A Collaborative Approach to InPlace Sensor Calibration

1
A Collaborative Approach to In-Place Sensor
Calibration

• Presented by Lee Hathcock

2
Introduction

• Sensors
• Calibration issues
• Want an algorithm to automatically calibrate

3
Assumptions

• Error types
• Systematic
• Random
• Phenomenon assumptions
• Known and limited spatial freq.
• High temporal freq.
• Sensor assumptions
• Dense deployment
• Slow sensing, no sizeable drift
• No angle-dependant sensor meas. Gains
• No hysteresis / delay in sensors

4
Algorithm (part 1)

• Collect time-series data (sync.)
• Weight potential data points
• Filter out irrelevant data points
• Fit a calibration function to the filtered data

5
Algorithm (part 2)

• Problems with different calibration paths
• Obtain CMCOF
• Enumerate calibration paths
• Specify maximum length of path
• Discard paths with low confidence threshold
• Readjust calibration function

6
Experiment

• MICA motes
• 9 motes, 3x3 grid, 5cm spacing
• Thermistors
• ground truth from thermocouple
• Hair dryer for heat source
• 80 of values within 5C
• 10 greater than 10C

7
Impressions

• Would like to see a larger scale test of this
• Also might want to try this coupled with the
  power-saving modes discussed earlier
• Earlier assumption about no delay may not be valid