Industrial Wireless Sensor Networks: Research Challenges and Novel Solutions

Informati a e delle Tele omuni azioni

DOTTORATO DI RICERCA IN

Ingegneria Informati a e delle Tele omuni azioni

XXIII Ci lo

Industrial Wireless Sensor Networks:

Resear h Challenges and

Novel Solutions

Ing. Emanuele Antonino Tos ano

Coordinatore

Prof. O. Mirabella

Tutor

1 Introdu tion 1

1.1 Overviewof the ommuni ation proto ols for lassi al WSNs 3

1.1.1 Optimization-based proto ols . . . 3

1.1.2 Data- entri proto ols . . . 5

1.1.3 Cluster-based proto ols . . . 5

1.1.4 Lo ation-based proto ols. . . 6

1.1.5 Topology management proto ols . . . 7

1.2 Dieren es between lassi alWSNs and industrialWSNs. . 8

1.2.1 Ar hite ture. . . 8

1.2.2 Requirements . . . 10

1.3 Resear h hallenges and possible solutions . . . 11

2 Assessment of ross- hannel interferen e in IEEE 802.15.4 networks 13 2.1 Coexisten e ofwireless networks . . . 14

2.2 On ross- hannelinterferen e inIEEE 802.15.4 . . . 16

2.3 Testbed andMethodology . . . 18

2.3.1 Methodology . . . 19

2.4 Case study and experimental results . . . 23

2.4.1 The IEEE802.15.4platform . . . 23

2.4.2 Preliminary Assessments . . . 25

2.4.3 WorstCase PER validation . . . 27

2.4.4 Interferen ebyasingle node . . . 28

2.4.5 Interferen efrom multiplenodes . . . 31

2.4.6 Inuen eof MACparameters . . . 34

3 Multi hannel SuperframeS heduling for IEEE 802.15.4 39

3.1 TheIEEE 802.15.4proto ol . . . 40

3.2 Cluster-treetopologies andbea onframe ollisions . . . 42

3.3 Approa hesfor bea onframe ollisionavoidan e. . . 44

3.4 TheSDS Algorithm . . . 46

3.5 Themulti hannel approa h . . . 48

3.6 Multi hannel SuperframeS heduling . . . 49

3.7 Analysisofthe MSS algorithm . . . 53

3.7.1 MSSS hedulability . . . 54

3.7.2 Frequen y onstraints . . . 57

3.8 Implementation issues . . . 58

3.8.1 Changesto theIEEE802.15.4MAC . . . 58

3.8.2 AddingMSS Supportto theupperlayers . . . 60

3.9 Experimental Testbed . . . 62

3.10 Con ludingremarks . . . 67

4 A Topology Management proto ol for RT-WSNs 69 4.1 Approa hesto improve WSNperforman e . . . 70

4.2 Network model . . . 72

4.3 DesignPrin iples . . . 73

4.4 Theproposedtopology management proto ol . . . 75

4.4.1 InitState. . . 76

4.4.2 Cluster Head. . . 78

4.4.3 RelayNode. . . 79

4.4.4 CommonNode. . . 80

4.5 Dis ussionand Proto olAnalysis . . . 80

4.6 Performan e Evaluation . . . 87

4.6.1 EnergyE ien y of theproposedsolution . . . 87

4.6.2 Ee tof thedelaybound onCH toRN transmissions 89 4.6.3 Ee t ofthe proposedtopology management me ha-nismon theSPEED routing proto ol . . . 91

4.7 Con ludingremarks . . . 94

5 An improved dynami topology management proto ol for RT-WSNs 95 5.1 Benetsandlimitations ofthe stati approa h . . . 96

5.2 Thedynami approa h . . . 98

5.2.2 Node Initalization . . . 99

5.2.3 AU organization . . . 102

5.2.4 Lifetime Estimation . . . 103

5.2.5 Dynami energy balan ing . . . 104

5.3 Performan eevaluation . . . 106

5.3.1 Ee t of thenode ex hange poli y onnetwork lifetime 106 5.3.2 Ee tofthetopologymanagementproto olonSPEED real-time performan e . . . 109

5.4 Con luding remarks . . . 111

6 A hain-based routing proto ol for industrial WSNs 113 6.1 Chain-based routing inWSNs . . . 114

6.1.1 Linear s heme. . . 114

6.1.2 Binary- ombining s heme . . . 115

6.1.3 Multiple- hain s heme . . . 115

6.2 The Cir ularChain DataForwarding (CCDF)proto ol . . . 116

6.3 Fault Toleran e Me hanisms . . . 119

6.4 Distributed Chain Creation . . . 120

6.4.1 Sub- hain reationalgorithm . . . 123

6.5 Proto ol Analysis . . . 124

6.5.1 Node TraversalTime (NTT) . . . 124

6.5.2 Chain Traversal Time . . . 126

6.5.3 AverageChain Trip Time . . . 128

6.6 Performan eevaluation . . . 131

6.6.1 Validation of theoreti alresults . . . 132

6.6.2 Comparative assessments . . . 135

6.7 Con luding remarks . . . 137

1.1 Ar hite ture of atypi al industrialnetwork. . . 9

2.1 Stru tureof the testbed. . . 19

2.2 Modelfor overlapping transmissionprobability. . . 22

2.3 PreliminaryAssessments . . . 27

2.4 Pa ket Lossand expe ted PERversus thevarying transmis-sion periodof the interferernode. . . 28

2.5 Pa ket Losswithequidistant nodes.. . . 29

2.6 Expe ted worst asePER withequidistant nodes. . . 30

2.7 EstimatedPERasafun tionofthedieren ebetween inter-ferer andsour e re eived powerlevel. . . 31

2.8 Spe traofthe threeinterferernodesseenbythere eiver an-tenna. . . 32

2.9 Pa ket Lossusing multiple interferernodes. . . 33

2.10 The ee t ofthe CCAThreshold. . . 35

2.11 Theee tofthe MinimumBa koExponentondelayunder ross- hannelinterferen e. . . 36

3.1 Network topology. . . 43

3.2 Dire t andindire t bea on ollisions. . . 44

3.3 Examples enariowhere twogrouped oordinators mayhave interfering nodes. . . 48

3.4 S heduling the lusters inalternate timesli es(TS1 andTS2). 49 3.5 MSS superframes heduling. . . 51

3.6 S heduling of asuperframeset thatisunfeasible usingSDS. 54 3.7 Superframestru ture. . . 60

3.8 Asso iationof a(non PAN-) oordinator. . . 61

3.10 Softwarear hite ture ofTKN15.4 . . . 63

3.11 Testbed S enario. . . 65

3.12 Temporal tra eof the experiment: asso iationofC5. . . 66

3.13 Temporal tra eof the experiment: asso iationofC4. . . 66

3.14 Temporal tra e of the experiment: steadystate network op-erations . . . 67

4.1 Statediagramoftheproposedtopologymanagement proto ol. 76 4.2 Theproposedtopology management proto ol . . . 81

4.3 Designspa eof theproposedtopology management proto ol 86 4.4 MeanAU power onsumption vs. varying super-framelength. 89 4.5 Su essRatiofor CH-to-RN syn hronous ommuni ations. . 91

4.6 End-to-end delay. . . 92

4.7 Pa ketlossratio. . . 93

4.8 SPEEDHit ratio. . . 93

5.1 Super-frame stru ture of the stati topology management proto ol. . . 96

5.2 Super-framestru ture ofthedynami topologymanagement proto ol. . . 99

5.3 Graphi al Representation ofAUlifetime. . . 104

5.4 Exampleof node ex hange between two AUs. . . 105

5.5 AU onguration after the initialization phase (a) and AU lifetimedistributionafter one minute(b). . . 107

5.6 Dynami evolutionofthe WSN intermsof AU lifetime. . . 108

5.7 Network lifetimewithdynami AU adaptation. . . 109

5.8 Performan e omparison.. . . 110

6.1 Network Ar hite ture. . . 117

6.2 Chain-based s hemes. . . 118

6.3 Network Setup. . . 121

6.4 Node Traversal Time. . . 125

6.5 ChainTraversalTime. . . 128

6.6 Timeto re overlost frame. . . 130

6.7 ChainTripTimes. . . 134

2.1 In-air jamming resistan e obtained with 2 m distan e from

transmitter to there eiver. . . 18

2.2 Basi Testbed onguration . . . 25

Introdu tion

A Wireless Sensor Network (WSN) is a olle tion of nodes organized into

a ooperative network, typi ally operating in an unattended environment.

Ea hnodeisequippedwithapro essingelement,aradiofrequen y

trans eiv-er(usuallywith asingleomnidire tionalantenna), anumberofsensorsand

a tuators, memories (data, program, ash) and a power sour e. From the

fun tionalpointofview,nodes anbe lassiedassour es,sinksandrouters.

Sour es aresensornodesthatmonitor adenedphenomenon (e.g.

temper-ature) and transmit data, whereas sinknodes are those whi h olle t and

pro ess data. Routersarenodesthatarein hargeof forwarding datafrom

sour estoward thesink(s). Nodes anplaymultiple rolesat thesametime,

e.g., sour es may also a t as routers. Moreover, multiple nodes an take

part indatapro essingbeforedelivering data to the naldestination.

When a WSN is designed, multiple oni ting requirements should be

metsimultaneously. Ontheonehand,nodesshouldhavesu ient

omput-ing and storage apabilities and enough bandwidth for transmission, they

should beable to work autonomouslyand mayhave dierent QoS

require-ments (e.g. limited end-to-end delay). On theother hand, devi es should

havelow ostsandlimitedenergy onsumption,sothatlonglifetime anbe

a hieved. Although the orre t trade-o between these oni ting

require-mentsis dependent on thespe i WSN appli ation, most of the resear h

on WSNs fo uses on howto in rease thenetwork lifetime.

Inordertomeetthelong-lastingrequirement, WSNnodestypi ally

fea-ture low-power pro essors and very small memories. However, this is not

su ient, as the energy onsumption in WSNs is typi ally dominated by

Inordertoprolongthenodes'lifetime, andthus thelifespanofthenetwork

asa whole, as mu h as possible, strategies aiming at redu ing energy

on-sumptionhave to be implementedat allthe dierent levels of thenetwork

proto ol sta k. This is why literature oers many ommuni ation

proto- ols aiming at redu ing energy onsumption, implemented at the various

levels of the proto ol sta k, from thephysi al up to theappli ation layer,

and even ross-layer approa hes to save energy are foundin theliterature.

An overview of the existing literature on WSN ommuni ation is given in

Se tion 1.1.

Industrial appli ations an take advantage of the lower ost and easier

deployment of WSNs as ompared to traditional industrial networks [1℄.

While the deployment of a traditional industrial network infrastru ture is

ostly and time- onsuming, WSNs only need a minimal infrastru ture, if

any. In addition, WSNs allow greater exibility and s alability than

tra-ditional industrial networks. In industrial s enarios a WSN may be used

to redu e the networking ost of less riti al ontrols and/or to onne t

dierent network ells (i.e., dedi ated eldbuses) for monitoring purposes.

However, industrial WSNs have both dierent requirements and dierent

ar hite tures than traditionalWSNs [2℄. In industrial WSNs themost

im-portant requirement isto a hieve a predi table behaviourand bounded

la-ten y. Energy stillplays arole,butisless ru ialthanintraditionalWSN,

as industrial WSNs are not supposed to be unattended for long periods.

Con erning network ar hite ture, unlike traditional WSNs whi h typi ally

have toworkwithout anyinfrastru ture,anindustrialWSNisusually

on-ne ted to a real-time wired ba kbone (e.g., Industrial Ethernet or a

real-time eldbus), be ause dataows required by riti al ontrol loops annot

be transmitted overthe wireless medium. A more detailed explanation of

thedieren esbetween lassi alandindustrialWSNsisgiveninSe tion1.2.

Su h dieren esmake the existingproto olsfor lassi alWSNs unsuitable,

or justin onvenient, for theimplementation of industrialWSNs.

Thisthesisinvestigatesnovelapproa hesatdierentlevelsofthe

proto- olsta k, whi hareexpli itly developed for industrialWSNs. Asitwill be

explained in Se tion 1.3, the proposed me hanisms and proto ols address

dierent hallenges(e.g., robustness to theinterferen es,better bandwidth

exploitation,energye ien y,boundedend-to-endtransmissiondelay),but

allofthempursuethe ommon obje tive ofmakingWSNte hnologyready

1.1 Overview of the ommuni ation proto ols for

lassi al WSNs

As sensor nodes are typi ally battery-operated, energy saving is a major

design issuein lassi alWSNs. It hasbeenproven thatthe ommuni ation

ost for sensor nodes is mu h higher than the omputational ost. For

this reason, when deploying a WSN, the network topology, and thus the

distan e between ommuni ating nodes, is a ru ial aspe t. In some ases

sensors an be put in pla e in a ontrolled way, so the WSN an be built

inan energy-e ient wayifa suitablenodepla ement strategy isfollowed.

However, inmost pra ti al ases sensornodesarerandomly s attered over

the eld, so WSNs are self-organizing and deployed in an ad ho fashion,

andthenetworktopology annotbeseta ording toanystrategytargeting

energy onsumption. Asaresult,inorderto prolongthenetwork's lifetime

as mu h as possible, approa hes aiming at redu ing energy onsumption

have to be implemented at all the dierent levels of the network proto ol

sta k, from the physi al up to the appli ation layer, and even ross-layer

approa hesto save energy arefound intheliterature.

The strategies working at the physi al layer try to redu e system-level

power onsumption throughhardwaredesign or bymeans of suitable

te h-niques, su h asDynami Voltage S aling or duty y leredu tion. The

ap-proa hesoperating at thedatalinklayer typi ally exploit low-power MAC

proto ols aimed at redu ing the main auses of energy wastage, i.e.,

ol-lisions, overhearing, idle listening and the proto ol overhead due to the

ex hange ofa highnumberof ontrol pa kets. At thenetwork layerenergy

onsumption is mainlydealt withindatarouting.

Energy-saving routing proto ols for WSNs an be lassied into four

main ategories, i.e., optimization-based, data- entri , luster-based, and

lo ation-based. Su h ategories are not ne essarily disjoint, and some

ex-amples of routing algorithms mat hing multiple ategories an be found.

Examples given here are the TEEN [3℄ and the APTEEN [4℄ proto ols,

whi h arebothdata- entri and luster-based.

1.1.1 Optimization-based proto ols

A broad spe trum of routing algorithms for WSNs aiming at redu ing the

energy onsumption of sensor nodes arepresent in the literature. Some of

most of them the main goal is the optimization of some metri . For this

reason, we will hen eforward refer to them as optimization-based

energy-awareroutingproto ols. Exampleofmetri stobeminimizedaretheenergy

onsumed per message, the varian e in the power level of ea h node, the

ost/pa ketratio,or themaximumenergy drain ofanynode.

Tryingto minimizetheenergy onsumed permessage mayleadtopoor

routing hoi es, assome nodes ould be unne essarilyoverloaded and thus

ouldqui klyextinguishtheirbatteries. Amoreee tiveoption,ifallnodes

areequallyimportant fortheWSNtooperate orre tly,istotrytobalan e

the battery power remaining in the nodes, as there is no point in having

battery power remaining in some nodes while the others have already run

out of power. Minimization of the ost/pa ket ratio involves labeling

dif-ferent links withdierent ostsand then hoosing the best option soasto

delay the o urren e of network partitioning as long as possible. On the

otherhand,the ideaofminimizing the maximumenergy drainofanynode

derivesfromthe onsideration thatnetworkoperationsstart tobe

ompro-misedwhentherstnodeexhaustsitsbattery,soitisadvisabletominimize

battery onsumption inthis node.

Anumberofoptimization-basedpower-aware routingapproa hestryto

maximizenetworklifetime. Theytargetnetworksurvivability,meaningthat

theirgoalistomaintainnetwork onne tivityaslongaspossible. Toa hieve

this goal, optimal routes that avoid nodes with low batteries and try to

balan ethetra loadare hosen[5℄. Theuseofoptimizationte hniquesto

ndthe minimum ost path, where the ost parameter takesenergy (alone

or ombined with other metri s) into a ount, is proposed. However, the

minimum ostpathapproa hhasadrawba kintermsofnetworklifetimein

thelongterm. Infa t,aproto olwhi h,on eithasfoundanoptimalpath,

uses only that path for routing will eventually deplete the energy of the

nodesalong the path. Aslarge dieren esinthe energy levels of theWSN

nodes ouldlead toundesiredee ts su hasnetwork partitioning,suitable

solutions have been developed. A notable example is the Energy-Aware

Routing proto ol [6℄, where network survivability is pursued by hoosing

nota singleoptimalroute, but aset ofgood routes,i.e.,sub-optimal paths

1.1.2 Data- entri proto ols

Unlike the optimization-based routing algorithms des ribed above, other

routing proto olsforWSNsobtainlowpower onsumption forsensornodes

withoutexpli itlydealingwithenergy onsiderationswhenperformingroute

sele tion,butimplementingme hanismswhi hredu eenergywastage. One

of the main auses of energy wastage in WSNs is data redundan y, whi h

derives froma ombinationof a la kof global identiers (as noIP-like

ad-dressingispossibleinWSNs)andtherandomdeploymentofsensors,whi h

in many ases makes it di ult, if not unfeasible, to sele t a spe ied set

of sensors within a given area. To solve this problem, data- entri routing

approa heswereintrodu ed. Intheseapproa hes,dataisnamedusing

high-level des riptors, alled meta-data, and data negotiation between nodes is

used to redu e redundan y. Another approa h to redu e data redundan y

(and the onsequent energy wastage)is byperforming data aggregation at

therelayingnodes, whi h onsistsof ombining datafromdierentsour es

and eliminating dupli ates, or applying fun tions su h as average,

mini-mum and maximum. Data aggregation also over omes the overlap

prob-lem, whi h arises when multiple sensors lo ated in the same region send

the same data to the same neighbour node. Thanks to data aggregation

signi ant energy savings an be a hieved,as omputation at sensornodes

is less energy- onsuming than ommuni ation. When performed through

signalpro essing te hniques, dataaggregationis referredto asdata fusion.

A ording to the kind of routing proto ol, dataaggregationmaybe a task

performed by spe ial nodes or any node in the network. Notable

exam-ples of data- entri routing proto ols whi h perform data aggregation for

energy-saving purposes are SPIN [7℄ and Dire ted Diusion [8℄, whi h in

turn inspiredseveral otherproto ols.

1.1.3 Cluster-based proto ols

Another riti al aspe t for energy onsumption is the presen e of nodes

whi h, being either loser to the sink or on the optimal (e.g.

minimum- ost) path to the sink, perform more relaying than the other nodes, thus

depletingtheir energy reservefaster thantheothers. Whensu h nodesrun

outofenergy,networksurvivabilityis ompromised,andwhenallthenodes

losest to thesink die, the sink itself be omes unrea hable. To avoid this

problem, hierar hi al or luster-based routing was introdu ed. In

thesink. Ea hofthem olle tsdatafromthesensorsbelongingtoits luster

and forwards itto the sink. In heterogeneous networks, luster heads may

be dierent from simple sensor nodes, being equipped with more powerful

energy reserves. In homogeneous networks, on the other hand, inorder to

avoid a qui kdepletion of luster heads, the luster head role rotates, i.e.,

ea h node works as a luster head for a limited period of time. Energy

savinginthese approa hes anbeobtainedinmanyways,in luding luster

formation, luster-headele tion,et . Someoftheseapproa hesalsoperform

dataaggregationat the luster-headnodesto redu edata redundan y and

thus save energy. Notable examples of luster-based routing proto ols are

LEACH [9℄ andits extensionssu h asTEEN[3℄ and APTEEN [4℄.

Derivedfromthe luster-basedproto olisa ommuni ationmodelwhere

nodesarenotexpli itlygroupedinto lusters,butea hnodeonly

ommuni- ateswitha loseneighbourandtakesturnstotransmittothebasestation,

thus redu ing the amount ofenergy spent perround. This is alled

hain-based approa h, asdata goes a ross a hain of nodes, from the sour es to

the nal destination. This lass of proto ols will be dis ussed in a more

detailedwayinChapter6,Se tion 6.1.

1.1.4 Lo ation-based proto ols

Lo ation-basedroutingproto olsusepositioninformationfordatarelaying.

Lo ation information an be exploited for energy-e ient data routing in

WSNs as, based on both the lo ation of sensors and on knowledge of the

sensedarea,adataquery anbesentonlytoaparti ularregionoftheWSN

ratherthan the wholenetwork. Thisfeatureof lo ation-based routing

pro-to olsmay allow for asigni ant redu tion inthe numberoftransmissions

andthus inthe power onsumptionof sensornodes.

The Geographi and Energy-Aware Routing (GEAR) proto ol,

de-s ribed in [10℄, whi h uses an energy-aware metri along with

geographi- al information to e iently disseminate data and queries a ross a WSN.

Unlike other geographi al proto ols not spe i ally devised for sensor

net-works,su hasthewell-knownGreedyPerimeterStatelessRouting(GPSR)

proto ol [11℄, this proto ol addresses the problem of forwarding data to

ea h node inside a target region. This feature enables GEAR to support

data- entri appli ations.

SPEED [12, 13℄ ia another well-known lo ation-based proto ol

forward-ing to a hieve to manage the QoS. Thebasi ideais to maintain a desired

delivery speed a ross the sensor network. A similar approa h is used in

RPAR[14℄,wheretransmissionpoweradaptation isusedtondatrade-o

between delivery speed and energy e ien y.

1.1.5 Topology management proto ols

Topology management proto ols areaslightly dierent approa h to saving

energy than standard routing proto ols, as they do not dire tly operate

dataforwarding. Theseproto ols runat alowerlevel ofthenetwork sta k,

i.e. just underthe network layer. Theirobje tive isto improve theenergy

e ien y of routing proto ols for wireless networks by oordinating the

sleep transitions of nodes. Several routing proto ols infa ttry to enhan e

networklifetimebyredu ingthenumberofdatatransmissionsor balan ing

the transmission power, but negle t idle power onsumption. However,

several measurements, e.g. in [15,16℄, show that idle power dissipation

should not be ignored, as it ould be omparable to the transmitting or

re eivingpower. Therefore,inordertooptimizeenergy onsumption,nodes

shouldturnotheirradios. Topology ontrol proto olsexploit redundan y

indensenetworksinorderto putnodesto sleepwhilemaintainingnetwork

onne tivity. They anbeapplied tostandard routingproto olsfor ad-ho

networksorforWSNsthatdonotdire tlyhandlesleeps hedules. Although

some ofthemaredesigned forwireless ad-ho networksrather thanWSNs,

the typi ally high redundan y of sensor nodes and the need for maximum

energy saving make WSNs perhaps themost suitable type of networks for

taking advantageof these proto ols.

TheGeographi AdaptiveFidelity (GAF)[17℄ proto ol, inorderto put

nodesinto low-powersleep stateswithout ex essively in reasing thepa ket

lossrate, identiesgroupsofnodesthatareequivalent intermsofrouting

ostandturnounne essarynodes. Thisisa hievedbydividingthewhole

area into virtualgrids,small enough thatea h node ina ell an hearea h

node froman adja ent ell. Nodes thatbelongto the same ell oordinate

a tive and sleep periods, so that at least one node per ell is a tive and

routing delity (whi h requiresthat inany ell at anyone time thereis at

leastone node ableto perform routing [18℄) is maintained.

In [19℄, another distributed oordination proto ol for wireless ad-ho

networks, alled Span,is presented. The obje tive of the Span proto ol is

a-pa ity or the onne tivity of a multi-hop network. To a hieve this, Span

ele ts inrotation some oordinators that stay awake and a tively perform

multi-hop data forwarding, while the other nodes remain in power-saving

mode and he k whether they should be ome oordinators at regular

in-tervals. Coordinators form a forwarding ba kbone that should provide as

mu h apa ityasthe original network.

The Sparse Topology and Energy Management (STEM) proto ol

pre-sentedin[20℄isatopology ontrol proto olspe i allydesignedfor WSNs.

Theassumption of STEM isthat nodesina WSN may spend most of the

timeonlysensingthesurroundingenvironmentwaitingforatargeteventto

happen. Thus,unlikeother topology management s hemesthat oordinate

thea tivationofnodesduringthetransmissionphase,STEMoptimizesthe

energy e ien y of nodesduring themonitoring state, i.e. when no one is

sending data. STEM exploits the fa t that, while waiting for events, the

network apa ity an be heavily redu ed, thus resulting inenergy savings.

1.2 Dieren es between lassi al WSNsand

indus-trial WSNs

There are important dieren es between lassi al WSNs, whi h are

ad-dressedbythe proto olsdis ussed inSe tion 1.1, andtheindustrialWSNs

whi hareaddressedinthiswork. Aspreviouslymentioned,su hdieren es

involve both the requirements and the ar hite ture of the networks. The

mostrelevantaspe ts on erningthedierentar hite tureandrequirements

aredis ussed inSe tions1.2.1 and1.2.2, respe tively.

1.2.1 Ar hite ture

Classi al WSNs are independent deployments of ad-ho networks, whi h

typi ally run justone ollaborative monitoring appli ation. Theytypi ally

omprise a large number of nodes apable of monitoring a ertain

phe-nomenon (e.g. temperature, luminosity, et .), pro essing the relative data

and ex hanging it amongst themselves as well as with a base station via

a Sink node. The nodes ina WSN are generally lo ated in the proximity

of or inside the phenomenon they are monitoring. The environments

in-volved areoften remote or hostile to humans and insome ases thenodes

Figure1.1: Ar hite tureof a typi al industrialnetwork.

predi table. AWSNthereforehastobeautonomous, andableto ongure

itselfautomati allyand tofun tionwithouthumanintervention foraslong

as possible. Moreover, typi al WSNs annot rely on any other

infrastru -ture.

IndustrialWSNs,onthe ontrary,arealways oupled withwired

indus-trialnetworks, su haseldbusesorindustrialEthernet. Thereason isthat

wireless networksdiersubstantiallyfromwiredeldbuses intwo respe ts.

Firstly, a wireless hannel experien es mu h higher bit error rate than a

wired one. Se ondly, the wireless medium is shared with other networks,

thus itissubje tto external interferen es. Asaresult, itisnot always

fea-sible to repla ewired networkswith urrent wireless te hnologies. Rather,

industrialWSNsintegratewithwirednetworks,asthey angreatlyimprove

exibility and open new possibilitiesfor industrial appli ations. These

in- ludedeploymentofsensornodesinsettingswhererealizingawirednetwork

isnot feasibleor itwouldneed prohibitively expensivesafety erti ations.

AsshowninFigure1.1,typi alindustrialnetworksarehybrid andexhibita

hierar hi al ar hite ture,withoneormultiplewiredsegmentsandone

wire-lesssegment whi hisusedfortheless riti almonitoring and ontroltasks

and/or to inter onne t multiple wired segments. The main onsequen e

is that industrial WSNs do not need to be independent and autonomous

like lassi al WSNs. Rather, industrial WSNs an exploit the presen e of

both laten y andpredi tability.

1.2.2 Requirements

As dis ussed in Se tion 1.1, the most important requirement in typi al

WSNsisenergye ien y,followedbytheself- ongurationand

self-adapta-tion apabilities whi harerequiredinunattendeddeployments. Other

om-monrequirements are high s alability and low ost of thenodes. All these

hara teristi s areappre iated also inindustrialWSNs,espe ially

s alabil-ity. In fa t, large fa tories may in lude a very large number of nodesand

highnodedensity. Moreover, whilesu hnetworksshould overa largearea

the radio overage of sensor nodes is typi ally small. As a result, sensor

nodes must be able to perform routing in order to inter onne t multiple

wireless ells. However, in order to make WSNs suitable for fa tory

om-muni ation, thereareother requirements thathave tobemet.

Predi tability isprobablythemostimportantrequirement forindustrial

ommuni ations. Anindustrialnetworkshallprovidetoolsallowingtheend

user to simulate his network environment and determine in advan e

end-to-endperforman es ofthe systemsu h asend-to-endlaten y,therelevant

absolutejitterandnetworkthroughput. Forthisreason,anindustrialWSN

hastomake itpossibleto obtain (atleaststatisti al)upper boundsonthe

delivery timefor appli ation dataover the network.

Resistan e to the interferen es is also a major on ern. In fa t,

indus-trial WSNs operate in harsh environments with large metalli parts

(ma- hines) and should onsiderfa torslike hightemperature,dust, vibrations,

humidity,metalli surroundings,et . Thenetworkshouldtoleratepotential

interferen es andhigh variation ofthe radio signalstrength.

Finally,itisworthre allingthatindustrialWSNs annot ompletely

su-persedewiredfa tory ommuni ationsystems,be ausethey annot ompete

withwirednetworksintermsofperforman eandpredi tability. Rather,the

aimofindustrialWSNsisto omplement themandtoallowaexible

wire-less extension of preexisting wired networks. As a onsequen e, another

important requirement of industrial WSNs is the ability to integrate with

1.3 Resear h hallenges and possible solutions

All the above mentioned requirements represent resear h hallenges, to

whi h urrentliteraturehasprovidedonlypartialsolutions,ifany. Be ause

of the variety and the omplexity of su h requirements, it is not possible

to address all of them within one single ommuni ation proto ol. On the

ontrary, a suite of proto ols working at dierent layers is needed whi h

ollaborate toa hieve ommon goals. Apossiblesolutionistheappli ation

oftheDivide andConquer paradigm,whereea hlayeroftheproto olsta k

addresses just one requirement, or a few of them,while the areful

ombi-nationofmultiplete hniquesworkingatdierentlevelsleadstothedesired

results. Thisworkgoesinthatdire tion,providingdierentte hniquesand

proto ols working at dierent layers of the proto ol sta k and addressing

on e at a timethe requirements dis ussedin Se tion1.2.2.

Chapter 2 addresses the physi al layer, in parti ular therobustness of

IEEE802.15.4networksto ross- hannelinterferen e. The hapterprovides

a better understanding of ross- hannel interferen e in o-lo ated IEEE

802.15.4industrialnetworksandproposesageneralmethodologyforthe

as-sessment ofIEEE802.15.4performan eunderdierent ross- hannel

inter-feren e onditions. Thismethodologyallows anetworkdesignertoperform

on-site but a urate assessmentsand an be easily deployed in real

indus-trial environments to perform measurements dire tly in the

environment-under-test. Finally, a ase study based on COTS IEEE 802.15.4 devi es

is presentedto showhowto apply our methodology to a real s enario and

to dis uss theresultsobtained withone or multiple interferers and varying

some MAClevelparameters.

Chapter 3 addresses the s alability problem at the MAC layer. The

hapter proposes a novel multi- hannel approa h to the bea on ollision

avoidan e problem. Thenovelapproa h enhan es s alabilityof luster-tree

IEEE 802.15.4 networks while allowing ontention-free s heduling, thanks

to the use of multiple radio hannels in the same network. Moreover, a

Multi hannel Superframe S heduling (MSS) algorithm is presented that,

followingthemulti hannelapproa h, anoutperformthealgorithmsoered

by urrent literature, whi h usejustone hannel.

Chapters 4and 5 addresstheproblem ofredu ing energy onsumption

while introdu ing a predi table delay and follow an innovative approa h

that is based on a topology management proto ol whi h resides between

proto ol presented inChapter4 rules both the a tive/sleep y leof sensor

node, taking are of the energy e ien y,and datatransmission s hedule,

avoiding ollisions and ensuringthat thedelay introdu ed bythesleep

y- les is predi table. It also provides routing delity, but it follows a stati

approa h. Chapter 5 extends su h work, presenting a dynami topology

management proto olthatover omesthelimitationsofthestati approa h

introdu ing support for event-driven data transmissions and node joining

at run-time and providing a novel adaptive te hnique for energy balan ing

among nodes to further in rease network lifetime. The hapter provides

a detailed des ription of the dynami proto ol and simulation results on

network lifetimeand routingperforman e with omparative assessments.

Finally, Chapter 6 addresses predi table data delivery at the Routing

layer and integration between the industrial WSN and the wired

indus-trialinfrastru ture. Inparti ular,this hapterproposes anetwork

ar hite -tureanda ommuni ationproto ol, alledCir ularChainDataForwarding

(CCDF),thatnot only supports integrationwitha wiredindustrial

infras-tru ture, but also takes advantage of su h integration to deliver real-time

performan e, eventonodesthat ouldnotbedire tly overedbyasink. To

a hieve this goal, a hain-based me hanism is used, whi h integrates data

forwarding withthe hannela ess strategy. Theoreti al results, onrmed

by in-depth simulations, are provided to analyze the performan e of the

Assessment of ross- hannel

interferen e in IEEE 802.15.4

networks

TheIEEE802.15.4proto ol[21,22℄isgenerally onsideredasoneofthemost

promising options for low- ost low-power ommuni ations inindustrial

en-vironments [23℄. As industrial WSNs usually omprise a large number of

sensors and a tuators and typi al appli ations require small delays,

s ala-bilityis akeyissue[24℄. A viable solution issplitting a large network into

severalsmallernetworks,inter onne tedthroughawiredorawireless

ba k-bone. In order to support the requirements of industrial appli ations and

obtain reliable ommuni ations, the interferen e between thedierent

net-workshasto betakenintoa ount. A possibleoptionistheuseofdierent

radio hannelsforthedierentnetworks,thusimplementinga ellular

ar hi-te ture. Asimilarapproa h hasbeen presentedin[25℄. TheIEEE802.15.4

standard issuitable forthis solution, asthephysi al layer anuseupto 26

dierent radio hannels on three dierent bands (although the majorityof

Commer ial O-The-Shelf (COTS)IEEE 802.15.4 radios only support the

16 hannels dened on the 2.4 GHz band). However, when a similar

solu-tion is implemented, it isimportant to estimatethe ee t of ross- hannel

interferen e. Although in IEEE 802.15.4 there is no overlapping between

adja ent radio hannels, the work [26℄ shows that a tually some

interfer-en e is present, due to spurious emissions aused by the O-QPSK oding.

exper-imental results and theoreti al onsiderations on the oding of the IEEE

802.15.4 physi al layer. The te hnique des ribed in this hapter is based

on the work in [26℄, but extends it in several respe ts. While [26℄ mainly

dis ussestheresultsofmeasurementsperformedinaspe i IEEE802.15.4

deployment,here thefollowing ontributions areprovided:

•

Adis ussiononthe urrentbestpra ti esto opewith ross- hannel interferen e inIEEE 802.15.4networks, thatpinpointsthemain

lim-itations of su h approa hes.

•

Ageneri methodologyfortheevaluationof ross- hannelinterferen e betweenIEEE802.15.4networksinindustrialenvironments,whi h

al-lowsforon-the-ybuta urateon-siteassessments. Asthis

methodol-ogyreliesonlyonstandardIEEE802.15.4primitivesand omponents,

itisgeneri and easy to adoptinrealdeployments.

•

A ase study, whi h shows how to apply the proposed methodology to a reals enario. The asestudy platform, whi h isbasedon COTS

IEEE 802.15.4 devi es, is des ribed and theresults obtained are

dis- ussed.

This hapter is organized as follows. Se tion 2.1 gives an overview of

relevant literature. Se tion 2.2 introdu es theproblemof ross- hannel

in-terferen ein802.15.4networksandthe urrent bestpra ti essuggestedby

IEEE802.15.4hardware manufa turers. Se tion 2.3des ribesthe

method-ology proposed in this hapter and the asso iated testbed. Se tion 2.4

presents and dis usses the results of measurements performed on a ase

studyplatformbasedonCOTSIEEE802.15.4devi es. Finally,Se tion 2.5

gives some on luding remarks.

2.1 Coexisten e of wireless networks

Interferen e between wireless networks has been extensively addressed in

re entliterature. In2003,theIEEEpublishedado umentofre ommended

pra ti es[27℄inwhi htheproblemof o-existing 802.15.1and802.11b

net-works is analyzed through both simulations and analyti al models. The

problemofwireless linkassessment inindustrialenvironments isaddressed

worksexistwhi haddressinterferen einBluetoothnetworksusedin

indus-trial environments [29,30℄. Delay performan e and the pa ket loss

proba-bility aused by a number of o-lo ated interfering pi onets are analyzed

in[31℄and anupperboundon thepa keterror rateisanalyti allyderived.

In [32℄ the ee t of transient interferen e under TDMA proto ols is

eval-uated for dependability purposes. In [33℄ the impa t of an IEEE 802.15.4

network on an IEEE 802.11b one is studied. In[34℄ theinuen e of IEEE

802.11 on IEEE 802.15.4 is analyzed and a model to estimate the pa ket

error rate obtainable ininterferen e onditions is given. In[35℄ themodel

isextended,derivingthepa keterrorrateofIEEE802.15.4networksunder

ombined interferen e from WLANs and Bluetooth networks. Empiri al

evaluations of the o-existen e of IEEE 802.15.4 with IEEE 802.11,

Blue-tooth and mi rowave ovens are presented in [36℄. The work [37℄ assesses

the impa t of CSMA/CA parameters on the IEEE 802.15.4 performan e

inthe presen eofinterferen e oming from IEEE802.11, Bluetooth or the

same IEEE 802.15.4, but it emulates a simple industrial ontrol task to

evaluate appli ation-spe i performan e and does not aim at providing a

general methodto obtaina urate on-siteperforman eassessments. In

ad-dition, it does not deal with ross- hannel interferen e, as the interfering

IEEE 802.15.4networksare deployed in the same hannel. In [38℄, a

sim-ulator thattakesinto a ount oexisten eissuesbetween IEEE 802.11and

IEEE 802.15.4 is used to al ulate the pa ket error rate of both networks.

Con erning ross- hannel interferen e, various experimental studies exist,

whi h mainly fo us on the IEEE 802.11 proto ol family [39,40℄. In [41℄

the impa t of ross- hannel interferen e and other fa tors(su h asbea on

frames and overhead aused by both a ess points and WLAN adapters)

on theperforman e of IEEE 802.11g networks is experimentally analyzed.

In [42℄theauthors investigate the orrelationbetween spatial distan eand

hannelspa ing todealwithinterferen ebetween on urrent transmissions

in a multi hannel WSN. Their results, although interesting, are

hardware-spe i , asthey refer to a proprietary platform. Moreover, the authors do

not target areal industrial s enario, so their results arenot dire tly

appli- able to IEEE 802.15.4 industrial networks. No methodologies are given

to obtain appli ation-related gures, su h as pa ket error rate or laten y

values,through on-siteassessments.

Cross- hannel interferen e inIEEE 802.15.4networks is alsoaddressed

bysome appli ationnotes[43,44℄relevant tospe i devi es(Texas

addressthe re eiver jamming resistan e (i.e.,the degree to whi h

interfer-ers will impa t the re eiver) and quantify the re eiver performan e in the

presen e of interferers through interferen e reje tion measurements, whi h

showthe omplian e of theaddressed radioswiththeIEEE802.15.4

spe i- ations. However, all the measurements areperformedin lab, onne ting

the transmitter and the re eiver through ables and attenuators to

elimi-nateall theother sour es of interferen e. Furthermore, no in-air testing is

performedin[43℄, whilesome in-airassessment isoutlinedin[44℄, butit is

onlya roughestimation oftheinterferen e reje tionobtained withvarying

frequen yosets(

< 25

Mhzor

> 25

MHz,respe tively)betweenthedesired arrierandtheinterferer. Onthe ontrary,thework[26℄givesaninsighton

theee tsof ross- hannel interferen e ina spe i IEEE 802.15.4

deploy-ment, providing both analyti al results and experimental measurements.

Dierently from [26℄, in this hapter we provide a generi methodology to

a uratelyassessthe ee t of ross- hannel interferen einindustrial IEEE

802.15.4networks. Thanksto the ombination of des riptive statisti sand

error propagation theory, our methodology allows to obtain not only a

re-alisti performan e assessment of real industrial networks through on-site

measurements, but also the a ura y of pa ket loss and worst- ase PER

measurementsintermsof onden eintervals. Theproposedmethodology

is truly generi , as it only relies on a simple testbed that uses only

stan-dard IEEE 802.15.4 features and that an be easily deployed on-site in

industrialenvironments.

2.2 On ross- hannelinterferen e inIEEE 802.15.4

The IEEE 802.15.4physi al layerdenes three dierent radio bands, ea h

withadierentdatarateandadierent odingte hnique. Today,themost

widelyusedis the 2.4GHz band, whi h belongs to the ISMband. Sixteen

dierentdata hannelsaredenedaroundthe2450 MHzfrequen y,ea h of

themhavinga2MHzbandwidth. Thedistan ebetweentwoadja ent

han-nelsis 5 MHz. Nevertheless, be ause of theOset QuadraturePhase Shift

Keying (O-QPSK) modulation used at the physi al layer, a small fra tion

of thesignalis spread asspurious emissionoutside the5 MHz bandwidth,

as shown in [26℄. In order to limit ross- hannel interferen e, the IEEE

802.15.4spe i ations[21℄imposeatransmit powerspe traldensity(PSD)

devi e measured with a 100 kHz resolution bandwidth in frequen ies

dis-tantmorethan 3.5MHzfromthe enterfrequen yas20dB(relativeto the

peek)and -30dBm(absolute limit),respe tively. The IEEE802.15.4

stan-dard alsodenes the minimum jammingresistan e for there eiverso that

the Pa ket Error Rate (PER) is less than 1% as 0 dB for an interferer in

theadja ent hannel and 30dB for aninterferer inthealternate hannel 1

,

respe tively. A ording to the IEEE 802.15.4 standard, su h a jamming

resistan e should be al ulated using 20 byte pa kets with a desired

sig-nal powerof

−82

dBm and only oneinterferer. Thepro edure to ompute the jamming resistan e for an IEEE 802.15.4 trans eiver a ording to the

standardisdes ribedinsomeappli ationnotes,su has[43℄and[44℄,whi h

refertospe i devi es. In[44℄thejammingresistan eobtainedfromin-lab

measurements is used to al ulate the minimum distan e of the interferer

so thatthe PER keeps under1%. Thisrelation is obtained using thepath

lossequationto al ulate thepowerofthe desiredsignalgiven thedistan e

between thetransmitter andthere eiver. Then, using theinverse formula,

thedistan eof the interferer thatresultsinthe desiredjammingresistan e

valueisobtainedforthe giventransmitter/re eiverdistan e. We omputed

the jamming resistan e for the adja ent hannel as des ribed in [44℄,

us-ing three Maxstream XBee modules, equipped with the same trans eiver

as in [44℄. The interferer transmitted a ontinuous 2

modulated pattern of

pseudo-random data. Dierently from [44℄, we performed in-air

measure-mentsinareals enarioreprodu ingtheworking onditionstypi allyfound

in industrial ontexts and used the path loss equation in [21℄ to ompute

thea tual attenuationof thesignals,i.e.,

L

p

(d) =

(

40.2 + 20 log d,

d < 8m

58.5 + 33 log

d

8

, d > 8m.

(2.1)

Thedistan ebetweentransmitterandre eiverwasxedto2m. Theresults

of ourmeasurements, giveninTable2.1,show thatthejamming resistan e

in reases with the distan e between the interferer and the re eiver. In all

our measurements the obtained jamming resistan e is far better than the

minimumvalueof0 dBimposedbythestandard. Inthe ase of1.5m

dis-tan e, wewerenot ableto al ulatethe exa tvalue,astheobtained pa ket

1

Theadja ent hannelis oneoneitherside ofthe desired hannel that is losest in

frequen y to thedesired hannel,and the alternate hannelis onemore removedfrom

theadja ent hannel[21,22℄.

2

Interferer Distan e (m) 1.50 1.25 1.00 0.63 0.50

Jamming Reje tion(dB)

>

23 23 19 15 8

Table2.1: In-airjammingresistan eobtainedwith2mdistan efrom

trans-mitterto there eiver.

errorrate(PER)waslessthan1%evenwiththemaximuminterfererpower.

Thismeansthatthejammingreje tionwas ertainlyhigherthanthe

23

dB value obtained with a 1.25 m distan e from the interferer. These results

also show that there is a signi ant dieren e between thejamming

resis-tan evaluesobtained throughin-labmeasurements,shownin[44℄,andthe

ones measured on site. We on lude that urrent best pra ti es that use

in-labjammingresistan eandthepathlossformulato obtaintheminimum

distan ebetween the PERand theinterferergive onlyarough information

tothenetwork designer. For thisreason, itisadvisabletoperform testing

in the real working s enario under realisti onditions. However, to

per-form on-site a urate assessments on ross- hannel interferen e, a suitable

methodologyhastobe arefullydevisedandthe orrespondingexperimental

testbed hastobedeployed. Thisisexa tlythemain ontributionprovided

bythis hapter.

2.3 Testbed and Methodology

The approa h proposed in this hapter requiresa simple testbed made up

of portableand aordable omponents. The testbed onsists ofa personal

omputer

(P C)

, in harge of ontrolling the transmitter

(T )

and re eiver

(R)

nodes through a serial onne tion, and one or more interferer nodes

(N

i

)

ongured in su h a way to autonomously send frames on dierent hannels at the same time. An auxiliary re eiving antenna onne ted to a

portablespe trumanalyzer

(S)

,ifavailable,maybeusefultodete texternal sour es of interferen e. Su h a testbed is generi , as it does not require

either a parti ular kind of radio modules or a spe i environment, asno

assumptions on the environment are made (e.g., on the presen e/absen e

of obsta les,on their shape,material, et .). It is possible to deploy su h a

testbed using anyIEEE 802.15.4 COTS modules, as long as they support

thestandard IEEE802.15.4primitives.

Figure2.1: Stru ture ofthetestbed.

reside through a USB or RS232 port and an send ommands to either

modifythenetworkparameters orsenddataframesorreadre eivedframes.

Asintypi alindustrials enariosthepresen eofperiodi interferingpa kets

is a realisti assumption [45℄, in our testbed interferer nodes periodi ally

transmit the same pa ket for the duration of the measurement ampaign,

without theneed to atta h a PCto theinterferernodes.

2.3.1 Methodology

The hoi eoftheparameters tobetakeninto a ountinthemeasurements

isbasedonthesensitivityassessmentsmadein[26℄,wherethesensitivityof

thetestbedtotheRSSIvaluereturnedbytheIEEE802.15.4moduleversus

distan e and the pa ket lossratio versus interferen e power level were

an-alyzed. The results obtained showed that theexperien edRSSI values are

dire tly relatedto the distan e and arealso quitestable, asthe oe ient

of variation wasbelow2%inalmostall theperformedmeasurements. This

agreeswithotherstudiesonthe hara terizationofIEEE802.15.4link

qual-ityandsignalstrength,su has[46℄. However,in[26℄itwasalsoshownthat

RSSIisnot agoodindi atorofthelinkqualityinnoisyenvironments, asit

doesnot distinguishbetween the signaland interferen e power. Moreover,

manda-tory and thus drives the WSN design hoi es. As a result, reliability and

timelinessarethe ru ialrequirementsto be taken into a ount. For this

reason,theperforman eindi atorsadoptedherearelaten y,pa ketlossand

worst asepa keterror rate(PER). They an be obtained asfollows:

Laten y estimation

When dealing with wireless industrial ommuni ations, given the typi al

time- riti alrequirements ofthe ex hanged tra , laten yis animportant

parametertobeassessed. Anestimateoftheone-waylaten iesofdataframe

transmissions an be obtained by omparing the logs of sent and re eived

frames. To guarantee thetemporal oheren e oftimestamps, the

measure-ments have to be performed on the same PC, therefore with a ommon

lo k referen e. Another important detail to be onsidered when

evaluat-inglaten ies isthat, asthetransmitterand re eivermodulesare onne ted

to the PC through a serial onne tion, an additional laten y is introdu ed

in both the transmission and the re eption of a frame. Asthe amount of

datato be transmittedisknownandthereisno ontention forthemedium

a ess,this delay anbeestimatedand subtra tedfromtheone-waydelay.

Inparti ular,ifa

L

data

o tetdataframehastobetransmittedthroughthe wireless onne tion, and a

L

ov

o tet overhead isneeded to send the trans-mission (or re eption) ommand, the time spent for the transmission (or

there eption) ofa frameovertheserial linkis

T

RS232

=



8 (L

ov

+ L

data

)

L

byte



(L

start

+ L

byte

+ L

parity

+ L

stop

)

D

RS232

(2.2)

where

L

byte

is the number of bits in every frame of the RS232 proto ol,

L

start

,

L

parity

and

L

stop

are the number of start, parity and stop bits

re-spe tively,and

D

RS232

isthebaudrateoftheserial onne tion. Considering thatthe propagation time an be negle ted, the laten y an be al ulated

as

T

f rame

= t

rx

− t

tx

− T

RS232

rx

− T

RS232

tx

(2.3) where

t

rx

and

t

tx

arethetimeinstantsoftheframere eptionand transmis-sion,respe tively,while

T

RS232

rx

and

T

RS232

tx

are theoverheads for trans-mitting and re eiving a frame,respe tively. However, thedelay al ulated

with(2.3)in ludessomeoverheadsintrodu ed bytheoperatingsystemand

ommuni ation ontrollers. To limit su h ajitter, itis advisableto redu e

theproperdatastru turestotra kthesendingandre eivingofdataframes,

so that the jitter aused by blo king I/O fun tions is avoided. Moreover,

when a very high degree of a ura y indelay measurements is required,it

is advisableto run thesoftwareundera real-time kernel.

Pa ket Loss estimation

In our testbed, experiments are run by repeatedly sending pa kets from

the transmitter

T

to the re eiver

R

and ounting the times a pa ket sent by

T

is not re eived by the re eiver

R

. Suppose that, given a dened transmitting power and a dened kind of interferen e, ea h pa ket has a

xed probability

(1 − P L)

to be su essfully re eived by the destination, andaprobability

P L

tobelost. Thisassumption anbe onsideredrealisti inawellair- onditionedenvironment withnomovingobsta les [46℄. Under

this assumption the pa ket lossevent will happen a ording to aBernoulli

distribution, where the

P L

parameter representsthe probability to have a pa ketloss.

Thebestapproximationofthe

P L

probabilityisgivenbythesamplemean

d

P L =

1

n

P

n

i=1

X

i

, where

n

is the number of pa kets transmitted in the whole experiment and

X

i

arethe resultsof a singlepa ket transmission(1 means that the pa ket has been lost, 0 means that the pa ket has been

su essfully re eived). Moreover, ifthe numberof pa kets that aresent in

ea h experiment is large, the onden e bounds for

P L

an be obtained through theformula

P L = d

_{P L ± z}

1

−

α

2

r

c

P L

(1 − c

P L

)

n

(2.4) where

z

1

−

α

2

is the z-s ore of the standard normal distribution that

deter-mines thedesiredinterval of onden e [47℄,e.g.,1.96 for 95% onden e.

Worst Case PER estimation

To obtain the worst- asepa ket error rate, a onstant ross- hannel

inter-feren e should be onsidered. As itis fully des ribed in[26℄, even withan

interferer node thattransmitsdatapa kets periodi ally,ourtestbed makes

itpossible,underproperassumptions,toapproximatelyassessthePER

un-der onstantinterferen e onditions. ConsideringanIEEE802.15.4network

Figure 2.2: Modelfor overlapping transmissionprobability.

node,

L

i

theinterferer framelengthand

L

p

the length ofthepa ketweare interested in, su h that

L

p

≤ L

i

, and

L

i

≪ T

i

. Referring to Figure 2.2, a pa ket

p

doesnotoverlapwithapa ketoftheinterfererif

L

i

< t < T

i

− L

p

, where

t

isthe arrival timeof

p

. Therefore, theprobabilitythat no overlap will o ur between these pa ket is

(T

i

− L

p

− L

i

) /T

i

. So, the probability thatapa ket willoverlap withaninterferer dataframe is

P (C) =

L

p

+ L

i

T

i

.

(2.5)

Let

L

be the lost pa ket event and

C

the ollision event. Assuming

L

as our event, and

C

together with any other ause than a ollision as our set of mutually-ex lusive and all-in lusive auses of the event, we

an al ulate the PER using the Bayes theorem. Under the assumption

thatevery transmission overlap auses a ollision event, irrespe tive of the

fra tionofpa ketoverlapping, we have

PER = P (L|C) =

P (C|L) · P (L)

_{P (C)}

.

(2.6) InFormula(2.6) ,

P (L)

is exa tlythepa ketlossobtained throughour measurements,

P (C)

istheprobabilityobtainedin(2.5)and

P (C|L)

repre-sentstheprobabilityofa pa ketbeinglostbe auseofa ollisiongiven that

thepa ket is lost. A pa ket lossmay be due to eithera ollision with the

interferernodeoradierent ause(anythingotherthana ollision). We an

assessthepa ketlossratioobtainedinthesame onditionsbutwithoutany

interferernode,namely

PL

0

,and al ulate

P (C|L)

as

1 − PL

0

. If

PL

isthe pa ket lossratioobtained withthoseparameters and

P L

0

thepa keterror

rate obtained without anyinterferer, theworst asePER, i.e., thePER in

the aseapa ket ollideswithaninterfererpa ket, anbeapproximated as

PER =

T

i

(1 − PL

0

) PL

L

p

+ L

i

.

(2.7)

Theestimationoftheworst asePERforagivens enario an beuseful

in ontextswhere a dened reliability hastobemaintained,su h as

indus-trial automation. However, inorderto be useful,even these results should

in lude the onden e intervals. As there are two dierent parameters in

(2.7) that arederived from measurements, theerror propagationhas to be

al ulated using the error propagation theory. As an impre ision in

P L

0

may also ae t the measurements of

P L

, it is safeto usethe onservative estimation ofthe onden e intervalfor aprodu t,given bythesum ofthe

relative onden e intervals of thetwo fa tors[47℄. As a result, if

u

c

(P L)

and

u

c

(P L

0

)

are the onden e intervals for

P L

and

P L

0

respe tively, a onservative estimation ofthe onden e intervalis

u

c

(P ER) =

T

i

L

p

+ L

i

[P L · u

c

(P L

0

) + u

_c

(P L) · (1 − P L

0

)] .

(2.8) In order to assess the ee tiveness of our methodology, we ran some

experiments using our testbed. The experimental results obtained, as it

will be shown in the ase study addressed in Se tion 2.4, are ompliant

withour estimations a ording to (2.7)and (2.8) .

2.4 Case study and experimental results

Using our testbed,abroad seriesof in-airmeasurements toexperimentally

assess the impa t of ross- hannel interferen e under dierent operating

onditions an be run. In the following, the methodology proposed in the

previous se tion is explained through a ase study. Several test s enarios

werebuiltinordertoreprodu ethetypi alworking onditionsofindustrial

environments. Results obtained in these s enarios with one or multiple

interferers will bepresented.

2.4.1 The IEEE 802.15.4 platform

In our ase study, measurements were performed using the MaxStream

standardspe i ationsandworkex lusivelywithinthe2.4GHzISMband.

Both these two types of modules are equipped with a MC9S08GT60

mi- ro ontroller and an MC13193 802.15.4 RF trans eiver. They are

pin- ompatible,soforthe onne tionwiththePCthesamedevelopmentboards,

i.e.,MaxStream XBIB-U-DEVs andMaxStream XBIB-R-DEVs,have been

used. Theonlydieren ebetween thesemodulesisthetransmittingpower,

whi hisupto0dBmfortheXBeemodules,whileitisupto18dBmforthe

XBeePro ones. The original XBeermware(ver. 10A5) inAPImode[48℄

was usedin the transmitter and the re eiver node, while for the interferer

we developed a ustomized rmware using the Frees ale Codewarrior for

HC(S)08, the implementation of IEEE 802.15.4 provided by the Frees ale

Beekit andthe XBeeDevelopment Toolkit publi ly available in[48℄.

How-ever,whentheFrees aleIEEE802.15.4implementationisusedontheXBee

Pro modules,the maximumtransmitting powerdoesnot oin idewiththe

one of 18 dBm obtainable using the original rmware. For this reason our

ustomized rmwarewasrunonly whenthe ontinuoustransmitmodewas

needed,whileinalltheother asestheoriginalXBeermwareintheT

rans-parent Operationmode wasused.

To oordinate the IEEE 802.15.4 wireless nodesand thePC, aspe i

softwarewasdeveloped. Thesoftwareallowsustosetvariousparameters of

the nodesthatmake up thetestbed (i.e.,transmissionperiod, datapa ket

size, hannel, presen eof the interferer, et .), aswell asto drive the

trans-mitternode andmonitorthe tra of ageneri re eivernode. Aspe trum

analyzerisusedformonitoringpurposes,toensurethatnointerferen efrom

un ontrolled wireless devi es o ur duringour measurement ampaigns.

In all the experiments the interferer nodes transmit periodi pa kets,

while

T

transmits pa kets almost periodi ally, i.e. with an interarrival timeof

100 ± δ

mswhere

δ

isarandom value hosenintheinterval

[−5, 5]

, introdu edtoavoidtheo urren eofrepetitivepatternsofinterferen e. On

the other hand, no jitter was expli itly added to the interferer period, to

keep a xed ollision probability. The default settings of all the nodes in

our testbed, when only one interferer is present, are shown in Table 2.2.

Both transmitter and interferer nodes always use the non bea on-enabled

mode. The 16-bit addressing mode is used, soa 17 byte header hasto be

added to the payload shown in Table 2.2. If not stated otherwise, the

T

and

R

nodesarexed1mapartfromea hother,whiletheinterferernodes areinthe middle, at adistan e of 0.5mfrom

R

. No obsta les arepresent between nodes. All the experiments omprise a large number of samples

Transmitter Re eiver Interferer

TXpower 0dBm 0dBm 0/18dBm

CCAThreshold 44dBm 44dBm 44dBm

ma MinBe 0 0 0

Channel 11 11 12

Tx. Period 100ms n.a. Variable

Jitter 5ms n.a. No

Payload 30bytes n.a. 100bytes

ACKs No No No

Table 2.2: Basi Testbed onguration

(3000pa kets sent byT, ifnotspe ied dierently)and wereperformedin

a real-life indoor environment. We tried to minimize all theother sour es

of interferen e, e.g. fromWLANs operating nearby, by shutting down any

ele troni equipment under our ontrol apable of emitting radio waves in

nearby areas. Moreover, we monitored theenvironment through a Wi-Spy

2.4x portable spe trum analyzer, in order to assure that no interferen e

from un ontrolled wireless devi es o ur duringour experiments.

2.4.2 Preliminary Assessments

In order to verify that the obtained results will not be ae ted by

hard-warefailures or imperfe tions, it isimportant to perform preliminary

test-ing of the testbed omponents. Several omponents may lead to biased

results, e.g., pa ket loss in the serial line onne ting the PC to either

the transmitter or the re eiver, imperfe tions on the trans eivers (or

non- omplian e tothe IEEE802.15.4standard)orevendierent orientationsof

non-omnidire tional antennas.

In our ase study, we used XBIB-R-DEVboards onne ted to the PC

throughaUSB-to-serialadaptorfeaturingaPL-2303HX hipsetand

XBIB-U-DEV boardsdire tly onne ted to thePC througha USBport. Inboth

ases,theserial onne tionwastestedbytransmitting 10000 pa ketsinthe

bestpossible onditions for the wireless hannel, i.e., Tand Rwerepla ed

at 1mwithno obsta lesinbetween andwithout anyinterferer. Theywere

set to use a 0 dBm transmitting power and a knowledged transmissions,

and the spe trum analyzer was used to verify that no other interferen e

o urred duringthe test. In su h onditions, therewasnopa ketloss.

ThetestingoftheMC13193trans eiverembeddedintheXBeeandXBee

tothestandardspe i ationsofourdevi es,intermsofboththePSDmask

and jamming resistan e. The results in terms of jamming resistan e were

alreadydis ussedinSe tion2.2. ThePSDmaskwasmeasuredinairsetting

the100kHzresolutionbandwidthasindi atedin[21℄,withbothanAnritsu

MS2668C and a Wi-Spy 2.4x portable spe trum analyzer. The output of

latter is shown in Figure 2.3a. In that gure it is easy to noti e that, for

frequen iesdistant 3.5 MHz or more from the arrier, themeasured signal

neverex eedsthe -20dBmrelative thresholdneitherthe-30dBm absolute

one. As a result, even the trans eiver su essfully passed the omplian e

test.

Two dierent typesof antennas were usedinour testbed, i.e. standard

SMA- onne torized monopoleantennas and integrated whip monopole

an-tennas. Inparti ular,a standardSMA- onne torized antennawasusedfor

there eiver, while the transmitter and theinterferer nodeswereequipped

withtheintegratedwhipantennas. Themeasuredradiationpatternofboth

typesofmonopoleantennasusedarepubli lyavailableon[49℄andare losed

to theideal ones, i.e., they are almost omnidire tional (rippleof

±10

dB), inthe monopoleH-plane asexpe ted. Howeversin e manyexternalfa tors

may inuen e the radiation pattern, we performed our pattern

measure-ments in our testbed environment. We used XBee modules for both the

transmitterand there eiver. Thetransmitter nodewaspla edat thesame

height, but 2 meters away from the re eiver. The transmitting power was

set to

−2

dBm. The re eiver was kept xed, while the transmitter angle was hanged in steps of 5 degrees s anning the antenna H-plane. For ea h

angle,100 samplepa ketsweresent,andtheaveragevaluewastaken. The

re eived poweris depi ted inFigure2.3b, whi h showsthat:

1. with the same nominal transmitted power, the re eived power level

(RSSI) was slightly higher when a standard SMA- onne torized

an-tenna wasused;

2. bothtypesofantennashaveradiationpatternsthat,withafairlygood

approximation, an be onsidered omnidire tional (ripple of about

±6

dB).

Thelast result isquite relevant, as it indi ates that small anglevariations

that might be introdu ed by rotating the interferer nodes do not have a

XBee Pro

XBee

(f-f

c

) MHz

0

-1

-2

-3

-4

1

2

3

4

(a) Power Spe tral Density of XBee

andXBeePromodules

0

−10

−20

−30

−40

−50

₀

30

60

90

120

150

180

210

240

270

300

330

SMA Antenna

Whip Antenna

θ (degrees)

Normalized

RSSI (dB)

(b) Radiation diagrams (re eived

powerunderdierentanglesinthe

H-plane)

Figure 2.3: PreliminaryAssessments

2.4.3 Worst Case PER validation

After the preliminaryassessment,some experiments were run through our

testbed to assess the ee tiveness of the model for the estimation of the

worst asePER.Thedefault ongurationofour testbedwasusedwithan

XBee Pro asthe interferernode. Theperiodof the interfererwas hanged,

and for ea h value both the experien ed

P L

and the expe ted

P ER ±

u

c

(P ER)

wereobtained a ordingtoformulas(2.7) and(2.8) . Inaddition, toexperimentallyassesstheworst asePER,weusedourmodiedrmware

that sends ontinuously a data frame, so that the hannel utilization is

lose to the worst ase, i.e., 100% hannel utilization. The results of this

experimentareshowninFigure2.4,wheretherstvalue(markedasnone

on thex-axis) is the one experien ed (i.e.,measured) withoutinterferen e,

while thelast one (marked as ContinuousTX on thex-axis) isthe value

experien edwith ontinuoustransmissionsfromtheinterferer(about98.8%

duty y le). Noti e that, in thelatter ase, no expe ted PER is given, as

the worst ase PER oin ides with the PL experien ed with ontinuous

interfering transmission. Figure 2.4shows thatthe numberof lost pa kets

in reases with the de reasingperiodoftheinterferer node. Thisis be ause

None

100

80

60

40

30

20

10

Continous TX

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

Period of the Interferer node (ms)

Probability

Packet Loss

Expected PER

Figure2.4: Pa ketLossand expe tedPERversus thevaryingtransmission

periodof theinterferernode.

in reases. However,theaveragePERvaluesareverysimilarinallthetrials.

Moreover, theanalyti al PER mat hes theexperimental one. Thisgives a

signi ant eviden e of the ee tiveness of the model we used to al ulate

the worst ase PER, although the size of the 95% onden e interval is

larger when the period of the interferer is large. This is expe ted, as the

onden eintervalisproportional to theinterfererperiod(

T

i

).

2.4.4 Interferen e by a single node

To assess the level of interferen e on a ommuni ation aused by an

in-terferer working on an adja ent hannel we used a simple s enario with a

transmitter,are eiverandaninterfererlo ated1 mapartfromea hother,

ea h of them being the vertexof an equilateral triangle with a side of one

meter. The onguration of the transmitter and re eiver nodes is that in

Tab.2.2, where an interferer node transmits apayload of 100 byteswith a

onstant period of 100 ms. Both the transmitter and theinterferer belong

totheXBeefamilyandtheirtransmissionpoweris0dBm. Sixexperiments

were run, inwhi h the transmitting hannel of thedisturbing node is

var-ied. Theresults,showninFigure2.5(witha95% onden einterval),show

that,althoughthe power ontributiononthe adja ent andonthefollowing

hannel isa verysmall fra tionofthatemitted bytheinterferer node, itis

enoughtodetermineanon-nullpa ketloss,whi hmeansthat ross- hannel

interferen eisnon-negligible. Wehighlightthatto al ulatesu h onden e

12

13

14

15

16

0

1

2

3

4

5

6

7

x 10

−3

Packet Loss

Radio channel of the interferer

Figure2.5: Pa ketLoss withequidistant nodes.

very lowobserved proportions[50℄,asinthe aseof theresultsobtained in

this experiment. For thisreason,the95% onden eintervalsinFigure2.5

wereobtained througha dierent method,givenin [50℄,i.e., thelowerand

theupperboundsare al ulatedas

(A −B)/C

and

(A + B)/C

,respe tively, where

A = 2 · n · d

P L + z

1

2

−

α

₂

,

(2.9)

B = z

1

−

α

2

r

z

2

1

−

α

2

+ 4 · n · d

P L(1 − d

P L),

(2.10)

C = 2(n + z

2

1

−

α

2

).

(2.11)

The expe ted value of the worst ase PER was al ulated using equation

(2.7) , andtheresults areshownin Figure2.6. Here we an noti ethat the

ee t of ross- hannel interferen e learly depends on the hannel of the

interfering node. This is an expe ted result, as spurious emissions of the

interfering signalde rease withthe hanneloset. However, aslong asthe

energy re eived bythe re eiverfrom thetransmitter and interferer node is

similar, only alimited pa ketlosso urs. Inthis ase theworst asePER

isalwayslowerthan4.5%,thatex eedsthe1%imposedbythestandardfor

0dBjammingreje tion. We underlinethat, asour purposeherewasnotto

assess the jamming resistan e, we did not use 20 byte pa kets as foreseen

in the IEEE 802.15.4 standard, but 47 byte pa kets (payload=30 bytes,