WSN Overview

7/28/2019 WSN Overview

1/4

20 0278-6648/03/$17.00 2003 IEEE IEEE POTENTIALS

dvances in hardwareand wireless networktechnologies havecreated low-cost,low-power, multi-functional miniature

sensor devices. These devices make uphundreds or thousands of ad hoc tinysensor nodes spread across a geographi-cal area. These sensor nodes collaborate

among themselves to establish a sens-ing network. A sensor network that canprovide access to information anytime,anywhere by collecting, processing,analyzing and disseminating data. Thus,the network actively participates in cre-ating a smart environment.

Sensor networks promise to revolu-tionize sensing in a wide range of appli-cation domains. This is because of theirreliability, accuracy, flexibility, cost-effectiveness and ease of deployment,according to Tilaket al. Smart sensors

can offer vigilant surveillance and candetect and collect data concerning anysign of machine(s) failure, earthquakes,floods and, even, a terrorist attack.Sensor networks enable: 1) informationgathering, 2) information processing,and 3) reliable monitoring of a varietyof environments for both civil and mili-tary applications.

The architecture of the sensor nodeshardware consists of five components:sensing hardware, processor, memory,power supply and transceiver. These

devices are easily deployed because noinfrastructure and human control areneeded. They sense, compute and actu-ate into the physical environments. Theycan self-organize and can adapt to sup-port several applications.

Each sensor node has wireless com-munication capability and sufficientintelligence for signal processing andfor disseminating the data. The limitedenergy, computational power, and com-munication resources of a sensor noderequires the use of a huge number ofsensor nodes in a wider region. Thislarge number also allows the sensornetwork to report with greater accuracythe exact speed, direction, size, andother characteristics of a moving objectthan is possible with a single sensor.

Since sensor networks have a largenumber of sensor nodes, the cost of a sin-gle node is very important to justify theoverall cost of the network. The cost of asensor node should be much less than $1(USD) in order for the sensor network tobe feasible according to Akyildiz et al.

Communication in sensor networks

is not typically end to end. Energy istypically more limited in sensor net-works than in other wireless networksbecause of the nature of the sensingdevices and the difficulty in rechargingtheir batteries. Lastly, studies haveshown that current commercialBluetooth devices are unsuitable forsensor network applications because oftheir energy requirements, state Tilaket

al and expected higher costs than sen-sor nodes (Akyildiz et al).

Intuitively, a denser infrastructurewould lead to a more effective sensornetwork. It can provide higher accuracyand has a larger aggregate amount ofenergy available. However, if not prop-erly managed, a denser network can

also lead to a larger number of colli-sions and potentially to congestion inthe network; this will increase latencyand reduce energy efficiency.Moreover, the large number of samplesreported by the sensors may vastlyexceed the data information required.

Examples ofpossible applications

Detecting environmental hazards,monitoring remote terrain, or even cus-tomer behavior surveillance are amongmany sensor network applications.Researchers are trying to adopt sensornetwork technology to problems hard tosolve with conventional wireless net-working.

Some examples are the following: Sensors are deployed to analyze

remote locations (the motion of a torna-do, fire detection in a forest);

Sensors are attached to taxi cabs ina large metropolitan area to study thetraffic conditions and plan routes effec-tively;

Wireless parking lot sensor net-works that determine which spots areoccupied and which spots are free;

Wireless surveillance sensor net-

works for providing security in a shop-ping mall, parking garage or at someother facility;

Military sensor networks to detect,locate or track enemy movements, and

Sensor networks can increase alert-ness to potential terrorist threats.

A hierarchical sensor networkWe depict a sensor network exam-

ple in military terms to show how sen-sors cooperate among themselves andhow they disseminate and aggregate

the data.The tactical military network archi-

tecture (see Fig. 1) consists of a groupof units (i.e., clusters) managed by com-manders (i.e., parent nodes). Thesecommanders receive orders from head-quarters (i.e., the sink node) and, inreturn, send back their report.

The commanders send the orderreceived from headquarters to their gen-erals (i.e., cluster heads). Every generalis responsible for a group of soldiers(i.e., children) in a unit. Soldiers com-

municate locally (i.e., within a unit)with their counterparts or their general.Soldiers in a unit cannot communicatewith generals from other units whereasgenerals can only communicate amongthemselves. After hearing the messagesfrom their soldiers, generals send theirobservations to their commanders.

In a battlefield, soldiers in a unitcontact their general to notify the gen-eral about a specific observation in theirunit. The general, then, can issue anorder to his soldiers to take an actionregarding their observation, or can con-tact his commander for an opinion. Incase of decisive actions, such as anattackcommand, only headquarters canorder a decisive action based on theinformation from the commanders.

Sensor network challengesChallenges in hardware design,

communication protocols and applica-tions design face sensor network tech-nology to make it a reality. Extendingthe lifetime of the sensor network andbuilding an intelligent data collecting

Sensor

networks:

an overview

Malik Tubaishat andSanjay Madria

collectively sendingback information

A

Authorized licensed use limited to: LIVERPOOL JOHN MOORES UNIVERSITY. Downloaded on February 20, 2009 at 03:56 from IEEE Xplore. Restrictions apply.


2/4

APRIL/MAY 2003 21

systems are two. Other challengesinclude:

Sensor networks topologychanges very frequently;

Sensors use a broadcast communi-cation paradigm whereas most net-works are based on point-to-point com-

munications; Sensors are very limited in power,computational capacities and memory;

Sensors are very prone to failures; Sensors may not have global iden-

tification (ID) because of the largeamount of overhead;

Sensors are densely deployed inlarge numbers. The problem can beviewed in terms of collision and conges-tion. To avoid collisions, sensors that arein the transmission range of each othershould not transmit simultaneously.

Ad hoc deployment requires thatthe system identifies and copes with theresulting distribution and connectivityof nodes, and

Dynamic environmental condi-tions require the system to adapt overtime to changing connectivity and sys-tem stimuli.

RequirementsSensor network requirements

include the following:Large number of sensors: To make

use of the cheap small-sized sensors,

sensor networks may contain thousandsof nodes. Scalability and managing thesehuge numbers of sensors is a majorissue. Clustering is one solution to thisproblem. In clustering, neighbor sensorsjoin to build one cluster (group) and electa cluster head to manage this group.

Low energy use: In many applica-tions, the sensor nodes will be deployedin a remote area in which case servic-ing a node may not be possible. Thus,the lifetime of a node may be deter-mined by the battery life, therebyrequiring minimal energy expenditure.(Recharging a large number of sensorbatteries would be expensive and timeconsuming.)

Efficient use of the small memory:When building sensor networks, issuessuch as routing-tables, data replication,security and such should be consideredto fit the small size of memory in thesensor nodes.

Data aggregation: The huge num-ber of sensing nodes may congest thenetwork with information. To solve thisproblem, some sensors such as the clus-ter heads can aggregate the data, dosome computation (e.g., average, sum-mation, highest, etc.), and then broad-cast the summarized new information.

Network self-organization: Given thelarge number of nodes and their potentialplacement in hostile locations, it is essen-

tial that the network be able to self-orga-nize itself. Moreover, nodes may fail(either from lack of energy or from phys-ical destruction), and new nodes mayneed to join the network. Therefore, thenetwork must be able to periodicallyreconfigure itself so that it can continue

to function. Individual nodes maybecome disconnected from the restof the network, but a high degree of

connectivity overall must be main-tained.

Collaborative signal process-ing: Yet another factor that distin-guishes these networks fromMobi le Ad-hoc Networks(MANETs) is that the end goal isthe detection/estimation of someevent(s) of interest, and not justcommunication. To improve thedetection performance, it is oftenquite useful to fuse data from mul-tiple sensors. This data fusion

requires the transmission of dataand control messages. This needmay put constraints on the networkarchitecture.

Querying ability: there are twotypes of addressing in sensor net-work; data-centric, and address-centric according toIntanagonwiwat et al. In data-cen-tric, a query will be sent to specificregion in the network. Whereas, inaddressing-centric, the query willbe sent to an individual node.

Potential advantages ofsensor networks over MANET

Although many protocols and algo-rithms have been proposed for tradi-tional wireless ad hoc networks, theyare not well suited to the unique fea-tures and applications requirements ofsensor networks. Yes, sensor nodes areprone to failures and may not haveglobal identification (ID). Still, sensornetworks have many advantages overthe traditional wireless ad hoc network.

Wireless sensor networks improvesensing accuracy by providing distrib-uted processing of vast quantities ofsensing information (e.g., seismicdata, acoustic data, high-resolutionimages, etc.). When networked, sen-sors can aggregate such data to pro-vide a rich, multi-dimensional view ofthe environment;

They can provide coverage of avery large area through the scatteringof thousands of sensors;

Networked sensors can continue tofunction accurately in the face of fail-

Sensed ObjectChild Node Cluster Head Parent Node

Commanders

Headquarters Level 3

Level 1

General

Level 2

Soldier

Fig. 1 Hierarchicalsensor networkarchitecture



3/4

ure of individual sensors. Thus, allow-ing greater fault tolerance through ahigh level of redundancy;

Wireless sensor networks can alsoimprove remote access to sensor databy providing sink nodes that connectthem to other networks, such as theInternet, using wide-area wireless links.

They can localize discrete phe-nomenon to save power consumption;

They can minimize human inter-vention and management;

They can work in hostile and unat-tended environments; and

They can dynamically react tochanging network conditions.

How ad hocsensor networks operate

An ad hoc sensor network is a collec-tion of sensor nodes form-ing a temporary networkwithout the aid of any

central administration orsupport services. In otherwords, there is no station-ary infrastructure such asbase stations.

In general, the sensornodes use wireless radiofrequency (RF) trans-ceivers as their networkinterface and communi-cate with each other usingmulti-hop wireless links.Each sensor node in the

network also acts as arouter, forwarding datapackets for its neighbor nodes.

Ad hoc networks must deal with fre-quent changes in topology. This isbecause sensor nodes are prone to fail-ure and also new sensor nodes may jointhe network to compensate the failednodes or to maximize the area of inter-est. Because of these characteristics, acentral challenge in the design of the adhoc sensor network is the developmentof self-organizing sensor network anddynamic routing protocols that can effi-ciently find routes between two com-municating nodes.

For the tiny sensors to coordinateamong themselves to achieve a largesensing task in a less power consump-tion, they should work in a cluster.Each cluster assigns a cluster head tomanage its sensors. The advantages ofcluster heads are:

Clustering allows sensors to effi-ciently coordinate their local interac-tions in order to achieve global goals;

Scalability;

Improved robustness; More efficient resource utilization; Lower energy consumption; and Robust link or node failures and

network partitionsIn Fig. 2, we show the general archi-

tecture of a sensor network. As shown inthe figure, there are three layers: the ser-vices-layer, the data-layer and the physi-cal-layer. The services include, but are

not restricted to, routing protocol, datadissemination and data aggregation.

The physical-layer consists of thephysical nodes. These nodes are thesinks, children nodes, the cluster headsand the parents. Parent nodes are thoseconnected to two or more cluster heads.All the messages are virtually modeledin the data-layer.

The sink node(s) broadcast a query

either to the entire sensor network ortoward a specific region depending ontype of query used. When the sensornodesclose to the sensed object, detectfor example a change in heat, location,speed, etcthen they broadcast this datato their neighboring sensor nodes.

Since each sensor (i.e., child) is con-nected to at least one cluster head, clus-ter head(s) will eventually receive thisdata. Cluster heads task is to processand aggregate this data and broadcast itto the sink node(s) through the neigh-boring nodes. This is because the clus-ter head receives many data packetsfrom its children. Hence, it is the clus-ter heads task to process and filter thisdata as information.

To compensate the hardware limita-tions in the sensor nodes such as mem-ory, battery and computation power,sensor applications deploy a large num-ber of sensor nodes in the targetedregion. These sensor nodes then collab-orate among themselves to perform as

one big wireless ad hoc network. Theclose distance between the nodes helpsalso in saving power by reducing theradius of transmission for each node.

Data versus address-centricNow we will explain why sensor

networks should be data-centricinstead of address-centric. The princi-ple idea of sensor networks is to

design very cheap and simple sensornodes. This way sensor applicationscan contain thousands of these dispos-able nodes are used without any bur-den. Giving a unique address for eachnode is costly especially when thou-sands of nodes are used in a sensornetwork application.

The limited memory and computa-tional power force one not to depend on

the contents of an indi-vidual sensor nodeitself. Rather, we are

interested in the obser-vation of a group ofsensors.

Data-centric appli-cations focus on datagenerated by sensors.So, instead of sendinga query say to sensor#45, the query will besent to say region #6which is known fromthe Global PositioningSystem (GPS) device

placed on the sensornodes. The idea ofusing GPS to easily locate sensors isvery important when disseminating thedata packet, as we can send the queryto a specific region by the help of theGPS embedded in some sensor nodes.

(Unfortunately, embedded GPS sen-sor nodes can sometimes be misleadingwhen their line of sight is blocked.Further, GPS gives the location withina range (i.e., not exact location). Hence,nodes very close to each other willhave the same GPS result.)

AggregationSome sensor nodes are assigned to

aggregate data received from theirneighbors. Aggregator nodes cancache, process and filter the data tomore meaningful information andresend to the sink nodes. Aggregationis useful for the following reasons:

Increase the circle of knowledge; Increase the level of accuracy; and Data redundancy to compensate

for sensor nodes failing.

22 IEEE POTENTIALS

Data-layer

Services

Information

Data

Query

Cluster Heads

Children Nodes

Sinks

Services-layer

Physical-layer

Fig. 2 Sensor network architecture



4/4

DisseminationData produced by the sensors usual-

ly has to be routed through severalintermediate nodes to reach its destina-tion. Problems arise when intermediatenodes fail to forward incoming mes-sages. Other problems are:

Routing protocol should find theshortest path;

Redundancy: a sensor may receive

the same data packet more than once.In sensor networks, two scenarios

for data dissemination exist: query dri-ven and continuous update. Each sce-nario is applicable to specific types ofsensor applications. The former is usedas a one-to-one relation. That is, thesink broadcasts a query and, in turn,receives from the sensor nodes onereport in response to this query. Forexample, the sink may query the firstpresence of an object such as seeing anenemy tank, or even an animal.

The second scenario is a one-to-many relation. That is, the sink nodebroadcasts a query and receives contin-uous updates for this query. For exam-ple, the sink may query for the direc-tion of a mobile object. In turn, the sen-sors will report to the sink the newlocation of the mobile object. The con-tinuously updated data disseminationscenario has a high rate of energydepletion; but its data is more reliableand accurate than the query driven.This is because more sensors are

involved in the query report.For instance, the sensor nodes in aparking lot network should be individu-ally addressable. This way one can eas-ily determine the locations of all thefree spaces. Another example is placingsensors above every passengers seat inan airplane to detect any unusual move-ment of any passenger. In case of anydanger, a sensor network can collabo-rate to take control of the airplane (e.g.,switching the lights off, closing thepilots cockpit door, etc.).

Last pointFinally, and most important, the

advantage of using these sensors istheir ability to maintain connectivity incase of movement. As these sensors arevery tiny, they are vulnerable to beingaccidentally moved. Hence, sensor net-works should maintain network con-nectivity even if some of their sensorsare moved. For example, sensors locat-ed in a forest may be vulnerable to anykind of mobility (e.g. human, animal,insect, rain, wind).

Read more about it Deborah Estrin, David Culler, Kris

Pister, Gaurav Sukhatme, Connectingthe Physical World with PervasiveNetworks, IE EE Pe rv as iveComputing, Volume 1, Number 1, Jan-Mar 2002.

S. Tilak, N. Abu-Ghazaleh, and W.Heinzelman, A Taxonomy of WirelessMicro-Sensor Network Models, ACM

Mo bi le Comp ut in g andCommunications Review (MC2R),Volume 6, Number 2, April 2002.

D. Estrin, R. Govindan, J.Heidemann, and S. Kumar. Next cen-tury challenges: Scalable coordinationin sensor networks. In the Proceedingsof the fifth annual ACM/IEEE interna-tional conference on mobile comput-ing and networking, pages 263-270,1999.

Joanna Kulik, Wendi RabinerHeinzelman, and Hari Balakrishnan,

Adaptive Protocols for Informat ionDi ss em in at io n in Wi re le ss Se nsorNe tw orks, Proc. 5th ACM/IEEEMobiCom Conference (MobiCom 99),Seattle, WA, August, 1999.

Joanna Kulik, Wendi RabinerHeinzelman, and Hari Balakrishnan,Ne go ti at io n-ba se d Pr ot oc ol s fo rDisseminating Information in WirelessSensor Networks, Proceedings of 5thACM/IEEE MobiCom Conference(MobiCom 1999), Seattle, WA,August, 1999

http://w3.antd.nist.gov/wctg/manet/ Deborah Estrin, Lewis Girod, GregPottie, Mani Srivastava, Instrumentingthe World with Wireless Networks, inproceedings of the internationalConference on Acoustics, Speech andSignal Processing (ICASSP 2001). SaltLake City, Utah, May 2001.

Chalermek Intanagonwiwat,Ramesh Govindan, Deborah Estrin,Direc ted Dif fus ion: A Scalable andRobust Communication Paradigm forSensor Networks, In Proceedings of theSixth Annual International Conferenceon Mobile Computing and Networks(MobiCOM 2000), August 2000.

Ian Akyildiz, Weilian Su, YogeshSankarasubramanian, Erdal Cayirci, ASurvey on Sensor Network, IEEECommunications Magazines, August,2002.

Budhaditya Deb, SudeepBhatnagar and Badri Nath, A TopologyDi sc ov ery Al go ri thm fo r Se nsorNetworks with Applications to NetworkManagement, in IEEE CAS workshop,September 2002.

About the authorsMalik Tubaishat currently is a Ph.D.

student in the Department of ComputerScience at the University of Missouri-Rolla. He received his master degree inComputer Science from the UniversitySains Malaysia, Malaysia in 2000. Heobtained his BSc in Computer Sciencefrom Yarmouk University, Jordan in1994. His research interests include

self-organizing sensor networks, androuting protocols in sensor networks.He has published papers inInternational conferences and journals.Contact him at .

Sanjay Kumar Madria is anAssistant Professor, Department ofComputer Science, at the University ofMissouri-Rolla, USA. Earlier he wasVisiting Assistant Professor in theDepartment of Computer Science,Purdue University, West Lafayette, IN.He has widely published papers in jour-

nals and conferences in his areas ofresearch interest that include sensornetworking, mobile computing, webdata management and transaction pro-cessing. He is an IEEE Senior Member.Contact him at .

APRIL/MAY 2003 23

Graduating this year?

Depending on your grad-

uation date, you either have

received a mailing, we like

to call the B59, or you will

be getting one soonto

complete and return.

We use the B59 to

update your mailing address

and to verify that the edu-

cational information on our

records is correct. (You are

indeed graduating.)

Or take the easy way, the

online version called, IEEE

Graduating Student Data

Sheet is at:

www.ieee.org/graduate

Thanks!

Authorized licensed use limited to: LIVERPOOL JOHN MOORES UNIVERSITY Downloaded on February 20 2009 at 03:56 from IEEE Xplore Restrictions apply

Documents

WSN Overview