The Design Space of Wireless Sensor Networks
∗
Kay R
¨
omer and Friedemann Mattern
Institute for Pervasive Computing
ETH Zurich
{roemer|mattern}@inf.ethz.ch
Abstract
In the recent past, wireless sensor networks have found their
way into a wide variety of applications and systems with
vastly varying requirements and characteristics. As a con-
sequence, it is becoming increasingly difficult to discuss
typical requirements regarding hardware issues and soft-
ware support. This is particularly problematic in a multi-
disciplinary research area such as wireless sensor networks,
where close collaboration between users, application do-
main experts, hardware designers, and software developers
is needed to implement efficient systems. In this paper we
discuss the consequences of this fact with regard to the de-
sign space of wireless sensor networks by considering its
various dimensions. We justify our view by demonstrating
that specific existing applications occupy different points in
the design space.
1 Introduction
In April 2004, the authors organized a workshop, funded
by the European Science Foundation (ESF), with a view
to carrying out coordinated research into wireless sensor
networks in Europe [18]. 24 experts from 11 European
countries including academic researchers and representa-
tives from industry were invited to discuss application ar-
eas with particular relevance for Europe as well as various
aspects of the hardware and software architectures required
to support these applications. Some of the more concrete
questions discussed at the workshop were:
• Which prospective application domains and concrete
applications are of particular value to Europe? What
are the requirements and challenges involved in imple-
menting these applications?
• What hardware requirements are needed to support
these applications? Are existing systems sufficient, or
is there a gap that needs additional research and devel-
opment?
∗
This work was partly supported by NCCR-MICS, a center supported
by the Swiss National Science Foundation under grant no. 5005-67322.
To appear in IEEE Wireless Communications, Dec. 2004.
• What type of software is needed (e.g., operating sys-
tems, programming abstractions, tools) to support
these applications and what requirements have to be
met?
• How can we better coordinate the mostly isolated and
disconnected research activities on sensor networks
across Europe?
During the discussions it was observed that wireless sen-
sor networks have found their way into a wide variety of
applications and systems with vastly varying requirements
and characteristics, and hence it was very difficult to dis-
cuss specific application requirements, research directions,
and challenges. In the past, a number of early, mostly US-
based research projects established a de facto definition of
a wireless sensor network as a large-scale ad hoc, multi-
hop, unpartitioned network of largely homogeneous, tiny,
resource-constrained, mostly immobile sensor nodes that
would be randomly deployed in the area of interest. While
this characterization is certainly valid for a large class of
applications (in particular from the military domain), an in-
creasing number of sensor-network applications cannot be
adequately characterized in this way.
As a result of this observation, it was suggested that the
sensor network design space and its various dimensions
should be characterized. Such an explicit design space
might not only prove helpful as a framework for discussing
and structuring coordinated research (e.g., analyzing mu-
tual dependencies between applications, software, and hard-
ware; avoiding duplicate work), but might also provide a
conceptual basis for the development of flexible software
frameworks that can be adapted to meet different applica-
tion needs.
This paper is a partial answer to the questions raised dur-
ing the above-mentioned workshop. We make an attempt to
specify important dimensions of the sensor network design
space and we justify our findings by showing that existing
sensor network applications occupy different points in the
design space. We build on earlier work [16] that classified
system models of sensor networks with respect to commu-
nication protocols but did not consider the diverse nature of
concrete applications.
1
2 Design Space
Initial research into wireless sensor networks was mainly
motivated by military applications, with DARPA continu-
ing to fund a number of prominent research projects (e.g.,
Smart Dust, NEST) that are commonly regarded as the
cradle of sensor network research. The type of applica-
tions considered by these projects led to a de facto defi-
nition of a wireless sensor network as a large-scale (thou-
sands of nodes, covering large geographical areas), wire-
less, ad hoc, multi-hop, unpartitioned network of homoge-
neous, tiny (hardly noticeable), mostly immobile (after de-
ployment) sensor nodes that would be randomly deployed
in the area of interest.
More recently, other, civilian application domains of
wireless sensor networks have been considered, such as en-
vironmental and species monitoring, agriculture, produc-
tion and delivery, healthcare, etc. (see Section 3). Concrete
projects targeting these application areas indicate that the
above definition of a wireless sensor network does not nec-
essarily apply for these applications – networks may con-
sist of heterogeneous and mobile sensor nodes, the network
topology may be as simple as a star topology, networks
may make use of existing communication infrastructures,
etc. To meet this general trend towards diversification, we
will discuss important dimensions of the sensor network de-
sign space in the following subsections. We will informally
characterize each of the dimensions and, where appropri-
ate, identify (possibly orthogonal) property classes in order
to support a coarse-grained classification of sensor network
applications.
It is certainly debatable which issues are important
enough to be explicitly considered as dimensions in the de-
sign space and one could argue in favor of adding more di-
mensions or removing some from our suggestions detailed
below. In fact, we expect that this might become reason-
able in the future as the field and its applications evolve.
However, we have tried to ensure that our initial sugges-
tion consisted of a sensible set of dimensions, by basing our
choice on the following two principles. Firstly, there should
be notable variability between applications with respect to
dimensions. Secondly, a dimension should have a signifi-
cant impact on the design and implementation of technical
solutions.
2.1 Deployment
The deployment of sensor nodes in the physical environ-
ment may take several forms. Nodes may be deployed at
random (e.g., by dropping them from an aircraft) or in-
stalled at deliberately chosen spots. Deployment may be a
one-time activity, where the installation and use of a sensor
network are strictly separate activities. However, deploy-
ment may also be a continuous process, with more nodes
being deployed at any time during the use of the network –
for example, to replace failed nodes or to improve coverage
at certain interesting locations.
The actual type of deployment affects important proper-
ties such as the expected node density, node locations, reg-
ular patterns in node locations, and the expected degree of
network dynamics.
Classes: random vs. manual; one-time vs. iterative.
2.2 Mobility
Sensor nodes may change their location after initial deploy-
ment. Mobility can result from environmental influences
such as wind or water, sensor nodes may be attached to
or carried by mobile entities, and sensor nodes may pos-
sess automotive capabilities. In other words, mobility may
be either an incidental side effect, or it may be a desired
property of the system (e.g., to move nodes to interesting
physical locations), in which case mobility may be either
active (i.e., automotive) or passive (e.g., attached to a mov-
ing object not under the control of the sensor node). Mo-
bility may apply to all nodes within a network or only to
subsets of nodes. The degree of mobility may also vary
from occasional movement with long periods of immobility
in between, to constant travel.
Mobility has a large impact on the expected degree of
network dynamics and hence influences the design of net-
working protocols and distributed algorithms. The actual
speed of movement may also have an impact, for example
on the amount of time during which nodes stay within com-
munication range of each other.
Classes: immobile vs. partly vs. all; occasional vs. con-
tinuous; active vs. passive.
2.3 Cost, Size, Resources, and Energy
Depending on the actual needs of the application, the form
factor of a single sensor node may vary from the size of a
shoe box (e.g., a weather station) to a microscopically small
particle (e.g., for military applications where sensor nodes
should be almost invisible). Similarly, the cost of a single
device may vary from hundreds of Euros (for networks of
very few, but powerful nodes) to a few cents (for large-scale
networks made up of very simple nodes).
Since sensor nodes are untethered autonomous devices,
their energy and other resources are limited by size and cost
constraints. Varying size and cost constraints directly re-
sult in corresponding varying limits on the energy available
(i.e., size, cost, and energy density of batteries or devices
for energy scavenging), as well as on computing, storage,
and communication resources. Hence, the energy and other
resources available on a sensor node may also vary greatly
from system to system. Power may be either stored (e.g., in
batteries) or scavenged from the environment (e.g., by solar
cells).
These resource constraints limit the complexity of the
software executed on sensor nodes. For our classification,
2
we have partitioned sensor nodes roughly into four classes
based on their physical size.
Classes: brick vs. matchbox vs. grain vs. dust.
2.4 Heterogeneity
Early sensor network visions anticipated that sensor net-
works would typically consist of homogeneous devices that
were mostly identical from a hardware and software point
of view. Some projects, such as Amorphous Computing [1],
even assumed that sensor nodes were indistinguishable, that
is, they did not even possess unique addresses or IDs within
their hardware. This view was based on the observation that
otherwise it would not be feasible to cheaply produce vast
quantities of sensor nodes.
However, in many prototypical systems available to-
day, sensor networks consist of a variety of different de-
vices. Nodes may differ in the type and number of attached
sensors; some computationally more powerful “compute”
nodes may collect, process, and route sensory data from
many more limited sensing nodes; some sensor nodes may
be equipped with special hardware such as a GPS receiver to
act as beacons for other nodes to infer their location; some
nodes may act as gateways to long-range data communica-
tion networks (e.g., GSM networks, satellite networks, or
the Internet).
The degree of heterogeneity in a sensor network is an im-
portant factor since it affects the complexity of the software
executed on the sensor nodes and also the management of
the whole system.
Classes: homogeneous vs. heterogeneous.
2.5 Communication Modality
For wireless communication among sensor nodes, a number
of communication modalities can be used such as radio, dif-
fuse light, laser, inductive and capacitive coupling, or even
sound.
Perhaps the most common modality is radio waves, since
these do not require a free line of sight, and communica-
tion over medium ranges can be implemented with rela-
tively low power consumption and relatively small anten-
nas (a few centimeters in the common sub-GHz frequency
bands). Using light beams for communication requires a
free line of sight and may interfere with ambient light and
daylight, but allows for much smaller and more energy-
efficient transceivers compared to radio communication.
Smart Dust [7], for example, uses laser beams for communi-
cation. Inductive and capacitive coupling only works over
small distances, but may be used to power a sensor node.
Most passive Radio Frequency Identification (RFID) sys-
tems use inductive coupling, for example. Sound or ultra-
sound is typically used for communication under water or
to measure distances based on time-of-flight measurements.
Sometimes, multiple modalities are used by a single sensor
network system.
The communication modality used obviously influences
the design of medium access protocols and communication
protocols, but also affects other properties that are relevant
to the application.
Classes: radio vs. light vs. inductive vs. capacitive vs.
sound.
2.6 Infrastructure
The various communication modalities can be used in dif-
ferent ways to construct an actual communication network.
Two common forms are so-called infrastructure-based net-
works on the one hand and ad hoc networks on the other
hand. In infrastructure-based networks, sensor nodes can
only directly communicate with so-called base station de-
vices. Communication between sensor nodes is relayed via
the base station. If there are multiple base stations, these
have to be able to communicate with each other. The num-
ber of base stations depends on the communication range
and the area covered by the sensor nodes. Mobile phone
networks and Smart Dust [7] are examples of this type of
network.
In ad hoc networks, nodes can directly communicate with
each other without an infrastructure. Nodes may act as
routers, forwarding messages over multiple hops on behalf
of other nodes.
Since the deployment of an infrastructure is a costly pro-
cess, and the installation of an infrastructure may often not
be feasible, ad hoc networks are preferred for many appli-
cations. However, if an infrastructure is already available
anyway (such as the GSM network), it might also be used
for certain sensor network applications.
Combinations of ad hoc networks and infrastructure-
based networks are sometimes used, where clusters of sen-
sor nodes are interconnected by a wide area infrastructure-
based network.
Note that the above arguments not only apply to commu-
nication, but also to other infrastructures, such as localiza-
tion or time synchronization (e.g., GPS satellites).
Classes: infrastructure vs. ad hoc.
2.7 Network Topology
One important property of a sensor network is its diame-
ter, that is, the maximum number of hops between any two
nodes in the network. In its simplest form, a sensor net-
work forms a single-hop network, with every sensor node
being able to directly communicate with every other node.
An infrastructure-based network with a single base station
forms a star network with a diameter of two. A multi-hop
network may form an arbitrary graph, but often an overlay
network with a simpler structure is constructed such as a
tree or a set of connected stars.
The topology affects many network characteristics such
as latency, robustness, and capacity. The complexity of data
routing and processing also depends on the topology.
3
Classes: single-hop vs. star vs. networked stars vs. tree
vs. graph.
2.8 Coverage
The effective range of the sensors attached to a sensor node
defines the coverage area of a sensor node. Network cover-
age measures the degree of coverage of the area of interest
by sensor nodes. With sparse coverage, only parts of the
area of interest are covered by the sensor nodes. With dense
coverage, the area of interest is completely (or almost com-
pletely) covered by sensors. With redundant coverage, mul-
tiple sensors cover the same physical location. The actual
degree of coverage is mainly determined by the observa-
tion accuracy and redundancy required. Coverage may vary
across the network. For example, nodes may be deployed
more densely at interesting physical locations.
The degree of coverage also influences information-
processing algorithms. High coverage is a key to robust
systems and may be exploited to extend the network life-
time by switching redundant nodes to power-saving sleep
modes.
Classes: sparse vs. dense vs. redundant.
2.9 Connectivity
The communication ranges and physical locations of indi-
vidual sensor nodes define the connectivity of a network. If
there is always a network connection (possibly over multi-
ple hops) between any two nodes, the network is said to be
connected. Connectivity is intermittent if the network may
be occasionally partitioned. If nodes are isolated most of
the time and enter the communication range of other nodes
only occasionally, we say that communication is sporadic.
Note that despite the existence of partitions, messages may
be transported across partitions by mobile nodes.
Connectivity mainly influences the design of communi-
cation protocols and methods of data gathering.
Classes: connected vs. intermittent vs. sporadic.
2.10 Network Size
The number of nodes participating in a sensor network is
mainly determined by requirements relating to network con-
nectivity and coverage, and by the size of the area of inter-
est. The network size may vary from a few nodes to thou-
sands of sensor nodes or even more. The network size deter-
mines the scalability requirements with regard to protocols
and algorithms.
2.11 Lifetime
Depending on the application, the required lifetime of a sen-
sor network may range from some hours to several years.
The necessary lifetime has a high impact on the required
degree of energy efficiency and robustness of the nodes.
2.12 Other QoS Requirements
Depending on the application, a sensor network must sup-
port certain quality-of-service aspects such as real-time con-
straints (e.g., a physical event must be reported within
a certain period of time), robustness (i.e., the network
should remain operational even if certain well-defined fail-
ures occur), tamper-resistance (i.e., the network should
remain operational even when subject to deliberate at-
tacks), eavesdropping-resistance (i.e., external entities can-
not eavesdrop on data traffic), unobtrusiveness or stealth
(i.e., the presence of the network must be hard to detect).
These requirements may impact on other dimensions of the
design space such as coverage and resources.
3 Applications
In this section we justify our design space model by locat-
ing a number of applications at different points in the de-
sign space. For this, we have selected concrete applications
that are well-documented and that have advanced beyond a
mere vision. Some of the applications listed are field exper-
iments, some are commercial products, and some are ad-
vanced research projects that use sensor networks as a tool.
For classification, we have used the reported parameters that
were actually used in practical settings and we have deliber-
ately refrained from speculation as to what else could have
been done.
Note that there are usually different technical solutions
for a single application, which means that the concrete
projects described below are only examples drawn from a
whole set of possible solutions. However, these examples
reflect what was technically possible and desirable at the
time the projects were set up. Therefore, we have decided
to base our discussion on these concrete examples rather
than speculating about the inherent characteristics of a cer-
tain type of application. Table 1 classifies the sample ap-
plications according to the dimensions of the design space
described in the previous section.
3.1 Bird Observation on Great Duck Island
A wireless sensor network (WSN) is being used to observe
the breeding behavior of a small bird called Leach’s Storm
Petrel [9] on Great Duck Island, Maine, USA. These birds
are easily disturbed by the presence of humans, hence WSN
seems an appropriate way of better understanding their be-
havior. The breeding season lasts for seven months from
April to October. The biologists are interested in the usage
pattern of their nesting burrows, changes in environmental
conditions outside and inside the burrows during the breed-
ing season, variations among breeding sites, and the param-
eters of preferred breeding sites.
Sensor nodes are installed inside the burrows and on the
surface. Nodes can measure humidity, pressure, tempera-
ture, and ambient light level. Burrow nodes are equipped
4
Deployment Mobility Resources Cost Energy Heterogeneity Modality Infrastructure Topology Coverage Connectivity Size Lifetime QoS
Great Duck manual,
onetime
immobile matchbox ∼ 200 USD battery, solar weather
stations,
burrow
nodes,
gateways
radio base station,
gateways
star of
clusters
dense (every
burrow)
connected tens –
hundreds (∼
100
deployed)
7 months
(breeding
period)
–
ZebraNet manual,
one-time
all,
continuous,
passive
matchbox – battery nodes,
gateway
radio base station,
GPS
graph dense (every
animal)
sporadic tens –
hundreds
one year –
Glacier manual,
one-time
all,
continuous,
passive
brick – battery nodes, base
station
radio base station,
GPS, GSM
star sparse connected tens –
hundreds (9
deployed)
several
months
–
Herding manual,
one-time
all,
continuous,
passive
brick ∼ 1000
USD
battery homogeneous radio base station,
GPS
graph dense (every
cow)
intermittent up to
hundreds (10
deployed)
days to
weeks
–
Bathymetry manual,
one-time
all,
occasional,
passive
brick – battery homogeneous radio GPS graph sparse (0.5 –
1km apart)
connected up to
hundreds (6
deployed, 50
planned)
several
months
–
Ocean random,
iterative
all,
continuous,
passive
brick ∼ 15000
USD
battery homogeneous radio satellite star sparse intermittent 1300
deployed,
3000
planned
4-5 years –
Grape manual,
one-time
immobile matchbox ∼ 200 USD battery sensors,
gateway,
base station
radio base station tree
(two-tiered
multi-hop)
sparse (20m
apart)
connected up to
hundreds (65
deployed)
several
months
(growth
period)
–
Cold Chain manual,
iterative
partly
(sensors),
occasional,
passive
matchbox
(sensors),
brick
(relays)
– battery sensors,
relays,
access
boxes,
warehouse
radio relays,
access boxes
tree
(three-tiered
multi-hop)
sparse intermittent up to
hundreds (55
sensors, 4
relays
deployed)
years –
Avalanche manual,
one-time
all,
continuous,
passive
matchbox – battery homogeneous radio rescuer’s
PDA
star dense (every
person)
connected tens –
hundreds
(number of
victims)
days
(duration of
a hike)
dependability
Vital Sign manual all,
continuous,
passive
matchbox – battery medical
sensors,
patient
identifier,
display
device, setup
pen
radio, IR
light (for
setup pen)
ad hoc single-hop dense connected tens days to
months
(hospital
stay)
real-time,
dependabil-
ity,
eaves-
dropping-
resistance
Power manual,
iterative
immobile matchbox – power grid sensor
nodes,
transceivers,
central unit
radio (sensor
unidirec-
tional)
transceivers layered
multi-hop
sparse
(selected
outlets)
connected tens –
hundreds
years
(building
lifecycle)
–
Assembly manual,
one-time
all,
occasional,
passive
matchbox ∼ 100 Euro battery different
sensors
radio ad hoc star sparse connected tens hours
(duration of
assembly)
–
Tracking random
(thrown
from
aircraft)
all,
occasional,
passive
matchbox ∼ 200 USD battery homogeneous radio UAV graph sparse intermittent
(UAV)
tens –
thousands (5
deployed)
weeks –
years
(conflict
duration)
stealth,
tamper-
resistance,
real-time
Mines manual all,
occasional,
active
brick – battery homogeneous radio,
ultrasound
(for
localization)
ad hoc graph dense connected up to
hundreds (20
deployed)
months –
years
tamper-
resistance
Sniper manual immobile matchbox
with FPGA
∼ 200 USD battery homogeneous radio ad hoc graph redundant
(multiple
nodes
recognize
shot)
connected up to
hundreds (60
deployed)
months –
years
real-time
Table 1: Classification of the sample applications according to the design space.
5
