Decimeter-LevelLocalization
withaSingleWiFi AccessPoint
DeepakVasisht
Swarun Kumar,DinaKatabi
IndoorLocalizationisCool!
SpotFi [SIGCOMM’15],ToneTrack [Mobicom’15],Phaser
[Mobicom’14],Tagoram [Mobicom’14],LTEye [SIGCOMM’14],
ArrayTrack [NSDI’13],PinPoint [NSDI’13],PinIt [SIGCOMM’13],
Zee[MobiCom’12],PinLoc [MobySys’12],EZ[MobiCom’10],….
• Locateoff-the-shelf devices
• Accuracyoftensofcm
But… TheyNeed4-5AccessPoints
HomesandsmallbusinesseshaveONE access
point(AP)
Application:Controlheatingbasedonoccupancy
Application:WiFi Geo-Fencing
Application:Device-to-deviceLocalization
Enabledevice-to-devicelocalizationwithout
infrastructuresupport
Chronos
• Enablesdecimeter-accuratelocalizationusinga
singleoff-the-shelfWiFi card
• A novelalgorithmtoestimatepropagationtimeto
sub-nanosecondaccuracyusingaWiFi card
• Implementedandevaluatedinpracticalsettings
WhypastworkneedsmultipleAP’s?
θ
SingleAccessPoint?
distance
θ
MeasuringDistance
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
0
5
10
15
20
25
Distance=speedoflightxpropagationdelay
MeasuringDistance
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
0
5
10
15
20
25
Distance=speedoflightxpropagationdelay
Howdowemeasurepropagation delay?
PropagationDelay
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
0
5
10
t
15
20
25
ϕ 2𝜋𝑓𝑡 𝑚𝑜𝑑2𝜋
Phaseofthesignal()=
PropagationDelay:Example
2
1.5
1
2.41GHz
0.5
0
-0.5
0
-1
-1.5
-2
0
5
0.5
10
1
t(ns)
15
𝜙 = 2𝜋𝑓𝑡𝑚𝑜𝑑2𝜋
20
1.5
25
PropagationDelay:Example
5.8GHz
5.18GHz
2.48GHz
2.41GHz
0
0.5
1
t(ns)
1.5
Mathematically
𝜙- = 2𝜋𝑓- 𝑡𝑚𝑜𝑑2𝜋
𝜙. = 2𝜋𝑓. 𝑡𝑚𝑜𝑑2𝜋
𝜙/ = 2𝜋𝑓/ 𝑡𝑚𝑜𝑑2𝜋
Mathematically
𝜙- = 2𝜋𝑓- 𝑡𝑚𝑜𝑑2𝜋
𝜙. = 2𝜋𝑓. 𝑡𝑚𝑜𝑑2𝜋
𝜙/ = 2𝜋𝑓/ 𝑡𝑚𝑜𝑑2𝜋
UseChineseRemainderTheoremtogetthe
propagationdelay
Can’tmeasurepropagationdelaywithout
detectiondelay
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
0
Distance=speedoflightxpropagationdelay
5
10
15
20
25
Measureddelay=propagationdelay+detectiondelay
PacketDetectionDelay
2
1.5
1
Detection
Delay
0.5
Detection
0
-0.5
Decoding
-1
-1.5
-2
0
5
10
15
20
25
• Detectiondelay>> Propagationdelay
§ Howdoweeliminatedetectiondelay?
DetectionDelay≈ 200ns,PropagationDelay≈ 20ns
• Detectiondelayisunpredictable
Problem:Separatedetectiondelayfrom
propagationdelay
Solution:Leveragethatpropagationdelayand
detectiondelayhappenatdifferentfrequencies
f
2
1.5
1
f-fc
0.5
Detection
0
-0.5
-1
-1.5
-2
0
5
Propagation
Delay
10
15
20
25
Detection
Delay
Decoding
2
1.5
f-fc
f
1
0.5
Detection
0
-0.5
-1
-1.5
-2
0
5
Propagation
Delay(t)
10
15
20
25
Detection
Delay(t’)
𝜙 = 2𝜋𝑓𝑡𝑚𝑜𝑑2𝜋
𝜙 = 2𝜋𝑓𝑡 + 2𝜋 𝑓 − 𝑓4 𝑡′𝑚𝑜𝑑2𝜋
Decoding
f
f-fc
Detection
Propagation
Delay(t)
Decoding
Detection
Delay(t’)
𝜙 = 2𝜋𝑓𝑡𝑚𝑜𝑑2𝜋
𝜙 = 2𝜋𝑓𝑡 + 2𝜋 𝑓 − 𝑓4 𝑡′𝑚𝑜𝑑2𝜋
0
Idea:UseOFDMtomeasurephaseatf=fc
ButWiFi doesnottransmitatf=fc
Solution:LeverageOFDM
Phase
𝜙4
𝜙 = 2𝜋𝑓𝑡 + 2𝜋 𝑓 − 𝑓4 𝑡′𝑚𝑜𝑑2𝜋
𝜙4 = 2𝜋𝑓4 𝑡 + 0𝑚𝑜𝑑2𝜋
Phaseat
f=fc
𝑓4 − 3𝛿 𝑓4 − 2𝛿 𝑓4 − 𝛿
𝑓4 𝑓4 + 𝛿
𝑓4 + 2𝛿 𝑓4 + 3𝛿
Mathematically
𝜙4,- = 2𝜋𝑓4,- 𝑡𝑚𝑜𝑑2𝜋
𝜙4,. = 2𝜋𝑓4,. 𝑡𝑚𝑜𝑑2𝜋
𝜙4,/ = 2𝜋𝑓4,/ 𝑡𝑚𝑜𝑑2𝜋
Chronos eliminatespacketdetectiondelayby
leveragingOFDMproperties
AdditionalSystemComponents
• InitialPhaseOffsetCompensation
• Multipathresolution
AdditionalSystemComponents
• InitialPhaseOffsetCompensation
• Multipathresolution
InitialPhaseOffsets
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
0
5
10
t
15
20
𝜙 = 2𝜋𝑓𝑡𝑚𝑜𝑑2𝜋
𝜙 = 2𝜋𝑓𝑡 + Δ𝜙𝑚𝑜𝑑2𝜋
25
Idea:UseAcknowledgements
2
2
1.5
1.5
1
1
0.5
0.5
0
0
-0.5
-0.5
-1
-1
-1.5
-1.5
-2
-2
0
0
5
5
10
10
t
15
15
20
20
𝜙- = 2𝜋𝑓𝑡 + Δ𝜙𝑚𝑜𝑑2𝜋
𝜙. = 2𝜋𝑓𝑡 − Δ𝜙𝑚𝑜𝑑2𝜋
𝜙- + 𝜙. = 4𝜋𝑓𝑡𝑚𝑜𝑑2𝜋
25
25
Idea:UseAcknowledgements
2
2
1.5
1.5
1
1
0.5
0.5
0
0
-0.5
-0.5
-1
-1
-1.5
-1.5
-2
-2
0
0
5
5
10
10
t
15
15
20
20
𝜙- = 2𝜋𝑓𝑡 + Δ𝜙𝑚𝑜𝑑2𝜋
𝜙. = 2𝜋𝑓𝑡 − Δ𝜙𝑚𝑜𝑑2𝜋
25
25
Chronos eliminatesphaseoffsetsbyusing
𝜙- + 𝜙. = 4𝜋𝑓𝑡𝑚𝑜𝑑2𝜋
acknowledgements
AdditionalSystemComponents
• InitialPhaseoffsetCompensation
• Multipathresolution
Problem:MultipathEffect
Solution:Finddelaysforeachpath
10ns
16ns
5.2ns
Distancetosourcecorrespondstothe
smallestdelay.
ExperimentalEvaluation
Implementation
• Evaluationwithoff-the-shelfIntelWiFi 5300cards
• Kernelmodificationstotheiwlwifi driverintheUbuntukernel
• Groundtruthmeasurementsusinglaserdistancemeasurement
device(1mmaccurate)
EvaluationTestbed:OfficeEnvironment
20m
20m
DistanceMeasurementAccuracy
1
CDF
0.8
0.6
21cm
14cm
0.4
LOS
NLOS
0.2
0
0
1
2
Error(m)
3
SpotFi (SIGCOMM’15)
LocalizationAccuracy
3AP’s
4AP’s
5AP’s
1
CDF
0.8
0.6
98cm
65cm
0.4
0.2
LOS
NLOS
0
0
2
Error(m)
190cm
80cm
60cm
4
SpotFi (SIGCOMM’15)
LocalizationAccuracy
3AP’s
4AP’s
5AP’s
1
CDF
0.8
0.6
0.4
0.2
190cm
80cm
60cm
98cm
65cm
Chronos
canachievestate-of-the-art
0
localizationaccuracywithasingleAP
0
2
4
Error(m)
LOS
NLOS
Applications
SmartHomes
WiFi Geo-fencing
DevicetoDevice
localization
Applications
SmartHomes
WiFi Geo-fencing
DevicetoDevice
localization
Application:SmartHomes
Bedroom1 Bedroom2
9m
Livingroom
Bath
Kitchen
13m
Application:SmartHomes
Bedroom1 Bedroom2
9m
Livingroom
ChronosBathdetectsthecorrectroomwith
Kitchen
accuracy94%. 13m
Applications
SmartHomes
WiFi Geo-fencing
DevicetoDevice
localization
Application:GeoFencing
7m
Coffee Station
9m
Application:GeoFencing
7m
Coffee Station
Chronos canaccuratelyauthenticateWiFi
9m
userswith97%accuracy.
Applications
SmartHomes
WiFi Geo-fencing
DevicetoDevice
localization
Application:TakeMyPicture Drone
Application:TakeMyPicture Drone
3
y(m)
2
1
0
User
-1
Drone
-2
-3
-2
0
x(m) 2
Application:TakeMyPicture Drone
1
CDF
0.8
0.6
0.4
4.2cm
0.2
0
0
5
10
Error(cm)
15
Application:TakeMyPicture Drone
1
CDF
0.8
0.6
0.4
0.2
4.2cm
Chronos 0enablesadronetofollowtheuser
0
5
10
15
withnoinfrastructuresupport.
Error(cm)
RelatedWork
• WiFi Localization:SpotFi [SIGCOMM’15],ToneTrack [Mobicom’
15],Phaser [Mobicom’ 14],Tagoram [Mobicom’14],….
• Closest Work:SAIL[MobiSys’14]
Conclusion
• Chronos isthefirstsystemtoenableaccurate
localizationonoff-the-shelfWiFi cards
• Itskeyenablerisanovelalgorithmthatcan
estimateaccuratepropagationdelay,byeliminating
thedetectiondelay