Lessons from a Sensor Network Expedition

1
Lessons from a Sensor Network Expedition

• Authors R. Szewczyk, J. Polastre, A.
  Mainwaring, D. Culler
• Presented By Steve Wilson

2
Introduction

• We want to deploy an actual sensor network into
  environment.
• Use the real life application to gain insight
  into things analysis and simulation cannot tell
  us.
• Use the sensor data and network metrics to gain a
  deeper understanding of WSNs.
• Habitat monitoring is a widely accepted
  application.

3
Application Design

• Deployed 43 nodes network to monitor nesting
  habits of Leachs Storm Petrel.
• Used Mica mote running TinyOS.
• Built integrated sensor board which could fit
  into burrowing tunnels.
• The failure of one sensor likely affects other
  sensors.

4
Application Design

• The sensors included
• Photoresistive light sensor
• Digital temperature sensor
• Capacitive Humidity sensor
• Digital barometric pressure sensor
• Passive infrared detector (thermopile
  thermistor)

5
Packing up the Motes

• Motes need to live through diverse weather
  conditions.
• Motes were sealed with parylene, necessarily
  leaving the sensors exposed.
• Above ground motes were placed in acrylic
  enclosures.
• In burrows, motes were without enclosures.

6
Example Mote
7
Network Architecture

• One hop sensor patches connected to a gateway.
• Gateway transmits through local transit network
  to base stations which provide database services.
• The final level consisted of remote servers to
  support analysis and visualization of the
  collected data.

8
Goals of Experiment

• Evaluate the following aspects
• The sealant used on the sensors.
• The radio performance in and out of burrows.
• The usefulness of the collected data for
  biologists.
• The system and network longevity.

9
Network Analysis

• Look at the patterns of packet loss.
• Initial loss rate high but improves over time.
• Look at patterns of loss within virtual time
  slices.
• Figure shows patterns of horizontal black lines,
  indicating loss over all nodes.

10
(No Transcript)
11
Network Dynamics

• Network utilization is low but collisions appear
  to be playing a role in nodes 45 and 49.
• The CSMA MAC layer should allow each node to back
  off until the channel is clear.
• Nodes work on a 70 second cycle.
• Clock drift and channel variations could cause
  slot reallocation, but infrequently.

12
Network dynamics

• We can look at the phase of each node relative to
  the 70 second sampling period.
• Ideally each mote should hold the same time slot
  but MAC delays and clock drift cause a change in
  phase.
• Could be caused by detecting radio noise and
  interpreting it as a packet.

13
(No Transcript)
14
Analysis of Sensors

• Light sensor provided a good base for
  establishing the proper functioning of sensor
  board.
• They had experience with the sensor and knew what
  to expect.
• Most reliable of the sensors.
• Its failure was highly correlated with failure of
  the other sensors.

15
Temperature Sensor

• Sensors resolution was 2 degrees Celsius.
• IR radiation from direct sunlight heated up
  enclosure and cause significantly higher
  readings.
• Sensor failed when contact with water occurred.
• Failure of temperature sensor was highly
  correlated with failure of humidity sensor.

16
Humidity Sensor

• Sensor was completely exposed.
• Protection of enclosure was inadequate.
• When the sensor got wet it created a
  low-resistance path between power terminals.
• Caused really high or low readings.
• Once the sensor dried up the high reading sensors
  recovered while the low ones crashed.
• Unknown why the low ones crashed.

17
(No Transcript)
18
Infrared Sensor

• Difficult to analyze.
• Thermistor readings track closely with the
  temperature but on highly different scales.
• The absolute readings were well out of range.
• Without an IR camera readings are basically
  useless.

19
Power Analysis

• Battery voltage was used as indirect measure of
  power consumption.
• Complicated by battery replacement, failed
  voltage indicators, failed sensors, and gaps in
  the data due to database crashes.
• Since only 5 nodes exhausted battery, analysis of
  just one was done.
• Indicated poor utilization, and that a different
  battery with constant voltage should be used.

20
Indicators of Failure

• Identified erroneous sensor readings and
  application phase skew.
• Humidity sensor was main indicator of a nodes
  health.
• Those with really low readings usually failed
  within two days.
• Nodes with slower clocks also showed high
  probability of failure within two days.
• These indicators showed very few false positives.

21
Conclusions

• One of the first outdoor deployments of WSNs has
  been analyzed.
• Presented features that correlate highly with
  node failure.
• Shown that great care must be taken in the
  designing of the WSN.
• The insights gained will aid in development in
  self-organizing and self-healing WSNs.

22
References

• http//www.greatduckisland.net
• The extensible sensing system http//www.cens.ucl
  a.edu