VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SENSOR NETWORKS

International Journal of Wireless & Mobile Networks (IJWMN) Vol. 9, No. 5, October 2017
DOI: 10.5121/ijwmn.2017.9507 67
VIRTUAL ARCHITECTURE AND ENERGY-
EFFICIENT ROUTING PROTOCOLS FOR 3D
WIRELESS SENSOR NETWORKS
Vianney Kengne Tchendji1
,Jean Frédéric Myoupo2
, Pauline Laure Fotso3
and
Ulrich Kenfack Zeukeng1
1
Department of Mathematics and Computer Science, University of Dschang, Cameroon
2
Lab-MIS, University of Picardie Jules Verne, 33 rue St Leu, 80039, Amiens, France
3
Department of Computer Science, University of Yaounde 1, Cameroon
ABSTRACT
This paper proposes a virtual architecture for three-dimensional (3D) wireless sensor networks (WSNs), a
dynamic coordinate system, and a scalable energy-efficient training protocol for collections of nodes
deployed in the space that are initially anonymous, asynchronous, and unaware of their initial location.
The 3D WSNs considered comprise massively deployed tiny energy-constrained commodity sensors and
one or more sink nodes that provide an interface to the outside world. The proposed architecture is a
generalization of a two-dimensional virtual architecture previously proposed in the literature, in which a
flexible and intuitive coordinate system is imposed onto the deployment area and the anonymous nodes are
partitioned into clusters where data can be gathered from the environment and synthesized under local
control. The architecture solves the hidden sensors problem that occurs because of irregularities in rugged
deployment areas or environments containing buildings by training the network of nodes arbitrarily
dispersed in the 3D space. In addition, we derive two simple and energy-efficient routing protocols,
respectively for dense and sparse networks, based on the proposed dynamic coordinate system. They are
used to minimize the power expended in collecting and routing data to the sink node, thus increasing the
lifetime of the network.
KEYWORDS
Wireless Sensor Network, Self-organization, Training protocol, Energy-efficient Routing Protocol.
1. INTRODUCTION
A wireless sensor network (WSN) is a distributed system comprising a number of tiny wireless
sensing devices that integrate signal processing and wireless communications capabilities ([1],
[2]). Each sensor is powered by a non-rechargeable and non-replaceable battery and has low
memory, computation, and transmission range capacities ([1]). Each sensor is able to harvest a set
of data in a certain environment, and transmit it in a multi-hop manner to a base station (BS)
where it is utilized. WSNs have a significant impact on various fields including military,
scientific, industrial, and healthcare. Further, the ubiquity of WSNs has resulted in them
pervading society and redefining the way in which we live and work ([3], [4], [5], [6], [7], [8],
[2]).

68
The fundamental goal of a sensor network is to produce, over an extended period of time,
globally meaningful information from raw local data obtained by individual sensor nodes.
Importantly, this goal must be achieved while prolonging as much as possible the useful lifetime
of the network and ensuring that it remains highly available and continues to provide accurate
information in the face of security attacks and hardware failure. The sheer number of sensor
nodes in a sensor network and the unique characteristics of their operating environment
(anonymity of individual sensors, limited power budget, and a possibly hostile environment) pose
several challenges for the designers of protocols. In fact, the limited power budget at the
individual sensor node level mandates the design of ultra-lightweight data gathering, fusion, and
communication protocols. An important guideline in this direction is to perform as much local
data processing at the sensor level as possible, avoiding the transmission of raw data through the
sensor network. Recent advances in hardware technology have resulted in the biggest challenge
of the sensor network community being the development of ultra-lightweight communication
protocols for activities such as training, self-organization, network maintenance, security, data
collection and fusion, and routing ([4], [9], [10]).
There are several possible models for WSNs. In this work, we consider WSNs in which all the
sensor nodes are fixed, short-ranged, and homogeneous. We assume that the BS and all the sensor
nodes have a local clock that keeps synchronous time, perhaps by interfacing with the BS.
Further, all sensor nodes run the same protocol and can perform computations on the data being
sensed. As is customary, time is assumed to be slotted and all transmissions take place at slotted
boundaries ([11], [12]). This simple model involves using one or more special sink nodes
deployed alongside the sensor nodes. Thus, the raw data collected by individual sensor nodes are
fused in stages and forwarded to the sink nodes, which provide the interface to the outside world.
However, in some applications, it is impossible or impractical to deploy sink nodes within the
sensor network. In such cases, the task of harvesting the information produced by the sensor
network and that of providing an interface to the outside world may be performed by aircraft
and/or helicopters overflying the sensor network, or by laser transmission to a satellite
constellation. In such a network, security is a crucial point that requires serious consideration. In
fact, WSNs have many constraints, including the communication medium, which is wireless:
nowadays it is very easy to read, intercept, and even modify the data transmitted, and to
compromise an entire network. In addition, the sensors’ application context is another problem, as
they are usually deployed in hostile environments. Thus, there is a need to secure the protocols in
order to guarantee authentication, confidential exchanges ([13], [14]), data integrity, and network
availability. Several security protocols have been proposed in the literature. They include TinySec
([15]) which ensure the authentication of the packets sent from a BS to all nodes (broadcast or
multicast). Ultimately, a good security system should be able to prevent external attacks (coming
from an attacker outside the network) as well as internal attacks (from an attacker inside the
network, possibly by compromising a node).
1.1 RELATED WORK
Once deployed, a fundamental prerequisite for self-organization is that sensors need to acquire
some form of location awareness, ([9], [16]). Almost all applications benefit that the sensed data
be supplemented with location information, but not all of them. The problem for a sensor to know
its location in the network is crucial in a anonymous network. One interesting paper on the self-
organization of two dimensional WSN is due to Wadaa et al., in [17]. A sensor is trained to
acquire its coordinates. A cluster is then the set of sensors having the same

69
coordinates and it results a 2D virtual WSN ([18], [19], [20]). Howard et al., ([20]) propose an
incremental algorithm for self-deployment of in 2D. However there are few papers on the design
of 3D WSN. A virtual 3D network architecture for Underwater Sensor Networks (UWSNs) is
introduced by Tamoghna et al., ([21]). The main goal of the so called EDETA (Energy-efficient
aDaptive hiErarchical and robusT Architecture) ([22]) which based on two-levels hieratical
architecture, is to optimize the save node’s power. EDETA is more suitable for the
implementation of safety applications such as a wireless fire detection system. Boufares et al.,
([16]) proposed a 3D architecture based on virtual force which is the generalization on the
cellular network in 3D. In the 3D architecture in ([23]) the sensors need to be powerful to carry
out the negotiation tactic that ensure network connectivity, and to manage a density control
strategy that is used to balance the node distribution, hence we are far from the sensors defined as
Micro Electro-Mechanical Systems (MEMS) – miniaturized low-power devices that integrate
sensing, special-purpose computing and wireless communications capabilities. Chen and Qian
([24]) derive an algorithm based on ideal fluid dynamics for deploying sensors in 3D. Huang et
al., ([25]) show how the coverage problem can be solved in 3D.
1.2 THE CONTRIBUTION OF THIS PAPER
In this paper, we focus on the concept underlying the 2D virtual architecture developed by Wadaa
et al., ([17]). We extend their work in 3D and introduce new routing protocol in case of sparse
network. The architecture is created and orchestrated by the BS, which is able to split the network
into a set of clusters according to the strength and direction of the broadcast it can perform. Our
proposed training protocol is a generalization to 3D sparse WSN of the one proposed by Wadaa et
al., ([17]) for 2D dense WSN. Our major contributions are as follows:
• Firstly, we propose a virtual architecture comprising a 3D dynamic coordinate system for
massively deployed collections of anonymous sensor nodes. This coordinate system
yields, at no extra cost, a clustering scheme: two nodes are in the same cluster only if
they have the same coordinates.
• We then show that while training the sensor nodes, the process through which nodes
learn their coordinates can be performed by a secure binary protocol. This protocol only
uses binary comparisons (that are less energy consuming compared to scalar operations)
to perform training. It is then called lightweight protocol.
• Next, we propose two energy-efficient routing protocols, respectively for dense WSN and
low-density (or sparse) WSN. The proposed protocols can be used to collect and forward
data from the sensors to the sink node. They use the dynamic coordinate system to
minimize the power expended in collecting and routing data.
The remainder of this paper is organized as follows: Section 2 presents the lightweight training
protocol proposed for virtual clusterisation. Sections 3 and 4 describe the proposed routing
protocols, respectively for dense WSN and low-density WSN, used to send the collected data to
the BS. Section 5 concludes this paper.

70
2. TRAINING A 3D WSN: VIRTUAL ARCHITECTURE
2.1. THE MAIN LINES
The practical deployment of many sensor networks will result in sensors initially being unaware
of their location: they must be trained with this vital information. Further, owing to limitations in
form factor, cost per unit, and energy budget, individual sensor nodes are not expected to be GPS-
enabled. Moreover, many probable application environments limit satellite access.
The localization problem is for individual sensor nodes to determine, as closely as possible, their
geographic coordinates in the area of deployment. The most striking solutions to the localization
problem are based on multilateration: sensor nodes receiving location messages from at least
three sources can approximate their own locations. Langendoen and Reijers ([26]) provide a good
survey of localization protocols for WSNs.) In some other applications, exact geographic location
is not necessary: all that the individual sensor node needs is coarse-grain location awareness.
There is an obvious tradeoff: coarse-grain location awareness is lightweight but the resulting
accuracy is only a rough approximation of the exact geographic coordinates. It can be made by an
overflying aircraft or helicopter. In this case, all that the individual sensor nodes need is to
determine their approximate distance to three different positions of the training agent.
Our approach is different: we obtain this coarse-grain location awareness by the training protocol
that imposes a coordinate system onto the sensor network. An interesting by-product of our
training protocol is that it provides a partitioning into clusters and a structured topology with
natural communication paths. The resulting topology makes it simple to avoid collisions between
transmissions of nodes in different clusters, between different paths, and also between nodes on
the same path. Our clustering protocol has the following desirable features:
• lightweight, as a by-product of training;
• organizes anonymous asynchronous nodes;
• a cluster is the locus of all nodes having the same coordinates; and
• individual nodes need not know the identity of other nodes in their cluster.
Hereafter we assume a WSN that consists of a sink and a set of sensors randomly deployed in its
3D broadcast range, as illustrated in Figure 1. For simplicity, we assume that the sink node is
centrally placed, although this is not actually necessary. The task of training refers to the
imposing of a coordinate system onto the sensor network in such a manner that each sensor
belongs to exactly one sector.
We assume that the sink node can make l omnidirectional transmissions, m horizontal directional
transmissions, and n vertical directional transmissions. The coordinate system divides the sensor
network area into equiangular wedges (or sections). In turn, these wedges are divided into sectors
by means of concentric spheres or coronas centered at the sink and whose radii are determined to
optimize the transmission efficiency of sensors-to-sink transmission. Sensors in a given sector are
mapped to a cluster; the mapping between clusters and sectors is one-to-one. With reference to
Figure 1, the task of training a sensor network involves establishment of the following:
1. Coronas: The deployment area is covered by l coronas determined by l concentric spheres of
radii centered at the sink node;

71
2. Horizontal wedges: The deployment area is ruled into m horizontal angular wedges centered
at the sink node;
3. Vertical wedges: The deployment area is ruled into n vertical angular wedges centered at the
sink node.
As illustrated in Figure 1, at the end of the training period, each sensor node has acquired three
coordinates: the identity of the corona in which it lies, and the identity of the horizontal and
vertical wedges to which it belongs. Importantly, the locus of all the sensor nodes that have the
same coordinates determines a cluster.
Figure 1: A trained 3D sensor network.
2.2. THE PROPOSED LIGHTWEIGHT TRAINING PROTOCOL
The main goal of this section is to present, in detail, our proposed lightweight scalable training
protocol for WSNs. The key advantage of this protocol is that each node participating in the
training, incurs an energy cost that is logarithmic in the number of clusters and wedges defined by
the protocol.
After deployment, nodes sleep until they are awakened by their individual timers. Thus, each
node sleeps for a random period of time, wakes up briefly and, if it hears no messages of interest,
selects a random number x and returns to sleep x time units. Clocks are not synchronized but over
any time interval [t, t + ∆t], a percentage directly proportional to t of the nodes are expected to
wake up briefly. During this time interval, the sink continuously repeats a call to training,
specifying the current time and a rendezvous time. Thus, in a probabilistic sense, a certain
percentage of the nodes will be selected for training. The time interval t can be adjusted to control
the percentage of nodes selected. Using the synchronization protocol proposed by Wadaa et al.,
([17]), the selected sensor nodes reset their clocks and set their timer appropriately before
returning to sleep. In fact, it is natural to assume that, just prior to deployment, the sensor nodes
are synchronized. However, because of natural clock drift, resynchronization is necessary. This is
performed with respect to the master clock running at the sink.
2.2.1. THE CORONA TRAINING PROTOCOL
The corona training protocol is similar to that developed by Wadaa et al., ([17]). However, the
corona training protocol operates as follows: formally, consider an l-leaf binary tree T and refer to
Figure 2 (borrowed from [17]). In the figure, the leaves are numbered left to right from one to l.

72
The edges of T are labeled by 0s and 1s in such a manner that an edge leading to a left subtree is
labeled 0, and an edge leading to a right subtree is labeled 1.
Figure 2: Illustration of corona training (Wadaa et al., ([17]).
In this corona training protocol, the preorder and inorder numbers of internal nodes in T
correspond, respectively, to time slots in the training protocol and to the transmission ranges used
by the sink.
2.2.2. THE WEDGES TRAINING PROTOCOL
In the wedges training protocol, the deployment area is divided into m horizontal angular wedges
with angles α and n vertical angular wedges with angles β. This is shown respectively in Figures
3(a) and 3(b). As in the corona training protocol, sensors must read a string of length(or log2n
bits for vertical angular wedges) they possess. The time is divided in slots, , and
at each slot, the sink node makes a directional horizontal transmission of angle
(respectively, a directional vertical transmission of angle
Figure 3: Horizontal and vertical wedges: (a) Horizontal wedges. (b) Vertical wedges

73
The vertical wedges are trained after the horizontal wedges. The wedge training protocol is
virtually similar to the corona training protocol, but the omnidirectional transmissions of the sink
node are replaced by the directional transmissions. The sink node uses the binary tree in Figure 2
to determine the values of the different angles to use in each slot. Figure 4 shows an example of
vertical directional transmissions for n = 8:
Figure 4: Directional transmissions of the sink node for n = 8.
In the next section, we describe how the data collected by the sensors arrive at the sink node.
3. ROUTING IN A DENSE TRAINED SENSOR NETWORK
In this section, we show how data collection is achieved in a network that we assume to be dense,
i.e., after the clusterisation protocol, we have the following property:
∀ (i, j, k) ≤ (l, m, n), cluster (i, j, k) is not empty (2)
We assume that the sensor network is already clustered according to the technique presented in
the previous section. This facilitates structuring of data routing at two levels: within a cluster and
between clusters. To achieve this goal, we introduce two concepts:
• Relay cluster: the data collected by sensors in a cluster (i, j, k) can go through many
clusters to reach the sink node. Cluster (i-1, j, k) is designated as a relay cluster of cluster
(i, j, k). It is the first cluster that transits the data collected by the sensors of cluster (i, j,
k);
• Cluster-head (or gateway node): This is a unique node in a given cluster that is
responsible for forwarding data to the outside (i.e., to the relay cluster). It is used to
conserve the energy of the sensors in a cluster and prevent unnecessary overloading of
the network.
In our routing protocol, we make the following assumptions:
1. Each node has a unique identifier ID throughout the network;
2. A sensor is able to assess its residual energy (which we denote Er);
3. A message sent by a sensor is received after a finite time (a slot) by all its neighbors;
4. The clustering is set in such a way that all the sensors of a given cluster can directly
communicate between them and with some sensors of their neighboring clusters.

74
We consider that all sensors communicate with the same frequency. Thus, there may be many
collisions if communications are not well managed. To avoid collisions between packets
transmitted, several communication channel management protocols exist, for example, frequency
division multiple access (FDMA), time division multiple access (TDMA), and code division
multiple access (CDMA). We chose the carrier sense multiple access/collision avoidance
(CSMA/CA) protocol ([27]) for the reservation of the communication channel. We chose it for
the following reasons:
1. The bandwidth is not subdivided, allowing fast transfer of data;
2. In most WSNs, the amount of data collected is not large, so they are unlikely to be
fragmented; even if that were the case, the fragmentation-reassembly
mechanism of CSMA/CA solves the problem;
3. It enables shared access to the channel, while solving interference problems, and
concurrent access of sensors;
4. Its back-off algorithm gives the same probability to all the sensors to access the channel.
As the sensors have a limited source of energy, we must prevent redundant data releases. Indeed,
it is important to reduce the network traffic to avoid collisions, conserve the energy of the
sensors, and prevent unnecessary overloading of the network. To achieve these objectives, data
collected by sensors must contain a minimum number of messages.
We use cluster-heads to satisfy these conditions. Cluster-heads are the collector nodes in a
cluster; more specifically, they gather data in their cluster before sending them to the relay
cluster. Therefore, the data held by cluster-heads can be compressed or aggregated before
transfers. The data can also be filtered to prevent the transfer of identical messages. Thus,
management of the routing within a cluster increases the energy consumption of the gateway
node compared to the ordinary nodes. Consequently, the role of gateway is given only to nodes
that have sufficient energy.
To start the routing protocol, the sink node repeats a call for routing, specifying the current time
and a rendezvous time. On that date, the sensors must collect data and send them towards the sink
node. Thus, our routing protocol is divided into two steps: (1) election of a cluster-head in each
cluster, and (2) transmission of collected data to the sink node. The second step is itself
subdivided into two phases: routing within a cluster, and routing between clusters.
3.1. DISTRIBUTED CLUSTER-HEAD ELECTION
Election of cluster-heads is conducted in a distributed manner, and at the same time in all the
clusters. A sensor can act as a cluster-head only if its residual energy Er is greater than a
threshold Es defined by a network administrator. For the gateway node (cluster-head) election
process, sensors in clusters (i, j, k), i >1, having a residual energy greater than the threshold Es,
send a Hello message1 to their relay cluster (i-1, j, k), and wait for an acknowledgment (a Hi
message2). Then, all the sensors that receive a Hi message will be sure that they can access the
sensors of their relay cluster. They postulate as gateway node by diffusing a Head message3 in
their cluster (i, j, k). The sensor having the highest residual energy Er among those that postulate
is elected gateway node. If several candidates have the same amount of residual energy, the one
with the highest identifier (ID) is elected.

75
1. The Hello message contains two pieces of information, ((i, j, k), (i-1, j, k)): the
coordinates of the current cluster and those of the relay cluster.
2. The Hi message contains two pieces of information, ((i, j, k), (i+1, j, k)): the coordinates
of the relay cluster and those of the current cluster.
3. The Head message contains three pieces of information, (ID, Er, (i, j, k)): identifier of the
current sensor, its residual energy, and the coordinates of its cluster.
The sensors in clusters (1, j, k), do not send any Hello message, instead they respond to Hello
messages sent from clusters (2, j, k) with Hi messages. Subsequently, the sensors with residual
energy greater than the threshold Es, send a Head message in their own cluster. The criteria for
selecting the gateway node are the same here as in clusters (i, j, k), i >1.
To conserve energy and prevent redundant transmission, each sensor reacts only once after the
reception of Hello and Hi messages. This first step of our protocol (distributed cluster-head
election) is performed by each sensor using pseudo-codes 1, 2, 3, 4, and 5 described in Annex 1.
At the end of this step, a cluster-head is elected in each cluster, and the next task (the first phase
of the second step) is to route the data collected towards the sink node through cluster-heads.
3.2. ROUTING WITHIN A CLUSTER
When the sensors in a given cluster wish to send messages to the sink, they send them to the
gateway node (or cluster-head), which will in turn send them to their relay cluster, and so on,
until the sink node is reached. When a sensor receives a message with a destination in its cluster,
it forwards it to the gateway node; otherwise, it ignores the message.
To prevent redundant messages being sent, each gateway node keeps track of the last message
received from each sensor within its cluster. Thus, if an identical message is received from
another sensor, it will not send it again.
When the residual energy of the gateway is below the threshold Es, it initiates re-election of the
gateway, but this time it does not postulate. If during this re-election, there is no sensor that has
residual energy greater than the threshold Es, two solutions are possible with the last active
cluster-head:
• It can decrease the threshold value (for example by half, i.e., Es/2), and inform the
sensors of its cluster. This allows the network to stay functional longer;
• It can also send a warning message to the sink node in order that, either new fixed sensors
are deployed in the cluster or a mobile sensor is sent there. In the case of a rechargeable
network, this alert message can initiate the charging mechanism of the sensors.
Remark: In our work, we assume that all the sensors in cluster (1,j,k) can directly communicate
with the sink node. In the opposite case, sensors in cluster (1,j,k) that are candidates to become
the gateway node must also verify that they can communicate with the sink node before sending
the Head message. In this case, the sink node must participate in the election process as the only
sensor in cluster (-1,-1,-1).
Once the data are forwarded to the cluster-head, it must route them wisely to sink node, possibly
through other clusters. The next task (second phase of the second step) in our protocol describes
how the messages collected in the different clusters are routed to the sink node.

76
3.3. ROUTING BETWEEN CLUSTERS
Property (2) is very interesting because cluster (i, j, k) can directly communicate with cluster
(i−1, j, k) (its relay cluster). However, the sensors in cluster (1, j, k) can directly communicate
with the sink node whose coordinates are (-1, -1, -1). The sink node is the final destination of all
the messages traveling in the network. Thus, data can successfully move from cluster (i, j, k) to
cluster (i-1, j, k), and then to (i-2, j, k),..., (1, j, k), until (-1, -1, -1).
In order to conserve energy, sensors are not awake all the time; they can automatically sleep for a
random period of time, and wake up for a period of time (defined by the network administrator)
to collect data and send them to the sink node. When sensors are awake, they can perform two
possible actions:
• An event occurs: Sensors in cluster (i, j, k) make a message containing the code of the
event and the coordinates of the cluster in which it lies, then send it to its cluster-head, for
it to forward to the relay cluster, or send it directly to the sink node if i = 1;
• Reception of a message: The sensor checks if it is the receiver; in this case, it forwards it
to its cluster-head, or to the sink node if its coordinates are (1,j,k). Otherwise, it simply
ignores the message.
This process is illustrated by Figure 5, in which it can be seen how the data are collected within a
cluster and between clusters. The dotted arrows show that ordinary sensors transfer the data they
hold only to their cluster-head, and never out of their cluster. The continuous line arrows show
that communication between clusters is made only by the cluster-heads, and always in the same
angular wedge.
Figure 5: Routing in a dense sensor network.
This protocol has a number of advantages:
1. The CSMA/CA protocol minimizes collisions during transmissions;
2. The data collected in a cluster can be grouped, aggregated, or compressed before routing,
which helps to reduce the amount of information flowing through the network, and the number of
messages generated by the sensors in a cluster;
3. Routing is carried out by cluster-heads, allowing ordinary sensors to conserve energy;
4. Data collected in an angular wedge never go through another, this prevents overall network
congestion;
5. The cluster-head re-election mechanism ensures the longevity of the network structure, and
provides a first step in fault tolerance.

77
3.4. SIMULATION RESULTS
In the simulation conducted, we focused on the energy percentages of the sensors nodes and the
cluster-head during the routing process in the 3D virtual architecture described above. Our
simulations were performed using the WSNet software ( [28]) based on a virtual architecture of 5
km in diameter, 10 coronas, eight horizontal wedges, and eight 45° vertical wedges, on which we
randomly generated dense networks of 1000 sensors with a range of 500 meters. The ideal case,
in which the BS is located at the center of the area, was used.
We performed tests repeatedly and averaged the results. The energy model used is one that has
been utilized in many efficiency studies ([29]): E = ET + ER + ES + EI, where ET and ER are,
respectively, the total energy used by the sensors for transmissions and receptions in the network;
ES and EI are, respectively, the energy expended by the sensors while sleeping or simply awake.
The curves were made with version 5.0 of the gnuplot software.
While ET ≥ ER ≥ ES ≥ EI, the cluster-heads lose much more energy than the ordinary nodes. In
fact, they make more transmissions in their cluster. This is illustrated in Figure 6, where the
energy of the cluster-head decreases faster than that of ordinary nodes. On the other hand, the
sharp drop in energy in the first slots corresponds to initialization of the communication phase,
which includes the election of cluster-heads. Once the cluster-heads are elected, routing is
performed with lower energy consumption.
Figure 6: Energy consumption of a sensor during routing.
For a deployment of 500 sensors in a virtual architecture containing four coronas, and 90° vertical
and horizontal wedges, there are 64 clusters. Thus, there are 64 cluster-heads, which have a larger
workload than ordinary nodes; i.e., a ratio of 12.8% of sensors. However, with 1000 sensors, that
ratio decreases to 6.4% of sensors. Therefore, this percentage decreases as the number of sensors
increases. Considering the re-election of cluster-heads when the energy of the current cluster-
head is less than the threshold, our protocol provides several cluster-head generations in each
cluster, and thus increases the lifetime of the network.
Figure 7 shows that without cluster-heads, the energy of the sensors decreases very rapidly,
whereas with cluster-heads, the overall energy of the sensors decreases very slowly. Our method
of reducing transmission of identical messages further conserves this energy.

78
The protocol presented in this section is not suitable for sparse networks, in which the sensors
partially occupy the deployment area. In this case, routing of data to the sink node can no longer
follow the clear and optimal path (angular wedge) used for dense networks.
Figure 7: Impact of the use of cluster-heads on the conservation of the energy of ordinary sensors
4. ROUTING IN A SPARSE TRAINED SENSOR NETWORK
Training does not allow the sink node to distinguish empty clusters and non-empty clusters. In a
sparse network, the following property holds:
In contrast to dense networks, in which the data collected by cluster (i, j, k) are routed to the sink
node through cluster (i-1,j,k), in a low-density network, there is no guarantee that cluster (i-1,j,k)
will contain sensors that can forward the messages. Thus, an empty clusters detection phase must
precede routing of data in order to define the best path for the data to follow. This empty clusters
detection phase also makes it possible to determine the area actually covered by the sensors after
the deployment, and therefore gives the possibility of reacting accordingly.
A cluster is considered empty if it contains no sensor, or if it contains a set of sensors
disconnected from the rest of the network.
4.1. DETECTION OF EMPTY CLUSTERS AND DISTRIBUTED CLUSTER-HEAD ELECTION
Knowledge of the distribution of the sensors in the space of interest enables the sink node to
determine the optimal path from a cluster to the sink (or to another cluster).
For l coronas, m horizontal wedges, and n vertical wedges, there are (l × m × n) clusters in the
virtual architecture; therefore, for each message received, the sink regularly updates two tables
h(l, m, n) and relay(l, m, n). Each entry (i, j, k) of table h(l, m, n) contains one if cluster (i, j, k) is
not empty and zero otherwise, allowing the sink to obtain a global view of the sensor’s
distribution; and each entry (i, j, k) of the table relay(l, m, n) contains the coordinates of the relay
cluster of the cluster (i, j, k).

79
Because cluster-heads significantly affect conservation of the energy of the ordinary sensors, as
well as the routing of data, they must be elected as soon as possible. We propose to realize
distributed cluster-heads election during the empty clusters detection phase.
4.1.1. THE SINK’S ALGORITHM
At the beginning, the sink periodically broadcasts the date on which the detection algorithm will
begin. All sensors are awake when the sink initiates the detection. It then transmits its
neighboring clusters (those of the first corona) a Detect message containing its coordinates (-1, -
1). Then, it waits for acknowledgments (ACK messages) that will allow it to update tables h(l, m,
n) and relay(l, m, n). Owing to network connectivity, it is certain that at least one sensor will
receive this message. During the process, each message transmitted by a sensor towards the sink
contains the coordinates of its cluster and those of its relay cluster. Table h(l, m, n) is also
initialized to zero. At each reception of a message from a sensor in a cluster (i, j, k), the sink node
puts one in h(i, j, k) and, in the entry relay(i, j, k), it assigns the coordinates of its relay cluster
obtained from the variable relay of the received message. At the end of the algorithm, cluster (i, j,
k) is considered empty when h(i, j, k) = 0, and the relay cluster of cluster (i, j, k) is determined by
the value of the entry relay(i, j, k).
4.1.2. THE SENSOR’S ALGORITHM
The network must be connected; therefore, for all clusters (i, j, k) considered non-empty, there is
always a path from it to the sink node. Isolated clusters cannot reach the sink and are considered
empty even if there are not. For the sensors, there are three main events in the detection of empty
clusters: reception of a Detect message asking sensors to indicate their coordinates (here, this is
equivalent to the reception of a Hi message in Section 4.1); reception of a Head message sent by
a sensor of the same cluster to postulate as cluster-head (it starts the cluster-head election
process); and reception of an ACK message sent by a cluster-head to the sink node to indicate its
coordinates and those of its relay cluster.
Reception of a Detect message
In order to reach the sink by the shortest path, the sensors of each cluster must choose their relay
cluster wisely. To accomplish this, each sensor has to route only the first Detect message that it
receives. When a sensor receives a Detect message, it checks if it has not already had to route a
Detect message. If it has not, it forwards it to enable the sensors in the other clusters to signal
their coordinates. Otherwise, it ignores the message. Subsequently, each sensor that received its
first Detect message, constructs a Head message and forwards it in its cluster to postulate as
cluster-head.
Reception of a Head message
This election is carried out as in Section 4.1. At the end of the cluster-head election, the elected
cluster-head sends an ACK message to the sink through their relay cluster. The cluster-heads of
the first corona send their ACK message directly to the sink, while those of the other coronas send
theirs on through the relay cluster.
Reception of an ACK message
A sensor that receives an ACK message from its neighbor node checks whether the message is for
its cluster and, if it is, it sends it to its gateway node (cluster-head), which checks if it has already
routed an ACK message from the same cluster. If it has not, it sends the message through its relay
cluster. Otherwise, it simply ignores the message.

80
4.2. ROUTING
The sink node uses tables h(l, m, n) and relay(l, m, n) to build a message propagation tree that
presents the path traveled by the messages from each cluster. This tree can be optimized (for
example, by decreasing its height) to decrease the distance between the sink and the different
clusters.
To send the messages to the sink node, the sensors in cluster (i, j, k) send them to their relay
cluster (through their cluster-head), which in turn transmits them to their relay cluster, and so on,
to the sink node. Figure 8 illustrates this communication process. Clusters (i-1, j, k), (2, j-1, k),
and (1, j-1, k) are empty. It can be seen that cluster (i, j-1, k) is the relay of cluster (i, j, k), and the
messages coming from cluster (i, j, k) pass through clusters (i,j-1,k), (i-1, j-1, k), (i-2, j-1, k), …,
(2, j-1, k), and (1, j-1, k) to reach the sink node. These messages pass alternately through two
angular wedges.
Figure 8: Routing in a sparse sensor network.
The methods used for channel access and communication within and outside of clusters are the
same as those used for routing in a dense sensor network.
4.3. SIMULATION RESULTS
For a virtual architecture comprising eight coronas and 45° horizontal and vertical angular
wedges, i.e., a total of 512 clusters, Figure 9 shows the average percentage of empty clusters
obtained after 10 executions of our empty clusters detection algorithm. These tests were
conducted with a random deployment of the sensors on the space of interest. The number of
sensors varied between 100 and 10000. It can be seen that for these tests, more than half of the
clusters are empty, which sufficiently illustrates the need for the empty cluster detection phase.
By depicting the deploying of sensors on approximately 50% of the space of interest, Figure 10
clearly shows that cluster-heads consume more energy than ordinary sensors. Because there are
fewer sensors in the network (compared to a dense network), the energy difference between
cluster-heads and ordinary sensors is lower than that in a dense network. The rapid decrease in
energy in the first slots is due to the empty clusters detection phase and cluster-heads election.
It is still important to note that the simulations were performed under ideal circumstances: the
area of the sensors is spherical, the BS is at the center, and environmental constraints were not
considered. Evidently these ideals are rarely met (the sink node can be at the network edge,
sensors have rectangular area, etc.).

81
Figure 9: Percentage of empty clusters.
Figure 10: Energy consumption during routing in a sparse WSN.
5. CONCLUDING REMARKS
This paper proposed a virtual architecture for 3D WSNs and a dynamic coordinate system for
collections of anonymous sensor nodes deployed in the space. The coordinate system provides, at
no extra cost, an interesting clustering scheme in which two nodes are in the same cluster only if
they have the same coordinates. It is important to note that this clustering scheme operates for
anonymous sensor nodes. Thus, sensor nodes do not know the identity of the other nodes in the
same cluster. A lightweight and secure protocol for use by the sensors, during training, to learn
their coordinates was also proposed. As it is energy efficient, this training can be repeated on
either a scheduled or an ad hoc basis to provide robustness and dynamic reorganization. We also
showed that in a trained WSN the tasks of routing and data fusion can be performed by very
simple and energy-efficient protocols.
Two simple and energy-efficient routing protocols for dense and sparse networks, based on the
dynamic coordinate system, and which minimize the power expended in collecting and routing
data to the sink node were also presented. Further, the concepts of cluster-head and relays cluster,

82
which facilitate centralized data compression or data aggregation in a cluster before transfers
were also introduced for more efficiency. They are also used to avoid network congestion, and to
preserve the overall energy of the sensors; thereby increasing the lifetime of the network.
However, despite these encouraging results, there is more work to be done. Firstly, in our model,
synchronization of sensors with the global clock of the sink node after a period of sleep, has not
yet been actualized. Further, the channel access method needs to be improved. Secondly, new
challenges have appeared in sparse sensor networks. These challenges include management of
network connectivity, routing optimization by seeking an optimal means of reaching the sink
node, and management of mobile sensors in areas not covered by fixed sensors. Finally, because
sensors are often deployed in unattended areas, they can be subject to external attack. Thus,
security is also a major issue for routing in this architecture.
APPENDIX
Cluster-heads election pseudo-code
The pseudo-codes presented in this section enable sensors to choose a gateway node in each
cluster. The notations presented in Table 1 are used.
Table 1: Parameters of the training and routing pseudo-code
The init() procedure (Algorithm 1) facilitates initialization of the cluster-head election process in
each cluster by broadcasting the message Hello. It helps the sensors that have sufficient energy to
test the connectivity to their relay cluster. The reception_Hello() procedure (Algorithm 2)
describes the behavior of the sensors while receiving the Hello message, it enables them to send
an acknowledgment (Hi message) to the sensors in the

83
cluster who want to become cluster-heads.
The reception_Hi() procedure (Algorithm 3) describes the actions performed by a sensor on
receiving a Hi message; it enables the sensor to initialize the coordinates of its relay cluster, as
well as the information to offer as cluster-head. The Head message containing id and the residual
energy of the sensor is diffused in the cluster to postulate the sensor as gateway node. The
reception_Head() procedure (Algorithm 4) is used on receiving a Head message to choose the
best cluster-head among the applicants.

84
The election_gateway() procedure (Algorithm 5) makes calls to all the other procedures to
complete the gateway node election process. It comprises two phases: In the first phase, sensors
with residual energy greater than Es verify whether they can communicate with the relay cluster
(procedures 1, 2, and 3). In the second phase, the choice for gateway node is made.
The variables already_receive_Hello and already_receive_Hi ensure that a sensor reacts only
once after reception of Hello and Hi messages.

85
Each sensor makes only one call to the init() procedure and reacts only once on receiving Hello
and Hi messages. As each sensor can be a candidate in the gateway node election, a sensor can
respond several times when a Head message is received. As we assumed the network is dense, we
cannot predict the number of sensors in each cluster. Consequently, we cannot predict the number
of times Procedure 4 will be called. Therefore, it is up to the network administrator to
parameterize the duration of each phase of Algorithm 5. Nevertheless, the duration of each phase
should enable the execution of Procedures 2 and 3 once, and the execution of Procedure 4 several
times.
Corollary: The times T1 and T2 of each phase of Algorithm 5 should be determined according to
the number of slots necessary for execution of Procedures 2, 3, and 4 and to the maximal number
of sensors in a cluster.
REFERENCES
[1] Akyildiz, I. F., Su, W., Y. Sankarasubramaniam, Y. and E. Cayirci, E. (2002) ‘Wireless sensor
networks: A survey, Comput. Networks’, 38(4), 393-422.
[2] Zhirnov, V. V. and Herr, D. J. (2001). ‘New frontiers: Self-assembly and nanoelectronics’.,
Computer, 34(1), pp. 34-43.
[3] Agre, J. and Clare L. (2000). ‘An integrated architecture for cooperative sensing networks’,
Computer, 33(5), pp. 106-108.
[4] Dargie, W. W. and Poellabaurer, C. (2010) ‘Fundamentals of Wireless Sensor Networks: Theory and
Practice,, Wiley.
[5] Intanagonwiwat, C., R. Govindan,R. and D. Estrin, D. (2000) ‘Directed diffusion: A scalable and
robust communication paradigm for sensor networks’. MOBICOM, pp. 56-67.
[6] Perrig, A., Szewczyk, R., Tygar, J. D., Wen, V. and Culler D. E. (2001) ‘Spins: Security protocols
for sensor networks’, In: Proceedings of the 7th Annual ACM/IEEE International Conference on
Mobile Computing and Networking, pp. 189-199.
[7] Shen, C. C., Srisathapornphat, C. and Jaikaeo, C. (2001) ‘Sensor information networking architecture
and applications’, IEEE Pers. Commun., 8(4), 52-59.
[8] Tilak, S., Abu-Ghazaleh, N. B. and Heinzelman, W.(2002) ‘A taxonomy of wireless micro-sensor
network models’. ACM SIGMOBILE Mobile Computer Communication Review., 6(2), 28-36.
[9] El Emary,I. M. M. and S. Ramakrishnan, S. (2013) ‘Wireless Sensor Networks: From Theory to
Applications’, CRC Press.
[10] Li, Y. and Thai, M. T. (2013). Wireless Sensor Networks and Applications, Springer.
[11] Bar-Yehuda, R., Goldreich, O. and Itai, A. (1991) ‘Efficient emulation of single-hop radio network
with collision detection on multi-hop radio network with no collision detection’. Distributed
Computing, 5(2), pp. 67-71.
[12] Bertzekas,D. and Gallager R. (1992) Data Networks, Second Edition, Prentice-Hall.
[13] Li, C., Wang, Z. and Yang, C. (2011) ‘Secure Routing for Wireless Mesh Networks’, International
Journal of Network Security, 13(2), 109-120.

86
[14] Saroit, I. A., El-Zoghdy, S. F. and Matar, M. (2011) ‘A scalable and distributed security protocol for
multicast communications’, International Journal of Network Security, 12(2), 61-74.
[15] Karlof, C., Sastry, N. and Wagner, D. (2000) ‘TinySec: A link layer security architecture for wireless
sensor networks’, Proceedings of the 2nd International Conference on Embedded Networked Sensor
Systems, pp. 162-175, ACM.
[16] Boufares, N., Khoufi, I., Minet, P., Saidane, L. and Ben Saied, Y. (2015) ‘Three dimensional mobile
wireless sensor networks redeployment based on virtual forces’, IEEE Wireless Communications and
Mobile Computing Conference (IWCMC), pp.563-568.
[17] A. Wadaa, A., S. Olariu, S., L. Wilson, L., M. Eltoweissy, M. and K. Jones K..(2005) ‘Training a
wireless sensor network’, Mobile networks and Applications, 10(1-2), 151-168, 2005.
[18] Bertossi, A. A., Olariu S., Pinotti M. C. (2008) ‘Efficient corona training protocols for sensor
networks’, Theoretical Computer Science, 402(1): 2-15.
[19] Olariu, S., Wadaa, A., Wilson, L. and Eltoweissy, M. (2004) ‘Wireless sensor networks: leveraging
the virtual infrastructure’. IEEE Network 18(4): 51-56.
[20] Howard, A., Mataric, M.J. and Sukhatme, G. S. (2002) ‘An Incremental Self-Deployment Algorithm
for Mobile Sensor Networks’, Autonomous Robots, Special Issue on Intelligent Embedded Systems,
13(2), pp. 113-126.
[21] Tamoghna, O., Manas, M. and Studip M..(2013) ‘Tic-Tac-Toe-Arch: a self-organising virtual
architecture for Underwater Sensor Networks’. IET Wireless Sensor Systems, vo. 3, Issue: 4, pp. 307-
316.
[22] Capella, J. V., Bonastre, A., Serrano, J. J. and Ors, R. (2009) ‘A New Robust, Energy-efficient and
Scalable Wireless Sensor Networks Architecture Applied to a Wireless Fire Detection System’,
International Conference on Wireless Networks and Information Systems, WNIS '09. pp. 395 – 398.
[23] Miao, C., Dai, G., Zhao, X., Tang, Z. and Chen, Q (2014) ‘3D Self-Deployment Algorithm in Mobile
Wireless Sensor Networks’, In: Sun L., Ma H., Fang D., Niu J., Wang W. (eds) Advances in Wireless
Sensor Networks-CWSN 2014. pp. 27-41.
[24] Chen, J., Qian, H. (2014) ‘Three-dimensional deployment algorithm based on ideal fluid dynamics
model for mobile sensor networks’, Computer Modelling & New Technologies 18(9)12-18
[25] Huang, C-F., Tseng, Y-C. and Lo L-C. (2007) ‘The coverage problem in three-dimensional wireless
sensor networks’, Journal of Interconnection Networks, 8, Issue 03, p. 209.
[26] Langendoen, K.. and Reijers, N. (2003). ‘Distributed localization in wireless sensor networks: A
quantitative comparison’, Computer Network, 43(4), 499-518.
[27] Tanenbaum A. and Wetherall D. (2011) ‘Computer Networks’, Pearson Education, Inc/ Prentice Hall.
[28] WSNet, [online]. http://wsnet.gforge.inria.fr. (2016 ).
[29] A. Allirani, A. and M. Suganthi M.(2008) An energy sorting protocol (ESP) with low energy and low
latency in wireless sensor networks. International Journal of Computer Science and Network
Security, 8(11), pp. 208-214.

87
AUTHORS
Vianney Kengne Tchendji is a Senior Lecturer of Computer Science at the University
of Dschang, Dschang, Cameroon. He received his PhD in Computer Science from the
University of Picardie-Jules Verne, Amiens, France, in 2014. His current research
interests include network virtualization, parallel algorithms and architectures,
scheduling, wireless communication, ad hoc and sensor networking.
Laure Pauline Fotso is Professor of Computer Science at the University of Yaounde 1,
Cameroon. She received her PhD in Computer Science from the Rensselaer
Polytechnic Institute, Troy, New York in 1981. His current research interests include
parallel algorithms, Multiple Objective Linear Programming, Optimization and
Constraints Satisfaction Problems.
Jean Frédéric Myoupo is Professor of Computer Science at the University of Picardie-
Jules Verne, Amiens, France. He was Dean of Faculty of Mathematics and Computer
Science from 199 to 202. He received his PhD in Applied Mathematics from the Paul
Sabatier University of Toulouse, France in 1983 and his Habilitation in Computer
Science from the University of Paris 11, Orsay, France in 1994. He has been lecturer at
the University of Sherbrooke, Canada, 1983-1985. Heis a past member of IEEE
Computer, ACM and SIAM. His current research interests include parallel algorithms
and architectures, wireless communication and mobile computing, ad hoc and sensor
networking.
Ulrich Kenfack Zeukeng is a PhD student at the university of Dschang, Cameroon. He
received his Master degree in Computer Science in 2015 from the university of
Dschang. His current research interests include wireless communication and ad hoc
networking.

VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SENSOR NETWORKS

More Related Content

What's hot (19)

Similar to VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SENSOR NETWORKS (20)

Recently uploaded (20)

VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SENSOR NETWORKS