Error Trends In Multihop Localization

1
Error Trends In Multihop Localization

• Andreas Savvides
• Networked and Embedded Systems Lab
• Preliminary Presentation for IPSN03

2
Multihop Ad-Hoc Localization

• Many types of localization, each with different
  requirements
• To support network functions e.g geo-routing,
  based services
• Sensor networks in different environments (more
  fine grained)
• No one fits all solution yet
• Ultrasonic, acoustic, laser, radio ToF, RSS
  Methods
• Centralized vs. Distributed
• Measurement quantities magnetic field, distance,
  angles
• What are the trends for large scale systems?
• How well can localization work as systems scale?

3
Problem Setup

• Randomly dispersed network where some nodes know
  their locations
• How scalable is this setup?
• What is the best one can effect
• Our experimental system based on small sensor
  nodes and ultrasonic distance measurements
• Many aspects and tradeoffs need to be considered
• Power consumption, computation, cost, latency,
  accuracy

4
Evolving Ranging Technologies
5
Error Trends in Multihop Networks

• Sensor measurements are noisy
• Additional error added before the result is
  computed
• Broad classification of error
• Channel error transducers are not perfect,
  channel variations introduce significant error
• Algorithmic and computation error algorithms
  make a set of underlying assumptions,
  approximations, computation error
• Setup error associated with node configuration
  and network topology parameters

6
Setup Error

• Induced by measurement error BUT also reflected
  in the network configuration parameters
• Deployment geometry
• Network density
• Beacon concentration
• Measurement technology accuracy
• Certainty in beacon locations
• Study these trends in different network
  configurations and topologies using the CR-Bound

7
Cramer-Rao Bound

• Classical result from statistics that gives a
  lower bound on the error covariance matrix of an
  unbiased estimate
• Our experiments assume that the underlying
  measurement error distribution is White-Gaussian
• Based on our ultrasonic localization system
  measurements
• Not always the case, but good enough to reveal
  some trends, useful to keep in mind during design
  and deployment time

8
Scenario Setup

• Scenarios designed to maintain uniform density
• Generated in a radial pattern while controlling
  the number of nodes/unit area
• Beacon nodes are placed in the outer ring

9
Geometry Effects
10
Geometry Effects II
11
Geometry Effects III
12
Geometry Effects IV
13
Density Trends
6 neighbors per node
RMS Error (m)
14
Density Effects with Different Ranging
Technologies
6 neighbors
12 neighbors
RMS Error(m)
15
Error propagation as network scales
10 beacons 6 neighbors per node
16
Effect of Adding More Beacons
RMS Error(m)
100 nodes 4 to 20 beacons
17
How does Collaborative Multilateration Compare to
the Bounds?
18
An Ultrasonic Localization System

• Why ultrasound
• Low cost
• Low power around 5mW for 5m range
• Unobtrusive 40KHz
• Accurate distance measurements O(1cm)
• Low latency between measurements
• Shooting for 10 measurements per second

19
Ultrasonic Ranging Latency
USND TX Start
Ultrasound Detected
Transmitter
Receiver
20
Ranging Characterization

• Lab characterization of ranging module, at 25
  pulses (temperature 21.4 Celcius)

21
Smart Beacon Calibration
22
Smart Beacon Calibration
23
Smart Beacon Calibration
24
Smart Beacon Calibration
25
Host PC Controller Software
3D GUI Client(s)

• Manager
• Packet based
• One to many switching
• Facilitate online
• processing of incoming
• data
• Allows direct use of
• MATLAB code

TCP Server
Other SW
Localizer
Calibration SW
Gateway Node
Serial I/O
26
Conclusions

• For long distances RF ToF will most probably
  prevail
• Multihop localization are promising in many
  setups
• Ultrasound measurement system works well for our
  purposes
• Use network level algorithms to improve
  positioning performance
• Do not have any comparison with relative
  localization schemes yet