Choir: Empowering Low-Power Wide-Area Networks in Urban ...swadhin/reading_group/... · Wide-Area...

Choir:EmpoweringLow-PowerWide-AreaNetworksinUrban

Settings

RashadEletreby

DianaZhang,Swarun KumarandOsmanYağan

1http://www.witechlab.com/LoRa/ChOIR.html

ImagineaworldwhereeverysingleobjectisconnectedtotheInternet…

2

Fewkbps

Severalmilesaway

10yearbatterySimpleandcheapRFinterface

3

SmartInfrastructure SmartHomes SmartVehicles

Low-PowerWide-AreaNetworking(LP-WAN)

LongRange

• Upto10KMsinruralareas

LowDatarate

• Orderofkilobitspersecond

LowCost

• <$5

LowPower

• Upto10yearsofbatterylife

Low-PowerWide-AreaNetworking(LP-WAN)

InitiativesfromIndustry(LoRa,SIGFOX)andstandardizationbodies(3GPPLTEM,NBIoT)

KeyChallenges

6

KeyChallengesInterference

Collisionsemergefromthesheer densityofnodesandthesimplicity ofthecurrentMACprotocols(e.g.,transmit

assoonaswakeup)

LPWANrangesdropby10xinurban areasduetoexcessivemultipath,shadowing,etc.

7

Range

8

WiFi/CellularWirelesssensor

networksLPWANs

LoRaWAN

Sigfox

MegaMIMO

SAM

ZigZag

Glossy

ACR

Pastwork

….

…. ….

ChoirScalability

• Decodes10’sofcollidedtransmissions

Range

• Extendstherangeofteamsofcooperatingnodes

Preservingsimplicity

• Fullyimplementedatasingle-antenna basestation

9

basestationoveranareaof10Km2 inPittsburgh

:Chirps

ChirpinT.D. Chirponaspectrogram

Data encoding

10

:1-bitencoding

𝒏 bits->dividetheBWto𝟐𝒏 initialfrequenciesIn general,

‘0’ ‘1’

11

:2-bitencoding

12

:2-bitencoding

13

ChoirinactionInterference

14

Range

CollisionofchirpsDifferentdata

+

15

Collisionofchirps

+

Samedata

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

16

80 100 120 140 160 180FFT Bin

0

100

200

300

400

500

Abs.

FFT

Hardwareimperfections

𝑓

𝑓 + 𝛿𝑓&

𝑓 + 𝛿𝑓'

Localoscillatormismatch

17

Hardwareimperfections

Packet1

Packet2

Sub-symboltimingoffsets

18

19

Time

Frequency

TO

Chirpsaresignalswhosefrequencyincreaseslinearlywithtime

20

Time

Frequency

TO

Anoffsetintimemapstoanoffsetinfrequency!FO

21

Time

Frequency

Twochirpswithanoffsetinfrequency!FO

Collisionofchirps

+

Samedata

80 100 120 140 160 180FFT Bin

0

100

200

300

400

500

Abs.

FFT

Hardwareoffsets!22

23

Decodingdata

U1data: U2data:

24

!U1data+U1hardwareoffsets=125U2data+U2hardwareoffsets=130

90 100 110 120 130 140 150 160 170FFT Bin

0

100

200

300

400

500

Abs.

FFT

Symbol 1

125 130

DecodingdataPreamble Sym.1 Sym.2 Sym.n…

Preamble Sym.1 Sym.2 Sym.n…

Peaklocationsareusedtoestimatehardwareoffsets

Hardwareoffsetsremainconstantacross

thepacket

25

U1data+U1hardwareoffsets =125U2data+U2hardwareoffsets=130

Symbol1:

DecodingdataPreamble Sym.1 Sym.2 Sym.n…

Preamble Sym.1 Sym.2 Sym.n…

Howtomeasureaccuratehardwareoffsetsacrossthepreamble?

Peaklocationsareusedtoestimatehardwareoffsets

Hardwareoffsetsremainconstantacross

thepacket

26

Decodingdata𝑓&∗, 𝑓'∗ = 𝑎𝑟𝑔𝑚𝑖𝑛{34∈ 3467,3487 ,39∈ 3967,3987 } 𝑦𝐶

6& − ℎ&𝑒?'@34A + ℎ'𝑒?'@39A'

𝑓B ->initialfrequencyoffsetestimateofuseriℎB ->channelestimateofuseriΔ ->binsizeoftheFFT𝐶6& ->conjugatenominalchirp𝑦 ->receivedsymbol𝑓B∗ ->correctfrequencyoffsetofuseri

27

28

Whichpeakcorrespondstowhichuser?

29

Whichpeakcorrespondstowhichuser?

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

Symbol 2

27.2 189.6

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

Symbol 1

352.2107.6

30

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

Symbol 2

27.2 189.6

Whichpeakcorrespondstowhichuser?

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

Symbol 1

352.2107.6

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

Symbol 2

27.2 189.6

Whichpeakcorrespondstowhichuser?

0 100 200 300 400 500 600FFT Bin

0

100

200

300

400

500

600

Abs.

FFT

Symbol 1

352.2107.6

31

User1 User2Integerpartdependsonbothdataandhardware

offsets

Fractionalpartdependsonlyonhardware

offsets

1 Near-fareffect

2 Inter-symbolinterference

3 Handlingageneralnumberofcollisions

32

Wegeneralizethissolutiontoaccountfor…

ChoirinactionInterference

33

Range

RangeExtension

62°F

67°F

65°F

Eachnodeisout-of-range!

34

RangeExtension

62°F

67°F

65°F

Eachnodeisout-of-range!

35

63°F

65°F

Canweexploitdatacorrelationstoobtainacoarse-grainedviewofthesenseddata?

36

0 100 200 300 400 500 600FFT Bin

0

5

10

15

20

25

Abs.

FFT

65 °F

62 °F 67 °F

Noisefloor

Objective

0 100 200 300 400 500 600FFT Bin

0

5

10

15

20

25

Abs.

FFT

64.5 °F

Noisefloor

37

0 100 200 300 400 500 600FFT Bin

0

5

10

15

20

25Ab

s. F

FT

65 °F

62 °F 67 °F

Noisefloor

Approach

0 100 200 300 400 500 600FFT Bin

0

5

10

15

20

25

Abs.

FFT

64.5 °F

Noisefloor

ChoirReceivefilter

38

0 100 200 300 400 500 600FFT Bin

0

5

10

15

20

25Ab

s. F

FT

65 °F

62 °F 67 °F

Noisefloor

0 100 200 300 400 500 600FFT Bin

0

5

10

15

20

25

Abs.

FFT

64.5 °F

Noisefloor

ChoirReceivefilter

Implementation

39

Evaluation

40

Hardwareoffsets

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100 120 140

CD

F

Observed CFO+TO (Hz)

ObservedIdeal

41

HardwareoffsetsaretrulydiverseacrossLPWANradios

Resolvinginterference

20000

40000

60000

80000

100000

120000

140000

2 3 4 5 6 7 8 9 10

Net

wor

k T

hrpt

(bi

ts/s

ec)

# Users

Ideal

42

Resolvinginterference

0

20000

40000

60000

80000

100000

120000

140000

2 3 4 5 6 7 8 9 10

Net

wor

k T

hrpt

(bi

ts/s

ec)

# Users

Ideal ALOHA

43

Resolvinginterference

0

20000

40000

60000

80000

100000

120000

140000

2 3 4 5 6 7 8 9 10

Net

wor

k T

hrpt

(bi

ts/s

ec)

# Users

Ideal ALOHA

ChOIR

44

29x

Extendingrange

45

Numberofcollaboratingnodes

Range

1 1Km

10 2.5Km

30 2.65Km

2.65X

ConclusionObjective

Results Scalability

• Decodes10’sofcollidedtransmissions

Range

• Extendstherangeofteamsofcooperatingnodes

Preservingsimplicity

• Fullyimplementedatasingle-antenna basestation

Platform

46