SOME NOTES AND PROPOSALS ON THE USE OF IP-BASED

APPROACHES IN WIRELESS SENSOR NETWORKS

Tiago Camilo, Jorge Sá Silva, Fernando Boavida

Department of Informatics Engineering,
University of Coimbra, Coimbra, Portugal
[email protected]

ABSTRACT
Wireless Sensor Networks (WSNs) are gaining visibility and importance in a
variety of fields and will certainly be part of our day-to-day lives in the near future.
This trend is, in effect, putting WSNs under the research community spotlight. The
feverish activity around WSNs has led to some myths and misconceptions over the
last years that, in some way, have blocked the way forward. This paper addresses
some of these myths and discusses a model for Wireless Mesh Sensor Networks
that go beyond them, showing that it is time to look at WSNs under a different
light. The paradigms that support the proposed model have a direct impact on the
addressing scheme, mobility support and route optimisation. These have been put
to the test both by simulation and prototyping, showing that they constitute solid
ground on which future Wireless Mesh Sensor Networks can be built.

Keywords: wireless sensor networks, IP, ant colony optimization, anycast.

1 INTRODUCTION nodes; being self-healing, since sensor nodes

cooperate in order to automatically re-route their
Composed of a potentially high number of very signals to the out-of-network node, thus ensuring a
small devices, Wireless Sensor Networks (WSNs) more reliable communication path; supporting multi-
are one of the most promising technologies for the hop routing, since data can be forwarded through
21st century. Emerging WSNs make use of recent multiple nodes before it reaches a sink-node.
advances in electronic sensors, communication The overall performance of a sensor node is
technologies and computation algorithms. affected by the features of its components: battery,
WSNs have unique characteristics, mainly due to memory, processor, sensors, receiver/transmitter.
their component devices, called sensor nodes. These Hill et al. [7], grouped sensor nodes into four classes,
nodes are typically small size devices with depending on their physical size, radio bandwidth
communication and monitoring capabilities, as well and memory size (Table 1).
as limited resources, namely in terms of memory,
energy and processing power. A node in a WSN Table 1: Sensor nodes classes [7]
consists of a sensor or an actuator that is connected
to a bidirectional radio transceiver. In contrast to Sensor Radio Memory
sensor nodes, sink nodes (special WSN nodes that Size
Type Bandwidth Flash RAM
act as central nodes which gather/distribute Specialized
information) have fewer restrictions, allowing them 0,1 <4
sensing mm3 <50 Kbps
to store relatively large amounts of information and Mb Kb
platform
perform highly demanding processing and routing Generic
tasks. These devices are responsible for managing 1-10 0,5 <10
sensing <100 Kbps
the communication between a sensor network and its cm3 Mb Kb
platform
respective base station (wireless or wired). High-
Wireless Mesh Sensor Networks (WMSNs) 1-10 10 <128
bandwidth ~500 Kbps
aggregate several types of sensor nodes under a cm3 Mb Kb
sensing
single working network, thus giving access to the >10 32 <512
data, information and/or services of each component Gateway >500 Kbps
cm3 Mb Kb
WSN platform. In a typical mesh network each
sensor node can communicate with more than one
Sensors like Spec
node, enabling better overall connectivity than in
(http://www.jlhlabs.com/jhill_cs/), which are
traditional star topologies. WMSNs are characterized
characterized by being extremely small, belong to
by: offering a combination (mesh) of several types of
the first class, the specialized sensing platform. This

Ubiquitous Computing and Communication Journal 1

class covers sensor nodes that only perform communication paradigm, contrary to the node-
monitoring and send their sensor data directly to the centric approach followed by most networks,
sink-node. Mote [9] devices are bigger and can including ad hoc networks and the Internet. The
perform forwarding tasks. Such devices belong to the node-centric approach is regarded as unsuitable to
generic sensing platform class. High-bandwidth WSNs, as it generally relies in complex
sensing nodes have more resources than the nodes signalling/routing protocols. In the following, each
belonging to the previous classes. However, the size communication paradigm is explained.
of the device also increases, mainly due to the power The data-centric communication paradigm is
supply. An example of an instance of such sensor built on the data gathered by the sensor. In this case
class is the iMote [9] sensor node. iMote nodes have the observer is not interested in knowing which
video, audio and air-monitoring capabilities. Lastly, particular sensor replies to a specific query. Instead,
the gateway class aggregates nodes that can execute the most important thing is to get the answer to the
all the above mentioned tasks and, additionally, query. Therefore in protocols such as Directed-
support the interaction between the sensor network Diffusion [8], the user only needs to specify a certain
and the infrastructured network, being strategically condition when querying the network. The returned
placed between them. The devices from this class data can be provided by one or more sensor nodes, or
(e.g., Stargate, http://platformx.sourceforge.net) even be an aggregation of sensor data gathered by a
typically have several interfaces and make use of a group of sensors. Data-centric communication
more powerful energy source. provides the ability to specify various parameters,
Security, traffic control, military strategy, such as rate of publication, rate of subscription,
industrial control, healthcare and habitat monitoring validity of the data and many other. In short, with
are examples of possible applications of WSNs and this communication paradigm routing is based on the
WMSNs. This wide range of applications requires data provided by a sensor, not on the identity or
that WMSN protocols are adaptable to their location of the sensor.
deployment environment. This paper proposes a The location-centric communication approach
WMSN architectural model, named IPSense, which uses the location of sensor nodes as a primary means
supports such adaptation requirements, namely by for addressing and routing (e.g. CODE [13]). The co-
using flexible addressing, enhanced mobility and operation between devices in a given area,
energy-efficient routing. IPSense also demonstrates performing local aggregation, takes a special role in
that some of the myths associated with wireless this kind of communication. This paradigm is
sensor networks – namely, that the use of multiple extremely dependent on positioning systems, such as
addressing schemes, the use of IP and the use of the Global Positioning System (GPS), which may not
complex routing protocols have high cost – do not always be feasible since it requires high amounts of
hold. energy. In addition, this type of communication
The following section identifies and discusses approach may lead to increased cost and may be
some of the WSN myths, their implications and restricted to outdoor use. The most important
possible ways to circumvent them. Section 3 presents advantage of this approach is related to the absence
the IPSense model, describing its key paradigms, of routing tables in the network. Each node forwards
features and approaches. These were put to the test packets based on the destination location. Such
by simulation and prototyping, and the results of that characteristic is very helpful in mobile scenarios,
evaluation are presented in Section 4. Section 5 since mechanisms such as route discovery and
presents some conclusions and guidelines for further routing updates are not required, decreasing the
work. energy spent on control messages. However each
sensor node has to know the correct location of
2 MYTHS IN WIRELESS SENSOR intermediate and destination nodes.
NETWORKS Lastly, in node-centric communication sensor
nodes are labelled with unique identifications, such
This section discusses some of the myths as IP addresses, which are used to populate the
normally associated with wireless sensor networks. routing tables and to perform packet forwarding.
These are common misconceptions that, in general, This communication paradigm is the “traditional”
limit the current use of this type of networks. The communication method nowadays used in the
section analyses each issue from several angles, Internet, throughout wired and wireless networks.
identifying real problems and possible ways forward. Due to its characteristics, it is possible to perform
hierarchical addressing, creating simpler routing
2.1 WSNs Should Be Data-Centric schemes. However this approach is the less used in
Three main communication paradigms can be WSNs/WMSNs environments since it can lead to
used in sensor networks: data-centric, location- considerable amount of control traffic, which will
centric and node-centric [12]. WSNs are application- dramatically reduce the network lifetime.
specific and, in general, use a data-centric Nevertheless, it has the advantage of enabling global

Ubiquitous Computing and Communication Journal 2

connectivity and the support of well know IP-based location and frequently access them.
protocols. The most common approach to integrate
Each of the presented communication paradigms WMSNs with IP-based networks is to incorporate a
has advantages and drawbacks and, contrary to proxy between the two networks (e.g. [15]). This
common belief, they all can be applied to WMSNs. It element acts as a protocol translator, allowing inter-
is up to the WMSN developer to choose the best network communication. The proxy needs to
approach that fulfils the application requirements. It understand the protocols of the networks it
is important to note that hybrid solutions can also be interconnects: the IP stack and the protocol stack
applied, where the conjunction of two or more used within the WMSN (Fig. 1). Strategically placed,
communication models can lead to specific benefits. the WMSN/IP proxy acts as a database (normally
through software), collecting and storing the
2.2 IP Is To Complex To WSNs monitored data provided by the sensors nodes.
The main motivation for using IP in WMSNs is
global connectivity. Such use would greatly increase
the potential for interaction between sensors and
observers, allowing the latter to access the gathered
data from virtually anywhere, in what is normally
known as a 4G scenario. Nevertheless, IP was not
designed for energy-restricted, low-memory and
low-processing-power devices, which has led to the
belief that IP is too complex for WSNs/WMSNs. In
spite of this, an increasing number of researchers are Figure 1: Proxy WMSN/IP architecture
looking into the use of IP in sensor networks due to
the potential that it represents. This architecture provides access to the collected
There exist several differences between IP-based data through the use of standard TCP/IP protocols,
networks and the current WSN/WMSN models. On without losing the capability to create application-
one hand, the broad applicability of WMSNs has led specific protocols for the WMSN. However, the
to architectural models that are application-specific. proxy architecture is based on a single, centralized
On the other hand, WMSNs are, by nature, data- access point, leading to problems in terms of
centric systems, where sensor data is the key for the reliability, scalability and mobility support.
routing mechanism. This is not the case in an IP- The main problems that are commonly pointed
based network, where devices/applications are out when direct support of IP in WMSNs is
reached through the use of IP addresses, which are concerned are the following: lack of available
application-independent. addresses, lack of configuration management, and
In IP-based networks the communication is energy efficiency due to heavy processing required
typically performed on a one-to-one basis, with the by the IP stack [3]. Nowadays, solutions for each of
possibility to support one-to-many interactions in the these problems exist, namely the use of NAT
case of multicast. On the other hand, in mechanisms, the use of the Dynamic Host
WSN/WMSN communication is usually of the Configuration Protocol and the use of efficient, thin
many-to-one or one-to-many types, since sensor IP stacks. Computational limitations and low
nodes usually transmit to a single point, the sink- memory resources are commonly pointed as
node which, in turn, queries sensor nodes. Such limitations to a full support of the IP stack in WSNs
behaviour considerably reduces the amount of data [2]. However the work performed in [5] with the µIP
transmitted over the network. TCP/IP, proves the feasibility of such integration.
In IP-based networks, bandwidth is the most Developed to be executed in 8-bit micro-controllers,
important obstacle/restriction, due to the amounts of this stack allows nodes to directly communicate with
data transferred between elements (video, sound, etc). full-IP devices, as illustrated in Fig. 2.
In WMSNs the amount of data transferred between
devices is typically small. Moreover, there can be
relatively long periods of time with no data
exchanges.
WMSNs are very peculiar regarding the
development phase. Contrasting with most networks,
the nodes location is very important, and in some
cases it has a direct impact on the network lifetime.
Typically, WMSNs need to be self-configured and
Figure 2: IP-based WMSN
unattended due to the fact that access to sensor nodes
is not always possible, contrary to IP-based networks
By endowing sensor nodes with IP addresses,
where administrators know the nodes physical
they will be able to benefit from the main capabilities

Ubiquitous Computing and Communication Journal 3

of this protocol. In this model de communication useful in WMSNs.
between mobile devices and mobile nodes is
completely transparent. Neither tunnel mechanisms 3 THE IPSENSE MODEL
nor intermediate devices are required to perform the
protocol translation. Sensor nodes are accessible This section presents a model – which, for ease
from any other IP-capable device, such as a PDA. of reference we will call IPSense – for WMSNs that
Although IP seems to be a viable protocol for uses and explores paradigms that are contrary to the
micro-controllers, the IP header is too large when myths and misconceptions presented in the previous
compared with the typical size of data packets sections. Specifically, IPSense has the following
exchanged between sensor nodes. The overhead can features:
reach up to 90% [6]. However it is possible to use  Global connectivity, through the use of IP-
compression mechanisms to reduce the total message enabled nodes;
size [10].  Support for all the addressing paradigms
mentioned in the previous section;
2.3 IPv6 Is Even More Complex And Can  Mobility support and Auto-configuration;
Hardly Be Used In WSNs  Optimised routing;
When developing a WMSN, it is important to  Reduced protocol overhead;
have scalability in mind. WSN/WMSN networks  Energy-efficiency.
with up to 1000 sensor devices will be quite common The following sections detail the IPSense model.
in the future. With these numbers, it is unfeasible to
use public IPv4 addresses on each device, which is 3.1 Sensor Router And Mobility Support
why NAT mechanisms may be a solution. However, IPSense explores sensor node aggregation in
such solution also has drawbacks in terms of end-to- clusters, through the use of Sensor Routers (SRs), in
end transparency, security, performance and order to manage the communication between the
processing time. cluster members and the access point, located in the
Using IPv6 in WMSNs would allow us to wired/wireless network (Fig. 3). Sensor Routers are
overcome some of the IPv4 limitations and would gateways between wireless sensor networks and the
open a whole new range of possibilities. The large outside world comprising a whole range of
address space available with IPv6 would allow us to heterogeneous networks (Fig. 4). SRs are special
assign public addresses to all sensor nodes, devices that have more powerful hardware
eliminating the need for NAT. Communication capabilities when compared to sensor nodes, in what
between sensor nodes and external devices would be concerns energy, memory and processing power.
done without intermediate nodes, protocol translation
or proxies. In contrast with IPv4, the IPv6 header is
designed to keep the overhead to a minimum, by
shipping unused fields to extension headers.
One important property in WMSNs is the lack of
network management. After the deployment of
sensor nodes, it becomes unfeasible (in most
scenarios) to perform administrative operations, such
as to choose the best configuration (e.g. routing) for
each node. Therefore it is important to use auto-
configuration mechanisms such as the Neighbour Figure 3: Sensor Router concept
Discovery Protocol and other auto-configuration
facilities natively found in IPv6. Sensor routers are responsible for configuring
Finally IPv6 provides a new type of address, the the sensor nodes in their own sub-network. Each
anycast address. The use of anycast addressing is node will be provided with one IPv6 unicast address
normally associated with fault tolerance mechanisms, and a set of IPv6 anycast addresses. Unicast
where the same service is provided by more than one addresses allow the unique identification of each
device, all using the same IP address. A packet sensor node. Anycast addresses are assigned
destined to an anycast address is delivered to the according to the device sensor properties, thus
“best” interface, from all those using the same providing an efficient way to address sensor nodes
anycast address. with similar capabilities. The Sensor Router concept
In spite of the general belief that IPv6 is too is fundamental to guarantee energy efficiency in
complex for WMSNs, part of the IPv6 features can WMSN networks. By concentrating energy-
be easily ported to this type of networks at reduced expensive communication between WMSN and
cost, boosting the usefulness of IP in WMSNs. wired/wireless networks in a dedicated element, it is
Mobility support and auto-configuration are just two possible to considerably relax the energy
examples of IPv6 features that can be extremely management and communication requirements of

Ubiquitous Computing and Communication Journal 4

sensor nodes, allowing them to efficiently deal with
sensor operations and local communication.

Figure 6: Sensor node mobility

3.2 Addressing Scheme

Figure 4: Connectivity model between IPSense and IPSense combines the three WSN
4G networks communication paradigms mentioned before. It
supports the ability to communicate and to interact
In addition, SRs can easily provide mobility with a specific sensor, using IPv6 addresses,
support inside the sub-networks, according to the according to the node-centric routing approach. The
network mobility model developed by IETF’s use of anycast addresses supports the two remaining
NEMO working group [17]. In this context, SRs play routing paradigms in WSNs: data-centric and
the role of Mobile Routers, managing several tasks: location-centric.
route IPv6 unicast addresses, forward IPv6 anycast 3.2.1 Anycast addressing
addresses, map unicast/anycast addresses, and act as IPSense enables the association of IPv6 anycast
gateway between the Internet and WMSNs. addresses to specific services or locations. Therefore
WSN mobility support opens the possibility to it is possible to associate an address to devices with
monitor different types of mobile phenomena such as the same sensing capabilities, such as video, sound
migrations or vehicle movement. In a typical mobile and temperature. Such addressing scheme can also
sensor environment, possible issues are the location be useful to determinate which functionality a
of each individual node or the location of the whole specific sensor node can offer, just by inspection of
sensor network. Such features are normally the list of its anycast addresses.
supported by complex protocols that require When an observer needs to know the values of
considerable power. To overcome this problem, some monitored parameter (e.g., humidity), a query
IPSense explores the use of some of the Mobile IPv6 is made with the IPv6 anycast address associated to
characteristics in Sensor Routers (Fig. 5). such service. The message is routed to the node that
Since SRs are responsible for a group of devices, is best positioned to respond to such query (Fig. 7),
whenever they move, the entire network has to e.g., the sensor node belonging to that specific
perform the same movement. However, IPSense also anycast group with highest remaining power or the
allows sensor nodes to change from one WSN to sensor node closest to the SR.
another, when a new SR is found (Fig. 6).
A sensor node only knows that it lost
connectivity with its SR when no Router
Advertisements (RAs) are received after a period of
time. On this occasion, the sensor device has two
possibilities: stays in idle state until the reception of
an RA from a new SR, or contacts directly a known
SR. After receiving the RA, the sensor node needs to
alert the SR of its presence and the auto-
configuration process will start.
Figure 7: Anycast addressing scheme

This procedure is based on the concept presented

by Ata et al. [1], where a new network architecture is
proposed to solve known inter-domain anycast
problems. The authors propose the addition of a new
element, the Home Anycast Router (HAR), which is
responsible for forwarding packets to its network
prefix (Fig. 8). This device has to be placed in the
incoming/outgoing network link of the group of
Figure 5: Sensor Router mobility devices configured with anycast addresses, the
Anycast Receivers (ARs). The HAR is configured

Ubiquitous Computing and Communication Journal 5

with one unicast address, the unique identification, intra-WSN communications should used MAC
and with a set of anycast addresses that are equal to addresses (64 bits) or the reduced address
the anycast addresses configured in the ARs. The identification (16 bits) proposed by the IETF
Anycast Initiator (AI) is the node who wishes to 6LoWPAN working group [11]. Whenever an
communicate with one AR. It can be located inside observer intends to communicate directly with one
or outside the network managed be the HAR. specific sensor node, the SR translates the
In IPSense we propose to delegate the HAR destination address from the 128-bit address to the
competences in the SR, which is the device corresponding internal address (64 or 16 bits).
responsible for a group of sensor nodes, the ARs. Header compression techniques are also necessary to
The AI will be the mobile/wired device used by the reduce the impact of IPv6 headers on WMSN
observer to collect monitoring data from the WMSN. packets. In [5], the authors suggest the use of an
additional field that codes the unnecessary fields in
the communication between two link-local nodes.
3.2.3 Auto-configuration
Sensor node configuration is one of the critical
factors in WMSNs. It is fundamental to find
configuration mechanisms that do not require human
intervention. IPv6 natively supports address and
network prefix configuration. This is based on IPv6
link-local addresses [16], which are plug-and-play
Figure 8: IPv6-based global anycasting architecture addresses formed by the combination of a network
prefix and a 64-bit MAC address.
IPv6 was initially designed to support In the IPSense model, a typical address is
geographic addresses [1]. By assigning IPv6 to well divided into three parts, as illustrated in Fig. 9:
known geographic places, it is possible to create a network prefix (which should be less than 64 bits),
networking map representing the specific place of MAC address, and an intermediate space that will be
each access point. This can be explored in the SR used to identify the type of address, anycast or
scenario by assigning a specific geographic network unicast. This intermediate address space should also
prefix to each SR (Fig. 9). This association is valid be used to identify the anycast service: location or
for fixed and mobile networks, since the network type of service. Unicast addresses (unique
management is located in the network border, being identification) are formed as illustrated in the figure:
the SR the only responsible for the forwarding of the the intermediate space should be field with zeros.
outgoing traffic.
The SR carries out a set of tasks: build and
manage routing tables, assign anycast addresses,
perform the correspondence between sensor types
and anycast addresses, forward unicast and anycast
packets in both directions, and finally manage its
sub-network in terms of energy levels. Figure 9: Formation of the unicast/anycast address
By combining sensor type and sensor location,
the IPSense model allows the observer to query a The anycast address configuration is a little more
specific type of sensor (e.g. humidity) from a complex than the unicast address configuration. Each
specific location (e.g. the garden) just by specifying node is assigned a group of anycast addresses
one anycast address. This approach has the potential according to its own capabilities. For example, if a
to deal with several disperse sub-networks at global sensor node is capable of monitoring three distinct
scale, without loosing contact with the sensor nodes, parameters, it will be assigned at least three anycast
adopting a uniform, universal, IP-based paradigm. addresses, each one representing one type of
3.2.2 Unicast Addresses parameter. The anycast assignment occurs only after
Another important aspect is the ability to the sensor node has acquired the unicast global
communicate with one specific node directly; this is address. It is the responsibility of each node to
only possible due to the unicast address inform the SR of all its capabilities. A sensor node
configuration. Each node will be assigned, as already requires a new unicast/anycast address in any of the
described above, not only anycast addresses, but also following situations:
one unique identifier, one unicast address, that will  The interface is initialized at system start-up
be used to identify the sensor node outside/inside the  The interface is restarted after a temporary
network. system fault
From an energy-efficiency perspective, it is not  The system administrator activates the interface
feasible to use the native IPv6 128-bit addresses for after an inactivity period.
communications between sensor nodes. Therefore

Ubiquitous Computing and Communication Journal 6

3.2.4 Ant colony route optimization nodes and es is the actual energy level of node s), and
IPSense proposes a route optimisation algorithm and are parameters that control the relative
tailored to the needs of WMSNs, called Ant Colony importance of trail versus visibility.
Route Optimisation (ACRO). This protocol is based The selection probability is a trade-off between
on the Ant Colony Optimization heuristic [4] that visibility (which means that nodes with more energy
uses a model based on the behaviour of ant colonies. should be chosen with high probability) and actual
Ants are insects with simple individual behaviours trail intensity (that means that if on the (r,s) path
and efficient processes for individual survival but, as there has been a lot of traffic then it is highly
a colony, they can create complex organizational desirable to use this path).
systems, where each ant plays a specific part that is Each sensor node records the travel data of each
essential to the survival of its anthill. forward ant: the previous node, the forward node, the
Ant Colony Optimization (ACO) is based on the ant identification and a timeout value. Whenever a
observation of real anthill behaviours, more forward ant is received, the node searches its
specifically in the way ants find the shortest path memory for a record of that specific ant, to detect
between the food and the anthill. To bring food to the possible loops. If no record is found, the node saves
anthill, the ant colony solves an interesting the required information, restarts a timer, and
optimization problem. Initially, ants randomly course forwards the ant to the next node. If a record
the region near the anthill searching for food. Each containing the ant identification is found, the ant is
ant, while travelling its path, places a chemical eliminated. When a node receives a backward ant, it
substance in the soil named pheromone, creating a searches its memory to find the next node to where
pheromone trail. The following ants detect the the ant must be sent. The timer is used to delete the
present of this substance and tend to choose the path record that identifies the backward ant if, for some
marked with the bigger concentration of pheromones. reason, the ant does not reach that node within the
These substances enable the formation of the return time period defined by the timer. Each ant k carries
path to the ant and inform other ants about the best information on the average energy available on the
paths to the food. After a period of time, the more path that ends in the current node (EAvgk), and the
efficient paths (paths with the shortest distance to the minimum energy level registered in that path
food) will have a bigger pheromone concentration. (EMink). These values are updated at each node that
Conversely, the less efficient paths will have a lower receives forward ants, and will be used to calculate
pheromone concentration, due to the smaller number the pheromone parameters. When a forward ant
of ants travelling those paths and also because of the reaches the SR, it carries information regarding the
natural pheromone evaporation. In the optimization travelled path. The SR uses this information to
problem that the anthill has to face, each ant is determinate the quality of the path, following an
capable of building a complete solution for the optimization function, which considers the main
problem. However, the best solution is achieved with WMSN limitations (i.e. energy). Before a backward
the information gathered by the colony as a whole. ant k starts its return journey, the destination node
In the IPSense model, each node belonging to an computes the amount of pheromone trail that the ant
SR sub-network generates an ant k (control packet) will drop during its journey, using (2):
at regular intervals, which travels through the
network, jumping from sensor to sensor until it 1 (2)
reaches the SR. At each sensor router i, the ant Tk 

C 
EMin k Fd k
chooses the next sensor node j. At each node r, a  EAvg k  Fd k 
forward ant selects the next hop using the
probabilistic rule presented in (1). The identifier of
where C is the maximum energy level of nodes
every visited node is stored in Mk and carried by the
and Fdk is the distance travelled by the forward ant k
ant.
(the number of nodes stored in its memory).
Whenever a node r receives a backward ant coming
 T r, s  E(s) from a neighbouring node s, it updates its routing
 if s  M k
pk r, s    T r, u  Eu 
   table by using (3).
(1)
 uM k
Tk
0 otherwise Tk (r , s )  (1   )  Tk (r , s )    (3)
   Bd k 
where pk(r,s) is the probability that ant k chooses
where ρ is a coefficient such that (1 - ρ)
to move from node r to node s, T is the routing table represents the evaporation of trail since the last time
at each node that stores the amount of pheromone Tk(s,r) was updated. φ is a coefficient and Bdk is the
trail on the (r,s) path, E is the visibility function travelled distance (the number of visited nodes), by
 
given by C  es 1 (C is the initial energy level of the backward ant k until node r. These two parameters

Ubiquitous Computing and Communication Journal 7

will force the ant to loose part of the pheromone behaviour that closely follows the one observed in
strength during its way to the source node. The idea groups of wild animals. In each scenario, well known
behind this behaviour is to build a better pheromone ad hoc protocols were used: the Destination
distribution (nodes near the sink-node will have more Sequence Distance Vector (DSDV) protocol,
pheromone levels) and will force remote nodes to representing the group of Table Driven Routing
find better paths. Such behaviour is extremely protocols, and the Ad hoc On-Demand Distance
important when the sink-node is able to move, since Vector Routing (AODV) protocol, from the group of
the pheromone level adaptation will be much quicker. On-Demand Routing Protocols. In each scenario one
The Ant Colony Route Optimisation approach of the sensor nodes was randomly elected as the
combines mobile agents (ants) with the Ant Colony sink-node. This device was responsible for collecting
Optimization heuristic, and finds the best routing all the monitored data provided by the entire WMSN.
path between two nodes based on a specific In each simulation run, eight network sizes were
optimization function. This routing algorithm is used, from 25 nodes up to 200 nodes per network.
extremely adaptable, since it uses an optimization The following parameters were used on each node.
function that the administrator/observer can change  sensing power = 0.00000175 mW;
according to the goals. It is possible to build paths  transmitting power = 0.175 mW;
based on energy-efficiency, as the example presented  receiving power = 0.145 mW;
above, but by changing (2) the algorithm can focus  idle power = 0.0 mW;
on QoS, for example, choosing paths according to  initial energy = 0.5 J;
the available bandwidth or other parameters. As previously mentioned, the main goal of this
study was to evaluate existing ad hoc protocols, in a
4 BREAKING THE MYTHS realistic WMSN deployment scenario, including
highly dynamic environments. Therefore, energy
In a way, IPSense intends to ‘break the myths’ consumption, being one of the most restrictive
presented in section 2. In order to do so, it is parameters in WMSNs, was part of the study. The
necessary to subject this model to a series of tests results are illustrated in Fig. 10, where static and
that prove the feasibility of the underlying ideas and mobile scenarios are compared. The remaining
shows that they do not lead to performance problems. energy is calculated by adding the energy levels from
This testing was, in fact, carried out, and the results all the sensor nodes at the end of the simulation.
are presented in this section. These were obtained by In mobile environments using the AODV
simulation and prototyping. protocol, the network became very unstable as we
reached 100 nodes, dropping until only 14% of total
4.1 Mobile IP Based Protocols In WSNs energy in the end of simulation. With more sensor
In a first phase it is important to understand the nodes the network became inactive during the
implications of using well-known mobile ad hoc simulation, due to the high number of nodes that
protocols in WSN environments. Therefore, realistic drained their energy. With 200 nodes the DSDV
scenarios were simulated in order to study the protocol still kept the energy level at 56% of the
behaviour of such protocols in static and dynamic initial energy in mobile scenarios, contrasting with
environments. the very good performance in static scenarios (the
In a first set of simulations, the sensor nodes remaining energy was 83% of the initial energy).
were randomly distributed over a square area
(1000mx1000m), they remained static during the
entire simulation runs (300 sec) and the only moving 100
90
entity was the phenomenon, here simulating a
Remaining Energy (%)

80

moving bird that stimulated the sensors by its chip. 70

60
In the second scenario all the sensors were moving 50

entities, simulating the movement of a group of birds 40 Static AODV

30 Static DSDV

(each sensor representing a moving bird). 20

Mobile AODV
Mobile DSDV

To simulate the movement of birds, the authors 10

0
adapted the boids model, produced by Reynolds 25 50 75 100 125 150 175 200
Nº of Nodes
(http://www.red3d.com/cwr/), to NS-2. This model
simulates the coordinated movement observed in
groups of animals, e.g. birds and fish. The model is Figure 10: Remaining energy vs. network size
based on a decentralized management, where each
individual provides instructions to its neighbours, In terms of packet delivery ratio, the DSDV
and the behaviour of the group results from the protocol, once again compared to the AODV
combined action of each individual. Reynolds protocol, led to better results. Fig. 11 presents the
identified three simple rules that, when followed by number of packet drops per protocol for each
each individual member of a group, result in a scenario. In the mobile scenario packet drops grow

Ubiquitous Computing and Communication Journal 8

exponentially. AODV with 125 nodes already deployed, each with a radio range of 141.4 units
presents 60.000 packet drops, and it stays at this (derived by 2d 2 ), and with up to 8 neighbours. The
level up to 200 nodes due to the energy behaviour phenomenon was simulated by using a moving
registered in the previous study. On the other hand, device that stimulated the sensor nodes every 2
DSDV presented better results in both scenarios. In seconds. After being stimulated by the phenomenon,
static environments the increase of sensor nodes each sensor node forwarded the data to the sink-node
resulted in a relatively small increase in the number using the well know ad hoc protocol AODV. The
of dropped packets. However in the mobile scenario, sink-node was placed in one of the edges of the grid,
an exponential growth was observed and it had no energy restrictions. The simulated time
was 100 seconds and the energy-related parameters
were the same as in the previous study. Different
levels of packet lengths were considered, from 32
bytes to 256 bytes (802.15.4-based devices use 127-
byte packets). The first set of results (Fig. 12)
compares the network remaining energy percentage
as a function of packet length. This was achieved by
summing the remaining energy in all nodes. In the
scenario with 32-byte packets the remaining energy
was 5.6% of the initial energy, contrasting with 3%
when using 256-byte packets.
Figure 11: Dropped packets vs. network size Fig. 13 illustrates the percentage of nodes that, at
the end of the simulation, still had energy levels
All the results suggest the need to create a new capable of monitoring the phenomenon. Due to the
protocol, capable of minimizing the number of lost nodes and sink-node geographic distribution and the
packets and of reducing energy consumption in phenomenon movement, the first sensors achieving
scenarios where the sensor nodes have mobility and null available energy were the devices placed in the
IP support. By analyzing the well known WSN middle of the grid. As the packet length grew, the
mobile protocols (SENMA and MULE [14]) it is number of sensor nodes unable to transmit also
possible to conclude that they present limited increased. The last graph presents packet losses in
capabilities and are not the most adequate to the simulated network (Fig. 14). Losses are mainly
dynamic WMSNs. In the SENMA architecture, the due to the protocol characteristics, but they also
integration of mobile agents, such as little airplanes, reflect the increasing number of dead nodes in the
reduces the WMSN applicability. On the other hand, communication path between the excited node and
the MULE model presents a new element in the the sink-node.
WMSN architecture, with specific characteristics,
which may be difficult to find in the phenomenon 6

environment. Therefore, there exists the need to 5

study new protocols that natively support mobility,

Remaining Power (%)

and from all the WMSN elements perspectives: 3

phenomenon, sensor, network and observer 2

movement. Only by supporting such characteristics 1

will it be possible to make a true integration of the
0
WMSNs in heterogeneous systems. 32 64 96 128 160 192 224 256
Packet Length (Bytes)

4.2 IPv6 In WMSNs

The second study addressed the use of IPv6 in Figure 12: Remaining energy vs. packet length
WMSNs, when compared to the use of IPv4. IPv6
packets have a 40-bytes header (without extension 100
fields), 20 bytes more than the IPv4 header. The
study was carried out both by simulation and by 90
Remaining Nodes (%)

prototyping. 80

In order to use a network topology that 70

minimized the number of lost packets, the sensor
nodes were strategically distributed in a grid layout 60

(1000m x 1000m). This deployment strategy is not 50

realistic. However, the main goal of this simulation 32 64 96 128 160

Packet Length (Bytes)
192 224 256

was to study the advantages and drawbacks of IPv6

in WMSNs, in a controlled environment. The grid
spacing was 100 units (d) and 100 sensor nodes were Figure 13: Remaining nodes vs. packet length

Ubiquitous Computing and Communication Journal 9

 1400 In what concerns energy consumption, it is
1200 possible to observe that the results obtained in the
prototype implementation do not differ from the ones
Nº of Dropped Packets

1000

800 obtained by simulation. In wireless environments,

600 IPv6 leads to an excess energy consumption of only
400 1% when compared to IPv4. In the wired scenario
200 the difference is even lower.
0
32 64 96 128 160 192 224 256
Packet reception is also different for both
Packet Length (Bytes) technologies. When connected via SLIP, packets are
sent without any validation or control, which causes
them to arrive with errors at the receiver in the final
Figure 14: Dropped packets vs. packet length
portion of the test. When connected via wireless,
packets either arrive correctly or are lost.
The use of an additional 32 bytes (i.e. the 32-64
From the results presented above we can
bytes interval) represents an increase of 9.4% in the
conclude that the use of IPv6 does not have a
packet losses, at the end of the simulation. Such
decisive effect on the ESB platform. All values
difference in packet length leads to a decrease in the
obtained have marginal differences, perfectly
remaining energy of only 4% of the initial energy.
acceptable when compared to the advantages IPv6
For the prototype, Embedded Sensor Boards
can bring to wireless sensor networks.
(ESB) (http://www.scatterweb.net) nodes were used,
running the Contiki operating system 4.3 Ant Colony Route Optimization Testing
(http://www.sics.se/~adam/contiki/). In this study, This section presents the experimental results
sensor nodes were excited by a phenomenon that obtained in the tests made to the Ant Colony Route
moved in a uniform way. Each time a sensor Optimisation (ACRO) protocol presented in section
detected movement an UDP packet was sent to the 3.3. In a WMSN, energy is a vital resource. This is
sink-node, in a point-to-point communication. Two why it is necessary to use a protocol that, in addition
types of connections were used: wired connection to being compatible with IP, is energy-efficient. In
using SLIP; and wireless connection. this study we compared the ACRO protocol with the
Fig. 15 and 16 present the results of the IP-aware well-known ad hoc protocols AODV and
experience in terms of number of sent packets and DSDV.
battery time, respectively. As can be seen, the In order to better understand the efficiency of
wireless communication case leads to better results, ACRO, two test cases were used. Both simulations
mainly due to the low energy consumption of the used the same node configuration, already presented
TR1001 transceiver used in this case, and also the in earlier tests. However, in order to have a more
relatively high number of SLIP control packets. realistic scenario, the nodes were charged with
different initial energy levels, namely 100, 75, and
2200 50 J (one third of the nodes were charged with each
IPv4
2000 IPv6
of the three energy levels). In each simulation run
different network sizes were used, from small
Packets (Thousands)

1800
networks with 10 nodes up to 100 nodes per network.
1600 In all cases nodes were deployed in a random fashion
1400 (in a 600mx600m field), since in most real WSN
1200
deployments sensor locations cannot be controlled
by an operator. The location of the phenomenon and
1000
Wireless SLIP the sink-node were not known. Nodes were
Link responsible for monitoring the phenomenon and
sending the relevant sensor data to the sink-node.
Figure 15: Number of packets sent by IPv4 vs. IPv6 Four metrics were used to compare the energy
180
performance of the protocols under study: average
IPv4
170
final energy (the average of the nodes’ energy at the
IPv6
160 end of simulation); minimum final energy (the
energy of the node with the lowest energy at the end
Time (hours)

150
140
of simulation); final energy standard deviation (the
130
120
standard deviation of the various energy levels of the
110
nodes); and energy efficiency (the ratio between the
100 total final energy and the number of data packets
Wireless SLIP
received by the sink-node).
Link
The first scenario simulates a randomly deployed
WMSN monitoring a static phenomenon, which
Figure 16: Battery time

Ubiquitous Computing and Communication Journal 10

excites one of the sensor nodes at 30000 bits/sec for adopting a flexible algorithm, such as ACRO, it is
100 seconds. At the end of the simulation, different possible to create routing paths based several criteria,
numbers of packets were delivered to the sink-node including the network characteristics and ambient
by each of the three routing protocols. ACRO conditions. Therefore, by creating a system capable
presented always the best ratio between the total to be adaptable to several conditions, we intend to
final energy and the number of delivered packets, as increase the lifetime of the WMSN, without the need
illustrated in Fig. 17 d). This is in agreement with the of human interaction.
values obtained for the average energy parameter:
once again ACRO presents the best results (Fig. 17 ACKNOWLEDGEMENT
a)), and the gain is more visible in bigger networks The work presented in this paper is partially financed
(e.g. 50 to 100 nodes). In networks where the energy by the Portuguese Foundation for Science and
level varies it is important that the used protocol has Technology, FCT, under the 6MNet
some form of balancing these levels, by using nodes POSI/REDES/44089/2002 project, and by the IST
with more energy more often than nodes with less FP6 CONTENT Network of Excellence (IST-FP6-
energy. The better performance of ACRO is also 0384239).
apparent in Fig. 17 c) (standard deviation), and Fig.
17 b) (minimum energy). 6 REFERENCES
The second study introduces mobility to the [1] S. Ata, M. Hashimoto, H. Kitamura, M. Murata,
phenomenon. Therefore, the nodes were excited by a ‘Mobile IPv6-based Global Anycasting’, draft-
mobile device. Once again, node and sink locations ata-anycast-mip6-00 (2006).
were unknown. The phenomenon moved randomly [2] M. Zuniga, B. Krishnamachari, ‘Integrating
in the field. In this scenario, the difference in the Future Large-scale Wireless Sensor Networks
final results, presented in Fig. 18, became smaller. with the Internet’ Department of Electrical
However, the ACRO protocol still led to the best Engineering, University of Southern California.
results in all categories. Moreover, similarly to what http://ceng.usc.edu/~bkrishna/research/papers/Zu
happened in the previous study, in this simulation the nigaKrishnamachari_sensorIP.pdf, (2005)
difference between the protocols’ performance [3] C. Chong, S. Kumar, ‘Sensor Networks
increased with the network size. Evolution, Opportunities,and Challenges’,
These results show that ACRO has clear Proceedings of The IEEE (2003).
advantages over the well-known AODV and DSDV [4] M. Dorigo, M. Caro, ‘Ant Colony System: A
routing protocols, in what concerns energy efficiency Cooperative Learning Approach to the
and packets delivery ratio. Travelling Salesman Problem’, IEEE
Transactions on Evolutionary Computation
5 CONCLUSIONS (1996).
[5] A. Dunkels, ‘Full TCP/IP for 8-bit architectures’,
In this paper we addressed several issues in Proceedings of The First Internacional
currently influencing the effective deployment of Conference on Mobile Systems, Applications,
Wireless Mesh Sensor Networks. Some of these and Services (MOBISYS03) (2003).
issues result from misconceptions or clichés that are [6] A. Dunkels, J. Alonso, T. Voigt, H. Ritter, J.
or will shortly become outdated. Schiller, ‘Connecting Wireless Sensornets with
We argued that WMSNs should have global TCP/IP Networks’, wwic2004 (2004).
connectivity, through IPv6 deployment in Sensor [7] J. Hill, M. Hoton, R. Kling L. Krishnamurthy,
Routers, mobility support (also through Sensor ‘The Platforms enabling Wireless Sensor
Routers), multiple addressing schemes (data-centric, Networks’, Communications of the ACM, vol.
node-centric, and location-centric), auto- 47, no. 6, pp. 41-46 (2004).
configuration, and energy-efficient routing. These [8] C. Intanagonwiwat, R. Govindan, D. Estrin,
features were combined in a proposed model, called ‘Directed Diffusion: A Scalable and Robust
IPSense. We have shown that these features are Communication Paradigm for Sensor Networks’,
worth exploring and are feasible, by simulating and Proc. ACM MobiCom’00, Boston, MA, pp. 56-
implementing them. The tests addressed the use of 67 (2000).
IP-based approaches in WSNs, the impact of IPv6 [9] L. Krishnamurthy, R. Adler, P. Buonadonna, M.
when compared to IPv4, and the energy and Yarvis, ‘Design and deployment of industrial
transmission performance of the Ant Colony Route sensor networks: experiences from a
Optimisation protocol relative to other well-known semiconductor plant and the north sea’. In
WSN routing protocols. The results have shown that SenSys ’05: Proceedings of the 3rd international
the proposed model has clear benefits with low cost. conference on Embedded networked sensor
Our future work will address new methods to adjust systems, pages 64–75 (2005).
WMSNs characteristics/capabilities with the [10] S. Mishra, R. Sridharan, R. Sridhar, ‘A Robust
environment (deployment scenario) variables. By Header Compression Technique for Wireless ad

Ubiquitous Computing and Communication Journal 11

 hoc Networks’, in MobiHoc2003 (2003). MULEs: Modeling and analysis of a three-tier
[11] G. Montenegro, N. Kushalnagar, ‘Transmission architecture for sparse sensor networks",
of IPv6 Packets over IEEE 802.15.4 Networks’, Elsevier Ad Hoc Networks Journal, vol. 1, issues
draft-montenegro-lowpan-ipv6-over-802.15.4-12, 2-3 (2003);
(2007). [15] R. Szewczyk, E. Osterweil, J. Polastre, D. Estrin,
[12] D. Niculescu, ‘Communication Paradigms for ‘Habitat Monitoring with Sensor Neworks’,
Sensor Networks’, IEEE Communications Communication of the ACM (2004).
Magazine (2005). [16] S. Thomson, T. Narten, T. Jinmei ‘“IPv6
[13] P. Ramanathan, ‘Location-centric Approach for Stateless Address Autoconfiguration’, draft-ietf-
Collaborative Target Detection, Classification, ipv6-rfc2462bis-08 (2005).
and Tracking’, IEEE CAS Workshop on [17] R. Wakikawa, A. Petrescu, P. Thubert, ‘Network
Wireless Communication and Networking Mobility (NEMO) Basic Support Protocol
(2002). RFC3963 (2005).
[14] R. Shah, S. Roy, S. Jain, W. Brunette ‘Data
100 100

95 90

Minimum Energy (%)

Average Energy (%)

90 80

85 70

80 60
ACRO
ACRO
50 AODV
75 AODV
DSDV
DSDV
40
70
10 20 30 40 50 60 70 80 90 100
10 20 30 40 50 60 70 80 90 100
Nº Nodes Nº Nodes

a) Average Energy b) Minimum Energy

38 7
ACRO ACRO
36
AODV 6 AODV
Standard Deviation (%)

34
DSDV DSDV
Energy Efficiency

5
32
30 4

28 3
26
2
24
1
22
20 0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Nº Nodes Nº Nodes

c) Standard Deviation d) Energy Efficiency

Figure 17: Energy performance comparison – static phenomenon

100 100

95 90
Minimum Energy (%)
Average Energy (%)

90 80

85 70

80 60
ACRO
ACRO
50 AODV
75 AODV
DSDV
DSDV
40
70
10 20 30 40 50 60 70 80 90 100
10 20 30 40 50 60 70 80 90 100
Nº Nodes Nº Nodes

a) Average Energy b) Minimum Energy

38 7
ACRO ACRO
36
AODV 6 AODV
Standard Deviation (%)

34
DSDV DSDV
Energy Efficiency

5
32
30 4

28 3
26
2
24
1
22
20 0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Nº Nodes Nº Nodes

c) Standard Deviation d) Energy Efficiency

Figure 18: Energy performance comparison – moving phenomenon

Ubiquitous Computing and Communication Journal 12