H2020-GANo.645274 D2.4

1

Wireless Software and Hardware platforms for Flexible and Unified radio and network controL

ProjectDeliverableD2.4Resultsofsecondsetofshowcases

Contractualdateofdelivery: 31-12-2016

Actualdateofdelivery: 23-12-2016

Beneficiaries: IMEC,TCD,CNIT,TUB,NCENTRIC,RUTGERS,SNU

Leadbeneficiary: TCD

Authors: DiarmuidCollins(TCD),MaiconKist(TCD),AlextianBartholomeuLiberato(TCD),IngridMoerman(IMEC),PeterRuckebusch(IMEC),SpiliosGiannoulis(IMEC),PieterBecue(IMEC),AnatolijZubow(TUB),MikolajChwalisz(TUB),PiotrGawłowicz(TUB),IleniaTinnirello(CNIT),PierluigiGallo(CNIT),DomenicoGarlisi(CNIT),RobinLeblon(NCENTRIC),SvenZehl(TUB,ChangmokYang(SNU),SunghyunChoi(SNU)

Reviewers: MitchDeGeest(NCENTRIC)andIvanSeskar(RUTGERS)

Workpackage: WP2–GeneralRequirementsandShowcases

Estimatedpersonmonths: 6

Nature: R

Disseminationlevel: PU

Version: 1.0

Abstract:This public deliverable reports on the results of the second set of showcases that have beenimplemented.ItalsoincludesthespecificationsofthethirdsetofshowcasestobeimplementedbytheendofYear3.ThisdeliverablereportsonactivitiesinWP3,WP4,WP5andWP6thatarerelatedtotheshowcases.

Keywords:Showcases,proof-of-concept,useofUPIs

H2020-GANo.645274 D2.4

2

Executive Summary This deliverable reports on the showcases implemented in the second year of the project, whichexhibitthefunctionalityoftheWiSHFULplatform.Eachshowcaseintroducestheproblemitaimstoaddress, presents an overview of the showcase demonstration, howWiSHFUL UPIs are used, theresultsorexpectedresults (forshowcasesbeing implemented),andfuturesteps.Furthermore, theworkoutlinedhereinprovidesthebasisofupcomingscientificpublicationsandotherdisseminationmaterial as a means of maximizing the impact of the project. Showcases in this deliverabledemonstratethefollowing:

• TheLTE-LAA-Wi-Ficoexistencewillshowhowamanaged802.11Wi-FinetworkmakesuseofWiPLUS to passivelyestimate the airtimeoccupied by LTE-U at eachAP in order to performbetter channelassignmentand to load-balanceclient stationsacross theAPs formaximizingtheoverallthroughput/servicequality.

• The GNU Radio showcase aims to demo the execution of two independent virtual radiossharingoneRFfront-endsimultaneously.

• The IRIS demo shows the integration with UPIs by changing the parameters on the fly atUSRPs.

• The Coexistence of IEEE 802.15.4e TSCH with IEEE 802.11 networks showcase is ademonstration and proof of the solution feasibility and applicability for a cross technologysynchronizationschemebetweenTSCHandWi-Finetworks.

• TheRadioSlicingforVirtualizedHomeWi-FiAccessPointsiscurrentlybeingimplementedwithanovelslicingtechnologytoguaranteebandwidthandtrafficisolationfortheprimaryusersofthehomeAPinuplinkanddownlink.

• MACadaptation inpresenceof legacystationsextendsresultspresented inD2.3toconsiderthe possibility that some wireless nodes cannot be directly controlled by WiSHFUL. ThisshowcaseprovesthattheoptimizationlogiccanbeimplementedonWiSHFULprogrammablenodeswithorwithoutpresenceoflegacynodes.

• MCSSelection shows that lowering frame loss inhighlymobilenetworksdemonstrateshowdynamicreconfigurationofframeaggregationandPHYrateadaptationparameterscanlowertheframeloss.

• The link estimator selection shows that detecting the optimal link estimation algorithm invarious network topology scenarios can increase the overall network performance bydynamically selecting the optimal link estimation algorithm supported by a global controlprogramandcontrollingasensornetwork.

• Multihop load awareMAC adaptations proposes and implements a very promising schemecalledREACTformitigatingtheperformanceimpairmentsofCSMA/CAprotocolsinmulti-hoptopologiesbasedonthedynamicadaptationofthecontentionprocessexperiencedbynodesinthewirelessnetwork.

Technicaldetailregardingtheimplementationofsupportforeachshowcaseislefttotheappropriatetechnicaldeliverables(D3.4,D4.4andD6.4).Showcasesareusedtopromotetheutilityoftheprojectby providing convincing scenarios that useWiSHFUL technologies and infrastructure for advancedwirelessresearchandexperimentation.Theyalsoactaseducational,trainingandtutorialmaterialtosupport third parties using developed WiSHFUL UPIs. In combination with D2.2, D2.3 and theupcoming D2.5, this deliverable provides a complete picture of WiSHFUL capabilities andeffectivenessbyapplyingthetechnologiesdevelopedbytheprojecttoareasofactiveresearchandproblemsrelevanttothecommunityinauser-friendlyandconsistentmanner.Finally,thisdocumentalsodefines,atahigh-level,alistofintelligenceshowcasestobeimplementedwithinthethirdyearoftheproject.Theseinclude:

• Overtheair(OTA)updatesusingGITARforWSNs

H2020-GANo.645274 D2.4

3

• Extension of MAC adaptation in multi-hop topologies based on directional antenna andmultiplepathreservations

• Radio-basedindoorlocalization• ExtensionofMACoptimizationsinhigh-densityscenarios,withonlinephase• Interference classification for Wi-Fi nodes on the basis of error patterns, using machine

learning• Radiovirtualizationwithsimultaneoustransmissionandreception• IEEE802.11OverlappingBSSmanagement• Closed-loopratecontrolforIEEE802.11infrastructurenetworks• ContextAwarenessinspectrummanagementsystem-aidedSUnetworks

H2020-GANo.645274 D2.4

4

List of Acronyms and Abbreviations

5G FifthGeneration

A-MPDU AggregateMPDU

AP AccessPoint

API ApplicationProgrammingInterface

AGC AutomaticGainControl

BSS BasicServiceSet

BlockAcks BlockAcknowledgements

CSMA CarrierSenseMultipleAccess

CCA ClearChannelAssessment

COM ComponentObjectModel

CSI ChannelStateInformation

CWopt CWOptimum

CPS Cyber-PhysicalSystems

DDOS DistributedDenialofService

DSL DigitalSubscriberLoop

DSR DynamicSourceRouting

EWMA ExponentialWeightedMovingAverage

ISM Industrial,ScientificandMedical

IEEE InstituteofElectricalandElectronicsEngineers

IoT InternetofThings

IP InternetProtocol

LQI LinkQualityIndicator

LQE LinkQualityEstimator

LTE LongTermEvolution

LBT ListenBeforeTalk

LSA LicensedSharedAccess

LR-WPANs

Low-RateWirelessPersonalAreaNetworks

LTE Long-TermEvolution

LTE-U LTEUnlicensed

MAC MediumAccessControl

MCS ModulationandCodingScheme

MCDs Measurement-CapableDevices

MEDCA ModeratedBackoff

MPDU MACprotocoldataunit

H2020-GANo.645274 D2.4

5

MSDU MACservicedataunit

NB-IoT NarrowBand-IoT

NRDMS NumberofRPLDIOmessagessent

NRPS NumberofRPLparentswitches

NTP NetworkTimeProtocol

OTA OverTheAir

PDR PacketDeliveryRatio

PLCP PhysicalLayerConvergenceProtocol

PTP PrecisionTimeProtocol

PUs PrimaryUsers

REM RadioEnvironmentMaps

ROT Radio-onTime

RR RetransmissionRatio

RSSI ReceivedSignalStrengthIndicator

RRH RemoteRadioHead

RF RadioFrequency

SFER SubframeErrorRate

SFs SlotFrames

SDR SoftwareDefinedRadio

SSIDs ServiceSetIdentifiers

SAS SpectrumAccessSystem

SMS SpectrumManagementSystems

TSCH Time-SlottedChannelHopping

TSF TimeSynchronizationFunction

Tx Transmitter

USRP UniversalSoftwareRadioPeripheral

VNFs VirtualizedNetworkFunctions

Wi-Fi WirelessFidelity

H2020-GANo.645274 D2.4

6

Table of contents

1 Introduction .......................................................................................................... 9

2 Results of the second set of Showcases implemented in Year 2 .................. 102.1 LTE-U-Wi-Ficoexistence..................................................................................................10

2.1.1 PresentationofUPIUsed............................................................................................................112.1.2 Application..................................................................................................................................112.1.3 Results.........................................................................................................................................122.1.4 NextSteps...................................................................................................................................12

2.2SDRCapabilities................................................................................................................122.2.1 GNURadioCapabilities:RadioVirtualization..............................................................................132.2.2 IRISRadioCapabilities.................................................................................................................14

2.3CoexistenceofIEEE802.15.4eTSCHwithIEEE802.11networks........................................162.3.1 Assumptions................................................................................................................................182.3.2 Possiblecoexistenceschemes.....................................................................................................182.3.3 Followedapproach......................................................................................................................192.3.4 Systemdesign.............................................................................................................................192.3.5 PresentationofUPIusedandnew..............................................................................................232.3.6 Results.........................................................................................................................................242.3.7 NextSteps...................................................................................................................................25

2.4RadioSlicingforVirtualizedHomeWi-FiAccessPoints......................................................252.4.1 PresentationofUPIUsedandNew.............................................................................................262.4.2 Application..................................................................................................................................262.4.3 NextSteps...................................................................................................................................26

2.5MACadaptationinpresenceoflegacystations.................................................................272.5.1 PresentationofUPIusedandnew..............................................................................................312.5.2 Results.........................................................................................................................................332.5.3 Nextsteps....................................................................................................................................35

2.6 Loadandtopologyawarenetworking...............................................................................352.6.1 MCSSelection.............................................................................................................................362.6.2 Linkestimatorselection..............................................................................................................392.6.3 MultihoploadawareMACadaptations......................................................................................51

3 Definition of Showcases to be implemented in Year 3 ................................... 633.1OTAupdatesusingGITARforWSNs..................................................................................63

3.1.1 Overview.....................................................................................................................................633.1.2 Goals............................................................................................................................................63

H2020-GANo.645274 D2.4

7

3.1.3 Breakthroughs.............................................................................................................................633.1.4 Methodology...............................................................................................................................643.1.5 UseofWiSHFULFunctionality.....................................................................................................64

3.2Extension ofMAC adaptation inmulti-hop topologies, based on directional antenna andmultiplepathreservations................................................................................................643.2.1 Overview.....................................................................................................................................643.2.2 Goals............................................................................................................................................643.2.3 Breakthroughs.............................................................................................................................653.2.4 Methodology...............................................................................................................................653.2.5 UseofWiSHFULFunctionality.....................................................................................................65

3.3Radio-basedindoorlocalization........................................................................................653.3.1 Overview.....................................................................................................................................653.3.2 Goals............................................................................................................................................663.3.3 Breakthroughs.............................................................................................................................663.3.4 Methodology...............................................................................................................................663.3.5 UseofWiSHFULFunctionality.....................................................................................................66

3.4ExtensionofMACoptimizationsinhigh-densityscenarios,withonlinephase...................663.4.1 Overview.....................................................................................................................................663.4.2 Goals............................................................................................................................................673.4.3 Breakthroughs.............................................................................................................................673.4.4 Methodology...............................................................................................................................673.4.5 UseofWiSHFULFunctionality.....................................................................................................67

3.5 Interference classification forWi-Fi nodes on the basis of error patterns using machinelearning............................................................................................................................673.5.1 Overview.....................................................................................................................................673.5.2 Goals............................................................................................................................................673.5.3 Breakthroughs.............................................................................................................................683.5.4 Methodology...............................................................................................................................683.5.5 UseofWiSHFULFunctionality.....................................................................................................68

3.6Radiovirtualizationwithsimultaneoustransmissionandreception..................................683.6.1 Goals............................................................................................................................................693.6.2 Breakthroughs.............................................................................................................................693.6.3 Methodology...............................................................................................................................693.6.4 UseofWiSHFULFunctionality.....................................................................................................70

3.7 IEEE802.11OverlappingBSSmanagement........................................................................703.7.1 Overview.....................................................................................................................................703.7.2 Goals............................................................................................................................................703.7.3 Breakthroughs.............................................................................................................................71

H2020-GANo.645274 D2.4

8

3.7.4 Methodology...............................................................................................................................713.7.5 UseofWiSHFULFunctionality.....................................................................................................71

3.8Closed-loopratecontrolforIEEE802.11infrastructurenetworks......................................713.8.1 Overview.....................................................................................................................................713.8.2 Goals............................................................................................................................................713.8.3 Breakthroughs.............................................................................................................................723.8.4 Methodology...............................................................................................................................723.8.5 UseofWiSHFULFunctionality.....................................................................................................72

3.9ContextAwarenessinspectrummanagementsystem-aidedSUnetworks........................723.9.1 Overview.....................................................................................................................................723.9.2 Goals............................................................................................................................................733.9.3 Breakthroughs.............................................................................................................................733.9.4 Methodology...............................................................................................................................733.9.5 UseofWiSHFULFunctionality.....................................................................................................74

4 Conclusion .......................................................................................................... 75

References ................................................................................................................ 76

H2020-GANo.645274 D2.4

9

1 IntroductionThisdeliverable reportson theoutcomeof thesecondsetofWiSHFULshowcasesplanned inWP2General requirements and showcases and define in the previous deliverable D2.3 [1]. Theseshowcases define relevant and convincing scenarios in view of promoting WiSHFUL frameworkcapabilities. The aim is to show third parties and experimenters the potential benefits of usingWiSHFUL infrastructure and software platforms forwireless innovation creation. For example, theIRIS SDR module, which is a highly reconfigurable radio, demonstrates WiSHFUL capabilities bychanging Frequency, Gain and bandwidth parameters. Additionally, we also demonstrate someadvancedWiSHFULexperimentationcapabilitiesdefinedbytheGNURadiovirtualizationcomponentin Section 2.2. These showcases also act as educational, training and tutorial material to supportthirdpartiesusingWiSHFULUPIs.

Section 2 starts by listing showcases implemented during Year 2 of the project. Showcases aregroupedintothefollowingcategories:

• LTE-LAA-Wi-Ficoexistence• SoftwareDefinedRadio(SDR)capabilities• CoexistenceofIEEE802.15.4eTSCHwithIEEE802.11networks• RadioSlicingforVirtualizedHomeWi-FiAccessPoints• MACadaptationinpresenceoflegacystations• Loadandtopologyawarenetworking

Section3definesalistofshowcasestobeimplementedinYear3oftheproject.Theseinclude:

• OverthatairupdatesusingGITARforWSNs• Extension of MAC adaptation in multi-hop topologies based on directional antenna and

multiplepathreservations• Radio-basedindoorlocalization• ExtensionofMACoptimizationsinhigh-densityscenarios,withonlinephase• Interference classification for Wi-Fi nodes on the basis of error patterns, using machine

learning• Radiovirtualizationwithsimultaneoustransmissionandreception• IEEE802.11OverlappingBSSmanagement• Closed-loopratecontrolforIEEE802.11infrastructurenetworks• ContextAwarenessinspectrummanagementsystem-aidedSUnetworks

Finally,section4concludesthisdeliverable.

H2020-GANo.645274 D2.4

10

2 ResultsofthesecondsetofShowcasesimplementedinYear2ThissectionpresentstheresultsfromthesecondsetofshowcasesproposedanddefinedinD2.2andimplementedinYear2.

2.1 LTE-U-Wi-FicoexistenceCellularnetworkoperatorsareoffloadingtrafficinunlicensed5GHzspectrumusingLTEUnlicensed(LTE-U[2]).However,thispartoftheradiospectrumisalsousedbyexistingIEEE802.11standards,e.g.802.11ac,orfutureWi-Fistandards,e.g.802.11ax.Hence,thesituationbecomessimilartotheonein2.4GHzISMbandwhereradiospectrumissharedbydifferentradiotechnologies,e.g.Wi-Fi,Bluetooth, ZigBee, etc., where numerous studies have demonstrated the adverse effects ofinterferencefromdifferentwirelesstechnologiesonWi-Fi.

InparticularJindaletal.[3]showedthat,asLTE-Uduty-cyclingdoesnotimplementlistenbeforetalkmechanisms (LBT, e.g. CSMA/CA), introduction of LTE-U can disproportionately reduce Wi-Fithroughput performance. Moreover, interference from LTE-U with moderate power can be evenmore harmful toWi-Fi than high-power interference [4]. Other studies confirm this [22]. Figure 1showsresultsfromourownexperimentswhereasinglehighqualityWi-FilinkwasaffectedbyaLTE-Usignalsource.Inparticular,itshowsthenormalizedUDPthroughputoftheWi-FilinkunderLTE-Uinterference, relative to the non-interferedWi-Fi link, as a function of different interfering signalstrengths,i.e.distances,andLTE-Uduty-cycles.WecanclearlyseetheimpactoftheLTE-UdutycycleontheWi-Fiperformance.Moreover,theWi-FiperformanceisinfluencedevenwhenLTE-Usignalisveryweak,i.e.LTE-Usourceisfaraway.

Figure1.DegradationinUDPthroughputofahighqualityWi-FilinkinthepresenceofLTE-Udevice

operatinginthesamebandandatdifferentsignalstrengths.

InWiPLUS[5]wepresentedapassivelowcomplexitymethodfordetectionofLTE-URFdevicesandtheir duty-cycles in real-time and using only commodity Wi-Fi hardware. The method accuratelyestimates theairtimeoccupiedby LTE-U, i.e.ONperiod, at awide rangeof LTE-U signal strengthsallowingtheAPtoassesstheavailableairtimeforWi-Ficommunication.

In this showcasewe demonstrate how amanaged 802.11Wi-Fi networkmakes use ofWiPLUS topassively estimate the airtime occupied by LTE-U at each AP in order to perform better channelassignment and to load-balance client stations across the APs for maximizing the overallthroughput/servicequality.

H2020-GANo.645274 D2.4

11

2.1.1 PresentationofUPIUsedIn order to support this showcase the Wishful control framework must provide the followingfunctionality:

• UPI_RforpassivedetectionofchanneloccupationbyLTE-UateachAP,• UPI_R/NforidentificationofownWi-FinetworkloadateachWi-FiAP,• GlobalWishfulcontrolprogramwhichperformsWi-FinetworkadaptationtodynamicsinLTE-

Uchanneloccupation(->adaptiveduty-cycling)andownWi-Finetworkload:

• Long-term:changingradiochannelsofAPs,• Short-term:handoverclientSTAstolessloadedadjacentAPs.

Figure2.EstimationofchanneloccupationbyLTE-U(left)andexampleWi-Finetwork(right).

For passive estimation of channel occupancy by LTE-U we implement an UPI_R function whichinforms the centralized Wi-Fi controller about the current LTE-U channel occupancy in real-timeusingevents.ChangesinthenetworkloadateachWi-FiAParealsoreportedusingevents.Fromthisinformationtheglobalcontroller isabletoadapttochangesinLTE-UdutycyclesandownnetworkloadbyperformingRFchannelassignmenttoAPsandbytriggeringSTAhandoveroperations.

2.1.2 ApplicationFigure 3 shows theWishful framework components involved in this showcase.Wehave two localcontrol apps reporting information about LTE-U channel occupation andWi-Fi network load. Thisinformation isprocessedby jointchannelassignmentandhandovercontrolapplicationresidingoncentralized compute node. Based on the reported data this application performs STA handoveroperationaswelladjuststheAP’schannel.

time

freq

LTE-U336

40

44

48

LTE-U2

LTE-U1

AP1

AP2 AP3

STA1

STA2

STA3

STA4

H2020-GANo.645274 D2.4

12

Figure3.LTE-UWi-Fico-existenceusingWishfulcontrolframework.

2.1.3 ResultsWewillcompareourLTE-UawarechannelassignmentandSTAload-balancingschemewithstate-of-the-art solution,which takes only own network load into account. As performancemetricwewillestimatetheaverageclientthroughput,throughputfairnessandothermetrics.

2.1.4 NextStepsWe are currently working on the implementation of the showcase and the setup for thedemonstration.

2.2 SDRCapabilitiesSoftware-DefinedRadio (SDR) refers toplatformswhere thebasebandprocessing is performedbysoftware modules running either on field programmable gate arrays, digital signal processors,general-purposeprocessorsoracombinationthereof.Asaconsequence,operationcharacteristicsofthe radio technology, such as coding algorithm,modulation type, and channel accessmechanism,canbeeasilychanged,simplybyloadingdifferentsoftware.Inaddition,multipleradiodeviceswithdifferentcharacteristicscanbereplacedbyasingleSDR.

A typical SDR ismodularized in two independent parts: the hardware (which is typically just a RFfront-end/RF channelizer) and the software (a set of libraries to ease the development of signalprocessingalgorithms).ThemostutilizedandcommercialRFfront-endforSDRisthefamilyUniversalSoftware Radio Peripheral (USRP) from the Ettus Research (National Instruments - NI) company.USRPs are connected with a general-purpose machine (a personal computer or a notebook),executingthesoftwareresponsibleforthesignalprocessing.Althoughseveralsoftwaresolutionstoperform such task are available, the two most known and utilized by the SDR developmentcommunityareGNURadioandIRIS.

WeplantoillustratethecapabilitiesofWiSHFULwithbothGNURadioandIRISintwoindependentshowcases. With GNU Radio, we will showcase the execution of two independent virtual radiossharingoneRFfront-endsimultaneously;whileforIRIS,wewillshowcasetheintegrationwithUPIsbychangingtheparametersontheflyontheUSRPsdevices.

NetworkLoad

AP

Jointchannelassign.&handover

HOApp

AP

LTE-Uestim.

STAbalance

HORequest

LocalNode

ComputeNode

HOReplyEvent

LTE-U/Load-ReportEvents

Gateway

NetworkLoad

LTE-Uestim.

LocalNode

Ch.assign. Handover

operation

H2020-GANo.645274 D2.4

13

2.2.1 GNURadioCapabilities:RadioVirtualizationThefifthgeneration(5G)ofmobilenetworksisenvisionedtoovercomethefundamentalchallengesofexistingcellularnetworksbyprovidinghigherdatarates,excellentend-to-endperformance,anduser coverage in hot-spots and crowded areas with lower latency, and energy consumption. Toefficiently support sucha vast rangeof targets, providing a single air-interface “one-size-fits-all” isnotadesirablechoice. Instead, futuremobilenetworksshouldbecomeflexible,providingdifferentair-interfaces for particular users and applications. Radio virtualization is considered a promisingsolutiontoreachthisflexibilitybyenablingdeploymentofmultiplevirtualradios,i.e.,air-interfaces,ontopofoneRFfront-end.

The basic premise to radio virtualization is the adoption of a RF front-end dedicated only tofrequency digitizing, and a virtualization manager, i.e., a Hypervisor in standard virtualizationnomenclature.TheHypervisorisresponsibleforabstractingtheRFfront-endintoanumberofvirtualfront-ends (similar to what is done with virtual machines), and spectrum, which can be used bydifferentradios.Inaddition,theHypervisorisalsoresponsibleforschedulingtheavailableresourcesbetween the multiple air-interfaces. Mobile operators can quickly introduce new or evenexperimental services in theirproductionnetworkswithoutaffectingexistingusersorcan followanumberofotherdeploymentscenarios.

In this showcase,wedemonstrate the implementationofaHypervisor implemented inGNURadioandmanagedthroughtheWiSHFULframework.OurshowcasedemonstrationisshowninFigure4.Itwillconsistof(I)oneUSRPactingasaRemoteRadioHead(RRH)whichtransmitsLTEandNB-IoTdatasimultaneously,(II)oneUSRPactingasaLTEmobilesubscriber,and(III)oneUSRPactingasaNB-IoThealthcare data receiver. At the RRH side, wewill have a dedicated computer executing our SDRHypervisorandthesoftwareradiosfortheLTEtransmitterandNB-IoTtransmitter.Similarly,wewillhaveonededicatedcomputerfortheLTEandNB-IoTreceivers.

Toprovideinteractivity inourdemo,wewillhavetheLTEvirtualradiotransmittingavideostreamand theNB-IoT virtual radio transmitting data froma healthcare sensorwornby participants.Thetwodisplaysshowthevideostreamreceivedinthemobilesubscriberandthehealthcaresensor intheNB-IoT.WiSHFULframeworkwillbeusedtomanagetheconfigurationoftheHypervisorandthevirtualradiosthroughasetofflexibleandextensibleUPIs.

Figure4GNURadioVirtualisationComponent

Dat

a D

ispl

ay

IoT SensorMobile subscriber

NB-IoT RX Waveform

LTE TX Waveform

NB-Iot TX Waveform

Hypervisor LTE RX Waveform

NB-IoTData Display

LTE DataDisplay

VideoStreaming

HealthcareSensor

Rad

io H

ardw

are

Softw

are

Mod

ules

Dat

a Ac

quis

ition

RRH

PC 2 PC 3PC 1

Wishful

Agent

Wishful

Agent

Wishful

Agent

WiSH

FUL C

ontroller

UPI_R

H2020-GANo.645274 D2.4

14

a. PresentationofUPIUsedTheUPIsprovidedbyWiSHFULtoset-upandcontrolourshowcase,areasfollows:

def setCenterFrequency(self, center_frequecy)

Configures the center frequencyof a SDRRF front-end. This function returns amessage (apythonstring)on thestatusof theoperation,e.g., “SUCCESS”,“ERROR:Frequencyoutof range”,“ERROR:Frequencyalreadyinuse”.

def setVirtualRadioCenterFrequency(self, virtual_radio, center_frequecy)

Configures thecenter frequencyofagivenvirtual radio.This functionreturnsamessage (apythonstring) informing the status of the operation, e.g., “SUCCESS”, “ERROR: Frequency out of range”,“ERROR:Frequencyalreadyinuse”.

def setBandwidth(self, virtual_radio, bandwidth)

Configures thebandwidthofaSDRRF front-end.This functionreturnsamessage (apythonstring)informing thestatusof theoperation,e.g., “SUCCESS”, “ERROR: thisbandwidth isnotacceptedbythecurrentRFfront-end”.

def setVirtualRadioBandwidth(self, virtual_radio, bandwidth)

Configuresthebandwidthofagivenvirtualradio.Thisfunctionreturnsamessage(apythonstring)informingthestatusof theoperation,e.g.,“SUCCESS”,“ERROR:bandwidthoutofrange”,“ERROR:spectrumoverlapswithanothervirtualradio”.

ThesefunctionsareusedtochangetheconfigurationoftheRFfront-end(inthiscaseaUSRP)andvirtual radios. As an illustrative example, consider the case of a NB-IoT virtual experiencing highinterferencefromtheLTEvirtualradio.UsingaforementionedWiSHFULUPIs,theNB-IoTvirtualradiocanbeconfiguredtochangeitscentralfrequencytoonewithlessinterference.

b. ResultsWiththisshowcaseweintendtoshowtheflexibilityofmanagingvirtualradiosthroughtheWiSHFULUPIs. During the showcase, we use the UPIs functionality to demonstrate the impact of differentbandwidthsinthedatarageofthereceivingdevices.Forexample,wecanchangethebandwidthoftheLTEtransmitter/receiverpairtoaccommodateahigh-resolutionvideostream.Similarly,wecanchangethecentral frequencyoftheNB-IoTtransmitter/receiverpairandevaluatethe interferencecausedduetheproximitywiththeLTEvirtualradiospectrum.

c. NextStepsThe next steps include the implementation of UPIs to create and remove virtual radios on theHypervisor on the fly. More advanced feature include the encapsulation of virtual radios asVirtualizedNetworkFunctions(VNFs)andtheirmanagementthroughputtheWiSHFULFramework.

2.2.2 IRISRadioCapabilitiesTheIrisSDRPlatformisasoftwareframeworkforbuildinghighlyreconfigurableradionetworks.Thisarchitecture was developed using heterogeneous processing platforms including general-purposeprocessors, field-programmable gate arrays and the Cell Broadband Engine [6]. Its objective is toprovideanenvironmentforflexiblereconfigurationofallthesoftwareradiocomponentsinreal-time

H2020-GANo.645274 D2.4

15

withexcellentperformance.Toachievethisgoal,theplatformoffersinterfacestosupporton-the-flymodification of radio flow-graphs. This model supports adding and developing new radiocomponents for the software platform.Due to the flexibility offeredby components developed insoftware,anypartoftheradiocanbereconfiguredonthefly.

While platforms such as Iris SDR offer the necessary capabilities to enable dynamic support andreconfigurability, they can be quite complex for users to learn and utilise. These systems requiredeepknowledgeofnetworkprotocolsandsoftwarearchitectures,signalprocessingchains,cognitiveradios, C++ libraries and code, and so forth. TheWiSHFUL project aims to reduce the knowledgebarrierstousingadvancedSDRsandexperimentationfacilities.

In this showcase, we aim to demonstrate the integration betweenWiSHFUL and the IRIS SDR bymodifying the frequencyon two IRISSDRnodes,one transmitter (TX)andone receiver (RX),whileexecutingapingcommand.OurshowcasedemonstrationisshownintheFigure5.Itconsistsof(1)twolaptopsconfiguredwithIRISSDRsoftwareandWiSHFULagentsoftware,(2)onelaptoprunningtheWiSHFULController, (3)twoN210swithfirmwareUHD_003.005.005-0,(4)oneEthernetswitchconnectingtheWiSHFULControllermachinewiththeTxandRxlaptops.

Figure5TopologyusedintheIRISshowcase

Wedemonstrate the followingscenario: (I)oneUSRP transmitsaOFDMwaveformusingTUN/TAPinterface connected to one laptop. This task represents the transmitter running the Iris SDRFrameworkandWiSHFULAgentsoftware.TheTxnodesends ICMPpacketstothereceiverwithanintervalconstantintime.

The other laptop running the Iris SDR Framework and WiSHFUL Agent will receive the OFDMwaveformusingTUN/TAPinterfaceviatheUSRP.Thisnoderepresentsthereceiver,whichitisgoingtoreceivetheICMPpacketstransmittedbytheTxnode.

TheWiSHFULControllerlaptopisresponsibleforsendingarequesttotheTxandRxWiSHFULAgentstochangetheFrequencyparameteronbothnodes.ThisnodeusestheIRISWiSHFULAPIstotransmitthe commands to the Tx andRx nodes. The controller node supports dynamic reconfigurability ofindividual parameters of signal processing functionality in real-time on the Tx and Rx nodes. Theresult is the Tx and Rx nodes changing communicating frequencies on-the-fly,while continuing tosendandreceiveICMPpackets.AvideodemonstrationofthisIRISWiSHFULshowcaseisavailablein[7].

H2020-GANo.645274 D2.4

16

a. PresentationofUPIUsedTheUPIsprovidedbyWiSHFULtoset-upandcontroltheshowcaseisasfollows:

@framework.bind_function(upis.radio.set_frequency)

def set_frequency(self, set_frequecy)

Configures the value of the frequency in SDR. This function returns a message (a python string)informing the status of the operation, e.g., “SET_FREQUENCY_OK”, “ERROR:SET_FREQUENCY_NOT_OK”.

@framework.bind_function(upis.radio.set_gain)

def set_gain(self, set_gain)

ConfiguresthevalueofthegaininSDR.Thisfunctionreturnsamessage(apythonstring)informingthestatusoftheoperation,e.g.,“SET_GAIN_OK”,“ERROR:SET_GAIN_NOT_OK”.

@framework.bind_function(upis.radio.set_bandwidth)

def set_bandwidth(self, set_bandwidth)

Configures the value of the bandwidth in SDR. This function returns amessage (a python string)informing the status of the operation, e.g., “SET_BANDWIDTH_OK”, “ERROR:SET_BANDWIDTH_NOT_OK”.

@framework.bind_function(upis.radio.set_rate)

def set_rate(self, set_rate)

ConfiguresthevalueoftherateinSDR.Thisfunctionreturnsamessage(apythonstring)informingthestatusoftheoperation,e.g.,“SET_RATE_OK”,“ERROR:SET_RATE_NOT_OK”.

b. ResultsIn this showcase we demonstrate the integration between WiSHFUL UPIs and the Iris SDRFramework.Our goal is tomodify the frequency (set_frequency) on the Iris SDR Tx and Rx nodesusingtheWiSHFULUPIs.Thisintegrationsupportsindividualparameterreconfigurabilityinreal-time.

c. NextStepsThenextsteps involve the implementationofUPIswithadditional featuressuchaschangingcyclicprefix,modulationdepth,subcarriersvalues,antennatype,andsoforth,onthefly.

2.3 CoexistenceofIEEE802.15.4eTSCHwithIEEE802.11networksTheIndustrial,ScientificandMedical(ISM)spectrumof2.4GHzisaverybusypopulatedfrequencyspectrumthatisbyintentionsharedamongnumerousaccesstechnologies.Averypopular,ifnotthedominating technology is the omnipresent IEEE 802.11 (or Wi-Fi). Wi-Fi is by design primarilyoptimizedandtunedforhighthroughputonadistributedbesteffortbasis.Ontheotherhand,thetrend towardsCyber-Physical Systems (CPSs) caused increasedpopularityofmanyWirelessSensorNetworks(WSNs)technologies–suchasWirelessHART,Zigbee,BluetoothLEetc.–whichalsosharethis ISMband.Thedesignofthosetechnologieshasbeenprimarilydrivenbydemandsfor longlifethusenergyefficiency,relativelylowthroughputratesanddiscontinuousdatabursts.Devicesusing

H2020-GANo.645274 D2.4

17

these communication technologies are commonly battery powered and expected to operate overlongperiods(uptoyears)withoutmaintenance.

The demand for low transmission power and limited communication ranges comes also as adisadvantageandvulnerabilityiftheyhavetoco-existwithother,muchstronger,transmitters(suchasWi-Fi),whichcanimpaircommunications.

Modern CPSs encompass more and more varied control applications (including industrialapplications)withdiversified,oftenverydemandingrequirementsforthequalityofcommunication.Especially in CPS contexts approaches face an increasing emphasis on reliability and latencyrequirements.

In real world settings it is foreseeable that different co-locatedwireless technologieswill have togracefully co-exist to fulfill own system goals. Unfortunately, this leads to interference betweentechnologiesanddegradedperformance.Thediscussiononthe interference isgiven inSection2.1.The related work addresses the problem to deduce something about the behavior of the othertechnologyinordertomaketheinterferencelessharmful,howeverwithlimitedsuccess.

Thesettingforthisworkismotivatedbyademandingversionoftheabovescenario.Inparticular,weconsider coexistence of two prominent technologies that fulfill different goals in the envisionedsetting.On theCPS side IEEE802.15.4eTime-SlottedChannelHopping (TSCH) [1],which is gainingmomentuminbothstandardizationanddeploymentandconstitutesatechnologyforhighlyreliablecontrolnetworks.On theother side thepopularWLANtechnology, IEEE802.11Wi-Fi [2],used forhigh throughput best-effort data streams. Such combination is quite typical for laboratories, andindustrial settingswhere a plant infrastructure is run over a highly reliable and latency constraintTSCHnetwork,whileWi-Finetworks,occupyingthesameISMbands,areusedforhighthroughputmultimedia data. The challenge is to ensure that existence ofWi-Fiwill not adversely impact anyTSCHtransmission.

Thecurrentlyestablishedandwell-knownmechanismsoftheexistingtechnologies–suchasCarrierSenseMultipleAccess(CSMA)–cannotbeappliedinsuchascenariobecauseoftworeasons:

• TSCH slotsmight be scheduled to fulfill hard timing deadlines that cannot not be delayed norinterrupted(byWi-Fi);

• Unicast transmissions in TSCH do not incorporate Clear channel assessment (CCA)mechanismsandmustbeavoidedcompletely(evenifthenextslotisyettobegin).

EvenwithWi-Fi’s listen-before-talk carrier sensing capabilities, it leads to interferencewith TSCH.The Wi-Fi node has no means of knowing that a scheduled TSCH transmission is about to startclaiming the channel. On the other hand, TSCH standard doesn’t incorporate means to delaydedicatedslots(noristhatdesiredfordeterministictransmissions),asthereisnoCCAforthose.Inorder not to compromise TSCH transmissions, Wi-Fi needs to be silent for the duration of TSCHtransmissions.

The two technologies,althoughsharing thesame ISMband,are incompatiblewitheachotherandareunable todecodeeachother’s transmissions. Thismeans there isno intrinsicway forWi-Fi tolearnfromTSCHtransmissionsdirectly.Somesortofexternal informationexchange isnecessary inordertoachieveco-existence.

In this studyweadvocate explicit informationexchange and coordinationof operation among theTSCH andWi-Fi networks.We feature an approach inwhich the schedule, of time-critical but lowvolumecommunicationinTSCHnetworks,iscommunicatedtotheWLANsystems,whichdelaytheirpotentialtransmissioninordertoavoidinterference.ThemuchlesscriticalWLANtraffic,beingthebest effort service, can tolerate such delays. Our approach opens a number of challenges. Mostcriticalisthenecessitytoassureaprecisetimesynchronizationacrossthetwoincompatiblesystems.

H2020-GANo.645274 D2.4

18

Wehavedesignedacompletesolutionforcoordinatedcross-technologycooperationbetweenbothtypesofnetworksincludingacrosstechnologysynchronizationalgorithmthatallowsaWi-FinetworktoacquiretimereferencesignalsfromrunningTSCHnetwork.Thistimereferencesignalisthenusedby aWi-Fi network to keep the channel free for the time of TSCH transmissions and use regular(CSMAbased) transmission during the rest of the time.We implement a prototype of the systemusingCOTSWi-Finodesequippedwithath9kbasedIEEE802.11wirelesschipset,andTinyOSbasedfullyTSCHcompatiblewirelesssensornodes.

2.3.1 AssumptionsWe assume that TSCH and Wi-Fi networks are co-located in the same space and time and arerequired to coexistence without interference. Both networks are significantly overlapping incoverageandareinternallyusingsinglehopcommunicationinthisscenario,i.e.wedonotconsiderspatialreuse(multihopscenariowillbediscussedinSection9.1).Additionally,weassumethatTSCHis running some sort of industrial application and requires high transmission reliability. Thus wewould liketomakesurethattransmissionsarenot interferedbyWi-Fi,whileco-locatedWi-Firunsonbest-effortbasisrequiringmechanismstodisableallWi-FiactivityduringTSCHtransactions.

We assume fully compliant TSCH devices [8], using the IEEE 802.15.4e Standard defaultconfigurations. Itthisscenario, it isnotnecessaryforTSCHtofollowanartificiallyspecificscheduleforourenvisionedsystemtowork.MoredetailsandnecessaryschedulepropertiesarediscussedinChapter5.Asthereisnopossibilitytodirectlyexchangedatabetweencross-technologydevices,itisnecessary toacquire theTSCH linkschedule throughaseparatecontrol channel.Solutionsonhowsuchdatacanbeexchanged,isnotafocusforthiswork.

2.3.2 PossiblecoexistenceschemesThereisanumberofpossibleapproachestoenablecoexistenceofbothtechnologies.Firstofall,wecanseparatetheminfrequencyandtemporaldimensions.Wealreadyassumetheysharethesamespaceandthat,aswearetalkingaboutincompatiblephysicallayertechnologies,wearenotabletotalkaboutcodedimension.

a. FrequencyseparationAverysimpleapproachtoletTSCHandWi-Fico-exististoseparatethemcompletelyinfrequency.ThefirstoptionistoexcludechannelsusedbyWi-FinetworkfromtheTSCHhoppingscheme,thusachievingfull frequencyseparationbetweenbothnetworks.Thisapproach isspeciallypromising incaseof thehighnetwork load inWi-Fi.On theotherhand, inbiggerdeploymentswecanobservethatmultipleWi-FiAPsareusingmultiplechannels,thustherewouldbenofreespectrumavailableforTSCHnetwork.

The second approach is to makeWi-Fi avoid the TSCH transmissions in frequency. It is not verypractical as the big advantage and interference resilience of TSCH is based on spreadingtransmissionsoverthewholeISMbandandWi-Fidoesnotemployfrequencyhoppingmechanisms.

b. TemporalseparationTheotherapproachistoenablebothtechnologiescoexistintime.BasicapproachisalreadythereatleastontheWi-Fiside.Wi-FistationsemployCSMAprotocolbydefaultandthuswillnotstartanytransmissioniftheydetectsufficientenergyinthechannel.Thisishowevernotaveryreliablewayofdetectingothertechnologies(likeIEEE802.15.4)andoftenleadstofalsenegatives.

It isalsonotpracticaltomakeaTSCHnetworkavoidWi-Fitransmissions intime.First,Wi-Fisendsdata opportunistically, so it is not possible to reliably predict when the transmission will occur.

H2020-GANo.645274 D2.4

19

Secondly, TSCH as Time Division Multiple Access (TDMA) protocol does not employ CCA beforetransmission.

TheonlypossibilityistoperforminterferenceavoidanceintheWi-Finetwork.Namely,lettheTSCHnetworkprovideschedulinginformationsothattheWi-FitransmissionscanbedelayedtopointsintimewherenocollisionwithtransmissionfromtheTSCHnetworkisguaranteed.

2.3.3 FollowedapproachWewillfocusonthecross-technologyTDMAprotocoltocoordinatethetransmissionbetweenbothtypes of nodes and reduce interference to a minimum. The system should run a hybrid TDMAschemeontheWi-Finodes,whichsynchronizestotheTSCHTDMAslots.Itthenadaptstimeslotstokeep free slots that are implicitly reserved for TSCH network, and uses the remainder for regular(CSMA based) transmission. This schedule, if executed perfectly, guarantees no cross-technologyinterference.

In our system we consider two networks. A TSCH based wireless sensor network running someindustrialapplicationandaWi-FiAPwithitsownstations.ExamplecanbeseenonFigure6.Duetothe incompatibility of the PHY layers between IEEE 802.11 and IEEE 802.15.4 it is not possible todirectlycapturepacketsfromothertechnology.

Figure6Exampleillustratingtwoco-locatedwirelessnetworks

2.3.4 SystemdesignIn the proposed scenario and from the time-keeping perspective, it is possible to distinguish 3separatetimereferences:

• TheIEEE802.15.4eTSCHbasedsensornetworkisatightlysynchronizedwirelesssystemwhereaparticularnodewilladapttheclockbasedonowntimeparentbuthasnodirectnotionoftheabsolutetimereference.

• Linux hosts keepown time reference inUNIX Epoch time and can synchronize itwith othernodes over Network Time Protocol (NTP) or over Precision Time Protocol (PTP)where hightimingaccuracyisrequired.

• The IEEE802.11 uses Time Synchronization Function (TSF) to allow synchronization betweendevices in one BSS (Basic Service Set) e.g. AP puts own TSF counter value in the beaconpackets,andallstationssynchronizetoit.

InordertoachievethetightcooperationbetweenWi-Fidevicesandsensornetworkrequiredfortheco-existenceusecase(asdescribedinD2.3),whereWi-Finetworkwillceasethetransmissionswhenit would collide in time and frequency with the sensor node transmissions, it is necessary to

H2020-GANo.645274 D2.4

20

synchronize both networks. For start let us assume simplified problem with the focus onsynchronization between one sensor node (NXP JN5168 running TinyOS version of TSCH) and oneLinuxhostmachineequippedwithWi-Ficard.Therearetwomainoptionstosynchronizethosetwodevices.Overthedirectwiredconnectionorindirectlyoverwirelesslink.Bothofthesolutionshaveadvantages and disadvantages. Thewired connection promisesmore accurate and easier solutionbut requires tight couplingbetween thedevices.On theotherhand, thewireless solution ismuchmoreflexibleintermsofcouplingbutrequiresmoreoverheadinthedetectionofthesignalssentbythe different and in general incompatible technologies. In the next sections we will analyze bothwiredandwirelessapproaches.

The required synchronization accuracy depends on the time slot duration of the IEEE802.15.4enetwork.Although it isnot fixed in thestandard, theusualvalue is10ms. Itmeans that theWi-Fishould be able to cease own transmissions for the time of a TSCH slot. The inaccuracies of thesynchronization can be adjusted with additional guard times, but it will additionally limit theperformanceofWi-Finetwork.

Thereisstillaproblemofschedulealignmentwiththesynchronizednetworks.Itisnecessarytobeable tocalculatewhen theoverlappingTSCHchannelwillbeused. In the simplecase, if there isausedcellthatisscheduledtobetransmittedonanoverlappingchannel,theWi-Fitransmissionwillbe deferred. In more advanced scenario, the system should be able to detect if the Wi-Fitransmission would harm TSCH nodes that are allowed to transmit in a given cell (i.e. are in theinterference range ofWi-Fi). Themore advanced scenario should improveWi-Fi performance (wereducethetimeinwhichtransmissionsarenotallowed)withoutaffectingtheTSCHperformance.

a. TSCHScheduleThegoalofthispartofthework istocreateaspectral fingerprintmodelthatcanbeusedtocrosscorrelate against the recorded spectral data from theWi-Fi radio. This requires two independentproblemstobesolved.Firstly,thescheduleofTSCHneedstobeavailabletoWi-Fi.Secondly,Wi-Fineeds to learn, andadapt to, the timingofTSCH topredictwhen thenextactive slotof theTSCHscheduleisstarting.Finally,thestructureofscheduleshasstrongimplicationsonthedetect-abilityofthepatternandhencetheabilitytosynchronizetoit.Notallchosenschedulesareequallyqualifiedfor our spectrum based synchronization approach. They can possess a hidden internal periodicitythatcanimpedetheoptionforsynchronization.

The logical representation of all links in the network is called a schedule. Links on the nodes areorganizedinSlotFrames(SFs).Individualnodesonlyhaveknowledgeoflinksthattheyparticipatein(eitherassourceordestination).ASFhasacertainlengthandrepeatsseamlessly intime.MultipleSFscanbecontainedinaschedule,butforthisworkweassumethatonlyoneSFisused.Theglobalschedule therefore repeats with the same periodicity as the SF. Figure 7 shows an exemplaryschedulecontainingonlyoneSFwithtwolinks.

Twofundamentalwaysexisthowtoconstructschedules,whichrepresentalsothenetworktopologyontheMAClevel.Eitherconnectionsarecomputedonacentralcomputationlogicanddistributedtothenodes,or linksarenegotiatedbetweennodesonapeerbasisand instantiated inadistributedmanner.Ofcoursehybridapproachesofbotharepossible.Inindustrialsettings,wherereliabilityandlatencyconstraintsarevital,thecentralizedapproachistypicallychosen,forbeingbettersuitedforoptimization towards certain goals and constraints. This fact eases the process to acquire theschedulefromthecentralcomputationlogictomakeitavailabletotheWi-Fisynchronization.

H2020-GANo.645274 D2.4

21

Figure7TSCHScheduleintimeandchannel

IntheTSCHstandardeachnodecanbepartofmultiplenetworks,andthushavemultipleschedulesoperating in parallel. This can of course lead to overlapping slots between multiple schedules.Collisionsareresolvedbasedontheschedulepriority.InthisworkweconsiderTSCHschedulesthatconsistofoneSlotFrame (SF)onallparticipatingnodes.This constraint is implemented to simplifythescheduleexchangeprocessbetweenTSCHandWi-FiandtomakethefrequencypatternofTSCHmoreprominent.ImplementingmultipleSFslikein[9] isper-sepossible,butthefrequencypatterngetspotentiallymuchlongeraswell(iftheyaremutuallyprimetoeachother).

The foundation of our approach to synchronize Wi-Fi to TSCH is based on the intrinsic uniquechannel-hoppingpatternofaTSCHnetwork. Inorder tosuccessfully synchronize to thispattern, ithastobeacquiredfirstbyextractingallnecessaryinformationfromtheTSCHnetwork.Thisincludesthelinklist,theSFdescriptionandtheorderedlistofchannelsusedforthechannelhopping.OutofthatitisthenpossibletocomputethesequenceofrealchannelsusedforthatspecificTSCHnetwork.WeleveragethefactthatTSCHappliesarandomhoppingmechanismthatisactuallyrepeatingaftersometime.Thiswaivestheneedtocomputethechannelhoppingsequenceforanyspecificpointintime,asitwouldbefortruerandomhopping.Insteadwegeneratetheentireuniquesequenceanduseittocorrelatethemeasurementsagainst.Theactualfunctioningofthepseudo-randomschemeanditsperiodicityisdiscussedbelowinPeriodicityintimeandfrequency.ThecomputedsequenceofchannelsforeachlinkintheschedulegivesadatasetliketheonedepictedinFigure7.

Wearespecificallyinterestedinthecomputedsequenceofpseudo-randomchannels,asweusetheresulting frequency-spectrum-pattern produced by TSCH to synchronize to the schedule withoutexternal synchronizationmechanism.Even though the schedule repeatswith totalnumberof timeslots in the SF, the resulting computed sequence of channels for the TSs repeats with a slowerfrequency.Thisiscausedbythewaythepseudo-randomchannelhoppingiscomputed.Themodulooperator in the channel formula causes the repeating SF to use different channels in successiveiterations. The exact shift of channels per iteration depends on the actual lengths of the SF andchannellist.Thereisanupperboundofuniquetime-channelpatternthatisreachedwhenSF-lengthandchannellistlengtharemutuallyprimetoeachother(orprimenumbersthemselves).

Thestartingpoint forschedulemodelcomputations is the listofactive links in theTSCHschedule.LinksofaspecificSFaredefinedwithsourcenode,destinationnode,slotoffset,channeloffset.ThesampleTSCHlinkdescription(code-nameolafusedforexperiments)canbeseeninFigure8.Itwascreated in such a way to be most detectable with a very specific pattern with very low auto-correlationvalue.

H2020-GANo.645274 D2.4

22

Figure8ExampleTSCHschedule

The relativeoffsetparametersof links canbe transformed into real channelsand slots. Therefore,theorderedchannel list,aswellas the lengthof theSFneedstobeknown.Theresultingdatasetcontainselementswithpropertiessourcenode,destinationnode,ASNandchannel.Theusedchannellistinthisexamplehadthefollowingchannels{23,22,24,21}andtheSFalengthof25.Both,numberof channels and SF length, are no prime number but mutually prime. Consequentially, length (inSlots)oftheresultinguniqueslot-channelschedulehereis100.

The final step is to transform the computed slot-channel pattern into the time-frequencydomain.Meaningthediscreteslotsneedtobespreadtoactualtimevalues,whilethechannelsarespreadtodiscrete frequency values. Both operations are orthogonal and hence commutative. The resultingdatastructurecontainsthendatapointwithpropertiesofsourcenode,destinationnode,timeandfrequency. This step isnecessary for the subsequent cross- correlationwithmeasuredFFTdata. Intime domain the slot wise data needs to be resampled to match the sampling rate of theWi-Fispectralscan.Thespreadinginthefrequencydomainisabitmorecomplicated.ThefrequenciestobeusedforspreadingthechannelsintoaredictatedbytheFFTsamplingoftheWi-Firadio.AstraightforwardandsimplesolutionistoignoreanyspectralshapeofIEEE802.15.4-PHYtransmissionsandreplace the discrete channel by a rectangularmask of all frequencies that fall into a bandwith +-1MHz around the channel center frequency. Another approach is to use a more sophisticatedspectralmodeltomimicthereceivedsignalstrengthontheparticularsetoffrequenciesthatisalsoreceivedovertheairbytheWi-Fi radio.Thefinalgenerateddataforthescheduleshownabove inFigure8isshowninFigure9.WeusethisgenerateddatalaterforcrosscorrelatingtothemeasuredFFTdatatosynchronizeWi-FitoTSCH.

Figure9Uniquetime-frequencyschedulemodelusedforcross-correlation

H2020-GANo.645274 D2.4

23

b. CrosstechnologysynchronizationOne of the most challenging problems that needs to be solved in order to achieve the crosstechnologyTDMAscheme isachievinga synchronizationbetween twosystems. In thisChapterwewillfocusonsolvingthatproblem,byusingthespectralfingerprintmodelandcorrelatingitwiththeWi-Fispectralmeasurementsacquiredfromtheath9kbaseddevice.

Atheros802.11nchipsetsincludeabuilt-inspectralanalysisfeatures[10].Theyareabletoreporttheabsolutemagnitudeof thesignal frombasebandFFTprocessingunit.Theuserof this functionalityhasanability tocontrolhowoften thespectraldata is reported to thekerneland thoughdebugfsinterface to the user land process. The samples are time stamped with the accurate TimingSynchronizationFunction(TSF)values,asdefinedinIEEE802.11.Thoseareusedtoextractaccuratetiminginformationfromthedata.

c. CrosstechnologyTDMAHaving the Wi-Fi node synchronized to TSCH network we still need to make use of thissynchronization in the hybrid TDMA scheme. As the result of cross-correlation we have achievedproper mapping between TSF, UNIX time and the TSCH slot start. The next step is to use thisinformationtocontrolthebehavioroftheWi-Finodes.

Webasetheapproachonhybridmediumaccessarchitecture[11]toconfigurebehavioroftheWi-Finodes.Thesolutionexploitsthe802.11powersavingfunctionalitytoenablecontrolofthesoftwarepacketqueues.

Weareusing theproposedandmodifiedath9kkernel driver to control thedevices andun/pausesoftware queues based on the detected TSCH schedule and timing.Wehave extended the PyRIC1Pythonlibrarytosupportnewkernelfunctionalities.Withsuchaddition,andwithusageofpersistentNetlinksockets,itispossibletoachievefastinterfacebetweencontrolprocessrunninginPythonandthekerneldriver.

The proposed solution carries the same limitation as in [11] in terms of accuracy of theNetlinkcommunication.Additionally,duetothecurrentscheduling limitations, itwasnecessarytoconvertTSFtimestampstotheUNIXhostclock.Thiscanleadtotheadditionalinaccuraciesanddelaysintheexecution.

2.3.5 PresentationofUPIusedandnewThemaincontributionofthisshowcaseisacrosstechnologysynchronizationschemebetweenTSCHandWi-Finetworks.InthefirstimplementationitusesstaticinformationabouttheTSCHschedule,sonaturalextensionistouseWiSHFULUPI’stoextractthis informationfromdynamicallychangingTSCHnetwork.Thesolutionalsousestheath9kspectralscanfunctionalitiesalreadyavailableintheWiSHFULframeworktogatherinformationnecessaryforthetimesynchronization.

Ontheotherhand,thesynchronizationschemeitselfcanbeintegratedintheWiSHFULframeworkasaservice.Itdoesn’tprovideanyadditionalUPIbyitselfbutallowsformorefine-grainedcontroloftwotechnologiesatthesametimeandinacoordinatedmanner.

The implementationof thescenarioalso leads to improvements in theWi-FihybridTDMAschemethat gives much more flexibility on the control of Wi-Fi node sleep timing, that is not any morecontrolled by fixed schedule. On the other hand, such solution requires much tighter (and local)controltokeepWi-Finodeoperational.Note,toolongsleepingperiodwithoutcontrolwillresultinthenodedisassociationfromAP,orevenAPlosingallofitsclients.

1https://github.com/wraith-wireless/PyRIC

H2020-GANo.645274 D2.4

24

2.3.6 ResultsTo prove the feasibility of the solutionwe have designed an experiment consisting of two nodesconnected though variable attenuator (simulatingwireless channel).One is a sensornode runningTinyOS implementation of TSCH standard and sending packets. Second is an Intel NUC nodeequippedwithath9kbasedWi-FicardthatistryingtodetectTSCHtransmissions.

Theexampledata set is shown in Figure10, andpresentsdatameasuredby the IntelNUCdevicewith ath9k based Wi-Fi card connected over wire with the NXP JN516x node with 93dB ofattenuation.NXPJN516Xwassendingpacketsaccordingtoolafschedule.

Figure10ath9kspectrumscanofTSCHnodeovercable

Result of the normalized correlation can be seen in Figure 11. If we do peak detection(marks)andmapittotheothertimestampsinthefigure(TSFbased)wegetfollowingtimepoints: 0.715s, 1.715s, and 2.715s. At this point of time the TSCH schedule (presented inFigure8)started.Thecorrelatedvaluesarealotlargerthanthreestandarddeviationsfrommeanandahighattenuationofthecablegivesastrongindicationoftheapplicabilityofthesolution.

Figure11Cross-correlationofmeasurementswithTSCHmodel

0:00:00

0:00:00.500000

0:00:01

0:00:01.500000

0:00:02

0:00:02.500000

0:00:03

−0.1

0.0

0.1

0.2

0.3

0.4

0.5Cross-correlationcorr3σpeak

H2020-GANo.645274 D2.4

25

2.3.7 NextStepsThe whole implementation part of this showcase is finished. The next steps naturally includemoreextensiveevaluationandexperimentation.Namely,measuringpacketlossesinbothTSCHandWi-Fiandcomparestatisticsofdefaultoperationandwithoursystemrunning.ThisshouldbedonewithdifferentTSCHschedulesandvaryingtraffictypesinWi-Fi(e.g.graduallyincreasingloadinWi-Finetwork).

2.4 RadioSlicingforVirtualizedHomeWi-FiAccessPointsUsingseveralvirtualwirelessnetworks,identifiedthroughdifferentservicesetidentifiers(SSIDs)andbasicserviceset identifiers (BSSIDs)onasinglephysicalWi-FiAccessPoint (AP) isverycommon intodays'homeWi-Finetworks.ThefirstvirtualnetworkusuallyprovidestheInternetconnectivityfortheresidentortheowneroftheAP.Allfurthervirtualnetworksare,forexample,usedforso-called"GuestNetworks",CommunityNetworksorPublicHotspotsofcellularnetworkproviders.Albeitallof them are either used by different target groups, or are used to provide Internet access tosecondaryusers.Theprimaryuser(APowner)inthebestcaseshouldnotevenbeawareofthefactprimarynetwork connection is shared.While current approaches, e.gmeSDN [12], are focusedontrafficshapingorslicingonthewirednetworksideforprovidingguaranteedbandwidthtoprimaryusers,tobestofourknowledge,allcurrentapproachesareignoringissuesrelatetothewirelessMACandPHYlayers.

In this showcasewepresent a solution that considers the characteristicsof today'shomewirelessaccessnetworksandappliesanovelslicingtechnologiestoguaranteebandwidthandtrafficisolationfortheprimaryusersofthehomeAPinuplinkanddownlink.Figure1showsanexampleusecaseforthis approach.The solution inbasedon the functionalitypresented inhMAC [11],whichallows topauseandun-pausesinglesoftwarequeuesoftheATH9kdriveronaperlinkbasis.

Figure12–MACLayerSlicinginvirtualizedhomeWi-Finetworks,enabledthroughslottedtransmissionon

MAClayer.DeviceslicesaredynamicallyadaptedusingMAClayerinformationtoguaranteebandwidth.

The hMAC utilizes the existing ATH9K driver power saving implementation for the per link queuemanagement. As the power saving functionality is running on the host, no modifications to theregistersoftheWi-FideviceneedtobedonewhichpreservesthestandardMACfunctionalitieson

H2020-GANo.645274 D2.4

26

thedevicesuchasCSMA/CA.Forthisreason,thehMACdoesnotintroduceadditionalunfairnessandisfullycompliantwiththeIEEE802.11standard.

We utilize this functionality to enable a slotted mode on top of ATH9K based Wi-Fi hardware.Multipleslotsarecombinedtoformaslicethatisthenassignedtoaprimarydevice.End-usersareabletosetfixedbandwidthguaranteesfortheirprimarydevices.Theradioslicerthentakescareofassigning sufficient slots to the slices of the primary devices to ensure that the guaranteedbandwidthisdelivered.Thisisdonebyconsideringthecurrentlyusedphysicalbitrateoftheprimarydeviceandthecorrespondingcurrentsuccessprobabilityplusthecurrent loadandinterferenceonthe wireless channel. Doing so, every primary device will get its own explicit slice, while theremainingslotscanbeusedbyallprimaryandsecondary(e.g.hotspotusers)devices.

2.4.1 PresentationofUPIUsedandNewIn order to support this showcase the Wishful control framework must provide the followingfunctionality:

• UPI_Rforpassivedetectionofcurrentradiochannelload,• UPI_RfortogetcurrentlyusedPHYrateofeachclientSTA,• UPI_Rfortogetprobabilityofsuccessfulframetransmissionforspecificphysicalrate,• UPI_RforcontrollingthehMAC(install,uninstall,updateslotallocation),• LocalWishful control program,which performs periodic slice adaptation based on channel

load, client physical rate, transmit success probability and allocated user bandwidth.Moreover, the local control program has to update the slot allocation according to thecomputedslicesbycallingthehMACUPI_Rs.

• NodeRedAddon for livevisualizationofthecurrentthroughput,currentlyusedPHYrateandslicesizeduringtheshowcasepresentationasshowninFigure14.

For passive estimation of channel usage, we implement an UPI_R function that can be called todeliverthecurrentlyusedairtimeoftheradiochannel.ThisvalueisextractedoutoftheregistersofAtheros basedWi-Fi cards.Moreover, we implement a UPI_R function that delivers the currentlyusedphysicalrateofallassociateddevices(achievedbyiwcalls)andaUPI_Rfunctionthatdeliversthe probability of a successful frame transmission for this physical rate (achieved by querying theMinstrel rate table). The local control program then dynamically adapts the slice size assigned toeachdevice,uniquely identified via itsMACaddress,during runtime,basedon this information tothedesiredbandwidth,pre-setbefore.

2.4.2 ApplicationFigure14showstheNode-REDconfigurationtovisualizetheshowcase.Duringtheshowcase,deviceswillbesequentiallyturnedonwhichwillbevisiblefortheaudienceinthenoderedvisualization,e.g.Figure13.Moreover,itispossibletoseethattheguaranteedbandwidthfortheprimaryuserswillalwaysstayabovetheguaranteedlevel,whilethethroughputofthesecondarydevicesmaydecreasedependingonthechannelconditionsandtrafficloads.

2.4.3 NextStepsWeare currentlyworking to complete the implementationof the showcaseand the setup for thedemonstration.

H2020-GANo.645274 D2.4

27

Figure13–NodeRedlivevisualizationshowingtroughput,PHYrateandslicesizeofsingleWi-Fidevices

Figure14–NodeRedgraphicalrepresentationoftheshowcasevisualization

2.5 MACadaptationinpresenceoflegacystationsThisshowcasegeneralizestheresultspresentedinD2.3onthepossibilityofdefiningMACadaptationlogic devised to perform optimizations under varying load conditions (Load & interference awareMACadaptationshowcase),bydynamicallytuningthecontentionwindowofthewirelessnodesasafunctionofthenumberofactivenodes,andbyswitchingtoatime-divisionaccessprotocolincaseofseverecongestionlevels.WhileinYear1weconsideredthatthesamecontrolprogramwasrunningon each contending node, supporting the WiSHFUL UPI, in this generalization we consider thepossibility that some wireless nodes cannot be directly controlled by WiSHFUL. In particular,althoughtheoriginalshowcasewasdemonstratedwithboth802.11and802.15.4.nodes,werefertowireless localareanetworksbasedonthe802.11technology. Indeed,the logictobedeployedforoptimizing network performance in presence of legacy nodes has to take into account the

H2020-GANo.645274 D2.4

28

peculiaritiesoftheradiotechnology,andtheindirectformofcontrolthatcanbeachievedonlegacynodes by using the standard control mechanisms (rather than the WiSHFUL UPI). Note that thecoexistencewithlegacynodesisaverycrucialaspectfortheimpactofthesolutionsproposedwithintheWiSHFULframework:infact,wecanreasonablyassumethatwirelessnodessupportingWiSHFULUPI in real deployments can be a negligible fraction of the total wireless devices in short-terms.However,thedemonstrationofWiSHFULbenefitseveninpresenceofafewcontrollablenodescanbeavalidargumentforincentivizingtheadoptionofWiSHFUL.

Scenarioandadaptation logic.We consider awireless local areanetworkwitha single contentiondomain, inwhichallnodesare in radiovisibilityand transmit framesaccording to the IEEE802.11format and modulation schemes. In this domain, some nodes, including the Access Point, areWiSHFULcompatible,i.e.exposetheWiSHFULUPIandareabletorunlocalcontrolprograms,whilesomeothernodesfollowthelegacy802.11MAC/PHYprotocols.ThenumberoflegacynodesNandWiSHFUL nodes M can be tuned in different experiments; the specific network topology can becompletelycharacterizedbythecoupleofvalues(N,M).

Figure15–Networktopologyusedintheshowcase

As widely documented in literature, different optimization functions and load metrics can beconsideredaccordingtothedesiredperformancemetric.Eachoptimizationcanbegeneralizedandadaptedinpresenceof legacynodes. Inourshowcase,welimittheanalysistooptimizationsbasedon the tuning of the contentionwindow values and discuss, froma functional point of view, howaddressing coexistence under heterogeneousMAC protocols. Indeed, nodes using heterogeneouscontentionwindows can easily coexist in thenetwork (i.e. theydonot disturb each thanks to thecarrier sense and backoff freezing mechanism). The only coexistence problem can be an unfairrepartitionofthechannelresourcebecausestationsreceiveanumberoftransmissiongrantsthat,inlongtermsareinverselyproportionaltotheemployedcontentionwindowvalue.Conversely,nodesrunning CSMA protocols cannot safely coexist with nodes running a TDMA protocol, becausescheduled-based transmissions cancollidewithcontention-based transmissions startingbefore thescheduledtime.

Case1:InYear1weusedafirsttuningfunctionofthecontentionwindow,calledModeratedEDCAbackoff(MEDCA),whosegoalistheminimizationofthedelayjittersonthechannelaccesstimes.Itis well known that these jitters depend on the exponential backoffmechanism, which introducesshort-term throughput unfairness among the stations, and significant variabilities on the timebetween two consecutive channel accesses performed by the same station. UnderMEDCA, delayjitters are avoided by adopting a fixed contention window. This mechanism can natively support

H2020-GANo.645274 D2.4

29

coexistencewithlegacynodes,ascurrentlyproposedinastandardamendment.Thisisbecausethefixedvalueissettotheaveragecontentionwindowvalueexperiencedunderexponentialbackoffbythesamenumberofcontendingnodes.Sincethethroughputperformanceofeachstationdependsonthechannelaccessprobability,whichinturnsisonlyfunctionoftheaveragecontentionwindow,the moderated EDCA scheme is able to minimize the delay jitters while guaranteeing the sameaveragethroughputoflegacyEDCAstations.TheonlydifferenceisthatWiSHFULnodesworkingwiththefixedcontentionwindowwillexperiencelowerdelayjittersthanlegacynodes,withoutincreasingtheirthroughput.

Case2:InYear1weusedasecondtuningfunctionforthecontentionwindow,calledCWoptimum(CWopt), that enforces all the contending nodes to use the theoretical valuewhich optimizes thethroughputperformance.Suchanoptimalvaluedependsonthenumberofnodescontendinginthenetwork,asindicatedbelow,wherethenumberofcontendingnodesis𝑛!"#$,andthecollisiontimeandbackoffslotaregivenby𝑡!"##$%$"&and𝑡!"#$.

𝐶𝑊 = 𝑛!"#$ ∙ ! !!"##$%$"&

!!"#$ (1)

For generalizing this optimization logic in presence of legacy nodes, we theoretically studied thethroughputresultsthatcanbeachievedinanetworkwithNlegacynodesandMWiSHFULnodes,byassumingthatthetransmissionprobabilityτWofWiSHFULnodes isan independentvariable(whichcan be tuned by opportunistically choosing a contention window value equal to 2/τW, while thetransmissionprobabilityτloflegacynodescannotbechosenanddependsontheexponentialbackoffmechanismandexperiencedcollisionprobability.TheresultsofthisstudyaresummarizedinFigure16,wherethegreencurvesrepresentthetotalthroughputachievedinthenetwork,thebluecurvesrepresent theper-stationthroughputachievedby legacystationsandtheredcurvesrepresent theper-stationthroughputachievedbyWiSHFULnodes.Thefiguresshowsdifferentnetworkscenarios,representedbythecoupleofvalues(N,M)whichquantifythenumberoflegacyandWiSHFULnodes.Obviously,the(0,20)casecorrespondstothecaseconsidered inYear1, inwhichallnodescanbecontrolledbyWiSHFUL.Fromthefigureit isevidentthatasimplethroughputmaximizationcannotbe a practical solution. Indeed, maximizing the total throughput can lead to minimizing channelaccessgrantsforWiSHFULnodes,thusreducingthecontentionleveltothelegacynodesonly(whichachievethroughputresultsmuchhigherthanWiSHFULnodes).Onlyforveryunbalancedcases(e.g.(1,19))withafewlegacynodes,throughputmaximizationleadstocomparableresultsforWiSHFULand legacynodes. The figurealso shows (with anexplicit arrow) theτWvalue selectedbyMEDCA,which corresponds to exactly the same throughput for legacy andWiSHFUL nodes.We concludedthisanalysisbyexcludingthepossibilitytoonlyworkonWiSHFULnodesforoptimizingthenetworkperformance,andbyconsideringalternativesolutionsforindirectlyaffectingthebehaviouroflegacynodes.Acloserlooktothe802.11standardrevealsthatlegacynodesarenotconstrainedtoperformexponential backoff rules from the usual CWmin=16 and CWmax=1024 values, but can employalternative contention window limits signalled by the Access Point in the beacon frame. Thecontentionwindowvaluescanbedifferentiatedamongheterogeneousserviceclasses,includingbesteffort serving class, by means of a dedicated information element. The beacon frame can beexploitedalso for implementinganalternative controlmechanism: legacynodes canbepreventedfrom accessing the wireless channel in some portions of the beacon interval or in some beaconintervals,byspecifyingthestartofacontention-freeperiodthatcanbeignoredbyWiSHFULnodes.Since this mechanism is compatible even with 802.11 legacy stations not supporting EDCA, wedecided to explore this solution for using the standard-based signalling mechanisms deployed bymeansofthebeaconframesasanindirectcontrolmechanismforlegacystations.TheideaisusingtheoptimalcontentionwindowvalueforWiSHFULnodes,computedconsideringonlyMcontendingnodes,inthebeaconintervalsallocatedtoWiSHFULnodes,andthelegacyexponentialbackoffinthebeacon intervals allocated to legacy nodes. The WiSHFUL enabled Access Point could further

H2020-GANo.645274 D2.4

30

guaranteefairnessamongWiSHFULand legacynodes,bychoosingthenumberofbeacon intervalsallocatedtolegacyandWiSHFULnodesproportionallytothenumberofper-classnodes.

Figure 17 shows a graphical representation of our envisioned scheme by illustrating the channelaccess operations performed in 4 consecutive beacon intervals. In this example, we assume thatN=M; therefore, the beacon intervals are alternatively allocated to legacy and WiSHFUL nodes(intervals1and3areallocatedfor legacynode,and intervals2and4areusedforWiSHFULnodesemployingafixedcontentionwindowevaluatedbyFigure16withnflow=M).

Figure16–ThroughputresultsastheWiSHFULnodesvarytheircontentionwindow,incaseofcoexistence

withlegacyDCFnodes:(x,y)labelreferstoxlegacynodesandyWiSHFULnodes.

Figure17-GraphicalrepresentationoftheimplementedCWoptalgorit

Case3: InYear1weconsidereda lastoptimizationstage, inwhichWiSHFULnodeswereforcedtoswitchfromrandom-basedcontentiontoaTDMAaccessschemebyaglobalcontroller.Althoughwehavenotimplementedthegeneralizationofthisoptimizationscheme,previousconsiderationsaboutcase 2 suggest an easy solution for also supporting TDMA protocols on WiSHFUL. Indeed, sinceWiSHFULnodesandlegacynodesaccessthechannelinindependentbeaconintervals,it ispossible

H2020-GANo.645274 D2.4

31

toalsodeployheterogeneousaccess ruleswithin theallocatedchannel time.TheAccessPointcanimplementahierarchicalschedulingscheme,devisedtofirstallocatebeaconintervalstolegacyandWiSHFUL nodes, and then to allocate channel slots within the TDMA beacon interval to eachWiSHFULnode.

2.5.1 PresentationofUPIusedandnewTable1reportsthelistofWiSHFULUPIfunctionsusedfortheimplementationofthisshowcase.

UPI Description/Usagehc.start_local_control_program Setupandruncustomlocalcontroller;controller.send / controller.recv

Send and receive message between global and localcontroller;

radio.get_measurements GetMAC/Radiomeasurement, extract info on freezing timeandpacketstatistics;

radio.set_parameters Set MAC/Radio parameters; here we set an optimizedCW==CWmin==Cwax;

Table1-UPIfunctionsusedinshowcase

ThefollowingcodeinTable2showsthecontrolprogram,whichisresponsibleofinjectingthelocalcontrol logic and performs the node bootstrap phase. The local logic is implemented in the localcontrolprogram(myargs) function,whichperformsthetuningofthecontentionwindowaccordingtothecase1orcase2mechanisms.InputargumentsaregivenbythesendmethodoflpcDescriptorobject. For a given list of legacy_nodes andwishful_nodes, the controller configures the testbednodesforrunninglegacyDCFortheMAC-enhancedscheme.

[…] lcpDescriptor_wishful_nodes = [] mac_mdl='MEDCA' #possible values: ‘legacy-CSMA, MEDCA, CWopt for ii in range(0,len(mytestbed.wishful_nodes)): lcpDescriptor_wishful_nodes.append( controller.node(mytestbed.wishful_nodes[ii]).hc.start_local_control_program(program=local_control_program)) lcpDescriptor_wishful_nodes[ii].send( {'interface' : 'wlan0', 'mac_logic' : mac_mdl}) lcpDescriptor_legacy_nodes = [] for ii in range(0,len(mytestbed.legacy_nodes)): lcpDescriptor_legacy_nodes.append( controller.node(mytestbed.legacy_nodes[ii]).hc.start_local_control_program(program=local_control_program)) lcpDescriptor_legacy_nodes[ii].send( {'interface' : 'wlan0', 'mac_logic' : 'legacy-CSMA'}) […]

Table2Controlprogramcode

The optimization strategy utilized here is implemented in the following code in Table 3. Twooptimizationalgorithmcanbeadoptedand,forsakeofsimplicity,wedeployedauniquelocalcontrolprograminwhichareimplementedboththetuningmechanisms.Themechanismselectionisdefinedduring the controller bootstrap by a configuration parameter. In this case the configuration

H2020-GANo.645274 D2.4

32

parameters can take threepossiblevalues: i) legacy-CSMA, ii)CWopt,and iii) (ModeratedBackoff)MEDCA.

ThemainfunctioniscustomLocalCtrlFunction;itperformsthreedifferentMACbehavior:

1) Legacy-CSMA: only measurement are provided and no enforcement on MAC behaviour isprovided;

2) ModeratedBackoff (MEDCA):the localcontrollerextractsperiodicallyacount_freezingnumberfromthelastmeasurementreturnedbytheUPI_Rmonitorfunction.Thelast_count_freezingisaprogram variable used for detecting the periodic overflow of the counter. After filtering themeasurement, thenovel contentionwindow is computedbyusinganheuristic formula,whichrelatestheaveragecontentionwindowoflegacyEDCAstationstothenumberofbackofffreezes,also called IPT (interrupts per transmission). The computed value is filtered to avoid suddenmodificationsonthestationaccessrates.

3) Optimum contention window (CWopt): local controller retrieves periodically the number ofactiveWiSHFUL nodes. The contentionwindow value is computed as a function of number ofWiSHFULnodesinvolvedintheexperimentanddependsalsoonthepacketlength.

The novel contention window value is enforced by using the UPI_R function responsible ofconfiguring lower layerparameters.Thetuning logic isexecuted independentlyand locallybyeachnode,byexecutingthefunctioncustomLocalCtrlFunction(myargs).

""" Custom function used to implement local WiSHFUL controller """ def customLocalCtrlFunction(controller, interface, mac_mdl): # import modules and library import time import logging import math # initialization algorithm variables and node parameters b = 0.3 a = 0.1 last_count_freezing = 0 CWMIN = 15 CWMAX = 1023 ipt = 0 cw_f = CWMIN cw = cw_f; ip_address = controller.net.get_iface_ip_addr(interface) #local controller loop while not controller.is_stopped(): #receive message from controller msg = controller.recv(timeout=1) if msg: # to retreive traffic numbers n_tx_sta = msg["traffic_number"] # to retreive used MAC logic alg = msg["mac_mdl"] log.warning("num_tx_nodes=%d" % n_tx_sta ) #get node statistics UPI_myargs = {'interface' : interface, 'measurements' : [UPI_R.COUNT_FREEZING, UPI_R.IPT, UPI_R.NUM_TX_DATA_FRAME, UPI_R.NUM_RX_ACK, UPI_R.NUM_RX_ACK_RAMATCH, UPI_R.BUSY_TYME , UPI_R.TSF, UPI_R.NUM_RX_MATCH] } node_measures = controller.radio.get_measurements(UPI_myargs)

H2020-GANo.645274 D2.4

33

[…] if alg == "MEDCA" : """"""""""""""""""""""""""""""""""""""""""""" execute MEDCA algorithm and find new CW value """"""""""""""""""""""""""""""""""""""""""""" delta_freezing = count_freezing - last_count_freezing last_count_freezing = count_freezing ipt = ipt + a * (delta_freezing - ipt); targetcw = -0.0106 * ipt ** 2 + 2.9933 * ipt + 18.5519 # determine the target CW for this IPT cw_f = cw_f + b * (targetcw - cw_f) cw = round(cw_f) cw = int(cw) cw = max(cw,CWMIN) cw = min(cw,CWMAX) if alg == "CW_OPT": """"""""""""""""""""""""""""""""""""""""""""" execute CWopt algorithm and find new CW value """"""""""""""""""""""""""""""""""""""""""""" Tc = t_data + EIFS; #Collision time cw_f = n_tx_sta * math.sqrt(2*Tc / tslot) cw = round(cw_f) cw = int(cw) cw = max(cw,CWMIN) cw = min(cw,CWMAX) #update CW if not (alg == "legacy-CSMA"): UPI_myargs = {'interface' : interface, UPI_R.CSMA_CW : cw, UPI_R.CSMA_CW_MIN : cw, UPI_R.CSMA_CW_MAX : cw } controller.radio.set_parameters(UPI_myargs) return 'Local WiSHFUL Controller END'

Table3Optimizationstrategyutilized

2.5.2 Results

a. ResultswithMEDCAalgorithm(Case1)This subsection reports the results of the experiments based on theMEDCAMAC adaptation.Weactivated6wirelessnodescontendingundergreedytrafficsourcestowardsacommonAccessPoint.Forthisexperiment,weconsideramodulationrateof24Mbpsandaframelengthsof200byte.Theexperiment duration is 60 seconds. We first consider a legacy CSMA protocol with exponentialbackoff. Figure 18 shows the throughput performance achieved by each station and some regularsamples of the contention window values (gathered by the WiSHFUL UPI_R) employed by thestations. Although the CSMA protocol in principle should provide an equal share of the networkthroughput toeach station,wecanobserve someshort-termand long-term throughputvariabilityduetotheexponentialbackoffmechanism(short-term)andtothelocation-dependentinterferenceconditionssufferedbyeachstation(long-term).

Forthreeofthenodes,wethenactivatethe localcontrol logic implementingthemoderatedEDCAbackoffscheme.Figure19showsthatthethreestationsachieveanaveragethroughputcomparabletotheoneexperiencedwhentheywereusingexponentialbackoff,butwithsmallerfluctuations.Thisisconfirmedbythesamplesofthecontentionwindowvalues,thatexhibitverysmallvariations(from20toabout25).

H2020-GANo.645274 D2.4

34

Figure18–ThroughputperformanceandContentionWindowsamplesinanexperimentwith6wireless

nodesexecutingCSMAwithexponentialbackoff.

Figure19ThroughputperformanceandContentionWindowsamplesinanexperimentwith3stations

employingmoderatedbackoffincontentionwith3legacystations.

b. ResultswithCWOPTalgorithm(Case2)ThissubsectionreportstheresultsofanexperimentwhenWiSHFULnodesemploytheCWoptMACadaptationlogicandwhentheyareforcedtoaccessthechannelduringthecontention-freeperiodssignalled by theAccess Point beacon frames. The network scenario includes a total number of 10wirelessnodes,5ofwhichemploylegacyDCFandother5areusingWiSHFULCWoptMAClogic.Forthis experiment we used a modulation rate of 24Mbps and a frame length of 1470 byte. Theexperimentdurationis60seconds.Figure20(left)showsthatallthewirelessnodesinvolvedintheexperiment achieve almost the same throughput, because the beacon intervals allocated to eachclass of nodes are the same. However, WiSHFUL nodes slightly optimize their throughoutperformancebyusingtheoptimalcontentionwindowvalue.Figure20(right)showsthevalueofthecontention window value (namely, 57) evaluated by the CWopt algorithm and used byWiSHFULnodes.

6 EDCA EDCA CW

H2020-GANo.645274 D2.4

35

Figure20-ThroughputperformanceandContentionWindowsamplesof5stationsemployingCWoptin

contentionwith5stationsemployingexponentialbackoff

2.5.3 NextstepsIn conclusion, in this showcase we proved that the optimization logic can be implemented onWiSHFUL programmable nodes with or without presence of legacy nodes. In particular, weconsidereda firstmechanism, calledModeratedbackoff,whichdoesnot require any coordinationwith legacy nodes, and a secondmechanism, called optimal CW,which requires indirect forms ofcoordination with legacy nodes based on control mechanisms defined in the standard. Nextgeneralizationsof this showcasewill involve thesolutiondescribedascase3, i.e. thepossibility toguarantee coexistence among heterogeneous access protocols, and the definition of schedulingmechanismstobeimplementedattheAccessPointfordynamicallyallocatingthebeaconintervalstoWiSHFULandlegacynodes.

2.6 LoadandtopologyawarenetworkingIndynamicwirelessnetworkstheapplicationrequirementsvaryover-time.Moreover,networkscangrow or shrink as a result of node mobility. Network protocols designed for such networks (e.g.6LowPanandRPL forsensornetworksand IEEE802.11eandOLSRforWi-Finetworks)havebuilt-insupport for allowing such dynamic behaviour. The standards, defining these protocols, allow fine-tuningtheprotocoloperationviaconfigurationparameters,enablingdifferentperformancemetricstrade-offs evaluation. The implementations of these protocols, however, do not provide a unifiedinterfaceforthispurpose.TheseshowcaseswilldemonstratehowtheWiSHFULUPIscanbeusedto(i) dynamically monitor the network performance and topology; (ii) change network protocolconfiguration;or(iii)switchroutingmodules.Twoshowcaseswereimplemented:

Lowering frame loss in highlymobile networks: illustrates howdynamic reconfiguration of frameaggregationandPHY rateadaptationparameters can lower the frame loss.Becausenodemobilityintensifies the time varying nature ofwireless channel, the network stack has to be re-configuredaccordingtothedegreeofthemobility.Forthispurpose,thenodemobilityismonitoredand,basedonthelevelofmobility,theaggregationleveland/ormodulationandcodingscheme(MCS)indexarereducedorincreasedinreal-time.Byloweringtheaggregationlevel(e.g.numberofframes)orMCSindex in case of higher nodemobility, the impact on bit errors, due to mobility, can be reducedduringframereception.

Detecting the optimal link estimation algorithm in various network topology scenario’s:demonstrates thataglobalcontrolprogram,controllingasensornetwork,can increasetheoverallnetwork performance by dynamically selecting the optimal link estimation algorithm. For thispurpose, the network topology and performance is monitored and, the optimal link estimator is

H2020-GANo.645274 D2.4

36

choosenineachscenario.Insparsenetworks,simplelinkestimators(e.g.objectivefunction0,ETX)are preferred; in dense network more complex link estimators (e.g. 4BIT, fuzzy LQE) are moreappropriate.

2.6.1 MCSSelectionIEEE 802.11 is evolving from 802.11a/b/g to 802.11n/ac in order tomeet themuch-needed high-throughput demand of smartphones, laptops, and tablet PCs. To achieve high throughput, IEEE802.11n defines two types of frame aggregation:MAC service data unit (MSDU) aggregation andMAC protocol data unit (MPDU) aggregation. The latter, named the aggregate MPDU (A-MPDU),amortizes PHY protocol overhead overmultiple frames by packing severalMPDUs into a singleA-MPDU. It isgenerallyconsidered thatA-MPDU ismoreefficient in !-proneenvironments thanks totheusageofblockacknowledgements(BlockAcks)whichalloweachoftheaggregatedMPDUs(i.e.,A-MPDUsubframes)tobeindividuallyacknowledgedandselectivelyretransmitted.

Understandingly, it has been believed that longer A-MPDU conveying more A-MPDU subframesalways achieves higher throughput by reducing protocol overheads. All the existing studies havefound the optimal length of MAC frames based on mathematical analysis and/or simulations,assumingauniformdistributionoferrorsacrossanentireA-MPDU.

However, our experimental results reveal strong evidence that the distribution of errors over theentireA-MPDUisnotuniform,especially,formobileusers.Forexample,weobservethatwhenlongA-MPDUframesareused,thethroughput isreducedbyuptotwothirdsregardlessofthechannelcondition at the receiver in time-varying channel environment, even if an appropriate PHY rate isselected. Furthermore, we find many scenarios where the performance actually degrades as thelengthofA-MPDUincreasesduetothelimitedchannelcompensationprocedureexecutedbyWi-Fidevices. In such cases, the channel state information (CSI) measured using the physical layerconvergenceprotocol (PLCP)preambleatthebeginningof theA-MPDUmayno longerbevalid forsubframes in the latterpartofA-MPDUunder the time-varyingchannel. Specifically, thesubframeerrorrate(SFER)mayincreaseasthetimegapbetweenthepreambleandsubframeincreases,sinceautomaticgaincontrol(AGC),timingacquisition,frequencyacquisition,andchannelestimationstepsare conducted only during the PLCP preamble reception. It therefore leads to higher SFER in thelatter part of A-MPDU than that in the beginning part when the channel condition substantiallychangesduringtheA-MPDUreception.

We analyze the wireless channel dynamics considering mobility in IEEE 802.11n WLAN throughextensive measurements. From this, we reveal the fundamental problem of existing frameaggregationschemesmanifestedoverawiderangeofmobilityandIEEE802.11nPHYfeatures.

a. PresentationofUPIUsedandNewThelistofWiSHFULUPIfunctionsusedfortheimplementationofthisshowcaseareshowninTable4.

UPI Usage

Wi-Fi.ampdu_length() determine A-MPDU length bound in time unit(microseconds)

Wi-Fi.fixed_rate() determinefixedvalueofMCSfortransmission

rateadaptationalgorithm(Minstrel)canalsobeturnedon

Table4–UPIfunctionsusedinthisshowcase

H2020-GANo.645274 D2.4

37

Inthecontrolprogram(Table5), inputsthatexceedthepermittedrangeareadjusted:forA-MPDUlength,from100to10240us,andforMCS,from0to15.ForMCSselectionUPI,theinput‘-1’meansthatthedefaultrateadaptationalgorithm(Minstrel)isused.

[…]

if length < 100:

length = 100

elif length > 10240:

length = 10240

if rate > 16:

rate = 15

elif rate < -2:

rate = -1

res_l = device.radio.ampdu_length(phy_dev, str(length))

res_r = device.radio.fixed_rate(phy_dev, str(rate))

[…]

Table5Controlprogram

b. ResultsThis subsection shows the measurement results with changing A-MPDU length and MCS in bothstaticandmobileenvironments.Wesetthespeedofadeviceto1m/s,whichissimilarwithwalkingspeed.

Figure21summarizes theaverage throughputandSFER forvaryingaggregation timeboundwhichdetermines the aggregation length of A-MPDU. The aggregation time of 0 µs represents thetransmission of a single MPDU without aggregation. As the length of A-MPDU increases, thethroughputofthestaticscenario(0m/s)increasesduetotheoverheadreduction.However,foranaveragespeedof1m/s,themaximumthroughputisachievedat2,048µsaggregationtimebound.Whentheaggregationtimeboundislargerthan2,048µs,theincreasedSFERinducedbythemobilityoverwhelms the gain from the overhead reduction. Accordingly, the throughput decreases as theaggregationtimeboundincreases.

H2020-GANo.645274 D2.4

38

Figure21Throughputwithdifferenttimebounds.

MCS is a critical factor determining receiver performance. Although higher order MCS achieveshigher data rate, it is vulnerable not only to the channel degradation but also to mobility, sinceconstellationpointsarecloserfromeachotherandcodinggainisgenerallysmaller.Toverifythis,weconductexperimentsbychangingMCSsforagivenmobility.Figure22showstheSFERdependingonthe subframe location. When a station holds its position and does not move (0 m/s), the SFERremainsalmostzeroinallsubframelocationsregardlessoftheemployedMCS,becausethechannelqualitybetweenAPand the station is considerably good.However,when the stationmoves at anaveragespeedof1m/s,MCS4andMCS7employingbothamplitudeandphasemodulation(i.e.,16-QAMand64-QAM, respectively) showhigherSFER in the latterpartofA-MPDU,whileMCS0andMCS 2 which use only phase modulation (i.e., BPSK and QPSK, respectively) achieve stable SFERacross the entire subframe locations. That is, MCSs which use amplitude modulation are highlysusceptibletomobility.AnA-MPDUusinghighorderMCSnotonlyuseslongerA-MPDUlengthforagivenA-MPDUduration,butalsosuffersfromhigherSFERatthe latterpartofA-MPDU,especially,whenthechannelvarieswithinthetransmissiontimeofA-MPDU.Therefore,thelengthofA-MPDUwithhighorderMCSshouldbecarefullydetermined.

Figure22SFERfordifferentMCSs.

H2020-GANo.645274 D2.4

39

c. NextStepsThenextstepincludestheimplementationofUPIswithadditionalfeaturessuchasextractingtheA-MPDU transmission results from BlockAck and enabling/disabling the mobility-aware algorithmwhichadaptsPHYrateandframeaggregationlengthinreal-time.

2.6.2 LinkestimatorselectionLink estimators are extremely important in multi-hop wireless networks for obtaining a goodnetwork performance because they drive the decisions made by the routing protocol. Manyestimators exist but the quality of their estimation depends on the scenario at hand. In thisshowcase, the impact of the estimator on the network performance for different networkingscenariosisinvestigated.Forthispurpose,severallinkestimationalgorithmswereaddedtotheRPLimplementation of Contiki. The following subsection gives a brief overview of the evaluated linkestimators.

a. EvaluatedlinkestimatorsLinkestimationalgorithmstakethelossynatureandtimevaryinglinkqualityofWSNsintoaccountwhendeterminingthebestneighbourtoforwarddatato.Asdiscussedin[13],linkestimatorscanbeclassified in hardware- and software-based algorithms. The hardware based link estimators usequality indicators (i.e.RSSIandLQI) setby the radiodriverafterpacket reception. SoftwarebasedlinkestimatorssuchasFourbit[14]andFuzzyLQE[15],combineinformationfrommultiplenetworklayersintoasinglemetric.Therearealsoverysimplelinkestimators,suchastheobjectivefunction0[17],thatselectneighboursbasedonhopcount.Inthissectionaclassificationoftheevaluatedlinkestimatorsisgiven.

Hardware based link estimation algorithms use information provided by the radio (e.g. ReceivedSignalStrengthIndicator(RSSI)andLinkQualityIndicator(LQI))tocalculatethelinkmetric.Bothlinklayer variables are highly correlatedwith the PacketDelivery Ratio (PDR) [16] and can be used tomodel link quality. Due to the unstable nature ofwireless communication, the raw values can beaggregated with an Exponential Weighted Moving Average (EWMA) filter [13] to improve theestimation.BecauseLQIandRSSIareonlycalculatedafterpacketreception,packet-lossisnottakenintoaccount.Forthisreason,thelinkqualityoflessreliablelinkscanbeoverestimated.

Software based link estimation algorithms The objective function 0 [17], also referred to as hopcount, isavery simplesoftwarebased linkestimator thatminimizes thenumberofhops fromthesource towards the destination resulting in a route where the preferred parent is the furthestreachablenodeinthedirectionofthesink. Ifthequalityofthe linkwiththisnodeispoor,a lotofretransmissionsandextrapacketlossorenergyconsumptioncanoccur.Ontheotherhand,instablenetworkstheoverheadislimitedformaintaininglinkinformation.Morecomplexsoftwarebasedlinkestimation algorithms will tackle the aforementioned issue by using historical and/or cross-layerinformationtomakemoreintelligentdecisionswhenselectingthebest links.ETXbasedalgorithms[18]estimate thenumberofexpected transmissionsneeded to successfully sendapacket toeachdestination by counting the number of attempts needed in previous transmissions and select thepathwithminimalETX.TheFourbitalgorithm[14]combinesinformationfrommultipleOSI-layerstocalculatelinkmetrics.ItusesLQIandRSSIvaluestoaccountlinkqualityandcombinesthiswiththeETXpathmetric.FuzzyLQE[15]combinesmultiple linkmetrics ina lessdeterministicwaybyusingmembershipfunctionswhichassignascorein[0,1]foreverylinkmetric.TheseindividualscoresarecombinedwithanaggregationfunctionsuchastheYaggeroperator.Multiplemetricsareaggregatedbycalculatingaweightedmean(βwithatypicalvalueof0.6)oftheworstandaveragevalueofeachmetric.Thisresultsinalinkscorewhichcanbeinvertedtoobtainalinkmetricforroutingpurposes.

Intheexperimentpresentedhere,sixdifferentlinkestimatorsareused:

H2020-GANo.645274 D2.4

40

• Hardwarebasedlinkestimators• RSSI• LQI

• Softwarebasedlinkestimators:• ObjectiveFunction0• ETX• Fourbit• FuzzyLQE

b. PresentationofUPIUsedandNewThemaincontributionofthisshowcaseistheabilitytochangethelinkestimatoratrun-time.IthasbeenexposedasaUPI_Nparameter forContikinodesandcaneasilybegeneralized formulti-hopwirelessroutingprotocolsdesignedforotherphysicallayertechnologies.

Beside theability to change the linkestimator, the following tasks, eachwithaddedWiSHFULUPIfunctionality,wererequiredwhileevaluatingthisshowcase:

• Monitornetworkperformanceandtopology.• Gatherknowledgeaboutthetopologybyobservingneighbortableupdates.• MeasurenetworkperformancebycollectingstatisticsintheIPandRPLlayer.• Measureenergyconsumptionbycollectingtheradioontimefromallnodes.

• Configureexperimentscenario.• (De-)Increasethenodedensityby(de-)activatingapplicationsoncertainnodes.• Modifythesendintervalboundariesfortrafficgeneratingnodes.

Table6andTable7list,respectively,theUPIfunctionsandattributesusedinthisshowcase.

Table6UPIfunctionsusedduringthelinkestimatorselectionshowcase.

Function UPI Usage

get_neighbor_table N Monitor node density by collecting neighbor tableinformationfromallnodes.

start_application N Start application on a particular node to increase nodedensity.

stop_application N Stop application on a particular node to decrease nodedensity.

set_parameters N UsedtoupdatethelinkestimatorinRPL.

get_measurements_periodic R/N Retrieve the radio-on-time, IP and RPL statistics in aperiodicmanner.

subscribe_events N CollectRX/TXeventsfromtheapplicationlayer.

Table7UPIattributesusedduringthelinkestimatorselectionshowcase.

Attribute UPI Type Description

RADIO_ON_TIME R Measurement Cumulativemeasurement indicatingthe amount of time the radio wasonsincebooting.

H2020-GANo.645274 D2.4

41

RPL_OBJECTIVE_FUNCTION N Parameter Objective function used by RPL toestimate the link quality andcalculatepathcost.

RPL_STATS N Measurement CumulativestatisticsspecifictoRPL.

IP_STATS

N Measurement CumulativeIPlayerstatistics.

APP_PER_PACKET_RX_STATS N Event Event triggered each time a packetisreceived.

APP_PER_PACKET_TX_STATS N Event Event triggered each time a packetistransmitted.

APP_SEND_INTERVAL_MIN_MAX N Parameter Allows todefine theboundaries forthe application send intervalbetween a minimum and amaximum.

c. ResultsTheresultswereobtainedusingtheCoojasimulator.First,astandardscenariowasdefinedinwhichwetestedall linkestimators tohaveabaseline forcomparing resultsobtained inotherscenario’s.During the experiment, the following performance metrics were gathered from the Contiki IPv6routinglayerviatheUPI_Ninterface:

• GenericUPI_Nmetrics• Packetdeliveryratio(PDR)atthedestination(sink)node.• Averageradio-ontime(ROT)overallnodesinthenetwork.• Averagenumberofhops(NH)foreachrouteinthenetwork.• Retransmissionratio(RR)atthesourcenode.

• RPLspecificUPI_Nmetrics• NumberofRPLparentswitches(NRPS).Thisgivesagoodindicationaboutthestabilityofthe

networktopology,i.e.ifalotofparentswitchesoccur,thenetworktopologyisnotverystable.Thegraphsbelowshowtheaveragenumberofparentswitchespernodeperminute.

• NumberofRPLDIOmessagessent(NRDMS).Thisgivesagoodindicationaboutthenumberofmessagesthatarerequiredtoinformnodesabouttopologychanges.Thisstatistic isstronglycorrelatedbythenumberofRPLparentswitches.ThegraphsbelowshowtheaveragenumberofDIOmessagespernodeperminute.

Beside thestandardscenario,alsoadense (morenodesperm2)andasparse (lessnodesperm2)scenarioswereusedtoevaluatedifferent linkestimators.The followingsubsection lists theresultsobtainedforeachofthecollectedmetricsina)thestandardscenario;b)thedensescenario;andc)thesparsescenario.Theresultsin(b)and(c)arecomparedwith(a).Anoverallconclusionisgiveninthelastsubsection.

Standardscenario

Thestandardscenario,illustratedinFigure23,reflectsatypicalscenarioforwirelesssensornetworksinwhichaconsiderablenumberofnodesreport their sensorvaluesperiodically toaso-calledsinknode, collecting all sensor information and exposing it to the outside world. The send intervalboundariesare20secsand40secs,andforeachperiod,anewtransmissiondelayischosenwithinthis interval.Thescenarioconsistsof64sensornodes inaneightbyeightgrid.Thepositionofthe

H2020-GANo.645274 D2.4

42

sink(atacornerofthegrid)andthesizeofthegrid(175mby175m)waschosentoresultinamulti-hoptopology.ThecommunicationrangeandinterferencerangesettingsinCoojawereconfiguredtorespectively 45m and 60m. Note that the same experiment can be easily repeatedwith differenttopologiesand/orsinklocations.

Figure 24 shows the results for each of the aforementionedmetrics gatheredwhile executing thestandardscenario. Itcanbeseenthatthere isalreadyacleartrade-offbetweennetworkreliability(chart (a): Packet Delivery Ratio, PDR) and energy consumption (chart (b): Radio On-Time) whenchoosingof0orETXoverthemorecomplexFourbitorFuzzyLQElinkestimators.Themorecomplexlinkestimators introducemoreRPLoverhead, i.e.requiremoreparentswitches(chart(e))andDIOmessages (chart (f)). The average number of hops (chart (c)) andMAC retransmissions per packet(chart (d)) are comparable for all link estimators except for the RSSI based estimatorwhich has ahigherhopcountandrequiresmoreRPLoverhead,resultinginahigherenergyconsumption.

Examining the results a bit closer, it can be seen that while the Four-bit estimator has a highernumberofRPLparentswitchescomparedtotheRSSIestimator,ithasalowerincreaseinsentDIOmessages.ThisisbecausetheRPLswitchesarespreadmoreintimefortheRSSIestimator.ForFour-bit algorithm, most RPL parent switches occurred at the beginning of the simulation. Due to thetrickletimermechanisminRPL,havingalotofparentswitchesinashorttimespanonlyintroducedalimitedDIOmessageoverhead.

Figure23illustratesthestandardWSNscenario.An8x8gridoveranareaof175x175m2.Thesinkislocated

ontheleftcornerofthegrid.Thegreencirclearoundanodeindicatesitscommunicationrange,thegreycircleitsinterferencerange.Theothernodesinthegreencirclehaveacertainpercentageofsuccessfulreceptionrate,definedbythedistanceaccordingtothedefaultUDGMmodelusedinCooja.

H2020-GANo.645274 D2.4

43

(a)PacketDeliveryRatio

(b)Radioon-time

(c)Numberofhops

H2020-GANo.645274 D2.4

44

(d)Numberofretransmissions

(e)RPLparentswitches

(f)NumberofDIOmessages

H2020-GANo.645274 D2.4

45

Figure24Resultsgatheredwhileexecutingthestandardscenariofordifferentlinkestimators.

Densescenario

Thedensescenarioconsistsof144sensornodesinatwelvebytwelvegrid,inthesame175x175m2area. The node density is clearly higher because there are more nodes in communication range(greencircle)andinterferencerange(greycircle)comparedtothestandardscenario,asillustratedinFigure 25. The communication range (45m) and interference range (60m) settings are identicalcomparedtothestandardscenario.Thepositionofthesink isagainchosen inacornerofthegrid(upperleft).Alsothesamesendintervalboundaries(20secsand40secs)areapplied.

Figure25illustratesthedenseWSNscenario.An12x12gridoveranareaof175x175m2.Thesinkisagain

locatedontheupperleftcornerofthegrid.Thenodedensity(e.g.numberofnodesincommunicationrange(greencircle)andinterferencerange(greycircle)isclearlyhigherinthisscenario.

TheresultsshowninFigure26comparetheperformancefordifferentlinkestimatorsinthestandardscenario vs. thedense scenario. Because there aremuchmorenodes in the samearea, there aremanymorepossiblepaths to thesink.Also internal interferencewillbehigher. Indeed, theresultsclearly show that the overall network performance is lower for all link estimators in the densescenarioduetothehigher interference.Therelativedifferencebetweenthe linkestimators isalsohigherinthedensescenariovs.thestandardscenarioasexpected.

Thepacketdeliveryratiodrops(chart(a))significantlyforalllinkestimators,especiallyfortheFour-bit and RSSI based estimators which is also reflected in the higher number of per packet MACretransmission(chart(d)).Thisexplainsthehigherradioon-time(chart(b)),andconsequentlyenergyconsumption.FuzzyLQEandETXshowthehighestPDRfora limitedincreaseinradioon-time.TheRPLoverheadw.r.t.parentswitchingiscomparableinthedensescenariovs.thestandardscenario,except for theFour-bitandRSSIbasedestimators (aswas thecase in standard scenario).Hereweagain see a substantial increase in RPL parent switches (chart (e)). This does not yield more DIOmessages (char (f)) however, indicating that the parent switches occurred in the beginning of thesimulation.

H2020-GANo.645274 D2.4

46

(a)PacketDeliveryRatio

(b)Radioon-time

(c)Numberofhops

H2020-GANo.645274 D2.4

47

(d)Numberofretransmissions

(e)RPLparentswitches

(f)NumberofDIOmessages

Figure26Comparisonofresultsbetweenthestandardanddensescenariofordifferentlinkestimators.

H2020-GANo.645274 D2.4

48

Sparsescenario

Inthesparsescenariothestandard8x8grid inanareaof175x175m2wasused.Thenodedensitywasreducedbyloweringthecommunicationrangeandinterferencerangesettingsto30mand45mrespectively.As illustrated inFigure27, thenumberofnodes incomm.and interferencerangeareclearlylower.Thesamesinkpositionandsendintervalboundarieswereapplied.

Figure27illustratesthesparseWSNscenario.Thestandard8x8gridoveranareaof175x175m2isusedbut

thecommunication(greencircle)andinterferenceranges(greycircle)areloweredtodecreasethenodedensity.

TheresultsshowninFigure28comparetheperformancefordifferentlinkestimatorsinthestandardscenariovs.thesparsescenario.Becausetherearemuchlessnodesinthesamearea,therearealsolesspossiblepathstothesink.Thisresultsinahighernumberofhopsrequiredtoreachthesinkasclearlycanbeseeninchart(d).Theinternalinterferenceshouldhoweverbelower.Becauseofthis,thechoiceofthelinkestimatorshouldmakelessdifference.Indeed,theresultsclearlyshowthattheoverallnetworkperformanceisverysimilarforalllinkestimatorsinthesparsescenario.Therelativedifference between the link estimators is alsomuch lower in the sparse scenario vs. the standardscenarioandthedensescenario,asexpected.

The packet delivery ratio (PDR, chart (a)) is nearly equal for all link estimators. The radio on-time(chart(b))isslightlylowerforthesimplerobjectivefunction0ortheETXlinkestimatorscomparedtothe more complex Four-bit and Fuzzy LQE. This can be related to the higher RPL overhead.SignificantlymoreRPLparentswitches(chart(e))arerequiredbeforethenetworkreachesastablestate.Again,becausemostparent switchesoccur in thebeginningof the simulation, theeffectontheDIOmessages(chart(f))islesssubstantial.

H2020-GANo.645274 D2.4

49

(a)PacketDeliveryRatio

(b)Radioon-time

(c)Numberofhops

H2020-GANo.645274 D2.4

50

(d)Numberofretransmissions

(e)RPLparentswitches

(f)NumberofDIOmessages

Figure28Comparisonofresultsbetweenthestandardandthesparsescenariofordifferentlinkestimators.

H2020-GANo.645274 D2.4

51

Overallconclusion

Overall,wecanseethatthemorecomplexlinkestimatorsyieldahigherPDR.ThisrequireshowevermoreRPLoverheadand,consequently,ahigherenergyconsumption.Table8ranksthedifferentlinkestimators based on packet delivery ratio (PDR) and radio on-time (ROT) for the three scenarios.Notethatthisonlygivesanindicationwhichlinkestimatorgivesthebesttrade-offbetweenPDRandROT because the relative differences between the different ranks is not visible anymore. In thestandard scenario and dense scenario, Fuzzy LQE could be chosen as best trade-off if energyconsumptionisslightlylessimportant,otherwiseETXisabetterchoice.Inthesparsescenario,morecomplexestimatorsfailtoproducesignificantlybetterPDR,hencetheobjectivefunction)orETXarethemostappropriatechoices.

Table8ranksthedifferentlinkestimatorsbasedonpacketdeliveryratio(PDR)andradioon-time(ROT)forthethreescenarios.

Scenario Metric 1 2 3 4 5 6

StandardPDR FuzzyLQE Four-bit RSSI OF0 ETX LQI

ROT OF0 ETX FuzzyLQE LQI Four-bit RSSI

DensePDR FuzzyLQE ETX OF0 Four-bit LQI RSSI

ROT OF0 LQI ETX FuzzyLQE Four-bit RSSI

SparsePDR FuzzyLQE ETX Four-bit OF0 LQI RSSI

ROT OF0 ETX LQI FuzzyLQE RSSI Four-bit

d. NextStepsNumerousotheraspectscouldbeaddedinthisshowcase,enablingustoinvestigatetheimpactwhilechangingthemandallowingtomakemorefoundedconclusionsw.r.t.themostoptimallinkestimator.Thefollowingnon-exhaustivelistgivessomepotentialaspectsthatshouldhaveanimpactonthelinkestimatorandthatcouldbeaddedtothisshowcase.

• Trafficpattern(periodic,asymmetrical,randombursts).• MACprotocol(TDMAvs.CSMA).• MACdutycycle(e.g.numberofslotsorchannelcheckrate).• MAXnumberofMACretransmissions.• Externalinterference.• Nodemobility.

Thankstotheadvancesmadeinyear1and2oftheproject,suchparameterscouldalsobeadded.Tomakethisevemoreusefulfortheresearchers,abenchmarkingtoolboxshouldbeaddedontopoftheUPIsthatallowstheexperimentertoexecutecontrolprogramswithdifferentparametersettings.2.6.3 MultihoploadawareMACadaptationsInrecentyears, ithasclearlyemergedthatWi-Finetworkperformancecandramaticallydegradeinmulti-hop connected topologies and in high-density node scenarios. These conditions are likely tooccurwhenmultiplenetworkscoexistor largeaccess infrastructure isdeployed.Themain reasonsforthedegradationincludethestarvationandunfairnessphenomenaofCSMA-basedprotocolsduetoamismatchinthelocalviewsofthewirelessmediumamongthenodes,andduetothehighlevelofcontentionwhenthenetworkisheavilycongested.

H2020-GANo.645274 D2.4

52

In this showcase,we propose a solution formitigating the performance impairments of CSMA/CAprotocols in multi-hop topologies based on the dynamic adaptation of the contention processexperienced by nodes in the wireless network. A distributed protocol, called REACT, is used tonegotiate the channel airtime for a node as a function of the traffic requirements of itsneighborhood, taking intoaccountbandwidth reserved for thecontroloperations.Amechanism isprovidedforanodetotuneitscontentionwindowdependingonitsallocatedairtime.Differentfrompreviousschemes,anode’scontentionwindowisfixed insizeunlessthetrafficrequirementsof itsneighborhood change. The scheme is implementedon legacycommercial 802.11devicesexposingWiSHFULUPIs.

REACT protocol: Channel allocation in wireless networks can be viewed as a resource allocationproblemwheretransmitterscorrespondtodemandsandreceiverstoresources.ThegeneralideaonthebasisofREACThasbeenproposedinFigure29orslottedsystems,fornegotiatinganallocationofchannelairtimesbymeansofanauctionmechanism.Eachnoderunsanauctioneerthatmaintainsanoffer, the maximum airtime consumed by any adjacent bidder. Similarly, each node also runs abidder that maintains a claim, the airtime the bidder intends to consume at adjacent auctions.Through updates of offers and claims, the auctioneers and bidders converge on an allocation ofairtimes.Thedetailsof theprotocolaredescribed in .Figure30.ConsideranetworkwithNnodes.Starting from the channel airtime demand (w1, w2, … wN) performed by each node of the networkaccording to the running applications, at the end of the auction mechanism the REACT protocolassignsthenormalizedairtimes(s1,s2,…sN)thateachnodeisallowedtoconsume.Inordertolimitthechannelairtimetotheassignedvalue,wedesignedaschemefordynamicallytuningthecontentionwindow as a function of the transmission grants observed in a givenmonitoring interval, channelbusytimesandcollisionrate.Figure29showsanexampleofresourceallocationperformedbyREACTinanetworktopologywith6nodesandaspecifictrafficdemand.ThemainideaofREACTisequallysharingtheavailablechanneltimeamongconflictingnodes(whichinclude2-hopsnodesconsumingchanneltimeforthedesiredreceivernode)untilthetrafficdemandissatisfied,andfurtherdividingtheexcesscapacitytothenodesforwhichthedemandisnotsatisfied.

Figure29-ExampleofREACTresourceallocation

Tuning of the contentionwindow. In our showcase,we implemented amechanism for tuning thecontentionwindowofeachnodeinordertoachieveadesiredairtimeallocation.Therationaleofthetuning mechanism is depicted in Figure 30, which represents a sequence of channel accessesperformedbyareferencenodeiinanobservationintervalC.Theairtimeofnodeiiscoloredblue.Itincludes RTS, CTS, DATA, and ACK transmissions, as well as the inter-frame spaces between theframes sent in the same channel access (also called transmission opportunity). The observationinterval C is divided into sub-intervals c(1), . . . , c(k) delimited by the start of a transmission

H2020-GANo.645274 D2.4

53

opportunitygrantedtonode i.Each intervalc(k) includestheairtimeofnode iandalso itsbackoffexpiration time,which in turnsdependson the initialbackoffvalueandon the timethebackoff isfrozenduetothetransmissionofothernodes.Duringtheseintervalsf(k),inwhichbackoffisfrozen(colouredingray),thebackoffcounterisnotdecremented.Suchbackofffreezingcanbetriggeredbythephysicalcarriersense,aswellasbythevirtualcarriersensemechanismenabledbythereceptionofCTSframes.Assumethatchannelaccessesaremanagedbymeansofa4-wayhandshake.Lettx,xin{RTS,CTS,DATA,ACK}bethetimetotransmitx.TheairtimeinanaccessintervaldependsontheoutcomeofRTSandDATAtransmissions. ItcanvaryfromaminimumoftRTS+DIFSwhentheRTSfails,toamaximumoftRTS+SIFS+tCTS+SIFS+tDATA+SIFS+tACK+DIFSwhenthetransmissionissuccessful.

Figure30-successivechanneltransmissionforataggednode-i

Fromrenewaltheory,theportionofchannelresourcesallocatedtonodeimaybeexpressedas:

s! =E[a]E[c]

=E[a]

E a + E f + E W /2 ∙ σ

whereE[a]istheaverageairtimeandE[c]istheaveragechannelaccessinterval.ThisintervalcanbeexpressedasasumofE[a],theaveragetimethebackoffisfrozenE[f],andtheaverageinitialbackoffvalueE[W]inslotsmultipliedbyσ,thelengthintimeofabackoffslot.

The average airtime can be computed by considering the probability pRTS of a successful RTStransmission,andtheprobabilitypDATAofasuccessfulDATAtransmissionastRTS+pRTS·(tCTS+tDATA+2·SIFS)+pDATA·tACK+DIFS.Ingeneral,afterasuccessfulhandshake,it isassumedthatdatapacketsareprotectedbythevirtualcarriersense.However,inmulti-hopscenarios,itmayhappenthatoneof the receiver’sneighborsexperiencesacollisionon the receptionofaCTSdue toa transmissionoriginatedbyanodehiddentothereceiver; inturn, itmayinterferewiththeDATAreception.Theaveragetimethebackoff is frozendependsonthenumberofneighborsandontheir traffic,whilethe average backoff value depends on the contention window settings. The channel accessprobability depends on the average contention window value rather than on the specific backoffalgorithm. Let Wi be the contention window value of node i configured during an observationintervalC.

LegacyWi-Finodes canestimate the currentallocated rate si as a functionof the totalnumberofchannelaccessesn,thetotaltimewiththebackofffrozenF,andtotalairtimeAobservedinC:

𝑠! =a(k)!

!!!a(k)!

!!! + f k +!!!! W!/2 ∙ σ ∙ n

=A

A + F +W!/2 ∙ σ ∙ n=AC

whereWi/2·σ·nisanapproximationofthetotaltimerequiredforthebackoffcountdown.

Let Wi+ be a new setting of the contention window, and let E[a]+ and F+, respectively, be the

predictionoftheaverageairtimeandtotaltimebackoffisfrozenexperiencedunderthenewsettingofthecontentionwindowinanewchannelobservationintervalequaltoC+.Iftheothernodesused

H2020-GANo.645274 D2.4

54

fixedcontentionwindows, then thedynamic tuningofWi+doesnotaffect the collisionprobability

experiencedby the targetnode. Indeed, thecollisionprobability isgivenby theprobability thatatleastoneinterferingnodeistransmitting,giventhatthetargetisalsotransmitting.ItfollowsthatthenewWi+valuehasan impactonthechannelaccessprobabilityofstation ibutnotontheaverageairtime in each channel access (i.e., E[a]+ = A/n), which only depends on the collision probabilityexperiencedbyRTSandDATAframes.Similarly,wecanassumethatthefractionofchanneltimenotallocated to node i that is sensed as busy is not affected by the contention window tuning andtherefore equals to the previously experienced one, i.e., to F/(C-A) . The total time the backoff isfrozenF+isobtainedastheproductbetweenthisratioandthetotaltimenotspentintransmissionduringtheintervalC+.

Toobtain thedesired rate si* in thenext tuning interval C+, it is required to achieve a numberofchannel accesses equal to n* = si*· C+/E[a]+ = si* · C+ · n/A. Therefore, the total time spent forn*countdownsofthebackoffhastoequalizethedifferencebetweenthechanneltimenotallocatedtonodei,i.e.,C+(1–si*),andthetimethebackoffisfrozen.Itfollowsthatthecontentionwindowcanbetunedas:

W!∗ =

2σ∙A(1 − s!∗)n ∙ s!∗

∙ 1 −F

C − A

Theabovealgorithmsformonitoringthechannelbusytimesandallocatedairtimeandfortuningthecontention window as a function of the desired rate can be easily implemented in theWiSHFULframeworkontopoftheradioUPI,asdescribedinwhatfollows.

a. PresentationofusedUPIsandthecontrolprogramThis section presents a detailed description of the software architecture used for implementingREACT in the WiSHFUL framework, by also specifying the UPIs utilized in the showcase, theexperiment setup and the experimental results. The architecture is summarized in Figure 31. ThemainfunctionalrequirementsforimplementingREACTare:i)enablingtheinjectionofcustomframesfromuser-space for sending themessagesusedby theauctionprotocol, ii) interactingwithdriver-level statistics formeasuring channelbusy timesandairtimes; iii) configuring thenode contentionparameters(namely,thecontentionwindow)dynamically.TheoverallmechanismisimplementedbydefiningaREACTcontrolprogram,which includes:a)theREACTprotocol logicformanagingclaimsandoffers,andb)thedynamictuningoftheContentionWindowvalueasafunctionoftheobservedchannelparameters. The interfacesbetween theuser-space controlprogramand thekernel-spaceprimitivesisnativelyprovidedbytheWiSHFULUPIs.

Figure31-REACT802.11Architecture

Thefollowingtable(Table9)summarizestheUPIsusedbytheREACTcontrolprogram:

H2020-GANo.645274 D2.4

55

UPI Usage

net.inject_frame() send custom broadcast frames containing REACTclaimsandoffers

net.sniff_layer2_traffic() receiveframesanddissectREACTcontrolmessages.

set_mac_access_parameters() set CWmin=CWmax=CW for the data queue,enforcingthedesiredCWvalue

radio.get_measurements() get packet transmission statistics: data_count andrts_count areused to estimate the current freezingtimeandpredictthefutureone.

Table9-UPIlist

OneachWi-Finode,alocalcontrolprogramnamedreact()isactivatedbytheglobalcontroller(Table12)withthefollowinginput:

{"iface":"wlan0","i_time":1,"bw_req":bw_req[i_flow],enable_react":True}

where i_time is the observation interval to be used for updating the statistics and the contentionwindowvalue,bw_req represents the amountof traffic desiredby thenodeexpressed in Kb/s. Inorder to increase the modularity of the control program, two configuration files are introduced:node_info.csv (Table 10) and experiment_info.csv (Table 13). These files contain a tabulardescriptionofthenodesandexperimentsetup,respectively.Theglobalcontrolprograminspectstheconfigurationfiles,setupad-hocconnectionsbetweennodes,activatestheREACTalgorithmoneachnodeandstartsthesourcetrafficbymeansofiperfapplications.#hostname,driver,eth0_ip,freq,txpower,wlan0_ip nodezotacb3,ath9k,10.11.16.22,5180,1,192.168.0.1 nodezotacb4,ath9k,10.11.16.33,5180,1,192.168.0.2 nodezotacd3,ath9k,10.11.16.24,5180,3,192.168.0.3 nodezotaci3,ath9k,10.11.16.29,5180,3,192.168.0.4 nodezotack3,ath9k,10.11.16.31,5180,1,192.168.0.5 nodezotack4,ath9k,10.11.16.42,5180,1,192.168.0.6

#src,dst,bw_req,port,t_start,t_stop 1,2,6000,5012,1,100 2,3,6000,5021,1,100 3,4,6000,5034,1,100 4,5,6000,5045,1,100 5,6,6000,5056,1,100 6,5,6000,5065,1,100

Table10-node_info.csv Table11-experiment_info.csv

For sake of simplicity, the following code snapshot highlights only some relevant parts of theexperimentsetup.[…] #SETUP NODES, RUN REACT nodes_info_path=args['--nodes'] [hosts,driver,eth_ip,freq,tx_power,wlan_ip]=set_hosts(nodes_info_path); experiment_info_path=args['--experiment_info'] if experiment_info_path: [src,dst,bw_req,port,t_start,t_stop]=experiment_setup(experiment_info_path) for i_flow in range(0,len(src)): for jj in range(0, len(nodes)): if nodes[jj].ip == eth_ip[int(dst[i_flow])-1]: lcpDescriptor = controller.node(nodes[jj]).hc.start_local_control_program(program=react) msg={"iface":"wlan0","i_time":1,"bw_req":bw_req[ i_flow],enable_react":True} lcpDescriptor.send(msg) […]

Table12-Globalcontrolprogram:thereactcontrolprogramisactivatedoneachnode.

Thelocalcontrolprogram(showninTable13)sentbyremotecontrollertolocalnodesiscomposedbythreemainthreadswhichruncontinuously:

H2020-GANo.645274 D2.4

56

- REACT control message sender: (send_REACT_msg ) this thread collects the informationrequired for specifying offers and claims and send REACT control messages by means ofcustombroadcast L2messages. Controlmessage are sent at regular intervals of 100msbyusingtheUPIfunctioninject_frame().TheL2messageisforgeddirectlybythelocalcontrolprogram.Ateachtransmission,thestateinternalparametersofthenode(offers/claims)areupdated.

- Controlmessage receiveranddissector: (sniffer_REACT) this thread receives L2 framesanddissects REACTmessages. Themessages are processed in bursts collected at each second,while L2 frame are received and buffered by using the UPI function sniff_layer2_traffic().Whenanewburstofcontrolframesisprocessed,thelocalcontrollerupdatesthelistofstateparameters(offers/claims)foreachneighbornode.

- Contentionwindowsupdate:(update_cw)thisthreadisbasedontwoindependentinputs:i)packetandchannelstatisticsretrievedbymeansoftheUPIfunctionget_measurement();ii)desiredratecomputedbytheREACTauctionandstoredintheinternalstateparameters.Thedesiredrateisthenmappedintoanewcontentionwindowvaluesasafunctionofcollisionrate and channel busy times, as described in the previous section. When a new CW iscomputedacall totheUPIfunctionset_mac_access_parameters() isperformedtoenforcethenewCWvalue.

def main(iface="wlan0",i_time=1,bw_req=0,enable_react=False): #INIT REACT INFO init(iface); try: #Thread transmitter _thread.start_new_thread( send_REACT_msg,(iface,i_time,bw_req,enable_react ) ) #thread receiver _thread.start_new_thread( sniffer_REACT,(iface,i_time ) ) #update CW _thread.start_new_thread(update_cw,(iface,i_time,enable_react,i_time)) except (Exception) as err: print ( "exception", err) pass while 1: pass

Table13–REACTlocalcontrolprogrammain

def send_REACT_msg(iface,i_time=1,bw_req,enable_react): #TX my_mac =str(netifaces.ifaddresses(iface)[netifaces.AF_LINK][0]['addr']) while True: rate = min((float)( C ),( (bw_req*C)/float(MAX_THR)) ); neigh_list[my_mac]['w']=rate try: pkt_to_send={}; neigh_list[my_mac]['t']=float(time.time()) pkt_to_send['t']=neigh_list[my_mac]['t'] pkt_to_send['claim']=neigh_list[my_mac]['claim'] pkt_to_send['offer']=neigh_list[my_mac]['offer'] json_data = json.dumps(pkt_to_send) timeout = 30 #sec for key,val in neigh_list.items(): if float(time.time())-val['t'] > timeout: neigh_list.pop(key) update_offer()

H2020-GANo.645274 D2.4

57

update_claim() # REACT variables updated, transmit! send_ctrl_msg(iface,json_data) tx_interval=i_time/10 time.sleep(tx_interval - ((time.time() - starttime) % tx_interval) def send_ctrl_msg(iface,json_data): a=scapy.all.RadioTap()/scapy.all.Dot11(addr1="ff:ff:ff:ff:ff:ff", addr2=my_mac, addr3="ff:ff:ff:ff:ff:ff")/json_data mon_iface="mon0" controller.net.inject_frame(iface=mon_iface,frame=a,is_layer_2_packet=True,tx_count=1, pkt_interval=0)

Table14–firstREACTthread:sendcontrolmessage

def sniffer_REACT(iface,i_time): #scapy.all.sniff(iface=mon_iface, prn=updateAction(iface,i_time),store=0) call_timeout=i_time/2 call_count=2000 while True: # pktlist = scapy.all.sniff(iface=mon_iface, timeout=call_timeout, count=call_count,store=1) # UPI pktlist = controller.net.sniff_layer2_traffic(iface=mon_iface, sniff_timeout=call_timeout, ipdst=None, ipsrc=None) for pkt in pktlist: try: rx_mac=str(pkt.addr2) if rx_mac == my_mac: pass else: payload=bytes(pkt[2]) if 'claim' in str(payload): payload='{'+re.search(r'\{(.*)\}', str(payload) ).group(1)+'}' curr_pkt=json.loads(payload) neigh_list[str(rx_mac)]=curr_pkt; curr_pkt['t'] = float(time.time()) update_offer(); update_claim(); except (Exception) as err: if debug: print ( "exception", err) pass

Table15–SecondREACTthread:controlmessagedissector

""" update CW decision based on ieee80211 stats values and virtual channel freezing estimation """ def update_cw_decision(iface,enable_react,sleep_time): #get stats global my_mac global cw global cw_ global data_count_ global rts_count_ CWMIN=15 CWMAX=2047 UPI_myargs = {'interface' : 'wlan0', 'measurements' : [UPI_R.dot11RTSSuccessCount,UPI_R.dot11RTSFailureCount] } pkt_stats=controller.radio.get_measurements(UPI_myargs) pkt_size=1534 if pkt_stats: if rts_count_ == 0 and data_count_ == 0:

H2020-GANo.645274 D2.4

58

data_count = pkt_stats['dot11RTSSuccessCount'] - data_count_ rts_count = pkt_stats['dot11RTSSuccessCount'] + pkt_stats['dot11RTSFailureCount'] - rts_count_ data_count_=pkt_stats['dot11RTSSuccessCount'] rts_count_=pkt_stats['dot11RTSSuccessCount'] + pkt_stats['dot11RTSFailureCount'] return data_count = pkt_stats['dot11RTSSuccessCount'] - data_count_ rts_count = pkt_stats['dot11RTSSuccessCount'] + pkt_stats['dot11RTSFailureCount'] - rts_count_ data_count_=pkt_stats['dot11RTSSuccessCount'] rts_count_=pkt_stats['dot11RTSSuccessCount'] + pkt_stats['dot11RTSFailureCount'] tx_goal=0 I=0 dd = sleep_time; gross_rate = float(CLAIM_CAPACITY)*float(neigh_list[my_mac]['claim']); busytx2 = 0.002071*float(data_count) + 0.000046*float(rts_count); #how much time the station spent in tx state during the last observation internval SIFS=16 #usec tslot=9e-6 #usec freeze2 = float(dd) - float(busytx2) - cw_/float(2)*float(tslot)*rts_count; #how long the backoff has been frozen; if rts_count > 0: avg_tx = float(busytx2)/float(rts_count); #average transmission time in a transmittion cycle psucc = float(data_count)/float(rts_count); else: avg_tx=0 psucc=0 if avg_tx > 0: tx_goal = float(dd*gross_rate)/float(avg_tx); else: tx_goal = 0 freeze_predict = float(freeze2)/float(dd-busytx2)*float(dd-dd*float(gross_rate)) ; if tx_goal > 0: cw = 2/float(0.000009) * (dd-tx_goal*avg_tx-freeze_predict)/float(tx_goal); if cw < CWMIN: cw_=CWMIN elif cw > CWMAX: cw_=CWMAX else: cw_=cw cw_= pow(2, round(math.log(cw_)/math.log(2)))-1; # ENFORCE CW qumId=1 #BE aifs=2 cwmin=int(cw_); cwmax=int(cw_); burst=0 if enable_react: setCW(iface,qumId,aifs,cwmin,cwmax,burst); thr=(data_count)*1470*8/float(dd*1e6); out_val="%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f" % (time.time(),dd,data_count,rts_count,busytx2,gross_rate,avg_tx,freeze2,freeze_predict,tx_goal,I,cw,cw_,psucc,thr) my_ip=str(netifaces.ifaddresses(iface)[netifaces.AF_INET][0]['addr']) #my_ip.replace(".","_")

H2020-GANo.645274 D2.4

59

out_file="{}.csv".format(my_ip); with open(out_file, "a") as myfile: myfile.write(out_val+"\n")

Table16–ThirdREACTthread:ContentionWindowtuning

def setCW(iface,qumId,aifs,cwmin,cwmax,burst): phy=getPHY(iface); #SETCW UPI edcaParams = edca.EdcaQueueParameters(aifs=aifs, cwmin=cwmin, cwmax=cwmax, txop=burst) edcaParams = edca.EdcaQueueParameters(aifs=1, cwmin=1, cwmax=1, txop=1) # UPI controller.radio.set_mac_access_parameters(iface=iface,queueId=qumId,queueParams=edcaParams)

Table17-CalltoUPItoenforceanewContentionWindowvalue

b. ExperimentsetupandresultsExperiments have been conducted in Wilab2 testbed, using ZOTAC Wi-Fi nodes equipped withAtheros chipset AR9xxx family and running the ath9k driver. Experiments aim to analyze theperformancegainprovidedbyREACTinpresenceofmulti-hoptopologieswithhiddennodes.Herewepresenttwodifferentexperiments:1)anexperimentrunningonatopologywiththreenodesinachain;2)amorecomplexexperimentwith6nodes,whoseconnectivitygraphincludesamaximumdistanceofthreehopsamongthenodes.Ineachexperiment,wecomparethepercentageofchannelairtime(alsocalledaccesspersistency)achievedbyeachnode,underlegacyDCFandinpresenceoftheREACTmechanism.

Figure32–Networktopologyofthefirstexperiment:chainofthreenodes(nodeA:ZotacD6,nodeB:

ZotacG6,nodeC:zotacJ6).

Figure32showstheabstractnetworktopologywithachainofthreenodes,wherethedashedgreylinesindicatethenodeavailablelinks,andthereallocationofthenodesinthewilabttestbed.EachnodesendsUDPtrafficinsaturationmode.Packetsizeisfixedto1534byteandthedatarateisfixedto6Mbps. In the initialexperiment,weactivate threetraffic flows,asdepictedFigure32,byusinglegacy802.11withRTS-CTS.Figure33theairtimeachievedbyeachnode:it isevidentthatchannelaccess performance exhibit a clear unfair behaviour. In particular, node B access the channel for

B" C"A"

H2020-GANo.645274 D2.4

60

about41%ofthetime,whilenodeAandnodeCsucceedinaccessingthechannelonlyforabout12%and 23% of the time.When the experiment is repeated using REACT, all the nodes achieve a fairrepartitionofthechannelcapacity,withabout30%ofchannelairtimegrantedtoeachone.

Figure33–NormalizedairtimeachievedunderlegacyDCFwithRTS/CTSenabledinthechaintopology.

Figure34–NormalizedairtimeachievedunderREACTinthechaintopology.

Inthesecondexperiment,weconsiderthemorecomplextopologyshowninFigure35,inwhichweagain illustrates theabstract topology, theavailable linksandtheactive traffic flows (topdiagram)andtherealpositioningofthenodeswithinthewilabtestbed(bottomdiagram).Inthisexperiment,wetesttheREACTperformanceunderdynamicloadconditions,bysequentiallyactivatingthetrafficflows starting at nodes A, B, C, D, E, and F. The effects of the protocol is tuning the contentionwindow as a function of the channel airtime allocated to eachnode: therefore, in dynamic trafficconditions,theaveragecontentionwindowvalueschangeaftertheactivationofeachtrafficflow,asshowninFigure36.

0 10 20 30 40 50 600

10

20

30

40

50

60

70

80

90

100

3 FLOWS−MESH Access Persistency

time [s]

airti

me[

%]

AVERAGE (STD−DEVIATION)nodeA=12.65(11.27)nodeB=40.84(18.38)nodeC=23.15(17.95)

nodeAnodeBnodeC

0 10 20 30 40 50 60 70 800

20

40

60

80

100

120

3FLOWS airtime

time [s]

airti

me[

%]

AVERAGE (STD−DEVIATION)nodeA = 27(3) 26(3)nodeB = 26(2) 26(2)nodeC = 26(2) 26(4)

nodeAnodeBnodeC

H2020-GANo.645274 D2.4

61

Figure35-6nodestopologynodeA:zotacB3,nodeB:zotacB4,nodeC:zotacD3,nodeD:zotacI3,nodeE:

zotacK3,nodeF:zotacK4

Figure36-REACTperformanceinmulti-hoptopologieswithdynamictrafficconditions:CWvaluesand

airtimesachievedbyeachnode.

c. NextstepsTheproposedREACTschemeseemsverypromisingforregulatingtheaccessrateofnodesinmulti-hop topologies, where it is well known that greedy access behaviors can lead to significantperformance impairments.Weareplanningtoconsiderseveralpotentialextensionsoftheschemeforrealnetworkdeployments.First,wewouldliketogeneralizetheconceptofchannelallocationsbyconsideringmulti-hoptrafficflows.Whilethecurrentschemeworksbyconsideringeachnodeasindependent and by sending claims, which depend only on the local application demand, in realtopologiesitmayhappenthatthesameapplicationflowhastotraversemultipleconsecutivelinks.Self-contention of multi-hop traffic flows is a very critical phenomenon for ad-hoc networks.Therefore, it could be very relevant to mitigate self-contention by means of channel allocations

C" D"

B"

A"

F"

E"

H2020-GANo.645274 D2.4

62

whichtakeintoaccounttheapplicationdemandacrossconsecutive links.Second,wewould liketoexplore another generalization of the auction mechanism in general topologies in which someneighbor nodes waste channel times of a given transmitted, but send packets that cannot becorrectly demodulated. In other words, we would like to improve the auction robustness in ascenarioinwhichthecarriersenserangeandthetransmissionrangecanbedifferent.Inthiscase,some claims have to be indirectly estimated by the nodes, which cannot demodulate the REACTcontrolmessages. Finally, wewant to study, from a theoretical point of view, the stability of theREACTallocations.

H2020-GANo.645274 D2.4

63

3 DefinitionofShowcasestobeimplementedinYear3Thefollowingshowcasesareenvisionedforimplementationinthethirdyearoftheproject.Theyaredescribedinearlyformhere.

3.1 OTAupdatesusingGITARforWSNsMoreandmore Internet-of-thing (IoT) devices are currently beingdeployedand connected to theInternet.While therearemanybenefits, therearealsonumerouspossiblepitfalls,especiallyw.r.t.security. Latest example of a serious security problem was the incidents where smart applianceswerehackedandrepurposedtoparticipateinamassivedistributeddenialofserviceattacks(DDOS).

Oneofthekeyshortcomingsofsuchdevicesisthatthefirmwareonmostofthesedevicescannotbeupdatedover-the-airsecurely.Thisimpliesthatitisuptotheendusertoensurethatthefirmwareofhis IoTdevices is up todatewith the latest securitypatches.Moreover, softwareupdates arenotonlynecessary for securityupdatesbut canalsobeused toperformbug fixes, featureadditionorperformanceimprovements.

TheGITAR framework [19]offersanumberofbuildingblockswhich canbeused toenablepartialsoftware updates of constrained devices in the IoT.Most importantly, GITAR provides a dynamiclinker that can update/add/remove software modules at run-time. Secondly, GITAR enables toconvert a static software module into a dynamic software module without requiring any codechanges. To facilitate these features, the GITAR framework automatically embeds a componentobjectmodel(COM)insidethesoftwaremodule.

3.1.1 OverviewInthisshowcase,thebuildingblocksofGITARwillbeintegratedinWiSHFULandusedtoenableover-the-air (partial) software updates of a WSN devices. The UPI_M interface will be extended withfunctions that enable it to distribute, install and remove softwaremodules in one ormore nodesafterdeployment.

GiventheconstrainednatureofWSNdevices,specialattentionisrequiredintheprotocolusedforover-the-air software distribution. It should be possible to evaluate different combinations ofmac/routing/transport layer protocols in different scenario’s (number of hops, link quality,interference,etc..). For thispurpose, thealreadyavailableUPI_R/N functionwillbeextended (e.g.switchingrouting/transportprotocoliscurrentlynotpossibleatrun-time).

To fullyexploit thenewfeaturesGITARbrings intoContiki, somemodificationsare required in theContiki network stack.More specifically, currently each layer is statically linkedwith itsupper andlowerlayermakingitimpossibletoupdatelayersseparately.Toovercomethis,thenetstackmoduleinContikineedstoberefactored.

3.1.2 GoalsThemain goal is to extend theWiSHFUL UPI_M interface with functions that enable over-the-airupdates of WSN nodes. The second goal is to use/extend the UPI_R/N interfaces for evaluatingdifferentsoftwaredistributionmethods.ThethirdgoalistorefactortheContikinetworkstackininseparatelayerssothateachcanbeindependentlyupdatedand/orreplaced.

3.1.3 BreakthroughsThefollowinglistiteratesthebreakthroughscreatedinthisshowcase:

• ThepossibilitytouseUPI_Mforsendingand installing(partial)softwareupdatestoconstrainedIoTdevices.

• TheabilitytoevaluatedifferentsoftwaredistributionmethodsusingUPI_R/N.• Enabling run-time switching between different versions of a protocol layer in Contiki using

UPI_R/N.

H2020-GANo.645274 D2.4

64

3.1.4 MethodologyThefollowingstepswillbetakentorealizethisshowcase:

1) Set-upabasicsystema) Addabasicsoftwaredistributionmethodb) IntegrateGITARdynamicupdatecapabilitiesintoWiSHFULframework.c) Extend theUPI_M interfacewith functions todistribute softwaremodules tooneormore

nodesd) ExtendtheUPI_Minterfacewithfunctionstocontrolthesoftwareinstallationprocess.

2) Evaluatebasicsystema) TheoreticallyanalysetheoverheadintroducedbyOTAs.b) Verifytheoreticalanalysisbyevaluatingsomescenario’sinreal-life.

3) Extendbasicsystema) Addingandevaluatingdifferentsoftwaredistributionalternatives.b) RefactorContikinetstacktoenableupdatingandswitchingseparateprotocollayers.

3.1.5 UseofWiSHFULFunctionalityDuringtheexperimentalevaluationofthisshowcase,WiSHFULUPI_R/Nfunctionalitywillbeusedtomonitor the overhead and optimize the behavior of OTA software distribution methods and theUPI_Minterfacewillbeextended.

3.2 ExtensionofMACadaptationinmulti-hoptopologies,basedondirectionalantennaandmultiplepathreservations

3.2.1 OverviewThisshowcaseisanextensionoftheYear2showcasebasedonthetopology-awareandload-awaredistributedallocationofairtimeinmulti-hoptopologies.Thebasicideaoftheoriginalshowcasewasusing an auction protocol, called REACT, for computing the airtimes required by the active trafficsources in the network, and a run-time tuning of the contention window for guaranteeing theallocatedairtimetoeachnode,whilereducingthecontentionlevelamonghiddennodes.

Weplantogeneralizetheairtimeallocationformoregeneraltrafficsources.Indeed,inYear2eachsourcedeliverstraffictowardsa1-hopneighbor,whileinrealad-hoctopologiesitislikelythatmulti-hopflowsareactive.Theseflowscausethewell-knownself-interferenceprobleminachainofnodestransmitting the flow frames. Therefore, it could be interesting to perform airtime allocations byconsideringthattheapplicationrequirementssignaledbyagivensourcenodeshouldbepropagatedalong a chain of nodes in a ‘multi-hop reservation chain’. A new auction protocol, as well as anestimatorofflowdemandsfortherelaynodes,havetobedesigned.

We plan to also consider another extension based on the availability of programmable antennas,integratedwithinanOC1extension.Theseantennascanbeusedforgeneratingmorecomplexmulti-hoptopologies(beingthecurrentdepthofthenetworklimitedtothreehops).Moreover,adynamicreprogramming of the antenna pattern, as a function of the transmission destination, could beperformedformitigating interference inthedirectionsdifferentfromthedesiredone.This featurecouldrequireafurtherprotocolextension,basedontheintroductionofdirectionairtimes.

3.2.2 GoalsOur goal is designing an effective and robust solution for delivery traffic in ad-hoc networks bymitigatinghiddennodeproblemsandself-interferencegeneratedbytrafficflowstraversingmultiplerelay nodes. Our solution is also an interesting cross-layer optimization, in which applicationrequirements andbandwidthdemands at relaynodes aremapped intoMAC layer (i.e. contentionwindow)andPHYlayer(i.e.antennabeam)configurationsatrun-time.

H2020-GANo.645274 D2.4

65

3.2.3 BreakthroughsTheproblemofchannelallocationsinmulti-hopnetworkhasbeenfacedfromdifferentperspectives,but no solution is currently prominent as the best performing one. Even the concept of channelreservationsincludedintothe802.11sextensionswiththemeshdeterministicaccessschemesisnotused in practical scenarios because of several technical limits (e.g. problem with nodesynchronizationinmeshnetworks).Weexpectthatoursolutioncanfillthisgap,betweentheoreticallimits and practical feasibility, because it is based on ‘loose’ node coordination, i.e. on a simpledistributed auction protocol which only requires infrequent messages exchanges among adjacentnodes.

3.2.4 MethodologyWewill study theextension toREACT,basedonmultipleairtimeallocationsofagiven traffic flowamongtheconsecutivenoderelays,bymeansofatheoreticaland/orsimulationapproach.Wewillalsoconsidertheimpactofdirectionalantennasforquantifyingchanneldemandsofeachnodealongaspecificdirection.Oncetheschemeisdefined,itwillbevalidatedandrefinedinrealexperiments.In parallel, we will study the physical topologies that can be configured in the WiSHFUL wilabttestbed,byusingprogrammableantennas.

3.2.5 UseofWiSHFULFunctionalityWiSHFULUPIfunctionswillbeusedfor:tuningthecontentionwindowofthenodesasafunctionoftheREACTallocations;programming theantennabeamsdynamicallyasa functionof thenext-hopdestination node; programming the antenna beams statically for specific topology requirements;implementing a non-standard control protocol for ad-hocnetworks basedon the extendedREACTprotocol.

3.3 Radio-basedindoorlocalization3.3.1 OverviewRadio-based indoor localizationtakesadvantageofthepervasivecoverageofWi-Fisignals.Despiteseveralsolutionsthathavebeenproposed in literature,precise localizationusingWi-Finetworks isstill an elusive target. This is because operational scenarios are fragmented and extremelyheterogeneous,thereforeaone-size-fits-allradiolocationsolutionisnotfeasible.WiSHFULflexibility,node programmability and controllability allow a programmable positioning scheme that can alsoexploitsnodecoordination.

In this showcase we validate this concept with a radio program that is specifically dedicated topositioning, taking inspiration from passive RADARs and from WIDAR [20]. The exemplary MACbehaviorandtheproposedscenarioaredepictedinFigure37,whereseveralAPscoordinatedbytheglobal controller running a radio program that, being specifically designed for positioning, has no

DATA

ACK

ACK

APsWMP

radioprogramDedicatedforposi7oning

TARGETLegacyIEEE802.11STA

PassiveprobeUSRP

Overheardtransmissions

Figure37Concepttopologyforexploitingdedicatedradioprogramsforlocalization

H2020-GANo.645274 D2.4

66

relevancefordataexchangebutpermitsevaluationofRTTintheoppositedirections.Thetargetrunsstandard DCF and it doesn’tmatter if it is associated to any of the APs. APs send periodic DATAframesthatsolicitthetarget.Foreachreceiveddataframe,thetargetwaitsforaSIFSandanswerswithoneACK.Asstandardized, thishappenseven if thetarget isnotassociatedto theAP.TheAPradioprogramisspeciallydesignedtoanswerthisACKwithanotherACK,sentafteraSIFS.ThesamehappensinturnswithallotherAPs.Theresultingframeexchangesareoverheardbyapassiveprobe,which analyzes the two-way propagation delay over several paths and compensate eventual SIFSheterogeneitybetweennodes.Moreover,weplantoexploitthepossibilitytorun-timeswitchfroman operating frequency to another in order to evaluate the RTT value experiencedwith differentcarrierfrequencies.

3.3.2 GoalsThe goal of this showcase is to demonstrate the advantages of using radio programs and controlprogramsthatarespecificallydesignedforlocalizationoveraprogrammablepositioning.Inordertokeepthescenariorealistic,weassumeWiSHFUL-enabledradioprogrammabilityonlyonthenetworkside.

3.3.3 BreakthroughsRealistic and applicable Wi-Fi based radio localization mechanisms exploit protocols of the IEEE802.11 standards. These protocols have been designed for guaranteeing performance in terms ofthroughputandfairnessamongnodes,but localizationpurposeswherenot initiallyaddressed.TheWMP platform, integrated in WiSHFUL facilities, permits to seamlessly switch between radioprograms fordata transportandradioprograms forpositioning,obtaining thebest fromthe latterwhilemaintainingbackwardcompatibilityonIEEE802.11protocolsonthetarget.

3.3.4 MethodologyThe proposed positioning-dedicated radio program resides onWMP-enabled APs in theWiSHFULtestbed,whilethetargetnodeisamobilerobotequippedwithlegacyDCF.DedicatedradioprogramswillbeusedforsolicitingthetargetandexploretheuseofmachinelearningapproachesexploitingtheWiSHFULintelligenceframework.

3.3.5 UseofWiSHFULFunctionalityThe showcase requires several WiSHFUL functionalities: the capability of changing the MAC andtuningitsparameters,ofaddingnewframetypes,ofgettinglow-levelstatisticsfromprogrammablenodesandlow-levelradioinformationfromUSRPs,usedasplatformsforadvancedsensing.

3.4 ExtensionofMACoptimizationsinhigh-densityscenarios,withonlinephase3.4.1 OverviewThisshowcaseisanextensionoftheYear2showcaseforMACadaptationinHigh-Densitynetworks.Thebasic ideaof theoriginal showcase is todetectpathological topologies indensenetworksandenforceappropriateactions. Intheoriginalshowcase,thedetectionphasewasrunofflineandwasbasedonsimulationdata,whiletheactionwasimplementedonlineintheWiSHFULtestbed.In this showcase we will include the detection phase as an online classification, exploiting theWiSHFULintelligentframeworkwhiletakingintoconsiderationthattheinterferenceconditionscanbeclassifiedintothreemainclassesdependingonthenetworkdensity[21].This showcase will extend the archetypal topology used in Year 2 showcase by including moregenericdensenetworksandwillexplore the improvementsdue the joint informationgivenby theadjacencymatrix(indicatingtheconnectivitybetweennodes)andthetransmissionmatrix(indicatingsourceanddestinationsoftrafficflows).

H2020-GANo.645274 D2.4

67

3.4.2 GoalsThe goal of this showcase is to validate advanced interference detection mechanisms in densenetworks, by extendingpreliminary results obtainedwith theYear 2 showcase.Wewill explore indepththefeaturespaceandincludetherecognitionphaseintheonlineexperiment,integratingwiththeinter-BSSMACanditstuning.

3.4.3 BreakthroughsTheproblemofinterferencecoordinationhasbeenextensivelyfacedatthenodelevel.Theapproachofthisshowcaseistoperformanadvanceddetectionoftopologicalissuessuchashidden,exposednodesandflowsinthemiddle.Theclassificationofsuchphenomenausingmachinelearningisahottopic that promises performance issues recognition with both centralized and distributedapproaches. Phenomena such as hidden and exposed nodes result in similar performanceimpairmentsbutrequireoppositeactions:limitingthetrafficofferedbythehiddennodesinonecaseand stimulating the suffering node in the second one. Classification of phenomena is thereforecrucialforbestmitigationstrategyselection.

3.4.4 MethodologyTheadvanceddetectionofpathologicalinterferenceconditionswillbeperformedthroughclassifiersandneuralnetworksthataredefinedoffline,aretrainedonlineand/orofflineandusedonlineduringtheexperiment.

3.4.5 UseofWiSHFULFunctionalityIn this showcase we will exploit WiSHFUL monitoring capabilities using the Monitoring andConfigurationEngine, the intelligent framework for recognizing special interferenceconditionsandcontrolcapabilitiestoenforceandtunetheassignedradioprogram.

3.5 Interference classification for Wi-Fi nodes on the basis of error patterns usingmachinelearning

3.5.1 OverviewThe possibility to detect exogenous interfering sources in Wi-Fi networks, such as ZigBee, LTE inunlicensed bands, microwave ovens, and Bluetooth, is a very interesting feature for improvingcoexistenceinISMbands.WeobservedthatreceivererrorsgeneratedbyexogenousRFsignals(i.e.non-Wi-Fimodulatedsignals)exhibit significantdifferences (in termsofoccurrenceprobabilityanderror intervals) from the ones generated by collisions with other Wi-Fi transmissions. The errorsgeneratedbycross-technologyinterferencehavesignificantlydifferentpatternscomparedtoerrorstypical of Wi-Fi transmissions. Indeed, in case of wide-band noise and exogenous interferencesignals,errorsmayappearrandomlyatanypointduringthetimethedemodulatorisactive,whileforWi-Fimodulatedsignalserrorstatisticsvaryduringtheframereceptionanddependonframelengthand rate.Moreover,multipleevents canbegeneratedby the receiverduring the same interferingtransmissioninaburstoferrorsthatwedefineaserrorpattern.Forexample,achecksumfailurecanfollowthedetectionofagoodPLCP,oranother(failedornot)synchronizationtrialcanbeperformedafterabadPLCPeventaccordingtoconsecutiveresetsofthereceiver.Inthisshowcase,weplantoestimatetheerrorstatisticsfordetectingthepresenceofaninterferingsource,andthentoactivateaclassifier,workingontheprocessingofconsecutiveerrorbursts, foridentifyingtheinterferingsourceamongasetofpre-definedsourcesorforsignalingthataspecificinterfering source is unknown. The classifier will be studied off-line, on the basis of experimentaltracescollectedundercontrolledinterferingsources,formodelingthereceiverbehavior.

3.5.2 GoalsThe goal of this showcase is demonstrating how theWiSHFUL functionalities can be exploited forcollectingandprocessingdatatraces indifferentnetworkexperimentsdevisedtobuilddata-driven

H2020-GANo.645274 D2.4

68

modelsofWi-Fireceiversinpresenceofvariousinterferingsourcesandhowthedata-drivenmodelcanbeeasilyimplementedinarealnetworkforinterferenceclassification.

3.5.3 BreakthroughsInterference detection of non-Wi-Fi modulated signals is usually based on SDR platforms orsimultaneousmulti-technologyreceivers.AnapproachworkingwithacommercialcardinexploredinAirshark by using an 802.11n PHY able to read RSSI values at different sub-carriers and bysequentiallymovingaWi-Fimonitoringcardtotheadjacentchannelswithstepsof5MHz.Incaseofsuddendisappearanceof theRFsignalswhenmoving fromonechannel to thenextone, it canbeassumed that interference was due to a narrow-band ZigBee channel. Complex algorithms areappliedtotheRSSIsamplesforcharacterizingspectral,energyandpulsesignalsthataremappedintoa technologyclassificationscheme.While thispreviousworks relieson theclassicalanalysisof thefrequency and time domains, in this showcasewe plan to study the error domain, i.e. the errorsproduced by the interfering technologies. We expect that such an approach can be much moregeneral, because we are able to train the classifier and to recognize a given interference sourcewithoutknowingthetechnicaldetails(bandwidth,pulsesignals,energy)ofthesourcephysicallayer.3.5.4 MethodologyThis showcase is devised to build a specific network intelligence by using machine-learningtechniques.Theideaisusingerrorstatisticsandspecificerrorpatternsforestimatingthetimingsandthe effects of the interfering technology on a givenWi-Fi receiver. The temporal analysis of thereceiverevents is affectedby the receiver implementationbecause thedemodulator reset time incase of false or bad preambles depends on the card internal design and results in a differentgranularity of consecutive events. It follows that we plan to first train a classifier on the basis ofexperimentaltracescollectedundercontrolledinterferingsourcesfordesigningadata-drivenmodeloftheWi-Fireceiver.Then,themodelwillbeusedinrealexperimentsunderunknowninterference.

3.5.5 UseofWiSHFULFunctionalityWeplantouseUPI_RfunctionsforcollectingtheerrorstatisticsanderrorpatternsexperiencedbyWi-Fi receivers based on theWMParchitecture.Wewill also useUPI_R for generating interferingsources,includingLTEmodulatedsignalsinISMbandsandnon-standardmodulations.TheWiSHFULintelligence frameworkwill be used for building the data-drivenmodel and for implementing thenewinterferenceclassifier.

3.6 Radiovirtualizationwithsimultaneoustransmissionandreception3GPP is currently standardizing NarrowBand-IoT (NB-IoT). This radio access technology aims toprovidecost-effectiveconnectivityservices forbillionsofdevicesaroundtheworld,supporting lowpowerconsumption,theuseoflow-costdevicesandprovisionofexcellentcoverage.NB-IoTistobedeployed in the same channels used by standard LTEmobile carriers; usually sharing the channelwithaLTEdeployment.Inapreviousshowcase(seesection2.2.1),weexploredthevirtualizationofLTEandNB-IoTbasestationstoeasethedeploymentofthesetechnologies.However,ourvirtualizedbase stations had the limitation of being able only to transmit signals towards their clients, i.e.,clientscouldnottransmittotheirvirtualbasestations.Inthisshowcase,wewillshowtheextensionoftheradiovirtualizationframeworkthataddscapabilitytovirtualradiostoalsoreceivesignalsfromtheclients,asillustratedinFigure38.

H2020-GANo.645274 D2.4

69

Figure38Radiovirtualizationshowcasewithtransmissionandreception

The interesting scenarios, as far as the type and volume of data between the devices for thisshowcase,canbedescribedas:

• LTE: In the downlink (from LTE Base station to mobile subscriber) we will have a video-streaming; in the uplink (from mobile subscriber to the LTE base station) we will have afeedbackregardingthevideoquality.Thevideo-streamingapplicationwillusetheinformationreceivedfromthemobilesubscribertoadjustthevideo-quality;

• NB-IoT: In the downlink (fromNB-IoT base station to theHealthcare sensor display)wewillhavethedatacollectedbytheHealthcaresensor;intheuplink(fromHealthcaredisplaytotheNB-IoT base station) we will have a configuration of some parameter of the sensor (timeintervalbetweensensorreadingsforexample).TheHealthcaresensorwillusethisinformationtoadjustitsconfigurationaccordingly.

3.6.1 GoalsThegoalsof this showcase is todemonstrate theadvantages that radio virtualization canbring tofuturemobilewirelessnetworks.WeuseWishful toenable the configurationof virtual radiosandUSRP devices. In particular, wewill provide access to several parameters of both LTE and NB-IoTradios,andcentralfrequencyandbandwidthoftheUSRP.

3.6.2 BreakthroughsRadiovirtualizationiscurrentlybeingconsideredinstate-of-artarchitecturesforthenextgenerationofmobilenetworks.Webelieve thatapractical implementationof suchmechanisms is thekey toanalyzeitsimpactonwirelesscommunications.

3.6.3 MethodologyTheproposedradiovirtualizationframeworkwillbeevaluatedbymeansofexperiments intheIRIStestbed.Astheperformancemetricwewillcomparethethroughputandlatencyofwaveformswithandwithoutthevirtualizationframework,thedifferenceinSNRwithandwithoutthevirtualizationframework,andthecomputationaloverheadofthevirtualizationframework.

Hea

lth S

enso

r

Mob

ile S

ubsc

riber

IoT SensorMobile subscriber

NB-IoT TX/RX Waveform

LTE RX/TX Waveform

NB-Iot RX/TX Waveform

HypervisorLTE TX/RX Waveform

Sensor display and configuration

Video display andconfiguration

VideoStreaming

HealthcareSensor

Rad

io H

ardw

are

Softw

are

Mod

ules

Dat

a Ac

quis

ition

RRH

PC 2 PC 3PC 1

Wishful

Agent

Wishful

Agent

Wishful

Agent

WiSH

FUL C

ontroller

UPI_R

H2020-GANo.645274 D2.4

70

3.6.4 UseofWiSHFULFunctionalityThefollowingWiSHFULfunctionalityfromtheGNURadiomodulewillbeused:

• ThefunctionalitytochangethecentralfrequencyandbandwidthoftheHypervisor.

• Thefunctionalitytochangespecificparametersofvirtualradiowaveforms(numberofcarriersandMCSinLTEforexample).

3.7 IEEE802.11OverlappingBSSmanagement3.7.1 OverviewIEEE 802.11 Basic Service Sets (BSSs) working on the same radio channel are becoming commonbecause of the wide diffusion of 802.11 networks and limited availability of channels. Althoughcarrier sensemechanism in principle does not require any frequency planning, it has been shownthatsevereperformanceimpairmentscanoccurduetotheneighborcaptureeffect.ItoccurswhenaBSSisbetweentwoBSSswhichdonotheareachother(Figure39).Inpresenceofgreedytraffic,theBSS in themiddle can be prevented fromaccessing the channel indefinitely because it senses themediumpermanentlybusy.

Figure39.NeighborcaptureeffectinIEEE802.11networks.

To limit the neighbor capture effect and extend admission control / scheduling decisions, a newmechanismcalledOBSSmanagementisrequired.

3.7.2 GoalsThe envisioned OBSS management mechanism is based on two main components. The firstcomponent isresponsibleformonitoring(quantifying)theloadandinterferencestatusofeachBSSandforsignalingthisinformationtotheneighboringBSSs.Thelattercanbedoneusingeitherbeaconframes or QLoad report frame as currently discussed in the 802.11ax working group. The secondcomponentperformstheactualOBSSmanagement.Inthisshowcasewewillhighlighttwodifferentapproaches. First, a channel selection which takes the neighbour capture effect explicitly intoaccount when calculating the frequency plan. Second, an approach where neighbouring BSSs arecooperatingwitheachotherforresourcesharingonthebasisofsuchinformation.Inparticular,weconsidered timeslotmediumaccesswhereexclusive timeslotsareassigned toAPs suffering fromneighborcaptureeffect.

AP1

STA2

AP2

STA1

CarrierSensingRange

AP1 STA1

STA3

AP3

AP3 STA3AP2

busychann

el

H2020-GANo.645274 D2.4

71

3.7.3 BreakthroughsOBSSmanagement isbeingactivelydiscussed inthe IEEE802.11axworkinggroup.Webelievethatthe proposed mechanisms and results from this showcase would be of great importance to the802.11 community and would allow significantly faster evaluation/comparison of the proposedcoordinationalgorithms.

3.7.4 MethodologyThe proposed algorithms forOBSSmanagementwill be evaluated bymeans of experiments in an802.11n testbed. As the performancemetric we will compute the throughput fairness (e.g. Jain'sfairnessindex)ofeachBSS.

3.7.5 UseofWiSHFULFunctionalityThe following WiSHFUL functionality from the 802.11 (Wi-Fi) module will be used: a.) thefunctionalitytochangetheradiochannelofBSSs(APs),andb.)thepossibilitytoperformtimeslottedmediumaccessusingthehybridCSMA/TDMAMACoftheAtherosplatform.

3.8 Closed-loopratecontrolforIEEE802.11infrastructurenetworks3.8.1 OverviewTheratecontrolalgorithmsoftoday’sIEEE802.11networksaremostlyopen-loop,e.g.LinuxMinstrelalgorithm. This means that the Wi-Fi transmitter adapts the bitrate (MCS) based on link-levelmeasurements(probing).Unfortunately,open-lopratecontrol,whilebeingsimpleto implement, isinefficientespeciallyinmobileenvironmentsascomparedtoclosed-loopapproaches.Inclosed-loopratecontrolthereceiverestimatestherate(MCS)basedonmeasuringtheactualchannelqualityandsignalingitsvaluetothetransmitter.

3.8.2 GoalsThe goal of this showcase is to demonstrate that Wishful enables the required functionality forclosed-loop rate control (Figure 40). In particular, we will provide access to Channel StateInformation(CSI)onaperpacketbasis,whichcanbeusedasinputforcomputationofeffectiveSNRvaluesatthereceiverside.FromtheeffectiveSNRtheproperbitrate(MCS)isselectedandsignaledtothetransmitternode,whichwilluseitforthenextframetransmission.

In this showcase the signaling will be out-of-band, i.e. using theWishful control framework. Thisallowsustostudytheimpactoffeedbackdelayandlosses.

H2020-GANo.645274 D2.4

72

Figure40.Closed-loopratecontrol.

3.8.3 BreakthroughsThisshowcasewillhighlightthesuitabilityofusingWishfulcontrolplatformforstudyingclosed-loopratecontrolalgorithmsforWi-Fi.

3.8.4 MethodologyWeplantoimplementaspecificclosed-loopratecontrolalgorithmforWi-Fiinfrastructurenetworks.Thealgorithmwillbeevaluatedbymeansofexperimentsinan802.11ntestbed.Astheperformancemetricwewill compute the throughput and compare it to state-of-the-art open-loop approaches,e.g.LinuxMinstrel.Furthermore,wewillanalyzetheimpactoffeedbackdelayandloss.

3.8.5 UseofWiSHFULFunctionalityThefollowingWiSHFULfunctionalityforIEEE802.11isrequired.First,thecollectionofCSIvaluesonreceiver sideonaperpacketbasis. Currently,we support this for theWi-Fi-Atherosplatform.WeplantosupportitalsoforIntel-Wi-Fiplatform.Second,functionalityforinspectingCSIandcomputingeffectiveSNRfromCSIvaluesisrequired.Third,thefunctionalityforsettingthewirelessdatarates(MCS)atthetransmittersideonaperpacketbasisisrequired.

3.9 ContextAwarenessinspectrummanagementsystem-aidedSUnetworks3.9.1 OverviewOver the last decade, regulators, academic and industry bodies have been focusing significantresearch efforts on the topic of spectrum sharing, as a way to overcome the spectrum crunchproblem. Cognitive Radio is one of the possible approaches to enable sharing between differenttechnologies. However, there has been an increasing agreement that cognitive techniques basedsolely on local sensingmay not be enough to protect primary users (PUs) from interference. As aresult,regulatorshaveshiftedtheirattentiontospectrumsharingapproachesthatrelyonnetwork-widespectrummanagementsystems(SMS).ExamplesofSMS includetheSpectrumAccessSystem(SAS)inUS,LicensedSharedAccess(LSA)andTVWSgeo-locationdatabases,andradioenvironmentmaps(REM).Ontheotherhand,questionsremainregardingtheuseofthesemanagementsystemsinradioenvironmentswherethePUsdisplayahighlydynamicchannelaccessbehaviour.

Inthiswork,westudydifferentlearningandagiledecisionmakingtechniquesthatSUscanemployindynamicradioenvironments,whilesupervisedandsupportedbyacentralizedintelligencespectrummanagement system.Our systemmodel comprisesmeasurement-capable devices (MCDs), a SMS,andSUnetworks.TheMCDscanbeeitherdedicatedsensorsorSUs,thatcollectlong-termstatistical

WiFiTx WiFiRx

packeti

CSIest.

eff.SNR

MCSsel.

MCSfeedback(out-of-band)

packeti+1

SetMCS

H2020-GANo.645274 D2.4

73

dataoftheradioenvironmentofthePUs’behaviour.Thecollectedstatistics,andfeaturesfromtheradio environment are then sent and aggregated at the SMS,whichwill recognize several distinctsharing scenarios SUs may encounter. For each scenario, the SMS will generate a list of channelaccessstrategiesthatSUsmayemploytoavoidinterferencewiththePUs.

3.9.2 Goals• Design of several long-term channel statistics gathering, and deep learning-based

representationalgorithmsthatwillbeemployedbyMCDstocharacterizePUs’behavior• Derivation of optimal channel access strategies/techniques and spectrum sensing

configurationsbytheSMS,basedongatheredandaggregateddatafromtheMCDs• ConfigurationofSUs’channelaccessandsensingalgorithmsbytheSMS

3.9.3 BreakthroughsOur vision is to offload power, time and hardware-demanding sensing, classification, and decisionmaking procedures from battery-powered SUs to dedicated sensor networks and SUs that arepluggedintopowersources(e.g.basestations),andtonetwork-wideintelligentSMSs,respectively.Battery-poweredSUswillbeconfiguredby theSMSandwillonlyhave toperformshort-term low-complexity sensing procedures to classify their environment and select the appropriate channelaccessstrategythatminimizesinterferencetoPUs.

Offloading complexity frommobile SUswill allow the use of advanced deep learning classificationand Markov Chain-based algorithms that succinctly characterize the PUs’ behaviour, and narrowdown the set of possible scenarios and channel access strategies a SUmay encounter and apply,respectively. Overall, this approach will result in a more spectrally efficient operation by SUnetworks.

3.9.4 MethodologyOur experimentation scenario will comprise a PU-Tx and a PU-Rx, which can switch betweendifferent waveforms and channel access patterns. The PU-Rx will also report to the SMS itsexperienced interferenceovertime.ASU-Tx inthesameradioenvironmentwillbe incontactwiththe SMS from where it collects the list of configurations (e.g. possible hopping patterns andparameters) usedby thePUs, and consults its caused interference. Basedon this information andshort-term sensing, the SU-Txwill adapt its channel hopping pattern, transmit power, bandwidth,modulationorder, andother parameters tomaintaining the interference at PUsbelowa specifiedthreshold, andmaximize the number of packets that reach a SU-Rx. Simultaneously, a sensorwillperform long-termsensing,andpopulate theSMSwithpossiblescenarios,eachcharacterizedbyasetofPU’sfeatures.

H2020-GANo.645274 D2.4

74

Figure41–Illustrationoftheseveralagentsandinteractionspresentinthecontext-awareSMS-aided

spectrumsharingframework.

3.9.5 UseofWiSHFULFunctionalityWiSHFUL UPIs will be employed to set each element of the experiment running, performconfigurationof thePUandSUparameters, suchasbandwidthandrangeof transmitpowers,andexchangeofcontrolinformationbetweenPUs,SMS,sensorandSUs.

H2020-GANo.645274 D2.4

75

4 ConclusionThisdocumentsummarizesthegoalsandresultsofshowcasespresentedinYear2oftheproject.ItextendsthefirstsetofshowcasesdescribedinD2.3.TheshowcasespresentedhereindefinerelevantandconvincingscenariosinviewofpromotingtheWiSHFULframeworkcapabilities.Eachshowcaseprovidesahigh-levelspecificationofopensoftwareplatformsforradioandnetworkcontroldrivenby domain-specific requirements from different relevant market segments. This documentdemonstrates the impact of WiSHFUL research and technologies in developing new solutions toresolvechallengesfoundonwirelesscommunicationnetworks.Demonstrationsinclude:

• TheIRISSDRdemointegrationwithUPIsthatchangeparametersontheflyatUSRPs.• The Coexistence of IEEE 802.15.4e TSCH with IEEE 802.11 networks proves the solution

feasibilityandapplicabilityforacrosstechnologysynchronizationschemebetweenTSCHandWi-Finetworks.

• WehaveprovedthattheoptimizationlogiccanbeimplementedonWiSHFULprogrammablenodeswithorwithoutpresenceoflegacynodes.

• TheloadandtopologyawarenetworkingshowcasesdemonstratehowtheWiSHFULUPIscanbeusedtodynamicallymonitornetworkperformanceandtopology,changenetworkprotocolconfiguration,orswitchroutingmodules,andmitigateperformanceimpairments.

Additionally, thisdocumentdiscussesextensionsor continuationsof thework completed so far tofurtherexploitthecontributionsoftheproject.Theseinclude:

• ImplementationoftheLTE-UWi-Ficoexistenceshowcase.• UsingGNURadiotofacilitateradioequipmentsharing.• Measurepacket loss inboth IEEE802.15.4eTSCHandWi-Finetworksandwith implemented

showcasesystemrunning• RadioSlicingforVirtualizedHomeWi-FiAccessPoints.• Investigatethepossibilityofguaranteeingcoexistenceamongheterogeneousaccessprotocols

andschedulingmechanisms.• The implementation of UPIs with additional features such as extracting the A-MPDU

transmissionresultsfromBlockAckandenabling/disablingthemobility-awarealgorithmwhichadaptsPHYrateandframeaggregationlengthinreal-time.

• Aspects such as traffic patterns, MAC protocols, node mobility, and so forth can beinvestigated further as part of the optimal link estimator showcase. Additionally abenchmarkingtoolboxcouldbeaddedtosupportprogramexecutionwithdifferentparametersettingsmightbedeveloped.

• Weare alsoplanning to consider several potential extensions to theREACT scheme for realnetwork deployments including: Generalizing the concept of channel allocations; improvingauctionrobustness;andthestabilityoftheREACTallocationsfromatheoreticalpointofview

ThisdocumentalsooutlinestheutilityandimpactofWiSHFULinprojectshowcases.Technicaldetailsof these showcases are outlined in the appropriate technical deliverables D3.4, D4.4, and D6.4.Finally,thisdocumentdefinesalistofintelligenceshowcasestobeimplementedinthefinalyearoftheproject.

H2020-GANo.645274 D2.4

76

References

[1] ProjectDeliverableD2.3 -Resultsof first setof showcases,http://www.wishfulproject.eu/sites/default/files/images/review/WiSHFUL_D2.3_TCD_R_PU_2016-01-04_Final.pdf,lastaccessDecember5th2016.

[2] National Instruments, Real-time LTE/Wi-Fi Coexistence Testbed, http://www.ni.com/white-paper/53044/en/,2016.

[3] Jindal,Nihar, Breslin,Don andNorman,Alan, “LTE-U andWi-Fi: A Coexistence StudybyGoogle”,Wi-FiLTE-UCoexistenceTestWorkshop,2015.

[4] National Instruments, Real-time LTE/Wi-Fi Coexistence Testbed, http://www.ni.com/white-paper/53044/en/,2016.

[5] Olbrich,Michael,Zubow,Anatolij,etal.,“WiPLUS:PassiveLTE-UCo-ChannelInterferenceDetectionusingCommodityWi-FiHardware”,TechnicalReport,2016.

[6] P.D.Suttonetal.,"Iris:anarchitectureforcognitiveradionetworkingtestbeds,"inIEEECommunicationsMagazine,vol.48,no.9,pp.114-122,Sept.2010.doi:10.1109/MCOM.2010.5560595

[7] Controlling the IRIS SDR Framework using WiSHFUL UPIs, http://www.wishful-project.eu/sites/default/files/Short-Final-WebSite.mp4,lastaccessDecember5th2016.

[8] “IEEE Standard for Local and metropolitan area networks–Part 15.4: Low-RateWireless Personal AreaNetworks (LR-WPANs)Amendment1:MACsublayer,” IEEEStd802.15.4e-2012 (Amendment to IEEEStd802.15.4-2011),pp.1–225,apr2012

[9] S.Duquennoy,B.AlNahasetal.,“Orchestra:RobustMeshNetworksThroughAutonomouslyScheduledTSCH,”inProc.13thACMConf.Embed.NetworkedSens.Syst.,ser.SenSys’15.NewYork,NY,USA:ACM,2015,pp.337–350

[10] “LinuxWireless:ath9kspectralscan,”2016.

[11] S. Zehl, A. Zubow, and A. Wolisz, “hMAC: Enabling Hybrid TDMA/CSMA on IEEE 802.11 Hardware,”TelecommunicationNetworksGroup,TechnischeUniversitätBerlin,Tech.Rep.TKN-16-004,nov2016

[12] Jeongkeun Lee,Mostafa Uddin, Jean Tourrilhes, Souvik Sen, Sujata Banerjee,Manfred Arndt, Kyu-HanKim,andTamerNadeem.2014.meSDN:mobileextensionofSDN.InProceedingsofthefifthinternationalworkshoponMobilecloudcomputing&services(MCS'14).ACM,NewYork,NY,USA,7-14.

[13] Baccour,N.,Koubˆaa,A.,Mottola,L.,Z´u˜niga,M.A.,Youssef,H.,Boano,C.A.,Alves,M.:Radiolinkqualityestimationinwirelesssensornetworks:Asurvey.ACMTrans.Sen.Netw.8(4),34:1–34:33(Sep2012)

[14] Fonseca,R.,Gnawali,O., Jamieson,K.,Philip,L.:Fourbitwireless linkestimation. In:ProceedingsoftheSixthWorkshoponHotTopicsinNetworks(HotNets(2007)

[15] Baccour,N.,Koubˆaa,A.,Youssef,H.,Jamˆaa,M.B.,DoRosario,D.,Alves,M.,Becker,L.B.:F-lqe:Afuzzylinkqualityestimatorforwirelesssensornetworks. In:WirelessSensorNetworks,pp.240–255.Springer(2010)

[16] Meier,A.,Rein,T.,Beutel,J.,Thiele,L.:Copingwithunreliablechannels:Efficientlinkestimationforlow-power wireless sensor networks. In: Networked Sensing Systems, 2008. INSS 2008. 5th InternationalConferenceon.pp.19–26(June2008)

[17] Thubert,P.:ObjectiveFunctionZero for theRoutingProtocol forLow-PowerandLossyNetworks (RPL).RFC6552(ProposedStandard)(Mar2012)

[18] Javaid,N., Javaid,A., Khan, I.,Djouani, K.: Performance studyof etxbasedwireless routingmetrics. In:Computer,ControlandCommunication(IC4)(2009)

[19] Ruckebusch,P.,DePoorter,E.,Fortuna,C.,&Moerman,I.(2016).GITAR:GenericextensionforInternet-of-ThingsARchitecturesenablingdynamicupdatesofnetworkandapplicationmodules.AdHocNetworks,Volume36,Part1,January2016,Pages127–151

H2020-GANo.645274 D2.4

77

[20]P. Gallo, S.Mangione, and G. Tarantino, “Widar: Bistatic wi-fi detection and ranging for off-the- shelfdevices,” in World of Wireless, Mobile and Multimedia Networks (WoWMoM), 2013 IEEE 14thInternationalSymposiumandWorkshopsona,pp.1–6,June2013.

[21]E. Khorov, A. Kiryanov, A. Krotov, P. Gallo, D. Garlisi, I. Tinnirello, “Joint Usage of Dynamic SensitivityControlandTimeDivisionMultipleAccessinDense802.11axNetworks”,MultipleAccessCommunicationsMACOM2016

[22] NationalInstrumentswhitepaper,“Real-timeLTE/Wi-FiCoexistenceTestbed”,http://www.ni.com/white-paper/53044/en/