Time Synchronization RBS, Elson et al.

1
Time Synchronization (RBS, Elson et al.)

• Presenter Peter Sibley

2
Traditional Synchronization Methods

• Server sends messages to client, containing
  servers current time.
• Common extension
• Client requests time from server
• Server sends current time.
• Client estimates one-way latency from the
  round-trip time.

3
NTP

• (1-50ms) accuracy, most common time protocol.
• Uses hierarchy attached to a external clock.
• At the LAN level, workstations may use
  information from peers .

Reference Clock GPS ,Atomic Clock
Stratum 1
Stratum 2

Stratum 15
See http//www.eecis.udel.edu/mills/ntp.html
4
Sources of Error

• Send Time
• Constructing message
• Variable OS delays in moving message to the
  interface
• Access Time
• Waiting to transmit message. (depends on MAC)
• Propagation Time
• To time get to receivers interface
• Receive Time
• Time for interface to generate a message
  reception signal

5
Observations (Elson et al.)

• Try to remove send/access time errors.
• Synchronize among receivers.
• Relative time is more important.
• Latency is less of an issue, determinism is what
  matters.

6
Example Phase Est.

• Node i at (0,0) is triggered at t4.
• Node j at (0,10) is triggered at t5.
• The moving object has velocity (0,10).
• Notice, no reference to a global time scale.

7
Estimation of Phase

• A transmitter sends m reference packets
• Each of the n receivers records the arrival times
  according to their local clock
• The receivers exchange their observations
• Receiver i computes phase offset to another other
  receiver j as average offsets.

8
Phase-Estimation Simulation Results
9
Estimation of Clock Skew

• Each devices crystal oscillator, has slightly
  different frequency.
• Frequency of each oscillator varies over time.
• Use Least-Squares fit, instead of averaging phase
  offsets.
• Assumes phase error changes at a constant rate

10
Implementations

• Mote
• Tested 5 motes, with periodic reference pulse.
• 2 micro-sec resolution clock
• Ipaq running linux 2.4, 802.11 wireless
• Userspace Unix daemon.
• Use UDP.

11
Results (Mote)
12
Results
13
Multi-hop extension (example)
14
Multi-hop algorithm
15
Performance of multihop extension
16
Information Driven Dynamic Sensor Collaboration
for Tracking Applications, Zhao et al.

• Presenter Peter Sibley

17
Scenario
18
Collaborative Tracking.
19
Sequential Bayesian Estimation

• Problem Picking the next sensor, should be local
  choice.
• Need to Pick the neighbor sensor that will
  improve the estimation the most.
• Rephrase as an optimization problem,
• Objective is Mixture of Information Gain and Cost

20
Utility/Cost.

• Different Utility functions can be used
• Mahalanobis Distance
• Entropy Based
• Estimated Likelihoods
• (Depends on distributional assumptions)
• Costs
• Euclidean and weighted Euclidean distance from
  the leader node.

21
Tracking Results
22
Tracking Results