Transcript Fuzzy Data Collection in Sensor Networks

Fuzzy Data Collection
in Sensor Networks
Lee Cranford
Marguerite Doman
July 27, 2006
Overview







Overview of Sensor Networks
Sensor Network Applications
Research Objective
Prototype Platform
Proposals
Our Modifications
Ongoing Work
Wireless Sensor Networks


A collection of small hardware devices that
collect data from their environment
Research challenges



Energy efficiency
Data collection
Communications
overhead
Wireless Sensor Networks

Common application:
Environmental monitoring

Example: Controlled
prairie burning


Sensors can report
major temperature
changes
The spread of a fire
can be monitored
Research Objective


Long-term: Use fuzzy query and
database management approach to
data collection
This summer: Modify the operating
system of prototype sensor motes to
support an approximate (“fuzzy”)
attribute
(value ± margin)
Prototype Hardware: MICA
Motes


Prototype sensors developed by UC Berkeley to
support sensor networking research
Sensors: Light, temperature, barometric pressure,
seismic, sound, magnetic, GPS, and others

RF Communications

TinyOS: “Lite” embedded OS

TinyDB: “Lite” DBMS
Prototype Platform: TinyDB


TinyDB is a sensor network data collection system
Allows for polling of sensors through Structured
Query Language (SQL)



The sensor network is therefore abstracted to resemble a
relational database in its interface to the user
TinyDB's SQL dialect is in a very stripped-down,
“working proof of concept” form called TinySQL
Benefits: Ease of use, eliminates the “API
approach” sensor polling, can poll the whole
network easily
Problem




In the prairie fire scenario, we want to know where
dramatic rises and falls in temperature occur
TinySQL supports polling of
a mote's temperature
and network averaging
However, it relies on
central processing to identify local
trends
The result is unnecessary transmission of data from areas
not undergoing a change
Proposals

What if we could tell the network to only
return results that were outside of
ordinary trends?

Push data processing to the mote

Develop local “threshold” values based on
long-term node measurements

Extend TinySQL to support fuzzy queries

This allows us to ask the network, “Where is it
hotter than usual? Where is it cooler?”
Development and Simulation


Installed and customized TinyOS on a
Linux platform
Installed and evaluated six simulators

Selected PowerTOSSIM
 Set up a simulation environment to
evaluate the energy efficiency of queries

Designed TinyFSQL’s syntax and
methods of operation
Code Modifications

Operating system additions

Utilized data storage at the mote level

Implemented mote routines to return data
only if present values are outside the
current range
Code Modifications

Extended TinyDB
 Added an attribute to TinySQL to
interface with local mote trends
 Implemented
the “UPDATE” keyword
to force changes of local averages
 Added
the “fuzzy equal” operator
Work in Progress



Completion of TinyFSQL
 Add a greater range of fuzzy operands to
the TinyDB parser generator source
Modification to the Java GUI to include userfriendly selection of fuzzy attributes
Extensive tests using the PowerTOSSIM
simulator

Compare the energy efficiency of TinyFSQL to
TinySQL