences and application requirements, as well as adaptation to network changes.

Ⅲ. THE PROPOSED APPROACH

Generally, classical routing schemes employ an approach intended to solve WSN challenges, based on a limited number of criteria without considering user preferences, the evolution of application requirements, or the dynamic features of new WSNs. In addition, there has been less prior work in the area of multi constraint-based routing protocols, which consider other criteria and performance goals like congestion and robustness[2]. However, most of these studies present only improvements on existing protocols, without addressing contemporary problems like the integration of user preferences, loss of closeness nodes to the sink, adjustment and reorganizing of the routing scheme in response to intra-network events and the progress of application requirements, throughout the exploitation phase.

In other words, due to the technological limits of WSNs, protocols must confront many challenges to design and maintain sensor networks at all layers of a networking protocol stack. Therefore, we apply centrality measures to define the importance and role of a node within the network. It also helps in calculating its importance and relationship with its neigh boring nodes[12,13,14]. To exploit the information offered by centrality measures, and benefit from the strength and efficiency of MADM methods, we present a new routing approach based on the use of multi-criteria algorithms to satisfy a multiplicity of requirements and constraints in WSNs.

Figure 1

Figure 1   Overview of the multi-criteria routing approach


To address some of these challenges, we present our new approach for multi-criteria routing based on data propagation from nodes to a central sink, which becomes the root of the routing tree. The new approach includes three stages.

• The first one starts after the deployment of nodes in the field, where a neighbor detection mechanism is initiated to discover the overall network and construct an initial routing tree.

• In the second stage, our multi-criteria routing approach provides a mechanism for route construction, which considers the best choice of the relaying node to the sink, based on the evaluation of the neighboring node metrics according to an automatically varying mix of criteria.

• The last stage covers the exploitation phase, where these criteria are supervised and value metrics are varied dynamically in response to changes in the network state and application-defined goals.

The updated criteria values in the database are used to perform a selection mechanism and upgrade the routing tree if necessary. Fig.1 presents an overall view of our approach. It accepts the user and application-specified goals as an input to the system, and it responds to changes in the network state by triggering a rebuild of the routing tree. Our approach considers various criterion Ci. In this paper, we specifically focus on only six criteria.

• The energy remaining at the sensor node.

• The establishment of a connection between a neighbor and the sink(direct or indirect through relay nodes).

• The betweenness centrality of each neighbor(given information about the node’s importance in the network, latency, congestion and estimated lifetime).

• The bridging centrality of each neighbor (indicated by the role of the node in linking between parts of the network, that have an impact on network connectivity).

• The transmission power consumed to reach every node neighbor.

• The distance between each neighbor and the sink (referred to as the number of hops to reach the sink from each neighbor).

The essential aim of our approach is to increase the network lifetime, reduce the risk of node failures as well as the ability of connections to breakdown on the network, while at the same time, to ensure a rapid and efficient exchange of data on the network. These objectives can be achieved through minimizing the use of shortest paths in the routing between the sink and others nodes. In doing so, we can reduce the overload on nodes with a low battery power level.

A. Centrality Measures and the Topological Analysis of the Network

There are a wide variety of applications in sensor networks where the necessity to identify crucial components is of great importance. The node centrality, or the identification of critical nodes, has been a key issue in social network analysis. Due to similarities between social networks and wireless networks from the standpoint of graph theory, centrality measures are involved in several areas, and can help to solve many problems depending upon the requirements of applications and the architectures of WSNs[10,15].

The centrality of nodes, or the determination of nodes which are more important than others, plays a crucial role in optimizing traffic management and consolidating topology robustness. Selecting a centrality measure involves considering the application requirements that the network must satisfy, in addition to the required specific measurements[4]. In our approach, we integrate both betweenness and bridging centralities. The first can be used to determine the estimated overload of each node[3], while the bridging centrality serves to quantify the importance of each node in the connection between dense areas[16].

•Betweenness Centrality: A node with high betweenness centrality is more likely located on the shortest paths between multiple node pairs in the network, and therefore more information needs to be passed through it[4]. Normalized betweenness centrality is defined as

Cbet(vi)=j=1Nk=1N(gjk(vi)gjk),(1)

where g jk(vi)represents the number of shortest paths from v j to vk that pass through vi, and g jk is the number of shortest paths from v j to vk.

•Bridging Centrality: Bridging centrality identifies bridging nodes, which are located between highly connected sets (regions)[16]. The bridging centrality of a node is calculated based on the betweenness centrality Cbet(calculated based on Eq. (1))and the bridging coefficient CB, which measure the global and local characteristics of a node, respectively. Precisely, the bridging centrality Cbri(v)for node v, is defined by

Cbri(y)=Cbet(v)CB(v).(2)

The bridging coefficient of a node identifies the extent of how well the node is located between high degree nodes. The bridging coefficient of a node v is defined as

BC(y)=d(v)1iN(v)(1d(i)),(3)

where d(v)is the degree(degree centrality)of node v and N(v) is the set of its neighbors.

B. Use of MADM Methods to Weigh the Importance Criteria nd Rank Neighbors

According to the nature of routing problems, MADM algorithms represent a promising solution to the selection of the most suitable solution (route, relay node) in terms of many challenges, such as the quality of service and network lifetime[17]. Generally, an MADM method is based on four main parts: alternatives(possible solution), attributes(criteria), the relative importance of attributes(weight), and measures of alternatives(alternative rate)with respect to each attribute.

The weighting assignment plays a crucial role in this process, as it provides a way to quantify user preferences as the final results of the MADM method which widely depends on these weights. Weighting methods can be classified into two categories: “objective weighting methods”and“subjective weighting methods”. In subjective weighting methods, the assignment of the weight to the criteria depends on user preferences. One of most popular weighting concept is the pairwise comparison, which can be efficient because it forces the network administrator to dedicate conscientious attention to each criterion of the decision problem. The concept necessitates only the comparison between two criteria at a time, and it has been theoretically and empirically tested for a variety of decision situations. AHP is one of the first and most popular MADM methods based on pairwise comparison, so we can use it to assign criteria weights.

In our work, we propose an intelligent policy based on an analytical hierarchy process(AHP)and the technique for order preference by similarity to an ideal solution (TOPSIS) method. The AHP method is applied to find the weights of each criterion, taking into consideration the dependency between each criteria pair, and the TOPSIS method is used to rank the alternatives(neighbor nodes). This approach guarantees the acceptability and consistency of our weight vector, given that many criteria are considered that are characterized by their heterogeneous nature, physical scale, and unit of measure.

•AHP: AHP is one of the first MADM methods developed by Saaty[18]. The AHP approach has been widely used in the network selection process to assign weights to different criteria.

The AHP approach is based on three steps.

1. Construction of pairwise comparisons: AHP builds pairwise matrix comparison such as

A=(x11x12x1nx21x22x2nxn1xn2xnn),(4)

where xii=1, x ji=1/xi j. Elements xi j are obtained from Tab.2, defined by Saaty[18], it contains a 1-9 preference scale. Each element xi j presents the preference between two different criteria.

Table 2   Saaty’s scale for pairwise comparison

Saaty’s scalethe relative importance of the two sub-elements
1equally important
3moderately important with one over another
5strongly important
7very strongly important
9extremely important
2,4,6,8intermediate values

New window| CSV


2. Construct the normalized decision matrix: Anorm is the normalized matrix of A(3), where Anorm(ai j) is given by A(xij) as follows:

aij=xiji=1nxij,(5)

Anorm=(a11a12a1na21a22a2nan1an2ann).(6)

3. Calculating the weights of criteria: The weights of the decision factor i can be calculated by

Wi=j=1naijnandj=1nWi=1(7)

with n representing the criteria number.

• TOPSIS: Technique for order preference by similarity to an ideal solution(TOPSIS), known as a classical MADM method, it was developed in 1981[19]. In the TOPSIS method, the optimal alternative selected(in our case, the neighbor routing node)should have the shortest distance from the positive ideal solution and the farthest distance from the negative ideal solution.

The procedure can be categorized in six steps.

1. Construction of the decision matrix: The decision matrix is expressed as

D=(d11d12d1md21d22d2mdn1dn2dnm),(8)

where di j is the rating of the alternative Ai with respect to the criterion Cj.

2. Construction of the normalized decision matrix: In our work, we employ many criteria but not all of these criteria are at the same physical scale and unit of measure. Therefore, we proceed to normalize the decision matrix where each element ri j is obtained via a Euclidean normalization:

rij=diji=1ndij2,i=1,,n,j=1,,m.(9)

3. Construct the weighted normalized decision matrix: The weighted normalized decision matrix vi j is computed as

vij=Wirijwherej=1mWj=1.(10)

4. Determination of the ideal solution Aand the anti-ideal solution A:

A=[V1,,Vm]andA=[V1,,Vm].(11)

For desirable criteria:

Vj=max{vij,i=1,,n},(12)

Vj=min{vij,i=1,,n},(13)

For undesirable criteria:

Vj=max{vij,i=1,,n},(14)

Vj=max{vij,i=1,,n},(15)

5. Calculation of the similarity distance:

Si=j=1m(vijvj)2,i=1,,n,(16)

and

Sj=j=1m(vijvj)2,i=1,,n.(17)

6. Ranking:

Ci=SiSi+Si,i=1,,n.(18)

A set of alternatives can be ranked according to the decreasing order of Ci.

Ⅳ. MECHANISMS USED

To improve the efficiency and feasibility of our approach in the practical field, we implement this approach involving a protocol that we have named Multi-criteria based centralities measures routing protocol(MCRP). In this protocol, the centrality measures are used as criteria to quantify the congestion probability of each node(betweenness centrality)as well as the impact of node failure(bridging centrality)in terms of network robustness and data exchange continuity. The AHP method is used to assign the weight of the criteria on the basis of user preferences and application objectives. Ranking alternatives and the choice of appropriate relay nodes are realized by applying the TOPSIS method.

A. Construction of a Routing Tree

A routing tree can be built in several ways. One relatively simple way is to try to create the tree in such a way that the distance between any two nodes is minimized. This can be achieved by having the first node reached as the parent. This protocol, used in many routing protocols like TAG and COUGAR, is called First-Heard-From (FHF)[20]. We have based our work on the Group-Aware Network Configuration (GaNC)standard[21]. GaNC operates with the same principle as the FHF mechanism. However, in the GaNC algorithm, a child node can switch to a better parent during the routing tree build without waiting for the end of a process. This switch is based on a set of fixed order tiebreaker conditions introducing the global and local network characteristics, in addition to the semantics of aggregation. Our network configuration mechanism considers the properties of individual sensors when dynamically building or rebuilding a routing tree. In order to select a node’s parent, we consider many criteria, such as the distance between nodes and the sink(number of hops)and the energy remaining at the nodes(in Joules).

Neighbor Detection: To establish the connection between nodes and the sink, we used a procedure based on three stages as follows.

• The sink diffuses a broadcast packet towards all motes in order to initialize the connection phase and detect each node’s neighbors.

• The neighbors of each node are detected through a mechanism based on the emission of an NREQ packet (NEIGHBORS REQUEST). The first time, the emission is realized with minimum power. The level of emission power is increased by a judicious step, until the establishment of a connection with the nodes under the cover of the emission range.

• These nodes reply with an NREP packet(NEIGHBORS REPLY), their addresses are stored in a table (memory), named a neighbors table.

This mechanism is periodically repeated in order to guarantee a connection with the network and avoids a total disconnection. Fig.2 presents the steps of this mechanism.

Figure 2

Figure 2   Neighbor detection mechanism: (a)emission of NREQ with minimum power; (b)variation of emission power and remission of NREQ,neighbor’s reply with an NREP


Initial Sink Connection: The initial connection between nodes and the construction of the routing tree starts with a tree build request initiated by the emission of two types of packets. The first is a notification ADV packet(Advertisement)sent by the sink towards its neighbors, in order to inform them of its role, such as the head of the network. This message contains an identifier for the sender, and a value representing the current level in the tree being constructed, L(sender). Given that the base station is the root node of the tree, L(root)=0.Each node in the network transmit another RREQ packet(ROAD REQUEST) to its neighbors, in order to identify those that have a path to the sink. Nodes having a path to the sink reply with an RREP packet(ROAD REPLY). The addresses of its nodes are stored in a table named, routing table. Fig.3 provides an overview of this stage.

B. A Core Concept Based on MADM

Our protocol uses a local per-node list of neighboring nodes. Local to each node, a list of neighboring node statuses are maintained. This list holds a status of the evaluation criteria for each direct neighbor, used for selecting a parent node during tree construction. The construction of the routing tree is initiated by the root node, which conveys the initial weightings of the criteria, and gathers a global view of the state of the sensor network through piggy-backed node status information. By maintaining the ordering information for each of the neighbors, it becomes possible for the node to efficiently select an individual node’s parent, and the selection is dependent on the criteria weightings and the values of these criteria for the neighboring nodes, without being forced to evaluate all nodes.

Figure 3

Figure 3   Establishment of connection between nodes and sink: (a)the Sink sends an ADV packet to its neighbors and each node sends an RREQ to its own neighbors; (b)the sink replies to its neighbors with RREP; (c)the distant nodes send an RREQ to its neighbors; (d)the neighbors connected to the sink reply with a PREP and the node becomes connected to the sink


In our work, we use two MADM methods. AHP is used to assign the weights to different criteria like battery level and TOPSIS to rank the alternatives(neighbors)and determine the optimal solution (parent). Applying MCRP passes through many steps. Initially, we define a set of goals that need to be satisfied. This is drawn from a pre-determined set of goals that the application might want to fulfill. For example, a possible goal may be based on the number of nodes alive such as a network lifetime of 50%. Our terminology and approach for the problem are as follows.

First, we fix the pairwise comparisons matrix, used in the AHP method, that represents the comparative values of criterion i with respect to criterion j according to a set of goals, user preferences, and application requirements. Using this matrix, we calculate the weight vector which is used to weigh the criteria in the ranking stage(TOPSIS method).

Second, we construct the data matrix that combines the values of each criterion, for each neighbor. We normalize the data matrix and through the use of vector weight, we attribute a weight to each criterion. The corresponding weight(for each criterion) is a representation of the suitability of the recommendations with regards to the desired goal.

Finally, by using the weighted normalized data matrix and applying the TOPSIS method, we select the parent according to our preferences, through the calculation of the distance between each neighbor and the anti-ideal solution. Using these distances, we can rank alternatives according to a decreasing order. Fig.4 presents a detailed flowchart of the process selection.

The relative merit of each criterion is based on how it is seen as contributing to the desired system goal[22]. For example, if the power consumption of a node increases, while its remaining energy is decreased, this node likely becomes a less desirable parent node for its neighbors, as it may now be preferable for the overall lifetime of the system for it to conserve its energy. We therefore define the distribution of weights depending on the current node status and desired goals. In other words, weights are in effect defining the combination of criteria that optimally achieve the objective. This is in opposition to static mechanisms such as GaNC, in which the order and the significance of criteria(i. e. , the tiebreaker conditions)are fixed.

C. Weights Criteria Policy

Initially, the weights are broadcasted among all nodes. This initial allocation of weights is specified in the build request packet that is transmitted from the sink to all nodes. Our multi-criteria routing algorithm decides on the parent for each node with a weighted vector of the criteria. Depending on the observed outcome(e. g. , an observed trend towards failing the goals), user preferences, application requirements and status of nodes(dead or alive), the base station may choose to update weights among criteria globally and retransmit criteria values depending on the global network scheme and topology(betweenness and bridging centralities). We now go on to describe the general mechanism of weight and centralities updates.

We periodically check whether the goals are satisfied. If a particular goal is not satisfied, the weights are predetermined, redistributed proportionately and the network is reconfigured. Otherwise, with every reconfiguration, the weights are sent out in the build request message. We have assumed here that the sink has a global view of alive and dead nodes. In actuality, the root node(sink)follows, controls and supervises the evolution of node centralities through the routing tree based on information collected from all nodes. If centralities of one or many nodes increase or decrease rapidly and abruptly, the sink recalculates and retransmits the centralities merits to all motes, guaranteeing that the results of the selection process will not be affected.

In this manner, the sink can periodically or opportunistically acquire such knowledge for all nodes. This global information is necessary for our protocol to improve optimally. Fig.5 presents an example of a routing tree reconfiguration. First, after the deployment, we start by creating the initial routing tree (Fig.5(b)). Secondly, by considering the user preferences, application requirements and the AHP method, the sink calculates the weight vector and transmits it to all nodes in the network. As a result, all nodes initiate the selection phase using the weight vector, the data matrix and the TOPSIS method (For example, node number“9”was con nected to node number“2”in the initial routing tree, but after applying the weight vector, it switches to node number“4”). We can observe that the actual routing tree(Fig.5(c))is more balanced in terms of congestion and node overload. Suddenly, the user changes his or her preferences(the user requires quick data collection more than sustainable energy consumption), so the sink recalculates the weight vector. This prompts the node to rebuild the selection mechanism(node“9”switches to node number“2”, Fig.5(d)). The last stage(Fig.5(e))presents a routing tree reconfiguration when a node(node number“2”) depletes it battery or disconnects from the network.

Figure 4

Figure 4   Selection process based on AHP and TOPSIS methods and change of transmission power


Ⅴ. TEST EVALUATION

A. Simulation Configuration

Before introducing the experimental design and discussing the simulation results, we make the following assumptions.

• The network is composed of homogeneous sensor nodes.

• All sensor nodes have the same initial energy.

• The sink is not limited in terms of energy and memory.

• The radio channel is symmetric such that the energy consumption of transmitting data from node X to node Y is the same as that of transmission from node Y to node X.

In most work in our field, authors assume that the power level can be adjusted to the exact needs and calculate the energy cost using these exact values. In reality, this is not the case, as a radio can only be adjusted to one of the associated power levels and not set to the exact transmission power needed[23].

We model our energy-constrained network as a sensor network. Specifically, we choose to utilize a CC2420 radio chip that offers eight transmission power levels[24]. As described in Ref. [23], these Tx power levels have energy consumption in the form of PT =V ·(PROON · TStartup+Cplevel ·(L/(Trate))), where PT is the total power consumption to transmit a packet of length L at power level plevel. Voltage, V, is determined by the current battery voltage of the transmitter. PROON is the current usage in Amperes when the radio is in the state of“Radio On, Oscillator On”. TStartup is the length of time needed to start up the radio oscillator in seconds. The term Cplevel denotes the current usage in mA at the power level, plevel. The cost of PT also depends upon the length of the packet being sent and the transmission data rate. Trate denotes the transmission rate in bits per second.

The reception cost is calculated based on the CC2420 specifications. The specifications provide the receiving cost CR as 19. 7 mA. The cost in joules to receive a packet is PR. PR=VCRx(L/(Trate)).

Figure 5

Figure 5   Example of routing tree reconfiguration and emission of weight vector: (a)-(b)the process of creating an initial routing tree; (c)an actual routing tree; (d)a routing tree where nodes rebuild the selection mechanism; (e)routing tree reconfiguration


We evaluate performance in terms of three performance metrics: network lifetime, network coverage and survivability of critical nodes.

• Network lifetime and energy efficiency. Network lifetime is defined as the maximum number of communication rounds during which the collected information can be effectively delivered to the base station[25]. We demonstrate the improvement of our protocol and its efficiency against some known protocols by comparing the lifetime for each of them and the rate of the alive nodes at particular lifetime instances (for example, at 50% and 75% of network lifetime).

• Network coverage. Network Coverage is defined with stricter requirements such that no sub-grid, across the range of all nodes, suffers a loss of more than a certain percentage of its nodes. The concentration of dead nodes in one geographic zone affects the protocol performance. In some applications, like forest fire surveillance, having a global view of the sensing area plays an important role.

• Survivability of critical. The survivability of critical nodes is defined as the percentage of critical nodes alive, where critical nodes are the set of nodes that should mainly be preserved in terms of importance. We specifically focus on two aspects. The first is the manner in which nodes with a high betweenness centrality are dying or disconnecting during the network lifetime. This effect impacts the congestion and latency, if critical nodes die rapidly, this can only be explained by the fact that congestion occurred in the network. The second treats the lifetime of nodes with a high bridging centrality. Similar to the first aspect, ensuring that critical nodes with high bridging centrality are dying in a balanced way can lead to better robustness and provide more path alternatives in the case of network changes.

In terms of energy efficiency and reliability, hierarchical clustering protocols are still one of the most optimum solutions, therefore, to improve the performance of our work, we compare MCRP to LEACH[8] and HCR[9] protocols presented in the related works section. For these protocols, we fix the number of clusters, cluster-heads and advanced nodes percentage to 10% of the network size. In order to collect homogeneous and comparable results, we applied the same radio model that we have presented.

The experimental scenario consists of a wireless sensor network with a size S that varies from 250 to 1 000 nodes. Each time we change the size, nodes will be randomly deployed and the three protocols will be applied for the same network topology. Many evaluation parameters will be stored, such as the number of dead nodes, battery level for each node, the position of dead nodes and centralities values. To quantify the impact of network density in protocol performances, we use two network areas, 250 × 250 m2 and 300 × 300 m2. Changing the network size and network area changes the density and therefore affects the distances between nodes, transmission power consumption and network lifetime. Tab.3 presents all of the parameter settings used in the simulation.

Table 3   Parameter settings used in the simulation

parametersvalues
network size in m2250 × 250,300 × 300]
numbers of nodes250,500,750,1 000]
sink location(125,125),(150,150)]
initial energy1 000 J
data packet size6 400 bits
control packet size800 bits
battery voltage3V
umber of power levels8 levels
current usage in mAfor each power level8.5,9.9,11.2,12.5,13.9,15.2,16.5,17.4]
maximum radius in mfor each power level10,15,25,35,45,50,55,65]
transmission rate250 kbit/s
reception cost in mA19.7 mA

New window| CSV


The simulation was implemented in programs developed in the C language. We executed these programs in the MATLAB environment, in order to store the data simulation, plot the results and optimize the view.

B. The Lifetime and Dead Nodes

For the simulations, the density of nodes was diversified from 250 to 1 000 to evaluate the impact of this density on network lifetime. The evaluation was also done for different network sizes(200×200 m2, 250×250 m2 and 300×300 m2) to examine the effect of distance between nodes on network performance.

Fig.6 presents the results of comparing the MCRP, LEACH and HCR protocols based on the network lifetime goal.

Figure 6

Figure 6   Comparison of network lifetime for a network size of 300 × 300 m2,using the LEACH,MCRP and HCR protocols: (a)with 500 nodes; (b)with 750 nodes; (c)with 1 000 nodes


In each sub-figure, the x-axis denotes the number of rounds, whereas the y-axis represents the number of dead nodes, assuming a full battery for all sensors, initially. The results indicate that for all values of deployed nodes, MCRP has higher lifetime values based on the last dead node criterion. For example, with 500 nodes, we obtain a network lifetime of around 1 200 rounds for LEACH, 1 400 for HCR and above 1 500 rounds for MCRP. Therefore, MCRP improved the lifetime by approximately 10% in regard to LEACH and HCR. Moreover, with an increase in node density, MCRP displays a consistent increase in lifetime. The improved lifetime rate rises to 15% when we increase the network size up to 1 000 nodes. Additionally, we observe that for the 25% dead nodes criterion, HCR displays a better result in minimal densities(500 nodes) but this gap was reduced with an increase in density.

Fig.7(a)-(c)illustrate the results for comparison in terms of energy consumption, where the y-axis indicates the total residual energy of the network(obtained by summing the remaining battery charge for each alive node in each round)and the x-axis denotes the number of rounds.

It can be observed that the mean residual energy of the system in the case of MCRP is higher than the other protocols. For example, in 7(c)at the round 500, the network holds approximately 40% (4 × 105 J) of the initial energy applying LEACH and HCR. However, at the same instant, MCRP allows the nodes to keep 60%(6 × 105 J)of their initial battery charge.

The performance improvement of our approach can be attributed to the fact that MCRP takes into account the importance of individual nodes in maintaining network connectivity, which accordingly offers more routing choices at each node. Consequently, by conserving the energy of highly connected nodes and reducing the use of shortest paths, MCRP achieves an improvement in the network lifetime.

In addition, MCRP realizes a more balanced energy consumption, compared to the LEACH and HCR protocols. The routing tree formation process in MCRP includes multiple parameters that influence energy consumption. These parameters include the communication cost from the sensor node to the parent and the parent’s residual energy, which helps sensor nodes achieve a balanced energy dissipation in the network.

In order to show the impact of density on MCRP performance and its significant effect on energy consumption, we measure lifetime and the energy consumption in two different areas(250 × 250 m2 and 300 × 300 m2)for the same network size(1 000 nodes). Figs. 8 and 9 present the results concerning network lifetime and energy consumption. It can be observed that with a variation of network size, which automatically leads to an augmentation in the transmission radius, there is an increase in the node energy consumption. However, deploying the same number of nodes in a smaller area reduces the overall energy consumption and increases the overall net work lifetime. In addition, decreasing the network size leads to a decrease in distance between the nodes and makes it possible to transmit directly to a relatively distant neighbor. Nevertheless, in comparing LEACH and HCR, MCRP still produces better results in both smaller and larger area networks.

Figure 7

Figure 7   Comparison of energy consumption for a network size of 300 × 300 m2,using the LEACH,MCRP and HCR protocols: (a)with 500 nodes; (b)with 750 nodes; (c)with 1 000 nodes


C. Network Coverage Goals

While it is evident that networks should be considered effectively alive as long as a certain percentage of all its nodes remain alive, it is often the case that some nodes are more valuable than others[26]. Conversely, the usefulness of the network depends on being able to collect measurements and data from a large physical region. In this case, it is important that for every sub-area across the network, there remain a sufficient number of active nodes.

Figure 8

Figure 8   Comparison of network lifetime using the LEACH protocol,MCRP protocol,and HCR protocol: (a)for a network size of 300 × 300 m2; (b)for a network size of 250 × 250 m2


Figure 9

Figure 9   Comparison of energy consumption using the LEACH protocol,MCRP protocol,and HCR protocol: (a)for a network size of 300 × 300 m2; (b)for a network size of 250 × 250 m2


A demonstration of the system lifetime, based on the position of dead node, is performed with different network sizes and densities. Fig.10 presents the results for each protocol, MCRP and LEACH. For LEACH, we can observe that dead node positions formed a circular cloud or orbit around the sink. In addition, distant nodes from the sink died first due to an increase in energy consumption resulting from the connection length between these nodes and the sink, especially when nodes perform like cluster heads. Conversely, using HCRP, dead nodes were dispersed in a balanced way such that the active nodes could maintain a global and efficient vision of the overall regions of the network and were able to collect data.

D. The Survivability of Critical Nodes

In Fig.11, we analyze the survivability of critical nodes based on the betweenness centrality for different network densities. In this set of experiments, we calculate the two centralities for each node in the network, save the dead time of these nodes(25%, 50%, 75% or 100% of the total number of nodes) and localize these nodes in the field. Specifically, we focus on nodes with a higher betweenness centrality as critical nodes in the network. We study the localization of critical nodes and the rate of dead nodes in each quarter of the network lifetime.

Through a comparison between the first and the last quarter of the lifetime, we can see that the loss of critical nodes is equitable during the life of the network. This fact plays an important role in keeping a network operational and maximizing its overall lifetime. The critical nodes are a key player in the communication between the nodes and the sink since they occupy positions near the central node. The loss or failure of these critical nodes can rapidly lead to a decrease in the lifetime of the network as well as the capability to collect data from the farthest nodes.

Figure 10

Figure 10   Network coverage(localisation of dead nodes)using MCRP and LEACH protocols: (a)MCRP in 300 × 300 m2 with 750 nodes; (b)LEACH in 300 × 300 m2 with 750 nodes


Figure 11

Figure 11   Survivability of critical nodes by network size 200 × 200 m2 with 400 nodes using the MCRP protocol based on betweenness centrality: (a)for the first 25% of dead nodes; (b)for the last 25% of dead nodes


In Fig.12, we present the results for dead node distribution in the field based on the bridging centrality of each node. We observe that critical nodes with a high bridging centrality die in an equitable manner during the lifetime of the network. This attainment demonstrates that our routing policy preserves the critical nodes and reduces the risk of their failure due to an excess in energy consumption. A node with a high bridging centrality plays a central role in the communication process inside the network by associating subgroups of nodes localized in separate regions amid a sensing field. Minimizing the failure of these nodes ensures a reliable robustness by maintaining alternatives paths to the sink especially when ordinary routing nodes(those with a high betweenness centrality)die or disconnect from the network.

Figure 12

Figure 12   Survivability of critical nodes by network size 200 × 200 m2 with 400 nodes using the MCRP protocol based on bridging centrality: (a)for the first 25% of dead nodes; (b)for the last 25% of dead nodes


In summary, we can conclude that the multi-criteria based centrality routing protocol (MCRP) offers significant improvements, compared to the LEACH and HCR protocols in terms of network lifetime, network coverage and critical node survivability. Moreover, based on the comparison of the power consumption for the tree construction between the three approaches and the improvements offered by each protocol, MCRP demonstrates the best performance. This result is ex plained by its capability to adapt, as well as its capacity to dynamically vary the criteria weights for the node selection process.

Ⅵ. CONCLUSION AND FUTURE RESEARCH

To address the key constraints of WSNs, we propose a new routing protocol based on MADM methods and centrality measures. Our proposed method optimizes the overall network lifetime by prolonging those of highly critical nodes. Furthermore, our approach regulates network congestion by favoring relay nodes with less congestion and avoiding positions with high congestion probability. This is done by permanently selecting the optimized relay nodes, which are characterized by a stable energy consumption and a balanced data emission and reception.

As specified in the proposed model, our protocol considers multiple constraints for routing tree formation, which is crucial for a reasonable and balanced energy consumption of the network. In this paper, we have considered many metrics such as energy consumption and topological metrics. Additional performance metrics can be used without adding significant complexity to the algorithm, which makes it a favorable choice, when compared to other techniques. An essential advantage of our work is the integration of user preferences as an input to the model, and the adaptation to network changes.

Simulation results demonstrate that MCRP achieves significant energy savings, and enhances the network lifetime when compared to LEACH and HCR protocols. The multiple parameters involved in the routing tree formation process, for MCRP, help sensor nodes to dissipate their energy at a much more balanced rate, as compared to other protocols. The ability of MCRP to work independently, to constrain the deployment area size and node density, makes it a viable energyefficient scheme for WSNs.

Adding metrics related to mobility into our current framework remains an area for future research[27]. In addition, we anticipate that further improvements, in terms of QoS, can be achieved by reducing the packet loss due to the buffer overflow. This objective can be realized through the integration of a congestion detection mechanism and buffer optimization, based on a policy that identifies nodes with a high congestion probability(high betweenness centrality)and reduces the incoming data-flow over those nodes. A further scope of improvement would be to integrate the multi-sink aspect and to measure its effect on the global performance of our protocol[28].

Owing to the large and fast expansion of wireless applications, the unlicensed spectrum(e. g. , the industrial, scientific and medical(ISM)radio bands)have become over-saturated, while many licensed frequency bands remain unused most of the time. To solve this issue, new dynamic access techniques must be integrated into WSNs by employing cognitive radios[29]. In cognitive WSNs(CWSNs), the network topology still changes constantly and dynamically because motes can adapt and regulate their transmission metrics, and can activate or turn off their radio transceivers based on the presence of a primary user(PU)[30]. Moreover, in the scalable environment of CWSNs, the frequency spectrum is not always accessible for data transmission to all the motes. This makes the routes between the sink and the motes quite varied and dynamic, and poses many new routing challenges in CWSNs[31]. Therefore, our solution with its ability to locally make decisions and adjust the routing tree based on many local criteria, can be a promising choice for CWSN routing with some adjustments and adaptations.

The results achieved by this solution and the relevant features implemented, such as the integration of user preferences, render this protocol a competitive solution when compared to existing research, as well as a promising research field that can be successfully employed in an actual field.

The authors have declared that no competing interests exist.
作者已声明无竞争性利益关系。

Reference

I. F. Akyildiz , M. C. Vuran .

Wireless sensor networks

[M]. Volume 4. John Wiley&Sons, 2010.

[Cited within: 2]

S. Rani , S. H. Ahmed .

Multi-hop routing in wireless sensor networks: an overview,taxonomy,and research challenges

[M]. Springer, 2015.

[Cited within: 6]

L. C. Freeman .

Centrality in social networks conceptual clarification

[J]. Social Networks, 1978, 1(3): 215-239.

[Cited within: 2]

A. Jain , B. Reddy .

Node centrality in wireless sensor networks: Importance,applications and advances

[C]// IEEE 3rd International Advance Computing Conference(IACC),Ghaziabad, 2013: 127-131.

[Cited within: 3]

A. Sarkar , T. S. Murugan .

Routing protocols for wireless sensor networks: What the literature says

[J]. Alexandria Engineering Journal, 2016, 55(4): 3173-3183.

[Cited within: 3]

N. A. Pantazis , S. A. Nikolidakis , D. D. Vergados .

Energy-efficient routing protocols in wireless sensor networks: A survey

[J]. IEEE Communications Surveys Tutorials, 2013, 15(2): 551-591.

[Cited within: 1]

S. Tanwar , N. Kumar , J. J. Rodrigues .

A systematic review on heterogeneous routing protocols for wireless sensor network

[J]. Journal of Network and Computer Applications, 2015, 53: 39-56.

[Cited within: 1]

S. Tyagi , N. Kumar .

A systematic review on clustering and routing techniques based upon leach protocol for wireless sensor networks

[J]. Journal of Network and Computer Applications, 2013, 36(2): 623-645.

[Cited within: 2]

S. Hussain , A. W. Matin , O. Islam .

Genetic algorithm for energy efficient clusters in wireless sensor networks

[C]// IEEE Fourth International Conference on Information Technology (ITNG’07),Dubai, 2007: 147-154.

[Cited within: 3]

S. A. Sert , H. Bagci , A. Yazici .

Mofca: Multi-objective fuzzy clustering algorithm for wireless sensor networks

[J]. Applied Soft Computing, 2015, 30: 151-165.

[Cited within: 2]

B. M. Khan , R. Bilal , R. Young .

Fuzzy-topsis based cluster head selection in mobile wireless sensor networks

[J]. Journal of Electrical Systems and Information Technology, 2017

[Cited within: 1]

X. Li , Z. Guan .

Energy-aware routing in wireless sensor networks using local betweenness centrality

[J]. International Journal of Distributed Sensor Networks, 2013, 9(5): 307-308.

[Cited within: 1]

J. Duan , D. Gao , C. H. Foh , et al.

Tc-bac: A trust and centrality degree based access control model in wireless sensor networks

[J]. Ad Hoc Networks, 2013, 11(8): 2675-2692.

[Cited within: 1]

P. H. Pathak , R. Dutta .

Centrality-based power control for hot-spot mitigation in multi-hop wireless networks

[J]. Computer Communications, 2012, 35(9): 1074-1085.

[Cited within: 1]

L. Sitanayah , K. N. Brown , C. J. Sreenan .

Fault-tolerant relay deployment based on length-constrained connectivity and rerouting centrality in wireless sensor networks

[C]// Proceedings of the 9th European Conference on Wireless Sensor Networks(EWSN’12),Trento, 2012: 115-130.

[Cited within: 1]

W. Hwang , Y. Cho , A. Zhang , et al.

Bridging centrality: identifying bridging nodes in scale-free networks

[C]// Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining,Philadelphia, 2006: 20-23.

[Cited within: 2]

T. Gal , T. Stewart , T. Hanne .

Multicriteria decision making: advances in MCDM models,algorithms,theory,and applications

[M]. Volume 21. Springer Science&Business Media, 2013.

[Cited within: 1]

T.L .

Saaty.Decision making—the analytic hierarchy and network processes(AHP/ANP)

[J]. Journal of Systems Science and Systems Engineering, 2004, 13(1): 1-35.

[Cited within: 2]

C. Hwang , K. Yoon .

Methods for multiple attribute decision making

[M]. In: Multiple Attribute Decision Making,Springer,1981, 1981, 58-191.

[Cited within: 1]

Y. Yao , J. Gehrke .

The cougar approach to in-network query processing in sensor networks

[J]. ACM Sigmod Record, 2002, 31(3): 9-18.

[Cited within: 1]

A. Sharaf , J. Beaver , A. Labrinidis , et al.

Balancing energy efficiencyand quality of aggregate data in sensor networks

[J]. The International Journal on Very Large Data Bases, 2004, 13(4): 384-403.

[Cited within: 1]

Y. H. Suh , K. T. Kim , D. R. Shin , et al.

Traffic-aware energy efficient routing(TEER)using multi-criteria decision making for wireless sensor network

[C]// 5th International Conference on IT Convergence and Security(ICITCS),Kuala Lumpur, 2015: 1-5.

[Cited within: 1]

M. Mallinson , P. Drane , S. Hussain .

Discrete radio power level consumption model in wireless sensor networks

[C]// IEEE International Conference on Mobile Adhoc and Sensor Systems (MASS 2007),Bologna, 2007: 1-6.

[Cited within: 2]

Texas Instruments.

Chipcon cc2420

[EB/OL]. .

URL     [Cited within: 1]

I. Dietrich , F. Dressler .

On the lifetime of wireless sensor networks

[J]. ACM Transactions on Sensor Networks, 2009, 5(1): 5.

[Cited within: 1]

V. Kumar , S. Kumar .

Energy balanced position-based routing for lifetime maximization of wireless sensor networks

[J]. Ad Hoc Networks, 2016, 52: 117-129.

[Cited within: 1]

Y. Gu , F. Ren , Y. Ji , et al.

The evolution of sink mobility management in wireless sensor networks: A survey

[J]. IEEE Communications Surveys&Tutorials, 2016, 18(1): 507-524.

[Cited within: 1]

A. Wichmann , T. Korkmaz .

Smooth path construction and adjustment for multiple mobile sinks in wireless sensor networks

[J]. Computer Communications, 2015, 72: 93-106.

[Cited within: 1]

S. Kumar , D. Singhal , R. M. Garimella .

Cognitive wireless sensor networks Intelligent Sensor Networks: The Integration of Sensor Networks

[J]. Signal Processing and Machine Learning, 2012: 205.

[Cited within: 1]

G. P. Joshi , S. Y. Nam , S. W. Kim .

Cognitive radio wireless sensor networks: applications,challenges and research trends

[J]. Sensors, 2013, 13(9): 11196-11228.

[Cited within: 1]

D. Jiang , X. Ying , Y. Han , et al.

Collaborative multi-hop routing in cognitive wireless networks

[J]. Wireless Personal Communications, 2016, 86(2): 901-923.

[Cited within: 1]

/