SPECIAL SECTION ON INTELLIGENT DATA SENSING, COLLECTION

AND DISSEMINATION IN MOBILE COMPUTING

Received April 13, 2019, accepted May 5, 2019, date of publication May 9, 2019, date of current version May 22, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2915957

QSDN-WISE: A New QoS-Based Routing Protocol

for Software-Defined Wireless Sensor Networks
XIAOBO TAN1,4 , HAI ZHAO1 , GUANGJIE HAN 2,3 , (Senior Member, IEEE),
WENBO ZHANG 4 , AND TENG ZHU4
1 School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China
2 School of Information Science and Technology, Qingdao University of Science and Technology, Qingdao 266061, China
3 Department of Information and Communication System, Hohai University, Changzhou 213022, China
4 School of Information Science and Engineering, Shenyang Ligong University, Shenyang 110159, China

Corresponding author: Guangjie Han ([email protected])

The work was supported in part by the National Key Research and Development Program under Grant YS2017YFGH001945, in part by
the National Natural Science Foundation of China-Guangdong Joint Fund under Grant U1801264, in part by the Six Talent Peaks Project in
Jiangsu Province under Grant XYDXXJS-007, in part by the General Project of Liaoning Education Department under Grant L2015465,
in part by the Key Discipline Open Fund of Shenyang Ligong University, and in part by the Liaoning Natural Science Foundation under
Grant 20170540793.

ABSTRACT Today, a wide variety of applications with different requirements are rapidly developed in
industrial wireless sensor networks, and providing the Quality of Service (QoS) for this kind of communi-
cation network is inevitable. It is difficult to solve the problems of poor adaptability and difficulty in the
implementation of the QoS-based network configuration and management in traditional network architec-
ture. We present a hierarchical software-defined network architecture for wireless sensor networks, which
makes the complex network management possible and the system more adaptable. Furthermore, we propose
a QoS-based routing protocol, called QSDN-WISE, which consists of a clustering algorithm, a routing
algorithm, and local network maintenance. A double-cluster head-based uneven clustering algorithm, called
DCHUC, avoids the energy hole phenomenon and reduces the workload of a single cluster head. The
centralized QSDN-WISE routing algorithm constructs two heterogeneous forwarding paths for nodes, which
meets the requirements for different data levels. Local network maintenance reduces the number of control
messages in the network. The simulation results indicate that the QSDN-WISE can provide the QoS support
for data with different requirements, balance the network energy consumption, and prolong the network’s
lifetime.

INDEX TERMS Wireless sensor networks, QSDN-WISE, DCHUC, clustering algorithm, energy hole,
routing algorithm, network lifetime.

I. INTRODUCTION tant research objective in the industrialized application of

In recent years, with the rapid development of wireless sensor wireless sensor networks.
network technology, there have been increasingly more appli- Researchers have proposed many improved QoS routing
cations for wireless sensor networks (WSNs) [1]–[4]. Dif- protocols based on early classical QoS routing protocols, such
ferent application data have different requirements for QoS, as SAR, SPEED, and EQR, and some have been applied
including time delay and reliability, in various applications in practice [6], [7]. Shortly after IETF proposed the RPL
of industrial wireless sensor networks. Even within the same routing protocol, [8]–[10] adopted the data classification
application, data of different types require sensor networks mechanism and designed independent objective functions for
to operate, store, and transmit the data in different ways [5]. different types of data to meet QoS requirements for wireless
Therefore, designing a reliable and adaptable network com- sensor network applications that was based on it. However,
munication protocol to meet QoS requirements for different the routing protocols mentioned above, which are based on
applications in industrialized environments, and improving the existing architecture, still face some bottlenecks in prac-
the overall performance of the network, has become an impor- tical applications:
(1) Distributed routing protocols are usually adopted
The associate editor coordinating the review of this manuscript and in large-scale deployment of WSNs, which can only
approving it for publication was Anfeng Liu. choose an appropriate path based on local network

2169-3536
2019 IEEE. Translations and content mining are permitted for academic research only.
61070 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 7, 2019
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

information [11], [12]. However, it is difficult to choose At present, the research on software-defined wireless sen-
the best path and achieve global optimization. Moreover, sor networks mainly focuses on the network management
the large number of control messages generated in distributed architectures. Research on routing protocol, especially QoS
routing protocols also has an impact on data forwarding. routing protocol based on SD-WSN, is still in the initial
(2) Many communication and computation resources are stage. In this paper, we propose a new QoS-based routing
required in the process of establishing and maintaining rout- protocol for software-defined wireless sensor networks called
ing for large-scale deployment of WSNs, which imposes QSDN-WISE that is based on SDN-WISE. The main contri-
great burdens on sensor nodes with limited energy, commu- butions of the current study are summarized as follows:
nication, and computation capabilities [13], [14]. (1) We develop a centralized architecture based on
(3) Energy, computation, and storage resources of sensor SDN-WISE, which makes the complex network management
nodes are limited, while the routing protocols satisfying the possible and the system more adaptable.
diversity of QoS for WSNs are often very complex [15], [16]. (2) We propose a double cluster head-based uneven clus-
Therefore, it is hard to achieve the configuration and manage- tering algorithm called DCHUC, which reduces the workload
ment of protocols. of cluster heads and avoids the energy hole problem.
(4) Routing protocols are developed and solidified in sen- (3) We present a QoS-based routing protocol based on
sor nodes by manufactures for specific applications of WSNs, SDN-WISE called QSDN-WISE, which can provide QoS
and they are therefore hard to upgrade or replace [17], [18]. support for data with different requirements.
This results in poor adaptability and reusability. The remainder of this paper is organized as follows.
Therefore, new technologies must be introduced into the In Section 2, some related work on software-defined WSNs
applications of industrial WSNs to remedy the shortcomings is examined. Section 3 explains the data classification and
mentioned above. proposed SDN-based architecture for hierarchical WSNs.
The Software Defined Network (SDN) originates from the Section 4 presents QSDN-WISE in detail. The simulation
Clean Slate project of Stanford University, and is a new net- and analysis are shown in Section 5. Section 6 concludes this
work architecture that is one of the most promising solutions paper.
for the future Internet. In the SDN network, the logically cen-
tralized control plane is separated from the forwarding plane. II. RELATED WORK
The control plane manages and configures the network, and Numerous recent studies have focused on software-defined
deploys new network protocols in a programmable manner wireless sensor networks. In 2012, Luo and Mahmud almost
through open and unified interfaces. Network equipment only simultaneously proposed that SDN and WSN should be
forwards data according to the strategy sent from the con- integrated to solve the inherent problems in traditional
troller. Via the centralized control of SDN, the control plane WSNs, and became the pioneers of Software Defined WSN
can acquire network information, such as network topology research [25], [26]. Luo designed Sensor OpenFlow based
and traffic in time, so as to achieve more flexible and effective on OpenFlow protocol to separate the control plane from
management than the traditional network architecture. The the data plane of sensor nodes. Mahmud proposed a pro-
network equipment in SDN uses general hardware, so it is grammable sensor called Flow-Sensor and replaced the tradi-
more convenient to deploy, maintain, and upgrade the net- tional WSNs with software-defined WSNs, thus constructing
work [19]. SD-WSN with programmability and customization. In 2015,
As soon as it was proposed, SDN attracted wide atten- the research on software-defined wireless sensor networks
tions from academia and business circles for its innova- became a prevalent topic. Reference [27] uses standard-
tion and advantages, and it has achieved some theoretical ized commercial hardware and SDN technology to design
research results and practical application cases in wired net- a repeatable configurable WSN framework, which discards
works [20], [21]. After it proved successful in traditional the OpenFlow protocol, adopts loose-coupling method, and
wired networks, SDN was introduced into WSNs, Wireless endows sensor nodes with control functions partly. Each
Mesh Networks, and other wireless networks to improve sensor node in the framework executes policies through a
the overall performance of the network and reduce the microcontroller, and the parameters for the whole network are
cost of network management [22]. After substantial exper- configured globally through a centralized controller. Refer-
imentation, researchers have proven that centralized man- ence [28] proposes a new state-based SDN-WISE framework,
agement based on Software Defined WSN (SD-WSN) has which is designed not only to reduce the amount of data
more advantages in the aspects of sensor node simplifi- exchanging between the controller and sensor, but also to
cation, efficient network management and diagnosis, and make sensors support a state-based programmable mode by
network parameter configuration, among others, compared finite state machine. In this way, sensor nodes in SDN-WISE
with the traditional distributed management of WSNs [23]. can run operations that are not supported in stateless solu-
Furthermore, centralized management based on SD-WSN tions. In addition, SDN-WISE provides intact communication
has been considered to better adapt to the future develop- protocols and API interfaces, which provides a potential basis
ment of WSNs, and this view has been widely accepted by for its further research and development. The SDN-WISE
academia [24]. protocol stack is illustrated in Fig.1. Reference [29] proposes

VOLUME 7, 2019 61071

 X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

TABLE 1. Data classification table.

FIGURE 1. SDN-WISE protocol stack.

an implementation method for sensors using MCUs and

FPGAs with ultra-low energy consumption to support
SDN, which lays a foundation for the practical application
of SD-WSN.
At present, a few researchers have focused on routing pro-
tocols based on SD-WSN architecture. Reference [30] pro-
poses a QoS-based mechanism that is based on SDN-WISE.
In the mechanism, the controller uses the state to give infor-
mation about the congestion condition at the node, which
can deal with traffic flows for different QoS requirements.
Reference [31] proposes a disjoint multipath routing protocol
based on SDN (SDN-DMRP). In SDN-DMRP, the controller
collects relevant information about sensor nodes, calculates
disjoint multipath forwarding routes for sensor nodes through
a centralized algorithm, and sends them to sensor nodes in
the form of flow tables. SDN-DMRP only uses hop count as
parameter to choose the next hop, and does not provide QoS
support for the network.

III. THE PROPOSED MODEL

A. DATA CLASSIFICATION
At present, sensor nodes can simultaneously collect event
FIGURE 2. SDN-based Architecture for hierarchical WSNs.
data, periodic data, and monitoring data of other types [32],
[33]. Generally, after a sensor node collects event data, such
as alarm data, it is necessary to ensure that the monitoring
data is transmitted to the sink node in a timely and reliable their different sensitivities of time delay and packet loss rate.
manner, which ensures that the users can take measures The result of data classification is provided in Table 1.
timely and effectively [34]. Periodic monitoring data col-
lected by sensor nodes, such as temperature and humidity, B. SDN-BASED ARCHITECTURE FOR HIERARCHICAL
need periodic updates, so they do not require real-time and WSNS
reliability. However, audio and video data collected by sensor The proposed architecture is outline in Fig.2, and consists of
nodes often need to ensure a certain real-time performance. three components: the application plane, the control plane,
For sensitive targets in the monitoring environment, batch and the data plane. The application plane can dynamically
monitoring data in a short time often needs to be acquired. configure the parameters of the clustering algorithm and
However, such monitoring data often does not require low routing algorithm through the configuration file. The control
time delay, but requires certain reliability. It is difficult for plane consists of a network link discovery module, a topology
sensor nodes to meet different QoS requirements for the data management module, and a routing management module.
if they deal with data of different types in the same way. The data plane consists of a sink node and sensor nodes
Therefore, a routing protocol supporting QoS needs to adopt that support the QSDN-WISE protocol. The control plane
a classification mechanism for monitoring data. interacts with the data plane through a socket.
In order to meet the different requirements of real-time In order to support the functions of SDN, the SDN-WISE
performance and reliability for different types of data, client and flow table are added into the sensor nodes. In order
the QSDN-WISE routing protocol proposed in this paper to adapt to hierarchical network topology, sensor nodes in the
classifies the monitoring data of different types according to data plane are divided into ordinary nodes and cluster-head

61072 VOLUME 7, 2019

X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

nodes according to their functions. Ordinary nodes only real-

ize the function of data acquisition and send monitoring data
to the cluster head, while cluster heads have the functions
of data fusion, maintenance of cluster members, and data
forwarding, in addition to monitoring the data acquisition
function. The cluster information and maintenance informa-
tion generated in the controller are only sent to the cluster
heads. Therefore, the maintenance function of a cluster head
is to receive cluster information and maintenance information
from the controller, and transmit them to its member nodes in
the manner of the flow table.
The control plane implements three main functions, includ- FIGURE 3. Process of clustering & routing algorithm.
ing network link discovery, clustering-based topology man-
agement and maintenance, and routing construction and
maintenance, based on backbone nodes.
A. NETWORK INITIALIZATION PROCESS
1) NETWORK LINK DISCOVERY The network initialization process of QSDN-WISE is illus-
The network link discovery module is responsible for dis- trated in Fig.3. The controller broadcasts a beacon control
covering and maintaining the status information of sensor message to the whole network through the sink node, and
nodes in the network, including node ID, node neighbor initiates the distributed topology discovery process (TD) to
table, node residual energy, node congestion, and link stabil- obtain effective information of sensor nodes, such as node
ity, which are provided for clustering-based topology man- congestion, link stability, node residual energy, node neigh-
agement and routing construction and maintenance. Sensor borhood information, node ID, and RSSI. The data flows
nodes in the network use a classical Link Layer Discov- from points 1 to 4 complete the process of network topology
ery Protocol (LLDP), such as IEEE 802.15.4, 802.11b, etc., discovery and form the initialization network topology. After
to implement the distributed link discovery algorithm, send completing the topology discovery process, the topology
information about the nodes and links to the controller, and management module in the controller conducts the DCHUC
then save them in the database. clustering algorithm to cluster the network based on the
global topology information of the network. The data flows
2) TOPOLOGY MANAGEMENT AND MAINTENANCE from points 5 to 8 complete the clustering process. After
Clustering is an effective method for the management of the network clustering is completed, the routing management
large-scale deployment of wireless sensor networks. There- module constructs a backbone routing tree supporting QoS
fore, the topology management and maintenance module for the backbone network. The data flows from points 9 to
utilizes the related status information of the nodes and links 10 complete the routing process.
obtained by the network link discovery module to execute the
centralized clustering algorithm and cluster maintenance, and B. DCHUC CLUSTERING ALGORITHM
saves the results to the database.
DCHUC is a centralized algorithm executed in the controller,
so it not only can realize global optimization, but can also
3) ROUTING MANAGEMENT reduce the number of control messages generated in the net-
The routing management module conducts a centralized rout- work, which further reduces the frequency of network storm
ing algorithm to construct and maintain QoS routing for the and network energy consumption. Furthermore, DCHUC
backbone network by utilizing the related status information adopts a non-uniform clustering mechanism, so it ensures
of the nodes and links, clustering result information, etc., and the formation of more clusters in the area near the sink to
saves the results to the database. forward network data, which avoids the energy hole prob-
lem and prolongs the network lifetime. In addition, DCHUC
IV. QSDN-WISE ROUTING PROTOCOL adopts the double cluster head mechanism, in which double
The QSDN-WISE routing protocol proposed in this paper cluster heads are elected with low node congestion and high
includes four parts: the clustering algorithm, routing algo- link stability, respectively, according to the factors of node
rithm, cluster maintenance, and routing maintenance. Dur- congestion, link stability, and node residual energy. This not
ing the initialization process, the controller conducts only reduces the workload of single cluster head, but also sup-
clustering and routing algorithms to complete initial net- ports QoS services intra-cluster. Although the double cluster
work construction. Then the controller realizes the event- head mechanism of DCHUC forms two different network
triggered cluster and local routing maintenance based topologies, which increases the difficulty in management and
on the global topology information, which is updated control of the network, the centralized management mode
regularly. based on SD-WSN architecture makes up for this defect.

VOLUME 7, 2019 61073

 X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

DCHUC includes two processes: the election of cluster head TABLE 2. Pseudo-code of cluster head election.
and the clustering of ordinary nodes.

1) ELECTION OF CLUSTER HEAD

In order to achieve non-uniform clustering, DCHUC applies
the Min-Max Scaling idea and calculates the competition
radius for each sensor node with eq. (1).
dmax − d(ni )
rcmp (ni ) = (1 − c × ) × rmax (1)
dmax − dmin
In eq. (1), rcmp (ni ) denotes the competition radius of node
ni ; rmax is a constant number which represents the preset
maximum competition radius; dmax and dmin denote the max-
imum and minimum distance of all nodes from the sink node;
d(ni ) denotes the distance between node ni and the sink
node; finally, c is an adjusting parameter to ensure that the
competition radius is more reasonable.
The QSDN-WISE proposed in this paper requires ordinary
nodes in a cluster to send data to the cluster head directly.
According to the energy consumption model [35], [36], it has successfully. M (ni ) denotes the number of packets sent by the
been concluded that the competition radius of clusters should node ni to its neighbor nodes.
be less than d0 as much as possible in order to save network After calculating the probability of becoming a low node
energy consumption. Therefore, we set the competition radius congestion cluster head and high link stability cluster head,
of clusters (rmax ) to rmax ≤ d0 . d0 is a constant value of about respectively, for all the nodes, the controller sorts all the nodes
87 m. Eq. (1) ensures that there are more cluster heads near by node congestion in ascending order and by link stability in
the sink node. descending order, respectively, and saves the corresponding
DCHUC considers the factors of node congestion, link IDs of the nodes to the node congestion cluster head proba-
stability, and node residual energy to elect the cluster head. bility list (LPNC ) and the link stability cluster head probability
The controller calculates the competition radius for each node list (LPLS ), respectively. Then, the controller successively
according to eq. (1), and maintains a neighbor list (Lneighbor ) stores the IDs in (LPNC ) and (LPLS ) o the cluster list with low
for each node to store the ID of the neighbor nodes that are node congestion (LCHNC ) and the cluster list with high link
within the competition radius of the node. Then, the controller stability (LCHLS ), respectively, and deletes the node ID and
calculates the probabilities of becoming a low node conges- its neighbor node IDs from (LPNC ) and (LPLS ) until (LPNC )
tion cluster head and a high link stability cluster head for each and (LPLS ) are empty, i.e. all nodes are covered by cluster
node according to eqs. (2) and (3), respectively. heads. The pseudo-code for the centralized election cluster
Eresidual (ni ) head process is presented in Table 2.
PNC (ni ) = × (1 − NC(ni )) (2) After cluster head election, two heterogeneous sets of
Ēneighbor (ni )
cluster heads, each of which covers the whole network, are
Eresidual (ni )
PLS (ni ) = × LS(ni ) (3) obtained: a low node congestion cluster head set (LCHNC )
Ēneighbor (ni ) and high link stability cluster head set (LCHLS ). In order to
In eq. (2), PNC (ni ) denotes that node ni becomes the cluster implement clustering for ordinary nodes, in the process of
head with a low node congestion degree. Eresidual (ni ) denotes cluster head election, the controller maintains two candidate
the residual energy of the node ni . Ēneighbor (ni ) denotes the cluster head lists for each ordinary node, which are the low
average residual energy of the node ni . NC(ni ) denotes the node congestion candidate cluster head list LCCHNC and the
congestion degree of the node ni . In eq. (3), PLS (ni ) denotes high link stability candidate cluster head list LCCHLS .
the link stability of the node ni . Node congestion and link
stability are shown in eq. (4) and eq. (5) [37]. 2) CLUSTERING OF ORDINARY NODES
Q(ni ) After cluster heads are elected, DCHUC goes into the clus-
NC(ni ) = (4) tering process of ordinary nodes. During the election of two
L(ni )
types of heterogeneous cluster heads according to eqs. (2)
S(ni )
LS(ni ) = (5) and (3), a phenomenon in which one node becomes the cluster
M (ni ) head of two types of heterogeneous clusters at the same time
In eq. (4), Q(ni ) denotes the buffer queue length of node may occur, and it is allowed in DCHUC. However, in order
ni which has been occupied. L(ni ) denotes the total buffer to avoid data transmission between two heterogeneous cluster
queue length of node ni . In eq. (5), S(ni ) denotes the number heads in one cluster, DCHUC does not allow a cluster head
of packets sent by the node ni that arrives to its neighbor nodes of one type to become a member node of the other type.

61074 VOLUME 7, 2019

X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

TABLE 3. Pseudo-code of clustering.

In the clustering process of a certain type of cluster, if there FIGURE 4. Non-uniform clustering map of DCHUC.
is only one cluster head in the competition radius of an
ordinary node, this ordinary node will directly become a
member of the cluster in which the cluster head is located.
However, there may be multiple cluster heads within the
competition radius of an ordinary node. In order to solve this
problem, this paper defines clustering strength as shown in
eq. (6). DCHUC calculates the clustering strength of each
cluster in its competition radius for the node in this situation,
and chooses to join the cluster with the highest clustering
strength.
1−θ
Sincluster = θEresidual (nj ) + , θ ∈ (0, 1) (6)
dni −nj

In eq. (6), Eresidual (nj ) denotes the residual energy of the

cluster head nj ; dni −nj denotes the distance between the node
ni at the intersection of multiple clusters and the cluster head
nj ; finally, θ is a weight parameter with a value between FIGURE 5. Undirected connection graph formed by cluster heads.
0 and 1. Eq. (6) shows that if the residual energy of cluster
head nj is higher, and the distance between the cluster head
nj and the ordinary node ni is smaller, the ordinary node ni
will be more likely to become a member of the cluster in C. QSDN-WISE ROUTING ALGORITHM
which nj is located. Pseudo-code of clustering is presented In the QSDN-WISE routing algorithm, the controller adopts
in Table 3. the same QoS routing algorithm for each hierarchical topol-
After the clustering process, the controller maintains two ogy generated by DCHUC. Therefore, we will introduce the
heterogeneous cluster head lists which are LCHNC and QSDN-WISE by using the backbone network composed of
LCHLS , two cluster member lists for each cluster head which low congestion cluster heads as an example in this section.
are LCMNC and LCMLS , and two cluster heads for each The controller establishes an undirected connected graph
ordinary node which are CHNC and CHLS . After DCHUC, for the backbone network composed of cluster heads and a
the controller generates a Cluster Head Notification mes- sink node. In order to ensure the connectivity of the net-
sage (CH _NP), a Cluster Member Notification message work, we set the communication radius of sensor nodes to be
(CM _NP), and corresponding flow table rules, and transmits 1.5 times that of the competition radius, and established an
them to the corresponding sensor nodes. Fig. 4 illustrates the undirected connected graph, as illustrated in Fig. 5. The sink
network topology generated by DCHUC. node acquires its first hop of downward data transmission
As shown in Fig. 4, there are two types of cluster heads in via the communication radius rmax−Sink to further alleviate
the network after the DCHUC clustering algorithm, namely the energy hole. The value of rmax−Sink is much larger than
the low node congestion cluster head and the high link stabil- the communication radius of an ordinary sensor node, as the
ity cluster head, and these two types of cluster heads form two resources of the sink node are not limited.
hierarchical network topologies with their member nodes, After the undirected connected graph is formed, the con-
respectively. troller chooses the best next hop for each node according

VOLUME 7, 2019 61075

 X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

TABLE 4. Weight values for different data levels.

to eq. (7).
αk × NC(nj ) + βk × (1 − LS(nj ))
OFk (ni ) = ,
Eresidual (nj )
(dnj −Sink < dni −Sink ) (7)
In eq. (7), Eresidual (nj ) indicates that the node with high
residual energy should be selected as its next hop; NC(nj ) FIGURE 6. Upward routing tree for data level 2.
indicates that the node with low node congestion should be
selected as its next hop if the data sent by the node requires
low latency; LS(nj ) indicates that the node with high link from the sensor node to the sink node by setting αk and
stability should be selected as its next hop if the data sent βk for different types of data. The control information sent
by the node requires high reliability; dnj −Sink < dni −Sink from the sink node to the sensor nodes requires high real-
indicates that the node that should be selected as its next hop time capability and reliability, so the downward routing from
is the node with a distance from the sink that is less than the the sink node to the sensor node is based on the routing tree
distance between ni and the sink. Finally, αk and βk denote of data level 1, and the reverse routing tree is established.
the weight parameters of the k th data type, and their values Considering the high energy consumption of sensor nodes
range from [0,1]. We can conclude that the value of αk should in the area near the sink node due to the requirement of
be greater than that of βk for an important or emergency forwarding much more data, the communication radius of the
data type, while the value of αk should be less than that of sink node is increased and the best next hop of the sink node
βk for a non-emergency data type. In this paper, we set the in one hop range is chosen by eq. (7), so as to complete the
weight parameters as follows: the value of αk is set to 1 for reverse routing. Fig. 6 shows the final upward routing tree for
the data type with real-time requirement, otherwise it is set data level 2.
to 0; the value of βk is set to 1 for the data type with high
reliability requirement, otherwise it is set to 0. Table 4 lists D. MAINTENANCE FOR CLUSTERS AND ROUTING
the corresponding values of αk and βk for the 4 data levels Distributed clustering and routing algorithms in traditional
that are adopted in this paper. network architecture often adopt periodic cluster head rota-
The controller uses the objective function (OF) to choose tion and backbone network reconstruction. This process
the best next hop for each type of data of each backbone results in the periodic generation of a large number of control
node. Routing algorithms supporting QoS for WSN usually messages, which not only consumes too much node energy,
adopt different OF formulas for different types of data, which but also interferes with the transmission of monitoring data.
increases the complexity of the routing algorithm and param- Although the IRPL routing protocol adopts event-triggered
eter configuration. In contrast, QSDN-WISE uses a unified cluster head rotation and a local routing maintenance mech-
OF formula and meets the QoS requirements of different anism, which reduces the number of periodically controlled
types of data by only adjusting the two weight parameters messages, a large number of control messages will still be
of αk and βk , which reduces the complexity of the routing generated in the network when the event is triggered. QSDN-
algorithm. WISE adopts event-triggered cluster head rotation and a local
QSDN-WISE satisfies different QoS requirements for dif- routing maintenance mechanism, and implements a central-
ferent types of data by setting different values of αk and βk . ized algorithm at the controller side, which avoids the gener-
In Table 4, the objective function of αk = 1 and βk = 0 ation of a large number of control messages when the event
satisfies the low-latency QoS requirement of data level 2, is triggered, thus improving network performance.
i.e. choosing the node with high residual energy and low After the network initialization process of QSDN-WISE is
node congestion degree as the next hop. If αk = 0 and completed, sensor nodes in the network regularly send state
βk = 0, it indicates that QSDN-WISE chooses the next hop information, such as node residual energy, node congestion,
while only considering the node residual energy factor, which and link stability, to the controller so that the controller
satisfies the QoS requirement of the data level 4 without can keep up-to-date global network information. Because of
requirements for time delay and packet loss rate. Therefore, the small amount of data of node status information and
the controller constructs an independent upward routing tree the absence of a broadcasting mode, the network storm

61076 VOLUME 7, 2019

X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

phenomenon that is easily produced by a traditional dis-

tributed algorithm is avoided. The implementation of event-
triggered cluster head rotation and local routing maintenance
mechanism by the controller is described below.

1) CLUSTER HEAD ROTATION BASED ON EVENT TRIGGER

If the residual energy and the congestion degree of the cluster
head is greater than the preset threshold, the link stability is
less than the preset threshold, or the information of the cluster FIGURE 7. QSDN-WISE packet header format.
head cannot be updated by the controller, the cluster head
rotation in the controller side will be triggered. The controller
chooses eq. (2) or (3), according to the type of cluster head
(low node congestion or high link stability, respectively),
to calculate the probability of becoming the cluster head of
this type, and chooses the node with the largest probability
value as the successor cluster head. After the successor cluster
head is elected, the controller generates the cluster head
notification message (CH _NP), cluster member notification
message (CM _NP), and the corresponding flow table rules,
and sends them to the sensor nodes of the cluster.

FIGURE 8. QSDN-WISE report packet format.

2) LOCAL ROUTING MAINTENANCE
Event-triggered cluster head rotation can lead to the breakage
of the global backbone network routing tree maintained by the
data, the data level field that occupies 4 bytes is added to the
controller. Therefore, after the controller performs the cluster
original SDN-WISE header to support up to 16 data levels.
head rotation, the backbone network needs to be partially
The report message format of SDN-WISE is modified as
repaired. First, the controller obtains the type of the original
presented in Fig. 8. Bytes from 0 to10 are the message header
cluster head to determine which routing tree to repair. Then,
of the QSDN-WISE. The 12th and 13th bytes are the hop
the controller obtains all the child nodes in the routing tree,
and node residual energy. The 14th and 15th bytes are the
chooses the best next hop for the node and all its child nodes at
node congestion degree and link stability degree. The 16th
each data level, and sends the corresponding flow table rules
byte is the number of neighbor nodes. Starting from the 18th
to the corresponding nodes. The process of choosing the best
byte, the node address, RSSI, and node residual energy of the
next hop has been described in detail in Section C.
neighbor nodes are defined.
In the QSDN-WISE protocol, the cluster head notification
E. QSDN-WISE PROTOCOL DESIGN message (CH _NP) and cluster member notification message
The QSDN-WISE proposed in this paper is an improvement (CM _NP) are added based on the original SDN-WISE mes-
of the SDN-WISE protocol, on which it is based. We will sage.
introduce the QSDN-WISE protocol in the following aspects:
protocol stack, protocol message, and flow table structure. a: CH_NP MESSAGE
The controller builds the CH _NP message according to clus-
1) QSDN-WISE PROTOCOL STACK tering result and notifies the corresponding node to become
The QSDN-WISE protocol stack only modifies SDN-WISE the cluster head node. The format of the CH _NP message is
on the controller side. The topology Manager (TM) of the exhibited in Fig. 9. Bytes from 0 to 10 are the message header
SDN-WISE controller only generates planar network topol- of the QSDN-WISE. Starting from the 12th byte, the ID and
ogy, which is difficult to adapt to WSN applications with address information of all the member nodes of the cluster
large-scale deployment of sensor nodes. DCHUC cluster are defined. The value of the Type field in the CH _NP header
management is added to the TM module of the controller is 8.
to generate a hierarchical network topology structure based
on the SDN-WISE protocol stack. In order to support QoS b: CM_NP MESSAGE
routing, the controller of QSDN-WISE protocol stack adds a The controller builds the CM _NP message according to
routing management module. the clustering result and notifies the corresponding mem-
ber nodes. The format of the CM _NP message is shown
2) QSDN-WISE PROTOCOL MESSAGE in Fig. 10. Bytes from 0 to 10 are the message header of
The header of SDN-WISE is modified as shown in Fig. 7. the QSDN-WISE. The 12th and 16th bytes are IDs of the
In order to meet the QoS requirements for different types of two cluster heads of different types. The 13th and 17th bytes

VOLUME 7, 2019 61077

 X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

TABLE 5. Simulation platform information table.

FIGURE 9. QSDN-WISE CH_NP packet format.

FIGURE 10. QSDN-WISE CM_NP packet format.

TABLE 6. Simulation parameter table.

FIGURE 11. QSDN-WISE flow table example.

are the type of the cluster heads. 0 is the cluster head with
low data congestion, and 1 is the cluster head with high link
stability. The 14th , 15th , 18th , and 19th bytes are the addresses
of two types of cluster heads.

c: QSDN-WISE FLOW TABLE

The QSDN-WISE protocol adds a data level field to the
matching rules based on the original SDN-WISE flow table
structure to match different data levels. Fig. 11 provides an
example of the modified QSDN-WISE matching rules and
behaviors. The destination addresses of the two flow table
entries are the same in the figure, but the matching data QSDN-WISE adds some parameters based on the simula-
levels are different and correspond to level 1 and level 2, tion parameters provided by SDN-WISE, such as node buffer
respectively. The matched actions are also different. The data size and d0 . Table 6 describes the specific parameters.
of matching level 1 will be forwarded to C, while the data of
matching level 2 will be forwarded to D. B. SIMULATION RESULTS ANALYSIS
We compare QSDN-WISE with SDN-DMRP and SDN-
V. SIMULATION & ANALYSIS WISE, which are based on SDN architecture, and IRPL,
A. SIMULATION SETTINGS AND PERFORMANCE HEED, and EEUC, which are based on the traditional net-
INDICATORS work architecture.
QSDN-WISE protocol is an improved protocol that is based Fig. 12 shows the effect of adjusting the parameters c
on SDN-WISE, while the SDN-WISE project is still under- and rmax on clustering results. In Fig. 12, the abscissa is the
going continuous improvement. At present, the JAVA version value of rmax and the ordinate is the number of cluster heads
of SDN-WISE has become an open source project and has after clustering. Each curve represents the variation of the
been published in GitHub [38]. The C version of SDN- number of cluster heads with rmax under the condition that
WISE has been used in the experimental environment of real c is a constant. It can be concluded that the number of cluster
sensors. The QSDN-WISE protocol proposed in this paper heads decreases with the increase of rmax if the adjustment
is improved and simulated based on the JAVA version. The parameter c is constant, while the number of cluster heads
specific information of the simulation platform is provided increases with the increase of the adjustment parameter c if
in Table 5. rmax is constant. It can also be concluded from the simulation

61078 VOLUME 7, 2019

X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

FIGURE 12. DCHUC clustering result.

FIGURE 14. Comparison of packet loss rate.

that of DCHUC. With the increase of time, cluster head rota-

tions are more frequent in IRPL, which requires more mes-
FIGURE 13. Comparisons of control message overhead. sage overhead than that in DEHUC. In the simulation, we can
also determine that the growth rate of the control messages
of EEUC and HEED is extremely slow, while the growth
result in Fig. 12 that there is no obvious jitter in the number rate of the control messages of IRPL and DCHUC increases
of cluster heads, which indicates that the cluster heads are significantly after the network operates about 700 rounds in
distributed evenly in the network. EEUC, and about 600 rounds in HEED. Starting from about
Fig. 13 shows the comparison of the control message 600 rounds, more dead nodes gradually appear in the network
overhead between the DCHUC, HEED, EEUC, and IRPL running HEED and EEUC, which results in the reduction
clustering algorithms under the condition that the number of the number of control messages. In contrast, cluster head
of network nodes is 200. HEED and EEUC clustering algo- rotations in IRPL and DCHUC become frequent, which sig-
rithms perform the clustering algorithm periodically, so we nificantly increases the number of control messages.
count message overhead in rounds. Fig. 14 illustrates the comparison of packet loss rate
It can be concluded that the control message overhead between QSDN-WISE, SDN-WISE, and SDN-DMRP pro-
of the above four clustering algorithms increases with time, tocols under the condition that the number of nodes is
and the control message overhead of DCHUC and IRPL is 200. In the simulation, the data are divided into four levels
significantly less than that of HEED and EEUC. HEED and according to Table 4, namely CLASS1, CLASS2, CLASS3,
EEUC periodically execute their respective clustering algo- and CLASS4. It can be concluded that the packet loss
rithms in a round-based manner, resulting in a large number rates of CLASS1 and CLASS3 in the network running the
of control messages, while both DCHUC and IRPL adopt an QSDN-WISE protocol are significantly less than those of
event-based clustering algorithm. Large numbers of control CLASS2 and CLASS4, because the data sensitive to packet
messages are mainly generated in the initial stage of the loss rate in CLASS1 and CLASS3 will choose the node with
network. After that, local topology maintenance is adopted, high link stability as its next hop, thus reducing the packet
and the network generates fewer control messages. HEED loss rate of data forwarding. However, SDN-WISE and SDN-
chooses cluster heads randomly, which reduces the number of DMRP only adopt the hop number to choose its next hop, and
generated control messages. EEUC needs to exchange a large do not support QoS service. Therefore, the packet loss rates in
number of control messages among nodes when choosing SDN-WISE and SDN-DMRP are close to the packet loss rates
cluster heads. Therefore, it is evident that the control message of CLASS3 and CLASS4 data. It is evident that QSDN-WISE
overhead of HEED is lower than that of EEUC. The control provides QoS support for data sensitive to packet loss
message overhead of DCHUC before 557 rounds in Fig. 13 is rate.
slightly greater than that of the IRPL clustering algorithm Fig. 15 displays the end-to-end delay comparison between
because the controller that implements centralized DCHUC QSDN-WISE, SDN-WISE, and SDN-DMRP protocols
after network initialization needs to periodically obtain the under the condition that the number of network nodes is
status information of all nodes in the whole network, and 200. It can be concluded that the end-to-end delays of
consumes a certain amount of message overhead. In contrast, CLASS1 and CLASS2 in QSDN-WISE are significantly less
IRPL only needs more control messages for topology main- than those of CLASS3 and CLASS4 data. The end-to-end
tenance, and there are no large-scale cluster head rotations delay in SDN-WISE is greater than that of CLASS1 and
before 557 rounds; therefore, the overhead is less than that of CLASS2 in QSDN-WISE, but less than that of CLASS3 and
DCHUC. Starting from about 558 rounds, the control mes- CLASS4 in QSDN-WISE. Although SDN-WISE considers
sage growth of the IRPL clustering algorithm is greater than the shortest path based on hop number, the cluster head must

VOLUME 7, 2019 61079

 X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

that in SDN-DMRP and SDN-WISE because the clustering

algorithm in QSDN-WISE considers the node residual energy
when choosing cluster-head nodes, and adopts the idea of
non-uniform clustering to balance the network energy con-
sumption. Additionally, QSDN-WISE considers node resid-
ual energy and adopts multi-path routing based on double
cluster heads, which further balances the energy consumption
of the network.

VI. CONCLUSION
A hierarchical software-defined wireless sensor network
FIGURE 15. Comparison of end-to-end delay.
architecture and a QoS-based clustering routing protocol
called QSDN-WISE are proposed in this paper to solve
the problems of poor adaptability and difficulty in network
configuration and management, and support QoS in tradi-
tional network architecture for wireless sensor networks. The
DCHUC clustering algorithm in QSDN-WISE adopts a non-
uniform clustering mechanism to avoid an energy hole in
the "hot spot" network area. The double cluster head mech-
anism with a low time delay and high reliability reduces
the workload of a single cluster head, and implements QoS
requirements for the data in clusters. Based on data clas-
sification, the centralized QSDN-WISE routing algorithm
considers residual energy, node congestion degree, link sta-
bility, and distance between nodes as parameters to choose
FIGURE 16. Comparison of network lifetime. the next hop for nodes, and builds two heterogeneous for-
warding paths for nodes, to meet the requirements of different
data classes. Simulation results show that the QSDN-WISE
forward data of all levels in the cluster to the same relay node. clustering routing protocol can not only balance the energy
When the relay node forwards data, the processing delay of consumption of WSN, but also provides QoS support for data
the node will increase, and the processing delay accumu- with different QoS requirements, and compared with SDN-
lated during multi-hop forwarding will increase significantly. WISE, SDN-DMRP, and IRPL, QSDN-WISE exhibits good
SDN-DMRP adopts multi-path routing based on the shortest performance in energy saving, end-to-end delay, packet loss
path, so the time delay is less than that in SDN-WISE, but rate, and control of the number of messages.
still greater than that of CLASS1 and CLASS2 in QSDN-
WISE. The member nodes in the cluster in QSDN-WISE will REFERENCES
forward data with low time delay to the cluster head with [1] G. Han, H. Wang, M. Guizani, S. Chan, and W. Zhang, ‘‘KCLP:
low node congestion degree, and the cluster head of this type A k-means cluster-based location privacy protection scheme in WSNs for
IoT,’’ IEEE Wireless Commun. Mag., vol. 25, no. 6, pp. 84–90, Dec. 2018.
will still choose the node with low node congestion degree doi: 10.1109/MWC.2017.1800061.
as its relay node. Therefore, compared with SDN-WISE and [2] W. Zhang, J. Wang, G. Han, X. Zhang, and Y. Feng, ‘‘A cluster sleep-wake
SDN-DMRP, QSDN-WISE provides better QoS support for scheduling algorithm based on 3D topology control in underwater sensor
networks,’’ Sensors (Basel), vol. 19, no. 1, Jan. 2019, Art. no. E156.
the data that needs low time delay. [3] G. Han, H. Wang, J. Jiang, W. Zhang, and S. Chan, ‘‘CASLP: A confused
Network lifetime is an important parameter for measur- arc-based source location privacy protection scheme in WSNs for IoT,’’
ing the performance of wireless sensor networks. In the IEEE Commun. Mag., vol. 56, no. 9, pp. 42–47, Sep. 2018.
[4] G. Han, X. Miao, H. Wang, M. Guizani, and W. Zhang, ‘‘CPSLP: A cloud-
simulation, each node sends data packets at the same rate, based scheme for protecting source location privacy in wireless sensor
and the simulation ends when the first node in the net- networks using multi-sinks,’’ IEEE Trans. Veh. Technol., vol. 68, no. 3,
work runs out of energy. Fig. 16 shows the comparison pp. 2739–2750, Mar. 2019.
between the network lifetime in different numbers of nodes [5] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, ‘‘Priority- and delay-
aware medium access for wireless sensor networks in the smart grid,’’ IEEE
in the network in the QSDN-WISE, SDN-WISE, and SDN- Syst. J., vol. 8, no. 2, pp. 608–618, Jun. 2014.
DMRP protocols. It can be concluded that under the con- [6] W. Zhang, L. Liu, G. Han, Y. Feng, and Y. Zhao, ‘‘An energy efficient
dition of the same number of nodes, the network lifetimes and QoS aware routing algorithm based on data classification for industrial
wireless sensor networks,’’ IEEE Access, vol. 6, pp. 46495–46504, 2018.
in QSDN-WISE and SDN-DMRP are greater than that in [7] H. Teng et al., ‘‘A novel code data dissemination scheme for Internet of
SDN-WISE. SDN-DMRP adopts multi-path routing, which Things through mobile vehicle of smart cities,’’ Future Gener. Comput.
differentiates the energy consumption, and its network life- Syst., vol. 94, pp. 351–367, May 2019.
[8] N. T. Long, M. P. Uwase, J. Tiberghien, and K. Steenhaut, ‘‘QoS-aware
time is therefore greater than that in SDN-WISE. The net- cross-layer mechanism for multiple instances RPL,’’ in Proc. Int. Conf.
work lifetime in QSDN-WISE is significantly greater than Adv. Technol. Commun., Oct. 2013, pp. 44–49.

61080 VOLUME 7, 2019

X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

[9] O. Gaddour, A. Koubâa, and M. Abid, ‘‘Quality-of-service aware [32] G. Han, X. Yang, L. Liu, S. Chan, and W. Zhang, ‘‘A coverage-aware
routing for static and mobile IPv6-based low-power and lossy sen- hierarchical charging algorithm in wireless rechargeable sensor networks,’’
sor networks using RPL,’’ Ad Hoc Netw., vol. 33, pp. 233–256, IEEE Netw., to be published. doi: 10.1109/MNET.2018.1800197.
Oct. 2015. [33] X. Liu and P. Zhang, ‘‘Data drainage: A novel load balancing strat-
[10] C. H. Barriquello, G. W. Denardin, and A. Campos, ‘‘A geographic routing egy for wireless sensor networks,’’ IEEE Commun. Lett., vol. 22, no. 1,
approach for IPv6 in large-scale low-power and lossy networks,’’ Comput. pp. 125–128, Jan. 2018.
Elect. Eng., vol. 45, pp. 182–191, Jul. 2015. [34] M. Huang, W. Liu, T. Wang, H. Song, X. Li, and A. Liu, ‘‘A queuing delay
[11] G. Han, H. Guan, J. Wu, S. Chan, L. Shu, and W. Zhang, ‘‘An utilization scheme for on-path service aggregation in services oriented
uneven cluster-based mobile charging algorithm for wireless computing networks,’’ IEEE Access, vol. 7, pp. 23816–23833, 2019.
rechargeable sensor networks,’’ IEEE Syst. J., to be published. [35] L. M. Feeney, ‘‘An energy consumption model for performance analysis
doi: 10.1109/JSYST.2018.2879084. of routing protocols for mobile Ad Hoc networks,’’ Mobile Netw. Appl.,
[12] J. Li et al., ‘‘Battery-friendly based relay selection scheme to prolong vol. 6, no. 3, pp. 239–249, Jun. 2001.
lifetime for sensor nodes in Internet of Things,’’ IEEE Access, vol. 7, no. 1, [36] W. Zhang, G. Han, Y. Feng, and J. Lloret, ‘‘IRPL: An energy efficient
pp. 33180–33201, 2019. routing protocol for wireless sensor networks,’’ J. Syst. Archit., vol. 75,
[13] W. Zhang, G. Han, J. Wang, and Y. Liu, ‘‘A BP neural network prediction pp. 35–49, Apr. 2017.
model based on dynamic cuckoo search optimization algorithm for indus- [37] P. Karkazis, P. Trakadas, H. C. Leligou, L. Sarakis, I. Papaefstathiou, and
trial equipment fault prediction,’’ IEEE Access, vol. 7, pp. 11736–11746, T. Zahariadis, ‘‘Evaluating routing metric composition approaches for QoS
2019. differentiation in low power and lossy networks,’’ Wireless Netw., vol. 19,
[14] X. Liu, T. Qiu, and T. Wang, ‘‘Load-balanced data dissemination for no. 6, pp. 1269–1284, Aug. 2013.
wireless sensor networks: A nature-inspired approach,’’ IEEE Internet [38] SDN-WISE-API. Accessed: Jun. 10, 2018. [Online]. Available: https://
Things J., to be published. doi: 10.1109/JIOT.2019.2900763. www.mvnjar.com/com.github.sdnwiselab/sdn-wise-api/1.0.7/detail.html
[15] W. Zhang, L. H. G. Li, and L. Zhang, ‘‘E2HRC: An energy-efficient het-
erogeneous ring clustering routing protocol for wireless sensor networks,’’
IEEE Access, vol. 5, pp. 1702–1713, 2017.
[16] C.-C. Lin, D.-J. Deng, Z.-Y. Chen, and K.-C. Chen, ‘‘Key design of driving
industry 4.0: Joint energy-efficient deployment and scheduling in group-
based industrial wireless sensor networks,’’ IEEE Commun. Mag., vol. 54,
no. 10, pp. 46–52, Oct. 2016.
[17] J. Tan et al., ‘‘A low redundancy data collection scheme to maximize life-
time using matrix completion technique,’’ EURASIP J. Wireless Commun.
Netw., vol. 2019, no. 5. doi: 10.1186/s13638-018-1313-0.
[18] X. Liu, ‘‘Node deployment based on extra path creation for wireless
sensor networks on mountain roads,’’ IEEE Commun. Lett., vol. 21, no. 11,
pp. 2376–2379, Nov. 2017.
[19] S. Sezer et al., ‘‘Are we ready for SDN? Implementation challenges XIAOBO TAN received the B.E. degree from
for software-defined networks,’’ IEEE Commun. Mag., vol. 51, no. 7,
Liaoning Normal University, in 2000, and the M.E.
pp. 36–43, Jul. 2013.
degree from Northeast University, China, in 2006,
[20] I. F. Akyildiz, A. Lee, P. Wang, M. Luo, and W. Chou, ‘‘A roadmap for
where he is currently pursuing the Ph.D. degree
traffic engineering in SDN-OpenFlow networks,’’ Comput. Netw., vol. 71,
pp. 1–30, Oct. 2014. with the School of Computer Science and Engi-
[21] Q.-Y. Zuo, M. Chen, G.-S. Zhao, C.-Y. Xing, G.-M. Zhang, and P.-C. Jiang, neering. He is currently an Associate Professor
‘‘Research on OpenFlow-based SDN technologies,’’ J. Softw., vol. 24, with the School of Information Science and Engi-
no. 5, pp. 1078–1097, May 2013. neering, Communication and Network Institute,
[22] W. Ejaz, M. Naeem, M. Basharat, S. Kandeepan, and A. Anpalagan, Shenyang Ligong University. His research inter-
‘‘Efficient wireless power transfer in software-defined wireless sensor ests mainly include wireless sensor networks and
networks,’’ IEEE Sensors J., vol. 16, no. 20, pp. 7409–7420, Oct. 2016. embedded systems.
[23] R. Huang, X. Chu, J. Zhang, and Y.-H. Hu, ‘‘Energy-efficient monitoring
in software defined wireless sensor networks using reinforcement learning:
A prototype,’’ Int. J. Distrib. Sensor Netw., vol. 11, no. 10, Oct. 2015,
Art. no. 360428. doi: 10.1155/2015/360428.
[24] D.-J. Deng et al., ‘‘IEEE 802.11ax: Highly efficient WLANs for intelli-
gent information infrastructure,’’ IEEE Commun. Mag., vol. 55, no. 12,
pp. 52–59, Dec. 2017.
[25] T. Luo, H.-P. Tan, and T. Q. S. Quek, ‘‘Sensor OpenFlow: Enabling
software-defined wireless sensor networks,’’ IEEE Commun. Lett., vol. 16,
no. 11, pp. 1896–1899, Nov. 2012.
[26] A. Mahmud and R. Rahmani, ‘‘Exploitation of OpenFlow in wireless sen-
sor networks,’’ in Proc. Int. Conf. Comput. Sci. Netw. Technol., Dec. 2012,
pp. 594–600.
[27] T. Miyazaki et al., ‘‘A software defined wireless sensor network,’’ in Proc. HAI ZHAO received the Ph.D. degree in com-
Int. Conf. Comput. Netw. Commun., Feb. 2014, pp. 847–852. puter application technology from Northeastern
[28] L. Galluccio, S. Milardo, G. Morabito, and S. Palazzo, ‘‘SDN-WISE: University, China, in 1985. He is currently a Pro-
Design, prototyping and experimentation of a stateful SDN solution fessor and a Doctoral Supervisor with the School
for wireless sensor networks,’’ in Proc. IEEE Conf. Comput. Commun.,
of Computer Science and Engineering, Northeast-
Apr./May 2015, pp. 513–521.
ern University. He established the International
[29] M. Feng, S. Mao, and T. Jiang, ‘‘Enhancing the performance of futurewire-
less networks with software-defined networking,’’ Frontiers Inf. Technol.
Association for Internet Data Analysis (CAIDA),
Electron. Eng., vol. 17, no. 7, pp. 606–619, Jul. 2016. China’s First Node, and maintained long-term
[30] D. D. Paolo et al., ‘‘Exploiting state information to support QoS in research and teaching cooperation with the Uni-
software-defined WSNs,’’ in Proc. Ad Hoc Netw. Workshop, Jun. 2016, versity of Waterloo, Canada, UTD, USA, and UTA
pp. 1–7. universities. He has applied for ten national patents and software copyrights,
[31] W. Junfeng, M. Yiming, Z. Ping, M. S. Hossain, and S. M. M. Rahman, and published over 200 papers in related international conferences and
‘‘A software defined network routing in wireless multihop network,’’ journals. His current research interests include embedded systems, sensor
J. Netw. Comput. Appl., vol. 85, pp. 76-83, May 2017. networks, and pervasive computing.

VOLUME 7, 2019 61081

 X. Tan et al.: QSDN-WISE: A New QoS-Based Routing Protocol for Software-Defined Wireless Sensor Networks

GUANGJIE HAN (S’03–M’05–SM’18) received WENBO ZHANG received the Ph.D. degree in
the Ph.D. degree from Northeastern University, computer science and technology from Northeast-
Shenyang, China, in 2004. From 2004 to 2005, ern University, China, in 2006. He is currently
he was a Product Manager with ZTE Company. a Professor with the School of Information Sci-
From 2005 to 2006, he was the Key Account ence and Engineering, Shenyang Ligong Univer-
Manager with Huawei Company. In 2008, he fin- sity, China. He has published over 100 papers
ished his work as a Postdoctoral Researcher with in related international conferences and journals.
the Department of Computer Science, Chonnam His current research interests are ad hoc networks,
National University, Gwangju, South Korea. From sensor networks, satellite networks, and embedded
2010 to 2011, he was a Visiting Research Scholar systems. He received the ICINIS 2011 Best Paper
with Osaka University, Suita, Japan. In 2017, he was a Visiting Professor Awards and up to nine Science and Technology Awards, including the
with City University of Hong Kong, Hong Kong. He is currently a Professor National Science and Technology Progress Award, and the Youth Science
with the Department of Information and Communication System, Hohai Uni- and Technology Awards from the China Ordnance Society. He has served on
versity, Changzhou, China, and a Distinguished Professor with the Qingdao the editorial boards of up to ten journals, including the Chinese Journal of
University of Science and Technology, Qingdao, China. He is the author of Electronics and the Journal of Astronautics.
over 330 papers published in related international conference proceedings
and journals, including the IEEE COMST, IEEE TMC, IEEE TIE, IEEE TII,
IEEE TCC, IEEE TPDS, IEEE TVT, IEEE TETC, IEEE IOT JOURNAL, IEEE
TETCI, IEEE SYSTEMS JOURNAL, IEEE SENSORS JOURNAL, IEEE WIRELESS
COMMUNICATIONS, IEEE Communications Magazine, and IEEE NETWORK and
the holder of 125 patents. His H-index is 32 and i10-index is 90 in Google
Citation (Google Scholar). Total citation of his papers by other people is more
than 4426 times. His current research interests include the Internet of Things,
the industrial Internet, machine learning and artificial intelligence, mobile
computing, security, and privacy. He received the ComManTel 2014, Com-
ComAP 2014, Chinacom 2014, and Qshine 2016 Best Paper Awards. He has TENG ZHU received the B.S. degree in computer
served as the Co-chair for more than 50 international conferences/workshops science and technology from Shenyang Ligong
and as a Technical Program Committee Member of more than 150 confer- University, China, in 2016, where he is currently
ences. He has served on the editorial boards of up to 16 international journals, pursuing the master’s degree with the School of
including the IEEE JSAC, IEEE NETWORK, IEEE SYSTEMS JOURNAL, IEEE Information Science and Engineering. His current
ACCESS, IEEE/CCA JAS, and Telecommunication Systems. He has guest research interest includes software-defined net-
edited a number of special issues in IEEE journals and magazines, includ- work (SDN)-based network management technol-
ing the IEEE Communications Magazine, IEEE WIRELESS COMMUNICATIONS, ogy for wireless sensor networks.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, and Computer Networks. He
has served as a Reviewer for more than 60 journals.

61082 VOLUME 7, 2019