Survey of Deployments Two in detail: Redwoods and ZebraNet Others –Great Duck Island –TurtleNet –James Reserve Forest –Volcanos & earthquakes –Aquatic observing systems –Localization, real-time tracking
Great Duck Island: Petrel Monitoring UCB Goal: build ecological models for breeding preferences of Leach’s Storm Petrel –Burrow (nest) occupancy during incubation –Differences in the micro-climates of active vs. inactive burrows –Environmental conditions during 7 month breeding season Inconspicuous Operation –Reduce the “observer effect” Unattended, off-the-grid operation Sensor network –26 burrow motes deployed –12 weather station motes deployed (+2 for monitoring the insides of the base station case) Burrow Occupancy Detector
TurtleNet (Corner, Umass) "Wetness" is a measure of current in the water sensor. This graph shows that the turtle came out of the water to sun itself for only brief periods and went back into the colder water. Mica2Dot hardware, GPS, Solar cells on the backs of snapping turtles.
Volcano Monitoring (Welsh, Harvard) Motes with seismic sensors deployed on active volcano in Ecuador Science dictates: high fidelity during events, large spatial separation, time synchronization. Nature of the application allows triggered data collection rather than continuous.
Both Logging & Transmission Both are good – compensate for the other’s failures –Flash running out of space but transmissions continue –Transmissions stopped but Flash retains those data points
Wildlife Tracking – ZebraNet Asplos 02 Juang et al Princeton
Biological Goal Long-term & wide ranging zebra herd migration tracking Associated with data on feeding behavior, heart- rate, body temp.
Why a Wireless Sensor Network Approach? Traditional radio collars – coarse grain information Sensor nodes (GPS), not networked – usually must retrieve collar to download stored data Satellite tracking – high energy costs, low bitrate
A Day in the Life of a Zebra Social structure can be exploited –Plains zebra form tight-knit harems (1 male, multiple females). Collar 1 individual and track the group –Sometimes form loose herds of multiple harems, often at watering holes Drink water on a daily basis Mostly moving 24 hours a day
Peer to Peer System Design zebraB 10010 11111 10001 10000 zebraA 10101 11101 10001 10000
Implications of Collar Design GPS provides precise synchronized clock –For avoiding short-range network collisions Assume 5 days battery life between recharging –Need 13.5AH to sample (6KB/day), search for peers (6hr/day), search for base station (3 hr/day), and transmitting 640KB of data. 640KB Flash = 300 days of data compressed, 110 days uncompressed –Need to accommodate redundancy of data stored from other nodes
Homing Success Rate Fraction of data successfully delivered to base station (goal to eventually get 100% data reported) Simulation study (single radio): –Flooding protocol – share data with everyone encountered –History protocol – send to “best” peer discovered based on their previous success in delivering to base –Direct protocol – not peer-to-peer, just to base
Energy in battery powered nodes. –Constrain lifetime of nodes, if not recharged –Energy harvesting, weight of solar collectors –Duty cycling necessary -> clock synchronization Data delivery –Missing data Connectivity –Routing issues –Unsynchronized duty cycles –Collisions Dead nodes –Outliers Calibration of sensors
Hierarchy, heterogeneity, mobility –Robotics, actuation Packaging –Weather effects = dead nodes –Weatherproofing – gets in the way of sensors How to deal with massive amounts of data Infrastructure –System behavior monitoring –Interactive remote control (retasking)
Breakouts Form 3 or 4 ad hoc multi-disciplinary groups (outside comfort zone: mix ECE+stat+CS+bio) Discuss one of two topics –Research question you might address with Duke Forest data –Research study you might design from scratch, its requirements and challenges. Report back at end of class (elect a spokesperson)