Optimization in Open System

Wenowimplementtheproposeddeterministicdistributedprimal-dualalgorithm. Specically,

we consider a simple 8-node wireless sensor network as shown inFigure 3.5. All thesensors

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0 1 2 3 4 5 6

Throughput

Node Density (actuators/sq. km.)

No Power Control PC Heuristic Algorithm

Figure3.3: Throughput vs. Actuator Density

0 100 200 300 400 500 600

0 200 400 600 800 1000

Energy Consumption (mJ)

Simulation Time (s)

Control Overhead: Static Case Control Overhead: Dynamic Case

Figure3.4: Energy Consumption for Control Overhead

sample data with

τ _i = 0.1

^{. We use a random access CSMA/CA like MAC without backo.}

We rst x the routing in the network, and thus, xing the arrival rate at each node. We

thenlookat the convergence ofprimal-dual algorithm. The resultsobtained bytheproposed

primal-dual algorithm, togetherwith thetheoreticaloptimal solution, are presentedin Table

3.1. It can be easily seen that the results obtained from the primal-dual algorithm is very

close to theoptimal solution.

s5

s6

s7 s1

s2

s3

s4

Figure3.5: A SimpleNetworkTopology

Table 3.1: Comparison between the results of the proposed primal-dual algorithm and the

theoretical optimal solution

Node

a i µ i − opt µ i−primal−dual

1 0.1 0.102 0.121

2 0.2 0.208 0.225

3 0.1 0.1220 0.125

4 0.2 0.241 0.256

5 0.35 0.383 0.412

6 0.7 0.719 0.743

7 1.05 1.058 1.072

We nowlook at theconvergence ofthe distributedprimal-dual algorithm for some nodes

inthenetwork w.r.t time. Figure 3.6, 3.7, 3.8, and 3.9shows the convergence of distributed

primal-dualalgorithm for node 3,4,5,and 6inthenetwork. It canbeseen thattheoptimal

valueof

µ ₃

^{isobtained bythe}distributedprimal-dual algorithm inlessthan 100 iterationsof the algorithm. Thisshows averyfastconvergence of thedistributedprimal-dual algorithm.

3.11 Conclusions and Future Work

Forwirelesssensor-actuatornetworkswithrandomchannelaccess,weproposethateachsensor

must transmit its readings toward one actuator only inorder to take theburden ofrelaying,

towarddierentactuators,awayfromenergy-constrainedsensorsinastraightforwardfashion.

The objective for theopen systemwasto minimizethe total delay inthenetwork where the

constraintsarethearrival-rateandservice-rateofanode. Particularly,wehaveshownthatthe

0 5 10 15 20 25 0

0.05 0.1 0.15 0.2 0.25

Time −> s (10 ² )

µ 3 primal−dual µ ₃ opt

a 3

Figure3.6: Convergence of

µ ₃

^usingdistributed primal-dualalgorithm

objective functionis strictlyconvexfor theentire network. We thenusetheLagrangian dual

decomposition method to devise a distributed primal-dual algorithm to minimize the delay

in the network. The deterministic distributed primal-dual algorithm requires no feedback

control and therefore converges almostsurely to the optimal solution. The results showthat

the required optimal value of service rate is achieved for every node in the network by the

distributedprimal-dualalgorithm. Itisimportanttopayequalattentiontoboththeobserved

delayinthenetworkandenergyconsumptionfordatatransmissions. Afastconvergencemeans

thatonlyalittleextraenergy isconsumedtoperformlocalcalculationstoachievethedesired

optimizations. Only energy-ecient routing might not serve any purpose for some sensor

network applications. Similarly for the stochasticdelay control algorithm, we have shown a

probabilityoneconvergence anditsrateofconvergencewhichisentirelydistributedinnature.

We then proposed an algorithm for an optimal actuator selection that provides a good

mapping between any sensor and an actuator in thenetwork. The selection algorithm nds

a delayoptimal actuator for eachsensorinpolynomialtime. We nallyproposeadistributed

actuationcontrolmechanismforSANETsthatisresponsibleforanecientactuationprocess.

The actuators can dynamically coordinate and perform power control to maintain a dened

level of connectivity subject to throughput constraints. The control overhead for static and

mobile actuator scenarios is analyzed using ns-2 simulations. The PC heuristic algorithm is

applicable to multihop SANETs to increase throughput,batterylife, andconnectivity.

In future, we will present a detailed simulation based study of PC heuristic algorithm

in dierent networking scenarios with some application specic actuation requirements and

practical evaluationof distributedmultiple-actuator actuation process. We will also work on

0 5 10 15 20 25 0.15

0.2 0.25 0.3

Time −> s (10 ² )

µ 4 primal−dual µ ₄ opt

a 4

Figure3.7: Convergence of

µ ₄

^usingdistributed primal-dualalgorithm

the development of PCheuristic algorithm to improve some MAC layerperformancemetrics

usingacross-layerapproach. Asaconsequenceofaveryfastconvergence toWardrop

equilib-rium,wearealsotemptedtoperformtheenergy-analysisoftheproposedlearningandrouting

schemeinthe context of network lifetime.

In thefollowing chapter, we will look at the energy eciency issues for data aggregation

inSANETs.

0 5 10 15 20 25 0.3

0.32 0.34 0.36 0.38 0.4 0.42 0.44 0.46 0.48 0.5

Time −> s (10 ² )

µ 5 primal−dual µ ₅ opt

a 5

Figure3.8: Convergence of

µ ₅

^usingdistributed primal-dualalgorithm

0 5 10 15 20 25

0.65 0.7 0.75 0.8

Time −> s (10 ² )

µ ₆ primal−dual µ ₆ opt

a 6

Figure3.9: Convergence of

µ ₆

^usingdistributed primal-dualalgorithm

The LEAD Cross-Layer Architecture

for SANETs

SANETsarecomposedofsensorsandactuatorslinkedtogetherbywirelessmediumtoperform

distributedsensing and acting tasks. Delay andenergy constraints have a signicant impact

on the design and operation of SANETs. We consider a sensor-actuator network in which

both energy and delayarehard constraintsand mustbejointly optimized.

In this chapter, we present the design, implementation, and performance evaluation of a

novellow-energy,adaptiveanddistributed(LEAD)self-organization framework. This

frame-work provides coordination, routing, and MAC layer protocols for network organization and

management. Weorganizetheheterogeneous sensor-actuatornetworkintoclusterswhereeach

cluster is managed by an actuator. To maximize the network lifetime and attain minimum

end-to-end delays, it is essential to optimally match each sensor node to an actuator and

nd an optimal routing scheme. We provide an actuator discovery protocol that nds out a

destination actuator for each sensorinthe network basedon theoutcome ofa cost function.

Further,oncethe destination actuatorsarexed,we provideanoptimal owrouting solution

withtheaim ofmaximizing network lifetime. We thenproposea delay-energy aware TDMA

based MAC protocol in compliance withthe routing algorithm. The actuator-selection,

op-timal routing, and TDMA MAC schemes together guarantees a near-optimal lifetime. The

proposal is validated bymeans ofanalysis andns-2 simulationresults.

Furthermore, preventing sensor nodes from being isolated is very critical. The problem

of sensorinactivity arises from the pathloss and fading that degrades the qualityof the

sig-nals transmitted from actuators to sensors, especially in anisotropic deployment areas, e.g.,

rough and hilly terrains. Sensor data transmission in SANETs heavily relies on the

schedul-ing information that each sensornode receives fromits associated actuator. Therefore ifthe

signal containing scheduling information is received at a very low power due to the

impair-mentsintroduced bythewireless channel, thesensornode might be unable to decode it and

consequently itwill remainisolated.

Each sensornode transmits its datato only one of theactuators. However, all actuators

cooperateandjointlytransmitschedulinginformationtosensorswiththeuseofbeamforming.

This results inan important reductioninthe numberof isolated sensorscomparing to single

actuatortransmissionforagivenleveloftransmitpower. Thereductionisduetotheresulting

arraygainandtheexploitationofmacrodiversitythatisprovidedbytheactuatorcooperation.

4.1 Introduction

Distributed systems based on networked sensors and actuators with embedded computation

capabilities enable an instrumentation of the physical world at an unprecedented scale and

density,thus enabling a newgeneration ofmonitoring and control applications. SANETs are

anemergingtechnology thathasawiderangeofpotentialapplicationsincludingenvironment

monitoring, medicalsystems,robotic exploration,andsmartspaces. Suchnetworksconsistof

alarge number of distributedsensor and fewactuator nodes thatorganize themselves into a

multihop wirelessnetwork. Each sensornodehasone or moresensors (includingmultimedia,

e.g., video and audio, or scalar data, e.g., temperature, pressure, light, infrared, and

mag-netometer), embedded processors, and low-power radios, and is normally battery operated.

Typically, thesenodescoordinateto perform a common task. Whereas,theactuators gather

this information and react accordingly.

Sensor-actuator networkshave thefollowing unique characteristics:

•

^Real-time requirement: Depending on the application there may be a need to rapidly respond to sensor input. Examples can be a re application where actions should be

initiated ontheeven areaassoonaspossible.

•

Coordination: UnlikeWSNswherethecentralentity(i.e.,sink)performsthefunctionsof datacollectionandcoordination,inSANETs,newnetworkingphenomenacalled

sensor-actuator and actuator-actuator coordination may occur. In particular, sensor-actuator

coordinationprovidesthetransmissionofeventfeaturesfromsensorstoactuators. After

receiving eventinformation, actuatorsmayneed to coordinatewitheachother (depend

on the acting application) in order to make decisions on the most appropriate way to

perform the actions.

In this chapter, we investigate a new self organizing framework for SANETs. We consider a

heterogeneous network thatconsistsof sensorsand actuator nodesrandomly deployed inthe

network. Eachsensormusttransmititsdatatoonlyoneoftheactuatorstoconservethescarce

energy resource. Thisarisesthe problemof actuator selection for eachsensorinthenetwork.

Inthelastchapter,wediscussedadelayoptimalactuatorselectionalgorithm. Whereas,inthis

chapter, welookattheenergyissueswhileselectinganactuator. Inparticular,weproposean

optimal actuator selectionand owrouting protocol(LEAD-RP)withtheaimof maximizing

the network lifetime. We show that the actuator-selection and ow routing problem with

energy constraints can be modeled asa mixed integer non-linear programming optimization

problem(MINLP)[61]. SinceMINLPisNP-hardingeneral,wedevelopadistributedapproach

which provides agoodapproximation of theoptimal solution. We usea relaxation technique

in order to decide on the optimal actuator and then optimize the ow routing toward this

actuator to extend network lifetime. For optimalactuator selection,wepropose an Actuator

Discovery Protocol (LEAD-ADP) that collects information about neighboring actuators for

each sensor node in the network. The destination actuator is decided as outcome of a cost

function. Once the destination actuators are xed, we nd out an optimal ow routing to

maximizethenetworklifetime. Bothofthesestepsarecarriedoutatthenetworklayer. Atthe

MAC layer, we propose an adaptive TDMA like MAC (LEAD-MAC) with minimized awake

periods (LEAD-Wakeup) to avoid the problem of synchronization during ow splitting and

to meet thedelay constraints inSANETs. The actuator selection, optimal ow routing, and

TDMA MACsolutiontogether guaranteea near-optimal lifetime for SANETs.

Depending on the application there may be a need to rapidly respond to sensor input.

Moreover, to provide right actions, sensor data must still be valid at the time of acting.

Therefore, the issue of real-time communication is very important in SANETs since actions

are performed on the environment after sensing occurs. Examples can be a re application

where actionsshould beinitiated on theevent area assoon aspossible. Unlike WSNs where

the central entity (i.e., sink) performs the functions of data collection and coordination, in

SANETs, new networking phenomena called sensor-actuator and actuator-actuator

coordi-nation may occur. In particular, sensor-actuator coordination provides the transmission of

event features from sensors to actuators. After receiving event information, actuators may

need tocoordinatewitheachotherinordertomakedecisions onthemostappropriatewayto

perform theactions. Eachsensornode isassociated withanactuator whichisthedestination

of the sensordata. In orderto prevent sensordata collisions,actuators transmit time

sched-ules which coordinate sensor multi-hop transmission. Therefore each sensor after receiving

the scheduling information from its associated actuator transmits its dataat the right time

slot. Ifthesignalcontainingtheschedulinginformation isreceivedataverylowpowerdueto

channel impairments, the sensornode might be unable to decode it and consequently it will

remain isolated.

To thebestof our knowledgethe potential problem ofisolated sensornodesina SANET

hasnot been investigated. Actuatorsreceive sensordataina multi-hopfashion andtransmit

theschedulinginformationtotheminasinglehopfashion. Asensornodeneedstodecodethe

received scheduling information fromtheactuator that itis associated with. Thisis inorder

to know its assigned time slot inwhich it should transmit its sensed data. However due to

the impairments introduced by the wireless channel (signal degradation due to pathloss and

fading), it is very likely that some sensor nodes, more likely the ones that are distant from

theactuator, wouldnotbeableto decode theirscheduling information. Thisisbecause some

sensornodeswouldprobablyreceivethesignalcontainingschedulinginformationataverylow

Signal-to-Noise Ratio(SNR).Consequently theywillremainisolated,afactthatcouldcreate

someisolatedzonesinthesensingeld. Thiswouldresulttoincompleteinformationreception,

a situation that needs to be overcome for the uniform monitoring of the sensing eld. A

potentialsolutiontothiswouldbetheuseofpositiveand/ornegativeacknowledgments(ACKs

and/orNACKs) withrespectto the reception of scheduling information. In this fashion, for

thesensornodesthat cannotdecode their scheduling information, multi-hop transmission of

theirschedulescanbeemployed. Howeverthiswouldresulttoasignicantoverheadburdenin

termsoftimeandenergywasteofthesensornodes,thatcanreducetheirlifetime. Furthermore

thattypeof solutionwouldincrease thecomplexityof theemployed protocols.

Forasensornetworkwithmultiplesinks(sinks/actuatorscanbethoughtofsimilarentities

for design purposes),the trac generated bysensor nodesmay besplit and sent to dierent

sinks [62, 25 ]. In the presence of multiple sinks, the problem of optimal sink selection with

theaimof extending lifetimeusing anycast routing isstudied in[63]. The authors proposea

heuristic solutionbasedon trac volumes sent to dierent basestations to selectan optimal

basestation. The proposed solution is based on ow splitting which follows dierent routes

from a source to its selected destination. The provided solution is elegant in the essence of

extending lifetime at routing layer. The only issue with this solution is the synchronization

(MAC layer) among dierent nodesto which a source (sensor) directs its ow. Theydo not

addressthissynchronizationprobleminthepaper. Simulationresultsshowbetterperformance

based onnumerical dataand theissuesrelatedto MACand synchronizationwereelevated.

Incases,whentherearemultipleactuatorsandmappingbetweenthesensorsandactuators

isnot given,the jointproblem ofndingan optimalactuator and extendingnetwork lifetime

with minimum end-to-end delay constraints is a challenging and interesting problem. This

problem is relevant from both the application's and wireless networking perspectives. From

an application requirement perspective,some real-timemultimedia sensingapplications(e.g.,

video surveillance )require tohave allthe trac generatedfrom asource sensorto be routed

to the same actuator (it may follow dierent routes) so that decoding and processing can

be properly completed because the information from the same source is highly correlated

and dependent. Froma wireless networkingperspective,the actuator chosen asa sinkcould

have asignicant impacton theend-to-enddelayswhich isahard constraint[C-3] for

sensor-actuator applications. Thisisbecausetheend-to-enddelaysaretopologydependent;actuator

selectionsimplybasedonenergyconstraintscannotguaranteeoptimalend-to-enddelays,and

therefore, itshould be basedon both delay-energy constraints. Asa result, there appears to

beavitalneedtounderstandhowtoperformoptimalroutingtojointlyachieveminimum

end-to-enddelayroutesandoptimizenetworklifetimeindelay-energyconstrainedsensor-actuator

networks.

Inthischapter, weproposeaPHY,RoutingandMACsolutionwiththeaimofeliminating

isolated zones inthe sensing eld, maximizing the network lifetime, and attaining minimum

end-to-end delays. The problem of sensor inactivity can be eectively faced on the physical

layer without increasing the protocol complexity and dissipating extra energy from sensor

nodes. Actuatorscancooperateandformadistributedantennaarray,aconceptthathasbeen

proposed for cellular communications [64 ]. The array jointly performs adaptive beamforming

anddistributesthetimescheduletoeachsensornode. Sensorsreceivethescheduleinformation

at a much higher power due to the array gain that results from beamforming and to the

exploitation of macro-diversity which isinherent tothedistributednature ofaSANET.This

results to asignicant reductioninthenumberofisolated sensorsfor a given transmit power

level. The cost is the need of Channel State Information at the transmitter (CSIT). It is

shown by Matlab simulations that this eectively faces the problem of isolated zones. It is

thenproposedthateachsensornodetransmits itsdatato onlyone actuator. Asensorselects

an actuatorwhich isminimum numberofhopsaway. Note thatthisactuator selectionisjust

todecideaterminalpointforsensordatatransmissionsandmulti-path routingisactuallyused

to transmit databetween a sensorand its associated actuator. An advantage of settingmin.

hop criteria for actuator selection is that the lower-tier (sensor-actuator coordination level)

of our heterogeneous network can be organized into clusters, where each cluster is centrally

managed by an actuator. It is also shown thatthe owrouting with energy constraints can

bemodeledasanon-linearprogrammingoptimizationproblem(NLP).Weusearelaxationto

optimize theowrouting towards this actuator to extend networklifetime. We thenpropose

to usean adaptive TDMA like MAC (that corresponds to therouting solution) to avoidthe

problem ofsynchronization during owsplitting andto meet thedelay criteria forSANETs.

The organization of this chapter is as follows. Section 4.2 highlights some interesting

related literature. The problem formulation is presented in Section 4.3. In Section 4.4, we

provide the network model underconsideration indetail. Section 4.5 focuses on the

coordi-nation framework. Section 4.6 details the design criteria of our proposed actuator-selection,

optimal-routing scheme and optimization algorithm. In Section 4.7, we present a distributed

networklearningframeworktosolvetheactuator-selectionproblem. InSection4.8,wepresent

a primal-dualalgorithmfor lifetimemaximization. The mediumaccessschemeisdiscussedin

Section4.9. InSection4.10,wepresentourLEAD-wakeupprotocol. Threedierent

actuator-to-sensor transmissionschemesaregiveninSection4.11. Thesimulationresultsarepresented

inSection 4.12. InSection 4.13, we conclude thechapter andoutlinethefuture directions.

4.2 Related Literature

To our knowledge, sensor-actuator networks have not been extensively studied in the

net-workingliterature. However, our work inthis direction has been informed and inuenced by

a varietyofprevious research eortsinthedomain of WSNs,which we now describe.

TSMP (Time Synchronized Mesh Protocol) [70] is a networking protocol that forms the

foundation of reliable, ultra low-power wireless sensor networking. TSMP provides

redun-dancy and fail-over in time, frequency and space to ensure very high reliability even in the

redun-dancy and fail-over in time, frequency and space to ensure very high reliability even in the

Dans le document Cross layer optimization of wireless sensor and sensor-actuator networks (Page 95-110)