EnergyEfficient localization for networks of underwater drifters

1
Energy-Efficient localization for networks of
underwater drifters

• Diba Mirza
• Curt Schurgers
• Department of Electrical and Computer Engineering

2
Underwater Sensing Applications

• Goals
• Understand various physical, chemical
  biological processes
• How do they interact ?
• How are they correlated in space and time?

• Focus Collect relevant data within the natural
  dynamics of the ocean.

3
Underwater Sensing System

• System features
• Localized sensing.
• Swarm deployments.
• Networked for collaborative sensing.

1 J. Jaffe, C. Schurgers. Sensor networks of
freely drifting autonomous underwater explorers.
In Proc. of WUWNET06, Los Angeles, CA, pp.
93-96, Sept 2006.
4
Network Localization

• Need position information to
• Interpret data
• Obtain spatial map of processes

• GPS not available underwater
• Obtain timely distance estimates (TOA )
• Localization can be done using existing methods.

Due to continuous motion induced by currents,
localization is a recurring cost.
5
Trade Localization Accuracy vs. Energy
Position accuracy depends on the extent of TOA
measurements
Can we select the minimum set of links to achieve
a desired position accuracy ?
6
Problem Setup
7
Determining the optimum link-set..
Localization algorithm is run offline
Can be computationally intensive for best
performance
To obtain optimum set of links

• Need a measure of localization error.
• Use the Cramer Rao Lower Bound (achieved by ML
  estimators)

8
Optimum link-set (contd.)

• Optimization problem

Constraint nodes exceeding error
threshold lt a.N
Error in node position estimates as computed from
the Cramer Rao Bound
9
Optimum link-set (contd.)

• Solution to Optimization Problem

(b)
(a)
Find node with maximum error allowance
Remove the link that causes min increase in total
error
What are the gains?
10
Spatial Gains
Simulation Scenario
Up to 40 reduction in measurements for ?
15 when protocol overhead is included.

• 3-Hop Network
• Average no. of neighbors, ?

Protocol Overhead

• Transmit distance estimates to a central
  location.
• Communicate policies to nodes.

11
Error Tolerance Topology Change

• Error depends on relative position of nodes
• Node positions continuously changing due to
  currents.
• What is the fidelity of the link-selection scheme
  over time?

Examine Geometric Dilution of Precision (GDOP) ,G
Total variance when all references are accurate
Conclusion Error is affected only by major
changes in topology.
12
Temporal Behavior
Simulations further validate error changes with
major changes in topology.
13
Adapting to changing requirements

• Suppose target error of a group of nodes changes
  over time.
• Say, L groups, each can choose any of K different
  target errors.
• How can the link-selection scheme be adapted ?

Possible Solutions

• Re-compute the link policy
• Involves collecting range estimates from all
  nodes.
• Over head can be large.
• Pre-compute all possible policies
• Gives rise to KL different policies.
• Setting a particular policy requires global
  communication.

Is there a better way?
14
Adapting to changing requirements

• A specific condition Nodes with known positions
  restricted to a single plane (surface)

Result Error primarily depends on that of
one-hop neighbors closer to the surface.

• If target error at some level changes,
  sufficient to
• Update link-policy with 1-hop neighbors at a
  lower level .
• Communicate the required target error to 1-hop
  neighbors.

How is this better than methods suggested earlier?
15
Adaptive Link-Selection

• Advantages of adaptive link-selection scheme
• Smaller number of policies only L.K.
• Localized communication (with only 1-hop
  neighbors).

Event occurs at hop 2. New target error for hop 2.
Figure 10. Adapting link selection policy
Nodes at hop 2 adapt locally by updating the
links selected for ranging with nodes at hop 1.
16
Conclusions

• Optimal link-selection results in fewer
  measurements for localization.

• Unless major topology change do not have to
  reselect links

• Scenario on the right is achievable.

Future Investigate the scalability of the
method .
Figure drawn roughly to scale