DTIC ADA446898: Achieving Long-Term Surveillance in VigilNet

0.56 MB·English

by  Defense Technical Information Center

Achieving Long-Term Surveillance in VigilNet
† ‡
Tian He , Pascal Vicaire, Ting Yan, Qing Cao , Gang Zhou, Lin Gu,
‡ ‡
Liqian Luo , Radu Stoleru, John A. Stankovic, Tarek F. Abdelzaher
Department of Computer Science
University of Virginia, Charlottesville 22903
Abstract—Energy efficiency is a fundamental issue for out- datacaching[21],topologymanagement[22],clustering[23],
door sensor network systems. This paper presents the design placement [24] [25] to energy-aware applications [26][27].
and implementation of multi-dimensional power management
Insteadoffocusingonasingleprotocol,ouranswertoenergy
strategies in VigilNet, a major recent effort to support long-
efficiency is an integrated multi-dimensional power manage-
term surveillance using power-constrained sensor devices. We
integrate a novel tripwire service with an effective sentry and ment system. Our contributions, presented in this paper, are
duty cycle scheduling in order to increase the system lifetime, identified in the following aspects: 1) Our design is vali-
collaboratively. Through extensive system implementation, we dated through an extensive system implementation: VigilNet
demonstrate the feasibility to achieve high surveillance perfor-
– a large-scale sensor network system delivered to military
mance and energy efficiency, simultaneously. We invest a fair
agencies. 2) VigilNet takes a systematic approach, and the
amount of effort to evaluate our architecture with a network
of 200 XSM motes in an outdoor environment, an extensive energyefficiencyisnotnarrowlyaccountedforwithinasingle
simulationwith10,000nodes,aswellasananalyticalprobabilistic protocol. We propose a novel tripwire service, integrated with
model. These evaluations demonstrate the effectiveness of our an effective sentry and duty cycle scheduling to increase the
integrated approach and identify many interesting lessons and
system lifetime, collaboratively. 3) Tradeoffs are investigated
guidelines, useful for the future development of energy-efficient
tomeetrequirementsofbothsurveillanceperformanceandthe
sensor systems.
network lifetime. We present a complete system with 40,000
I. INTRODUCTION linesofcode,runningonmotes,thatachievesperformanceand
VigilNet is a recent major effort to support long-term energy efficiency simultaneously. 4) We devote considerable
militarysurveillance,usinglarge-scalemicro-sensornetworks. effort to evaluate the system with 200 XSM motes in an
Besides requirements of accurate target tracking and classifi- outdoor environment and an extensive simulation of 10,000
cation[1],oneofthekeydesigngoalsofVigilNetistoachieve nodes,inordertoidentifyasetofusefullessonsandguidelines
long-term surveillance in a realistic mission deployment. Due for future research.
to the small form factor and low-cost requirements, sensor The remainder of the paper is organized as follows: Sec-
devices such as the XSM motes [2] are normally equipped tion II categorizes power management features for different
withlimitedpowersources(e.g.,twoAAbatteries).Moreover, applicationscenarios.SectionIIIdescribesthepowermanage-
because of the hostile environment and a large number of ment requirements in VigilNet. Section IV introduces three
nodesdeployed,currentlyitisnotoperationallyandeconomi- power management strategies utilized in VigilNet namely,
cally feasible to replace the power source without introducing sentryservice,tripwireserviceanddutycyclescheduling.Sec-
enormouseffortandelementsofrisktothemilitarypersonnel. tionVdescribestheintegratedpowermanagementarchitecture
In addition, the static nature of the nodes in the field prevents in VigilNet. Section VI briefly discusses some additional
the scavenging of the power from ambient motion or vibra- energy efficient techniques applied in VigilNet. Section VII
tion [3][4]. The small form factor and possible lack of the addresses the tradeoff between energy efficiency and network
line of sight (e.g., deployment in the forest) make it difficult performance. Section VIII details the system implementation.
toharvestsolarpower.Ontheotherhand,a3∼6monthsystem Section IX provides the evaluation of a network of 200 XSM
life span is essential to guarantee the effectiveness of normal motes as well as an extensive hybrid simulation with 10,000
militaryoperations,whichnecessitatesa12∼24foldextension nodes. Finally, Section X concludes the paper.
ofthenormallifetimeofactivesensornodes.Consequently,it
II. BACKGROUND
is critical to investigate practical approaches of spending the
power budget effectively. Power management is by no means a stand-alone research
Many solutions have been proposed for energy effi- issue.Itcanbedramaticallyaffectedbytheunderlyingsystem
ciency at various levels of the system architecture, rang- configuration and by the application requirements. These in-
ing from the hardware design [5][2], coverage [6][7][8][9] cludetheform factor[28], hardware capability[5], possibility
MAC [10][11][12], routing [13][14][15], data dissemina- of energy scavenging [4][29], network/sensing topology and
tion [16], data gathering [17][18], data aggregation [19][20], density [6], link quality [30], event patterns, node mobility,
availability and accuracy of time synchronization [31], real-
†[email protected]
‡QingCao,LiqianLuoandTarekAbdelzaherarenowwithUniversityof time requirements and the nature of the applications [26]. At
Illinois,Urbana-Champaign the hardware level, multi-level sleep modes in the low power
Report Documentation Page Form Approved
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.
1. REPORT DATE 3. DATES COVERED
2006 2. REPORT TYPE 00-00-2006 to 00-00-2006
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Achieving Long-Term Surveillance in VigilNet
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
University of Virginia,Department of Computer Science,151 Engineer’s REPORT NUMBER
Way,Cahrlottesville,VA,22094-4740
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
The original document contains color images.
14. ABSTRACT
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON

a. REPORT b. ABSTRACT c. THIS PAGE 12
unclassified unclassified unclassified
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
microcontroller[5]enablesoftwaretocontroltherateofpower B. Power Management in Surveillance System
dissipation.Fine-grainedpowercontrol[2]allowsapplications
On the other hand, operations in surveillance sys-
to activate hardware modules incrementally. Radio wakeup
tems [35][36][37] [38], such as VigilNet, are event-driven in
circuits [32] achieve passive vigilance with a minimal power
nature. In surveillance systems, we are more interested in the
draw. Energy scavenging [3] is also possible for some ap-
data profile between inception and conclusion of the transient
plication scenarios, where ambient energy can be harvested.
events. These systems should remain dormant in the absence
Sensing coverage schemes [6][7] exploit redundancy in the
oftheeventsofinterests,andswitchtoanactivestatetoobtain
node deployment to activate only a subset of nodes. The
high fidelity in detection. Normally, the surveillance systems
coordinatedschedulingofthesensordutycycle[33]increases
improvethesystemlifetimethroughthefollowingapproaches:
the probability of detection and reduces the detection delay
• Coverage control: Surveillance systems are normally
withaminimalpowerconsumption.Communicationprotocols
deployed with a high density (For instance, the default
turnofftheradiowhenanodeisnottheintendedreceiver[12].
configuration of VigilNet [38] has 28 nodes per nominal
Though many individual solutions are proposed, few real
radiorange(30m))forthesakeofrobustnessindetection
systems actually achieve power efficiency comprehensively,
andfine-grainedsensingduringtracking.Wecanincrease
which makes the integrated approach in VigilNet novel and
the system lifetime by activating only a subset of nodes
practically useful. Considering the diversity of the different
at a given point of time, waiting for potential targets.
approaches, we categorize power management strategies in
• Duty cycle scheduling: The duration of transient events
the context of two types of systems: sampling systems and
withintheareaofsurveillanceisnormallynon-negligible.
surveillance systems.
Bycoordinatingnodes’sleepschedules,wecanconserve
energy without noticeably reducing the chance of de-
tection. Duty cycle scheduling is different from sample
A. Power Management in Sampling Systems
scheduling in the sense that duty cycle scheduling is
at the micro-scale (milliseconds vs. minutes) and it is
Great Duck Island [26] and Structural Monitoring [27] are
strongly affected by the dynamics of the events (e.g.,
typical sampling systems, which are deployed as distributed
target velocity).
large-scale data acquisition instruments. Power management
• Incremental activation: The sampling systems are nor-
strategiesinthesesystemsnormallymakeuseofthefollowing
mallydesignedfordatalogging.Ateachsampleinstance,
techniques:
all sensors should be activated to obtain a complete data
• Predefined sampling schedules: Most environmental profile. In contrast, surveillance systems are designed
phenomena, such as temperature, exist ubiquitously over to detect transient events of interest. It is sufficient to
space and continuously over time. The static nature of activate only a subset of sensors for the initial detection.
these phenomena makes it sufficient to construct the After the initial detection, we can activate other sensors
data profile by sampling the environment within discrete to achieve a higher sensing fidelity and to perform
time and space. Nodes can conserve energy by turning classification.
themselves off, according to a predefined schedule.
III. POWER MANAGEMENT REQUIREMENTS IN VIGILNET
• Synchronized and coordinated operations: Once the
sampling interval is defined a priori, nodes can commu- Ourpowermanagementstrategiesaremotivatedbyatypical
nicateinasynchronizedfashion.Withaprecisetimesyn- military surveillance application. The mission objective of
chronization[31],areceivercanturnontheradiomodule such a system is to conduct remote, persistent, clandestine
right before the message payload arrives. Consequently, surveillance to a certain geographic region to acquire and
wecanavoidlow-powerlisteningoverradio[10]duringa verify enemy capabilities and transmit summarized intelli-
non-active period. In addition, with the knowledge about gence worldwide in a near-real time manner. Several system
the sending rate of individual nodes, we are able to requirements affect our power management design within
estimate the link quality without control messages [34]. VigilNet:
• Data aggregation and compression: Since channel me- • Continuous surveillance: Due to the dynamic/transient
dia access is costly, especially when the receiver is in nature of the event, VigilNet is required to provide
a deep-sleep state [10], it is beneficial to send out one continuous surveillance. This requirement significantly
aggregate containing multiple sensing readings [19][20]. affects the overall architecture of power management
In addition, due to the value locality of the sensed data, strategiesandthedegreeofenergyconservationVigilNet
we can compress the total number of bits to be sent over can achieve.
the air. Since both aggregation and compression need • Real-time: As a real-time online system for target track-
to buffer a relatively large number of readings, which ing, VigilNet is required to cope with fast changing
introduces a certain delay, they are not quite suitable for events in a responsive manner. The delays introduced
time-critical surveillance systems. However, they match by the power management directly affect the maximum
most sampling systems very well. target speed our VigilNet can track. It is an essential
(cid:0) (cid:2) (cid:4) (cid:6)(cid:7)(cid:8) (cid:10) (cid:12)(cid:14) (cid:30) (cid:7)(cid:8) (cid:7)(cid:31) (cid:4) (cid:31) ! (cid:2) # (cid:18) (cid:0) (cid:2) (cid:4)(cid:6)(cid:7)(cid:8) (cid:10) (cid:12) (cid:14) (cid:16) (cid:7)(cid:18)(cid:6)(cid:20) (cid:8) (cid:21)(cid:23)
designtradeofftobalancebetweennetworklongevityand
responsiveness.
$%(’ $%(’ $%(’ $%(’
• Rare and critical event detection: Due to the nature of ) ) ) )
military surveillance, VigilNet deals with the rare event
model.Inthismodel,thetotaldurationofeventsissmall,
compared to the overall system lifetime. On the other
hand, events are so critical that the power management
becomes a secondary consideration in the presence of
events. $%&’ $%(’ $%(’ $%(’
$ $ $ $
• Stealthiness:Deployedinhostileenvironments,itisvital
for VigilNet to have a very low profile. Miniaturization
(cid:24) (cid:20) (cid:27)(cid:6)(cid:7)(cid:6)(cid:7)(cid:2) (cid:8) (cid:12) (cid:14) (cid:16) (cid:7)(cid:18)(cid:6)(cid:20) (cid:8) (cid:21)(cid:23)
(cid:24) (cid:20) (cid:27)(cid:6)(cid:7)(cid:6)(cid:7)(cid:2) (cid:8) (cid:12) (cid:14) (cid:30) (cid:7)(cid:8) (cid:7)(cid:31) (cid:4) (cid:31) ! (cid:2) #(cid:18)
makes nodes hard to detect, physically; however, radio
* , - , . , / ,
messages can be easily intercepted if nodes frequently
Fig.1. FourDifferentPartitionMethods
communicate.Powermanagementprotocolsdesignedfor
VigilNet should maintain silence during surveillance in
the absence of significant events. 1) Tripwirepartition: VigilNetimplementsitstripwirepar-
• Flexibility: We envision the deployment of VigilNet tition policy based on the Voronoi diagram. A network with
underdifferentdensities,topologies,sensingandcommu- n bases is partitioned into n tripwire sections such that each
nication capabilities. Therefore, it is essential to design tripwire section contains exactlyone basei and every node in
a power management architecture that is flexible enough thattripwiresectionisclosertoitsbaseithantoanyotherbase
to accommodate various system scenarios. inside the sensor field. Every node in the network uniquely
belongs to one and only one tripwire section. The rational
IV. KEY POWER MANAGEMENTSTRATEGIES IN VIGILNET behind Voronoi partition is to reduce the energy consumption
In order to achieve long-term surveillance that meets the and the end-to-end delay in data delivery.
militaryrequirement(e.g.,3∼6months),anaggressive12∼24 The positionsofbasesdirectlydetermine the layoutoftrip-
fold life-time extension is essential. Our initial investiga- wire sections and affect the routing path length for individual
tion [38] indicates that a single power management strategy nodes. The optimal base placement method to minimize the
is neither sufficient nor flexible. Therefore we restructure averagepathlengthtothenearestbasecanbefoundat[40].In
our prototype system described in [38] by adding a new practice, the base placement strategy is normally determined
combination of tripwire service and duty cycle scheduling. by the mission plan and topology.
Webelievethisistherightdirectiontopursue.Inthissection, 2) Tripwire partition mechanism: This section describes
we detail three main strategies, namely the tripwire service, the mechanism to enforce the tripwire partition policy. At the
sentryserviceanddutycyclescheduling,beforepresentingan beginningofthetripwire partitionoperation,eachbasebroad-
overarchingarchitectureinthenextsection.Inordertosupport casts one initialization beacon, to its neighbors with a hop-
these strategies, all nodes within VigilNet find their positions count parameter initialized to one. Link Symmetry Detection
with an accuracy of 1∼2 meters and they synchronize with [41] is used to ensure beacons can only be received through
each other within 1∼10 milliseconds using the techniques high quality symmetric links. Each receiving node maintains
described in [39] and [31], respectively. Long-range commu- the minimum hop-count value of all beacons it received from
nication devices are deployed as bases to relay sensor reports thenearestbase,intermsofthephysicaldistance,andignores
outside of the sensor field. beaconswithhigherhop-countvaluesandthosebeaconsfrom
other bases. Beacons are flooded outward with hop-count
A. Tripwire Services
values incremented at every intermediate hop. Through this
Thissectionproposesanovelnetwork-widepowermanage- mechanism, all nodes in the network get the shortest high
mentstrategycalledTripwireService.This servicedividesthe qualitypath,inhops,tothe nearestbase,in physicaldistance.
sensorfieldintomultiplesections,calledtripwiresections,and While the above mechanism is intuitive, the design deserves
appliesdifferentworkingschedulestoeachtripwiresection.A some further clarification. First, the boundaries betweenparti-
tripwiresectioncanbeeitherinanactiveoradormantstate,at tions are well delimited if we partition the network according
a given point of time. When a tripwire section is dormant, all to the physical distance between sensor nodes and bases
nodeswithinthissectionareputintoadeep-sleepstatetosave (Figure1Aand 1C).Ifthecommunicationhopisusedinstead,
energy.Surveillanceinactivetripwiresectionscanbedoneby the radio irregularity and the interference cause partitions to
either turning all nodes on or applying coverage algorithms interleave with each other (Figure 1B and 1D). This brings
such as the sentry service discussed later in Section IV-B. complexity and uncertainty to the design of optimal tripwire
The rationale behind the tripwire service is the existence of placementstrategies.Second,itisbeneficialtousehopcounts
roads in the area of interest. By deploying the tripwire along to build diffusion trees within each partition, because 1) the
theroad,wecanguaranteethe detectionwithoutactivatingall normal geographic-based routing does not guarantee high-
sensors in the area. quality shortest path to the root. 2) Due to the existence of
high-quality long links, a smaller number of nodes become 100%
90%
agcetoivgerapbhaicck-bboanseednrooduetisngin. Fthinealhlyo,pt-hbiassdedesirgonutpinrogvitdheasncienrttahine obability 678000%%%
Pr 50%
robustness to the base failure. If a base fails, the sensor field on 40% SensingRanger=2m
can be easily repartitioned without this base. Detecti 2300%% SSSeeennnsssiiinnngggRRRaaannngggeeerrr===812m40mm
3) Tripwire scheduling: A tripwire section can be either in 10%
0%
an active or a dormant state. We configure the state of each 1.00E-04 1.00E-03 Density(#/m*m) 1.00E-02
tripwire section by setting a 16 bits schedule at the corre-
Fig.2. DetectionProb.Vs.SentryDensity
spondingbase.Eachbitinthescheduledenotesthestateofthis
tripwiresectionineachround(rotation)upto16rounds.After
16 rounds, the pattern is repeated. With this design, we can Since, theoretically, there is at most one sentry within each
assign 65536 different schedules to each tripwire and assign ROVrange,accordingtothecirclecoveringtheorem[42],the
65536N (N is the numberoftripwires.)different schedulesto sentry density is upper bounded by √ 2π . Given the area
the network. The schedule can be predetermined or randomly size, sensing range and sentry density27,RwOeV2get the detection
generated.RandomschedulingisdonebysettingtheTripwire probability (Figures 2) according to the derived model. For a
Duty Cycle (TDC), which is the percentage of active rounds typical deployment with 1000 nodes in 100×1000 m2 area,
in the schedule. Figure 2 indicates how to choose the right combination of
systemconfigurations.Forexample,inordertoachievea99%
B. Sentry Services
detectionprobability, we canchoose either a sentrydensity of
Inordertoexploitthehighnodedensitywithinthesections, 0.008 nodes/m2 (ROV= 6 meters) with 8 meter sensing range
we design and implement a section-wide power management or a lower density of 0.004 nodes/m2 (ROV=8.5 meters) with
strategy,called sentryservice.The mainpurpose ofthe sentry
14meterssensingrange.WenotethatwhenROVis settothe
service is to select a subset of nodes, which we define as
sensingrange,wecanguarantee100%detection,assumingno
sentries, in charge of surveillance. Sentry selection contains
voids.
two phases. Nodes first exchange neighboring information
2) HowtoenforceROV: AfterwechooseaROVvalue,we
through hello messages. In each hello message, a sender
need to enforce it during the sentry selection phase. Since the
attaches its node ID, position, number of neighbors and its
sensingrangeisnormallysmallerthantheradiorange,directly
ownenergyreadings.Afterthefirstphase,eachnodebuildsup
usingtheradiorangeastheROVcannotguaranteeaneffective
a one-hop neighbor table. In the secondphase, each node sets
coverage of the area. For example, the HMC1002 dual-axis
a delay timer. The duration of the timer is calculated based
magnetometerusedbyMICA2hasonly30-feeteffectiverange
on the weighted Energy rank Renergy and weighted Cover for a moving car. If we use the Chipcon radio (>100 feet) to
rankR asshowninEquation1.TheenergyrankR
cover energy define the range of vicinity, less than 10% of area is sensing
is assigned according to energy readings among neighboring
covered. There are two approaches to address these issues.
nodes (e.g., the node with the highest energy reading within
The first approach is to reduce the radio sending power to
a neighborhood has a rank of 1. ) Similarly, the cover rank
emulate the ROV range. The power setting can be chosen
R is assigned according to the number of neighbors
cover in such a way that there is about one sentry within each
within a node’s sensing range. As for current implementation,
sensing range. The second approach is to discard declaration
we assign equal weights to both ranks.
messages from any sentry beyond the distance of ROV. The
first approach achieves sensing coverage, without the location
W ×R +W ×R
T = e energy c coverMaxDelay+Jitter information of the nodes [43], while the second approach
timer (W +W )×#Neighbors
e c provides a more predictable sentry distribution, because the
(1)
emulated ROV would be affected by the radio irregularity in
Afterthedelaytimerfiresinonenode,thisnodeannounces the environment. Consequently, we adopt the second solution
itself as sentry by sending out a declaration message. While inoursystem,giventhefactthatlocalization[39]issupported
other nodes, in the vicinity of the declaring node, cancel their in VigilNet.
timers and become dormant non-sentry nodes. The effective
C. Sentry Duty Cycle Scheduling
range, in physical distance, of a sentry’s declaration message
is defined as the Range of Vicinity (ROV). While the sentry The requirement for continuous sensing coverage in the
selection can be straightforwardly implemented, the challeng- sentryserviceimposesatheoreticalupperboundonthesystem
ing part is to choose and to enforce the appropriate range lifetime. This upper bound is decided by the total number of
of vicinity (ROV). This parameter directly affects the sentry nodes deployed. Since a target normally stays in the sensing
density, hence affects the lifetime of the network. area of a sentry node for a non-negligible period of time, it is
1) HowtochooseROV: TheappropriateROVvaluecanbe not necessary to turn sentry nodes on all the time. By using
chosen by the analytical intrusion detection model detailed in dutycyclescheduling,weareabletobreakthetheoreticalup-
Appendix. This model describes the relationship between the perboundimposedbythefullcoveragealgorithms[7].LetT
on
detectionprobability,thesensingrangeandthesentrydensity. be the active duration and T be the inactive duration, then
off
than the sensing range leads to a partial coverage and a lower
(cid:55)(cid:85)(cid:76)(cid:83)(cid:90)(cid:76)(cid:85)(cid:72)(cid:3)(cid:54)(cid:72)(cid:85)(cid:89)(cid:76)(cid:70)(cid:72) (cid:55)(cid:39)(cid:38)
ROV value than the sensing range leads to redundancy in the
(cid:54)(cid:70)(cid:75)(cid:72)(cid:71)(cid:88)(cid:79)(cid:76)(cid:81)(cid:74)
(cid:49)(cid:72)(cid:87)(cid:90)(cid:82)(cid:85)(cid:78)(cid:16)(cid:47)(cid:72)(cid:89)(cid:72)(cid:79) coverage. When ROV equals 0 meter, the sentry service is
(cid:58) actually disabled and all nodes with the section are awake,
(cid:68)
(cid:78)(cid:72) providing the highest degree of coverage. At the third-level,
(cid:54)(cid:72)(cid:81)(cid:87)(cid:85)(cid:92)(cid:3)(cid:54)(cid:72)(cid:85)(cid:89)(cid:76)(cid:70)(cid:72) (cid:53)(cid:82)(cid:87)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81) (cid:53)(cid:50)(cid:57) (cid:88)(cid:83)(cid:3)(cid:86) duty cycle scheduling controls the energy-burning rate of
(cid:72)
AAccttiivvee (cid:85)(cid:89) individual sentry nodes by manipulating their wakeup/sleep
(cid:76)(cid:70)
(cid:72) schedule. The Sentry Duty Cycle (SDC) parameter is used to
(cid:54)(cid:72)(cid:70)(cid:87)(cid:76)(cid:82)(cid:81)(cid:16)(cid:47)(cid:72)(cid:89)(cid:72)(cid:79)
controltheawarenessofsentrynodes,whichisthepercentage
(cid:39)(cid:88)(cid:87)(cid:92)(cid:3)(cid:38)(cid:92)(cid:70)(cid:79)(cid:72)(cid:3)
(cid:54)(cid:70)(cid:75)(cid:72)(cid:71)(cid:88)(cid:79)(cid:76)(cid:81)(cid:74) (cid:58)(cid:68)(cid:78)(cid:72)(cid:88)(cid:83) (cid:54)(cid:39)(cid:38) of active time. The duty cycle scheduling can be disabled by
(cid:49)(cid:82)(cid:71)(cid:72)(cid:16)(cid:47)(cid:72)(cid:89)(cid:72)(cid:79) setting SDC to 100%. By adopting different values for TDC,
ROVandSDC,wecanflexiblyadjustourpowermanagement
Fig.3. IntegratedPMArchitecture
to accommodate different system scenarios.
theSentryTogglePeriod(STP)isdefinedas(T + T ),and
on off VI. OTHER ENERGY CONSERVATION TECHNIQUES
theSentryDutyCycle(SDC)isdefinedas Ton .Theoretically,
STP
the duty cyclescheduling canachieveunbounded energycon- Besides the three main power management strategies, sev-
servation by lowering the SDC value. The paramount concern eralothertechniqueshavebeenintegratedintovariousaspects
of this technique is that lowering the SDC value increases the of the VigilNet system. Similar techniques [20][44][27][10]
detection delay and reduces the detection probability. We can have been proposed in the literature and we provide this sec-
either effectively implement random duty cycle scheduling or tion for the completeness of the VigilNet power management
more sophisticated scheduling algorithms to coordinate node design and implementation.
activities to maximize performance. In [33], we demonstrate
• Minimum connected dominating tree: To ensure a
a local optimal scheduling coordination algorithm to reduce
swift delivery of messages, VigilNet requires an active
the detection delay and increase the detection probability.
diffusion tree over active tripwire section. Since the
We prove that, at relatively large SDC (e.g 5% <SDC), the
communication range is normally much larger than the
differencebetweenrandomschedulingandoptimalscheduling
sensingrange[5][2],itispossibletobuildadiffusiontree
can be practically ignored. Since the random scheduling does
ontopofsentrynodes.Toreducetheenergyspentduring
notneedcontrolmessagesforcoordination(morestealthy)and
idle listening, VigilNet desires a tree with the minimum
it is not affected by time drift, we choose random scheduling
connected dominating set (a tree with minimum non-
over the coordinated one in the system implementation.
leaf nodes). Since it is a NP-Complete problem to find
the minimum connected dominating set of a graph, we
V. INTEGRATED SOLUTION: TRIPWIRE-BASED POWER
adopt a localized approximation as follows: during the
MANAGEMENT WITH SENTRY SCHEDULING
building process, each node rebroadcasts the hop-count
To achieve an aggressive network lifetime extension, the beacon after a certain time delay. The delay in one node
VigilNet power management subsystem integrates the three is inversely proportional to the number of neighbors and
strategies mentioned in previous sections into a multi-level the energy remaining. By doing so, the node with more
architecture, as shown in Figure 3. At the top level, the neighbors and more energy left has a higher chance to
tripwire service controls the network-wide distribution of become the parent node within the diffusion tree.
power consumption among sections; the uniform discharge • Data aggregation:Thechannelmediaaccessinwireless
of energy across sections is achieved through the scheduling sensor network is relatively expensive. For example, in
mechanismwediscussedinsectionIV-A.3.WeuseaTripwire the Chipcon radio implementation for MICA2, to deliver
Duty Cycle (TDC), which is the percentage of active time a default payload size of 29 bytes, the total overhead
for each tripwire section, to control the network-wide energy- is 17 bytes (37%!), including 8 bytes preamble, 2 bytes
burning rate. There are two special cases: when TDC equals synchronization, 5 bytes header and 2 bytes CRC. This
100%, the whole network becomes active and the tripwire motivates us to utilize various kinds of aggregation tech-
service is merely a network partition service. When TDC niques. The first technique we use is called Application-
equals 0%, the whole network is in dormant status and it Independent Aggregation, which concatenate data from
can only be awaken by external sources. At the second level, different modules into one aggregate, regardless of their
the sentry service controls the power distribution within each semantics. For example, system-wide parameters can be
section. The uniform discharge of energy in a section is sent with time synchronization messages. The second
achievedthroughautomaticrotationstrategiesaccordingtothe technique we use is called Application-Dependent Ag-
remaining power within individual nodes. We use the Range gregation. The tracking subsystem in VigilNet performs
of Vicinity (ROV) parameter to control the energy-burning the in-network aggregation by organizing the nodes into
rate of active sections. When ROV equals the sensing range groups. Instead of each node reporting its position sep-
of nodes, the section is fully covered. A higher ROV value arately, a leader node calculates the weighted center
(cid:36)(cid:83)(cid:83)(cid:79)(cid:76)(cid:70)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81)(cid:3)(cid:47)(cid:68)(cid:92)(cid:72)(cid:85) 100% Application
Application
(cid:40)(cid:81)(cid:89)(cid:76)(cid:85)(cid:82)(cid:55)(cid:85)(cid:68)(cid:70)(cid:78) (cid:38)(cid:79)(cid:68)(cid:86)(cid:86)(cid:76)(cid:73)(cid:76)(cid:70)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81) (cid:57)(cid:72)(cid:79)(cid:82)(cid:70)(cid:76)(cid:87)(cid:92)(cid:3) (cid:41)(cid:68)(cid:79)(cid:86)(cid:72)(cid:3)(cid:36)(cid:79)(cid:68)(cid:85)(cid:80)(cid:3) (cid:53)(cid:72)(cid:79)(cid:68)(cid:92) 90%
(cid:53)(cid:72)(cid:74)(cid:85)(cid:72)(cid:86)(cid:86)(cid:76)(cid:82)(cid:81) (cid:51)(cid:85)(cid:82)(cid:70)(cid:72)(cid:86)(cid:86)(cid:76)(cid:81)(cid:74)
(cid:48)(cid:76)(cid:71)(cid:71)(cid:79)(cid:72)(cid:90)(cid:68)(cid:85)(cid:72)(cid:3)(cid:47)(cid:68)(cid:92)(cid:72)(cid:85) 80% Middleware
(cid:39)(cid:88)(cid:87)(cid:92) (cid:47)(cid:82)(cid:70)(cid:68)(cid:79) 70% Middleware
(cid:42)(cid:85)(cid:82)(cid:88)(cid:83)(cid:3) (cid:39)(cid:92)(cid:81)(cid:68)(cid:80)(cid:76)(cid:70)(cid:3) (cid:38)(cid:92)(cid:70)(cid:79)(cid:72) (cid:54)(cid:72)(cid:81)(cid:87)(cid:85)(cid:92)(cid:3) (cid:55)(cid:85)(cid:76)(cid:83)(cid:90)(cid:76)(cid:85)(cid:72) (cid:53)(cid:72)(cid:83)(cid:82)(cid:85)(cid:87)
(cid:55)(cid:76)(cid:80)(cid:72)(cid:3) (cid:76)(cid:93)(cid:68)(cid:87)(cid:76)(cid:82)
(cid:54)(cid:92)(cid:81)(cid:70) (cid:48)(cid:74)(cid:80)(cid:87) (cid:38)(cid:82)(cid:81)(cid:73)(cid:76)(cid:74)(cid:3) (cid:54)(cid:70)(cid:75)(cid:72)(cid:71)(cid:88)(cid:79)(cid:76) (cid:54)(cid:72)(cid:85)(cid:89)(cid:76)(cid:70)(cid:72) (cid:48)(cid:81)(cid:74)(cid:87) (cid:81) (cid:40)(cid:81)(cid:74)(cid:76)(cid:81)(cid:72) 60%
(cid:81)(cid:74)
Networking
50%
(cid:49)(cid:72)(cid:87)(cid:90)(cid:82)(cid:85)(cid:78)(cid:3)(cid:47)(cid:68)(cid:92)(cid:72)(cid:85) (cid:54)(cid:72)(cid:81)(cid:86)(cid:76)(cid:81)(cid:74)(cid:3)(cid:47)(cid:68)(cid:92)(cid:72)(cid:85)
Networking
(cid:39)(cid:76)(cid:73)(cid:73)(cid:53)(cid:88)(cid:86)(cid:82)(cid:76)(cid:69)(cid:82)(cid:88)(cid:81)(cid:86)(cid:3)(cid:55)(cid:87)(cid:3)(cid:85)(cid:72)(cid:72) (cid:36)(cid:39)(cid:86)(cid:92)(cid:72)(cid:80)(cid:87)(cid:72)(cid:80)(cid:70)(cid:87)(cid:76)(cid:72)(cid:82)(cid:87)(cid:85)(cid:81)(cid:76)(cid:70)(cid:3) (cid:53)(cid:58)(cid:68)(cid:71)(cid:68)(cid:76)(cid:82)(cid:78)(cid:16)(cid:72)(cid:37)(cid:88)(cid:68)(cid:83)(cid:86)(cid:72)(cid:3) (cid:41)(cid:85)(cid:72)(cid:84)(cid:88)(cid:72)(cid:81)(cid:70)(cid:92)(cid:16)(cid:41)(cid:76)(cid:79)(cid:87)(cid:72)(cid:85)(cid:3) (cid:38)(cid:38)(cid:82)(cid:68)(cid:81)(cid:79)(cid:76)(cid:87)(cid:69)(cid:76)(cid:81)(cid:85)(cid:88)(cid:68)(cid:82)(cid:87)(cid:82)(cid:88)(cid:85)(cid:86)(cid:3)(cid:3)(cid:3) 40% Utilities Utilities
30% DataLink
(cid:39)(cid:68)(cid:87)(cid:68)(cid:3)(cid:47)(cid:76)(cid:81)(cid:78)(cid:3)(cid:47)(cid:68)(cid:92)(cid:72)(cid:85) 20% Classfication DataLink
(cid:48)(cid:36)(cid:38) (cid:44)(cid:81)(cid:87)(cid:72)(cid:85)(cid:73)(cid:72)(cid:85)(cid:72)(cid:81)(cid:70)(cid:72)(cid:3)(cid:68)(cid:89)(cid:82)(cid:76)(cid:71)(cid:68)(cid:81)(cid:70)(cid:72) (cid:54)(cid:72)(cid:81)(cid:86)(cid:82)(cid:85)(cid:3)(cid:39)(cid:85)(cid:76)(cid:89)(cid:72)(cid:85)(cid:86)
10% Classfication
Driver
0% Driver
(cid:48)(cid:44)(cid:38)(cid:36)(cid:21)(cid:3)(cid:18)(cid:59)(cid:54)(cid:48)(cid:3)(cid:18)(cid:59)(cid:54)(cid:48)(cid:21)(cid:3)(cid:18)(cid:3)(cid:48)(cid:44)(cid:38)(cid:36)(cid:21)(cid:39)(cid:50)(cid:55)(cid:3)(cid:48)(cid:82)(cid:87)(cid:72)(cid:86) (cid:39)(cid:76)(cid:86)(cid:83)(cid:79)(cid:68)(cid:92)(cid:3)(cid:68)(cid:87)(cid:3)(cid:38)(cid:9)(cid:38) Text Data
Fig.4. TheVigilNetSystemArchitecture Fig.5. MemoryLayout
of gravity from multiple inputs and reports only one events, such as temperature and humidity, are not directly
aggregate back to the base. correlated with the responsiveness of the system. While in
• Implicit acknowledgement: Given that the sensor pay- thesurveillancesystem,responsivenessandawarenessdirectly
load is very small, it might not be energy efficient to affect the system performance including tracking and classifi-
acknowledge every packet explicitly. Implicit acknowl- cation. The former can be measured in terms of the detection
edgement can be achieved through several approaches. probabilityanddelay,andthelatercanbemeasuredintermsof
They differ in functionality and overhead. B-MAC [10] thenumberofnodesdetectingexternalevents,simultaneously.
providesanefficientimplementationoftheCSMAproto- WehaveinvestigatedresponsivenessinprevioussectionsIV-B
col with radio-layer acknowledgement support. Observ- and IV-C. This section focuses on how to improve the system
ing that most of the packets need to be forwarded for awareness. In VigilNet, awareness is supported by the on-
routing, we alternatively implemented the acknowledge- demand wakeup service. The on-demand control is stealth-
ment as a special field in outgoing packets. When there ier compared to the periodic control [38], because wakeup
are no outgoing packets for a period of time, a special beacons are sent only when events occur. To support the on-
acknowledgement packet is sent. demand control, we need to guarantee the delivery of wakeup
• Incremental detection: Multi-sensingmodalities are de- beacons. Because of the special stealthiness requirement, the
sired for achieving target classification. However, it is non-sentriescannotsynchronizetheirclockswiththeirsentries
not necessary to activate all sensors only for detection. by exchanging messages. Therefore, neighboring non-sentry
Among the three types of sensors in XSM motes, the motesmaynolongerhaveasleep-wakeupcyclesynchronized
optic TR230 PIR sensor has the longest detection range witheachotherduetotheclockdrift,andasentrycannotkeep
and a relatively low power consumption,i.e., 0.88mW. trackwhichofits neighbors are awake.Toguarantee delivery,
We use this sensor to support the initial detection and a non-sentry periodically wakes up and checks radio activity
to incrementally wakeup other sensors for classification (detectspreamble bytes)oncepercheckingperiod(e.g.,every
purposes. second). If no radio activity is detected, this node goes back
• Passive wakeup circuitry : Several efforts [2][45][32] to sleep, otherwise it remains active for a period of time,
have been made to support low-power passive wakeup preparingforincomingtargets.Ifasentrynodewantstowake
by using an acoustic detector [45], infrared sensor [2] up all neighboring nodes, it only needs to send out a message
or radio [32]. Currently, the hardware-event-driven de- withalongpreamblewithalengthequaltoorlongerthanthe
sign[2]ofXSMmotesisnotmatureenoughforVigilNet checking period of non-sentry nodes. Since in the rare event
to exploit this aspect. However, this is a very promising model, the wakeup operations are done very infrequently, the
direction. long preamble doesn’t introduce much energy consumption
in sentry nodes. On the other hand, since the amount of time
VII. TRADEOFF: PERFORMANCE VS.ENERGY EFFICIENCY takentochecktheradioactivityisconstantforaspecificradio
One key research challenge for VigilNet is to reconcile the hardware,thelengthofcheckingperioddeterminestheenergy
needfornetworklongevitywiththeneedforfastandaccurate consumption in non-sentry nodes. In general, a long checking
target detection and classification. The former requires most period leads to a lower energy consumption. However, to
sensor nodes to remain inactive, while the later desires many ensure that a sentry node wakes up neighboring non-sentry
activesensornodes.Aswementionedbefore,theeventmodel nodes before a target moves out of their sensing range, the
directly affects the design of the power management. Energy checking period can not be arbitrarily long. Theoretically, the
√
efficiency can be comparatively easy to achieve if events of upperbound of checking period is R2−r2, where R is radio
S
interests are ubiquitously present. The data quality of some
Fig.6. LocationofDeploymentandADeployedXSMMote Fig.7. TripwirePartition
range, r is sensing range of sentries and S is the speed of serviceformthebasisforpowermanagementsubsystem.Their
target. Due to the other delays, such as sensor warm-up time, functionalities are supported by other services. For instances,
the checking period should be smaller than this theoretical the localization service provides the basis for the tripwire
bound.Inourimplementation,non-sentrynodeshave1%duty partition and sentry selection. The group management service
cycle with 1 second checking period. allows power-efficient data aggregation. The configuration
service facilitates the online tuning of the power management
VIII. IMPLEMENTATION parameters. Multilevel sleep modes in the ATmega128 permit
a high-granularity control of power dissipation. Selectable
ThepowermanagementarchitecturedescribedinSectionV
transmission power settings (255 levels) in CC1000 enable
hasbeenintegratedintotheVigilNetsystem.Wehavesuccess-
us to adjustthe effective range of sentry declaration messages
fully transferred VigilNet to a military agencyfor deployment
dynamically.
by the end of 2004. The overarching architecture of VigilNet
is shown in Figure 4. The components in gray are specially
IX. SYSTEM EVALUATION
designed for the power management purpose. Other compo-
nents provide extra energy-aware features, as mentioned in This section presents experimental results that evaluate
section VI. the performance of the power management subsystem. The
VigilNet is built on top of the TinyOS operating sys- experimental results in Section IX-A are obtained through
tem. TinyOS supports a lightweight event driven computation an actual deployment of 200 XSM motes, focusing on the
model with two-level scheduling. VigilNet is mostly written sentry selection, tripwire partition and tracking delays. Other
in NesC, a derivative language from C specially designed for experiments in Section IX-B, especially those related to the
embedded programming. This language enables the program- system lifetime, require a significant amount of time. Un-
merstodefinetheinterfaces,functionsofcomponentsandthe fortunately, we currently can not afford to deploy such a
relations (dependencies) among them. The size of VigilNet is large system unattended for a long time. We have to conduct
about 40,000 lines of code, supporting multiple existing mote thoseevaluationsthroughahybridapproach,whichusesbasic
platforms including MICA2 and XSM. The compiled image measurements from a smaller number of motes as input to
occupies 83,963 bytes of code memory and 3,586 bytes of a simulator. By doing so, we can investigate the impact of
data memory. The code and data memory maps are shown in different system configurations on the performance of power
Figure 5. management.
We categorize the system components into seven groups;
Data link and sensor driver layers use default components A. Field Evaluation
in TinyOS; Network layer consists of three major compo-
The field evaluation was done as part of a technical transi-
nents:robustnessdiffusiontree,asymmetriclinkdetection[41]
tion on December 2004, when we deployed 200 XSM motes
and radio-based wakeup service. The sensing layer provides
on a dirt T-shape road (200 meters by 300 meters). The XSM
detection and classification with continuous calibrator and
mote is designed by the joint efforts of Ohio State University
frequency filters [1]. We note that it is very critical to have a
[2]andCrossBowInc, whichfeaturesanAtmelATmega128L
sensingsubsystemwithminimalfalsealarmsinanoutdooren-
microcontroller and a Chipcon 433MHz CC1000 radio. Its
vironment.Otherwise,thenetworklifetimeisseverelyreduced
sensing suite includes magnetic, acoustic, photo, temperature
due to unnecessary wakeup operations. The application layer
and passive infrared sensors (PIR). Figure 6 displays the
focuses on tracking and high-level classification [46]. The
environment where our system was located and the picture of
middleware layer occupies most code (40%) and data mem-
oneoftheXSMmotes.Nodesarerandomlyplacedroughly10
ory (35%). Among all the middleware services, the tripwire
meters apart, covering one 300-meter road and one 200-meter
service, sentry selection, duty cycle scheduling and wakeup
road.
100% (cid:51)(cid:75)(cid:68)(cid:86)(cid:72)(cid:3)(cid:44)(cid:44) (cid:51)(cid:75)(cid:68)(cid:86)(cid:72)(cid:3)(cid:44)(cid:44)(cid:44) (cid:51)(cid:75)(cid:68)(cid:86)(cid:72)(cid:3)(cid:44)(cid:57) (cid:51)(cid:75)(cid:68)(cid:86)(cid:72)(cid:3)(cid:57)
of (cid:55)(cid:76)(cid:80)(cid:72)(cid:3)(cid:54)(cid:92)(cid:81)(cid:70) (cid:47)(cid:82)(cid:70)(cid:68)(cid:79)(cid:76)(cid:93)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81) (cid:36)(cid:86)(cid:92)(cid:80)(cid:80)(cid:72)(cid:87)(cid:85)(cid:92)(cid:3)(cid:39)(cid:72)(cid:87)(cid:72)(cid:70)(cid:87)(cid:76)(cid:82)(cid:81) (cid:37)(cid:68)(cid:70)(cid:78)(cid:69)(cid:82)(cid:81)(cid:72)(cid:3)(cid:38)(cid:85)(cid:72)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81)
mulativeFractionNodes 24680000%%%% SNeonnt-rSyentry (cid:54)(cid:87)(cid:68)(cid:85)(cid:87) (cid:54)(cid:92)(cid:86)(cid:87)(cid:72)(cid:80)(cid:51)(cid:3)(cid:44)(cid:75)(cid:81)(cid:68)(cid:76)(cid:86)(cid:87)(cid:72)(cid:76)(cid:68)(cid:3)(cid:44)(cid:79)(cid:76)(cid:93)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81) (cid:39)(cid:40)(cid:82)(cid:85)(cid:89)(cid:51)(cid:80)(cid:72)(cid:75)(cid:81)(cid:68)(cid:68)(cid:87)(cid:3)(cid:86)(cid:81)(cid:55)(cid:72)(cid:87)(cid:3)(cid:85)(cid:3)(cid:57)(cid:68)(cid:54)(cid:70)(cid:72)(cid:44)(cid:78)(cid:44)(cid:70)(cid:44)(cid:76)(cid:87)(cid:81)(cid:76)(cid:82)(cid:74)(cid:81) (cid:54)(cid:72)(cid:81)(cid:51)(cid:87)(cid:85)(cid:75)(cid:92)(cid:68)(cid:3)(cid:54)(cid:86)(cid:72)(cid:72)(cid:3)(cid:79)(cid:57)(cid:72)(cid:70)(cid:44)(cid:87)(cid:76)(cid:82)(cid:81)
u
C (cid:51)(cid:82)(cid:90)(cid:72)(cid:85)(cid:3)(cid:48)(cid:74)(cid:80)(cid:87)
0%
2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400
VoltageValue(mv) (cid:53)(cid:82)(cid:87)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81) (cid:58)(cid:68)(cid:78)(cid:72)(cid:88)(cid:83)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:54)(cid:72)(cid:85)(cid:89)(cid:76)(cid:70)(cid:72) (cid:51)(cid:75)(cid:68)(cid:86)(cid:72)(cid:3)(cid:57)(cid:44)(cid:44)
(cid:43)(cid:72)(cid:68)(cid:79)(cid:87)(cid:75)(cid:3)(cid:53)(cid:72)(cid:83)(cid:82)(cid:85)(cid:87)
(cid:51)(cid:82)(cid:90)(cid:72)(cid:85)(cid:3)(cid:48)(cid:74)(cid:80)(cid:87)
Fig.8. EffectivenessofSentrySelection
(cid:51)(cid:75)(cid:68)(cid:86)(cid:72)(cid:3)(cid:57)(cid:44)(cid:44)(cid:44)
(cid:40)(cid:89)(cid:72)(cid:81)(cid:87)(cid:3)(cid:55)(cid:85)(cid:68)(cid:70)(cid:78)(cid:76)(cid:81)(cid:74)
1
of (cid:55)(cid:85)(cid:76)(cid:83)(cid:90)(cid:76)(cid:85)(cid:72)(cid:3)(cid:54)(cid:72)(cid:70)(cid:87)(cid:76)(cid:82)(cid:81)
FractionmROV00..68 Fig.11. PhaseTransitionandRotation
mulativeMinimu00..24 TABLEI
u POWERCONSUMPTIONACCORDINGTOTHEMOTESTATE.
C
0
2 4 6 8 10 12 14 16 Nodestate RadioState Processor Sensors Total
ROV(meters) (Messages State State Power
persecond)
Fig.9. ROVEnforceResults Init receive(2) active off 49.449mW
SentrySleep off(0) sleep off 42µW
100% NonSentrySleep LPL(0) sleep off 450µW
mulativeFractionofDelays 24680000%%%% DCVCeeolaltmosecscmiittfiuyiocnnaEitcsDioatientmiloaDanyteiDolaneylDayelay AAAwwwaaakkkeeeSCCeoonmmsimmngSensing rrreeeccceeeiiivvveee(((202))) aaaccctttiiivvveee ooonnff 477910...4404519mmmWWW
u
C
0% the radio irregularity introduced by the ground effect in the
0 2 4 6 8 10 12 14
Delay(seconds) outdoor environments, a small percentage of sentry nodes can
not reach each other, even when they are very close (<5
Fig.10. DistributionofDifferentDelays
meters) to each other.
4) Delays under power management: In this experiment,
1) Effectiveness of the tripwire partition: One snapshot of weinvestigatevariousdelaysunderpowermanagement.When
thenetworklayoutcollectedbyourGUIisshowninFigure7. a target enters the surveillance area, a detection report is
We placed 200 XSM field motes and 3 mica2dot base motes issued first, followed by classification reports. Finally, after
in the field. Accordingly, the network is divided into three sufficient information is gathered, velocity reports are issued.
sections. The layout indicates that the Voronoi-based tripwire Figure 10 illustrates the cumulative distribution of different
partitioning is very effective and that all nodes attach to the delays. The communication delay (leftmost curve) is much
nearest base nodes through the shortest path. smallercompared with other delays. About 80% of detections
2) Effectiveness of the sentry selection: In this experiment, are done within2seconds.Over80%of the classificationand
we evaluate the effectiveness of sentry selection. Figure 8 velocityestimationsaremadewithin4seconds.Thisempirical
plots the cumulative distribution function of the voltages of result indicates that our power management does not degrade
nodes within the network. The left curve is the voltage CDF the tracking performance significantly.
of non-sentry nodes and the right curve is the voltage CDF
for sentry nodes. It confirms that our sentry selection process
TABLEII
is effective and that nodes with high remaining energy have a
KEYSYSTEMPARAMETERS
highprobabilitytobechosenassentries.Forinstance,noneof
Parameter Definition DefaultValue
nodeswithvoltagebelow2.65Vischosenasasentry.Figure8 SDC Sentrydutycycle(see IV-C) 25%
further confirms that it is not the case that nodes with high STP Sentrytoggleperiod(see IV-C) 1second
SSA Sentryserviceactivation True
voltages are always selected as sentries, due to the random
TN Numberoftripwirepartitionsinthenetwork 1
jitter introduced in Equation 1 and to the localized selection TDC Tripwiredutycyclepercentage(seeIV-A.3) 100%
VS TargetSpeed 4m/s
process on a non-uniform distribution of XSM motes.
RN Numberofsystemrotationsperday 1
3) Effectiveness of ROV enforcement: We also investigate SR SensingRange 10m
RR RadioRange 30m
the effectivenessof enforcing Range of Vicinity (ROV), when
wesetthesystemparameterROVas10meters.Figure9shows
B. Hybrid Evaluation
the cumulative distribution function of minimum distances
between sentry-pairs. The average minimum is 9.57 meters In the hybrid evaluation, we use the experimental measure-
with 1.88 meters standard deviation. We note that due to ments from the XSM platform as inputs to a discrete event
SSA=false SSA=true,STP=1s,SDC=100% 35000 SSA=true,STP=6400s,SDC=25%
Averagedetectiondelayforoneday(ms)11225050500000000000000000 20 SSSSAA==4tt0rruuee,,SSTTPP60==11ss,,SSDD8CC0==6225.%5%100 120SSSSAA==1ttrr4uu0ee,,SSTTPP1==6110ss,,SSDDCC1==831072..55%%200 Averagedetectiondelayforoneday(ms)1122350505000000000000000000000 SSSSSSSSSSAAAAA=====2ftttt0rrrrauuuulseeeee,,,,SSSSTTTTPPPP====5211050s,sss4,,,S0SSSDDDDCCCC====25222%555%%% 60 80 100 120
Duration(days)
Duration(days)
Fig. 12. Influence of sentry duty cycle (SDC) on average detection Fig. 15. Influence of sentry toggle period (STP) on average detection
delay(ADD). delay(ADD).
Averagedetectiondelayforoneday(ms)11111468024680000000000000000 SSSSSSSSSSSSAAAAAA======ftttttrrrrrauuuuulseeeeee,,,,,SSSSSTTTTTPPPPP=====11111sssss,,,,,SSSSSDDDDDCCCCC=====1362127250%...0555%%%% Detectionprobability(%)1166778899000505050505 SSSSSSSSSSSSAAAAAA======ftttttrrrrrauuuuulseeeeee,,,,,SSSSSTTTTTPPPPP=====12516050s4,sss0,,,S0SSSsD,DDDCSCCC=D===2C2225=555%%%%25%
0 5 10 15 20 25 0 20 40 Durati6o0n(days) 80 100 120
Duration(days)
Fig.13. Influenceofsentrydutycycle(SDC)onaveragedetectiondelay Fig.16. Influenceofsentrytoggleperiod(STP)ondetectiondelay(DP).
(ADD)(thesecondview)
SSA=true,STP=1s,SDC=12.5% SSA=true,STP=1s,SDC=25% sixstatesin Table I. Whena messageis transmitted, the radio
SSA=true,STP=1s,SDC=37.5% SSA=true,STP=1s,SDC=62.5%
(%)105 SSA=true,STP=1s,SDC=100% SSA=false switchestothetransmitstatefor30ms(atypicaltimerequired
bility19050 byXSMnodestosendamessageundertheMACcontention).
oba 90 The indicatednumber of messagesper secondin Table I is an
pr 85
on 80 upper bound result from the empirical observations.
Detecti 77050 20 40 60 80 100 120 140 160 180 200 tig2at)ePtherrefoermmaajnocrepmereftorricmsaanncde msyesttreimcspuanrdaemredteifrfse:reWntesiynsvteems-
Duration(days)
configurations. 1) Detection Probability (DP), which is the
Fig.14. Influence ofsentrydutycycle (SDC)ondetectionprobability percentage of successful detections among all targets that
(DP). enter into the system during one day. 2) Average detection
Delay (ADD), which is the average time elapsed between the
entrance of a target into the area and its detection by one of
simulator. This simulator emulates the multi-phase VigilNet sensor nodes. 3) Network lifetime (NL), which is defined as
operations as shown in Figure 11. We distribute 10,000 nodes the number of days for which the detection probability of a
randomly within a 1,000,000 m2 square. VigilNet initializes target remains greater than 90%. The key system parameters
in three minutes with a sequence of phases (from Phase I to are listed in Table II. Unless mentioned otherwise, the default
VII). After that, VigilNet enters the surveillance phase (Phase values in Table II are used in all experiments. The baseline
VIII). The system rotates periodically to refresh system-wide for comparison is VigilNet without any power management.
soft-states and balance the power consumption. The number 3) Impact of the sentry service and duty cycle scheduling:
of rotations per day is defined as RN as shown in Table II. In this section, we evaluate the energy savings achieved by
A target enters the network randomly at one of the edges and thesentryserviceandthe dutycyclescheduling.In particular,
exitsrandomlyattheoppositeedgeofthearea.Toemulatethe we study the influence of the activation of the sentry service
sensing delay we experienced in the real testbed, we consider (SSA), of the sentry duty cycle (SDC), and of the sentry
that a target is detected when it is within the sensing range of toggle period (STP) on energy consumption. One hundred
an active node for at least 5 milliseconds and when that node targets are simulated during each rotation to obtain statistics
can reach its tripwire base station to report the event. in detection probability, but we take into consideration the
1) Battery model: We obtained similar empirical power power consumption of only ten of them (VN=10) as the real
consumption results as reported in [2], which provides very workload. As previously mentioned, we use a network of
complete analysis of XSM motes. XSM motes use two stan- 10,000 nodes randomly distributed within a square of 1km
dard AA (A91) batteries. Each battery has an energy capacity edge length. Each node has a radio range of 30 meters. This
uniformly chosen between 2,848mAh and 2,852mAh [47]. configuration matches our real system requirements dictated
However,tomodel reality better[48], we suppose thata mote by the military: nodes have an average of 27.5 neighbors
dies when it has used 85% of the available energy. within their communication range, and an average of 3.1
Thesensornodesareinoneofsixpowerconsumptionstates neighbors within their sensing range.
atanytime.Welistanddetailthepowerconsumptionofthese Figures 12, 13 and 14 show the variations of the average