Localization protocols for wireless sensor networks Stefan Dulman Email: [email protected].

Localization protocols for wirelesssensor networks

Stefan DulmanEmail: [email protected]

energy efficient sensor networks

Presentation Overview

Introduction and motivation

Lateration – a simple approach

ClassificationCentralized methods

One hop positioning

Distributed methods

Relative positioning

Mobility schemes

Conclusions


Wireless Sensor Networks

Nodes with VERY limited resources16 bit processor, 2KB RAM, 60KB FLASH

Low data-rate radio (115.200 bits/sec)

Limited energy available (1-2 small batteries)

Small physical size

Networks characteristicsDistributed network, mobile unreliable nodes

Deployed in a harsh environment

Self-organizing and self-healing


Ideal Sensor Networks

“Smart Dust”Tens of thousands of sensor nodes

Node lifetime longer than 2 years

Node size smaller than 1 mm3

Node price smaller than 5 cents


Typical Applications

A broad range of applications, all possibilities definitely not explored yet!Environment and wildlife monitoring

Remote study of the birds on Great Duck IslandZebra monitoring in Kenya

… and even more monitoringIndustryAgricultureDisaster controlSurveillance and security systems


Is positioning necessary?

YES!Can be the mean or the goal of a WSN applicationApplication examples:

Meteorological and environmental monitoringData has no meaning if not stamped with time + location

Package trackingLibrary archivingPosition tracking (e.g. military applications)

Is used as building block in:Routing protocolsData dissemination protocolsLocalization as application of WSN


Example: geographic routing

Allows development of algorithms with better scalability

Position centric addressing first proposed in 1970’sRecent growing interest for it

Nodes are addressed by their location instead of ID

No additional job required to support routing

State of the packet (position) and destination position are sufficient

Simplest algorithm: Cartesian routing

Stojmenovic (IEEE Commun.Magazine 2002) presents several strategies for geographical routing


Problem statement

Regular assumptions for WSN protocol test scenarios:Large number of nodes

Random deployment in a (known shape) given area

Known (identical) transmission range for all nodes

Static/not very dynamic networks

Question: What are the geographical positions of the nodes?

Absolute positioning



A possible solution?

Usage of Global Positioning System (GPS) devices

Not a feasible solution for WSN:High cost of the device (value/energy/computation power/space)

Unavailability/poor precision of the service in special environments (indoors, underground, etc.)

Conclusion: other approaches need to be developed and deployed


Lateration - A simple solution


Lateration description

Example: 2D space

Given:Three points with known positions

Distances to all three of them

Position can be determined by intersecting three circle centered in the points with radius the known distances


Lateration as localization technique


Lateration’

The concept can be easily applied to multihop networks

The method as such is not too useful:Imprecise position information

Imprecise distance estimates

The three circles usually do not intersect in a point (or at all!)

Several algorithms developed on this simple idea (e.g. APS schemes)


Classification


Classification

Different aspects of localization studied in vision, robotics, signal processing, networking, etc.

Solutions can be classified in several manners:One-hop or multi-hop schemes

Range free or range based schemes

Absolute, relative or local coordinates

Centralized, distributed or localized algorithms


Centralized methods


Centralized methods

All the data is collected at a central point and a global map is computed at once

Advantages:High quality solutions (in terms of the average distance error)

Global maps available

Disadvantages:Data needs to travel to a central point

High computation power required

Methods usually do not scale with the network size


Centralized methods

Convex optimizationOne of the first schemes available

Treats the localization problem from the point of view of linear programming and semi-definite programming

Various constraints are represented as linear matrix inequalities


Convex optimization’

Advantages:It is simple to model the distance and angle information

The solutions provided are optimal

Efficient computational methods have already been developed

Disadvantages:All the disadvantages of the centralized methods class

Computation complexities:Linear programming is quadratic in the number of connections

Semi-definite is cubic in the number of connections


Convex optimization’

Conclusions:Increase in connectivity results in increase in accuracy^2 but also traffic overhead

Precision of distance and angle information have a direct influence on the output

Bad data (missing/false connections) leads to algorithm failure


Centralized methods

Multidimensional scalingMDS-MAP method, makes use of connectivity information (and also distance information) to compute a relative (global) map

Finds an embedding in a lower dimensional space for a set of objects characterized by pair-wise distances between them

Recently the centralized method has been extended to a localized algorithm


Avig- Biar- Bor-

Brest Caen Calais

Cher- Clermont

Dijon

le

non ritz deaux bourg Havre

Avignon — 388 352 735 570 611 645 238 261 547

Biarritz 625 — 118 510 502 646 460 437 634 598

Bordeaux 567 190 — 391 379 528 342 227 475 479

Brest 1183 820 630 — 234 449 249 560 539 298

Caen 918 808 610 377 — 217 75 398 339 68

Calais 983 1040 850 722 350 — 293 442 355 174

Cherbourg 1038 740 550 401 120 471 — 473 413 142

Clermont 383 703 366 901 641 711 761 — 214 378

Dijon 420 1021 765 867 545 571 665 345 — 323

le Havre 881 963 771 480 109 280 229 608 520 —


Multidimensional scaling

DisadvantagesThe regular ones for centralized methods

AdvantagesIt is a quite precise method (one of the best so far)

Can work only with connectivity (and distance) information

Only 3 anchors are needed for a global map (in 2D)

Theoretical bound on the complexity of the method(SVD is cubic in the number of nodes)

A recent paper describes a localized scheme based on MDSCan deal with topologies containing holes

Eases the complexity of the computation




One hop positioning


One hop positioning

Nodes can directly contact the landmarks (e.g. GPS)

AdvantagesElegant solutions with precise results

DisadvantagesLine of sight is needed between the nodes and the landmarks

Landmarks need to be powerful devices


One hop positioning’

The lighthouse systemPositioning of an entire field of sensors may be achieved with only a single lighthouse device capable of “seeing” all the nodes

This device is able to collect all the data and at the same time help nodes localize themselves

The system requires each node to be equipped with a photo detector and a clock



Light House System’

Advantages:Simple theoretical method

The prototype used a single lighthouse device

For the 2D experiments: nodes situated at 14 meters positioned themselves with a relative accuracy of 2.2% and relative standard deviation of 0.68%.

Disadvantages:The line of sight assumption is a very strong one

Solution specific to the hardware


Distributed methods


Distributed methods

These methods allow nodes to compute their position by communicating to their neighbors only

Advantages:No need of global knowledge

Simple methods, majority of algorithms fit the hardware

Lower communication overhead

DisadvantagesHigh number of anchors needed

Not all the nodes can compute their position

The resulting positions are less precise


Distributed methods

Ad hoc localization system (AhLOS)Defines and combines several types of multilateration

Its main strong point is that it is a completely distributed protocol

Its weakest point is that the number of needed anchors should be large for a good result


AhLOS’

Initial phaseSome nodes can compute their position directly using lateration

These nodes behave as anchors for all the others, algorithms goes on iteratively

The position precisions degrade with the number of steps

Additional phasesSome nodes cannot find the position Collaboration groups are identified

Position is identified in a collaborative manner


AhLOS’

Algorithm might not have a convergence point

In general it fails for collaboration groups different than the presented one


AhLOS’

Advantages:AhLOS may produce very good results if accurate distance measurement hardware is present

Major disadvantage: very large number of beacons is required

Example: av. connectivity ~6.28

90% of regular nodes positions to be resolved

45% landmarks required


Distributed methods

Ad-Hoc positioning systems (APS or DV…)It is a combination between two major ideas:

Distance vector routing (DV)(information is forwarded hop by hop from each anchor in the network)

Global positioning system (GPS)(eventually each node will compute its position based on anchors positions and distance estimates)

The schemes adapt connectivity, distance, angle of arrival and compass information(6 possible combinations make sense and were studied)


APS’

DVHopThe simplest protocol available

Makes use only of connectivity information

Basic ideas:Number of hops between anchors and nodes are computed

Average hop distance is estimated

Position is computed via lateration - (W)LS method


DVHop practice - theory


APS’

DVDistanceIdentical with DVHop, but shortest path distance is propagated instead of hopcount

A new parameter has to be taken into consideration: the time to live (TTL) of the messages

There is a close connection between TTL and the final precision

-0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4-0.2

0

0.2

0.4

0.6

0.8

1

1.2

-0.2 0 0.2 0.4 0.6 0.8 1 1.2-0.2

0

0.2

0.4

0.6

0.8

1


APS’

EuclideanIf accurate distance measurements are available, nodes can estimate exact distance to anchors

DVBearingAngle estimates are used in order to determine the relative positions

In both methods, lateration is applied as the last step independently at each node



APS conclusions

Advantages:Distributed and localized protocols

Support some limited mobility (periodic schemes)

Can deal with various combinations of connectivity, distance estimates, angle estimates, compass information

DisadvantagesUniform distribution of anchors required

The DV component will ask for a high cost in case mobile scenarios





Relative positioning schemes generate a relative map, in a local coordinate system

Obtained positions are coherent all over the network (position based services are able to work)


Relative positioning’

The self positioning algorithm (SPA)The coordinate system is determined by a location reference group (LRG)

Nodes exchange info with neighbors to determine second neighborhood information (connectivity + distances)

Local maps are constructed

LRG helps orienting all the maps by aligning all the coordinate systems


SPA’

The approach is quite similar to the distributed version of multidimensional scaling algorithm

Advantages:Network-wide coherence is provided

No landmarks (anchors) are needed

Disadvantages:Existence of a location reference group is expensive if mobility is taken into account (even limited mobility and even if particular conditions apply)


Mobility schemes


Localization and mobility

All the presented schemes might work in presence of limited mobility

The basic mechanism would be to run periodically the algorithms

The computed positions will almost never reflect reality (computation takes time) but trajectories could be estimated, etc.


Distinct approach

Sequential Monte Carlo localization (Mobicom 2004)Adaptation of Monte Carlo localization used in robotics

Discrete time model

Posterior distributions are computed based on a set of weighted samples

Each step is divided in two:Prediction phase (new position estimates are computed)

Filtering phase (position estimates are filtered and space is resampled)

SMC already applied to target tracking, robot localization and computer vision


SMCL

Initial phase: initially, position is unknown so the sample space is randomly chosen

Step n:Phase 1: location set is updated based on observations and possible movements

Speed information is used (speed distribution assumed known)

Phase 2: computed positions inconsistent with observations are removed. Re-sampling is used to assure constant number of samples

Observations = anchors (‘seeds’) are present one or two hops away


SMCL’


Conclusion SMCL

Completely counter-intuitive!!!

Mobility improves localization results and reduces localization cost

More amazing: this is a range-free scheme (no distance estimates are used, only connectivity information)

A lot of work to be done: testing of all assumption, different mobility models, security, etc.


Conclusions


Conclusions

Localization is an essential service (both goal and mean)Line-of-sight solutions (GPS) less attractive in WSN

New schemes need to be developed

Large variety of assumptionsno “one size fits all” protocol exists

We described just the main ideas used, combinations lead to a huge number of protocols !

Large majority of protocols “proven by simulation”

Latest publications show increased interest for theoretical results

Localization protocols for wireless sensor networks Stefan Dulman Email: [email protected].

Documents

deployed slide

sensor nodes node lifetime

application of wsn slide

geographical routing

cents slide

dynamic networks question

simple solution slide

selfhealing slide