strategies and techniques for node placement in wireless

Available online at www.sciencedirect.com

Ad Hoc Networks 6 (2008) 621–655

www.elsevier.com/locate/adhoc

Strategies and techniques for node placement in wirelesssensor networks: A survey

Mohamed Younis a,*, Kemal Akkaya b

a Department of Computer Science and Electrical Engineering, University of Maryland, Baltimore County,

Baltimore, MD 21250, United Statesb Department of Computer Science, Southern Illinois University, Carbondale, IL 62901, United States

Received 13 December 2006; received in revised form 16 April 2007; accepted 11 May 2007Available online 24 May 2007

Abstract

The major challenge in designing wireless sensor networks (WSNs) is the support of the functional, such as data latency,and the non-functional, such as data integrity, requirements while coping with the computation, energy and communica-tion constraints. Careful node placement can be a very effective optimization means for achieving the desired design goals.In this paper, we report on the current state of the research on optimized node placement in WSNs. We highlight the issues,identify the various objectives and enumerate the different models and formulations. We categorize the placement strate-gies into static and dynamic depending on whether the optimization is performed at the time of deployment or while thenetwork is operational, respectively. We further classify the published techniques based on the role that the node plays inthe network and the primary performance objective considered. The paper also highlights open problems in this area ofresearch.� 2007 Elsevier B.V. All rights reserved.

Keywords: Node placement; Positioning; Wireless sensor networks; Node relocation

1. Introduction

Recent years have witnessed an increased interestin the use of wireless sensor networks (WSNs) innumerous applications such as forest monitoring,disaster management, space exploration, factoryautomation, secure installation, border protection,and battlefield surveillance [1,2]. In these applica-

1570-8705/$ - see front matter � 2007 Elsevier B.V. All rights reserved

doi:10.1016/j.adhoc.2007.05.003

* Corresponding author. Tel.: +1 410 455 3968; fax: +1 410 4553969.

E-mail addresses: [email protected] (M. Younis), [email protected] (K. Akkaya).

tions, miniaturized sensor nodes are deployed tooperate autonomously in unattended environments.In addition to the ability to probe its surroundings,each sensor has an onboard radio to be used forsending the collected data to a base-station eitherdirectly or over a multi-hop path. Fig. 1 depicts atypical sensor network architecture. For manysetups, it is envisioned that WSNs will consist ofhundreds of nodes that operate on small batteries.A sensor stops working when it runs out of energyand thus a WSN may be structurally damaged ifmany sensors exhaust their onboard energy supply.Therefore, WSNs should be carefully managed in

.

mailto:[email protected]



Command Node

Command Node

Command Node

Sensor nodes

Base-station Node

Command Node

Command Node

Command Node

Fig. 1. A sensor network for a combat filed surveillanceapplication. The base-station is deployed in the vicinity of thesensors to interface the network to remote command centers.

622 M. Younis, K. Akkaya / Ad Hoc Networks 6 (2008) 621–655

order to meet applications’ requirements while con-serving energy.

The bulk of the research on WSNs has focusedon the effective support of the functional, such asdata latency, and the non-functional, such as dataintegrity, requirements while coping with theresource constraints and on the conservation ofavailable energy in order to prolong the life of thenetwork. Contemporary design schemes for WSNspursue optimization at the various layers of thecommunication protocol stack. Popular optimiza-tion techniques at the network layer include multi-hop route setup, in network data aggregation andhierarchical network topology [3]. In the mediumaccess control layer, collision avoidance, outputpower control, and minimizing idle listening timeof radio receivers are a sample of the proposedschemes [1,4]. At the application layer, examplesinclude adaptive activation of nodes, lightweightdata authentication and encryption, load balancingand query optimization [5,6].

One of the design optimization strategies is todeterministically place the sensor nodes in order tomeet the desired performance goals. In such case,the coverage of the monitored region can be ensuredthrough careful planning of node densities and fieldsof view and thus the network topology can be estab-lished at setup time. However, in many WSNs appli-cations sensors deployment is random and littlecontrol can be exerted in order to ensure coverageand yield uniform node density while achievingstrongly connected network topology. Therefore,controlled placement is often pursued for only aselected subset of the employed nodes with the goal

of structuring the network topology in a way thatachieves the desired application requirements. Inaddition to coverage, the nodes’ positions affectnumerous network performance metrics such asenergy consumption, delay and throughput. Forexample, large distances between nodes weakenthe communication links, lower the throughputand increase energy consumption.

Optimal node placement is a very challengingproblem that has been proven to be NP-Hard formost of the formulations of sensor deployment[7–9]. To tackle such complexity, several heuristicshave been proposed to find sub-optimal solutions[7,10–12]. However, the context of these optimiza-tion strategies is mainly static in the sense thatassessing the quality of candidate positions is basedon a structural quality metric such as distance, net-work connectivity and/or basing the analysis on afixed topology. Therefore, we classify them as staticapproaches. On the other hand, some schemes haveadvocated dynamic adjustment of nodes’ locationsince the optimality of the initial positions maybecome void during the operation of the networkdepending on the network state and various externalfactors [13–15]. For example, traffic patterns canchange based on the monitored events, or the loadmay not be balanced among the nodes, causing bot-tlenecks. Also, application-level interest can varyover time and the available network resources maychange as new nodes join the network, or as existingnodes run out of energy.

In this paper we opt to categorize the variousstrategies for positioning nodes in WSNs. We con-trast a number of published approaches highlightingtheir strengths and limitations. We analyze theissues, identify the various objectives and enumeratethe different models and formulations. We catego-rize the placement strategies into static and dynamicdepending on whether the optimization is per-formed at the time of deployment or while the net-work is operational, respectively. We further classifythe published techniques based on the role that thenode plays in the network and the primary perfor-mance objective considered. Our aim is to helpapplication designers identify alternative solutionsand select appropriate strategies. The paper alsooutlines open research problems.

The paper is organized as follows. The nextsection is dedicated to static strategies for nodepositioning. The different techniques are classifiedaccording to the deployment scheme, the primaryoptimization metrics and the role that the nodes

M. Younis, K. Akkaya / Ad Hoc Networks 6 (2008) 621–655 623

play. In Section 3 we turn our attention to dynamicpositioning schemes. We highlight the technicalissues and describe published techniques whichexploit node repositioning to enhance network per-formance and operation. Section 4 discusses openresearch problems; highlighting the challenges ofcoordinated repositioning of multiple nodes andnode placement in three-dimensional applicationsetups and describes a few attempts to tackle thesechallenges. Finally, Section 5 concludes the paper.

2. Static positioning of nodes

As mentioned before, the position of nodes havea dramatic impact on the effectiveness of the WSNand the efficiency of its operation. Node placementschemes prior to network startup usually base theirchoice of the particular nodes’ positions on metricsthat are independent of the network state or assumea fixed network operation pattern that staysunchanged throughout the lifetime of the network.Examples of such static metrics are area coverageand inter-node distance, among others. Static net-work operation models often assume periodic datacollection over preset routes. In this section we dis-cuss contemporary node placement strategies andtechniques in the literature. We classify themaccording to the deployment methodology, the opti-mization objective of the placement and roles of thenodes. Fig. 2 summarizes the different categories ofnode placement strategies. By deployment method-ologies, as we elaborate in the next sub-section, we

Node's Rolein WSN

Sensor

Relay

Cluster-Head

Base-Station

OptimizationObjective

Area Coverage

Network Connectivity

Network Longevity

Data Fidelity

DeploymentMethodology

Controlled

Random

Sta

tic

No

de

Pla

cem

ent

Fig. 2. Different classifications of static strategies for nodeplacement in WSN.

mean how nodes are distributed and handled inthe placement process and the implication of thesensor field of view, i.e., two or three dimensions.Section 2.2 discusses the contemporary objectivesthat node placement is usually geared to optimizeand compares the different approaches in the litera-ture. Section 2.3 looks at how the placement strate-gies vary based on the role that a node plays in thenetwork. We note four types of nodes: sensors(which monitor their surroundings and are consid-ered data sources), relays, cluster-heads and base-stations. Since the placement of sensors is discussedin detail in Section 2.2, the focus of Section 2.3 is onthe other three types. Given the similarity betweenplacement techniques for cluster-heads and base-stations, we cover them under the category of whatwe call data collectors. A comparative summary ofall techniques discussed will be provided in a tableat the end of the section.

2.1. Deployment methodology

Sensors can generally be placed in an area ofinterest either deterministically or randomly. Thechoice of the deployment scheme depends highlyon the type of sensors, application and the environ-ment that the sensors will operate in. Controllednode deployment is viable and often necessary whensensors are expensive or when their operation is sig-nificantly affected by their position. Such scenariosinclude populating an area with highly precise seis-mic nodes, underwater WSN applications, and plac-ing imaging and video sensors. On the other hand,in some applications random distribution of nodesis the only feasible option. This is particularly truefor harsh environments such as a battle field or adisaster region. Depending on the node distributionand the level of redundancy, random node deploy-ment can achieve the required performance goals.

2.1.1. Controlled node deployment

Controlled deployment is usually pursued forindoor applications of WSNs. Examples of indoornetworks include the Active Sensor Network(ASN) project at the University of Sydney in Austra-lia [16], the Multiple Sensor Indoor Surveillance(MSIS) project at Accenture Technology Labs, inChicago [17] and the Sensor Network Technologyprojects at Intel [18]. The ASN and MSIS projectsare geared toward serving surveillance applicationssuch as secure installations and enterprise asset man-agement. At Intel, the main focus is on applications


in manufacturing plants and engineering facilities,e.g., preventative equipment maintenance (Fig. 3).Hand-placed sensors are also used to monitor thehealth of large buildings in order to detect corrosionsand overstressed beams that can endanger the struc-ture’s integrity [19,20]. Another notable effort is theSandia Water Initiative at Sandia National Labwhich addresses the problem of placing sensors inorder to detect and identify the source of contamina-tion in air or water supplies [21,22].

Deterministic placement is also very popular inapplications of range-finders, underwater acoustics,imaging and video sensors. In general, these sensorsare involved in three-dimensional (3-D) applicationscenarios, which is much more difficult to analyzecompared to two-dimensional deployment regions.Poduri et al. have investigated the applicability ofcontemporary coverage analysis and placement strat-egies pursued for 2-D space to 3-D setups [9]. Theyhave concluded that many of the popular formula-tions, such as art-gallery and sphere-packing prob-lems, which are optimally solvable in 2-D, becomeNP-Hard in 3-D. Most placement approaches forthese types of sensors strive to enhance the quality

Fig. 3. Sensors are mounted to analyze the vibration and assessthe health of equipment at a semiconductor fabrication plant(picture is from [18]).

of visual images and/or accuracy of the assessmentof the detected objects. For underwater applications,acoustic signals are often used as data carriers, andthe placement of nodes must also ensure that commu-nicating nodes are in each other’s the line-of-sight[23].

As we elaborate in Section 4, optimized deploy-ment of large-scale WSNs in 3-D applications isan emerging area of research that has just startedto receive attention. Most of the publications on3-D WSNs have considered very small networks[17], restricted the scope of the placement to a 2-Dplane [24], and/or pursued the fulfillment of simpledesign goals [25]. For example, Gonzalez-Banosand Latombe have studied the problem of findingthe minimum number of range-finders needed toestimate the proximity of a target, and their locationin order to cover an area [24]. Unlike other sensors,such as acoustic or temperature, etc., the authorshave to account for the restricted capabilities ofthe range-finders, which provide lower and upperbounds only, and for the difficulty of detectingobjects at grazing angles. The problem is formulatedas an art-gallery model for which the fewest guardsare to be placed to monitor a gallery. Similarly thework of Navarro et al. [25] addresses the orientationof video cameras for indoor surveillance so thathigh quality images of target objects are captured.

2.1.2. Random node distribution

Randomized sensor placement often becomes theonly option. For example, in applications of WSNsin reconnaissance missions during combat, disasterrecovery and forest fire detection, deterministicdeployment of sensors is very risky and/or infeasi-ble. It is widely expected that sensors will bedropped by helicopter, grenade launchers or clus-tered bombs. Such means of deployment lead torandom spreading of sensors; although the nodedensity can be controlled to some extent. Althoughit is somewhat unrealistic, many research projects,such as [26], have assumed uniform node distribu-tion when evaluating the network performance.The rationale is that with the continual decrease incost and size of micro-sensors, a large populationof nodes is expected and thus a uniform distributionbecomes a reasonable approximation.

Ishizuka and Aida [27] have investigated randomnode distribution functions, trying to capture thefault-tolerant properties of stochastic placement.They have compared three deployment patterns(Fig. 4, from [27]): simple diffusion (two-dimensional

Fig. 4. (a) Simple diffusion, (b) constant (uniform) placement and (c) R-random placement.

Fig. 5. An illustration of weighted random deployment (from[28]). The density of relay nodes inside the circle (close to thebase-station) is lower and the connectivity is weaker than outside.


normal distribution), uniform, and R-random,where the nodes are uniformly scattered with respectto the radial and angular directions from the base-station. The R-random node distribution patternresembles the effect of an exploded shell and followsthe following probability density function for sensorpositions in polar coordinates within a distance R

from the base-station:

f ðr; hÞ ¼ 1

2pR; 0 6 r 6 R; 0 6 h < 2p ð1Þ

The experiments have tracked coverage and nodereachability as well as data loss in a target trackingapplication. The simulation results indicate that theinitial placement of sensors has a significant effecton network dependability measured in terms oftolerance of a node failure that may be caused bydamage and battery exhaustion. The results alsoshow that the R-random deployment is a betterplacement strategy in terms of fault-tolerance. Thesuperiority of the R-random deployment is due tothe fact that it employs more nodes close the base-station. In a multi-hop network topology, sensorsnear the base-station tend to relay a lot of trafficand thus exhaust their battery rather quickly. There-fore, the increased sensor population close to thebase-station ensures the availability of spares forreplacing faulty relay nodes and thus sustains thenetwork connectivity.

While a flat architecture is assumed in [27], Xuet al. consider a two-tier network architecture inwhich sensors are grouped around relaying nodesthat directly communicate with the base-station[28]. The goal of the investigation is to identify themost appropriate node distribution in order to max-imize the network lifetime. They first show that uni-form node distribution often does not extend thenetwork lifetime since relay nodes will consumeenergy at different rates depending on their proxim-ity to the base-station. Basically the further away

the relays are from the base-station, the more theenergy they deplete in transmission. To counter thisshortcoming, a weighted random node distributionis then proposed to account for the variation inenergy consumption rate in the different regions.The weighted random distribution, as depicted inFig. 5, increases the density of relays away fromthe base-station to split the load among more relaysand thus extends their average lifetime. Although ithas a positive impact on the network lifetime, theweighted random distribution may leave some relaynodes disjoint from the base-station since somerelays may be placed so far that the base-stationbecomes out of their transmission range. Finally, ahybrid deployment strategy is introduced to balancethe network lifetime and connectivity goals. Theanalysis is further extended in [29] for the case whererelay nodes reach the base-station through a multi-hop communication path. The conclusion regardingthe three strategies was found to hold in the multi-hop case as well.

1

5

9

13

2

6

10

14

3

7

11

15

4

8

12

16

Fig. 6. Knowing the coordinate of obstacles, sensor’s field ofview is adjusted. In the shown example, redrawn from [11], thesensor on grid point 14 is not sufficient to cover the area indirection to point 11. The same applies for (2,7), (6,3) and (10,15).


2.2. Primary objectives for deployment

Application developers surely like the sensors tobe deployed in a way that aligns with the overalldesign goals. Therefore, most of the proposed nodeplacement schemes in the literature have focused onincreasing the coverage, achieving strong networkconnectivity, extending the network lifetime and/or boosting the data fidelity. A number of secondaryobjectives such as tolerance of node failure and loadbalancing have also been considered. Most of thework strives to maximize the design objectives usingthe least amount of resources, e.g., number ofnodes. Obviously, meeting the design objectivesthrough random node distribution is an utmostchallenge. Meanwhile, although intuitively deter-ministic placement can theoretically meet all pri-mary and secondary objectives, the quest forminimizing the required network resources keepsthe problem very hard. In this section, we categorizepublished work according to the optimizationobjective of the sensor placement. As we mentionedearlier the focus will be on sensor nodes that probethe environment and report their findings. Section2.3 will cover placement strategies for nodes thatplay other roles, i.e., relaying and data collection.

2.2.1. Area coverageMaximal coverage of the monitored area is the

objective that has received the most attention inthe literature. Assessing the coverage varies basedon the underlying model of each sensor’s field ofview and the metric used to measure the collectivecoverage of deployed sensors. The bulk of the pub-lished work, e.g., [30], assumes a disk coverage zonecentered at the sensor with a radius that equals itssensing range. However, some recent work hasstarted to employ more practical models of the sen-sor’s field of view in the form of irregular polygons[31]. Some of the published papers, especially earlyones, use the ratio of the covered area to the sizeof the overall deployment region as a metric forthe quality of coverage [30]. Since 2001, however,most work has focused on the worst case coverage,usually referred to as least exposure, measuring theprobability that a target would travel across an areaor an event would happen without being detected[32]. The advantage of exposure-based coverageassessment is the inclusion of a practical objectdetection probability that is based on signal process-ing formulations, e.g., signal distortion, as applica-ble to specific sensor types.

As mentioned earlier, optimized sensor place-ment is not an easy problem, even for deterministicdeployment scenarios. Complexity is often intro-duced by the quest to employ the least number ofsensors in order to meet the application require-ments and by the uncertainty in a sensor’s abilityto detect an object due to distortion that may becaused by terrain or the sensor’s presence in a harshenvironment. Dhillon and Chakrabarty haveconsidered the placement of sensors on a gridapproximation of the deployment region [11]. Theyformulate a sensing model that factors in the effectof terrain in the sensor’s surroundings and inaccu-racy in the sensed data (Fig. 6). Basically the prob-ability of detecting a target is assumed to diminishat an exponential rate with the increase in distancebetween a sensor and that target. A sensor candetect targets that lie in its line of sight. An obstaclemay thus make a target undetectable. The sensingmodel is then used to identify the grid points onwhich sensors are to be placed, so that an applica-tion-specific minimum confidence level on objectdetection is met. They propose a greedy heuristicthat strives to achieve the coverage goal throughthe least number of sensors. The algorithm is itera-tive. In each iteration, one sensor is placed at thegrid point with the least coverage. The algorithmterminates when the coverage goal is met or abound on the sensor count is reached.

Clouqueur et al. also studied the problem of pop-ulating an area of interest with the least number ofsensors so that targets can be detected with the high-est probability [12]. Unlike [11], random deployment


is assumed in this work. The authors propose a met-ric called path exposure to assess the quality of sen-sor coverage. The idea is to model the sensing rangeof deployed nodes and establish a collective coveragemap of all sensors based on a preset probability offalse-alarm (detection error). The map is thenchecked in order to identify the least exposure path,on which a target may slip by with the lowest prob-ability of detection. Fig. 7, which is redrawn from[12], illustrates the idea on a grid structure. Employ-ing such a metric, the authors further introduced aheuristic for incremental node deployment so thatevery target can be detected with a desired confi-dence level using the lowest sensor count. The ideais to randomly deploy a subset of the available sen-sors. Assuming that the sensors can determine andreport their positions, the least exposure path is iden-tified and the probability of detection is calculated. Ifthe probability is below a threshold, additionalnodes are deployed in order to fill holes in the cover-age along the least exposure path. This procedurewould be repeated until the required coverage isreached. The paper also tried to answer the questionof how many additional nodes are deployed per iter-ation. On the one hand, it is desirable to use the least

0.07

0.53

0.31.10

0.03

2.24

1.88

2.29

3.36

0.89

1.48

0.03

0.49

0.61

0.01.18

0.10

3.00

1.14

2.55

2.01

1.12

1.36

0.05

0.19

0.23

0.00.06

0.02

0.16

0.11

0.83

0.17

0.85

0.82

0.86

0.95

1.21

0.00.07

0.04

0.12

0.24

0.17

1.42

0.24

0.63

1.61

0.07

0.39

0.00.07

0.03

0.07

0.54

0.07

1.50

0.07

0.86

0.74

0.97

0.85

0.00.43

0.03

0.98

0.45

1.32

1.05

1.00

0.80

1.38

1.89

1.36

0.00.21

0.07

1.54

3.60

1.89

41.4

1.60

40.5

1.65

1.13

0.36

0.03.45

0.06

43.4

8.83

80.0

80.0

40.2

44.9

0.92

0.16

0.12

0.03.44

0.05

43.4

3.39

44.9

42.5

3.01

5.17

0.09

0.06

0.19

0.00.20

0.03

2.69

0.32

2.75

3.44

0.29

0.35

0.04

1.26 0.00.050.520.530.65

0.17 0.00.141.061.050.25

Fictitious

Point A

Fig. 7. The dark circle marks the location of each sensor. The weights otarget is detected by all sensors (combined exposure). The least exposualgorithm.

number of sensors. On the other hand, the means forsensor deployment may be expensive or risky, e.g.,sending a helicopter. The authors derive a formula-tion that account for the cost of deploying nodesand the expected coverage as a function of sensorcount. The formulation can be used to guide thedesigner for the most effective way to populate thearea.

Pompili et al. [33] have investigated the problemof achieving maximal coverage with the lowestsensor count in the context of underwater WSNs.Sensors are to be deployed on the bottom of theocean along with a few gateway nodes. The sensorssend their data to nearby gateways which forward itover vertical communication links to floating buoysat the surface. To achieve maximal coverage withthe least number of sensors, a triangular grid hasbeen proposed. The idea, as depicted in Fig. 8, isto pursue a circle packing such that any threeadjacent and non-collinear sensors form an equilat-eral triangle. In this way, one can control coverageof the targeted region by adjusting the distanced between two adjacent sensors. The authorshave proven that achieving 100% coverage is possi-ble if d ¼

ffiffiffi3p

r where r is the sensing range. The

0.01

0.01

0.01

0.00

0.01

0.01

0.01

0.02

3

0.02

0.00

0.01

0.00

0.00

0.01

0.00

0.02

0.01

9

0.04

0.01

0.01

0.00

0.02

0.00

0.01

0.04

0.02

5

0.05

0.02

0.06

0.02

0.24

0.02

0.01

0.03

0.04

2

0.03

0.07

0.01

0.07

0.01

0.07

0.01

0.07

0.04

7

0.01

0.01

0.01

0.01

0.00

0.01

0.01

0.01

0.01

2

0.01

0.28

0.49

0.25

0.44

0.06

0.04

0.04

0.07

5

0.05

0.65

0.48

0.49

0.43

0.32

0.33

0.07

0.05

6

0.05

0.25

0.05

0.24

0.22

0.04

0.04

0.03

0.02

6

0.05

0.02

0.01

0.02

0.01

0.02

0.01

0.05

0.04

6

0.05

0.000.000.010.011

0.010.010.030.042

FictitiousPoint B

Sensor

0.00 Edge Weight

n the individual line segments are based on the probability that are path, marked with a dark line, is found by applying Dijkstra’s

Fig. 8. Sensor placement based on a triangular grid. Coveragecan be controlled by adjusting the inter-node distance ‘‘d’’. Thefigure is redrawn from [33].

regionboundary

r-strip #0

a

b c


communication range of sensor nodes is assumed tobe much larger than r and thus connectivity is notan issue. The authors further study how nodes canpractically reach their assigned spots. Assumingnodes are to be dropped from the surface and sinkuntil reaching the bottom of the ocean, the effectof water current is modeled and a technique is pro-posed to predict the trajectory of the sensor nodesand to make the necessary adjustment for the droppoint at the surface.

The sensor placement problem considered byBiagioni and Sasaki [34] is more difficult. They optto find a placement of nodes that achieves thecoverage goals using the least number of sensorsand also maintain a strongly connected networktopology even if one node fails. The authors reviewa variety of regular deployment topologies, e.g.,hexagonal, ring, star, etc. and study their coverageand connectivity properties under normal andpartial failure conditions. They argue that regularnode placement simplifies the analysis due to theirsymmetry despite the fact that they often do notlead to optimal configurations. The provided ana-lytical formulation can be helpful in crafting aplacement as a mix of these regular topologies andin estimating aggregate coverage of nodes.

Even r-stripsare vertically

alligned

Odd r-stripsare shifted

r-strip #2

r-strip #4

r-strip #6

r-strip #1

r-strip #3

r-strip #5

Fig. 9. Illustration of the placement algorithm in a plane and afinite size region [35].

2.2.2. Network connectivity

Unlike coverage, which has constantly been anobjective or constraint for node placement, networkconnectivity has been deemed a non-issue in some ofthe early work based on the assumption that thetransmission range Tr of a node is much longer thanits sensing range Sr. The premise is that good cover-age will yield a connected network when Tr is a multi-

ple of Sr. However, if the communication range islimited, e.g., Tr = Sr, connectivity becomes an issueunless substantial redundancy in coverage is provi-sioned. It is worth noting that some work tackledthe connectivity concern through deploying relaynodes that have long haul communication capabili-ties. Such approaches will be discussed in Section 2.3.

Kar and Banerjee have considered sensor place-ment for complete coverage and connectivity [35].Assuming that the sensing and radio ranges areequal, the authors first define an r-strip as shownin Fig. 9(a). In an r-strip, nodes are placed so thatneighbors of a sensor along the x-axis are locatedon the circumstance of the circle that defines theboundary of its sensing and communication range.Obviously, nodes on an r-strip are connected. Theauthors then tile the entire plane with r-strips onlines y ¼ kð0:5

ffiffiffi3pþ 1Þr such that the r-strips are

aligned for even values of the integer k and shiftedhorizontally r/2 for odd values of k, as illustratedin Fig. 9(b). The goal is to fill gaps in coverage withthe least overlap among the r-disks that define theboundary of the sensing range. To establish connec-tivity among nodes in different r-strips, additionalsensors are placed along the y-axis (the shaded disksin Fig. 9(b)). For every odd value of the integer k,two sensors are placed at ½0; kð0:5

ffiffiffi3pþ 1Þr�

0:5ffiffiffi3p

r� to establish connectivity between every pairof r-strips. For a general convex-shaped finite-sizeregion, connectivity among nodes in horizontal r-

strips is established by another r-strip placed diago-nally within the boundary of the region (Fig. 9(c)).The authors generalize their scheme for the casewhere points of interest are to be covered ratherthan the whole area. However, unless the base-station is mobile and can interface with the WSNthrough any node, establishing a strongly connected


network is not essential in WSNs since data aregathered at the base-station. Therefore, ensuringthe presence of a data route from a node to thebase-station would be sufficient and thus fewernodes can be employed to achieve network connec-tivity than the presented approach would use. Inaddition, vertically placed nodes or diagonal r-strips

can become a communication bottleneck since theyact as gateways among horizontal r-strips, whichmay require the deployment of more sensors to splitthe traffic.

The focus of [36] is on forming K-connectedWSNs. K-connectivity implies that there are K inde-pendent paths among every pair of nodes. For K > 1,the network can tolerate some node and link failuresand guarantee certain communication capacityamong nodes. The authors study the problem ofplacing nodes to achieve K-connectivity at the net-work setup time or to repair a disconnected network.They formulate the problem as an optimizationmodel that strives to minimize the number of addi-tional nodes required to maintain K-connectivity.They show that the problem is NP-Hard and proposetwo approaches with varying degrees of complexityand closeness to optimality. Both approaches arebased on graph-theory. The idea is to model the net-work as a graph whose vertices are the current or ini-tial set of sensors and the edges are the existing linksamong these sensor nodes. A complete graph G forthe same set of vertices (nodes) is then formed andeach added edge is associated a weight. The weightof the edge between nodes u and v is set to uv

r � 1� �

,where uv is the Euclidean distance between u and v,and r is the radio range of a node. The weight basi-cally indicates the number of nodes to be placed toestablish connectivity between u and v. The problemis then mapped to finding a minimum-weight K-vertex-connected sub-graph ‘‘g’’. Finally, missinglinks (edges) in g are established by deploying theleast number of nodes. The authors proposedemploying one of the approximation algorithms inthe literature for finding the minimum-weight K-vertex-connected sub-graph, which often involvessignificant computation. An alternative greedyheuristic has been also proposed for resourceconstrained setups. The heuristic simply constructsg by including links from G in a greedy fashion andthen prunes g to remove the links that are unneces-sary for K-connectivity. Again in most WSNs, it isnot necessary to achieve K-connectivity amongsensors unless the base-station changes its locationfrequently.

On the other hand, the authors of [37–39]promote the view that in massively dense sensornetworks, it is unrealistic to model the network atthe node level; something they call the microscopiclevel. Instead, they promote studying the nodedeployment problem in terms of macroscopicparameters such as node density. For example, in[38,39] they consider a network of many sensorsreporting their data to a spatially dispersed set ofbase-stations and opt to find a distribution functionfor sensor nodes so that data flows to base-stationsover short routes and the traffic is spread. The goalis to minimize communication energy, limit interfer-ence among the nodes’ transmission and avoid traf-fic bottlenecks. In [38], they assume that relayingnodes do not generate data, and that the amountof traffic going from one part of the network toanother across a very small line segment is bounded.The latter assumption captures the effect of band-width limitation. They then formulate the nodeplacement problem as an optimization function tofind the best spatial density of nodes, i.e., probabil-ity density function. Analytical results indicate thatthe traffic flow between sensors and base-stationsresembles the electrostatic field induced betweenpositive and negative charges. The work is furtherextended in [39] by dropping the two assumptionsabout bandwidth limitation and the use of dedicatedrelay nodes and by employing a more elaboratephysical layer model for the radio transmissionand reception. The goal of the optimization in thenew formulation is to minimize the number of sen-sors and find their optimal spatial density for deliv-ering all data to the base-stations. Calculus ofvariations is pursued to derive an analytical solu-tion. It is worth noting that coverage goals are notconsidered in the formulation and the authors seemto implicitly assume that massively dense networksensure good area coverage.

2.2.3. Network longevity

Extending network lifetime has been the optimi-zation objective for most of the published communi-cation protocols for WSNs. The positions of nodessignificantly impact the network lifetime. For exam-ple, variations in node density throughout the areacan eventually lead to unbalanced traffic load andcause bottlenecks [28]. In addition, a uniform nodedistribution may lead to depleting the energy ofnodes that are close to the base-station at a higherrate than other nodes and thus shorten the networklifetime [40]. Some of the published work, such as


[29] which we discussed earlier, has focused on pro-longing the network lifetime rather than area cover-age. The implicit assumption is that there is asufficient number of nodes or the sensing range islarge enough such that no holes in coverage mayresult.

The maximum lifetime sensor deployment prob-lem with coverage constraints has been investigatedin [41]. The authors assume a network operationmodel in which every sensor periodically sends itsdata report to the base-station. The network isrequired to cover a number of points of interestfor the longest time. The average energy consump-tion per data collection round is used as a metricfor measuring the sensor’s lifetime. The problem isthen transformed to minimizing the average energyconsumption by a sensor per round by balancing theload among sensors. The idea is to spread theresponsibility of probing the points of interestamong the largest number of sensors and carefullyassign relays so that the data is disseminated usingthe least amount of energy. A heuristic is proposedthat tries to relocate sensors in order to form themost efficient topology. First, sensors are sorted indescending order according to their proximity tothe point of interest that they cover. Starting fromthe top of the sorted list, the algorithm iterates onall sensors. In each iteration, the sensor is checkedfor whether it can move to another location to serveas a relay. The new location is picked based on thetraffic flow and the data path that this node is partof or will be joining. Basically, the relocating nodeshould reduce its energy consumption by gettingclose to its downstream neighbor. A sensor reposi-tioning is allowed only if it does not risk a loss incoverage.

Chen et al. have studied the effect of node densityon network lifetime [42]. Considering the one-dimensional placement scenario, the authors derivean analytical formulation for the network lifetimeper unit cost (deployed sensor). They also argue thatnetwork lifetime does not grow proportionally tothe increased node population and thus a carefulselection of the number of sensors is necessary tobalance the cost and lifetime goals. Consideringthe network to be functional until the first node dies,an optimization problem is defined with the objec-tive of identifying the least number of sensors andtheir positions so that the network stays operationalfor the longest time. An approximate two-step solu-tion is proposed. In the first step, the number of sen-sors is fixed and their placement is optimized for

maximum network lifetime. They formulate thisoptimization as a multi-variant non-linear problemand solve it numerically. In the second step, thenumber of sensors is minimized in order to achievethe highest network lifetime per unit cost. A closedform solution is analytically derived for the secondstep. A similar problem is also studied by Chenget al. [43]. However, the number of sensors is fixedand the sensor positions are determined in orderto form a linear network topology with maximallifetime.

2.2.4. Data fidelity

Ensuring the credibility of the gathered data isobviously an important design goal of WSNs. Asensor network basically provides a collectiveassessment of the detected phenomena by fusingthe readings of multiple independent (and some-times heterogeneous) sensors. Data fusion booststhe fidelity of the reported incidents by loweringthe probability of false alarms and of missing adetectable object. From a signal processing pointof view, data fusion tries to minimize the effect ofthe distortion by considering reports from multiplesensors so that an accurate assessment can be maderegarding the detected phenomena. Increasing thenumber of sensors reporting in a particular regionwill surely boost the accuracy of the fused data.However, redundancy in coverage would requirean increased node density, which can be undesirabledue to cost and other constraints such as the poten-tial of detecting the sensors in a combat field.

Zhang and Wicker have looked at the sensorplacement problem from a data fusion point of view[44]. They note that there is always an estimationdistortion associated with a sensor reading whichis usually countered by getting many samples. Thus,they map the problem of finding the appropriatesampling points in an area to that of determiningthe optimal sampling rate for achieving a minimaldistortion, which is extensively studied in the signalprocessing literature. In other words, the problem istransformed from the space to the time domain. Theset of optimal sensor locations corresponds directlyto the optimal signal sampling rate. The approach isto partition the deployment area into small cells,then determine the optimal sampling rate per cellfor minimal distortion. Assuming that all sensorshave the same sampling rate, the number of sensorsper cell is determined.

Similar to [44], Ganesan et al. have studiedsensor placement in order to meet some application


quality goals [45]. The problem considered is to findnode positions so that the fused data at the base-sta-tion meets some desired level of fidelity. Unlike [44],the objective of the optimization formulation is tominimize the energy consumption during communi-cation; a tolerable distortion rate is imposed as aconstraint to this optimization Also, in this workthe number of sensors is fixed and their position isto be determined. Given the consideration of energyconsumption, data paths are modeled in the formu-lation, making the problem significantly harder. Theauthors first provide a closed form solution for theone-dimensional node placement case and then useit to propose an approximation algorithm for nodeplacement in a circular region. Extending theapproach to handle other regular and irregularstructures is noted as future work.

Wang et al. have also exploited similar ideas for aWSN that monitors a number of points of interest[46]. Practical sensing models indicate that theability to detect targets or events diminishes withincreased distance. One way to increase the credibil-ity of the fused data is to place sensors so that apoint of interest would be in the high-fidelity sensingrange of multiple nodes. Given a fixed number ofsensors, there is a trade-off between deploying a sen-sor in the vicinity of one point of interest to enhancethe probability of event detection and the need tocover other points of interest. The probability ofevent detection by a sensor is called the utility.The utility per point of interest is thus the collectiveutility of all sensors that cover that point. Theauthors formulate a non-linear optimization modelto identify the sensor locations that maximize theaverage utility per point of interest. To limit thesearch space, the area is represented as a grid withonly intersection points considered as candidatepositions.

Finally we would like to note that the work ofClouqueur et al. [12], which we discussed earlier,can also be classified under data fidelity based sen-sor placement. They estimate the credibility of fuseddata from multiple sensors and use it to identify theposition of sensors for maximizing the probabilityof target detection.

2.3. Role-based placement strategies

The positions of nodes not only affect coveragebut also significantly impact the properties of thenetwork topology. Some of the published workhas focused on architecting the network in order

to optimize some performance metrics, for example,to prolong the network lifetime or minimize packetdelay. These architectures often define roles for theemployed nodes and pursue a node-specific posi-tioning strategy that is dependent on the role thatthe node plays. In this section, we opt to categorizerole-based node placement strategies. Generally, anode can be a regular sensor, relay, cluster-heador base-station. Since the previous section coveredthe published work on sensor placement, we limitthe scope in this section to surveying strategies forrelay, cluster-head and base-station positioning.Since cluster-heads and base-stations often act asdata collection agents for sensors within their reach,we collectively refer to them as data collectors.

2.3.1. Relay node placement

Positioning of relay nodes has also been consid-ered as a means for establishing an efficient networktopology. Contemporary topology managementschemes, such as [47–49], assume redundant deploy-ment of nodes and selectively engage sensors inorder to prolong the network lifetime. Unlike theseschemes, on-demand and careful node placement isexploited in order to shape the network topology tomeet the desired performance goals. Many variantsof the relay node placement problem have been pur-sued; each takes a different view depending on therelationship between the communication ranges ofsensor and relay nodes, allowing a sensor to act asa hop on a data path, considering a flat or tierednetwork architecture and the objective of the opti-mization formulation. The assumed capabilities ofthe relay nodes vary widely. Some work considersthe relay node (RN) to be just a sensor; especiallyfor flat network architectures. In two-tier networks,RNs usually play the role of a gateway for one ormultiple sensors to the other nodes in the network.The transmission range of RNs is often assumed lar-ger than sensors. When RNs do not directly trans-mit to the base-station, the placement problembecomes harder since it involves inter-RN network-ing issues.

Hou et al. [50] have considered a two-tier sensornetwork architecture where sensors are split intogroups; each is led by an aggregation-and-forward-ing (AFN) node (Fig. 10). A sensor sends its reportdirectly to the assigned AFN, which aggregates thedata from all sensors in its group. The AFNs andthe base-station form a second-tier network inwhich an AFN sends the aggregated data reportto the base-station over a multi-hop path. The

BaseStation

Upper tier

Lower tier

MSN

AFN

Fig. 10. Logical view of the assumed two-tier network architec-ture (redrawn from [50]). MSN stands for micro-sensor nodes.


authors argue that AFNs can be very critical to thenetwork operation and their lifetime should bemaximized. Two approaches have been suggestedto prolong the AFNs’ lifetime. The first is to provi-sion more energy to AFNs. The second is to deployrelay nodes (RNs) in order to reduce the communi-cation energy consumed by an AFN in sending thedata to the base-station. The RN placement andenergy provisioning problem is formulated as amixed-integer non-linear programming optimiza-tion. For a pool with an E energy budget and Mrelay nodes, the objective of the optimization is tofind the best allocation of the additional energy toexisting AFNs and the best positions for placingthe M relays. To efficiently solve the optimizationproblem, the formulation is further simplifiedthrough a two-phase procedure. In the first phasea heuristic is proposed for optimized placement ofthe M relay nodes. Given the known positions ofthe RNs, in the second phase the energy budget isallocated to the combined AFN and RN popula-tion, which is a linear programming optimization.

Unlike the work described earlier on random-ized deployment of relay nodes (RNs) [28,29], thefocus in [51–54] is on deterministic placement. Theconsidered system model fits indoor applicationsor friendly outdoor setups. In [51], the authors con-sider the placement of relay nodes that can directlyreach the base-station. Given a deployment of sen-sor nodes (SNs), the problem is to find the minimumnumber of RNs and where they can be placed inorder to meet the constraints of network lifetimeand connectivity. The network’s lifetime is mea-sured in terms of the time for the first node to die.Connectivity is defined as the ability of every SN

to reach the base-station, implying that every SNhas to have a designated RN. The problem is shownto be equivalent to finding the minimum set cover-ing, which is an NP-Hard problem. Therefore, arecursive algorithm is proposed to provide a sub-optimal solution. The algorithm pursues a divide-and-conquer strategy and is applied locally. Sensorscollaboratively find the intersections of their trans-mission ranges. Relay nodes are placed in the inter-sections of the largest number of sensors so that allsensors are served by a relay node.

This work has been further extended in [52–54] toaddress the problem of deploying a second-tier ofrelay nodes so that the traffic is balanced and theleast number of additional relay nodes are used.In [52], a lower bound on the number of requirednodes is derived. The authors have also proposedtwo heuristics. The first is very simple and places asecond level relay (SLR) at a distance from the firstlevel relay (FLR) so that FLRs stay operational forthe longest time. In the second heuristic, SLRs areallocated only to those FLRs that cannot reachthe base-station. The number of SLRs is thenreduced by removing redundancy. Basically, if SLRi

can serve both FLRi and FLRj, SLRj is eliminatedsince FLRj is already covered. The validation resultsshow that the second heuristic provides near opti-mal solutions in a number of scenarios. Moresophisticated approaches have been proposed in[53,54]. The main idea is that an existing FLR thatis close to the base-station is considered as a candi-date SLR before deploying new RNs to serve asSLRs.

The objective of the relay placement in [55] is toform a fault-tolerant network topology in whichthere exist at least two distinct paths between everypair of sensor nodes. All relay nodes are assumed tohave the same communication range R that is atleast twice the range of a sensor node. A sensor issaid to be covered by a relay if it can reach thatrelay. The authors formulate the placement problemas an optimization model called ‘‘2-ConnectedRelay Node Double Cover (2CRNDC)’’. The prob-lem is shown to be NP-Hard and a polynomial timeapproximation algorithm is proposed. The algo-rithm simply identifies positions that cover themaximum number of sensors. Such positions canbe found at the intersections of the communicationranges of neighboring sensors. Relay nodes arevirtually placed at these positions. The analysis thenshifts to the inter-relay network. Relays with themost coverage are picked and the algorithm checks


whether the relays form a 2-connected graph andevery sensor can reach at least two relays. If not,more relays are switched from virtual to real andthe connectivity and coverage are rechecked. Thelatter step is repeated till the objective is achieved.

Tang et al. have also studied the RN placementproblem in a two-tiered network model [26]. Theobjective is again to deploy the fewest RNs suchthat every SN can reach at least one RN node andthe RNs form a connected network. In addition,the authors consider a variant of the problem inwhich each SN should reach at least 2 RNs andthe inter-RN network is 2-connected in order forthe network to be resilient to RN failures. Twopolynomial time approximation algorithms are pro-posed for each problem and their complexity is ana-lyzed based on a uniform SN deployment and onthe assumption that the communication range R

of a relay node is at least 4 times the communicationrange r of a SN. The idea is to divide the area intocells. For each cell an exhaustive search is per-formed to find positions that are reachable to allSNs in the cell. These positions can be determinedusing analytical geometry on the circumference ofthe circles that define the communication range ofSNs. The authors argue that for small cells this isfeasible since few sensors are expected to be in a cell.These positions are the candidates for placing RNs.The final RN positions are selected so that the RNof a cell is reachable to those RNs in neighboringcells. If this is not possible, additional RNs areplaced at grid points to form a path between disjointneighboring RNs. To support tolerance to RN fail-ures, two RNs positions are picked per cell such thatthere are at least two non-overlapping pathsbetween every pair of relay nodes. More RNs maybe required to enable forming such independentroutes.

On the other hand, Cheng et al. have considereda class of WSNs, such as biomedical sensor net-works, in which the sensors positions are deter-mined prior to deployment [8]. To boost thenetwork lifetime and limit interference, it is desiredto maintain network connectivity with minimumtransmission power per node. The authors havetried to achieve this design goal by construction.They consider a homogeneous network where sen-sors can act as a data source and a relay. Given aset of deployed sensors in an area, they opt to placethe minimum number of relay nodes for maintain-ing network connectivity such that the transmissionrange of each sensor does not exceed a constant r.

They formulate the optimization problem as a Stei-ner Minimum Tree with Minimum number of Stei-ner Points (SMT-MSP), which is NP-Hard. Totackle the high complexity, they propose a polyno-mial time approximation algorithm. This algorithmruns in two phases. In the first, a minimum-costspanning tree T is formed using the edge length asthe cost. If an edge length exceeds the transmissionrange, RNs are placed on that edge to maintain con-nectivity. In the second phase, the transmissionpower of each node is reduced to the minimum levelneeded to maintain the link to the next node on thetree T.

Lloyd and Xue have considered two RN place-ment problems [56]. The first is a generalization ofthe model of [8], discussed above, allowing RNs tohave longer transmission range than SNs and allow-ing both SNs and RNs to act as data forwarders ona particular data path. This network architecture isstill flat. In the second placement problem, a two-tier architecture is considered. The placement objec-tive in that case is to deploy the least number ofRNs in order to ensure inter-sensor connectivity.In other words, a sensor must be able to reach atleast one RN, and the inter-RN network must bestrongly connected. The second problem can beviewed as a generalization of [26] with R > r. Theauthors have crafted approximation algorithms forboth problems based on finding the minimum span-ning tree for the first problem and a combinedSMT-MSP and Geometric Disk Cover algorithmfor the second problem. The approach is furtherextended in [57] to form a k-connected networktopology in order to tolerate occasional failure ofrelay nodes.

Table 1 categorizes the relay node placementproblems and mechanisms discussed above.

2.3.2. Placement of data collectors

Clustering is a popular methodology for enablinga scalable WSN design [58]. Every cluster usuallyhas a designated cluster-head (CH). Unlike RNs,which forward data from some sensors as we dis-cussed earlier, CHs usually collect and aggregatethe data in their individual clusters and often con-duct cluster management duties consistent withtheir capabilities. When empowered with sufficientcomputational resources, CHs can create and main-tain a multi-hop intra-cluster routing tree, arbitratemedium access among the sensors in the individualclusters, assign sensors to tasks, etc. When inter-CHs coordination is necessary, or CHs have too

Table 1A comparison between the various approaches for relay node placement

Paper Tiers Objective Radioranges

Nodes onpath

Connectivity goals/constraints

[8] Single Minimal # of relays R = r Sensors Path between every pair of sensorsMinimal per nodetransmission power

Relays

[26] Two Minimal # of relays R P 4 r Relays Every sensor can reach at least two relaysTwo vertex-disjoint paths between every pair ofrelays

[50] Two Max. AFN lifetime N/A Relays Path from each AFN to base-station[51] Two Minimal # of relays R > r Relays Relays directly reach base-station

Path from each sensor to base-stationTime to first node to die must exceed a thresholdvalue

[52] Two Minimal # of relays R > r Relays Balance traffic loads among relays[53] Relay to base-station connectivity[54] Time to first relay node to die must exceed a

threshold value[55] Two Minimal # of relays R P 2r Relays Two vertex-disjoint paths between every pair of

sensors[56]-1 Single Minimal # of relays R P r Sensors Path between every pair of sensors

Relays[56]-2 Two Minimal # of relays R P r Relays Path between every pair of sensors[57] Single Minimal # of relays R 5 r Relays K vertex-disjoint paths between every pair of nodes

(sensors and relays)


limited communication range to directly reach thebase-station, CHs may need to form a multi-hopinter-CH network, i.e., a second-tier network. CHscan be also picked among the deployed sensor pop-ulation. In that case, little control can be exertedunless the nodes are moveable.

Careful positioning of cluster-heads in a hierar-chical network has been deemed an effective strategyfor establishing an efficient network topology [59–66]. The same applies to base-stations. In fact, mostpublished approaches for CH placement apply tobase-stations for the same network model, and viceversa. The similarity is mostly due to the fact thatboth CHs and base-stations collect the data fromsensor nodes and take some leadership role in set-ting up and managing the network. In order to sim-plify the discussion in this section and better definethe scope and limitations of the presented placementapproaches, we will refer to both cluster-heads andbase-stations as data collectors (DCs).

In general, the complexity of the DC placementproblem varies based on the planned network archi-tecture. If the sensors are assigned to distinct clus-ters prior to placing or finalizing the positions ofDC nodes, the scope of the problem becomes localto the individual cluster and only concerns eachDC independently from the others [60]. However,

if the DC placement precedes the clustering process,the complexity is NP-Hard, proven in [59] through areduction to the dominating set problem on unitdisk graphs. We first discuss those approacheswhere clusters are formed before the DCs are posi-tioned. In this case, the theme is typically to groupsensors based on a static metric like physical prox-imity, then place a head node for every group ofsensors in order to optimize some design objective.There is always an implicit assumption on how thenetwork will operate. Usually sensor nodes areexpected to transmit data continuously at a con-stant rate.

Oyman and Ersoy [60] employ the popular k-

means clustering algorithm to form disjoint groupsof sensors so that the average Euclidian distancebetween sensors in a cluster and its assigned DC isminimized. The algorithm is repeated for differentnumbers of clusters until a lower bound on networklifetime is reached with the least DC count. Thenetwork lifetime is estimated based on the energyconsumed by sensors to reach their respective DC.A DC is placed at the centroid of the sensors ofthe individual clusters. The Genetic Algorithm forHop count Optimization (GAHO) approachproposed in [61] follows a similar process. However,GAHO employs artificial neural networks for


optimal sensor grouping and strives to minimize thedata collection latency by reducing the number ofhops that sensor readings have to pass throughbefore reaching a DC. Instead of placing a DC atthe centroid of its cluster, a search is conductedaround the centroid to find a nearby position suchthat the DC is in the communication range of thelargest number of sensors in its cluster. A light-weight version of GAHO, called Genetic Algorithmfor Distance Optimization (GADO), is also pro-posed to achieve the same objective using theEuclidian distance between sensors and theirassigned DC. GADO essentially trades the optimal-ity of the DC positions for the complexity of thealgorithm.

Data latency is also the optimization objective ofCOLA, an actor (DC) placement algorithm pre-sented in [62]. In some application scenarios likedisaster management and combat field reconnais-sance, DCs not only collect and process the databut also can do some reactive actions, such as extin-guishing a fire or de-mining a travel path. That iswhy DCs are referred to as actors. In this case,the design objective is to minimize the deliverylatency of sensor data and the time for an actor toreach the spot that needs attention. Initially theactors are positioned uniformly in order to maxi-mize the coverage (i.e., minimize the overlap amongthe action ranges) of the area as shown in Fig. 11.Sensors are then grouped into clusters; each is ledby an actor. After clustering, each actor considersthe positions of its assigned sensors as vertices andcomputes the vertex 1-center [7]. Relocating theactor at the vertex 1-center location ensures mini-mum delay from the farthest sensor node, i.e., min-imize the maximum latency for data delivery.

However, when relocating an actor to its 1-centerlocation, it may lose its connection with the otheractors in the network. In order to also ensureinter-actor connectivity, the approach in [62] is fur-

3 actors 4 actors

Fig. 11. Initially, data-collectors (actors) are uniformly placed inthe area of interest. Circles define the acting range of an actor.

ther extended in [63]. Connectivity is maintained bymoving an actor close to the vertex 1-center of itscluster as much as possible without breaking thelinks with its neighbors. The relocation of the indi-vidual actors is organized following a global orderbased on the IDs of actors so as not to disconnectthe network with simultaneous relocations of neigh-boring actors.

There are also some approaches which deploy theDCs before grouping the sensor nodes. The cluster-ing process in this case strives to set up an optimalnetwork topology; picking the right DC for sensorsto send their data to. Given the scope of the paper,we focus on those approaches that pre-select thepositions of DC nodes; that is, they pursue a con-trolled placement of DCs. The placement problemin this case is more challenging and is shown to beNP-Hard [59]. Published solutions usually tacklethe complexity of the optimization by restrictingthe search space. For instance, in [7] the searchspace is restricted to the sensor locations and thebest position ‘‘s’’ among them is picked in termsof network lifetime. This solution is shown to be aconstant approximation of the optimal solution,e.g., achieves a fraction of the optimal network life-span. The approximation ratio is further improvedto (1-e), where e > 0 is any desired error bound, byfactoring in the routes and transmission schedule.However, the improvement comes at the cost ofincreased computation for solving multiple linearprograms. To limit such high computation, a tech-nique is proposed which explores the potential over-lap among the elements of the search space [64]. Theidea is to replace an infinite search space for eachvariable by a finite-element search space with aguaranteed bound on the possible loss in perfor-mance. Specifically, the search space grows expo-nentially with the increase in the number ofvariables and such growth can be reduced by explor-ing the potential overlap among the elements of thesearch space. In order to determine the potentialoverlap, the variables are expressed in the form ofa geometric progression and a common factoramong these geometric progressions is identified.

Bogdanov et al. have studied the problem ofdetermining the optimal network topology for mul-tiple DCs with the aim of maximizing the data ratewhile minimizing the energy consumption at theindividual sensors [59]. Such optimization arises insetups where the sensors’ batteries can be rechargedat a fixed rate and most data is to be collectedbetween successive energy scavenging cycles. For


networks with a fixed data rate, the problembecomes equivalent to minimizing the power usedto collect the data. The data rate mainly varies atrelaying nodes, which forward other sensor’s datain addition to sending their own. The authors arguethat appropriate placement of data collectors canshorten the data routes and prevent the overloadingof relaying nodes beyond their maximum achievabledata rate, which is determined based on the capacityof their onboard battery. Given the complexity ofthe placement problem, the solution space is alsolimited by allowing DCs to be located only at sensorpositions. Once the DCs are placed, each sensor des-ignates the DC that can reach it over the shortestpath. There is no explicit clustering performed.Two different heuristics, namely greedy and localsearch, were presented. The greedy scheme deploysDCs incrementally. Basically, the first DC is placedso that the data rate is maximized and then the sec-ond is placed assuming the position of the first DCis fixed and so on. On the other hand, the localsearch starts with a random placement of DCs.Each DC then tries to relocate to the position of aneighboring sensor in order to maximize the datarate. If no improvement is possible, the algorithmrecords the best achievable data rate and stops. Thisprocess is repeated several times for different ran-dom configurations and the one with the highestdata rate is finally picked. Fig. 12 shows the optimallayout for 4 DCs for a 10 · 10 grid of sensors whosetransmission ranges are 2.2 units. In the figure, thedark circles designate the DCs white the smaller cir-cles designate the sensor nodes.

Fig. 12. Optimal DC locations (black circles) achieved by thelocal search.

In [65], optimizing the placement of the DCnodes is formulated as an integer linear program-ming (ILP) model. The objective function of theILP formulation is to minimize the maximumenergy consumption at the individual sensors whileminimizing the total communication energy. Theconstraints of the ILP formulation include a boundon the total energy consumed by a node in a datacollection round and a restriction on the candidateDC location to be picked from a set of pre-deter-mined positions. Other constraints are also specifiedto ensure a balanced flow through the individualnodes and to allow transmission of messages to afeasible site only if a DC is to be placed at that site.The optimization is conducted by one of the DCs,i.e., the approach is centralized. After the DC loca-tions are determined, a flow-based routing algo-rithm is used in order to pick the right DC foreach sensor and determine the data paths. Unlike[59], DC positions are re-computed periodically atthe beginning of each data collection round in orderto cope with changes in the network state and dis-tribute the routing load among the sensors evenly.

The approach of [66] does not consider datarelaying and thus the problem becomes solvable inpolynomial time. In order to minimize the totalcommunication power, the DC node is to be locatedsuch that the maximum distance to a sensor node isminimized [66]. A computational geometry basedalgorithm, whose complexity is linear in the numberof nodes n, is proposed. This algorithm tries todetermine the circle with the smallest diameter thatencloses the nodes. Such a circle can be formed withat most three points picked among the locations ofthe sensors. The DC will then be positioned at thecenter of the circle [66]. The same algorithm is fur-ther extended for a two-tiered WSN where specialapplication nodes are designated as first-tier DCs[10]. An application node interfaces its cluster withthe base-station. The minimum enclosing circle isthus found for the application nodes as shown inFig. 13, redrawn from [10].

Table 2 provides a comparative summary of thecharacteristics of the static node placement mecha-nisms discussed in this section.

3. Dynamic repositioning of nodes

Most of the protocols described above initiallycompute the optimal location for the nodes anddo not consider moving them once they have beenpositioned. Moreover, the context of the pursued

01

2 3

01

2

01

Fig. 13. Finding the minimum enclosing circle for six application nodes [10]. The data-collector (black triangle) is placed at the center ofthe smallest disk that contains all application nodes (small circles).


optimization strategies is mainly static in the sensethat assessing the quality of candidate positionsare based on performance metrics like the data rate,sensing range, path length in terms of the number ofhops from a sensor node to the base-station, etc. Inaddition, the placement decision is made at the timeof network setup and does not consider dynamicchanges during the network operation. For exam-ple, traffic patterns can change based on the moni-tored events, or the load may not be balancedamong the nodes, causing bottlenecks. Also, appli-cation-level interest can vary over time, and theavailable network resources may change as newnodes join the network, or as older nodes run outof energy.

Therefore, dynamically repositioning the nodeswhile the network is operational is necessary to fur-ther improve the performance of the network. Forinstance, when many of the sensors in the vicinityof the base-station stop functional due to theexhaustion of their batteries, some redundant sen-sors from other parts of the monitored region canbe identified and relocated to replace the dead sen-sors in order to improve the network lifetime. Suchdynamic relocation can also be very beneficial in atarget tracking application where the target ismobile. For instance, some of the sensors can berelocated close to the target to increase the fidelityof the sensor’s data. Moreover, in some applicationsit may be wise to keep the base-station a safe dis-tance from harmful targets, e.g., an enemy tank,by relocating it to safer areas in order to ensure itsavailability.

Relocating the nodes during regular networkoperation is very challenging. Unlike initial place-ment, such relocation is pursued in response to anetwork- or environment-based stimulus. It thusrequires continual monitoring of the network stateand performance as well as analysis of events hap-

pening in the vicinity of the node. In addition, therelocation process needs careful handling since itcan potentially cause disruption in data delivery.The basic issues can be enumerated as follows: whendoes it make sense for a node to relocate, whereshould it go and how will the data be routed whilethe node is moving? In this section we discuss theseissues in detail and survey published approaches ondynamic node repositioning. We group publishedwork according to whether the node being reposi-tioned is a sensor or a data collector. In all the tech-niques covered in this section, no coordinationamong relocated nodes is provisioned. Collabora-tive multi-node relocation is an emerging area ofresearch and is exclusively covered in Section 4.

3.1. Relocation issues

When to consider relocation: The decision for anode movement has to be motivated by either anunacceptable performance measure (despite settingup the most efficient network topology) or a desireto boost such measures beyond what is achievableat the present node position. Motives vary basedon the targeted design attributes. Examples includethe observation of bottlenecks in data relaying,decreases in node coverage in an area, increases inpacket latency or excessive energy consumptionper delivered packet. A weighted average may alsobe pursued to combine multiple metrics based onthe application at hand.

Once a node has its motive, it will consider mov-ing to a new position. Such consideration does notnecessarily lead to an actual relocation. The nodefirst needs to qualify the impact of repositioning atthe new location on the network performance andoperation. Therefore the ‘‘when’’ and ‘‘where’’issues of node movement are very closely inter-related. In addition, the node must assess the

Table 2A comparison between the various approaches for nodes placement

Paper Application Space Deployment Nodetype

Primaryobjective

Secondary objective Constraint

[7] Generic 2-D Deterministic DataCollector

Network lifetime – –

[8] Biomedicalsensor networks

2-D Deterministic Relay Network lifetime Min. relay count Connectivity

[10] Generic 2-D Deterministic Datacollector


[11] Surveillance 2-D Deterministic Sensor Coverage Min. sensor count –[12] Outdoor 2-D Random Sensor Data fidelity &

coverageMin. sensor count –

[16] Surveillance 3-D Deterministic Sensor Coverage Connectivity –[17] Surveillance 3-D Deterministic Sensor Coverage Data fidelity –[18] Manufacturing 3-D Deterministic Sensor Data fidelity Connectivity[19] Structural health

monitoring3-D Deterministic Sensor Data fidelity Connectivity –

[20] Structural healthmonitoring

3-D Deterministic Sensor Data fidelity Connectivity & Fault-tolerance

–

[21] Contaminationdetection

2-D Deterministic Sensor Coverage – Fixed sensors count

[22] Contaminationdetection

2-D Deterministic Sensor Coverage Delay Fixed sensors count

[24] Generic 3-D Deterministic Sensor Data fidelity Min. sensor count –[25] Generic 3-D Deterministic Sensor Data fidelity – –[26] Outdoor 2-D Deterministic Sensor Min. relay count Fault-tolerance Connectivity[27] Outdoor 2-D Random Sensor Coverage &

connectivityFault-tolerance –

[28] Outdoor 2-D Random Relay Network lifetime – –[29] Outdoor 2-D Random Relay Network lifetime – –[33] Underwater 2-D Deterministic Sensor Coverage Min. sensor count –[34] Outdoor 2-D Deterministic Sensor Coverage &

connectivity& Fault-tolerance –

[35] Outdoor 2-D Deterministic Sensor Coverage &connectivity

Min. Sensor count –

[36] Outdoor 2-D Deterministic Sensor Connectivity Fault-tolerance –[38] Massively dense

networks2-D Random Relay Min. sensor

countDelay and energy Bandwidth

[39] Massively densenetworks

2-D Random Sensor Min. sensorcount

Delay and energy –

[41] Surveillance 2-D Controlled(nodes move)

Sensor Network lifetime – Coverage

[42] Generic 1-D Deterministic Sensor Min. relay count – CoverageNetwork lifetime

[43] Generic 1-D Deterministic Sensor Network lifetime – Connectivity[44] Outdoor 2-D Random Sensor Data fidelity – –[45] Generic 1-D Deterministic Sensor Data fidelity Minimal energy

consumption incommunication

– Lower bound ontolerable distortion

2-D – Fixed sensors count[46] Surveillance 2-D Deterministic Sensor Data fidelity – –[50] Generic 2-D Controlled

(nodes move)Relay Network lifetime – Fixed relays count

[51] Indoor or non-harsh outdoor

2-D Deterministic Relay Min. relay count – – Network lifetime– Connectivity


2-D Deterministic Relay Network lifetime Min. relay count –






Table 2 (continued)

Paper Application Space Deployment Node type Primary objective Secondary objective Constraint

[55] Generic 2-D Deterministic Relay Connectivity & fault-tolerance

Min. relay count –

[56] Generic 2-D Deterministic Relay Min. relay count – Connectivity[57] Generic 2-D Deterministic Relay Fault-tolerance Min. relay count Connectivity[59] Generic 2-D Deterministic Data

collectorMax. data flow Min. energy –


Network lifetime Min. CH count –


Delay Energy –

[62] Generic 2-D Random Datacollector

Coverage Delay –

[63] Generic 2-D Random Datacollector

Coverage Delay Connectivity




Network lifetime Load balancing –

[66] Surveillance 2-D Deterministic Datacollector



relocation overhead. Such overhead can be incurredby the node and the network. For example, if thenode is a robot, the energy consumed by the mechani-cal parts during the movement is a significant over-head to the lifetime of the robot’s battery and thusshould be minimized. Moreover, when energy andtimeliness metrics are of utmost concern, the impacton the lifetime of individual sensors and on routemaintenance has to be considered respectively.

Where to relocate: When having a motive torelocate, the node needs to identify a new positionthat would satisfy the motive, e.g., boost overallnetwork performance. Again, the qualification ofthe new position and possibly the search criteriamay vary based on the design attributes. Findingan optimal location for the node in a multi-hop net-work is a very complex problem. The complexity ismainly resulting from two factors. The first is thepotentially infinite number of possible positionsthat a node can be moved to. The second factoris the overhead of keeping track of the networkand the node state information for determiningthe new location. In addition, for every interimsolution considered during the search for an opti-mal position, a new multi-hop network topologymay need to be established in order to compare thatinterim solution to the current or previously pickedpositions.

A mathematical formulation of the node reloca-tion problem may involve huge numbers of param-

eters including the positions of all deployed nodes,their state information (including energy reserve,transmission range, etc.) and the data sources inthe network. In addition, a node may need to knowthe boundaries of the monitored region, the currentcoverage ratio of the network, the location of deadsensor nodes or other information in order to deter-mine its new location. Given the large number ofnodes typically involved in applications of WSNs,the pursuance of exhaustive search will be impracti-cal. In addition, the dynamic nature of the networkmakes the node state and sources of data maychange rapidly; thus the optimization process mayhave to be repeated frequently. Moreover, it maybe undesirable to involve the nodes in complex com-putation since this diverts both the computationcapacity needed for application-level processing,e.g., data fusion, and the energy needed for move-ment of the node. Therefore, approximate and localsolutions, or search heuristics, are more popular inthe context of WSNs [7,13,14].

Managing and justifying the move: Once the newlocation of the node has been picked and confirmedto enhance some desired design attributes, the nodeshould identify a travel path to the new location.The main contributing factors to the path selectionare the total distance to be traveled, the suitabilityof the terrain, the path safety and the risk of disrupt-ing the network operation. Minimizing the traveldistance for the nodes is crucial since the energy


consumed by the mechanical parts in such a move-ment is much more than the communication andcomputation energy. Therefore, the shortest possiblepath should be identified to reach the new location.However, the node also has to pick a path that isphysically feasible to travel over. The node may needto consult a terrain map or rely on special onboardequipments, e.g., cameras, to avoid obstacles anddead ends. The other concern is protecting the nodeduring the move. Since a WSN is usually deployedin harsh environments to detect and track dangeroustargets or events, the node should avoid exposure toharm or getting trapped. For example, the nodeshould not go through a fire to reach the newlocation.

The node should also minimize any negativeimpact on network operation. While the node ison the move, it must ensure that data continues toflow. For example, a sensor may have to increaseits transmission power to cover the planned travelpath in order to make sure that packets will con-tinue to reach it. Continual data delivery preventsthe node from missing important reports duringthe relocation, which may cause an application-levelfailure. Such application-level robustness is a designattribute in itself. Therefore, it is desirable to restrictchanges to the network topology. Avoiding radicalchanges to the data routes limits the disruption ofongoing data traffic, and also curtails the overheadthat the relocation introduces. Again, the node per-forms a trade-off analysis between the gain achievedby going to a new location and the overhead interms of additional energy consumption that themotion imposes on the other nodes. If the motionis justified, the node can physically relocate.

The last issue is whether there are constraints onthe time duration that the node budgets for the move.These constraints may arise in very dynamic environ-ments in which the traffic pattern changes frequently.In these cases, the gains achieved by going to a loca-tion may be lost or degraded very quickly and thenode would find out that it has to move yet againto a third location or even return to the old position.In the worst case, the node continues to move backand forth among these locations. Therefore, a grad-ual approach to the new location may be advisablein order to prevent this scenario.

3.2. Sensor repositioning schemes

While the bulk of published work envisions sen-sors to be stationary, some investigate the possibil-

ity of attaching sensors to moveable entities suchas robots [67,68]. Sensor mobility has been exploitedto boost the performance of WSNs. For example,mobile sensors can re-spread in an area to ensureuniform coverage, move closer to loaded nodes inorder to prevent bottlenecks or increase bandwidthby carrying data to the base-station [15,69–71]. Pro-posed schemes for dynamic sensor positioning in theliterature can be categorized into two groups, basedon when relocation is exploited, into post-deploy-ment or on-demand relocation. We discuss thesetwo categories of relocation in detail in the follow-ing sub-sections.

3.2.1. Post-deployment sensor relocation

This type of relocation is pursued at the conclu-sion of the sensor deployment phase when thesensor nodes are being positioned in the area. Aswe discussed earlier, in most WSN applications,sensor deployment is performed randomly due tothe inaccessibility of the monitored areas. However,this random configuration usually does not provideadequate coverage of the area unless an excessivenumber of nodes are deployed. Alternatively, thecoverage quality can be improved by moving thesensor nodes if they are able to do so. In that case,sensor nodes can be relocated to the regions withinadequate coverage, or no coverage at all. Giventhe energy cost of mechanical movement and thecommunication messages involved in directing themotion, the relocation process should be lightweightand should conclude in a reasonable time.

Wang et al. utilizes each sensor’s ability to movein order to distribute them as evenly as possible inthe region [69]. The goal is to maximize the areacovered within the shortest time duration and withminimal overhead in terms of travel distances andinter-sensor message traffic. The main idea is thateach sensor assesses the coverage in its vicinity afterdeployment and decides whether it should move toboost the coverage. To assess the coverage, a sensornode creates a Voronoi polygon with respect toneighboring sensors, as illustrated in Fig. 14. Everypoint inside a Voronoi polygon is closer to the sen-sor of that polygon, i.e., Si in Fig. 14, than any othersensor. The intersection of the disk that defines thesensing range and the Voronoi polygon representsthe area the sensor can cover. If there are uncoveredareas within the polygon, the sensor should move tocover them.

In order to decide where to reposition a sensor,three methods have been proposed: vector-based

B

Sensing

Range for Si

Si

Voronoi Polygonfor Si

A

Fig. 14. Every sensor Si forms a Voronoi polygon with respect tothe position of its neighboring sensors. The part of the polygonthat lies outside the sensing range is not covered by Si.


(VEC), Voronoi-based (VOR) and Minimax. Themain idea of the VEC method is borrowed fromelectromagnetic theory where nearby particles aresubject to an expelling force that keeps them apart.In the context of WSNs, virtual forces are applied toa sensor node by its neighbors and by the bound-aries of its Voronoi polygon in order to change itslocation. While in VEC the nodes are pushed awayfrom the densely populated areas, VOR pulls thesensors to the sparsely populated areas. In VOR,the sensor node is pulled towards the farthest Voro-noi vertex to fix the coverage hole in the polygon,which is point A for sensor Si in Fig. 14. However,the sensor will be allowed to travel only a distancethat equals half of its communication range, point‘‘B’’ in Fig. 14. This prevents the node from step-ping into the area handled by another sensor thatwas out of reach prior to the move (in other words,not a current neighbor of Si), which can lead to anunnecessary move backward later on. In the Mini-max method, a sensor also gets closer to its farthest

Fig. 15. Sensors pursue relocation iteratively until movem

Voronoi vertex. However, unlike VOR, the Mini-max approach strives to keep most of the other ver-tices of the Voronoi polygon within the sensingrange. It thus relocates the sensor to a point insidethe Voronoi polygon whose distance to the farthestVoronoi vertex is minimized. The Minimax schemeis more conservative in the sense that it avoids cre-ating coverage holes by going far from the closestvertices, leading to a more regularly shaped Voronoipolygon.

The conserved departure from the current sensorposition leads to a gradual relocation, round byround, as shown in Fig. 15. This usually causesthe sensor to zigzag rather than move directly tothe final destination. In order to shorten the totaltravel distance, a proxy-based approach is proposedin [72]. In this approach, the sensor nodes do notmove physically unless their final destination iscomputed. The authors consider a network with sta-tionary and mobile sensors. Mobile sensors are usedto fill coverage holes identified in a distributed wayby stationary nodes. Thus, mobile sensors onlymove logically and designate the stationary sensornodes as their proxies. This approach significantlyreduces the total and average distance traveled bymobile nodes while maintaining the same level ofcoverage as [69]. The approach only increases themessage complexity. However, given that movementis more costly in terms of energy, this increase canbe justified. Nonetheless, this process still can bevery slow and hence prolong the deployment time.

With the objective of reducing the overall deploy-ment time, Wu et al. have proposed another solutionto the same problem based on two-dimensionalscanning of clustered networks, called SMART[15]. The approach adopts a popular scheme for bal-ancing load among nodes in parallel processing

ent no longer improves coverage (taken from [69]).

31 2 32 1

32 22 187 56

24 12 530 1

23 7 31 2

191 5 773 21

106 7 510 4

22 2 22 2

1818 18 1818 18

99 9 99 9

33 3 33 3

2121 21 2121 21

77 7 77 7

1010 10 1010 10

1010 10 1010 10

1010 10 1010 10

1010 10 1010 10

1010 10 1010 10

1010 10 1010 10

a b c

Fig. 16. Steps for SMART: (a) initial 2D mesh, (b) row scan and (c) column scan.


architectures by assigning equal number of tasks toeach processor. This idea is applied to a multi-clusterWSN where each cluster is represented by a squarecell, forming a 2D mesh. An example is shown inFig. 16, which is redrawn from [15]. The number ofsensors annotated on every cell represents the loadof that cluster. Each cluster-head knows only itslocation within the mesh, i.e., row and column indi-ces, and the number of sensors in its cluster. It isassumed that a cluster-head can only communicatewith its counterparts in neighboring cells. Achievinguniform coverage is then mapped to the load-balanc-ing problem with a goal of evening the distributionof sensors among the clusters. To achieve this goal,each cluster-head performs both row-based andcolumn-based scans to exchange load information.In a row-based scan, the leftmost cluster-head for-wards its load, i.e., number of sensors, to its rightneighbor. Each neighbor in the row adds thereceived load to its own load and forwards it untilthe rightmost cluster-head is reached. This cluster-head computes the average load for its row andsends a message back until the left most cluster-headreceives it (Fig. 16b). After the scan process, thesensors are relocated to match the desired nodecount per cluster. That is, the overloaded clustersgive sensors and the underloaded clusters takesensors. The same procedure is also applied withineach column (Fig. 16c). The approach also handlespossible holes in the network when there are clusterswith no sensors. The simulation results in [15] com-pared SMART to VOR, discussed above, withrespect to the number of moves made by the sensorsand the number of rounds before termination.SMART was shown to use the minimum numberof moves. Although it was also shown that SMARTconverges in a fewer number of rounds for densely

populated WSNs, VOR was found to be superiorfor sparsely populated networks.

Another similar post-deployment relocationwork for improving the initial coverage and provid-ing uniform distribution of sensors is presented in[73]. Although the idea is similar to the VEC mech-anism of [69], this time it is inspired by the equilib-rium of particles in Physics. The particles followCoulomb’s law, pushing against each other to reachequilibrium in an environment. Therefore, theauthors define forces for each sensor node in the net-work based on the inter-node distances and the localdensity of nodes. The partial force on a node i fromnode j at time t is expressed as follows:

f i;jt ¼

Dit

l2ðRjpi

t � pjt jÞ

pjt � pi

t

jpjt � pi

tj; ð2Þ

where Dit is the node density in the vicinity of node i

at time t, l is the average node density in the net-work, R is the transmission range and pt is the loca-tion at time t. In Eq. (2), the

Dit

l2 ðRjpit � pj

t jÞ termdefines the force strength. The authors note thatincluding the average node density l is found to

expedite the convergence. The termpj

t�pit

jpjt�pi

t jindicates

the direction of the force. Each sensor’s movementis decided by the combined force applied to that sen-sor by all neighboring nodes. A sensor node is onlyallowed to move a certain distance per time step Insome cases, the node can move back and forth be-tween two locations leading to an oscillation. If anoscillation continues for longer than a preset limit,the node stays at the center of gravity between theoscillation points. To validate the performance,the approach is implemented and compared to asimulated annealing based solution which providesoptimal coverage. The validation results indicate

Choice 1:

Choice 2:

S3

S2

S1S0

S3

S2

S1S0

Fig. 17. Cascaded movement of sensors; S3 replaces S2, S2 settlesin S1’s position and S1 move to where S0 is.


that the proposed self-spreading approach providesnearly optimal coverage and delivers superior per-formance in terms of the total distance traveledand the time to converge.

3.2.2. On-demand repositioning of sensors

Instead of relocating the nodes at the deploymentphase, sensors can be relocated on demand toimprove certain performance metrics such as cover-age or network lifetime. This can be decided duringthe network operation based on the changes ineither application-level needs or the network state.For instance, the application can be tracking afast moving target which may require reposition-ing of some sensor nodes based on the new locationof the target. Furthermore, in some applicationsthere can be an increasing number of non-function-ing nodes in a particular part of the area, necessitat-ing the redistribution of available sensors. Inaddition to improving coverage, the energy con-sumption can be reduced through on-demandrelocation of sensors in order to reach the best effi-cient topology.

The approach presented in [13] performs sensorrelocation to counter holes in coverage caused bysensors failure. The idea is simply to identify somespare sensors from different parts of the networkthat can be repositioned in the vicinity of the faultynodes. The selection of the most appropriate choiceamong multiple candidate spare nodes is based onthe recovery time and overhead imposed. Both crite-ria would favor close-by spares over distant ones.Minimizing the recovery time can be particularlycrucial for delay sensitive applications. The over-head can be in the form of energy consumptiondue to the node’s travel and due to the messageexchange. The latter is especially significant if sparesare picked in a distributed manner. In order todetect the closest redundant sensor with low mes-sage complexity, a grid-based approach is proposed.The region is divided into cells with a designatedhead for each cell. Each cell-head advertises/requests redundant nodes for its cell. A quorum-based solution is proposed to detect the intersectionof advertisements and requests within the grid. Oncethe redundant sensor is located, it is moved to thedesired cell without disrupting the data traffic oraffecting the network topology.

Since moving a node over a relatively long dis-tance can drain a significant amount of energy, acascaded movement is proposed. The idea is todetermine intermediate sensor nodes on the path

and replace those nodes gradually. That is, theredundant sensor will replace the first sensor nodeon the path. That node also is now redundant andcan move to replace the second sensor node, andso on. For the example shown in Fig. 17, which isredrawn from [13], rather than directly moving S3

to the location of S0, in choice 2 all sensors S3, S2,and S1 move at the same time and replace S2, S1,and S0, respectively in order to minimize the reloca-tion time. The path is selected such that it will min-imize the total mechanical movement energy and atthe same time maximize the remaining energy ofsensor nodes. In order to determine such a path,Dijkstra’s least cost path algorithm is used. Theoverall solution is also revisited to provide a distrib-uted approach for determining the best cascadingschedule.

When validated, the approach has outperformedVOR of [69] with respect to the number of sensorsinvolved in the relocation, the total consumedenergy and the total remaining energy. In addition,cascaded movement has delivered much better per-formance in terms of relocation time, energy costand remaining energy than direct movement. How-ever, obviously the cost of maintaining a grid, selec-tion of cell-heads and identifying redundant nodeswill grow dramatically as the number of nodesincreases. For scalability, a hierarchical solutionmight be needed to restrict the size of the regionand the cost of movement.

Coverage improvement is also the objective ofrelocating imaging sensors in [70]. Stationary cam-eras may not provide the desired coverage whenthere are environmental peculiarities, e.g., movingobstacles, in the event area as seen in Fig. 18. Thus,moving the cameras in order to avoid obstacles thatblock their vision would increase the coverage. Themobility for the camera nodes is made possiblethrough providing a traction mechanism under eachcamera that will enable motion in one dimension, asis also implemented in Robomote [67]. This mobility

Fig. 19. Nodes close to the data collector die rather quickly dueto traffic overload; forcing the more distant nodes to relay thedata to the data collector.

Camera A Camera B

Obstacle

Fig. 18. An obstacle reduces the coverage of two camerasdirected at different areas.


feature on cameras is actuated when the coverageof the monitored areas falls below a certain ratio.The experiments performed with real-life targettracking applications have verified that the mobilecameras can increase the coverage of the monitoredarea and thus decrease the target miss-ratio signifi-cantly when compared to stationary-node basedsetups.

Finally, we would like to note that the work ofDasgupta et al. [41], which we discussed in Section2.2, can potentially fit under the category ofdynamic positioning if sensors are relocated afterdeployment. However, the repositioning is done atthe network planning stage to form the most effi-cient topology and thus we categorized the schemeas static.

3.3. Repositioning data collectors

As discussed earlier, sensor data is gathered ateither the base-station or CHs for aggregation andpossibly additional processing consistent with thecomputational capabilities of such data collectors(DCs). Dynamic repositioning of DCs has also beenpursued as a means for boosting network perfor-mance, dealing with traffic bottlenecks and prevent-ing interruptions in network operation. Unlikesensor repositioning, the goal for relocating DCs isusually not local to the individual node and involvesnumerous network state parameters. In this section,we focus on approaches that consider a single DCor uncoordinated repositioning of multiple DCs ifmore than one exist. Section 4 discusses collabora-

tive relocation of nodes, which has not receivedmuch attention in the literature and has numerousopen research problems.

3.3.1. Repositioning for increased network longevity

Although energy-aware multi-hop routing doesdynamically adapt to changes in sensor energy andtraffic pattern, sensors near a data collector (DC)die quickly as they are the most utilized nodes inthe network [40,74]. Consequently, nodes that arefurther away from the DC are picked as substituterelays, as depicted in Fig. 19. However, the amountof energy these nodes spend to communicate withthe DC is considerably higher. This effect can spreadin a spiral manner, draining the sensors energy andhence shortening the lifetime of the network. Tostop such a pattern of energy depletion, the DC isrepositioned [75].

The main idea is to move the DC towards thesources of highest traffic. The traffic density (P)times the transmission power (ETR) is used as ametric for monitoring the network operation andsearching for the best DC location. The idea is totrack changes in the nodes that act as the closesthop to the DC and the traffic density going throughthese hops. If the distance between the DC andsome of the nodes that are in direct communicationis smaller than a threshold value d, the DC will qual-ify the impact of these nodes on the overall networklifetime by considering the number of packets rou-ted through them. If the total P · ETR decreasesby more than a certain threshold D, the DC willconsider relocating to a new position. Mathemati-cally, the relocation takes place if:X

8i2SR

EðTRiÞ � P i �X

8j2SnewR

EðTRjÞ � P j > D; ð3Þ


where

• SR is the set of sensors that are one hop awayfrom the data collector on active routes,

• SnewR is the set of sensors that are one hop away

from the data collector at a new location,• Pi is the packet traffic going through node i, mea-

sured as the packet count per frame,• EðTRiÞ is the energy consumed by node i for trans-

mitting a packet to the next hop, and• the summations measure the total energy con-

sumed for packet transmission through all thenodes at the old location and new location ofthe data collector, respectively.

While such positioning will be ideal for high traf-fic paths, it can worsen the performance on pathswith lower traffic density or which are topologicallyopposite to the direction of the DC’s motion. There-fore, before confirming the move, it has been recom-mended that the DC validates the overall impact ontransmission energy by factoring in the possibleextension of some data paths and the cost of signal-ing overhead to those sensor nodes affected by themove. In addition, when the DC starts to move,the data gathering process still continues and thusthe routes should be adjusted before the DC getsout of range of some sensors (if any). Unlike the sta-tic approaches discussed in Section 2, route adjust-ment is an issue in dynamic DC positioning. Thisissue is handled in [75] by either increasing the trans-mission power or designating additional forwardersensors. The change in DC position may also intro-duce shadowing or multi-path fading to some links.Slow or gradual advance towards the new positionis shown to be effective in avoiding unexpected linkfailures that may cause negative performanceimpacts, since it allows the DC to rethink the suit-ability of the newly-selected position and/or thedecision to move further.

Through simulation, this approach for relocatingthe DC is shown to not only increase the networklongevity but also to enhance other performancemetrics like latency and throughput. First, commu-nication-related energy consumption and theaverage energy per packet are reduced as the DCmoves closer to the area where more nodes are col-lecting data. In addition, nodes collecting the mostdata are closer to the DC and fewer hops areinvolved, lowering the overall latency time for datacollection. Moreover, the packet throughput growssince most messages pass through fewer hops and

travel shorter distances, making them less likely tobe dropped.

3.3.2. Enhancing timeliness of delay-constrainedtraffic

In addition to boosting network longevity, repo-sitioning the DC is useful when real-time traffic withcertain end-to-end delay requirements is involved[76]. For instance, when routes to the DC get con-gested, most requests for establishing paths forreal-time data may be denied or the deadline missrate of real-time packets may increase significantly.Traffic congestion can be caused by an increase inthe number of real-time data packets coming fromnodes close to a recent event. In such circumstances,it may be infeasible to meet the requirements forreal-time data delivery. Therefore, repositioningthe DC has been recommended in order to spreadthe traffic on additional hops and increase the feasi-bility of meeting the timeliness requirements.

A trigger for such relocation can be an unaccept-able increase in the miss rate of real-time packets, orjust a desire to increase timeliness even if the missrate is at a level that is tolerable by the application.To boost timeliness, the DC is moved to the loca-tion of, or close to, the most heavily loaded node[76]. The rationale is to split the incoming trafficpassing through that node without extending thedelay experienced by real-time packets over otherroutes, as shown in Fig. 20. Such loaded nodes arepicked based on the real-time traffic service rate,often determined during route setup in order to allo-cate bandwidth to both real-time and non-real-timetraffic. Again, the advantages of the relocation haveto be qualified to make sure that the overhead is jus-tified. It is also worth noting that in this approachthe impact on link quality is not a major concernwhen the DC moves close to a heavily loaded node.This is because it is unlikely that the node is experi-encing a disruptive level of interference while beingable to relay a high volume of real-time packets.

Handling the DC motion is similar to theapproach described in Section 3.3.1. As long asthe DC remains within the transmission range ofall the last-hop nodes, the current routes can bemaintained by only adjusting the transmissionpower. If the new location places the DC out ofthe transmission range of some of the last hop nodesin the current routes, new forwarder nodes that arenot involved in any routing activity are selected. Itwas argued that those unused nodes introduce verylittle queuing delay, which is desirable for on-time

Fig. 20. (a) The DC (denoted as B) is relocated to the location of A if the delay is not extended on the path from C to B (b) If real-timetraffic through C is affected, B is relocated close to C while still splitting traffic at A.

High Risk

Medium RiskMedium/Low Active Data Source

Highly Active Data Source


delivery of all real-time packets that use these nodesas relays. We note, however, that designating newforwarder nodes is still more costly than just adjust-ing the transmission power. Therefore, if the newlocation imposes excessive topology changes, alter-native positions which will cause no or minimaltopology changes should be considered.

Low Risk

DataCollector

Fig. 21. Performance-centric relocation may move a data collec-tor dangerously close to serious events in the environment andthus risk the loss of such critical asset.

3.3.3. Maintaining uninterrupted operation

Relocating data collectors has also been pursuedto keep WSNs operational without interruption.Noting that some nodes may be isolated (the net-work may even become partitioned) if one of theDC nodes is damaged or becomes unreachable tosensors, some of the published schemes repositionedDC nodes in order to protect them and sustain net-work connectivity.

Keeping the DC away from harm is the goal ofthe GRENN approach proposed in [77]. Two mainscenarios motivate GRENN to reposition a DC.First, the DC may be in the way of an approachingserious event, such as a fast spreading fire, or aharmful target like an enemy tank. The second sce-nario may be caused by a desire to boost networkperformance by moving the DC node towards thedata sources, similar to the schemes discussed inthe previous two sub-sections. The latter scenario,as depicted in Fig. 21, may actually expose theDC to hazards and put it at high risk. Thus, han-dling such scenarios would be subject to perfor-mance and safety trade-offs.

To assess the safety implication of repositioningthe DC, an evolutionary neural networks based for-mulation is pursued. The idea is to track the DCsafety levels at different locations and use this infor-

mation to define the parameters of the DC safetymodel. Then the threat implication is estimated asa function of the proximity to reported events andthe severity of those events. If the location of theevent is not accurately known, the data volumerelated to that event and the location of the report-ing sensors are factored in. An objective function isthen formed to balance safety and performancegoals and used to guide the search for the new loca-tion of the DC. GRENN has been further extendedin [78] to identify a safe route for the DC to its newposition. The same safety assessment model isemployed to guide the selection of the travel path.The effect of the DC’s motion on the network per-formance is also factored in by measuring thethroughput at various spots the DC will visit along


the way. The objective function is extended to notonly trade-off the safety of the DC and the networkperformance but also to minimize the traveldistance.

Shen et al. [79] have investigated the placement ofmobile access points in order to connect nodes inisolated networks through airborne units or satel-lites. The deployed nodes usually do not haveexpensive radios for long haul communication andusually serve limited geographical areas. The limitedcommunication range may result in partitioning thenetwork, leaving some nodes unreachable to someothers. To overcome such structural weaknesses inthe network, mobile base-stations are employed tointerconnect isolated sub-networks through an air-borne relay, such as an unattended air vehicle(UAV) or satellite. As depicted in Fig. 22, which isredrawn from [79], a mobile base-station acts asan access point for the nodes in its neighborhood.The approach is to place the mobile access point(MAP) at the centroid of the sub-network. Nodes

Mobile access point

Airborne unit

Regular ground node

Fig. 22. Mobile access points are placed to connect isolated sub-networks through an airborne unit so that the end-to-end delayamong nodes in different sub-networks is minimized.

Table 3A comparison of dynamic node repositioning schemes

Paper Node type Relocation trigger Primary object

[13] Sensor Sensor failures Coverage of ho[15] Sensor Poor coverage Coverage[41] Sensor High traffic Network lifetim[69] Sensor Poor coverage Coverage[70] Sensor Poor image coverage Image coverag[72] Sensor Poor coverage Coverage[73] Sensor Poor coverage Coverage[75] Data collector High traffic Total power[76] Data collector Poor timeliness Timeliness[77] Data collector Application dependent Physical securi[78] Data collector Application dependent Physical securi[79] Data collector Network partitioning Connectivity[80] Data collector Node mobility Connectivity

will send messages which get flooded through thesub-network until they reach the MAP. As theMAP receives the packets, it records the paths eachtook to reach it. From this information the MAPbuilds a tree modeling the network, with itself atthe root. The MAP then moves to the centroid ofthis tree. The approach shares some similarities tothat of [80].

Table 3 summarizes the dynamic positioningschemes covered in this section.

4. Open research problems

While significant progress has been made inresearching the optimization of node positioningin WSNs, many challenging problems remain. Inthis section, we highlight open research problems,identify the issues involved and report on ongoingwork and preliminary results. We categorize openproblems into coordinated dynamic placement ofmultiple nodes and the positioning of sensors inthree-dimensional application setups.

4.1. Coordinated multi-node relocation

In many application setups, nodes coordinateamong themselves in order to efficiently and effec-tively handle application-level requirements. Exam-ples of such applications include robotic-basedland-mine detection and deactivation, multi-roverexploration of distance planets, employing a sen-sor-fleet for oceanic studies, etc. However unlikethe case discussed in the previous section, suchapplication-level coordination requires the reloca-tion process to care for inter-node networkingissues. For instance, consider the scenario depicted

ive Secondary objective Constraint

les Travel distance Convergence timeTravel distance Convergence time

e – CoverageTravel distance Convergence time

e – –Travel distance Convergence timeTravel distance Convergence timeThroughput –Avg. delay –

ty Network lifetime –ty and throughput Travel distance –

– –– –

Fig. 23(b). CH3 moves to maintain its communication link withCH4, probably loosing connectivity to CH2. However, cluster-heads still form a connected graph.

CH1

CH4

CH2

CH3

Sensor nodes

Cluster-head

Event of Interest

Event of Interest

Sj

Si

Fig. 23(a). A sample multi-cluster sensor network architecture,where each cluster is handled/managed by a distinct cluster-head.


in Fig. 23(a). A set of sensors are deployed in anarea of interest. The sensors have been partitionedinto non-overlapping clusters; each is managed bya distinct cluster-head (CH). To simplify the discus-sion, let us assume that CHs are the only nodesallowed or capable to relocate.

Given the commonly uneven distribution of sen-sor nodes or the battery exhaustion or failure of sen-sors in a certain region, some events would be hardto monitor or would overburden the scarce networkresources in the proximity of the event. In Fig. 23(a)two sample events articulate such an issue. Thedepicted events, may be targets or fires, are reportedby very few sensors for which the events happen tobe within their detection range. Clearly, the intra-cluster network topology for both clusters 1 and 4are not efficient since the data is routed over manyhops and may not be arriving at CH1 and CH4 reli-ably and/or on-time. In addition, some of the nodesthat are involved in relaying data may not haveabundant energy reserve to keep the path stablefor an extended duration. Furthermore, CH1 andCH4 may have advanced sensing instruments thatcan be employed to boost the fidelity of the assess-ment. The following are some of the challenges thatCHs may face when trying to boost the efficiency ofthe data collection and the operation of the intra-cluster network:

1. CH1 may decide to relocate to better serve theevent tracked by sensors in its cluster; that is,to monitor the event in a timely manner or tominimize the energy consumption at relaying sen-sor nodes. However, this repositioning may placeCH1 out of the communication range of CH4.Such a move can only be acceptable if CH1 does

not need to interact with CH4, which is unlikely,or when an alternative path can be establishedwith an acceptable delay bound. Referring toFig. 23(a), relocating CH1 would be feasible ifthe communication path (CH1, CH3, CH4) meetsthe timeliness requirements.

2. The network must consider the possibility that adifferent cluster could handle the event. Forexample, while CH1 can relocate to better servethe event, the data can also be routed from Si toCH2 over a shorter and more energy efficientpath. That would require changing the associa-tion of Si from cluster 1 to cluster 2, at least tem-porarily. This modification of cluster membershipwould surely impose an overhead. Such a trade-off is unavoidable.

3. When relocation of a CH is the only option forboosting the efficiency of the operation within acluster, a ripple effect may be caused throughoutthe network. Consider, for example, the relocationof CH4 close to Sj. Although such a close proxim-ity to the event would have a positive impact onthe operation in cluster 4, moving CH4 that farmakes it unreachable to other CHs. This scenariois not acceptable in a collaborative computingenvironment. Also, CH4 may become isolatedfrom the base-station if it cannot directly reachit. A more complex alternative is to relocate multi-ple CHs in order to maintain connectivity.Fig. 23(b) illustrates a possible multi-CH reloca-tion that better serves the event reported by Sj.Basically, CH3 gets closer to the new location ofCH4 in order to prevent their communication linkfrom getting broken, possibly at the expense ofloosing connectivity to CH2. This solution couldhave been more complex if CH1 did not have a


link to CH2, forcing it to move in order to be in therange of CH3 and CH2.

Now imagine all nodes can move subject to inter-node topology constraints. Obviously, coordinatedmulti-node relocation will be even more complexand can introduce lots of trade-offs. In addition,simultaneous relocation of nodes raises the issue ofconvergence. Repositioning one node can make themove that was concurrently made by another nodeeither redundant and/or incompatible. For example,an autonomous decision by multiple nodes for relo-cating toward an event may lower the coverage inanother region while increasing the fidelity of moni-toring the event more than necessary. Lightweightsynchronization protocols and/or localized schemesmay be required, especially for large networks, inorder to ensure convergence of the relocation pro-cess to an optimized network state, consistent withthe objectives and constraints of the move.

We envision coordinated multi-node relocation tobe a promising research direction. Only a little atten-tion has been paid to tackling the challenges ofcoordinated multi-node relocation. One of the fewattempts is reported in [81], where a dynamic multi-node positioning is proposed in order to improvethe network longevity. The motion of CHs isrestricted to maintain the connectivity of the inter-CH network, as seen in Fig. 24. When a CH is tobe relocated, its links with its neighbors are checkedfirst before the relocation is performed. If changingthe position of a CH will partition the network, itsneighbors are to move to restore broken links. Forexample in Fig. 24, if CH3 is to move toward CH2,CH5 would follow along to avoid being disconnectedfrom the network. In order to prevent simultaneous

CH

CH

CH

CH

CH

CLuster # 5

CLuster # 4

CLuster # 3

CLuster # 2

CLuster # 1

Fig. 24. The CORE approach strives to maintain a connectedinter-CH network by restricting the motion of the individual CHnodes.

relocations, a mutual-exclusion based mechanism isused. The idea is to require exclusive access to a glo-bal token in order to perform the relocation. A CHwill cease all motion until a token is granted. Eachsensor’s association to a cluster changes based onthe proximity of CHs nodes to the individual sensor.

Another notable work is the COCOLA approach[63], which has been discussed in section 2.3. UnlikeCORE [81], mutual exclusion is supported in a dis-tributed fashion. The goal is again to maintain con-nectivity among CH nodes when some of themmove. The synchronization protocol is simply tocoordinate with immediate neighbors based on apredefined priority system. The priority can be sta-tic, e.g., ID-based like CORE, or a dynamic figurethat relates to the application or the state of CHs.Assuming distinct node priorities, it was proven thatmultiple CHs can simultaneously relocate as long asthey are not neighbors.

Although not particularly geared toward WSNs,the approach [82] also addresses inter-CH coordina-tion. The authors consider a network of robotsthat collaborate on application tasks and move ondemand. The network is assumed to be 2-connectedat startup, i.e., every robot can reach every otherrobot over at least two vertex-disjoint paths. Thegoal is to sustain this level of connectivity even underlink or node failure. The approach is based on relo-cating some of the robots in order to reestablish the2-connectivity. The authors modeled the recovery asan optimization formulation in which nodes are toreposition using the minimal total travel distance.Two distributed algorithms were proposed. The firstalgorithm is called contraction and strives to moverobots inward to the centroid of the deployed nodes(robots) in order to boost connectivity. To avoidexcessive contraction, a gradual motion is proposed.The second approach employs graph theory to iden-tify cut vertices and orchestrate block movements.The idea is to pick nodes that have a degree of 1 tomove towards a cut vertex in order to establishnew links that serve as alternatives to the criticallinks between cut vertices. Although this work doesnot account for coverage requirements and sensor-base-station connectivity, the approach is worthexploring in the context of WSNs.

4.2. Node positioning in three-dimensional

application setups

The scope of the majority of published papers onstatic and dynamic node positioning strategies is

Fig. 26. Applying the truncated octahedron based node place-ment strategy in a 20 m · 20 m · 20 m 3-D space. Figure is from[84].


limited to terrestrial networks, where covered spaceis two-dimensional (2-D), and to small indoor set-ups. Even for applications that employ range find-ers, imagers and video sensors the focus has beenon the quality of the collected data and coverageof some 2-D space, basically by managing the angu-lar orientation of the sensors in a plane. As we men-tioned in Section 2.1, many of the popular coverageanalysis and placement strategies pursued for 2-Dspace become NP-Hard in 3-D [9,83]. However,with the increased interest in applications of sensornetworks in space exploration, airborne and under-water surveillance, oceanic studies, storm tracking,etc. tackling the contemporary design issues suchas coverage and connectivity in 3-D has become anecessity. We expect optimization strategies fornode positioning in WSN applications in large-scale3-D setups to be one of the hot topics of researchin the next few years. In fact, some preliminaryresearch results have started to emerge.

Alam and Haas [84] have investigated the prob-lem of achieving maximal 3-D coverage with theleast number of sensors. In the conventional 2-Dscenario, sensor coverage is modeled as a circleand the maximal coverage problem is mapped to acircle packing formulation which has a polynomialtime solution. Given the high complexity of thesphere-packing problem [9], the authors argue thatspace filling polyhedrons would be more suitablefor 3-D applications. The idea is to fill the 3-Dapplication space with the least number of polyhe-drons in order to provide maximal coverage, ideally100%. The metric used for coverage was called a vol-

umetric quotient, which is defined as the ratio of thevolume of the shape to be used to that of its enclos-ing sphere. The paper compares the truncated octa-

Fig. 25. A truncated octahedron has 14 faces; 8 are hexagonaland 6 are square. The length of the edges of hexagons and squaresare the same. Figure is from [85].

hedron, the rhombic dodecahedron, the hexagonalprism, and the cube in terms of volumetric quotient.The conclusion is that truncated octahedrons(Fig. 25), created through the use of the Voronoitessellation of 3-D space (Fig. 26), yield the bestresults. In addition to coverage, the paper studiesthe relationship between the sensing and transmis-sion ranges so that a connected topology is estab-lished. For truncated octahedron it has beenconcluded that connectivity is ensured if the trans-mission range of the employed nodes is at least1.7889 times the sensing range.

Ravelomanana [86] has studied the properties ofthe network topologies that result from randomdeployment of nodes in a 3-D region of interest.The main goal is to analyze the implications of thesensing and communication ranges on coverageand connectivity, respectively. Considering a uni-form distribution of nodes, the author derives con-ditions for the node transmission range r requiredfor achieving a degree of connectivity d, where everynode has at least d neighbors. In addition, the aver-age path length between two nodes in the network isformulated as a function of d and r. The same anal-ysis is performed for coverage, basically estimatingthe sensing range required to achieve a certaindegree of coverage in a region using a pre-deter-mined number of nodes. The work strives to providetheoretical bounds that can help in preliminarydesign and feasibility studies of 3-D WSNs. Pompiliet al. [33], have used these bounds to validate theeffectiveness of their random node deployment


scheme. The idea is to randomly drop sensors fromthe ocean surface and control their depth below thesurface by wires that attach them to some anchorsat the bottom of the ocean. The validation experi-ments have been very consistent with the theoreticalresults of [86] by identifying the right combinationof network size and sensing range for achieving100% coverage.

5. Conclusion

Wireless sensor networks (WSNs) have attractedlots of attention in recent years due to their potentialin many applications such as border protection andcombat field surveillance. Given the criticality ofsuch applications, maintaining efficient networkoperation is a fundamental objective. However, theresource-constrained nature of sensor nodes andthe ad-hoc formation of the network, often coupledwith unattended deployment, pose non-conven-tional challenges and motivate the need for specialtechniques for designing and managing WSNs. Inthis paper, we have discussed the effect of nodeplacement strategies on the operation and perfor-mance of WSNs. We categorized the variousapproaches in the literature on node positioning intostatic and dynamic. In static approaches, optimizednode placement is pursued in order to achieve somedesired properties for the network topology and cov-erage. On the other hand, dynamic repositioning ofnodes after deployment or during the normal net-work operation can be a viable means for boostingthe performance. Unlike the initial careful place-ment, node repositioning can assist in dealing withdynamic variations in network resources and thesurrounding environment. We have identified thetechnical issues pertaining to relocating the nodes;namely when to reposition a node, where to moveit and how to manage the network while the nodeis in motion. We have surveyed published techniquesfor node positioning and compared them accordingto their objectives, methodologies and applications.

In our opinion, static strategies are more practi-cal when a deterministic node placement is feasibleand when the cost of nodes is not an issue. Forexample, when populating a sensor network tomonitor an oil pipeline or to conduct security sur-veillance, pursuing a static strategy would yieldthe best coverage and ensure connectivity usingthe fewest number of sensors. When random distri-bution of nodes is the only option, a static strategyis not a recommended choice unless the cost of the

individual sensors is so insignificant that a very largepopulation of sensors can be employed without con-cern. On other hand, the sole use of mobile nodeswould not be practical due to the increased costand management overhead. Instead, we envisionthat a mix of static and dynamic schemes wouldbe the most effective and efficient placement strategyfor large-scale and mission-critical WSN applica-tions. Low to medium density of stationary nodescan be employed in the monitored region throughrandom distribution. In addition, significantly fewermovable nodes should be added. Such movablenodes can be dynamically relocated to fill coveragegaps, deal with changes in user interests, or recoverfrom connectivity problems.

We have identified the coordinated multi-noderepositioning problem as an open research area. Lit-tle attention has been paid to such a challenging andinteresting area. We also envision that the nodeplacement problem in 3-D will need an increasedattention from the research community in order totackle practical deployment scenarios. As we indi-cated, most of the published work considered 2-Dregions, and numerous issues are not tackled inthe 3-D space. Applications like video and underwa-ter surveillance raise a number of interesting nodeplacement challenges. Finally, we expect domainand node-specific positioning to gain increasedattention with the growing list of WSN applicationsand the availability of sophisticated models of thecapabilities of sensor nodes.

References

[1] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci,Wireless sensor networks: a survey, Computer Networks 38(2002) 393–422.

[2] C-Y. Chong, S.P. Kumar, Sensor networks: evolution,opportunities, and challenges, Proceedings of the IEEE 91(8) (2003) 1247–1256.

[3] K. Akkaya, M. Younis, A survey on routing protocols forwireless sensor networks, Elsevier Journal of Ad HocNetworks 3 (3) (2005) 325–349.

[4] P. Naik, K.M. Sivalingam, A survey of MAC protocols forsensor networks, in: C.S. Raghavendra, K.M. Sivalingam,T. Znati (Eds.), Wireless Sensor Networks, Kluwer Aca-demic Publishers, Norwell, MA, 2005, pp. 93–107.

[5] N. Sadagopan, B. Krishnamachari, A. Helmy, Active queryforwarding in sensor networks (ACQUIRE), Journal of AdHoc Networks 3 (1) (2005) 91–113.

[6] A. Cerpa, D. Estrin, ASCENT: Adaptive self-configuringsensor networks topologies, in: Proceedings of the 21stInternational Annual Joint Conference of the IEEE Com-puter and Communications Societies (INFOCOM’02), NewYork, NY, June 2002.


[7] A. Efrat, S. Har-Peled, J.S.B. Mitchell, Approximationalgorithms for two optimal location problems in sensornetworks, in: Proceedings of the 3rd International Confer-ence on Broadband Communications, Networks and Sys-tems (Broadnets’05), Boston, Massachusetts , October 2005.

[8] X. Cheng, DZ Du, L. Wang, B. Xu, Relay sensor placementin wireless sensor networks, ACM/Springer Journal ofWireless Networks, in press.

[9] S. Poduri, S. Pattem, B. Krishnamachari, G.S. Sukhatme,Sensor network configuration and the curse of dimension-ality, in: Proceedings of the 3rd IEEE Workshop onEmbedded Networked Sensors, Cambridge, MA, May 2006.

[10] J. Pan, L. Cai, Y.T. Hou, Y. Shi, S.X. Shen, Optimal base-station locations in two-tiered wireless sensor networks, IEEETransactions on Mobile Computing 4 (5) (2005) 458–473.

[11] S.S. Dhillon, K. Chakrabarty, Sensor placement for effectivecoverage and surveillance in distributed sensor networks, in:Proceedings of the IEEE Wireless Communications andNetworking Conference (WCNC’03), New Orleans, LA,March 2003.

[12] T. Clouqueur, V. Phipatanasuphorn, P. Ramanathan andK.K. Saluja, Sensor deployment strategy for target detec-tion, in: Proceedings of the 1st ACM international Work-shop on Wireless Sensor Networks and Applications(WSNA ’02), Atlanta, Georgia, September 2002.

[13] K. Akkaya, M. Younis, M. Bangad, Sink repositioning forenhanced performance in wireless sensor networks, Com-puter Networks 49 (2005) 434–512.

[14] G. Wang, G. Cao, T. La Porta, W. Zhang, Sensor relocationin mobile sensor networks, in: Proceedings of the 24thInternational Annual Joint Conference of the IEEE Com-puter and Communications Societies (INFOCOM05)Miami, FL, March 2005.

[15] J. Wu, S. Yang, SMART: A scan-based movement assistedsensor deployment method in wireless sensor networks, in:Proceedings of the 24th International Annual Joint Confer-ence of the IEEE Computer and Communications Societies(INFOCOM’05), Miami, FL, March 2005.

[16] A. Brooks, A. Makarenko, T. Kaupp, S. Williams, H.Durrant-Whyte, Implementation of an indoor active sensornetwork, in: Proceedings of the 9th International Symposiumon Experimental Robotics Singapore, June 2004.

[17] V.A. Petrushin, G. Wei, O. Shakil, D. Roqueiro, V.Gershman, Multiple-sensor indoor surveillance system, in:Proceedings of the 3rd Canadian Conference on Computerand Robot Vision (CRV’06), Quebec city, June 2006.

[18] L. Krishnamurthy et al., Design and deployment of indus-trial sensor networks: experiences from a semiconductorplant and the North Sea, in: Proceedings of the 3rd ACMConference on Embedded Networked Sensor Systems (Sen-Sys’05), San Diego, CA, November 2005.

[19] J. Paek, K. Chintalapudi, J. Cafferey, R. Govindan, SamiMasri, A wireless sensor network for structural healthmonitoring: performance and experience, in: Proceedings ofthe Second IEEE Workshop on Embedded NetworkedSensors (EmNetS-II), Syndney, Australia, May 2005.

[20] K. Mechitov, W. Kim, G. Agha, T. Nagayama, High-frequency distributed sensing for structure monitoring, in:Proceedings of the First International Conference on Net-worked Sensing Systems, Tokyo, Japan, June 2004.

[21] J. Berry, L. Fleischer, W.E. Hart, C.A. Phillips, Sensorplacement in municipal water networks, in: Proceedings of

the World Water and Environmental Resources Conference,Philadelphia, Pennsylvania, June 2003.

[22] J.-P. Watson, H. Greenberg, W.E. Hart, A multiple-objec-tive analysis of sensor placement optimization in waternetworks, in: Proceedings of the World Water and Environ-ment Resources Congress, Salt Lake City, June 2004.

[23] I.F. Akyildiz, D. Pompili, T. Melodia, Underwater acousticsensor networks: research challenges, Journal of Ad HocNetworks 3 (3) (2005) 257–279.

[24] H.H. Gonzalez-Banos, J.C. Latombe, A randomized art-gallery algorithm for sensor placement, in: Proceedings ofthe 17th ACM Symposium on Computational Geometry(SoCG’01), Boston, MA, June 2001.

[25] L. Navarro, J. Dolan, P. Khosla, Optimal sensor placementfor cooperative distributed vision, in: Proceedings of theInternational Conference on Robotics and Automation(ICRA’04), New Orleans, LA, April 2004.

[26] J. Tang, B. Hao, A. Sen, Relay node placement in large scalewireless sensor networks, Computer Communications, spe-cial issue on Wireless Sensor Networks 29 (2006) 490–501.

[27] M. Ishizuka, M. Aida, Performance study of node placementin sensor networks, in: Proceedings of the 24th InternationalConference on Distributed Computing Systems Workshops– W7: EC (Icdcsw’04) – Volume 7, Hachioji, Tokyo, Japan,March 2004.

[28] K. Xu, H. Hassanein, G. Takahara, W. Wang, Relay nodedeployment strategies in heterogeneous wireless sensornetworks: single-hop communication case, in: Proceedingsof the IEEE Global Telecommunication Conference (Glo-becom’05), St. Louis, MO, November 2005.

[29] K. Xu, H. Hassanein, G. Takahara, Q. Wang, Relay nodedeployment strategies in heterogeneous wireless sensornetworks: multiple-hop communication case, in: Proceedingsof 2nd IEEE Conference on Sensor and Ad Hoc Commu-nications and Networks (SECON’05), Santa Clara, CA,September 2005.

[30] C.-F. Huang, Y.-C. Tseng, The coverage problem in awireless sensor network, in: Proceedings of the ACM 9thAnnual International Conference on Mobile Computing andNetworking (MobiCom’03), San Diego, CA, September2003.

[31] A. Boukerche, X. Fei, R.B. Araujo, A coverage preservingand fault tolerant based scheme for irregular sensing range inwireless sensor networks, in: Proceedings of the 49th AnnualIEEE Global Communication Conference (Globecom’06),San Francisco, CA, November 2006.

[32] S. Meguerdichian, F. Koushanfar, M. Potkonjak, M.B.Srivastava, Coverage problems in wireless ad-hoc sensornetworks, in: Proceedings of the 20th International AnnualJoint Conference of the IEEE Computer and Communica-tions Societies (INFOCOM’01), Anchorage, Alaska, April2001.

[33] D. Pompili, T. Melodia, I.F. Akyildiz, Deployment analysisin underwater acoustic wireless sensor networks, in: Pro-ceedings of the ACM International Workshop on Under-Water Networks (WUWNet), Los Angeles, CA, September2006.

[34] E.S. Biagioni, G. Sasaki, Wireless sensor placement forreliable and efficient data collection, in: Proceedings of the36th Annual Hawaii international Conference on SystemSciences (HICSS’03) – Track 5 – Volume 5, Big Island,Hawaii, January 2003.


[35] K. Kar, S. Banerjee, Node placement for connected coveragein sensor networks, in: Proceedings of the Workshop onModeling and Optimization in Mobile, Ad Hoc and WirelessNetworks (WiOpt’03), Sophia Antipolis, France, 2003.

[36] J. Bredin, E. Demaine, M. Taghi Hajiaghayi, D. Rus,Deploying sensor networks with guaranteed capacity andfault tolerance, in: Proceedings of the 6th ACM Interna-tional Symposium on Mobile Ad Hoc Networking andComputing (MobiHOC’05), Urbana-Champaign, Illinois,May 2005.

[37] P. Jacquet, Geometry of information propagation in mas-sively dense ad hoc networks, in: Proceedings of the 5thACM Symposium on Mobile Ad-Hoc Networking andComputing (MobiHOC’04), Roppongi Hills Tokyo, Japan,May 2004.

[38] S. Toumpis, L. Tassiulas, Packetostatics: deployment ofmassively dense sensor networks as an electrostatics prob-lem, in: Proceedings of the 24th IEEE Conference onComputer Communications and Networking (INFO-COM’05), Miami, FL, March 2005.

[39] S. Toumpis, G.A. Gupta, Optimal placement of nodes inlarge sensor networks under a general physical layer model,in: Proceedings of 2nd IEEE Conference on Sensor and AdHoc Communications and Networks (SECON’05), SantaClara, CA, September 2005.

[40] M. Younis, M. Youssef, K. Arisha, Energy-aware manage-ment in cluster-based sensor networks, Computer Networks43 (5) (2003) 649–668.

[41] K. Dasgupta, M. Kukreja, K. Kalpakis, Topology-awareplacement and role assignment for energy-ecient informationgathering in sensor networks, in: Proceedings of the 8thIEEE Symposium on Computers and Communication(ISCC’03), Kemer-Antalya, Turkey, June 2003.

[42] Y. Chen, C. Chuan, Q. Zhao, Sensor placement formaximizing lifetime per unit cost in wireless sensor networks,in: Proceedings of the IEEE Military CommunicationConference (MILCOM’05), Atlantic City, NJ, October 2005.

[43] P. Cheng, C.-N. Chuah, Xin Liu, Energy-aware nodeplacement in wireless sensor networks, in: Proceedings ofthe 47th IEEE Global Telecommunications Conference(Globecom’04), Dallas, TX, November 2004.

[44] X. Zhang, S.B. Wicker, How to distribute sensors in arandom field? in: Proceedings of the 3rd InternationalSymposium on Information Processing in Sensor Networks(IPSN ’04), Berkeley, CA, April 2004.

[45] D. Ganesan, R. Cristescu, B. Beferull-Lozano, Power-efficientsensor placement and transmission structure for data gather-ing under distortion constraints, in: Proceedings of the 3rdInternational Symposium on Information Processing in Sen-sor Networks (IPSN ’04), Berkeley, CA, April 2004.

[46] Q. Wang, K. Xu, G. Takahara, H. Hassanein, Deploymentfor information oriented sensing coverage in wireless sensornetworks, in: Proceedings of the 49th IEEE Global Tele-communication Conference (Globecom’06), San Francisco,CA, November 2006.

[47] D. Cerpa, D. Estrin, ASCENT: adaptive self-configuringsensor networks topologies, in: Proceedings of the 21stInternational Annual Joint Conference of the IEEE Com-puter and Communications Societies ((INFOCOM’02), NewYork, NY, June 2002.

[48] B. Chen et al., Span: an energy-efficient coordinationalgorithm for topology maintenance in ad hoc wireless

networks, in: Proceedings of the ACM 7th Annual Interna-tional Conference on Mobile Computing and Networking(MobiCom’01), Rome, Italy, July 2001.

[49] Y. Xu, J. Heidemann, D. Estrin, Geography-informedenergy conservation for ad hoc routing, in: Proceedings ofthe ACM 7th Annual International Conference on MobileComputing and Networking (MobiCom’01), Rome, Italy,July 2001.

[50] Y.T. Hou, Y. Shi, H.D. Sherali, On energy provisioning andrelay node placement for wireless sensor networks, IEEETransactions on Wireless Communications 4 (5) (2005)2579–2590.

[51] K. Xu, Q. Wang, H.S. Hassanein, G. Takahara, Optimalwireless sensor networks (wsns) deployment: minimum costwith lifetime constraint, in: Proceedings of the IEEEInternational Conference on Wireless and Mobile Comput-ing, Networking and Communications (WiMob’05), Mon-treal, Canada, August 2005.

[52] Q. Wang, G. Takahara, H. Hassanein, K. Xu, On relay nodeplacement and locally optimal traffic allocation in heteroge-neous wireless sensor networks, in: Proceedings of the IEEEConference on Local Computer Networks (LCN’05), Syd-ney, Australia, November 2005.

[53] Q. Wang, K. Xu, H. Hassanein, G. Takahara, Minimumcost guaranteed lifetime design for heterogeneous wirelesssensor networks (WSNs), in: Proceedings of the 24th IEEEInternational Performance, Computing, and Communi-cations Conference (IPCCC’05), Phoenix, Arizona, April2005.

[54] Q. Wang, K. Xu, G. Takahara, H. Hassanein, Locallyoptimal relay node placement in heterogeneous wirelesssensor networks, in: Proceedings of the 48th Annual IEEEGlobal Telecommunications Conference (Globecom’05), St.Louis, Missouri, November 2005.

[55] B. Hao, H. Tang, G. Xue, Fault-tolerant relay nodeplacement in wireless sensor networks: formulation andapproximation, in Proceeding of the Workshop on HighPerformance Switching and Routing (HPSR’04), Phoenix,Arizona, April 2004.

[56] E.L. Lloyd, G. Xue, Relay node placement in wireless sensornetworks, IEEE Transactions on Computers 56 (1) (2007)134–138.

[57] X. Han, X. Cao, E.L. Lloyd, C.-C. Shen, Fault-tolerant relaynodes placement in heterogeneous wireless sensor networks,in: Proceeding of the 26th IEEE/ACM Joint Conference onComputers and Communications (INFOCOM’07), Anchor-age AK, May 2007.

[58] A. Abbasi, M. Younis, A survey on clustering algorithms forwireless sensor networks, Journal of Computer Communi-cations, Special Issue on Network Coverage and RoutingSchemes for Wireless Sensor Networks, in press.

[59] A. Bogdanov E. Maneva, S. Riesenfeld, Power-aware basestation positioning for sensor networks, in: Proceedings ofthe 23rd International Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM’04),Hong Kong, March 2004.

[60] E.I. Oyman, C. Ersoy, Multiple sink network design problemin large scale wireless sensor networks, in: Proceedings of theIEEE International Conference on Communications(ICC’04), Paris, June 2004.

[61] W. Youssef, M. Younis, Intelligent Gateways Placement forReduced Data Latency in Wireless Sensor Networks, in:


Proceedings of the IEEE International Conference onCommunications (ICC’07), Glasgow, Scotland, June 2007.

[62] K. Akkaya, M. Younis, COLA: A coverage and latencyaware actor placement for wireless sensor and actor net-works, in: Proceedings of IEEE Vehicular TechnologyConference (VTC-Fall’06), Montreal, Canada, September2006.

[63] K. Akkaya, M. Younis, Coverage and latency aware actorplacement for wireless sensor and actor networks, Interna-tional Journal of Sensor Networks, in press.

[64] Shi, Y.T. Hou, A. Efrat, Algorithm design for base stationplacement problems in sensor networks, in: Proceedings ofIEEE International Conference on Quality of Service inHeterogeneous Wired/Wireless Networks, Waterloo,Ontario, Canada, August 2006.

[65] S.R. Gandham, M. Dawande, R. Prakash, S. Venkatesan,Energy efficient schemes for wireless sensor networks withmultiple mobile base stations, in: Proceedings of the IEEEGlobecom’03, San Francisco, CA, December 2003.

[66] J. Pan, et. al, Locating base-stations for video sensornetworks, in: Proceedings of the IEEE Vehicular TechnologyConference (VTC-Fall’03), Orlando, FL, October 2003.

[67] <http://www-robotics.usc.edu/~robomote/>.[68] <http://www.symbio.jst.go.jp/symbio/morph.html>.[69] G. Wang, G. Cao, T. La Porta, Movement-assisted sensor

deployment, in: Proceedings of the 23rd InternationalAnnual Joint Conference of the IEEE Computer andCommunications Societies (INFOCOM’04), Hong Kong,March 2004.

[70] A. Kansal, E. Yuen, W.J. Kaiser, G.J. Pottie, M.B.Srivastava., Sensing uncertainty reduction using low com-plexity actuation, in: Proceedings of Information Processingin Sensor Networks (IPSN 2004), Berkeley, CA, April 2004.

[71] A. Kansal et al., Controlled mobility for sustainable wirelesssensor networks, in: Proceedings of IEEE Sensor and AdHoc Communications and Networks (SECON’04), SantaClara, CA, October 2004.

[72] G. Wang, G. Cao, T. La Porta. Proxy-based sensordeployment for mobile sensor networks, in: Proceedings ofthe 1st IEEE International Conference on Mobile Ad-hocand Sensor Systems (MASS’04), Fort Lauderdale, Florida,October 2004.

[73] N. Heo, P.K. Varshney, Energy-efficient deployment ofintelligent mobile sensor networks, IEEE Transactions onSystems, Man, Cybernetics, Part A 35 (1) (2005) 78–92.

[74] W. Heinzelman, A. Chandrakasan, H. Balakrishnan,Energy-efficient communication protocol for wireless sensornetworks, in: Proceeding of the Hawaii International Con-ference System Sciences (HICSS’00), Hawaii, January 2000.

[75] M. Younis, M. Bangad, K. Akkaya, Base-station reposi-tioning for optimized performance of sensor networks, in:Proceedings of the IEEE Vehicular Technology Conference(VTC-Fall’03), Orlando, Florida, October 2003.

[76] K. Akkaya, M. Younis, Relocation of gateway for enhancedtimeliness in wireless sensor networks, in: Proceedings of theIEEE Workshop on Energy-Efficient Wireless Communica-tions and Networks (EWCN 2004), in Conjunction with the23rd IEEE International Performance Computing andCommunications Conference (IPCCC’04), Phoenix, Ari-zona, April 2004.

[77] W. Youssef, M. Younis, K. Akkaya, An intelligent safety-aware gateway relocation scheme for wireless sensor net-

works, in: Proceedings of the IEEE International Conferenceon Communications (ICC’06), Istanbul, Turkey, June 2006.

[78] W. Youssef, M. Younis, M. Eltoweissy, Intelligent discoveryof safe paths in wireless sensor network, in: Proceedings ofthe 23rd IEEE Biennial Symposium on Communication(QBSC’06), Queen’s University Kingston, Ontario, Canada,2006.

[79] C.-C. Shen, O.Koc, C. Jaikaeo, Z. Huang, Trajectory controlof mobile access points in MANET, in: Proceedings of the48th Annual IEEE Global Telecommunications Conference(GLOBECOM ’05), St. Louis, MO, November 2005.

[80] M. Ahmed et al., Positioning range extension gateways inmobile ad hoc wireless networks to improve connectivity andthroughput, in: Proceedings of IEEE Military Communica-tions Conference (MILCOM‘01), Washington, DC, October2001.

[81] J. English, M. Wiacek, M. Younis, CORE: Coordinatedrelocation of sink nodes in wireless sensor networks, in:Proceedings of the 23rd IEEE Biennial Symposium onCommunications (QBSC’06), Kingston, Ontario, Canada,May 2006.

[82] P. Basu, J. Redi, Movement control algorithms for realiza-tion of fault-tolerant ad hoc robot networks, IEEE Networks(2004) 2–10, July/August.

[83] I.F. Akyildiz, T. Melodia, K.R. Chowdhury, A survey onwireless multimedia sensor networks, Computer Networks51 (4) (2007) 921–960.

[84] S.N. Alam, Z. Haas, Coverage and connectivity in three-dimensional networks, in: Proceedings of the 12th AnnualACM International Conference on Mobile Computing andNetworking (MobiCom’06), Los Angeles, CA, September2006.

[85] <http://en.wikipedia.org/wiki/Truncated_octahedron>.[86] V. Ravelomanana, Extremal properties of three-dimensional

sensor networks with applications, IEEE Transactions onMobile Computing 3 (3) (2004) 246–257, July/September.

Mohamed F. Younis received B.S. degreein computer science and M.S. in engi-neering mathematics from AlexandriaUniversity in Egypt in 1987 and 1992,respectively. In 1996, he received hisPh.D. in computer science from NewJersey Institute of Technology. He iscurrently an associate professor in thedepartment of computer science andelectrical engineering at the university ofMaryland Baltimore County (UMBC).

Before joining UMBC, he was with the Advanced SystemsTechnology Group, an Aerospace Electronic Systems R&D
organization of Honeywell International Inc. While at Honeywellhe led multiple projects for building integrated fault tolerantavionics, in which a novel architecture and an operating systemwere developed. This new technology has been incorporated byHoneywell in multiple products and has received worldwiderecognition by both the research and the engineering communi-ties. He also participated in the development the RedundancyManagement System, which is a key component of the Vehicleand Mission Computer for NASA’s X-33 space launch vehicle.His technical interest includes network architectures and proto-cols, embedded systems, fault tolerant computing and distributed
http://www-robotics.usc.edu/~robomote/

http://www.symbio.jst.go.jp/symbio/morph.html

http://en.wikipedia.org/wiki/Truncated_octahedron


real-time systems. He has four granted and three pending patents.He served on multiple technical committees and published over85 technical papers in referred conferences and journals.

Kemal Akkaya received his BS degree inComputer Science from Bilkent Univer-sity, Ankara, Turkey in 1997, MS degreein Computer Engineering from MiddleEast Technical University (METU),Ankara, Turkey in 1999 and PhD inComputer Science from University ofMaryland Baltimore County (UMBC),Baltimore, MD in 2005. He worked as aSoftware Developer in A World BankProject in Ankara, Turkey in 2000. He iscurrently an assistant professor in the

Department of Computer Science at Southern Illinois UniversityCarbondale, IL. His research interests include energy-awareprotocols for wireless sensor networks, security and quality ofservice issues in ad hoc wireless and sensor networks.

strategies and techniques for node placement in wireless

Documents