Analysis of methods for reducing topology in wireless sensor networks

Full text

Analysis of Methods for Reducing Topology in
Wireless Sensor Networks
Oleksii Popov
Infocommunication System Department
Kharkiv National University of Radio Electronics
Kharkiv, Ukraine
ukrbarbarian@gmail.com
Ievgeniia Kuzminykh
Computer Science Department
Blekinge Institute of Technology
Karlskrona, Sweden
ievgeniia.kuzminykh@bth.se
Abstract—This paper describes the phases of deployment of
wireless sensor networks, and more attention is given to the phase
of reducing topology and to methods by which this can be
achieved. A comparison of hierarchy-based topology construction
algorithms is presented. Algorithms use a simple, distributed and
energy efficient topology mechanism that discovers a optimal
connected dominant set (CDS) to disable unnecessary nodes,
providing important wireless sensor network characteristics such
as full coverage and connectivity.
Keywords—Wireless Sensor Network, Topology Control,
Topology Construction Algorithm, Critical Transmission Range,
WSN, TC, CTR
I. INTRODUCTION
Nowadays the task of building distributed systems for data
collection, management and monitoring is more relevant than
ever in a wide variety of application areas. Collecting data from
places that are not easily accessible or that are dangerous to
humans requires the right technology. Using wireless sensor
networks within small devices allows create a system that
provides continuous access to information about the status of
the objects served.
Since nodes are generally randomly located over the huge
territory and not under control, they must be co-located by the
location system and self-organized during utilization of sensor
network. They can be located by:
- dropping from the helicopter;
- placing them in a rocket;
- thrown by a catapult;
- individually by a human or robot.
From hundreds to several thousand nodes can be placed on
the territory of the sensor network depending on application
where wireless sensor network is deployed. It could be nature
and agriculture monitoring, bridge and building health
monitoring, smart city and smart home applications [1-3].
Nodes can be of ten meters from each other. The density of
nodes can be higher than 20 motes per meter. The dense
arrangement of multiple nodes requires careful network
maintenance because too much dense network is not good.
Very dense network will increase the number of message
collisions and will provide several copies of the same
information from similarly located nodes. To avoid this
problem, the number of active nodes and active links for saving
resources for future maintenance should be reduced. The
reducing the initial topology is the main goal of topology
control algorithms.
In this paper we consider existing methods of reducing
initial topology and further control of it. Using simulator tool
we also will make testing of different topology control
algorithms with such parameters of connectivity and coverage,
as critical transmission range, activity of nodes and energy
consumption.
II. PHASES OF TOPOLOGY DEPLOYMENT
There are three phases of topology deployment. Two of
them are described well in [4] and have names of topology
construction and topology maintenance phases respectively.
Below one more phase is represented that provides including
of new node in network.
A. Initial deployment
It can consist of a mass spread of nodes or the installation of
each separately. In this phase all nodes have just been deployed
and they are exploring their neighborhood, all the sensors are
set to transmit at their maximum power. The resultant graph
called the Maximum Power Graph, or MaxPower graph is
shown on Fig. 1, (a). This graph represents the maximum
topology of the network in terms of active nodes and active
links.
a) b)
Fig.1. Comparison of topologies: the (a) MaxPowerGraph versus (b) the
reduced topology.
This paper is part of research project at Blekinge Institute of Technology
sponsored by Swedish Institute scholarship program.

Fig. 1, (a) represents MaxPower graph for 500 nodes
uniformly distributed in an area of 500x500m with transmission
ranges of 80m.
The MaxPower graph is not a good graph for
communication model. It is cause of such negative effects as:
- excess of interference and collision;
- reduced capacity;
- redundant information due the fact that nodes from the
same location are sensing the same events, and thus, sending
the same information.
These problems are a major source of energy consumption
and topology needs to be changed with keeping important
characteristics like connectivity and coverage. On fig.1, (b) it is
represented reduced topology that provides connectivity and
coverage for the same topology as on fig.1, (a).
There are two main methods to reduce the topology of a
network [5], and therefore save energy and increase the
network lifetime:
1. Method based on TX range changing.
2. Method of creating a hierarchy.
First category can be related to such algorithms of changing
TX range as Spanning Tree Based (LMST, iMST), Graph based
(Gabriel graph, Relative neighborhood graph, Delaunay
triangulation of a graph), Neighbor based (KNeigh, XTC),
Routing based (COMPOW), Location based (GAF), Cone
based topology control.
Second category can be related to such algorithms of
creating hierarchy as building of Minimal Connected
Dominating Set (Energy Efficient Connected Dominating Set,
EECDS, A3, CDS-Rule-K), Cluster based (Low-Energy
Adaptive Clustering Hierarchy, LEACH), Backbone based
(SPAN).
By modifying such parameters as the state of the nodes
(active, inactive) and the radio transmission power the topology
of the network can be changed. The first parameter allows the
user to reduce the number of active nodes, which affects density
in certain areas, and thus reduce the number of collisions and
redundant data. Inactive nodes disconnect their transceivers and
go into a mode with very low power consumption, from which
they can be turned on again to be part of the active network, if
necessary in the phase of maintenance. The second parameter
has a direct impact on energy consumption and collisions, given
that radio transmission is the most expensive operation in terms
of energy consumption and one of the most frequently
performed. In other words, reducing the energy needed to
transmit the message will be an important saving.
B. Phase of updating the reduced network topology
In this phase the network starts consuming its energy until
the point it stops being optimal, network must restore the
reduced topology periodically in order to preserve connectivity,
coverage, density and any other metric that the application
requires.
After the deployment of the network, changing its topology
is associated with changes in the characteristics of the nodes
[4]. Topology could change such characteristics of node as
position, availability (due to interference, noise, moving
obstacles, etc.), battery charge, malfunctions, change of tasks.
Having the fact that nodes can become inaccessible and subject
to frequent failures due to battery discharge or destruction
makes network maintenance a difficult task. Sensor networks
with high mobility of nodes are possible. In addition, nodes and
networks perform various tasks and may be subject to
deliberate interference. Thus, the structure of the sensor
network is prone to frequent changes after deployment.
There are several methods to change the reduced topology:
1. Recreation
- create a new one “on the fly” (Dynamic);
- global or local.
2. Rotation
- rotate among several pre-calculated reduced
topologies (Static).
3. Rotate until it is possible, then restore (Hybrid).
Static methods build various reduced topologies during the
first initial phase of topology constructing and store them in
memory. Depending on the triggering criteria, the topology is
replaced by the next in the list. Criteria can be Time-based,
Energy-based, Random, Failure-based or Density-based. The
main purpose of this approach is to spread the network activity
among disjoint sets of nodes. Static methods take extra time in
the initial stage of building a topology, but it works faster in the
switching process than building a new topology on the fly.
The methods of static rotation of topology may have some
disadvantages. For example, it is difficult to know the pre-
consumed energy of each topology. Therefore, the method can
select some nodes in more than one topology that may not be
available or will make the topology shorter time period than
expected. Another drawback is that including of new nodes to
topology is not considered when switching topologies.
Dynamic Methods don’t make apriori calculations, but
change topology "on the fly." Thus, dynamic methods usually
require more time and energy, as they may need to run the
topology construction process several times. Therefore, the
energy efficiency of the algorithm for constructing the initial
topology plays an important role. Dynamic methods consider
the current status of the network when performing their
calculations that leads to a better or more adequate updated
topology in comparison with static methods. Example of
applying rotation algorithm to topology of area 300x300 m with
200 nodes is shown on Fig.2.
Fig.2. Rotation method is applied to reduced topology.

Hybrid methods are a mixture of Static and Dynamic
approaches. In this case the network works with a number of
predefined reduced topologies that rotate during the lifetime of
the network, just like in the Static. When a reduced topology
determines that it is not working effectively, it will invoke a
Dynamic technique to update and rebuild the reduced topology
with the current resources. Allows update of the predefined
static topologies and reduced the overhead of the dynamic
approach. The triggering criteria has a big responsibility on
determining when the network is not working properly.
C. Phase of including of new nodes
Additional nodes can be included at any time to replace
faulty nodes or due to changing tasks. Adding new nodes
creates the need to reorganize the network. Struggling with
frequent changes in the peer-to-peer network topology, which
contains many nodes and has very strict limitations on power
consumption, requires special routing protocols.
Despite the fact that a huge number of sensors and their
automatic deployment usually excludes their placement in
accordance with a carefully designed plan, the schemes for
initial deployment should: reduce the cost of installation; To
eliminate necessity, in any preliminary organization and
preliminary planning; Increase placement flexibility; Promote
self-organization and resiliency.
III. ANALYSIS OF ALGORITHMS FOR REDUCING TOPOLOGY
Under analysis in the paper hierarchy-based topology
construction algorithms with CDS-based solutions for reducing
topology.
The idea of such type of algorithm is creating a backbone
that supports communication. Also, the nodes that are not part
of the backbone can be turned off. Consider what algorithm can
be used for backbone-based topology construction.
1. Grow a Tree: A3
Start at one point and add new nodes to the existing tree like
in Prim´s algorithm. A3 calculates a Connected Dominating Set
(CDS) on the initial MaxPower topology, leaving in active state
only the dominating nodes which provide connectivity and
coverage in the network, and turning off all dominated
(redundant) nodes, which are considered unnecessary for the
correct activity of the network at the time of execution.
The algorithm is based on the method of a growing tree and
uses a selection metric based on the remaining energy at the
nodes and distance between them. The metric allows the
network to choose between a more reliable short-lived network
with a large number of nodes and a less reliable long-lived tree
with fewer nodes.
2. Maximal Independent Set (MIS) based: EECDS
The algorithm creates asset of maximum number of
independent nodes(which do not share links) in the first phase ,
and then selects gateway or extra nodes to connect the
independent sets during the second phase.
3. Pruning based: CDS Rule K
The algorithm calculates a topology that guarantees
connectivity by including most of the nodes, and then turning
unnecessary nodes off. The algorithm works also in two phases.
The first phase involves creating an initial CDS tree. For the
second phase, the algorithm turns off unnecessary nodes
applying one of the pruning rules The algorithm can discards
nodes whose neighbors form a connected graph (there is a path
between any pair of nodes) or can discards all nodes whose
neighbors are covered by other active nodes with higher priority
(priority pre-assigned).
IV. SIMULATION RESULTS
In order to provide a comparison among topology control
algorithms, the following performance metrics were chosen to
estimate the performance:
1) number of active nodes;
2) number of messages used in the CDS building process;
3) amount of energy used in the process.
Number of active nodes reflects the quality of nodes
choosing policy and has a direct impact on the network lifetime.
Number of messages shows complexity of message
exchange to build initial and after reduced topology that has
direct impact on energy consumption and on scalability of the
protocol.
Ratio of spent energy shows how much energy use
algorithm to build reduced topology, in other words shows a
cost of algorithm.
Three algorithms mentioned in chapter III and also
operation of sensor network without TC were implemented in
the Java-based event modeling tool called Atarraya [6],
designed to test topology control algorithms. Table I
summarizes the simulation variables used in the experiments.
The first experiment involves a fixed number of devices, but
changing the communication range of the nodes. The
communication ranges used in the experiment are calculated
according to equation of Penrose [7] and depend on the defined
number of nodes n and deployment area. In this case, Penrose
found that for networks with high density the length of the
longest edge, that is proportional to the value of the critical
transmission range (CTR), can be calculated, is determined
with high probability by Equation (1):
, (1)
,
where is an increasing function of , and is the
natural logarithm of ( ).
An experiment was also conducted when the number of
nodes changed and the communication range remained fixed. In
these two experiments, the nodes are distributed uniformly in
the deployment area. To improve accurancy of simulation
results all of the experiments show the average result of 50
random scenarios. Eelec is the energy used by the electronic

components of the radio, and Eamp is the energy used by the
radio amplifier.
TABLE I. INPUT DATA FOR SIMULATION
Experiment 1
Experiment 2
Deployment area
200mx200m
200mx200m
Number of nodes
100
10, 20, 40, 60, 80
and 100
Transmission
Range
1, 1.5, 2, 2.5 and
3xCTR(100) is
equivalent to: 27m,
41m, 54m, 68m and
81m
60m is equivalent to
1xCTR(10)
Node Distribution
Uniform (200,200)
Uniform (200,200)
Instances per
topology
50 instances
Emax
1 Joule
A3 Weights
WE = 0.5;WD = 0.5
Energy
Consumption
Eelec = 50nJ/bit; Eamp = 10pJ/bit/m2
Short Messages = 25Bytes (Hello, Parent
Recognition and Sleeping Messages)
Long Messages = 100Bytes (Children
Recognition and Data messages)
Idle state assumed negligible
A. Experiment 1
The main purpose of first experiment is to compare the
algorithms when the transmission range is changed. Since
these algorithms work on the basis of information from
neighbors, it is important to measure their performance with
transmission range of different sizes. The results are performed
on Fig.3. The curves with results for number of sent message
are omitted for brevity but they show the same dependencies as
for spent energy results.
Fig.3. Changing the transmission range of nodes.
B. Experiment 2
The main purpose of second experiment is to compare the
results obtained with algorithms when the network density is
changed by changing the number of nodes in a given area with
a fixed communication range of 60 m. This value is the value
calculated by Penrose formula (1) for 10 nodes. This
experiment is important to show how scalable algorithms are
in dense topologies and how the energy utilization depends on
the number of nodes. Results of simulation are shown on Fig.4.
Fig.4. Changing the network density.
V. CONCLUSIONS
Comparison of known mechanisms of reducing topology
shows that algorithm A3 shows the best results from the point
of view of energy efficiency. Simulation experiments show
that to achieve full communication coverage all algorithms
require approximately the same number of active nodes: 5-8%
and 41-42% of total number of nodes for dense and sparse
topologies, respectively. Algorithm A3 shows less number of
messages to build a reduced topology in comparison with in
EECDS and CDS Rule K algorithms, which limits overhead
and energy use. Energy is very important characteristic in
order to use this algorithm in a complete topology control
solution where the CDS tree have to be changed many times.
This message complexity is particularly noticed for EECDS in
dense networks because of the network congestion and
collisions that it generates.
REFERENCES
[1] D. Chaudhary, S. Nayse and L. Waghmare, "Application of Wireless
Sensor Networks for Greenhouse Parameter Control in Precision
Agriculture", International Journal of Wireless & Mobile Networks, vol.
3, no. 1, pp. 140-149, 2011.
[2] G. Zhou and T. Yi, "Recent Developments on Wireless Sensor Networks
Technology for Bridge Health Monitoring", Mathematical Problems in
Engineering, vol. 2013, pp. 1-33, 2013.
[3] I. Kuzminykh and S. Kabali, "Development of Smart
Infocommunication Networks for Intellectual Municipal Services",
in Second International Scientific-Practical IEEE Conference PIC S&T,
2015, Kharkiv, Ukraine, 2015, pp. 128-130.
[4] M. Labrador and P. Wightman, Topology control in wireless sensor
networks. Netherlands: Springer, 2009, p. 209.
[5] H. Karl and A. Willig, "Topology control", in Protocols and
Architectures for Wireless Sensor Networks, Chichester: John Wiley &
Sons Ltd, 2005, pp. 251-288.
[6] "Atarraya - A Topology control simulator for Wireless Ad-hoc and
Sensor Networks", Csee.usf.edu, 2017. [Online]. Available:
http://www.csee.usf.edu/~mlabrador/Atarraya/. [Accessed: 26- Dec-
2017].
[7] M. Penrose, "A Strong Law for the Longest Edge of the Minimal
Spanning Tree", The Annals of Probab., vol.27, no.1, pp. 246-260, 1999.