ESE 343 & 543

Localization Techniques

Lecture 3

Shan Lin
 What is Localization

A mechanism for discovering

spatial relationships between
objects
 Why is Localization Important?
Large scale mobile systems introduce many
fascinating applications with difficult problems…

Localization measures the spatial coupling

Temperature readings  temperature map
Asset tagging  asset tracking
“Smart spaces”  context dependent behavior
Google map services  navigation and traffic prediction
 The History of GPS

 Feasibility studies begun in 1960’s.

 Pentagon appropriates funding in 1973.
 First satellite launched in 1978.
 System declared fully operational in April, 1995.
 Department of Defense finally said:
“we need something better: all-day and all-night; all terrain ”

end-product is Global Positioning System (GPS)

• system (constellation) of 24 satellites in high altitude orbits

(cost ~ $12 billion)
• coded satellite signals that can be processed in a GPS
receiver to compute position, velocity, and time
• parts of system include:
space (GPS satellite vehciles, or SVs)
control (tracking stations)
users

first one launched in 1978

….June 26, 1993
Air Force launched 24th SV
 GSP satellite vehicles (SVs):
two generations: block I and block II

GPS block II
weigh ~1900 lbs.

GPS block I built by Rockwell

basic concept is that the GPS constellation replaces “stars” and
gives us reference points for navigation

examples of some applications (users):

• navigation (very important for ocean travel)
• zero-visibility landing for aircraft
• collision avoidance
• surveying
• precision agriculture
• delivery vehicles
• emergency vehicles
• electronic maps
• Earth sciences (volcano monitoring; seismic hazard)
• tropospheric water vapor

anything that involves location, motion, or navigation

step 1: using satellite ranging
step 2: measuring distance from satellite
step 3: getting perfect timing
step 4: knowing where a satellite is in space
step 5: identifying errors
Position is Based on Time

Signal leaves satellite

at time “T”

Signal is picked up by
T + 3 the receiver at time “T +
3”
Distance between satellite
and receiver = “3 times the
speed of light”
 Pseudo Random Code

Time
Difference

Satellite PRN

Receiver PRN
Signal From One Satellite

The receiver
is somewhere
on this
sphere.
Signals From Two Satellites
Three Satellites (2D Positioning)
Three Dimensional (3D) Positioning
 Sources of GPS Error
Standard Positioning Service (SPS ): Civilian Users
 Source Amount of Error
 Satellite clocks: 1.5 to 3.6 meters
 Orbital errors: < 1 meter
 Ionosphere: 5.0 to 7.0 meters
 Troposphere: 0.5 to 0.7 meters
 Receiver noise: 0.3 to 1.5 meters
 Multipath: 0.6 to 1.2 meters
 Selective Availability (see notes)
 User clock error: Up to a kilometer Errors
are cumulative and increased by PDOP.
Receiver Errors are Cumulative!

System and other flaws = < 9

meters

User error = +- 1
km
 Sources of Signal Interference

Earth’s Atmosphere

Solid
Structures

Metal Electro-magnetic
Fields
 Four Basic Functions of GPS

 Position and coordinates.

 Thedistance and direction between any two

waypoints, or a position and a waypoint.

 Travel progress reports.

 Accurate time measurement.

 Position Fix

A position is based on real-time satellite tracking.

 It’s defined by a set of coordinates.
 It has no name.
 A position represents only an approximation of the

receiver’s true location.

 A position is not static. It changes constantly as the

GPS receiver moves (or wanders due to random

errors).
 A receiver must be in 2D or 3D mode (at least 3 or 4

satellites acquired) in order to provide a position fix.

 3D mode dramatically improves position accuracy.
 Waypoint
A waypoint is based on coordinates entered into a
GPS receiver’s memory.
 It can be either a saved position fix, or user entered

coordinates.
 It can be created for any remote point on earth.
 It must have a receiver designated code or number,

or a user supplied name.

 Once entered and saved, a waypoint remains

unchanged in the receiver’s memory until edited or

deleted.
 Planning a Navigation Route

= Waypoint
Start
How A Receiver Sees Your Route
GPS Waypoint Circle of Error

X
 Real Time Differential GPS

x+5, y-
3
x+30,
y+60
x-5,
y+3

Receiver DGPS Receiver

DGPS Site
DGPS correction = x+(30-5) True coordinates =
and y+(60+3) x+0, y+0
True coordinates = x+25, Correction = x-5,
y+63 y+3
 NDGPS Ground Stations
National Differential Global Positioning System

Yellow areas show overlap between NDGPS stations. Green areas are little to
no coverage. Topography may also limit some areas of coverage depicted here.
 Variety of Applications

Two applications:

Passive habitat monitoring: Asset tracking:

Where is the bird? Where is the projector?
What kind of bird is it? Why is it leaving the room?
 Variety of Application
Requirements
 Very different requirements!
Outdoor operation Indoor operation
Weather Multipath
problems problems
Bird is not tagged Projector is tagged
Birdcall is characteristic Signals from projector
but not exactly known tag can be engineered
Accurate enough to Accurate enough to track
photograph bird through building
Infrastructure: Infrastructure:
Several acoustic Room-granularity
sensors, with tag identification
known relative and localization;
locations; coordination with
 Multidimensional Requirement
Space

Granularity & Scale

Accuracy & Precision
Relative vs. Absolute Positioning
Dynamic vs. Static (Mobile vs. Fixed)
Cost & Form Factor
Infrastructure & Installation Cost
Communications Requirements
Environmental Sensitivity
Cooperative or Passive Target
 Axes of Application Requirements
Granularity and scale of measurements:
What is the smallest and largest measurable distance?
e.g. cm/50m (acoustics) vs. m/25000km (GPS)
Accuracy and precision:
How close is the answer to “ground truth”
(accuracy)?
How consistent are the answers (precision)?
Relation to established coordinate system:
GPS? Campus map? Building map?
Dynamics:
Refresh rate? Motion estimation?
 Axes of Application Requirements
Cost:
Node cost: Power? $? Time?
Infrastructure cost? Installation cost?
Form factor:
Baseline of sensor array
Communications Requirements:
Network topology: cluster head vs. local determination
What kind of coordination among nodes?
Environment:
Indoor? Outdoor? On Mars?
Is the target known? Is it cooperating?
 Returning to our two Applications…

Choice of mechanisms differs:

Passive habitat monitoring: Asset tracking:

Minimize environ. interference Controlled environment
No two birds are alike We know exactly what tag is like
 Variety of Localization Mechanisms

 Very different mechanisms indicated!

Bird is not tagged Projector is tagged
Passive detection Projector might
of bird presence know it had
Birdcall is characteristic moved
but not exactly known Signals from projector
Bird does not have tag can be engineered
radio; TDOA Tag can use radio signal to
measurement enable TOF measurement
Passive target Cooperative Localization
localization
Requires
Requires
Basic correlator
Sophisticated
Simple
detection
triangulation
 Taxonomy of Localization Mechanisms
Active Localization
System sends signals to localize target
Cooperative Localization
The target cooperates with the system
Passive Localization
System deduces location from observation
of signals that are “already present”
Blind Localization
System deduces location of target without
a priori knowledge of its characteristics
 Active Mechanisms
Target
Synchronization channel
Ranging channel

Non-cooperative
System emits signal, deduces target location
from distortions in signal returns
e.g. radar and reflective sonar systems
Cooperative Target
Target emits a signal with known
characteristics; system deduces location by
detecting signal
e.g. ORL Active Bat, GALORE Panel, AHLoS
Cooperative Infrastructure
Elements of infrastructure emit signals;
target deduces location from detection of
signals
e.g. GPS, MIT Cricket
 Passive Mechanisms
Target
Synchronization channel
Ranging channel
Passive Target Localization
Signals normally emitted by the target are
detected (e.g. birdcall)
Several nodes detect candidate events and
cooperate to localize it by cross-correlation
Passive Self-Localization
A single node estimates distance to a set
of beacons (e.g. 802.11 bases in RADAR
[Bahl et al.], Ricochet in Bulusu et al.)
Blind Localization
Passive localization without a priori
knowledge of target characteristics ?
Acoustic “blind beamforming” (Yao et al.)
 Active vs. Passive

Active techniques tend to work best

Signal is well characterized, can be engineered for noise
and interference rejection
Cooperative systems can synchronize with the target to
enable accurate time-of-flight estimation
Passive techniques
Detection quality depends on characterization of signal
Time difference of arrivals only; must surround target
with sensors or sensor clusters
 TDOA requires precise knowledge of sensor positions
 Example of a Localization System
Unattended Ground Sensor and acoustic
localization system.

Each node has 4 speaker/

microphone pairs, arranged
Microphone along the circumference of the
Speaker enclosure. The node also has a
radio system and an orientation
sensor.
12 cm
 System Architecture
 Ranging between nodes based on detection of coded acoustic
signals, with radio synchronization to measure time of flight
 Angle of arrival is determined through TDOA and is used to
estimate bearing, referenced from the absolute orientation
sensor
 An onboard temperature sensor is used to compensate for the
effect of environmental conditions on the speed of sound
 Active and Cooperative Ranging

Measurement of distance between two points

Acoustic
 Point-to-point time-of-flight, using RF synchronization
 Narrowband (typ. ultrasound) vs. Wideband (typ. audible)
RF
 RSSI from multiple beacons
 Transponder tags (rebroadcast on second frequency),
measure round-trip time-of-flight.
 UWB ranging (averages many round trips)
 Psuedoranges from phase offsets (GPS)
 TDOA to find bearing, triangulation from multiple stations
Visible light
 Stereo vision algorithms
 Need not be cooperative, but cooperation simplifies the problem
 Passive and Non-cooperative
Ranging
Generally less accurate than active/cooperative
Acoustic
 Reflective time-of-flight (SONAR)
 Coherent beamforming (Yao et al.)
RF
 Reflective time-of-flight (RADAR systems)
 “Database” techniques
 RADAR (Bahl et al.) looks up RSSI values in database
 “RadioCamera” is a technique used in cellular infrastructure;
measures multipath signature observed at a base station
Visible light
 Laser ranging systems
 Commonly used in robotics; very accurate
 Main disadvantage is directionality, no positive ID of target
 Using RF for Ranging

Disadvantages of RF techniques
Measuring TOF requires fast clocks to achieve high
precision
Building accurate, deterministic transponders is very
difficult
 Temperature-dependence problems in timing of path from
receiver to transmitter
Systems based on relative phase offsets (e.g. GPS)
require very tight synchronization between transmitters
 RSSI…? Don’t Bother

 RSSI is extremely problematic

Path loss characteristics depend on
environment (1/rn) Path loss
Shadowing depends on environment Shadowing
Fading
Short-scale fading due to multipath
adds random high frequency

RSSI
component with huge amplitude
(30-60dB) – very bad indoors
 Mobile nodes might average out
fading.. But static nodes can be stuck
in a deep fade forever
 Possible applications
Crude localization of mobile nodes Distance
“Database” techniques (RADAR)
 Using Acoustics for Ranging

Key observation: Sound travels slowly!

Tight synchronization can easily be achieved using RF
signaling
Slow clocks are sufficient (v = 1 ft/ms)
With LOS, high accuracy can be achieved cheaply
Coherent beamforming can be achieved with low sample
rates
Disadvantages
Acoustic emitters are power-hungry (must move air)
Obstructions block sound completely  detector picks
up reflections
Existing ultrasound transducers are narrowband
 Typical Time-of-Flight AR System

Radio channel is used to synchronize the sender

and receiver
Coded acoustic signal is emitted at the sender and
detected at the emitter. TOF determined by
comparing arrival of RF and acoustic signals

Radio Radio

CPU CPU
Speaker Microphone
 Typical Angle-of-Arrival AR
System

TOF AR system with multiple receiver channels

Time difference of arrivals at receiver used to
estimate angle of arrival

Radio Radio

CPU CPU
Microphone
Speaker Microphone
Microphone
Microphone
Array
 Is GPS the solution?

Pros: Good accuracy Cons: Poor battery lifetime

Is GSM the solution?

Pros: Long battery lifetime Cons: Poor accuracy

 What about WiFi Localization?

E.g., SkyHook:

Basic
Basic Idea:
Idea:
1.
1. Several
Several trucks
trucks war-drive
war-drive aa place
place
2.
2. Create
Create Radio
Radio map
map == <Location:
<Location: WiFi
WiFi IDs>
IDs>
3.
3. Distribute
Distribute map
map to
to phones
phones
4.
4. Phone
Phone user
user goes
goes to
to war-driven
war-driven region,
region, overhears
overhears WiFi
WiFi IDs
IDs
5.
5. Reverse
Reverse Look
Look Up
Up IDs
IDs against
against radio
radio map
map
6.
6. Obtains
Obtains location
location
Is Skyhook the solution?
Middle Ground
Lower Accuracy than GPS, Longer Battery
lifetime
Better Accuracy than GSM, Shorter Battery
lifetime
but …

At the cost of:

Degraded location accuracy: walking paths ~
60m
Reliance on infrastructure (APs)
War-driving ($$ + carbon footprint)
“NYTimes: Skyhook fleet 500 trucks/drivers”
 No Eng. Eff. &
Acc. Solution

Skyhook
Better than
GPS Eng. Eff.
SkyHook

GPS unusable with phone

battery
GSM too inaccurate
 Eng. Eff. & Acc.
Solution

CompAcc

GPS unusable with phone

GPS unusable
batterywith
phone battery
GSM too inaccurate
 Contents
 CompAcc

 Evaluation

 Limitations and Future Work

 Conclusion
 Goals
 No War-Driving
 Cannot drive walking paths (campus, parks,
…)
 Expensive / Environment unfriendly

 No reliance on WiFi infrastructure

 Rural regions / developing countries

 Good accuracy (~GPS)

 Improve energy-efficiency
 Better than Skyhook, GPS
 CompAcc: Basic Idea
 Direction(compass) + Displacement(accelerometer) = User’s
directional trail

Directional Trail
 CompAcc: Basic Idea
 Direction(compass) + Displacement(accelerometer) = User’s
directional trail

…
u re
Si gnat
Pat h

 Compute path signatures

Derived from a local electronic map (Google Maps)
 CompAcc: Basic Idea
 Direction(compass) + Displacement(accelerometer) = User’s
directional trail

Directional Trail

…
u re
Si gnat
Pat h

 Compute path signatures

Derived from a local electronic map (Google Maps)

 Compare directional trail with path signatures

Best match provides the user location
 Correct location errors at
turns

al Trail
ection
…
Dir

gn ature
Si
Path
 Correct location errors at
turns

al Trail
ection
…
Dir

gn ature
Si
Path

Directional Trail

…
ture
h Sig n a
Pat
 Advantages

 No war-driving
 No reliance on WiFi infrastructure
 Maps available ubiquitously

 Improves battery lifetime

 GPS ~10h
 Skyhook ~16h
 Accelerometer ~ 39h
 Compass ~48h
 Architecture

Tile
1. Initial location GPS: Database
(lat X, long Y)

Tile
3. Obtain paths in the
user vicinity 6. Current location
(lat A, long B)

2. Report initial location Initial location

(lat X, long Y) Directional trail
4. Direction Current location
(Compass)

5. Displacement
CompAcc
(Accelerometer)
 Directional trail: displacement
Accelerometer based step count
displacement = step_count * step_size
Directional trail: direction
Directional trail: direction
Matching Directional Trail with Path
Signatures

Directional Trail

Path Signature

Dissimilarity Metric:

ci = compass readings
pi = path computed direction
N = directional trail size
 Contents
 CompAcc

 Evaluation

 Limitations and Future Work

 Conclusion
 Results

Compared 3 localization schemes

CompAcc
Skyhook
Wifi-War-Walk (We war-droved walking paths in
campus)

Metrics
Instantaneous Error = distance(estimated, real)
Average Localization Error (ALE) = Average Instantaneous Error
CompAcc Instantaneous Error
CompAcc Instantaneous Error
Results
Results
Results

Average ALE
GPS: 10m
CompAcc: 11m
WiFi-War-Walk: 30m
Skyhook: 70m

Energy
GPS: 10h
CompAcc: 23h
WiFi-War-Walk:16h
Skyhook:16h
 Location-Based Applications (LBAs)

For Example:
GeoLife shows grocery list when near Walmart
MicroBlog queries users at a museum
Location-based ad: Phone gets coupon at
Starbucks

iPhone AppStore: 3000 LBAs, Android: 500 LBAs

 Most emerging location based apps
do not care about the physical location

GPS: Latitude, Longitude

 Most emerging location based apps
do not care about the physical location

GPS: Latitude, Longitude

Instead, they need the user’s logical location

Starbucks, RadioShack,
Museum, Library
 Physical Vs Logical

Unfortunately, most existing solutions are physical

GPS
GSM based
SkyHook
Google Latitude

RADAR
Cricket
…
 Given this rich literature,

Why not convert from

Physical to Logical Locations?
Physical Location
Error
 Starbucks Pizza Hut

Physical Location
Error
 Starbucks Pizza Hut

Physical Location
Error

The dividing-wall problem

 SurroundSense:
A Logical Localization Solution
 Hypothesis

It is possible to localize phones by

sensing the ambience

such as sound, light, color, movement, WiFi …

 Hypothesis

It is possible to localize phones by

sensing the ambience

such as sound, light, color, movement, WiFi …

Multi-dimensional sensing extracts more
ambient information

Any one dimension may not be unique,

but put together, they may provide a
unique fingerprint
 SurroundSense
Multi-dimensional fingerprint
Based on ambient
sound/light/color/movement/WiFi

Starbucks Pizza Hut

Wall
Should Ambiences be Unique Worldwide?

J
H I P
A Q
K CB D
E L
Q
R N M
O F
G
GSM provides
Should macro
Ambiences locationWorldwide?
be Unique (strip mall)
SurroundSense refines to Starbucks

J
H I P
A Q
K CB D
E L
Q
R N M
O F
G
 Why does it work?

The Intuition:
Economics forces nearby businesses to be diverse

Not profitable to have 3 adjascent coffee shops

with same lighting, music, color, layout, etc.

SurroundSense exploits this ambience diversity

 SurroundSense Architecture

Ambience Fingerprinting Matching

Sound
Test
Color/Light Fingerprint
+
Acc.
=
WiFi

Logical
Fingerprint
GSM Macro Location
Database
Location

Candidate Fingerprints
 Fingerprints Acoustic fingerprint
(amplitude distribution)

 Sound:
0.14

(via phone

Normalized Count
0.12

0.1

microphone) 0.08

0.06

0.04

0.02
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
0
 Color:
Amplitude Values

(via phone Color and light fingerprints on HSL space

camera)
Lightness 1

0.5

00
1
0.8
0.5 0.6
0.4
0.2
Hue 1 0 Saturation
 Fingerprints
 Movement: (via phone accelerometer)

Cafeteria Clothes Store Grocery Store

Moving

Static
 Fingerprints
 Movement: (via phone accelerometer)

Cafeteria Clothes Store Grocery Store

Moving

Static

Queuing Seated
 Fingerprints
 Movement: (via phone accelerometer)

Cafeteria Clothes Store Grocery Store

Moving

Static

Pause for product Short walks between

browsing product browsing
 Fingerprints
 Movement: (via phone accelerometer)

Cafeteria Clothes Store Grocery Store

Moving

Static

Walk more Quicker stops

 Fingerprints
 Movement: (via phone accelerometer)

Cafeteria Clothes Store Grocery Store

Moving

Static

 WiFi: (via phone wireless card)

ƒ(overheard WiFi APs)

 Discussion
Time varying ambience
Collect ambience fingerprints over different
time windows

What if phones are in pockets?

Use sound/WiFi/movement
Opportunistically take pictures

Fingerprint Database
War-sensing
 Evaluation Methodology
51 business locations
46 in Durham, NC
5 in India

Data collected by 4 people

12 tests per location

Mimicked customer behavior

 Evaluation: Per-Cluster Accuracy

Cluster 1 2 3 4 5 6 7 8 9 10
No. of Shops 4 7 3 7 4 5 5 6 5 5

Localization accuracy per cluster

Accuracy (%)

Cluster
 Evaluation: Per-Cluster Accuracy

Cluster 1 2 3 4 5 6 7 8 9 10
No. of Shops 4 7 3 7 4 5 5 6 5 5

Localization accuracy per cluster

Fault tolerance
Accuracy (%)

Cluster
 Evaluation: Per-Cluster Accuracy

Cluster 1 2 3 4 5 6 7 8 9 10
No. of Shops 4 7 3 7 4 5 5 6 5 5

Sparse WiFi
Localization accuracy APs
per cluster
Accuracy (%)

Cluster
 Evaluation: Per-Cluster Accuracy

Cluster 1 2 3 4 5 6 7 8 9 10
No. of Shops 4 7 3 7 4 5 5 6 5 5

Localization accuracy per cluster

Accuracy (%)

No WiFi APs

Cluster
 Evaluation: Per-Scheme Accuracy

Mode WiFi Snd-Acc-WiFi Snd-Acc-Lt-Clr SS

Accuracy 70% 74% 76% 87%
 Evaluation: User Experience

Random Person Accuracy

1 WiFI
0.9 Snd-Acc-WiFi
Snd-Acc-Clr-Lt
0.8 SurroundSense
CDF

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70 80 90 100
Average Accuracy (%)
 Limitations and Future Work

Energy-Efficiency
Continuous sensing likely to have a large energy
draw
 Limitations and Future Work

Energy-Efficiency
Continuous sensing likely to have a large energy
draw

Localization in Real Time

User’s movement requires time to converge
 Limitations and Future Work

Energy-Efficiency
Continuous sensing likely to have a large energy
draw

Localization in Real Time

User’s movement requires time to converge

Non-business locations
Ambiences may be less diverse
 Conclusion
Ambience can be a great clue about location
Ambient Sound, light, color, movement …

None of the individual sensors good enough

Combined they may be unique

Uniqueness facilitated by economic incentive

Businesses benefit if they are mutually diverse in
ambience

Ambience diversity helps SurroundSense

Current accuracy of 89%