Data gathering in wireless sensor networks using mobile elements

International Journal of Wireless & Mobile Networks (IJWMN) Vol. 6, No. 4, August 2014
DATA-GATHERING IN WIRELESS SENSOR
NETWORKS USING MOBILE ELEMENTS
Khaled A. Almi’ani, Ali Al_ghonmein, Khaldun Al-Moghrabi and Muder
Almi’ani
Al-Hussein Bin Talal University
ABSTRACT
Using mobile gateway as mechanical data carrier has emerged as a promising approach to prolong the
network lifetime and relaying information in partition networks. This mobile gateway periodically travels
the network to gather the sensor data, where the gateway tour start and end at the sink. The gateway’s tour
length must be bounded by pre-defined time constant to avoid buffer overflow. In this paper, we investigate
the problem of scheduling the mobile gateway tour in which the tour length satisfies bounding constraints
while the sensors lifetime is also increased. We present an algorithmic approach that schedule gateway
path by partitioning the network into clusters, so that one node from each cluster must be visited by a
mobile gateway. Our experiment results demonstrate that the proposed approach significantly increases
the network lifetime compare to networks with static-sink. Also, the quality of the obtained gateway is
within 3/2 of the optimal solution tour length.
1. INTRODUCTION
Wireless Sensor Networks (WSNs) have recently witnessed increasing effort to explore
applications in various environments [1-3]. A WSN consists of hundred of battery-powered
devices deployed in a fly for unattended operations. Because energy is the main concern in WSN,
once the network is deployed, recharging the sensors batteries becomes impractical. One of the
major energy expenditures is communicating the sensors to deliver their readings to sinks. This
communication pattern results in hotspots whereby affected sensors around the sink(s) die earlier
than the others as all traffic is funnelled through these sensors.
To address this problem, using mobile gateways has emerged as a promising option [4]. Every
mobile gateway travels the network to gather the sensors data and returns to the sink (departure
point) to upload the data. By embedding the network with mobile gateways, the sensors data
forwarding traffic will be reduced significantly. In addition, the network no longer needs to be
connected, since mobile gateways could works as bridges to connect network partitions.
In the literature, many proposals have investigated the beneficial aspects of employing mobile
gateways. Based on the gateway’s motion strategy, these proposals can be categorized as follows:
• Random motion: Gateways can be mounted on entities that travel the network in an
unplanned fashion. For instance in [5, 6], humans or animals act as “data mules” travel
the network and opportunistically visit sensors to upload their data. In this motion
strategy, the End-to-End delay cannot be bounded as providing a reliable
communication is very difficult.
• Fixed motion: Gateways travel the network in a fixed path. For instance in [7], a
gateway is mounted on a public bus, which moves in a pre-determined schedule. The
DOI : 10.5121/ijwmn.2014.6409 113

International Journal of Wireless & Mobile Networks (IJWMN) Vol. 6, No. 4, August 2014
sensors on the street monitor the bus movements to determine when they can
communicate with the gateway.
• Controlled motion: Gateway path is determined based on an objective function. For
instance in [8, 9], gateway path is scheduled in which each sensor must be visited
before its buffer becomes full.
In this paper, we investigate the problem of scheduling mobile gateway paths in the
controlled motion paradigm. Here, we assumed that sensor network is deployed over a
large terrain. Sensors have the same sampling rate with a limited buffer size. The
network is then equipped with a mobile gateway that can travel the network to gather
the sensors data. The gateway tour start and end at the sink. Sensor data must be
uploaded to the sink at a pre-determined rate. This rate is determined based on the
sensors buffer size and the end-user interest. Satisfying this frequency constraint results
in bounding the gateway tour length, and therefore limits the number of sensors that the
gateway can visit in its tour. A problem that naturally crops up is to determine which
sensors the gateway must visit in its path, and how the sensors data should be routed
and stored in the sensors the gateway will visit during its tour.We refer to this problem
as the Mobile Gateway Scheduling with Visiting Deadline (MGS-VD).
We address this problem by dividing it into two steps: (1) clustering and (2) path planning. The
ground concept of these two steps is to partition the network into clusters so that the gateway path
can be constructed from only one sensor from each cluster. The sensors that will be involved in
the gateway tour will work as cluster heads as they are responsible to store other sensors data.
When the gateway enters the physical transmission range for every cluster head, the data stored in
the cluster head will be transferred into the gateway’s memory. These two steps will work
recursively to ensure that number of clusters is maximal and the established tour also satisfies the
visiting frequency constraint.
The rest of the paper is organised as follows. Section 2, provides a formal definition of the MGS-VD
problem. In Section 3, the related work of this research area is presented. Section 4 presents
our algorithmic approach. A number of experiments and the corresponding analysis and results
are presented In Section 5. The paper is concluded in Section 6.
114
2. PROBLEM DEFINITION
Assume an undirected graph
to represent a network topology. Here, V is a set of
sensor nodes as vertices, is a set of edges denoting a communication link between two sensor
nodes, and is a unique vertex in the network to represent its base-station/sink. We also
assumed that all sensors have the same sampling rate with a limited buffer size. Sensor data must
be delivered to the sink once every t time steps. The value of t can be determined as a
combination of the end‐user interest and sensors overflow restriction. The network is equipped
with a mobile gateway to gather network data by traveling the network at a constant speed. Every
time the gateway reaches a sensor, it downloads the sensor data into its memory. The mobile
gateway tour starts and ends at , while the travelling time of the gateway tour must be bounded
by t. The MGS-VD problem can now be defined as follows:
• Partitioning the set V into k disjointed sets

, so that
,

, and the vertices in
are connected.
• Finding the minimum travelling time tour for the mobile gateway that starts and ends at
, and also contains exactly one element from each of the groups
, the travelling time
of this tour must be less than or equal to t.
• The partitioning
is determined so the average sensor forwarding traffic is minimized

International Journal of Wireless Mobile Networks (IJWMN) Vol. 6, No. 4, August 2014
115
3. RELATED WORK
The use of mobile gateway as data carries has recently been explored in the literature. In [4], an
investigation that discussed several advantages of using mobile gateways have been presented,
whereas this investigation have mainly focused on communication protocols and reliability. In
[10, 11], radio-tagged zebras and whales are used as mobile gateways. These animal-based
gateways move randomly in the network terrain and exchange messages opportunistically. In
[12], the message ferries approach are used to route the data in a sparse network. The main
concept of this approach is to determine the mobile gateway path that minimizes the average
delay. In [5] , the investigation explored the benefit of employing mobile gateways, which travel
the network in parallel straight lines. To reduce delay, sensors away from the gateway path must
forward there packets to nearby sensors.
The mobile element scheduling (MES) problem [8, 9] has some similarity with the MGSVD
problem. This problem deals with determining the gateway path in which there is no data loss due
to sensor node buffer overflow. By adopting the assumption that the sensors must be visited
before their buffers become full, MES and MGS-VD share the property that the sensors must be
visited based on a deterministic frequency. However, MGS-VD addresses the situation where
constructing the gateway tour to visit all sensors without violating the visiting frequency
constraint is unachievable.
In situations where the gateway tour can be constructing to include all sensors without violating
the visiting frequency constraint, the MGS-VD becomes an extend of the well-known
Orienteering problem [13]. This problem is defined as determining the minimum tour length for a
vehicle to visit n-cities before a pre-determined time deadline. In this situations both of these
problems share the property that the visiting must be done before a pre-determined time deadline.
4. ALGORITHMIC SOLUTION
In this section we present an algorithmic approach to handle MGS-VD problem. Our goal is to
determine the gateway tour that satisfy the visiting frequency constraint and maximally reduces
the energy expenditures due to packets relaying. Here, the ground concept is to partition the
network into energy-aware clusters before establishing the gateway’s tour to visit the constructed
clusters.
Partitioning the network aims to construct clusters that have approximately the same number of
nodes. Such construction balances the sensors energy consumption since they will have
approximately the same forwarding load. Once the clusters are constructed, the gateway tour will
be established to involve one sensor from each cluster, whereas these selected sensors will work
as clusters heads, and they will store the other sensors packets.
The partitioning and the tour constructing steps will work recursively to find the maximum
possible number of clusters that satisfy the visiting frequency constraint. At the beginning, n
number of clusters will be constructed, n is equal to the number of nodes in the networks. Then,
in each round, if the tour that connects the cluster heads doesnot satisfy the visiting constraints,
the clustering process will be re-triggered and the number of clusters will be divided by two. This
process will stop and the gateway path will be obtained when the maximum number of clusters
that satisfy the visiting constraint is found.
In the partitioning step, the sensors will be grouped into clusters, where minimizing the distance
between nodes belong to the same cluster is the construction criteria. In this context, distance is
defined as the number of hops in the shortest path to connect two nodes. Figure 1 outlines the

process of this step. This process starts by selecting random nodes as the initial clusters. Once
these nodes are identified, each node will be assigned to its nearest cluster. After all nodes are
assigned to clusters, the centre node (Cn) for each cluster is determined. In a cluster, the centre
node is the node that has the minimum distance to all nodes in this cluster. In situations where
these identified nodes do not match the previous nodes, the process will be repeated. This process
is terminated to obtain clusters when the identified centre nodes are similar to the nodes identified
in the previous iteration.
116
Input: G,
1 Clusters
2 Cn [ ] select random nodes from G
3 O__node [ ] 0
4 hops[][n]
5 for i !#
6 do add C_CENTER [i] to
7 Stable false
8 While not stable
9 do for i !#
10 do for j !#$
11 do hops[Cn [i]][j] number of hops in the shortest path between i and j
12 for each I %
13 do add I to if Cn [j] is the closet Cn to i
14 O__node Cn
15 for each
16 do Cn [i] findCns()
17 If O__node = Cn then Stable true
Figure 1: the clustering process
Once the clusters are identified, the path planning step will be triggered to construct the gateway
tour. The path planning step aims to determine the minimum tour length that visits exactly one
sensor from each cluster, where this tour start and end at the sink. This description results in
considering this problem as a variant of the One-of-a-Set TSP [14], which is also been referred to
as the Errand scheduling problem[15]. The One-of-a-Set TSP deals with determining the
minimum tour length that visits at least one node from each set, where the Errand scheduling
problem deals with determining the best order of performing specific errands, each in which can
be performed at different nodes in the graph. The only difference in definition between these two
problems and the problem of planning the gateway path is in the number of nodes that the
constructed tour can visit from each set. In One-of-a-Set TSP and Errand scheduling problem, the
lower bound of this number is one, where there is no restriction about the upper bound. In the
problem of planning the gateway path, the number of nodes the tour must include from each set is
exactly one. This consideration emphasis the inherited relation between the TSP[16] and the path
planning problem investigated in this work. TSP deals with determining the minimum tour length
for a salesman to visit n-cities. Therefore, it is reasonable to construct the gateway tour based on
the TSP tour.

The process of constructing the tour is divided into two steps, nodes-identification and tour-constructing.
In nodes-identification, the identity of the nodes that will participate in the tour will
be identified. In tour-constructing, the TSP tour for these identified nodes will be constructed. To
construct the TSP tour, we employ Christofides algorithm [16], which is well-citied practical
algorithm and it has been used as a benchmark whenever a new algorithm is proposed for the
TSP. There are many other sophisticated algorithms that outperform Christofides but since we are
using a heuristic approach, starting from a simple and very robust algorithm has the advantage of
simplifying the implementation. The nodes-dentification step aims to select the nodes to form the
gateway path. In each round, this step tags the closest node to the partial-constructed tour as a
selected node. This process will stop when a node from each cluster is selected. Once these nodes
are identified, the tour-constructing will construct the TSP tour for the selected nodes using
Christofides algorithm.
117
5. EXPERIMENT EVALUATION
In this section, we have conducted extensive set of experiments to evaluate the performance of the
presented approach, namely MG. This validation is performed using J-Sim simulator [17]. We
observe the network lifetime and gateway tour length as the evaluation metrics. In the lifetime
evaluation, MG performance will be benchmarked against two other schemes; mobile-sink and
static-sink. In mobile-sink scheme, the sink will visit each sensor to download its data, where in
static-sink scheme; the sensors have to forward their data to reach the sink in a multi-hop fashion.
In this evaluation, only the cost of transmitting and receiving the actual data will be counted as
the energy expenditure. This consideration aims to emphasis the actual influence of MG on
network lifetime, since counting other sources of energy expenditure will trivially shows the
benefits of MG on network lifetime. To evaluate the quality of the tour length obtained by MG,
we compare its solution against the optimal result obtained by CPLEX .
For the purpose of this simulation, we adopt the two-ray propagation model. With transmission
power set to 21 mW, and receiving power set to 15 mW. The data packet size is set to 50 byte
and the data rate to 115 Kbps. Sensors are assumed to sample their reading once every second,
and they have 5 K-byte storage capability. Unless mentioned otherwise, each simulation is run on
a network with 200 node randomly deployed across 200×200'. The gateway is considered to
move in the network at 1 m/s speed. Each experiment is repeated 10 times and the average is
obtained.
5.1 Network Lifetime
Now we evaluate MG performance in term of network lifetime. In this evaluation, we consider
the x percent network lifetime metric, which is defined as the time until x percent of nodes run out
of energy. The values of x used in this evaluation are 10% and 50%. To simplify the analysis, the
value of MG tour length bound is mapped to facilitate controlling the number of obtained clusters.
Figures 2 (a) and (b) show the results for this evaluation. From these figures we can see that when
the number of clusters is greater than one, MG is able to significantly increase the network
lifetime. These figures also clearly show that increasing the number of clusters substantially
increases the gap between MG and static-sink scheme. As expected, this behaviour is due to the
forwarding cost, which becomes evident in this situation.

118
Figure 2(a): Number of clusters against 10% network lifetime
Figure 2(b): Number of clusters against 50 % network lifetime
5.2 Gateway Tour Length
To evaluate MG tour, we compared its quality against the optimal solution, which is obtained
using CPLEX. To obtain the optimal solution we modified TSP formulation to incorporate the
tour length bound. We also modified the formulation input parameters to have the clusters
obtained by MG as an input, where one node from each cluster must be involved in the tour. Due
to the NP-hardness of TSP, we limit the maximum number clusters used in this evaluation to 14
clusters and the total number of nodes to 40. Figure 3 depcits the impact of varying the number
of clusters on the tour’s travelling time. The result shows that reducing the number of clusters
reduces the gap between MGs performance and the optimal solution. This is due to the fact that
reducing the number of clusters reduces the valid tours permutation, and therefore increases the
probability that MG obtains a near-optimal solution. Also, we can see that MG is within 3/2 factor
of the optimal result.

119
Figure 3: Number of clusters against travelling time
To evalute the impact of the clusters size on the tour travelling time, we ran experiment for a 10
clusters network. Figure 4 depicts the result. The result shows that increasing the number of
sensors increases the gap between MG and the optimal solution. This is due to the fact that
increasing the number of nodes increases the space of the valid tours.
Figure 4: Number of nodes against travelling time
6. CONCLUSION
Using mobile gateway as mechanical data carriers has emerged as a promising approach wherein
sensor nodes do not need to form a connected network due to energy restrictions. In this paper,
we consider the situations where the travelling time of the gateway tour must be bounded by time
constraint to avoid sensors buffer overflow. We presented the problem of scheduling the mobile
gateways, so the time bounding is satisfied and sensors lifetime is maximized. To address this
problem, we presented an algorithmic approach, which works by partitioning the network into
clusters. The gateway tour is then planned to visit one node from each cluster. The experiments
showed that in term of network lifetime, the proposed approach significantly increases the

Data gathering in wireless sensor networks using mobile elements

More Related Content

What's hot (17)

Viewers also liked (15)

Similar to Data gathering in wireless sensor networks using mobile elements (20)

Recently uploaded (20)

Data gathering in wireless sensor networks using mobile elements