Lost in Spaceor

Positioning in Sensor Networks

Michael O’DellRegina O’Dell

Mirjam WattenhoferRoger Wattenhofer

RealWSN 2005 Positioning in Sensor Networks 2

Positioning

• What is positioning (a.k.a. localization)?– Deduce coordinates– GPS “software version”

• Why positioning?– Sensible sensor networks– Heavy/costly localization hardware– Geometric routing benefits

• Idea:– (Small) set of anchors– Others: location = f(network,communication,measurements)

RealWSN 2005 Positioning in Sensor Networks 3

Sensor Networks

RealWSN 2005 Positioning in Sensor Networks 4

Our Perspective

Theory Practice

RealWSN 2005 Positioning in Sensor Networks 5

Positioning – As We See It

• Models of Sensor Networks

• Positioning Algorithms

• Hardware Description

• Experiments

• Lessons

• Future Work

Theory

Practice

RealWSN 2005 Positioning in Sensor Networks 6

Part I: Theory

RealWSN 2005 Positioning in Sensor Networks 7

Models of Sensor Networks

• Unit Disk Graph (UDG)– [Clark et al, 1990]– Widely used abstraction

• Quasi-Unit Disk Graph (qUDG)– [Krumke et al, 2001]– [Barriere et al, 2003]– [Kuhn et al, 2003]– More realistic?

• “well-behaved” ) allow proofs

1

1

d

?

RealWSN 2005 Positioning in Sensor Networks 8

Available Information

• T[D]oA:time– GPS– Cricket [Priyantha et al, 2000]

• RSS: signal strength– RADAR [Bahl, Padmanabhan, 2000]

• Imply distance

• AoA: angle– APS using AoA [Niculescu, Nath, 2003]

• Relative distance to anchors– APiT [He et al, 2003]

RealWSN 2005 Positioning in Sensor Networks 9

Positioning Algorithms

• Based on (q)UDG– (sometimes) provable statements– Abstraction ) rough idea

• Virtual Coordinates Algorithm [Moscibroda et al, 2004]– Linear programming– Complex, time-consuming– 100-node network: several minutes on desktop

• GHoST, HS [Bischoff, W., 2004]– Dense networks– Optimal in 1D– UDG crucial

RealWSN 2005 Positioning in Sensor Networks 10

Positioning Algorithms… cont’d

• APS [Niculescu, Nath, 2001 & 2003]– Hop or distance based– Given distance estimate, use GPS triangulation– Least-squares optimization– Isotropic network helpful

• General graphs– Given inter-node distances– Also: Internet graph (latencies)

• Example: Spring Algorithms– Internet: Vivaldi [Dabek et al, 2004]– Ad hoc [Rao et al, 2003]

RealWSN 2005 Positioning in Sensor Networks 11

Spring Algorithm

• Most practical?• Originally: graph drawing• Idea

– Edge = spring– Rest length = distance– Embedding = minimal power configuration

• Algorithm– Steepest descent, numerical methods– Simple:

• New position = average of neighbors• Iterate

Local vs. global

RealWSN 2005 Positioning in Sensor Networks 12

Our View = Assumptions

• Minimal hardware– Low storage– Low computing power– Basic RSS measurements

• Short range– Few meters– (RADAR: building – several dekameters)

RealWSN 2005 Positioning in Sensor Networks 13

Part II: Practice

RealWSN 2005 Positioning in Sensor Networks 14

Hardware Description

• ESB– scatterweb.com– 32kHz CPU– 2kB RAM– Sensors and actuators

• RSS:– Indirectly via packet loss

• New version:– Actual RSS measurable at receiver

• “Battery with Antenna”

Desktop:

– 3GHz– 512MB– Factor 105

RealWSN 2005 Positioning in Sensor Networks 15

“software version” RSS

• Older ESB (software) version– @sender: vary transmission power

• Via potentiometer controlling current to tranceiver

• Value s between 0 and 99

• Write s into packet

• Repeat x times

– @receiver: count number received packets• per s

– Measurement: packet loss• Requirement

– Distance increase → power increase– Correlation: to be determined

• New version (software):– Direct read out

Future work!

RealWSN 2005 Positioning in Sensor Networks 16

Experiment 1 – “Laboratory”

• Power vs. Distance– A sends at power level s– x = 100 times– d = 1..120cm

• Minimum• 90%

A B

ds

0

5

10

15

20

25

1 11 21 31 41 51 61 71 81 91 101 111 121

Distance (in cm)

Min

imu

m P

ow

er R

ecei

ved

0

5

10

15

20

25

30

35

1 11 21 31 41 51 61 71 81 91 101 111 121

Distance (in cm)

po

wer

at

90%

of

pac

kets

rec

eive

d

Coffee machine?

RealWSN 2005 Positioning in Sensor Networks 17

Original Algorithm

• Spring Embedding– Good for “easy” networks

• Power-to-distance– Inverse of previous experiments

• Results– Unusable!

RealWSN 2005 Positioning in Sensor Networks 18

Experiment 2 – “Room”

• Localization in the plane– Rectangle: 4m x 3m– 4 anchors: corners– Test node: inside

• Each anchor Ai

– Send packet s = 0..99– Next anchor

• Test node N– Record packets received

A0 A2

A3 A1

N15: 278

14: 365

16: 302

11: 139

RealWSN 2005 Positioning in Sensor Networks 19

Experiment 2 – “Room” … Results

Anchor 1

0

100

200

300

400

500

600

700

800

900

1 11 21 31 41 51 61 71 81 91

Received Power

Fre

qu

ency

Anchor 1

0

50

100

150

200

250

300

1 6 11 16 21 26 31 36

Minimum Power Received

Fre

qu

ency

Anchor 1

0

50

100

150

200

250

300

1 6 11 16 21 26 31 36

Minimum Power Received with/without Obstacles

Fre

qu

ency

Anchor 1

0

50

100

150

200

250

300

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Hole Size

Fre

qu

ency

anchor distance avg. min power

A2 1.39 11

A0 2.78 15

A1 3.02 16

A3 3.65 14

(without obstacles)

RealWSN 2005 Positioning in Sensor Networks 20

Experiment 3 – “Network”

• 9 nodes in a room– Distances: 1..6m

• 1 sender at a time– Send 1 packet at each level– Others: record minimum received– Report previous minima

• Round robin

• Minima:– Good approximation– Storage: save factor 100 per round

RealWSN 2005 Positioning in Sensor Networks 21

Experiment 3 – “Network” … Results

• Error– Almost 30 units for same distance– Exp. 1: “nicer” curve

• Longer range effects?

• Symmetry!Symmetry

0

100

200

300

400

500

600

700

800

900

1 6 11 16 21 26 31 36

Minimum Power Received

Fre

qu

ency

Symmetry

0

2

4

6

8

10

12

14

16

1 2 3 4 5 6

Rounded Diffence in Average Minimum Received Power

Fre

qu

ency

0

5

10

15

20

25

30

35

40

45

0 100 200 300 400 500 600

Distance (in cm)

Ave

rag

e M

inim

um

Po

wer

Rec

eive

d

RealWSN 2005 Positioning in Sensor Networks 22

Lessons

• Average minimum– Stable– Good approximation– Saves storage

• Symmetric links

• Power versus Distance– Strongly environment dependant– Measurements between two nodes

Not generalizable

• RSSI in sensor networks: good, but not for “reasonable” localization

RealWSN 2005 Positioning in Sensor Networks 23

Future Work

• Here: more questions than answers

• Hardware RSS measurements– Indication given by reviewer

• Same experiments – different hardware– Same results/trend?

• Long range vs. short range

• More environments

• New models

mica2:

in progress

Similar results

RealWSN 2005 Positioning in Sensor Networks

DistributedComputing

Group

Questions?Comments?