Effect of Topology on the Performance of Mobile
Heterogeneous Sensor Networks
Ashish Gupta, Alexandre Delye de Clauzade de Mazieux, Monique Becker
GET/INT, SAMOVAR, Institut National des Telecommunications
91011 Evry CEDEX, France
Email:{Ashish.Gupta, Alexandre.Delye, Monique.Becker}@int-edu.eu
AbstractNetwork topology plays a significant role in
the design and performance of sensor networks. The
IEEE.802.15.4 standard is designed for Low-Rate Wireless
Personal Area Networks (LR-WPAN) with most applica-
tions in the automation sector. Zigbee networks placed at
the top of the above standard use two classes of devices:
Reduced Functional Devices (RFDs), where a node can
simply act as an end device with no other capabilities,
and the all powerful Fully Functional Devices (FFDs),
where a node can become an end device, router or the co-
coordinator as well. Therefore, traditional Mobile Ad hoc
Network (MANET) routing algorithms are not applicable
in these types of networks. Furthermore, it is important
to understand the implication of network topology over
the overall performance of the system. We used Network
Simulator-2 (NS2) to simulate different mobile sensor
network topologies with equal energy. The topologies were
different in terms of node types and hierarchy inside the
network. We observed that, in a heterogeneous mobile
sensor networks, increase in network hierarchy coupled
with the node heterogeneity has a detrimental effect on
the network performance.
I. I NTRODUCTION
Developments in the wireless technologies are opening
up new application avenues in the automation of tradi-
tionally labour intensive work. With the infinite number
of small and cheap devices available today, the question
remains, how should they communicate in an effective
manner? One of the major constraints of these small
devices is their limited battery capacity. In fact, the
degree of success of any kind of automation lies on
the level of human interference; the lower the human
interference is, the better the system is, while providing
seamless services. Therefore, not only it is important
that these devices should communicate in an efficient
manner but also independent of any fixed infrastructure.
Nevertheless, sensor networks are different from simple
ad hoc networks, as sensor networks are designed for
The Sixth Annual Mediterranean Ad Hoc Networking WorkShop, Corfu, Greece, June 12-15, 2007
100
much longer life time frame. Eventually, data collected
by each device should be communicated to certain Base
Station (BS) which might be fixed or mobile.
Sensor networks are divided into two types of cat-
egories: - homogeneous networks where each device in
the network has equal capabilities and the heterogeneous
networks where some devices are more powerful than the
other devices. A Zigbee network [1] is an example of
heterogeneous networks. Zigbee networks are launched
by an industrial alliance called Zigbee Alliance [2]. Like
any other network protocol Zigbee is also a layered
protocol and is placed at the top of the IEEE.802.15.4
standard [3]. Therefore, the role of the Zigbee Alliance is
to define the network and application layer. Application
scenarios for these types of networks range from the
system monitoring to the full fledged home and industry
automation. Zigbee Alliance members share the vision of
providing the customers with an industrial standard with
options of interoperability between the various devices
delivered by different vendors. Subsequently, several
Zigbee compliant solutions are currently available.
A node in a Zigbee Network can either be a FFD or
a RFD. A FFD node can act as a router, simple end
device or as a coordinator of the network, where as, the
RFD can only act as an end device. Therefore, RFDs
cannot play any intermediary role in communication
for other devices. This leads to a kind of network
where very few nodes can participate in the routing.
Hence, in this case, classical MANET algorithms such
as, [4], [5], [6], [7], [8] are not applicable. Similarly,
much of the comparisons between the heterogeneous
networks with homogeneous networks are not applicable
in mobile Zigbee networks. The authors in [9] made
comparison between the homogeneous networks and
heterogeneous networks in terms of overall network
deployment cost. They studied a multi-hop variant of
LEACH (M-LEACH) for the intra-cluster routing and
compared with the simple LEACH. However, M-LEACH
is applicable to the static networks and it also assumed
that nodes inside the cluster will also participate in the
intra-cluster routing. Similarly, most work in [10], [11]
and [12] relates to the static networks.
In this paper, we have studied the behaviour of Zigbee
like networks in the tree based topology. Especially,
our area of interest lies in understanding the network
performance based on different node types, radio ranges
and topologies. To compare different topologies, we
assumed that each and every topology should have equal
energy. We modified LEACH algorithm to select the
Cluster Head in the mobile network. We observed that
performance of the network degrades with the introduc-
tion of more hierarchy in the topology and it degrades
even further in case of networks similar to Zigbee.
II. OVERVIEW
Routing is one of the key design issues in the wireless
sensor networks. A sensor can either be a single hop or
multi hop distance away from the BS. In case of single
hop networks, nodes farthest from the BS spend more
energy than the nearest ones and opposite stands for
multi hop networks, since the information to BS is routed
by the nearest nodes. Consequently, depending upon the
topology, nodes either nearest or farthest to the BS will
die much earlier with respect to other nodes. However,
low energy routing is only easy for static networks and
with the introduction of fair amount of mobility in the
system; things can become even more complicated.
Mesh and clustering are the two most commonly em-
ployed techniques in multi hop as well as in the Zigbee
networks. Mesh networks are self-healing. The mesh
networks allow for continuous connections and reconfig-
uration over the broken paths by hopping from node to
another node until the final destination is reached. The
Ad hoc On-Demand Distance Vector (AODV) [4] routing
protocol is a reactive protocol designed for mobile ad
hoc networks. The Zigbee Mesh routing is similar to
AODV. In [13], the authors have discussed the behaviour
of mesh routing in Zigbee networks. They used the set
of FFD and RFD devices and studied the performance
of the network by varying the number of RFD devices.
They used the scenario where around 20% of the devices
were mobile. They observed that, the increase in node
heterogeneity in the network i.e. increase in the number
of RFD nodes, leads to the significant drop in routing
performance of the whole system when compared with
the simple AODV routing mechanism. However, authors
did not take into the account the energy implication
The Sixth Annual Mediterranean Ad Hoc Networking WorkShop, Corfu, Greece, June 12-15, 2007
101
for the above system. Increase in the number of RFD
nodes should decrease the overall energy of the network.
Therefore, different scenarios simulated by the authors
should have different energy levels. In this paper, we
will take more critical view of the toplogy on the
overall performance of the system using energy as the
constraint and how topology coupled by the energy plays
an intrinsic role in the network performance.
Clustering is a model where a sensor network is
subdivided into smaller units called clusters. A subset
of nodes in the network is elected as the cluster heads
(CHs), while the other nodes will join those clusters as
members. Zigbee tree algorithm is an example of cluster
tree algorithm. Depending upon the network parameters
set in the Zigbee application layer, e.g. nwkMaxDepth,
nwkMaxChildren and nwkMaxRouter, a tree is con-
structed and only a FFD node can become the router
or CH. In case of Zigbee tree routing, each CH has its
own 16 bits address space, called network address and
it further sub-divides/allocate the address space between
its children. Therefore, instead of 64 bit MAC addresses,
these 16 bit network addresses can be used for routing
purpose. This helps in auto routing as each CH is aware
of its own address space and its parent address space.
Members will forward data packets to the CHs which,
acting as relays, forward the packets to the sink or Base
Station (BS). However, tree routing is not much feasible
in mobile sensor networks, as CH itself might have
changed its position, therefore, its own address space
and hence the address space of its children do not remain
static at all. Another, way of selecting the CH is LEACH
algorithm. LEACH [14] is the self-organizing, dynamic
algorithm to decide CHs in the network. LEACH has
two phases; the Set up phase and the steady phase. In
the set-up phase nodes organize themselves into local
clusters and the appointed CHs communicates directly
to the BS. After fixed interval of time, the process of
local clustering is repeated again and the new cluster
formation takes place. Therefore, in this paper we used
LEACH algorithm over the Zigbee Tree algorithm to
select the CH and implement the routing scheme. In the
LEACH, at the beginning of each interval a sensor node
can become the CH as the function of some predefined
probability. Since, CH requires quite a large amount
of energy to communicate with Base Station, this CH
rotation policy helps to distribute energy evenly in the
whole network. Once a node declares itself the CH, the
nearby nodes send the join request and once the node
has joined the CH, it sends data to CH in the given time
slot.
 III. N ETWORK M ODEL
We used Fig.1 as the basis of our studies. We
took into consideration that, there can be few orphan
nodes present inside the network. LEACH algorithm
was mainly used to select the Cluster Head and the
following changes were made in implementation of
LEACH algorithm [15] in the Network Simulator (NS2).
Every node is explicitly defined as FFD or RFD,
with higher energy levels for FFD nodes.
For the ease of routing, it is assumed that each and
every node is location aware node with respect to
the Base Station.
Depending on the distance of the node from the
Base Station and its radio range, a node can calcu-
late number of hops required to connect to BS in
terms of NETWORK DEPTH.
NETWORK DEPTH of BS is assumed to be equal
to 0.
Every FFD node will periodically advertise itself as
the Cluster Head and then, wait for the join request
from the other FFD or RFD nodes.
If the FFD node has NETWORK DEPTH equal to
0, it will connect directly to BS, else, it will send
the join request to other FFDs.
Each node including FFD will also send periodi-
cally the join request, if they have not yet joined
any cluster head.
Decision to choose the parent for a FFD is based
on received signal strength and the NETWORK
DEPTH from the advertising FFD node.
Node will never join the cluster whose cluster
head has NETWORK DEPTH higher than the
NETWORK DEPTH of node itself.
IV. S IMULATION
We used 2.4GHz frequency range in our simulation.
In most of the sensor networks, if a sensor is out of
range of its CH or it is an orphan node, i.e. has no CH,
the node switches itself into the energy saving mode.
The node sleeps for some time and then again scans
the network if any device is available to become the
CH. However, it is not true in each and every case.
E.g. in case of automation of the logistics and the
supply chain management of the frozen items inside
the data-ware house, where strong emphasis is put on
temperature control, any significant or minor change in
the temperature must be communicated. If the sensor is
unable to report the same, it may modify the efficiency
The Sixth Annual Mediterranean Ad Hoc Networking WorkShop, Corfu, Greece, June 12-15, 2007
102
Fig. 1. Network Model
of the automation of the system. Therefore, during the
simulation, it was assumed, that node will not get into
the sleep mode if CH is not selected. Hence, the node
can communicate its messages/information at the very
first opportunity.
LEACH algorithm is based on the rotation policy of
the CH. This enables the proper energy utilization in
the sensor network, as the same nodes are not burdened
again and again to become CH. The LEACH was de-
signed for static networks. However, the use of rotation
policy makes it an ideal choice to be used to select
the CH even in the mobile sensor network, as we were
required to select CH again and again. The simulation
ended when the numbers of nodes inside the network
were less than 5 or the time limit exceeded 3700s.
We simulated four different network scenarios. Since,
the motivation for having an heterogeneous network is
to deploy an energy efficient system, therefore, all the
scenarios and simulations had equal initial energy. We
assumed that, a node can either be a FFD or RFD in
case of heterogeneous networks. Also, 75% of the nodes
were RFD and others were FFD ones. In both cases of
heterogeneous networks, FFD nodes were 333% more
powerful than the RFD nodes. As discussed earlier, in
case of a homogeneous network, each and every node
had equal initial energy i.e. 4.864J. However, in case
of a heterogeneous network, we had two energy levels,
i.e. 10J for the FFD nodes and 3J for the RFD nodes.
However, average energy per node remained same in
all the scenarios. Rationale for having heterogeneous
network is to let some nodes act as routers or Cluster
Heads and other nodes just being the end devices.
Therefore, the FFD nodes were made more powerful as
they do data routing consistently. consistently or might just have joined the cluster head
who itself had become orphan. This was not happening
TABLE I
S IMULATION S CENARIO
in Scenarios {1, 2} as all nodes were capable of routing
and hence, no orphaning problem happened. The use of
Network Topology Type Node Range Random Waypoint mobility model (RWP), added a fair
1 Simple LEACH Homogeneous 80m degree of randomness in the network. The reason behind
2 Hierarchical Clustering Homogeneous 50m
3 Hierarchical Clustering Heterogeneous 50m
the selection of radio range in Scenarios {3, 4} were the
4 Hierarchical Clustering Heterogeneous 30m type of topologies they exhibited. While in the scenario
{3} the choice of radio range helped not only to compare
the performance of the heterogeneous network but also
TABLE II to compare the performance with homogenous networks.
G ENERAL S IMULATION PARAMETERS
The scenarios {3, 4} provided with an opportunity to
Simulation Area 150*150 contrast between the different networks. Furthermore,
Location of Base station 67.745, 92.58 radio range of scenario {3} was higher then the radio
Number of Nodes 40 (including Base Station) range of scenario {2} but lower than the radio range of
Mobility model Random Way point
scenario {1}. This provided an interesting proposition to
Node speed 1m/s
Node Pause time 8s compare the networks, which not only differ in terms of
CH Rotation frequency 10s the device types but also in terms of the radio range.
Time out time for Child 5s
Packet Size 127 byte V. R ESULTS
Traffic CBR 0.2 interval time
Energy of Base Station 5000
Minimum thresh hold Energy 1
Total Energy in the Network 187 (excluding BS)
Average Energy per node 4.864
Simulation Time 3700 s

TABLE III
H ETEROGENEOUS NETWORK PARAMETERS

Number of Fully Functional Device 10

Number of Reduced Functional Device 29
Energy of Fully Functional Device 10J
Energy of Reduced Functional Device 3J

Simulation scenarios as listed in Table I represent

the different network topologies. Scenarios {1, 2} were
using homogeneous nodes and every node was capable
of routing. In case of Simple LEACH a node was at most
two hops away from the Base Station. On the other hand,
by decreasing the radio range, we had increased the level
of hierarchy in the network. Now, some nodes require
more than two hops to communicate with the BS and also
not every CH was single hop from the BS. Scenarios {3,
4} represent heterogeneous networks i.e. very few nodes Fig. 2. Number of Messages Received at BS
were capable of routing. Therefore, different radio ranges
were used to represent different hierarchical topologies.
The higher the number of intermediate hops, the higher This section studies the performance of the network
the probability of the route breakdown, or vice versa. scenarios based on the simulation parameters as listed
The other major difference between Scenarios {1, 2} in Table II and Table III. We have examined the perfor-
and Scenarios {3, 4} was that, the nodes in Scenario mance of networks on the basis of three given criteria.
{3, 4} might be orphan and searching the cluster head 1) Number of messages received at the BS.

The Sixth Annual Mediterranean Ad Hoc Networking WorkShop, Corfu, Greece, June 12-15, 2007
103
 2) Number of nodes alive.
3) Longevity of the network.
In case of homogenous networks, every node has more
or less the equal probability of becoming a cluster head.
On the contrary, very few nodes designated as FFDs can
act as the CH in the heterogeneous networks (25% in
this case). Therefore, if the whole set of FFDs are out of
the radio range of any given RFD node, the RFD node
becomes orphan and therefore, unable to communicate
with the BS.
Thus, nodes in the heterogeneous networks may suffer
repeatedly from the orphaning phenomenon. However, it
is not true in case of homogeneous networks. Therefore,
number of messages received at the BS is much higher
in case of the homogenous networks. Moreover, if CH is
more than one hop away from the BS, it requires another
CH as intermediate node to connect itself to the BS; and,
in case of heterogeneous networks, a CH can itself be
an orphan node.
Fig.2 compares the total number of message received
at the BS at any given time. We observed, as the
hierarchy of the network increases, the performance of
Fig. 3. Total Energy Consumed the network with respect to the number of messages
delivered at BS, decreases substantially. Performance is
at its best in case of homogeneous single hop network
and worst in case of multi hop heterogeneous networks;
the higher the number of hops is, the poorer the per-
formance is. Due to mobility, it is not only a simple
node which changes its position but also the CH itself
might have changed its positions as well. At any given
time there is also the probability that the selected CH
may also have moved out of the radio range of its
own cluster. Same phenomenon may occur during the
communication between the two CHs. Therefore, as the
degree of hierarchy increases the performance of the
network degrades. Performance degrades even more in
heterogeneous networks, as very few nodes are available
to become CH.
Every node consumes energy while transmitting or re-
ceiving the messages. Even the CH selection mechanism
consumes energy. From Fig.3 it can be seen that as the
hierarchy and the heterogeneity of the system increases
more energy is consumed by the system. It is higher
in case of the heterogeneous topology. This is due to
the fact that, the network suffers from the orphaning
phenomenon and more energy is utilized by a RFD node
to select a CH. However, in case of Scenarios {1, 2} we
Fig. 4. Number of Alive Nodes
do save energy, but due to mobility inside the network,
there have to be a trade-off between the data and the
energy.

The Sixth Annual Mediterranean Ad Hoc Networking WorkShop, Corfu, Greece, June 12-15, 2007
104
 Efficient energy utilization is one of the main reasons
for having heterogeneous networks. Therefore, all the
simulated scenarios had equal initial energy levels. This
helped to understand the performance of the system
on the basis of energy. Fig.4 gives an idea about the
longevity of the network. It compares the number of
nodes remained inside the network at any given time.
We observed that, the RFD nodes die more quickly than
that of nodes in homogeneous networks. This happened
due to the fact that, the RFD nodes had less initial energy
than that of homogeneous nodes and RFD nodes lost lot
of energy while searching a CH. Finally, only FFDs re-
mained in the network as they had the maximum energy.
This rapid network degradation, in case of heterogeneous
networks, also contributed to the cause of lower number
of messages received at BS.
VI. C ONCLUSION
The paper studied the performance of sensor networks
consisting of two types of devices, i.e. FFDs, the devices
which can route the packet as well as can act as the CH,
and the RFDs, which can only connect to the FFDs but
incapable of routing or becoming the CH themselves.
The heterogeneous networks discussed above are similar
to the networks proposed by the Zigbee. We simulated
four different scenarios, two each for the homogenous
and heterogeneous networks. The four network scenarios
varied in terms of topologies and device types. With
respect to number of messages received at BS, our results
are similar to that of authors in [13], however, we studied
a different problem: the assumptions are different. We
used energy as the constraint and examined the role
played by topology while designing heterogeneous net-
works. We observed that, the performance of the network
is deeply intertwined with the network topology. In the
mobile sensor networks, increase in the network hierar-
chy increases the energy consumption in the network as
well as degrades the performance of network in terms of
number of received messages. Furthermore, due to the
inability of RFDs to participate in routing, performance
of the system degrades even further in case of the
heterogeneous hierarchical networks. Zigbee networks
looks good and indeed they are good for the static
networks. However, for the networks where mobility
is a regular feature, the deployment of heterogeneous
networks like Zigbee does not seem to be correct idea
and at least, they need further investigations.
R EFERENCES
[1] Zigbee Specification V1, June 2005.
The Sixth Annual Mediterranean Ad Hoc Networking WorkShop, Corfu, Greece, June 12-15, 2007
105
[2] Zigbee alliance. http://www.zigbee.org.
[3] J. Zheng and M. J. Lee. A Comprehensive Performance Study
of IEEE 802.15.4, 2004.
[4] C.E. Perkins and E.E. Royer. Ad hoc on demand distance
vector routing. In WMCSA 99 2nd IEEE Workshop on Mobile
Computing Systems and Applications, New Orleans, LA, page
90100, Feb 1999.
[5] V. Park and S. Corson. Temporally-ordered routing algorithm
(tora)version 1.
[6] A. Qayyum A. Laouiti L. Viennot P. Jacquet, P. Muhlethaler and
T. Clausen. Optimized link state routing protocol (olsr). Internet
Draft, draft-ietf-manet-dymo-03.txt,RFC 3626, October 2003.
[7] E. Belding-Royer Chakeres and C. Perkins. Dynamic manet
on-demand routing protocol (dymo). Internet Draft, draft-ietf-
manet-dymo-03.txt,, October 2005.
[8] D. B. Johnson and D. A. Maltz. Dynamic source routing in ad
hoc wireless networks. volume 353, chapter 5, page 153181.
Kluwer Academic Publishers, 1996.
[9] Vivek P. Mhatre and Catherine Rosenberg. Homogeneous vs
heterogeneous clustered sensor networks: A comparative study.
In IEEE International Conference on Communications (ICC
2004), volume 27, pages 36463651, June 2004.
[10] Stanislava Soro and Wendi B. Heinzelman. Prolonging the
lifetime of wireless sensor networks via unequal clustering.
In Proceedings for 5th IEEE International Workshop on Al-
gorithms for Wireless, Mobile, Ad Hoc and Sensor Networks
2005 (WMAN05), Denver, Colorado, April 2005.
[11] Daniel Kofman Vivek P. Mhatre, Catherine Rosenberg and Ness
Shroff. A minimum cost heterogeneous sensor network with a
lifetime constraint. IEEE Transactions on Mobile Computing,
4(1), January/February 2005.
[12] Enrique J. Duarte-Melo and Mingyan Liu. Analysis of energy
consumption and lifetime of heterogeneous wireless sensor
networks. In Proceedings of IEEE Globecom, Taipei, Taiwan,
November 2002.
[13] Tony Sun Guang Yang Ling-Jyh Chen Nia-Chiang Liang, Ping-
Chieh Chen and Mario Gerla. Impact of node heterogeneity in
zigbee mesh network routing. In IEEE International Conference
on Systems, Man, and Cybernetics (SMC06), Taipei, Taiwan,
2006.
[14] A.Chandrakasan W.R.Heinzelman and H. Balakrishnan.
Energy-efficient communication protocol for wireless micro
sensor networks. In 33rd annual Hawaii International
Conference on System Sciences, HICSS, page 30053014, 2000.
[15] Network simulator 2 leach implementation.
www.internetworkflow.com/resources/ns2leach.pdf.