Journal of Network and Computer Applications ] (]]]]) ]]]–]]]

Contents lists available at ScienceDirect

Journal of Network and Computer Applications

journal homepage: www.elsevier.com/locate/jnca

An agent-assisted QoS-based routing algorithm for wireless sensor networks

Min Liu n, Shijun Xu, Siyi Sun
Key Laboratory of Embedded System and Service Computing, and CIMS Research Center, School of Electronic and Information Engineering, Tongji University,
Shanghai 201804, China

a r t i c l e i n f o abstract

Article history: Existing routing algorithms are not effective in supporting the dynamic characteristics of wireless
Received 15 September 2010 sensor networks (WSNs) and cannot ensure sufficient quality of service in WSN applications. This paper
Received in revised form proposes a novel agent-assisted QoS-based routing algorithm for wireless sensor networks. In the
8 January 2011
proposed algorithm, the synthetic QoS of WSNs is chosen as the adaptive value of a Particle Swarm
Accepted 29 March 2011
Optimization algorithm to improve the overall performance of network. Intelligent software agents are
used to monitor changes in network topology, network communication flow, and each node’s routing
Keywords: state. These agents can then participate in network routing and network maintenance. Experiment
Wireless sensor network results show that the proposed algorithm can ensure better quality of service in wireless sensor
Routing algorithm
networks compared with traditional algorithms.
Synthetic QoS
Crown Copyright & 2011 Published by Elsevier Ltd. All rights reserved.
Agent

1. Introduction algorithm uses the individual collaboration in groups to complete

the optimal search of NP problem, and it can effectively solve the
A wireless sensor network (WSN) is a multi-hop and ad hoc complex optimizing tasks in large space, and have got great
wireless network with a group of low-priced embedded sensor development in the past several years. Considering the dynamic
nodes randomly distributed in a monitoring area. These nodes characteristics of routing in wireless sensor networks, we propose
collaboratively perceive, collect, and process information of the a novel agent-assisted QoS-based routing algorithm for wireless
perceived objects in the geographical areas over the network, and sensor networks in this paper. The synthetic QoS of WSNs is
send the above information to the observers (Akyildiz et al., chosen as the adaptive value of PSO algorithm to improve the
2002). Data collected in WSNs are transmitted among all nodes overall property of network. An agent model is applied to
through routing, and then converged into the sink node. There- apperceive the changes in network topology, the network com-
fore, the routing algorithm determines whether the information munication flow, and each node’s energy state. Multi-agent can
can be transmitted completely or not, and greatly affect the also participate in network routing and network maintenance.
performance of the network. Comparing with the traditional algorithm, the proposed algo-
Recently, research of WSNs’ routing algorithm is mainly rithm can effectively improve the QoS metrics of WSN.
focused on some routing metrics, such as the congestion situa- The rest of the paper is organized as follows. Section 2 provides
tion, network lifetime, energy consumption, etc. With the great a brief review of some closely related works. In Section 3, we will
development of hardware designing technologies, the quality of introduce the QoS parameters in WSN and the synthetic QoS. An
services of the network is getting more attention. The question to agent model for QoS based routing is described in Section 4. The
find a path meeting the QoS guarantees for WSNs can be PSO and multi-agent based algorithm for the above routing model
formulated as a Non-deterministic Polynomial (NP) problem. are detailed in Section 5. In Section 6, four groups of experiments
However, the existing routing algorithms weakly support the are provided to test the algorithm. At last, we make some
dynamic characteristics of network, and cannot satisfy the conclusions.
requirements of the wireless sensor networks’ applications for
routing and QoS.
Particle Swarm Optimization (PSO), presented by Kennedy and 2. Related work
Eberhartin in 1995 (Eberhart and Shi, 1995), is a group-searching
based self-adaptive evolutionary computation method. The PSO Routing techniques are divided into three categories based on
their underlying network structures: flat, hierarchical, and location-
based (Al-Karaki and Kamal, 2004). Furthermore, these protocols
n
Corresponding author. can be classified into multipath-based, query-based, negotiation-
E-mail address: [email protected] (M. Liu). based, QoS-based, and coherent-based. QoS-based routing protocols

1084-8045/$ - see front matter Crown Copyright & 2011 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.jnca.2011.03.031

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
2 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]]

allow sensor nodes to balance energy consumption and certain algorithm, these agents can then participate in network routing and
predetermined QoS metrics before they deliver data to the sink network maintenance. Thus, algorithm performance can be obviously
node (Houngbadji and Pierre, 2010). For example, the end-to-end improved in delay, packet loss, and the synthetic QoS, respectively,
delay remains within an acceptable range, the link bandwidth with little energy consumption.
pertains to the tolerable compression ratio, the jitter is satisfactory,
and there is a low packet loss rate (Akyildiz et al., 2007).
Providing QoS guarantees in wireless sensor networks consists 3. A synthetic QoS based routing model for WSNs
of a very challenging problem, but several approaches have been
proposed in the literature for QoS support. Sequential Assignment 3.1. Quality of service (QoS) parameters in WSNs
Routing (SAR) (Sohrabi et al., 2000) for wireless sensor networks
builds multiple paths from a source node to the sink node. Path QoS based routing generally means that the QoS parameter
selection considers both QoS metrics (the flow delay require- constrains are added into the specific routing protocol. It is
ments and the source load balancing intentions) and energy available to provide a better way of data transfer performance
resources, to avoid nodes with low QoS and energy reserves. in this routing because QoS based routing protocol can search the
A Weighted Fair Queuing (WFQ) approach is used at every node to sufficient resources path meeting QoS under the conditions of the
provide the required share of bandwidth for both traffic classes. dynamic topology, limited resources, etc. In WSNs, a QoS based
SPEED was proposed by He et al. (2003) to provide a soft real-time routing protocol needs to perform the following features: (1) Path
end-to-end timeliness guarantee. Agrawal et al. (2006) utilizes determining—to determine the optimal routing path meeting the
the geographic location of sensor nodes as well; this protocol demand of the QoS from the numerous paths. (2) Resource
assigns an urgency factor to every packet depending on the reliability—to use network resource even without the ability of
remaining distance and the time left to deliver the packet. pre-reserving network resources. Usually, we can design some
Felemban et al. (2006) proposes the multipath multispeed proto- resource reservation mechanisms in high-level protocol to control
col (MMSPEED) for QoS guarantee of reliability and timeliness in the QoS, such as Resource Reservation Protocol (RSVP), etc.
wireless sensor networks. ReInForM (Deb et al., 2003) was (3) Path keeping—to prevent the sudden dropping of the QoS
proposed to address end-to-end reliability issues. routing performance.
The newly proposed solution to QoS routing in WSNs is based Different network applications will have their own QoS
on the swarm intelligence (SI) (Bonabeau et al., 1999; Kennedy demands. QoS in WSNs is mainly relative to some parameters
et al., 2001; Celik et al., 2010). SI was originally defined as ‘‘Any (Crawley et al., 1998; Constandinos et al., 2006) such as delay,
attempt to design algorithms or distributed problem-solving bandwidth, energy efficiency, robustness of network self-organi-
devices inspired by the collective behavior of social insects and zation, etc. Therefore, only one QoS parameter is not enough to
other animal societies’’ (Bonabeau et al., 1999). SI emphasizes the optimize routing path in a wireless sensor network, and a
bottom-up design of autonomous distributed systems that can synthetic QoS (multi-parameters) is necessary. The purpose of
show adaptive, robust, and scalable behaviors. Saleem et al. (2010) the synthetic QoS based routing is to get the optimal path,
have surveyed some existing SI algorithms proposed for wireless departing from the source node, ending in destination node, and
sensor networks, discussed the general principles of SI and its meeting all the QoS constraints, such as delay, jitter,
application to routing, and introduced a novel taxonomy for routing bandwidth, etc.
protocols in wireless sensor networks and use it to classify the Delay: also called latency, it is the transmission delay between
surveyed protocols. The SI framework encompasses other popular the two reference points. In the cable network, delay is mainly
frameworks such as Ant Colony Optimization (ACO) (Dorigo and caused by congestion. While in the wireless sensor networks,
Stützle, 2004; Liao et al., 2008), Bee Colony Optimization some other events may trigger data packet delay, including cohort
(Teodorovic, 2009), and Particle Swarm Optimization (PSO) delay, propagation delay, data flow competition, queue delay, etc.
(Bergh and Engelbrecht, 2006; Zhang and Xu, 2006). Zhang et al. Bandwidth: bandwidth is one of the most important metrics of
(Zhang 2004) developed three ant-based routing algorithms for QoS routing protocol; it is specifically the available bandwidth of
WSNs—sensor-driven and cost-aware ant routing (SC), flooded for- path from the source node to the destination node. However, the
ward ant routing (FF), and flooded piggybacked ant routing (FP). bandwidth will change with the movement of nodes. Many
These algorithms, with initial pheromone settings, have a good system routing protocols are trying to search the largest bandwidth, but
start-up, but the SC and FF algorithms are not quite effective in the least delay path should be chosen when several paths are
latency. Besides, the FP algorithm, while providing high success rates synchronously available.
of data delivery, consumes much higher energy. Cobo (2010) pre- Jitter: also called alterable delay; it is the time difference
sented an ant-based routing for Wireless Multimedia Sensor Networks between packets in a group of data flows sent by the same
(WMSN) using multiple QoS metrics; this approach built a hierarch- routing. Usually, the application impacted by delay will also be
ical structure on the network before choosing suitable paths to meet impacted by jitter.
various QoS requirements from different kinds of traffic, thus it could Packet Delivery Ratio and Loss: namely the highest rate of
maximized network utilization. Liang et al. (2006) proposed a packet loss in the network. Packet loss is usually caused by the
cluster-based algorithm, which used PSO to optimize clustering network congestion and the node’s mobility in wireless sensor
process under the condition of location and energy reserved about networks.
candidates and their neighbors. This algorithm may settle the existing Power Control and Conservation: in most cases, the sensor
algorithms’ problems of dying early due to ignoring the state of network nodes are powered by capacity-limited batteries, which
neighbors in the process of cluster-heads decision. restrict the lifetime of the nodes. When the energy of a node gets
In this paper, PSO is applied to optimize QoS routing, and a novel exhausted, the topology of network must be changed and the
agent-assisted QoS-based routing algorithm for wireless sensor net- routing needs to be re-established.
works is proposed. The synthetic QoS of WSNs is chosen as the
adaptive value of a Particle Swarm Optimization algorithm to 3.2. Synthetic QoS
improve the overall performance of network. Intelligent software
agents are used to monitor changes in network topology, network In this paper, we mainly consider the synthetic effect of QoS
communication flow, and each node’s routing state. In the proposed parameters, including delay, bandwidth, and packet loss.

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]] 3

The WSNs can be expressed as a weighted directed graph G In this way, fdelay will be very small, and it will be between
(V,E), where V is a set of sensor nodes with a wireless connection. 0 and 0.1.
If there are n þ1 nodes V, V¼{v0,v1,v2,v3,y,vn}. The communica- For another two metrics, bandwidth and packet loss, the same
tion radius of each node is ri, its communication area is Avi, and processing method can be applied. Thus, the value of the syn-
the edge e¼ (vi,vj)AE denotes the two-way wireless connection thetic QoS is the multiplicative function of the above three QoS
between two nodes (vi,vj). The path P(v1,vn) in G is an orderly metrics as follows:
compositing sequence of edges:
QoS ¼ fdelay  fbandwidth  fpacket_loss ð3Þ
Pðv1 ,vn Þ ¼ ððv1 ,v2 Þ,ðv2 ,v3 Þ    ðvi1 ,vi Þ    ðvn1 ,vn ÞÞ, vi A V, 2 rn r 9V9
The synthetic QoS value is between 0 and 1.
P(v1,vn) is a multi-hop path, the number of edges in P(v1,vn)
represents the hop distance between node v1 and node vn. Each
node in the path can be regarded as an independent router. The
4. An agent model for QoS based routing
first node of the path is the source node, and the final node is the
destination node, respectively called as vs and vd. Each node has
4.1. Routing table
its adjacent nodes. Each edge e¼(vi,vj)AE denotes that vi and vj
are the mutual adjacent nodes. Nvi ¼{vj9e¼ (vi,vj)AE, iaj} is a set
Routing table is a table stored in router, and plays the role of
of adjacent nodes of vi; it is found by the discovery mechanism of
path discovery in node routing (Deng Huang, 2009). Because
the adjacent nodes, which is also called HELLO information
every node in WSNs acts as a router, every node contains a
exchange. After exchanging HELLO message, the node adds its
routing table. Table 1 shows an example of the routing table for
QoS parameters to HELLO information.
the node A as depicted in Fig. 1. In Table 1, the first column
Given a path P(vs,vd), its synthetic QoS metrics can be defined
corresponds to the destination nodes and the first row corre-
by the delay, bandwidth, and packet loss, and can be reflected on
sponds to the adjacent nodes of the source node A. The values in
the node v and the link e. For every node vAV, the metrics are
Table 1 represent the synthetic QoS metrics from the source node
delay function—Delay(v), bandwidth function—Bandwidth(v),
A to the destination nodes through its adjacent nodes, such as
packet loss function—Packet_loss(v), and energy function—
0.84 for A to B. Table 1 shows the source node A has three
Energy(v). Accordingly, in the network, every link e¼(vi,vj)AE
neighbors B, C, and D. Data packets are sent from the source node
has its corresponding QoS metrics, which are respectively delay
A to the destination nodes X, Y, and Z, respectively.
function—Delay(e), bandwidth function—Bandwidth(e), packet
How to update the above routing table is very important. In
loss function—Packet_loss(e), and energy function—Energy(e).
order to get the information of packet transmission, a multi-agent
After defining the QoS metrics of the node and the link, the
model needs to be designed.
QoS metrics of the path P(vs,vd) can be calculated. Given the
source node vsAV and the destination node vdAV, the correspond-
ing QoS metrics of path P(vs,vd) are computed as the following: 4.2. Multi-agent model for QoS based routing
P X
DelayðPðvs ,vd ÞÞ ¼ v A Pðvs ,vd Þ DelayðvÞ þ DelayðeÞ
e A Pðvs ,vd Þ The intelligent agent (Shakshuki and Malik, 2006; Wu and
BandwidthðPðvs ,vd ÞÞ ¼ min fBandwidthðeÞg Kim, 2010; Chen et al., 2007) can be considered as an appropriate
e A Pðvs ,vd Þ
Y carrier to use the intelligent algorithm in WSNs; the agent can be
Packet_lossðPðvs ,vd ÞÞ ¼ 1 ð1Packet_lossðeÞÞ applied to apperceive the changes in network topology, the
e A Pðvs ,vd Þ
network communication flow, and each node’s energy state, and
If the path P(vs,vd) is a path satisfying all the QoS metrics, it must can also participate in network routing and network mainte-
meet the following requirements: nance, as shown in Fig. 2. Agent is objective-oriented and
P X abundant with communication languages; it can provide more
DelayðPðvs ,vd ÞÞ ¼ v A Pðvs ,vd Þ DelayðvÞ þ DelayðeÞ oD
flexible interaction and cooperation mode, and can satisfy the
e A Pðvs ,vd Þ
interaction of node in distributed network environment.
BandwidthðPðvs ,vd ÞÞ ¼ min fBandwidthðeÞg 4 B
e A Pðvs ,vd Þ
Y
Packet_lossðPðvs ,vd ÞÞ ¼ 1 ð1Packet_lossðeÞÞ oPL Table 1
e A Pðvs ,vd Þ The routing table of node A.

where D, B, and PL are the QoS guarantees of the WSN network. The source node A The adjacent The adjacent The adjacent
After defining every QoS function of the routing model, we can node B node C node D
establish the synthetic QoS model for every path. In the synthetic
QoS model, in which every QoS indicator must satisfy the QoS The destination node X 0.84 0.57 0.70
The destination node Y 0.71 0.44 0.85
constrain, any inconformity will greatly cut down the metrics’ The destination node Z 0.73 0.38 0.29
contribution and bring the negative and punitive influence to the
synthetic QoS.
For example, if Delay(P(vs,vd)) oD, the delay of the path may
satisfy the constraint conditions, then
ð1kÞDelayðPðvs ,vd ÞÞ
fdelay ¼ 1 ð1Þ
D
Take k close to 1, such as 0.9, then the value of fdelay will be
between 0.9 and 1.
If Delay(P(vs,vd)) 4D, it denotes that the delay indicator of path
cannot satisfy the constraint demands for delay application, then

DelayðPðvs ,vd ÞÞ
fdelay ¼ ð1kÞ ð2Þ
D Fig. 1. Topological structure of the node A.

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
4 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]]

D
A

S B

Current node Local time Travel record

S tS S
A tA S,A
B tB S,A,B
D tD S,A,B,D
Fig. 2. Behavior abstraction of the agent.
Fig. 3. The change of the data structure in the routing process of the forward
agent.
In QoS based routing, the synthetic QoS metrics are added into
the data structure of agent. Therefore, the data structure of the
agent consists of the agent ID and its type, the source node ID, the If broadcasting the forward agent, all of the adjacent nodes of
destination node ID, the current node ID, the hop distance of the node vk will receive the message from the forward agent.
agent, the start time and reach time, etc., and also includes the Finally, the forward agent will arrive at the destination node. The
mobile records of the agent. In the structure of the mobile forward agent may be deleted when the value of the routing hop
records, the QoS metrics are defined, such as delay, bandwidth, counter is beyond the setting value. By the flooding communica-
packet loss, energy, etc. tion, the intermediate nodes can copy and broadcast the forward
In the QoS based routing algorithm, the forward agent and the agent. Thus, the destination node may indirectly receive more
reverse agent are adopted to establish the routing strategy of than one forward agent from the source node vs, which forms the
WSN nodes. Routing tables are updated as the forward agent different routing paths between the source node vs and the
sends a packet from the source node to the destination node. destination node vd.
Once it reaches its destination, each forward agent tells the Adopting the routing-choose method to select the next node
traveling time information and other QoS parameter to of the forward agent, the routing table of the node vk will determine
the reverse agent, which updates the routing tables as it traces the next node, to which the forward agent will go. Figure 3 shows
the path of the forward agent in reverse. the change of the agent data structure in the routing process of the
forward agent. The source node S produces a forward agent, and
establishes the initial travel record in node S. Through the routing
4.2.1. Forward agent table of the nodes S, it transfers the forward agent to node A and
If node s hopes to establish routing with other nodes, the node then adds the travel record in. In the same way, the forward agent
s creates a forward agent in the creation time of the forward agent can arrive at the node B, and finally arrive at the destination node D.
message and writes into its own address, and then continuously The travel record only lists the node identification.
sends the forward agent to each adjacent node in flooding mode.
When node vk receives a forward agent, it implements the
following tasks. 4.2.2. Reverse agent
At the outset, the forward agent checks whether there exist The task of the reverse agent is to return to the source node vs
some visited nodes in its travel records. If exists, it shows that along the path of the forward agent, and to implement the
circulation appears in agent travel, and deletes it from the stacks. corresponding routing algorithm. When a forward agent arrives
If the travel information of the forward agent is stored by a stack, at the destination node vd, the node vd will initialize the forward
the delete operation can be done by simple stack pop-up agent into a reverse agent through changing some signs, and the
operation. reverse agent inherits the travel records of the forward agent.
Then, the forward agent adds a new data item into the mobile According to the path message in the original travel records, the
records, which point to a dynamically increasing list. Data item reverse agent will transmit the travel records to the source
includes the identification and relative information of the node vk node along the anti-direction with the forward agent. Meanwhile,
according to the algorithm requirements. the reverse agent will adjust the routing table along the travel
Next, if the node vk is not the destination node, the value of the path according to the travel records. Considering the reverse
routing counter in the forward agent adds 1. Every source node agent needs to follow its predecessor (the forward agent) to
address and its serial number contained in the forward agent are return to the source node; the reverse agent will no longer
only certain, and every intermediate node will record the source transmit message in flooding mode. In order to quickly adjust
node address and the maximum serial number (the maximum the routing table information along the travel path, the reverse
serial number is the serial number of the latest forward agent agent is given a higher priority, and the adjustment to the routing
received by the current node and sent from the same source table by the reverse agent is decided by the different routing
node). If the serial number received by the current node is not algorithms. When the reverse agent arrives at some nodes on the
bigger than the maximum serial number from the same source back path to the source node, some jobs need to be finished as
node, the forward agent will be discarded. Since the forward follows: (1) the reverse agent modifies the routing table of node
agent and the data message have the same priority and follow the vk using the travel time in the memory. (2) If the node vk is not
First-In First-Out (FIFO), the forward agent will experience the the source node, the reverse agent decides the next node accord-
same delay and congestion as the data message, by which the QoS ing to its own original travel records, and sends message to the
based routing can be built. next node.

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]] 5

5. PSO and agent based algorithm for the QoS routing model P is a particle swarm of the usable routing path, and Pi(vs,vd) is
a particle. Pi(vs,vd) contains the information of PSO algorithm:
The QoS based routing of WSNs is a Non-deterministic Poly-
nomial (NP) problem. Particle Swarm Optimization (PSO) can Pi(vs,vd)¼(vs,v1,v2,v3,y,vi  1,vi,vi þ 1,y,vn,vd) is the position
effectively solve the complex NP optimizing tasks (Kennedy and information of particles;
Eberhart, 1995; Liang and Yu, 2005) and PSO iteration formulae Ppbest i(vs,vd) denotes the optimal target value of the particle
can be represented by the following formula: Pi(vs,vd);
Pgbest(vs,vd) denotes the optimal target value of the optimum
vi ¼ ovi þ Z1 r and ðÞðpbesti xi Þ þ Z2 r and ðÞðpgbest xi Þ target.

xi ¼ xi þ vi Dt Then, the QoS-PSO algorithm can be performed by the follow-

ing steps:
where, o is the inertia weight, Z1 and Z2 are two random
variables subjected to the uniform distribution, vi is the velocity (1) Select the source node and the destination node, and apply the
of particle i, xi is the current position of particle i, Pbesti is the best forward agent and the reverse agent to establish the path set
position of particle i, and Gbest is the best position of the group. (swarm) between the source node and the destination node.
Using the above equation, a certain velocity that gradually gets (2) Calculate the synthetic QoS for each path (particle) to obtain
closer to Pbesti and Gbest can be calculated. Pbest, and select the best Pbest as Pgbest.
In WSNs, the paths between the source node and the destina- (3) Use the iteration formulas of PSO to get a new routing path.
tion node can be regarded as a set with multiple feasible (4) Calculate its synthetic QoS.
solutions; the set is a swarm in PSO, and every feasible path is (5) When QoS of a path (particle) is better than Pbest, use it to
regarded as a particle. The dynamic adjustment of path can be replace Pbest. When the best Pbest is better than Pgbest, use it to
considered as the adjustment of path node. The PSO and replace Pgbest.
agent based algorithm for the QoS routing model (remarked by (6) If the iteration time of the algorithm is beyond the designated
QoS-PSO algorithm) applies the synthetic QoS metrics as the maximum or the QoS computing value meets the require-
objective function of PSO to address an optimal path for node ments, the algorithm stops. Otherwise, turn to Step (3).
routing. (7) Determine the path, and modify the routing table.
In QoS-PSO algorithm, the agent based routing provides an
initial path for PSO. Every node in the network creates and
maintains a routing table for all the other nodes, and multiple 6. Experiments and analysis
routes between the source node s and destination node d are built
in the routing table. When a source node needs to send a data In order to monitor the cotton’s growth environment in coastal
packet to the destination node, the source node hopes to build a saline land, a farmland information acquisition system is estab-
routing with the destination node and starts an agent based lished to constantly monitor the environment parameters in the
routing discovery mechanism if no corresponding routing infor- target land, such as water, soil, salt, temperature, etc. The system
mation can be found in the routing table. By routing discovery applies the ZigBee-based WSNs technology and GPRS-based
mechanism, the source node generates a forward agent, writes its communication technology to collect the environment para-
own address and data creating time to the agent, and then meters, by which farmers can analyze the characteristics of the
broadcasts the agent to every adjacent node. The forward agent saline-alkali soil and get the useful guidance for cotton growth.
in routing discovery process is transmitted on the way of The system architecture is depicted in Fig. 4.
broadcasting. The system is composed by WSN data acquisition subsystem,
When the forward agent arrives at the destination node, a GPRS telecommunication subsystem, and data monitoring center.
route path is found. The forward agent will be converted to a The data acquisition subsystem, a ZigBee based wireless sensor
reverse agent. The reverse agent returns to the source node along network, finishes the monitoring of parameters of each node (Ai)
the way of the forward agent broadcasting, and meanwhile in the target area and transmits data packet to the sink node
adjusts the node routing tables it passed according to the network through the middle route node (A0i). The sink node sends the
situation. This process is called the multi-agent based routing monitoring data to the data monitoring center by GPRS module.
discovery. In the next routing experiments, the Ad hoc On-Demand Distance
The routing of the forward agent uses the way of broadcasting; Vector Routing (AODV) and the Energy-Efficient Ant-Based Routing
thus, multiple forward agents will arrive at the destination node. Algorithm (EEABR) are chosen as comparative algorithms to verify
That is to say, there will be more paths between the source node system performance. AODV is one of the classical WSN routing
and the destination node, which can be regarded as the initial algorithms; it can be applied to all kinds of network environments
swarm of PSO algorithm. and support service quality requirements. EEABR is another well
Assuming a path from the source node vs to the destination known SI-ACO based WSN routing algorithm (Camilo et al., 2006).
node vd: P(vs,vd)¼(vs,v1,v2,v3,y,vi  1,vi,vi þ 1,y,vn,vd). In the first experiments, WSN parameters were designed as
Assume that vi is a randomly selected node in the path P(vs,vd). shown in Table 2. The number of nodes is from 10 to 100 and the
If there exists a node v, a neighboring node of vi, and it is also iterative number is 10. The metrics, such as end-to-end delay,
adjacent with another two nodes vm and vn in the path, packet loss, and the synthetic QoS, were calculated by taking an
mA{s,y,1,y,i 1}, nA{iþ1,y,n,d}, then P0 (vs,vd) ¼(vs,v1,v2, average of the 10 times’ value, respectively, from AODV, EEABR
v3,y,v,vn,y,vd) can be defined as an adjacent path of the path and QoS-PSO algorithm with the increase in the node number.
P(vs,vd). Figures 5–7 show, respectively, the mean delay curve, the packet
After establishing a random adjacent path between the source loss curve, and the synthetic QoS curve with the increase of nodes.
node vs and the destination node vd, the QoS-PSO algorithm for From Fig. 5, we can find that the improved QoS-PSO algorithm
WSNs can be established. reduces 10–50% than AODV algorithm at end-to-end delay, and
Given the path set from the source node vs to the destination reduces up to 10% than EEABR algorithm. The result obviously
node vd: P¼ {Pi(vs,vd)9i¼1,2,3,y,n}. shows that the QoS-PSO algorithm needs a large number of nodes

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
6 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]]

Fig. 4. System architecture for farmland information acquisition system.

Table 2
0.6
Experiment parameters.
0.5
Experiment parameters list

Number of nodes 10,20,30,y,90,100 0.4

Packet Loss
Network range 1000 m  800 m
Communication range 50 m
Maximum rate of data 150 Kbps 0.3
Message size 64 bytes
Bandwidth of link 1 Mbps
Node mobile model BSAM
0.2
AODV
0.1 EEABR
6 QoS- PSO
0
10 20 30 40 50 60 70 80 90 100
5
number of node
Mean delay [s]

4 Fig. 6. Packet loss vs. the number of node.

3 1
0.9
2
0.8
AODV
1 EEABR 0.7
QoS- PSO 0.6
QoS

0 0.5
10 20 30 40 50 60 70 80 90 100
number of node 0.4
0.3 AODV
Fig. 5. Mean delay vs. the number of node.
0.2 EEABR
to obtain good performance; with the number of nodes decreas- 0.1 QoS- PSO
ing, the performance of algorithm will depress. 0
A comparison of packet loss (PL) between AODV algorithm, EEABR 10 20 30 40 50 60 70 80 90 100
algorithm, and QoS-PSO algorithm was provided in Fig. 6. It shows number of node
that the packet loss of QoS-PSO algorithm is similar to that of AODV
and EEABR when the number of nodes in the network is small. Fig. 7. The synthetic QoS vs. the number of node.
However, with the number of nodes increasing in the network, the
communication flows will surge. Under this condition, the packet loss
caused by the network congestion will significantly increase, and the ratio of delay and packet loss are only 10–50%, but the rising ratio of
packet loss of AODV is more than that of EEABR and QoS-PSO the synthetic QoS will reach high to 25–100%.
algorithms, between which QoS-PSO algorithm is obviously better. The second experiments with more nodes are done to verify
When the number of node is 100, the packet loss of AODV algorithm the optimization effect of service quality metrics in a relatively
is as high as 0.52, EEABR’s PL is 0.47, and QoS-PSO’s PL is only 0.38, broader wireless sensor network environment. The network
which is reduced by about 30% to AODV and about 20% to EEABR. experiment parameters are listed in Table 3.
In Fig. 7, all of the synthetic QoS metrics will decline with Figure 8 shows that the packet loss will increase with increase of
increase of the node number, but the synthetic QoS metrics of time, but the packet loss value of QoS-PSO algorithm is obviously
QoS-PSO algorithm is obviously higher than AODV and EEABR. It is lower than that value of AODV algorithm and EEABR algorithm, and
worth noting that the improvement percentage grows even up to its packet loss will finally tend to be a constant and keeps below 0.48,
100% than AODV with the increase of the number. while AODV and EEABR are 0.64 and 0.56, respectively.
In QoS-PSO algorithm, we paid more attention to the synthetic From Fig. 9, the QoS metrics of network will basically maintain
QoS metrics. Through considering the synthetic QoS, the reduction a certain value with the increase of time, but the QoS-PSO

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]] 7

Table 3 1
Experiment parameters.
0.9
Parameter list 0.8
Number of nodes 500 0.7
Network range 5000 m  5000 m
Communication range 200 m
0.6

QoS
Maximum rate of data 512 Kbps 0.5
Message size 64 bytes
Node mobile model RWM 0.4
Range of node moving speed 0.2 m/s,451/s
0.3 AODV
Bandwidth of link 10 Kbps
0.2 EEABR
0.1 QoS- PSO

0.7 0
0 2 4 6 8 10 12 14 16 18 20 22 24
0.6 Time

Fig. 10. The synthetic QoS vs. TIME when environment suddenly changes.
0.5
Packet Loss

0.4 1
0.9
0.3

Average Residual Energy

0.8
0.2 0.7
AODV
EEABR 0.6
0.1
QoS- PSO 0.5
0 0.4
0 2 4 6 8 10 12 14 16 18 20 22 24
0.3 AODV
Time
0.2 EEABR
Fig. 8. Packet loss vs. time. 0.1 QoS- PSO
0
1 10 20 30 40 50 60 70 80 90 100
0.9 number of node
0.8
Fig. 11. Average residual energy after 256 packets are received.
0.7
0.6 the field environment, and the recovery time is much shorter than
QoS

0.5 that of AODV and EEABR.

Finally, since WSN nodes have limited resources and the proposed
0.4
algorithm requires maintaining data structure of several fields in
0.3 each node, creating agents, continuously sending agent to each
AODV
0.2 adjacent node in flooding mode, etc., we monitor the average residual
EEABR
energy to analyze the impact of the resource in each node incurred by
0.1 QoS- PSO
the algorithm in the fourth experiments. The average residual energy
0 of the three algorithms after 256 packets is depicted in Fig. 11.
0 2 4 6 8 10 12 14 16 18 20 22 24 Figure 11 shows that the AODV algorithm plays worst, and QoS-
Time PSO and EEABR are almost the same. It is enough to say that the
proposed algorithm is practically effective.
Fig. 9. The synthetic QoS vs. time.

algorithm improves the metrics higher than AODV by even 60%, 7. Conclusion
and over 20% than EEABR.
In the field experiments, the possible environment changes, In this paper, we present a QoS-PSO algorithm for the syn-
such as sudden wind and electromagnetism impact, may cause thetic QoS routing model to increase the QoS level of WSN. The
the changes of the WSN network parameters, which will take an model applies the synthetic QoS parameters as the objective
important effect to the routing algorithms. To test the response function of PSO to address an optimal path for node routing,
ability of the algorithm to the environment change, movement and the multi-agent based routing table provides an initial path
speed of the node is suddenly increased to 1 m/s and the time for QoS-PSO algorithm. Finally, compared with the traditional
persists for 1 min after the third experiments implement 12 h; AODV algorithm and EEABR algorithm, the QoS-PSO algorithm
the QoS curves of both routing algorithms are depicted in Fig. 10. obviously shows its improvement in the quality of service of WSN
In Fig. 10, these experimental results show that all algorithms including delay, packet loss, and the synthetic QoS. The QoS-PSO is
demonstrate good robustness and have the ability to restore the also proved to be practically possible with the average residual
original level of QoS after a short fluctuating process. Furthermore, energy. In the future work, the proposed routing algorithm needs to
the QoS-PSO algorithm has a stronger adaptive capacity to changes in be improved effectively for those nodes with high mobility by setting

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031
8 M. Liu et al. / Journal of Network and Computer Applications ] (]]]]) ]]]–]]]

some proper WSN parameters and for the routing networks with Deb B, Bhatnagar S, Nath B. ReinforM: reliable information forwarding using
multiple sink nodes. multiple paths in sensor networks. In: Proceedings of the 28th Annual IEEE
international conference on local computer networks, Los Alamitos, CA, USA;
2003. p. 406–15.
Deng BW, Huang GM. Algorithm of creating routing table for WSNs. In:
Acknowledgments Proceedings—5th international conference on wireless communications, net-
working and mobile computing, WiCOM 2009; 2009.
Dorigo M, Stützle T, editors. Ant colony optimization. MIT Press; 2004.
The research work presented in this paper was partially Eberhart RC, Shi Y.Comparing inertia weights and constriction factors in particle
supported by the NSFC (Grant nos. 61034004 and 61073090), swarm optimization. In: Proceedings of the IEEE international conference on
the Guangdong Science and Technology Department of China neural networks. Piscataway: IEEE Service Center; 1995. p. 1942–48.
Felemban E, Chang-Gun L, Ekici E. MMSPEED: multipath multispeed protocol for
(Grant nos. 2009GJE00026 and 2009B090300429), the SMST QoS guarantee of reliability and timeliness in wireless sensor networks. IEEE
Commission (Grant no. 09510701300), and the SRF for ROCS Transactions on Mobile Computing 2006;5(6):738–54.
of SEM. He T, Stankovic JA, Lu C, Abdelzaher T. SPEED: a stateless protocol for real-time
communication in sensor networks. In: Proceedings of the 23rd IEEE interna-
tional conference on distributed computing systems (ICDCS-03), Providence,
References RI, United States; 2003. pp. 46–55.
Houngbadji T, Pierre. S. QoSNET: an integrated QoS network for routing protocols
in large scale wireless sensor networks. Computer Communications 2010;33:
Agrawal DP, Mahapatra A, Anand K. QoS and energy aware routing for real-time 1334–42.
traffic in wireless sensor networks. Computer Communications 2006;29(4): Kennedy J, Eberhart RC. Particle Swarm Optimization. In: Proceedings of the IEEE
437–45. international conference on neural networks, vol 4, Perth, Australia; 1995,
Akyildiz IF, Su W, Sankarasubramaniam Y, Cayirci E. Wireless sensor networks: a p. 1942–48.
survey. Computer Networks 2002;38(4):393–422. Kennedy J, Eberhart RC, Shi Y. Swarm intelligence. San Francisco, USA: Morgan
Akyildiz IF, Melodia T, Chowdhury KR. A survey on wireless multimedia sensor Kaufman; 2001.
networks. Computer Networks 2007;51(4):921–60. Liang Y, Yu HB. PSO-based energy efficient Gatheringin sensor networks. Mobile
Al-Karaki JN, Kamal AE. Routing techniques in wireless sensor networks: a survey. ad-hoc and sensor network.Berlin Heidelberg: Springer-Verlag; 2005 .
IEEE Wireless Communications 2004;11:6–28. Liang Y, Yu HB, Zeng P. Optimization of cluster-based routing protocols in wireless
Bergh FVD, Engelbrecht A. A study of particle swarm optimization particle sensor network using PSO. Control and Decision 2006;21:453–6. [in Chinese].
trajectories. Information Sciences 2006;176(8):937–71. Liao WH, Kao YC, Fan CM. Data aggregation in wireless sensor networks using ant
Bonabeau E, Dorigo M, Theraulaz. G. Swarm intelligence: from natural to artificial colony algorithm. Journal of Network and Computer Applications 2008;31:
systems. New York, USA: Oxford University Press; 1999. 387–401.
Camilo T, Carreto C, Silva J, et al. An energy-efficient ant base routing algorithm for Saleem M, Caro GAD, Farooq. M. Swarm intelligence based routing protocol for
wireless sensor networks. In: Proceedings of the fifth international workshop wireless sensor networks: survey and future directions. Information Sciences
on ant colony optimization and swarm intelligence, vol 4150; 2006. p. 49–59. 2010. doi:10.1016/j.ins.2010.07.005.
Celik F, Zengin A, Tuncel. S. A survey on swarm intelligence based routing Shakshuki E, Malik H. Multi-agent Based Clustering Approach to Wireless Sensor
protocols in wireless sensor networks. International Journal of the Physical Networks. International Journal of Wireless and Mobile Computing (IJWMC)
Sciences 2010;5(14):2118–26. 4. 2006. Inderscience Publisher.
Chen HS, Ji ZZ, Hu MZ, et al. Design and performance evaluation of a multi-agent- Sohrabi K, Gao J, Ailawadhi V, Pottie GJ. Protocols for selforganization of a wireless
based dynamic lifetime security scheme for AODV routing protocol. Journal of sensor network. IEEE Personal Communications 2000;7(5):16–27.
Network and Computer Applications 2007;30:145–66. Teodorovic D. Bee colony optimization (bco). In: Lim CP, Jain LC, Dehuri S, editors.
Cobo L, Quintero A, Pierre. S. Ant-based routing for wireless multimedia sensor Innovations in Swarm Intelligence. Springer-Verlag; 2009 .
networks using multiple QoS metrics. Computer networks 2010;54(17): Wu M, Kim. CG. A cost matrix agent for shortest path routing in ad hoc networks.
2991–3010. Journal of Network and Computer Applications 2010;33:646–52.
Constandinos X, Mavromoustakis H, Karatza D. Optimized QoS priority routing for Zhang XH, Xu. WB. QoS based routing in wireless sensor network with Particle
service tenability and overhead reduction using swarm based active network Swarm Optimization. Lecture Notes in Computer Science 2006;4088:602–7.
scheme. Computer Communications 2006;29:765–80. Zhang Y, Kuhn L, Fromherz M. Improvements on Ant Routing for Sensor Networks.
Crawley E, Nair R, Rajagopalan B, et al. A framework for QoS-Based routing in the In: Ants 2004, workshop on Ant Colony Optimization and Swarm Intelligence.
internet. RFC2386. /http://www.ietf.org/rfc/rfc.2386.txtS; August 1998. p. 154–65.

Please cite this article as: Liu M, et al. An agent-assisted QoS-based routing algorithm for wireless sensor networks. J Network Comput
Appl (2011), doi:10.1016/j.jnca.2011.03.031