Location Estimation Media, Algorithms and Systems

Stephane BoreuxAlexander Hörmandinger

Introduction- Explosive trend towards mobile computing- Location is a fundamental factor of the

application / usage context- Location can be:

(i) Physical location(ii) Symbolic location

- Location can be expressed:(i) Absolute location: one reference grid for all located objects(ii) Relative location: objects have own / different reference grids

Agenda• Location estimation media• Proximity• Triangulation• Indoor systems:

- Cricket system- Hitachi Air System

• Outdoor systems:- Galileo GNSS

• Scene Analysis- RADAR system

• Positioning in 2G / 3G networks

Location Estimation media• Radio frequency

– Multipath propagation– Shadowing– Large transmission range

• Infrared– Interference with ambient light– Limited transmission range

• Ultrasound– Medium dependent– Lower speed Higher accuracy– Limited coverage

Proximity

• Symbolic informations• Intelligents Services• Example

– IP subnet detection– Active badge

Triangulation (1)

• Time of Arrival (TOA)– Distances– Errors Corrections

• Geometric middle• Clock correction

– Examples• GPS• Active BAT• Cricket System

Cricket system

• RF and ultrasound• Units can be use as beacon or receiver• Mobile unit perform the algorithm

– Privacy– Independent of users number

• Critiques– Number of units needed– Integration in existing system

• Enhanced cricket compass

Mobile device

ZX

Y

Triangulation (2)

• Time Difference of Arrival (TDOA)– Differences between distances– Hyperbolic– Remove clock offset– Examples

• Hitachi AirLocation• TruePosition

– U-TDOA• 3GPP standard

Hitachi AirLocationTM

• Capabilities– Indoor and outdoor– 1 to 3 meters accuracy– Same range as Wifi (30 to 100 m)

• Wireless LAN based System• 3*TDOA• Main purposes

– Distribution and Warehouses– Company offices – Stations, airports, and underground arcades

Hitachi AirLocationTM

Triangulation (3)

• Angle of Arrival– Measure angle of Arrival– Directional antennas– Strong influence of environment– Mostly used in hybrid system– Examples

• TruePosition

Galileo (1)• GALILEO is an European initiative for a state-of-the-art

global navigation satellite system (GNSS) driven by the European Commission and ESA

• Strategic political decision towards independence of GPS and GLONASS

• Under civilian control• Programme phases:

- Development and In-Orbit Validation (2002-2005)- Deployment (2006-2007) -> first satellite launch in December 2005

- Commercial Operations (from 2008)

Galileo (2)• Technically a refined version of GPS - > uses Time of Arrival

approach

3 equal planes with 9 operational satellites + 1 on active spare in each plane

6 equal planes with 4 satellites in each plane

Orbital planes

56°55°Inclination angle of planes (with reference to equatorial plane)

23 222 km22 200 kmOrbital altitude

30 (27 operational + 3 in reserve)

24#satellites

GALILEOGPSCharacteristic

Galileo (3)

Open Service (OS):- Will be free for anyone to access- 2 signal bands - Accuracy: <4m horiz., <8m vertic. when using both signal bandsCommerical service (CS):- Commercial fee- Encrypted, 3 signal bands- Accuracy: < 1m horiz. and vertic. / <10cm when locally augmentedPublic Related Services (PRS)Savety Of Live Services (SoL)

Precise positioning service (PPS):- Originally only for military use - Accuracy: 22m horizontally, 27.7m vertically, 200 ns temporallyStandard positioning service (SPS):- For civil usage- Originally built-in variable error to degrade the accuracy („selective availability“) Original accuracy: 100m horiz., 156m vert., 340ns temp.- Selective availability was deactivated in 2000 -> same accuracy as PPS now

GALILEO servicesGPS services

Scene AnalysisIdea: Include information / characteristics of the scene to infer

the location of an object

General steps:(i) Profiling:

- Coverage area is divided into regions- Multiple samplings of parameters at each observation point are taken and stored in location database -> „fingerprints“

(ii) Matching:- Compares real-time measurments with entries in the location database and finds the „best“ match(es)

(iii) Estimation:- derive location estimation from the „best“ match(es)

Scene Analysis – RADAR (1)

- RADAR is an RF-based system for locating users inside buildings

- Developed by a Microsoft research group- Median resolution is 2 to 3m

RADAR (2) -Testbed

• 3 WirelessLAN Access points (= basestations)

• Mobile host is a laptop with WLAN network card

• Clocks on mobile host and on the base stations are synchronized

• Mobile host broadcasts UDP packets

• Signal strength (SS) is recorded by the basestations

Picture source: 1) (see References)

Profiling - RADAR (3)- Fingerprints: at least 20 samples for each user orientation at 70 distinct

physical locations on the floor (#fingerprints >= 20 * 4 * 70 = 5600)- Location database includes tuples of the form:

),,,( issdyx }3,2,1{∈i

- For each tuple the mean, the standard deviation and the median of the SS values for each of the base stations is computed

Matching - RADAR (4)Empirical method: Compute the distance (in signal space) between the

observed set of SS measurements and the recorded SSDistance function: p

pii

n

i ip em

wnL

1

1||11⎟⎟⎠

⎞⎜⎜⎝

⎛−= ∑

=with p=2 -> Euclidian distance

Random method: Estimate the users location by picking one of the 70 locations at random

Strongest base station method: Guess the user‘s location to be the same as the location of the base station that recorded the strongest signal

Cumulative distribution func. of the error distance of the methods

Picture source: 1) (see References)

Estimation - RADAR (5)- Question: Multiple or single nearest neighbor(s) ?- Multiple neighbors because:(i) Inherent variability in the measured SS -> No reason

to pick only the closest neighbor and reject others that are almost as close

(ii) Error vector (in physical space) to each neighbor is oriented different

- Result: for small k k-neighbors has some benefit- For instance: k=5: 25th percentile of error distance is 1.5m (22% better than using single neig.)- But for large k accuracy degrades rapidly

Other aspects - RADAR (6)Impact of the number of data points?-> For n=20 the median error distance

is less than 33% worse, for n=40 it is less than 10% worse

Impact of number of samples ?-> only a small number of samples are

needed: 1 sample -> median error distance 30% worse, 2 samp. -> 11% worse, 3 samp. -> 4% worse

Impact of user orientation ?High !

Tracking ?Reduce the problem to a sequence of location determination problems-> median error distance is 19% worse than for a stationary user.

Picture source: 1) (see References)

Propagation model - RADAR (7)- Motivation: reduce dependence on empirical data

- Floor Attenuation Factor propagation model:

⎭⎬⎫

⎩⎨⎧

≥<

−⎟⎟⎠

⎞⎜⎜⎝

⎛−=

CnWCnW

WAFCWAFnW

ddndPdBmdP

**

log10)(])[(0

0

n ... the increase of path loss with distance -> derived empirically)( 0dP ... signal power at some reference distance -> empir. or from specification

d ... transmitter-receiver distance

C ... maximum number of walls up to which the attenuation factor makes a difference

nW ... the number of walls between sender and receiver

WAF ... wall attenuation factor -> derived empirically

- Amazing results !

Picture source: 1) (see References)

Location estimation with 2G / 3G networks (1)

- Driving forces:(i) public: emergency services, intelligence(ii) customers ?

- Same basic techniques as in other positioning systems:(i) Time of arrival(ii) Angle of arrival(iii) Signal strength(iv) proximity (cell identification)

- Positioning system classification:(i) Self-Positioning: location estimation is done by MS(ii) Remote Positioning: BSs measure signals from the MS, these measurements are processed at a central site to estimate the location (network-based)

2G / 3G networks – Ericsson MPS- server-based GMPC (Gate-

way mobile positioning center) for positioning requests (PR) from an application

- server-based SMPC receives PR and returns loc. estimation of MS

- Uses standardized Mobile Location Protocol (MLP) between App. and GMPC

- System according to 3GPP standards

- Positioning methods:(i) CGI-TA and ATI for low accuracy(ii) Assisted GPS for high accuracy

Picture source: 12) (see References)

2G / 3G networks – Nokia mPosition

Position tech. in terminal:

(i) Standalone GPS

(ii) Assisted GPS

Position tech. in network:

(i) Cell identification (CI)

(ii) Enhanced CI (CI + Signal strength or Timing Advance

(iii) Assisted GPSPicture source: 13) (see References)

Conclusions

- Same basic techniques used in all types of systems

- A given infrastructure can be used to gain high location accuracy (sometimes)

- 2G/3G LBS architectures are available- 2G/3G systems are in need of a hybrid approach

(GPS) to achieve high accuracy- Current usage of this systems ?- Where is the killer application ?

References1) RADAR: An In-Building RF-based User Location and Tracking System, Paramvir Bahl,

IEEE Infocom 20002) On Indoor Position Location With Wireless Lan, P. Prasithsangaree, IEEE 20023) http://europa.eu.int/comm/dgs/energy_transport/galileo/4) http://www.esa.int/esaNA/5) www.wikipedia.org6) http://www.trueposition.com7) The Enhanced Cricket Compass with Ad Hoc Network, Kisuk Kweon, Computer

Architecture Laboratory, Department of Computer Science, KAIST8) The Cricket Location-Support System, Nissanka B. Priyantha, Anit Chakraborty, and

Hari Balakrishnan, MIT Laboratory for Computer Science, Cambridge9) Position Information System by Local Mobile Phone Network, Yuji Yamamoto,

Hirokazu Sato, Institute of Technology, Shimizu Corporation, Tokyo10) Overview of Radiolocation in CDMA Cellular Systems, James Caffery, IEEE

Communication Magazine April, 199811) Positioning GSM Telephones, Christopher Drane, IEEE Communications Magazine,

April 199812) http://www.ericsson.com/mobilityworld/13) http://www.nokia.com/link?cid=EDITORIAL_357