Received June 2, 2021, accepted June 26, 2021, date of publication July 16, 2021, date of current version

July 29, 2021.

Digital Object Identifier 10.1109/ACCESS.2021.3097942

Game Theory Based Congestion Control for

Routing in Wireless Sensor Networks
ZHI HU 1, XIAOWEI WANG1 , AND YUXIA BIE 2
1 Software College, Shenyang Normal University, Shenyang 110034, China
2 College of Electronic Information and Engineering, Shenyang Aerospace University, Shenyang 110136, China

Corresponding author: Xiaowei Wang ([email protected])

This work was supported in part by the National Natural Science Foundation of China under Grant 61901284, in part by the Basic Research
Project of Liaoning Provincial Committee of Education under Grant LJC201909, and in part by the Open Project of Key Laboratory of
Intelligent Computing in Medical Image, Ministry of Education.

ABSTRACT The Congestion in wireless sensor networks can lead to the loss and retransmissions of packets,
and excessive energy consumption. Since network congestion can deteriorate the throughput, the quality of
service (QoS) in data transmission, and network lifetime, congestion control is a critical problem. In this
paper, the data transmission process is firstly presented based on the data type. The cooperation game theory
is used to model the transfer behavior of different type data. Further, a cooperation game based congestion
control (CGCC) is proposed. The different type data cooperates with each other to use link capacity. And then
Ant colony Routing algorithm combined with CGCC (ARCGCC) is designed. According to the congestion
degree and QoS requirement of each type data, the routing path is built in the routing algorithm. When
congestion is detected, the congestion information is exchange. In the light of the data traffic and priority,
the nodes adjust the sending rate to the next hop node to mitigate the congestion. Experimental results show
that our mechanism and algorithm have better performance than some other mechanism and algorithms in
terms of the throughput, reliability, delay, and energy consumption.

INDEX TERMS Wireless senor networks, congestion control, game theory, routing algorithm, QoS.

I. INTRODUCTION in the current time interval, and these sensed data need to
Wireless sensor networks (WSNs) consist of a number of be transferred to the Sink as soon as possible. However,
the sensor devices that are equipped with the communication since there are a lot of sensor nodes in the monitored terrain,
module. These sensor nodes collect the data from the envi- it is available that the reliability of these data is not high,
ronment and transport them to the Sink node by the ad-hoc on the condition that the data is effective for the application.
network protocol [1], [2]. WSNs are seen as one of the most The non-periodic sensed data, such as the triggered event,
important elements in Internet of Things (IoT) [3]. WSNs is generated because the monitored value is larger or lower
have been deployed in a wide range of application domains, than the predetermined threshold value. The transmission of
such as habitat monitoring, disaster prevention, automation these data does not only need to have low latency but also high
control and infrastructural security, medical health. There is reliability. Similarly, the data, which is generated by the user
a large number of applications in WSNs connected to IoT. query command, has the same properties. In addition, some
With the requirement increasingly developing from the media data, such as captured picture, needs to be transferred
diversified domains, multiple application tasks need to be to the Sink at higher reliability. Thus, the current WSNs is
concurrently executed for the different function at the same a shared network, which the multiple application tasks are
time in the same WSNs. Such as the soil monitor system concurrently run on it [4]–[6]. And these different applica-
in modern agriculture, some physical data of environment tions generate the different type of data traffic, which have
is collected periodically and then transferred to the Sink. the different QoS transmission requirement.
The delay limitation is the main requirement for this kind of These data traffic from the sensor nodes to the Sink is
data. The temperature data reflects on the fact environment mainly a many-to-one communication in WSNs. Generally,
each sensor node selects its parent node based on the rout-
The associate editor coordinating the review of this manuscript and ing algorithm. The network topology is a tree. The data
approving it for publication was Chan Hwang See . of all nodes are routed and forwarded to the Sink on the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
103862 VOLUME 9, 2021
Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

topology tree, and the network congestion is easy to appear algorithm can avoid the packet retransmission so as to
when the data load exceeds the available bandwidth capa- decrease the energy waste. The network lifetime is extended.
bility. The congestion leads to lower throughput, unfairness The organizational structure of the remainder of the paper
of data transmission, higher packet loss rate, and longer is described as follows: The related works are discussed
delay. The energy of node wastes because of the loss packet in Section II. The data transmission process is analyzed
retransmissions. And the lifetime of entire network will be in Section III. Section IV presents the congestion control
shortened. Because the different data traffic wants to send mechanism in detail. Section V describes the routing algo-
packets as high as it can without considering the link capacity, rithm. The evaluation and simulation results are displayed in
the congestion occurs. However, if these different data traffic section VI. Section VII summarizes this paper.
cooperates with each other to use link capacity, the congestion
is removed and avoid. This problem can be formulated as a II. RELATED WORK
cooperative game where each type data traffic is modelled as In the past few years, the several mechanisms and algo-
a player in the game. rithms have been proposed to alleviate the congestion in
In this paper, the transmission process of data is firstly WSNs [7]–[10]. Generally speaking, the presence of con-
modeled by the view of node, and then the transmission gestion means that the load is temporarily greater than the
process is further modeled based on data type. And the data transmission capability of network resource. Thus, there are
traffic is classified and prioritized according to the delay two approaches to control the congestion: resource man-
and reliability of QoS parameters. These different types of agement and traffic control. In the first approach, the data
data traffic cooperate to use the bandwidth. The cooperation packets are forwarded to the Sink by the alternative paths that
game theory is used to calculate the bandwidth and rate in are not congested, when the congestion occurs. In wireless
order to relieve congestion and guarantee QoS requirement network, the congestion is weakened by the increasing the
in this transmission process. Cooperation game theory based transmission resource of networks, such as the active nodes,
congestion control (CGCC) mechanism is proposed. By the routing paths, bandwidth and larger transmission distance.
amount of data successfully transmitted with the shortest For example, the core of TARA [11] estimates the capacity
delay, the utility function is defined and the optimal solution of various topologies by the capacity analysis model, and
of rate is calculated in CGCC. And then the congestion then alleviates the congestion by using the topology-aware
degree is defined by the rate of input and output. In order resource adaptation strategy. With this method, the main chal-
to control the congestion and improve the performance in lenge stems from the fact that the resource control requires not
the entire network, Ant colony Routing algorithm combined only the local knowledge, but also knowledge about the entire
with CGCC (ARCGCC) is designed. The ARCGCC takes network resource. However, this is a hard task in wireless
the congestion degree of routing path and QoS requirement environments.
of data as the heuristic factor, such that the congestion is Unlike the approach of resource control, the approach of
avoided as much as possible. The Path-finding ant is cre- traffic control deals with the congestion by adjusting traffic
ated for each type data. The optimal routing with conges- rate at source nodes or intermediate nodes. When the resource
tion control is built after the several rounds of iteration in demands are near, or exceed, the capacity of network, the traf-
ARCGCC. If the congestion is detected, the nodes adjust fic control reduces the number of injected packets to allevi-
the sending and forwarding rate to the next hop node to ate the congestion. In [12], the end-to-end control method
mitigate the congestion, according to the data traffic and is applied. Since the end-to-end control method is slower
priority. reactive and depends heavily on the round-trip time from the
The main contributions of this paper are described as fol- source node to the Sink, this kind of method is used in few
lows: protocols and algorithms. The current main method is the
(1) The transmission model based on data type is set up. hop-by-hop control mechanism, which has faster response
On this basis, cooperation game theory based congestion and is a kind of distributed control mode. The PCCP [13]
control (CGCC) mechanism is proposed. The high through- is a priority-based hop-by-hop congestion control protocol
put and the fairness of the same priority data are achieved. for wireless sensor networks. The degree of congestion is
Under the condition of ensuring transmission requirements, computed by packet inter-arrival time and packet service
the high priority data gets the high bandwidth and rate. The times, and then imposes the hop-by-hop congestion control
rest bandwidth is allocated among the low priority data. depending on the measured congestion degree and the priority
(2) Ant colony routing algorithm with CGCC (ARCGCC) index. The different index is set to each node. The node
is designed. ARCGCC can provide the guarantee on the with higher priority index enjoys higher transmission band-
throughput, delay and reliability about QoS requirements. width. In HRTC [14], when the node detects the congestion,
By avoiding and mitigating the congestion, the throughput the back pressure message is sent out. The nodes in the path
can be improved on the Sink. The packet loss rate may be alleviate the congestion by decreasing the data throughput.
reduced, and the network performance is enhanced. However, game theory is recently introduced into the hop-
(3) Reducing the energy waste and prolonging the net- by-hop control method, which provides the optimal solution
work lifetime. The congestion control mechanism and routing for congestion control. GTCC [15], [16] is a game theory

VOLUME 9, 2021 103863

 Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

based congestion control protocol. By using packet flow rate,

the congestion is detected. The child nodes will be informed
about the presence of congestion by their parent node, and
then the child node uses game theory strategy to decide
whether it changes its parent. This control protocol changes
the routing topology path, which is hard to combined into the
routing algorithm. GTCCF [17] proposes the game theory
based congestion control framework. When the child nodes
send high data rate packets to the parent node in a selfish
FIGURE 1. Network topology.
way, the network congestion occurs. Thus, this congestion
problem is formulated as a non-cooperative game where each
selfish child node is modelled as a player. The rate of each
child node is maximized by non-cooperative game. But this
non-cooperative game does not consider the minimum rate
and QoS requirements. The rate solution of Nash equilibrium
is sometimes less than the minimum rate, and then has no way
to meet the QoS requirement.
Though the hop-by-hop congestion control is an effective
solution that mitigates the congestion in the network, the QoS
requirements and avoidance of congestion are seldom consid- FIGURE 2. Node traffic based transmission model.
ered in advance. Swarm intelligence Optimization algorithm
is tried to solve the combination optimization problem of
these two aspects. In [18], the solution is based on the con- transmission model, the new data type-based transmission
cept of bird flocking to guide packets through uncongested model is proposed.
nodes and avoid congested ones. The proposed mechanism
selects the least congested routes to deliver packets to the A. NETWORK TOPOLOGY
Sink. But this bird flocking optimization has frequent path The data traffic from the sensor nodes to the Sink is many-to-
change and high overhead, and has no way to find the optimal one multi-hop convergent in wireless sensor networks. Thus,
route. Another better optimization algorithm is Ant Colony the network topology is seen as the multi-hop collecting tree,
Optimization (ACO) [19], [20], which is a heuristic stochastic and the root of tree is the Sink, as is shown in Figure 1. Each
optimal algorithm with excellent performance. The positive node is the forwarding node except for the source node at the
feedback mechanism is adapted in the ACO. By continuous edge leaf node of tree. The forwarding nodes route data from
updating of pheromones, the most optimal path is found. its child node to the next hop node as the parent node.
The ACO is widely applied in the combinatorial problem, In this form of network topology tree, we use C(i) to denote
such as global multi-object optimization and complex routing the number of its child nodes. In the Figure 1, the Node
of network. The ACO has the characteristics of distributed 3 has 3 children nodes, C(Node 3) = 3. And the Node 9 has
computing mode and low computation complexity, which is 0 children nodes, C(Node 9) = 0, because it is the edge leaf
suitably applied in the solution to the combination problem of node.
congestion control and routing in wireless sensor networks.
The MS-ACO [21] is an algorithm that guarantees QoS B. NODE TRAFFIC BASED TRANSMISSION MODEL
requirements and controls congestion. The MS-ACO takes a Generally, the nodes have two classes of data traffic: source
very pessimistic view and then changes the path when the and transit. The source data traffic is generated by the local
congestion is detected at any time on the path. It wastes the node, and transit data traffic is from its child nodes. The data
energy and is mainly used in the next generation network. transmission process is depicted by the node traffic based
The ARCC [22] mitigates the congestion by adopting the transmission model in Figure 2.
above PCCP protocol. Firstly, the optimal path from the The rs denotes the rate of source data traffic that the node i
source to the Sink is found by ACO. Secondly, the generated generates. The rtr denotes the rate of transit data traffic from
ants from the Sink detect the congestion by the congestion the child node h of node i, by MAC Layer. The transit data
degree, and mitigate the congestion by using PCCP. This traffic and source data traffic are converged at the network
algorithm merely combines the ACO with PCCP, which does layer, and then these data traffic is transferred into the MAC
not consider the data classification problem. Layer. The rin is defined as the sum of source data traffic rate
rs and transmit data traffic rate rtr , which is rin = rtr + rs .
III. DATA TRANSMISSION The rf is the transmission rate of MAC Layer. The rf is influ-
By analyzing the process of data transmission, the trans- enced by the MAC protocol, electromagnetic environment,
mission model is setup, which is the base for controlling the number of active nodes in the adjacent domain. The rout
the congestion. Based on analyzing the node traffic-based denotes the practical transmission rate from node i to the next

103864 VOLUME 9, 2021

Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

hop node j:
rout = min{rin , rf } (1)
If rin < rf , that means the sum of source data traffic rate
rs and transmit data traffic rate rtr is lower than the rate rf
provided by the MAC Layer, because of rin = rtr + rs .
Therefore rout = rin . Otherwise, the transmission rate rf of
current wireless link of node i has no way to carry the total
input rate rin of node i. Then rout is close to rf .
When a sensor node has source data traffic and transit data
traffic, it is the forwarding node. On the other hand, when a
sensor node has only source data traffic, it is only the source
node. The transmitted data and source data are converged and
sent to the next hop node. Therefore, it is seen that the output FIGURE 3. Data type based transmission model.
traffic of node i is part of transmit traffic at the next hop
node j.
The above transmission model is analyzed by the send- Monitor Module. When these data meet the monitoring rate
ing nodes. This kind of transmission model reflects on the requirements, they are passed to Router Module and sent out
source node and forwarding node of the data traffic. How- to the next hop node.
ever, the rough degree of classification is high, which does When the congestion occurs, the rtr and rs of each type
not reflect on the characteristic of data. Especially, the QoS data are adjusted by the predetermined priority of data. The
requirement of data is not considered. The data should be different type data has the different QoS requirements for
classified by the application type and characteristic of QoS the data transmission. Because the Type-1 data has no high
requirement. requirement on the latency, the priority of Type-1 data can be
set to the lowest level, which is transmitted in the end. The
C. DATA TYPE BASED TRANSMISSION MODEL data of QoS-3 is normally the queried data or event report
For the different applications on shared WSNs in IoT, their data, which both the latency and reliability are necessary.
QoS requirement has three features: Latency, Reliability, and Thus, the priority of Type-3 data is set to the highest level,
Energy Cost [23]. In these features, the latency and relia- which is the first to be sent out. Though the data of Type-2
bility are main requirement for data transmission. Energy has the requirement on the delay, the redundancy of periodic
cost is considered in the network performance. The less Type-2 data is high. We set the priority of Type-2 data to the
energy consumes in data transmission, the longer network level in between. Their relationship is below.
lifetime. Therefore, according to the latency and reliability,
the sensed data can be classified into the multiple categories. P(Type-1) < P(Type-2) < P(Type-3) (2)
Take 3 common types as examples, which is below: Once the congestion is detected by the node, the rtr and rs
1) Type-1. The reliability is the main requirement for this of Type-1 data are decreased firstly, and then the rtr and rs of
kind of data, and the latency of data transmission is not Type-2 data are adjusted. At last, the rtr and rs of Type-3 data
required, or is rarely required for the applications, such as are modified.
captured monitoring picture. The sending rate can be reduced Based on the source rate rs , transit rate rtr , and output rate
to 0, when the congestion occurs. rout in the time interval T , the Congestion Degree (CDi ) of
2) Type-2. The latency is the main requirement for this kind node i is computed as follows:
of data, and the reliability of data transmission is not required,
Ns + Ntr
or is rarely required for the applications, such as periodic rs + rtr = (1 − ωa ) ∗ (rs + rtr ) + ωa ∗ (3)
sensed temperature data. The sending rate can be reduced to T
Nout
a lower level, when the congestion occurs. rout = (1 − ωs ) ∗ rout + ωs ∗ (4)
3) Type-3. The latency and reliability are essential for this T
rs + rtr
kind of data, such as the queried data and event report data. CDi = (5)
When the congestion occurs, the sending rate also needs to rout
be guaranteed. where the rs , rtr and rout are computed by the Exponentially
Based on this kind of classification, Classifier Module, Weighted Moving Average (EWMA) method; the Ns and Ntr
Monitor Module and Router Module are set in the network are the number of source data packets and forwarded data
layer. The transmission process of data is depicted by the data packet in the time interval T , respectively; and the Nout is the
type based transmission model in the Figure 3. number of data packets that are sent out in the time interval T ;
When the data enter the network layer, the data is firstly The ωa (0 < ωa < 1) and ωs (0 < ωs < 1) are set to 0.1. Thus,
classified in Classifier Module, and then the rate of each the congestion degree (CDi ) of node i is computed, which is
type data and their total rate are monitored and calculated in intended to reflect the congestion level. When the sum of rs

VOLUME 9, 2021 103865

 Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

and rtr is larger than rout , CDi ≥ 1, the congestion occurs. newly arrived data will be lost. Thus, the
Prate ri of each Type-i
When CDi < 1, there is no congestion. data needs to be controlled to ensure N i=1 ri < rout . At this
time, the newly arrived packet of each Type-i data is not
IV. GAME THEORY BASED CONGESTION CONTROL lost. The sending process of data is approximately simplified
The data traffic is firstly modelled, based on Cooperation as M/M/1 for the simplicity of calculation in the node. The
Game. And then the rate is computed and fairly allocated. average delay of transmitted data is below:
Cooperation Game based Congestion Control (CGCC) mech-
1
 XN
anism is proposed. 
 PN ri < rout
D = rout − i=1 ri Xi=1 (6)
A. GAME THEORY MODEL  N
 ∞ ri ≥ rout
By analyzing the data type based transmission model above, i=1
XN
it is seen that there are N types data traffic in the network, ri = rs + rtr (7)
i=1
which they compete to use the link bandwidth to the next hop
node. If there is no mandatory agreement mechanism among where the rate of Type-i data is denoted by ri . The rate is
these types of data traffic, each type data arbitrarily adjusts calculated according to the data type in the model. When the
the rate for maximizing its own interests. The network con- congestion occurs, the newly arrived data packet is directly
gestion will happen. The congestion will increase the packet lost. The delay of lost packet is not calculated. If there is no
loss rate, transmission delay and energy consumption in the congestion, the packet is mainly lost in wireless channel. The
network routing. The congestion will also reduce throughput packet loss rate of wireless channel is normally constant L.
and QoS performance of network routing. Therefore, in order According to the above analysis, the Characteristic function
to effectively control the congestion, the cooperation agree- U is defined below:
ment must be reached among these types of data traffic. XN(ri )αi · (1 − L)
Their rate needs to be reasonably allocated. For this problem, U =
cooperation game theory is a very suitable solution. i=1 D  
XN XN
αi
In the transmission process based on the view of data type, = (ri ) · (1 − L) · rout − ri (8)
i=1 i=1
these types of data traffic is modelled as Cooperation Game
CG = (N , S, U ), where:
where αi is balance factor. The utility value of Characteristic
1) The N are players in the game. Each type data traffic is
function is the amount of data per unit time, which is success-
modelled as a player in the game. The players set is presents
fully transmitted with the shortest delay.
by {Type-1, Type-2, · · · , Type-i, · · · , Type-N }.
Lemma 1: In the process of data transmission, the CG(N ,
2) The S denotes the Strategy of players, which is rate of
S, U ) has the Core 0, and the 0 is not empty set.
each player. The strategy space of rate ri for each player Type-
Proof: For the N players cooperative game, all of players
i is [li , rout ], which li is denoted by the minimum rate of Type-
participate in the same game coalition. Thus, this is grand
i data. In the transmission process based on game theory,
coalition. The allocation and negotiation of bandwidth and
each Type-i data needs to have the minimum rate that meets
rate is executed and agreed on the parent node. The rtr and
the QoS requirement of routing. When no other type data is
rs of each Type-i data are computed by the parent node.
transmitted or their rate is 0, Type-i data can monopolize this
Each Type-i data on the parent node has the right to veto the
communication link and then the rate ri is rout . The strategy
allocation solution of bandwidth and rate. The necessary and
space of all players is [l1 , rout ] × [l2 , rout ] × · · · × [li , rout ]
sufficient condition for Core non-empty is that there is the
× · · · × [lN , rout ].
veto player. Thus, The Core 0 existsand is not empty. There
3) The U is the Characteristic function. It is found that the
is always a set of strategic solutions r10 , r20 , · · · , ri0 , · · · , rN0
transmission delay and sending rate of the packet are mutually
that allow the system to get the optimized allocation of band-
contradictory in the network. Each Type-i data always wants
width and rate.
to be sent out to the receiving node at the maximum through-
After analyzing the above game model and proving the
put rate, with the minimum delay and the packet loss rate.
existence of Core, the solution of game strategy needs to
However, the increase in the sending rate of each Type-i data
be computed. For the N players, the value of Characteristic
will inevitably lead to the increase in the congestion degree
function is maximum, which the traffic rate strategy meets
of node. As a result, the transmission delay is increased. The
the following conditions:
Characteristic function needs to find the max value of data
packets successfully transmitted with the shortest delay. max U (9)
Each Type-i data is sent or forwarded to the next hop
node by MAC Layer and communication link. This sending s.t. ri ≥ li , rout ≥ li ≥ 0 (10)
XN
process is similar to the queen theory system [24], [25]. ri < rout (11)
i=1
Assuming that theP Type-i data traffic is a Poission flow with
the rate ri . When N i=1 ri ≥ rout , which the total arrival rate This is the optimization problem. The Lagrange multipli-
is larger than the link bandwidth, the congestion occurs. The ers λi and µi are introduced into the solution process. The

103866 VOLUME 9, 2021

Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

Lagrangian function F(ri , λi , µi ) is defined below: the special location, they can also be assigned with the more
XN  XN  weight. The data weight of node is designed to guarantee that:
αi
F (ri , λi , µi ) = (ri ) · (1 − L) · rout − ri (1) the node with higher data weight gets more bandwidth and
i=1 i=1
XN  XN  rate to transfer data about this type data; (2) the nodes with
+ λi (ri − li )+µi rout − ri the same weight get equal rate or bandwidth. On the node,
i=1 i=1
the weight of each Type-i data provides flexibility in realizing
(12) proportional fairness.
According to the Kuhn-Tucker conditions, there are For the Type-i data, the source data weight of node m is
Type−i
following: denoted by SWm , which represents the relative weight of
Type−i
∂F XN local source data at node m. SWm is independent of its
= αi ·riαi −1 (1−L)·rout − αi · riαi −1 (1−L) · rj child nodes. TWn
Type−i
is defined as the data weight of node
∂ri j=1,j6=i
n that is the member of C(m). The child nodes set of node m
− (1 − L) · (αi + 1) · riαi + λi − µi = 0 (13) Type−i
is denoted by C(m). TWn means the relative weight of
λi (ri − li ) = 0 (14) transit data routed through node n, which equals the sum of
 XN 
µi rout − ri = 0 (15) global data weight of all its child node and itself. The weight
i=1 ratio of node m and its child node n are calculated as follows:
λi , µi ≥ 0 (16) Type−i
SWm
PN PRType−i = P (20)
Since there is i=1 ri < rout , µi = 0. The link bandwidth m Type−i Type−i
+ SWm
n∈C(m) TWn
is sufficient for the allocation of all Type-i data rate ri . In order Type−i
to ensure QoS requirement of these types data, the ri is large TWn
PRType−i
nm = (21)
to li . There is λi = 0. Thus, the rate of each Type-i data is Type−i Type−i
P
n∈C(m) TWn + SWm
calculated below:
rout · αi Based on this kind of ratio relationship above, the source
ri = , i = 1, 2, · · · , N (17) Type−i
rate rm
Type−i
of node m and transmit rate rnm from its child
1+ N j=1 αj
P
node n can be allocated as follows:
When N
P
i=1 ri ≥ rout , which is the congestion degree of rmType−i = ri × PRType−i
m (22)
current node m is large to 1, CDm ≥ 1. Since the Type- Type−i
rnm = ri × PRType−i
nm (23)
1 data has the lowest priority, P the r1 is firstly decreased to
the lowest rate level l1 . If N i=2 ri is still larger than rout , In this way, high weight nodes get high rate and bandwidth,
the r2 is decreased to the lowest rate levelPl2 . Until when rk and the nodes with same weight get the same rate allocation.
is decreased to the lowest rate level lk , N i=k+1 ri < rout , The rate and bandwidth of nodes is fairly allocated according
the remaining bandwidth is allocated by the higher priority to the weight. The fairness of nodes is guaranteed.
type data. The rate of higher Type-i data is computed below:
  C. CONGESTION CONTROL MECHANISM
rout − kj=1 lj · αi
P
The node m monitors the wireless link utilization status, and
ri = , i = k + 1, · · · , N (18)
1+ N j=k+1 αj
sends an Inquiry Message to the children node set C(m) to
P
 Xk−1 XN  query each type data rate information and relative weight of
rk = lk + ϕ · rout − lj − ri (19) all children nodes. If the node n is a child node of node m, that
j=1 i=k+1
is n ∈ C(m), the node n returns a Report Message to send
where ϕ is reduction factor. The ϕ is set to 0.9, which means its weight and rate information to the parent node m. After
assigning 90% of the last remaining bandwidth to the Type-k the node m receives all of weight and rate information from
data. its child nodes, it will calculate the allocated sending rate
Through the above calculations, it is realized to allocate as for itself and its child nodes according to the above analysis
much bandwidth as possible to high priority data, under the process. And then the node m sends the rate allocation result
premise of guaranteeing the lowest QoS transmission rate of to its child nodes using the Notification Message. The data
each type data packet and routing packet can piggyback these messages. The
message exchange process is shown in Figure 4. The relative
B. RATE ALLOCATION AMONG THE NODES rate allocation and adjustment is executed on the parent node
By the game analysis above, the rate of each type data is and its child nodes.
calculated. And then the rate of Type-i data is fairly allocated From the perspective of feedback control, the rate alloca-
for the current node and its child nodes. This kind of fair tion and adjustment plan of the parent node is determined by
data allocation depends on the weight of each Type-i data the weight and rate information of itself and its child nodes.
on the nodes. Generally, each node has one unit weight for If the transmission capacity of parent node m can carry the
each Type-i data. The bandwidth or rate is fairly allocated input traffic of itself and all child nodes, which is CDm < 1,
among the nodes. However, if some nodes are deployed at the rate allocation is directly computed. Otherwise, the parent

VOLUME 9, 2021 103867

 Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

Algorithm 2 Rate Adjustment on the Child Node

1: Wait to receive the message from the parent node.
2: if Receive Notification Message
Type−i
3: The transmission rate rnm is set.
4: Goto step 1.
5: end if
6: if Receive Inquiry Message
7: Compute the relative weight information.
8: Send back to the parent node by Report Message.
9: Goto step 1.
FIGURE 4. Congestion control process. 10: end if

Algorithm 1 Rate Adjustment on the Parent Node

1: Monitor the link status, and compute the CDm within design of network routing. In the design of routing, two
the period T . aspects need to be considered. The first aspect implies to
2: if CDm < 1 check each Type-i data QoS value of the nodes and path. Only
3: Compute the rate ri according to equation (17). the nodes and path, which have low latency and packet loss
4: Goto step 11. ratio, are placed into the candidate set of the route decision.
5: end if The second aspect includes the congestion metric. In other
6: if CDm > 1 words, the routing with congestion control is built. The rout-
7: Compute the rate ri according to equation (18). ing decision is based on the combination value of these two
8: Send the Inquiry Message to its child nodes. metrics.
9: After receiving the Report Message from all The congestion control can be applied in many rout-
Type−i Type−i ing protocol and algorithm. And Ant Colony Algorithm is
child nodes, compute the SWm , T Wn ,
Type−i Type−i widely applied in the routing design. So, taking the Ant
PRm , and PRnm according to equation (20)
Colony based routing protocol as example, we add Coop-
and (21).
eration Game based Congestion Control (CGCC) mecha-
10: end if
Type−i Type−i nism into it, which is Ant Colony Routing with CGCC
11: Compute the rate rm and rnm via equation
(ARCGCC). By computing the average value of congestion
(22) and (23).
Type−i degree that the node gets, the congestion degree of path
12: Send the allocated rate rnm to its child nodes by
PC(P) is defined below, where the length of P is presented
using the Notification Message.
by Len(P).
13: Goto step 1.
X
PC(P) = CDm /Len(P) (24)
m∈P

node m needs to query the weight and rate information of its Since both the congestion information and QoS require-
child nodes, and recalculate the current rate of itself and its ment of each Type-i data are considered, the expected heuris-
child nodes, in order to control the congestion status. And tic value ηType−i that combines the congestion degree and
then the node m sends the result to the child nodes. The QoS requirement is designed as follow:
operating steps on the parent node m are given in Algorithm 1: ηType−i = γq · QnP
Type−i
+ γc · (1 − PC(P)) (25)
The rate and traffic of the child node are set by the received
control information from the parent node m. If receiving Type−i D − DnP L − LnP
QnP = αd · + αl · + αe
the Notification Message from the parent node m, the child D L
nodes modify the corresponding sending rate of each Type- P
Ere (vk )
i data. If receiving the Inquiry Message, the current weight · P k∈nP (26)
information of child node is calculated and returned to the k∈nP Einit (vk )
parent node. The operating steps on the child node n are where the γq and γc are the weight about the QoS value
written in Algorithm 2: and the congestion degree, where γq + γc = 1 and γq =
After that, the child node n allocates the rate of its child γc = 0.5. The QoS value of path P through the next hop
Type−i
nodes in the same way. Through this kind of step by step back node n is presented by QnP above. The αd , αl and αe
pressure mode, the congestion is controlled and the rate is also are the weight of the latency, reliability and energy, where
fairly allocated. αd + αl + αe = 1. The D is the maximum transmission
delay, and the L is the maximum packet loss rate. The DnP
V. ROUTING WITH CGCC and LnP denotes delay and packet loss rate of path P through
In order to control the congestion and improve the perfor- the next hop node n, respectively. The Ere (vk ) and Einit (vk ) are
mance in the entire network, CGCC is combined into the the residual energy and initial energy of node. According to

103868 VOLUME 9, 2021

Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

the parameters above, the important operations of ARCGCC Algorithm 3 Ant Colony Routing Algorithm With CGCC
is designed below: Initialization:
1: Set initial pheromone of edge to the constant value
A. PATH FINDING Phe .
When a Path-Finding Ant starts routing, it moves from the 2: Set the max iterations Ncmax = maxValue.
node m to node n, with the probability below: 3: Set the current iteration Nc = 0.
  α  β  4: Set the route searching timer Tr = TimerValue.
Type−i
 arg max

 τ Type−i
mn · η mn , q ≤ q0 5: Starts CGCC scheme: Algorithm 1 and Algorithm 2.

 n6∈Tabu Core Procedure:
  α  β
Type−i Type−i
τmn · ηmn (27) 6: while Nc 6 Ncmax do
q > q 7: Create the Path-Finding Ants of each Type-i data.

0
Type−i α Type−i β

    

l6∈Tabu τml · ηml ,
P

8: for each Ant do
9: if Sink is NOT in the neighbor node set do
where the selection parameter is q0 (0 ≤ q0 ≤ 1), the param- 10: Choose the next node n via equation (27).
eter α and β are pheromone and expected heuristic factor 11: Add the current node m to the Tabu table.
whose values determine the path that is favored. The Path- 12: Update pheromone of m → n edge equation (28).
Finding Ant of Type-i data retains pheromone of the edge 13: else
Type−i
from the node m to the node n is indicated by τmn . 14: Move to Sink.
15: Compute the heuristic value of current Ant.
B. LOCAL UPDATE 16: end if
When the Ant goes through the neighboring node m and node 17: end for
n, the pheromone of the edge between them is updated by the 18: Find the path with maximum heuristic value.
equation below: 19: Update the pheromone of path via equation (29).
20: Update the routing table of Type-i data.
τmn
Type−i
= (1 − ρL ) · τmn
Type−i
+ ρL · 1τmn
Type−i
(28) 21: Nc = Nc + 1.
where the ρL (0 < ρL < 1) is the parameter of local updating, 22: endPwhile
Type−i
1τmn is set to the initial value. 23: if N i=1 li ≥ rout do
24: Find out lowest priority Type-i, and li 6 = 0
C. GLOBAL UPDATE 25: Set Nc = 0, Goto step 6.
After all ants complete the one iteration of routing path 26: end if
finding, for the path of most heuristic value, the pheromone 27: if Tr expired do
of edges in this path is computed below: 28: Reset Tr .
29: Set Nc = 0, Goto step 6.
τmax
τmn
Type−i
= (1 − ρG ) · τmn
Type−i
+ ρG · (29) 30: end if
L(P)
where the ρG (0 < ρG < 1) is the parameter of global
updating. The τmax is the maximum pheromone value, and
L(P) is the length of the path with most heuristic value. After the Nc rounds of iteration, the optimal routing path is
The Cooperation Game Based Congestion Control (CGCC) found out. Suppose the number of data types is N , and the
is combined into Ant colony Routing (AR) algorithm, Ant number of Path-finding Ants is K . The time complexity on
colony Routing with CGCC (ARCGCC) is designed. The the nodes and Sink is O(Nc · N · K ).
steps of ARCGCC are given in Algorithm 3. Lemma 2: The rate allocation is the Pareto optimal solution
Firstly, the related parameters are initialized and Coopera- in ARCGCC.
tion Game based Congestion Control mechanism is started. Proof: Each Type-i data traffic firstly pursues the indi-
Secondly, the routing path is searched and established, vidual rationality. In ARCGCC, the Type-i data firstly finds
according to the routing algorithm. Thirdly, the rate of node the optimal routing path, that is ri ≥ li and the utility value
is adjusted when the congestion occurs. If the current band- of path is the largest.
width has no way to meet the requirements of all types data, And then when the congestion occurs, all of type data traf-
the lowest priority type data re-initiates the routing. fic need to achieve the group rationality. On the congestion
The ARCGCC algorithm uses distributed computing mode node, though the rate of each type data is adjusted to alleviate
to search the routing path without the congestion, and fur- the congestion, the sum of each type dataP rate is greater
PNthan
ther optimize the routing path. The nodes firstly initiate the the sum of their lowest rate, that is N r
i=1 i ≥ i=1 li .
routing searching by creating the Path-finding Ants of each If the lowest priority data cannot get rate allocation, it is
Type-i data. And then after the Path-finding Ants reach the necessary to re-initiate routing. According to ARCGCC, the
Sink, the Ant with the largest heuristic value is selected for high priority data gets more bandwidth and rate, and the rate
each Type-i data, and then its routing path is updated again. of low priority data is also guaranteed. Under the condition

VOLUME 9, 2021 103869

 Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

TABLE 1. Simulation parameters.

of relieving the congestion, the utility value generated by

{r1 , r2 , · · · , ri , · · · , rN } is the largest on congestion node.
Thus, the rate allocation {r1 , r2 , · · · , ri , · · · , rN } is Pareto FIGURE 5. Normalized throughput in network.

optimal solution.

VI. SIMULATION EXPERIMENT AND ANALYSIS

Based on TOSSIM network simulation platform, the con-
gestion control mechanism and routing algorithm is imple-
mented by using TinyOS and NesC, and the network topology
is generated by using Java tool in TOSSIM. The length of
data packet is 100 bytes. The communication radius of node is
about 40m. The main parameters of simulation are presented
in the Table 1.

A. SCENARIO 1
The network topology is set as Figure 1. In the multi-to-one
communication mode, each node generates different Type-i
data and sends them to Sink node. Three types of data traffic
are deployed on the network. The priority levels of Type-
1 to Type-3 data increase sequentially. The initial rate of FIGURE 6. Normalized throughput of each Type-i data.
each Type-i data is set to 3 packets/s, and then is reduced
to the proper rate according to network congestion status.
The network topology is relatively simple, and CGCC is used in the transmission process. Although PCCP increases the
to solve the congestion problem. CGCC is compared with throughput by adjusting the rate, the equal throughput of each
CCF and PCCP. CCF is the basic congestion control protocol, Type-i data is not changed. This is because the priority based
which is the benchmark for comparison. on node is used in PCCP. However, the priority based on the
Figure 5 shows the change of normalized throughput in data type is utilized in CGCC. CGCC can not only provide
network, when the node 1 enters the hibernation during the the high throughput, but also provide the high throughput
period of [21 sec, 40 sec]. The normalized throughput means for the high priority Type-i data. Thus, both Type-2 data
the ratio of sent data packets from the nodes and received and Type-3 data get higher throughput when the bandwidth
data packets on the Sink. The remaining bandwidth cannot allows.
be allocated effectively and there is lower throughput in CCF, Figure 7 shows the fairness of each Type-i data. The
when the node 1 is not active. This is because the work- node 7 is only activated within the time interval of [21 sec,
conservation algorithm is used in it. However, the hop-by-hop 40 sec], while the remaining 9 nodes are always active. The
back pressure rate control algorithm is used in PCCP and congestion occurs by increasing the number of nodes in the
CGCC. And the remaining bandwidth and rate is reallocated network. At the beginning, the priority sum of Type-i data
in them. The higher bandwidth utilization is obtained. The is 9, and the average throughput of each node accounts for
throughput is also increased. about 1/9. And then the node 7 is activated and added into
The throughput relationship of each Type-i data is showed the network topology. The average throughput of each node
in the Figure 6. Since each node has the same priority in CCF drops to 1/10 for Type-3 and Type-2 data. For Type-1 data,
and PCCP, the throughput of each Type-i data is basically the average throughput of each node is slightly lower than
equal. This is because they do not differentiate the data type 0.06. This is because the rate of high priority type data is

103870 VOLUME 9, 2021

Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

FIGURE 9. Throughput in routing.

FIGURE 7. Fairness.

node is deployed at the edge of the network in simulation

terrain.
The 40 nodes are randomly selected as the data source
node, and then their sequence ID is increased by the distance
to the Sink. Each source node sends data at the rate of
one packet a second for each Type-i data. The simulation
time is 60 seconds. By increasing the number of source
node, the congestion occurs in this simulation testing. The
30 sensor nodes are activated between [21 sec, 40 sec] and
enters the network topology. The traffic of each type data
is initially equal. The algorithms used to compare the per-
formance of ARCGCC are the MMSPEED and ARCC [22].
The MMSPEED is normal QoS routing algorithm, which is
base for testing the routing algorithm with congestion control
above, when the congestion occurs. In the algorithm design,
the parameter α and β are 1 and 3 respectively in ARCGCC.
FIGURE 8. Rate change. The object function is the heuristic value. And the iteration
times Nc are 35 in the algorithm.
The relationship of throughput about each Type-i data is
allocated preferentially in CGCC. The remaining bandwidth presented in the Figure 9. The throughput of Type-2 data is
and rate are allocated for the low priority type data, which lower than the Type-1 and Type-3, since MMSPEED calcu-
the rate of low priority type data is greater than or equal to 0. lates routing path according to the transmission reliability.
Thus, the average throughput of each node drops more. The throughput of each Type-i data is basically equal in the
Figure 8 shows the rate change on the node. The control ARCC, since the priority of node is set to same. For the
congestion time interval in CGCC is set to10 seconds. When ARCGCC, the throughput of Type-3 data is highest, since
the rate of high priority Type-3 data is increased, the rate of it has the highest priority level. The Type-3 data has the
low priority Type-1 data is decreased because of congestion maximum bandwidth and rate. The Type-1 data has the lowest
control mechanism. The rate of each type data is continuously priority level, which its rate is adjusted to the smaller value,
adjusted over time, under the condition of bandwidth permit- so its throughput is the lowest.
ting. The high priority type data gets the high rate, and the Figure 10 shows the transmission reliability, which is
low priority type data gets the low rate. reflected by the Packet Loss Ratio. The congestion occurs
during the period of [21 sec, 40 sec], when the number of
B. SCENARIO 2 source nodes is increased. In Figure 10(a), the packet loss
The network consists of 100 nodes. Terrain dimension is rate of Type-1 data increases as the distance between the
100m × 100m. The network nodes are deployed by using node and the Sink increases. The packet loss ratio of ARCC
uniform random distribution. The distance between any two and ARCGCC is significantly lower than that of MMSPEED.
nodes is more than 7 meters. The length of packet is 100 bytes. This is because the network-wide routing optimization is
The communication radius of node is about 40m. The Sink used in them. In ARCC, PCCP reduces the congestion and

VOLUME 9, 2021 103871

 Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

FIGURE 10. Packet Loss Ratio when increasing the source nodes. FIGURE 11. Delay when increasing the source nodes.

improves the transmission reliability. In ARCGCC, the data from the Sink, have the more hops number and then the
traffic is further classified. When the congestion occurs, greater transmission delay. ARCC uses the route optimization
the Type-1 data is transmitted as little as possible. After to reduce the latency. On the base of routing optimization,
the congestion is alleviated, the Type-1 data is restored to ARCGCC allocates more bandwidth and rate for Type-2 data
the corresponding sending rate. By this way, the packet loss by setting higher priority. Thus, ARCGCC has less delay than
ratio is reduced. Figure 10(b) shows the packet loss ratio MMSPEED and ARCC. Figure 11(b) shows the latency of
of Type-3 data. Under the condition that the delay is less Type-3 data transmission, under the condition that the packet
than 300ms, the packet loss ratio of ARCC and ARCGCC is loss ratio is less than 25%. Type-3 data has the highest prior-
lower. ARCC adjusts the sending rate of nodes to relieve the ity, and gets the maximum bandwidth and rate in ARCGCC.
network congestion and improve the transmission reliability. Thus, the transmission delay of Type-3 data in ARCGCC is
While adjusting the sending rate, ARCGCC allocates the also smallest.
more transmission bandwidth and rate to the high priority After the simulation testing is over, the residual energy of
Type-3 data. Thus, ARCGCC further reduces packet loss nodes is shown in the Figure 12. For MMSPEED, the aver-
ratio, and has better performance than them. age value of residual energy is 0.59, and the ARCC and
The data transmission delay is displayed in Figure 11. The ARCGCC are respectively 0.63 and 0.67. MMSPEED has no
delay is changed during the interval [21 sec, 40 sec], when routing optimization about energy, which leads to excessive
the congestion occurs by increasing the number of source energy consumption on some nodes. ARCGCC and ARCC
nodes in the network. Figure 11(a) shows the transmission are different from MMSPEED. Because the packet loss rate
delay of Type-2 data. The nodes, which is farther away is reduced in ARCGCC, the retransmission times of lost

103872 VOLUME 9, 2021

Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

[3] S. A. Dehkordi, K. Farajzadeh, J. Rezazadeh, R. Farahbakhsh,

K. Sandrasegaran, and M. A. Dehkordi, ‘‘A survey on data aggregation
techniques in IoT sensor networks,’’ Wireless Netw., vol. 26, no. 2,
pp. 1243–1263, Feb. 2020.
[4] Z. Zhou, D. Zhao, G. Hancke, L. Shu, and Y. Sun, ‘‘Cache-aware query
optimization in multiapplication sharing wireless sensor networks,’’ IEEE
Trans. Syst., Man, Cybern. Syst., vol. 48, no. 3, pp. 401–417, Mar. 2018.
[5] F. Derakhshan and S. Yousefi, ‘‘A review on the application of multiagent
systems in wireless sensor networks,’’ Int. J. Distrib. Sensor Netw., vol. 15,
no. 5, pp. 1–19, Mar. 2020.
[6] A. Majeed and T. Zia, ‘‘Multi-layer network architecture for supporting
multiple applications in wireless sensor networks,’’ J. Wireless Mobile
Netw., Ubiquitous Comput., Dependable Appl., vol. 8, no. 3, pp. 26–56,
Sep. 2017.
[7] M. H. Homaei, F. Soleimani, S. Shamshirband, A. Mosavi, N. Nabipour,
and A. R. Varkonyi-Koczy, ‘‘An enhanced distributed congestion control
method for classical 6LowPAN protocols using fuzzy decision system,’’
IEEE Access, vol. 8, pp. 20628–20645, Jan. 2020.
[8] C. Lim, ‘‘A survey on congestion control for RPL-based wireless sensor
networks,’’ Sensors, vol. 19, no. 11, p. 2567, Jun. 2019.
[9] H. A. A. Al-Kashoash, H. Kharrufa, Y. Al-Nidawi, and A. H. Kemp,
FIGURE 12. Residual energy. ‘‘Congestion control in wireless sensor and 6LoWPAN networks: Toward
the Internet of Things,’’ Wireless Netw., vol. 25, no. 8, pp. 4493–4522,
2019.
[10] D. Pandey and V. Kushwaha, ‘‘An exploratory study of congestion control
packet are decreased and the less energy wastes. The average techniques in wireless sensor networks,’’ Comput. Commun., vol. 157,
residual energy of ARCGCC is higher than ARCC. Thus, its no. 1, pp. 257–283, May 2020.
network lifetime is longer. [11] J. Kang, Y. Zhang, and B. Nath, ‘‘TARA: Topology-aware resource adap-
tation to alleviate congestion in sensor networks,’’ IEEE Trans. Parallel
Distrib. Syst., vol. 18, no. 7, pp. 919–931, Jul. 2007.
VII. CONCLUSION [12] J. Paek and R. Govindan, ‘‘RCRT: Rate-controlled reliable transport proto-
In this paper, the process of data transmission is firstly ana- col for wireless sensor networks,’’ ACM Trans. Sensor Netw., vol. 7, no. 3,
pp. 20:1–20:45, Oct. 2010.
lyzed in wireless sensor networks. Based on the data classifi- [13] C. Wang, B. Li, K. Sohraby, M. Daneshmand, and Y. Hu, ‘‘Upstream
cation and their priority, the new transmission model of data congestion control in wireless sensor networks through cross-layer opti-
is designed. And a mechanism of congestion control based on mization,’’ IEEE J. Sel. Areas Commun., vol. 25, no. 4, pp. 786–795,
May 2007.
cooperation game is proposed. This congestion control mech- [14] C. Sergiou and V. Vassiliou, ‘‘HRTC: A hybrid algorithm for efficient
anism is combined with ant colony algorithm. The congestion congestion control in wireless sensor networks,’’ in Proc. 6th Int. Conf.
degree and QoS value are added into the heuristic value of New Technol., Mobility Secur. (NTMS), Mar. 2014, pp. 1–5.
[15] J.-P. Sheu, C.-X. Hsu, and C. Ma, ‘‘A game theory based congestion control
algorithm. In the process of routing finding and selecting, protocol for wireless personal area networks,’’ in Proc. IEEE 39th Annu.
the congestion is avoided as much as possible. The optimal Comput. Softw. Appl. Conf., vol. 2, Jul. 2015, pp. 659–664.
routing path is built. When the congestion is detected, the [16] C. Ma, J.-P. Sheu, and C.-X. Hsu, ‘‘A game theory based congestion con-
trol protocol for wireless personal area networks,’’ J. Sensors, vol. 2016,
sending rate is adjusted to alleviate the congestion. Compared pp. 1–13, Dec. 2016.
to other congestion control algorithms, our proposed conges- [17] H. Al-Kashoash, M. Hafeez, and A. Kemp, ‘‘Congestion control for 6LoW-
tion control mechanism and algorithm show better perfor- PAN networks: A game theoretic framework,’’ IEEE Internet Things J.,
vol. 4, no. 3, pp. 760–771, Jun. 2017.
mance in terms of the throughput, residual energy, reliability [18] O. Chughtai, N. Badruddin, A. Awang, and M. Rehan, ‘‘Congestion-aware
and latency. As a result, our algorithm provides a novel and and traffic load balancing scheme for routing in WSNs,’’ Telecommun.
efficient way to improve the performance of wireless sensor Syst., vol. 63, no. 4, pp. 481–504, Dec. 2016.
[19] X. Liu, ‘‘Routing protocols based on ant colony optimization in wireless
networks. sensor networks: A survey,’’ IEEE Access, vol. 5, pp. 26303–26317, 2017.
[20] F. Li, M. Liu, and G. Xu, ‘‘A quantum ant colony multi-objective routing
ACKNOWLEDGMENT algorithm in WSN and its application in a manufacturing environment,’’
Sensors, vol. 19, no. 15, p. 3334, Jul. 2019.
The authors would like to thank for the help and sup- [21] Y. Li, Z. Ma, and Z. Cao, ‘‘A mitigating stagnation-based ant colony
port of Shenyang Normal University, Shenyang Aerospace optimization routing algorithm,’’ in Proc. IEEE Int. Symp. Commun. Inf.
University, and Key Laboratory of Intelligent Computing Technol. (ISCIT), Jan. 2006, pp. 36–39.
[22] S. K. Dhurandher, S. Misra, H. Mittal, A. Agarwal, and I. Woungang,
in Medical Image, Ministry of Education, for research ‘‘Using ant-based agents for congestion control in ad-hoc wireless sensor
work. networks,’’ Cluster Comput., vol. 14, no. 1, pp. 41–53, Mar. 2011.
[23] M. Asif, S. Khan, R. Ahmad, M. Sohail, and D. Singh, ‘‘Quality of service
of routing protocols in wireless sensor networks: A review,’’ IEEE Access,
REFERENCES vol. 5, pp. 1846–1871, Jan. 2017.
[1] D. Lin, Q. Wang, W. Min, J. Xu, and Z. Zhang, ‘‘A survey on energy- [24] S. Hafidi, N. Gharbi, and L. Mokdad, ‘‘Queuing and service management
efficient strategies in static wireless sensor networks,’’ ACM Trans. Sensor for congestion control in wireless sensor networks using Markov chains,’’
Netw., vol. 17, no. 1, Jan. 2021, Art. no. 3. in Proc. IEEE Symp. Comput. Commun. (ISCC), Jul. 2017, pp. 176–181.
[2] N. Hirosawa, H. Iimori, K. Ishibashi, and G. T. F. D. Abreu, ‘‘Minimizing [25] S. Lall, A. S. Alfa, and B. T. Maharaj, ‘‘The role of queueing theory in
age of information in energy harvesting wireless sensor networks,’’ IEEE the design and analysis of wireless sensor networks: An insight,’’ in Proc.
Access, vol. 8, pp. 219934–219945, Nov. 2020. IEEE 14th Int. Conf. Ind. Informat. (INDIN), Jul. 2016, pp. 1191–1194.

VOLUME 9, 2021 103873

 Z. Hu et al.: Game Theory Based Congestion Control for Routing in WSNs

ZHI HU received the B.S. degree in computer YUXIA BIE received the B.S. and M.S. degrees
science and technology from Northeastern Univer- in communication engineering from Shenyang
sity, Shenyang, China, and the M.S. degree in com- Ligong University, Shenyang, China, and the
puter software and theory from Shenyang Ligong Ph.D. degree in control science and engineering
University, Shenyang, and the Ph.D. degree in from the Nanjing University of Science and Tech-
technology of computer application from North- nology, Nanjing, China, in 2014. From 2008 to
eastern University, in 2016. From 2006 to 2016, 2016, she worked with the Key Laboratory of
he was a Researcher with Neusoft Research, Communication and Network, Dalian University.
Shenyang. Since 2017, he has been a Teacher with Since 2017, she has been a Teacher with the Col-
the Software College, Shenyang Normal Univer- lege of Electronic and Information Engineering,
sity. He is the author of more than ten articles and holds five patents. His Shenyang Aerospace University. She presides a National Natural Science
research interests include wireless networks, network routing, and machine Foundation project of China, a Doctoral Startup Foundation of Liaoning
learning. He was a recipient of the Liaoning Prize for Science and Technol- Province, and a General Project of Liaoning Provincial Committee of Edu-
ogy, in 2012. cation. She is the author of more than 20 articles and holds six patents.
Her research interests include satellite networks, data multiplexing, machine
learning, and UAV communication.

XIAOWEI WANG received the M.S. degree in

technology of computer application from North-
eastern University, Shenyang, China. She is cur-
rently a Full Professor, the Vice-Dean, and a
Teacher with the Software College, Shenyang Nor-
mal University. She is the host of several educa-
tional research projects. She is the author of more
than 20 articles. Her research interests include the
Internet of Things, computer architecture, and arti-
ficial intelligence.

103874 VOLUME 9, 2021