Indoor Localization without the Pain

Indoor Localization Without the Pain
Krishna Kant Chintalapudi, Anand
Padmanabha Iyer, and Venkat
Mobicom 2010
Motivation from existing works
• Schemes requiring specialized infra: Active Badge,
Cricket, Active Bat, etc.
 requires infrastructure deployment
• Schemes requiring RF signal maps: RADAR, Place
Labs, etc.
 takes too much time; laborious!
• RF propagation model based (e.g., TIX, ARIADNE)
 much less efforts than RF map; but still need a
lot of work to fit the models
Motivation from existing works
• Ad hoc localization (multi-hop wireless comm from known
wireless anchor nodes)
 requires enough node density to enable multi-hopping
• Robot navigation (called Simultaneous Localization and
Mapping, SLAM where a robot builds a map and determine
its location)
 requires special sensors (e.g., ultra-sound, LADAR, etc.)
• Indoor navigation (similar to robot navigation) – mostly
using compass (or gyroscope) + accelerometer
 requires an indoor map for accurate localization (+
accelerometer and compass error must be accommodated)
• Question? Can we do indoor localization
without such limitations??
Key Idea: Localizable?
• Localizable (globally rigid)
– Can determine unique positions (w/o distorting measured distance);
possible to rotate/flip, though
• Idea: if enough distance measurements are available, we can
“localize” devices/APs; orientation can be opportunistically fixed
using external input (e.g., GPS feed when entering a building)
Not localizable (rigid)
• RF propagation model between i and j
– pij = Pi – 10*ϒi log dij + R
• pij: recv signal strength, dij: distance, Pi: tx power, ϒi: path loss
– dij = 10 ^(Pi-pij)/10 ϒi
• Equations:
– Unknowns: m APs + n locations
• 4 unknown variables for each AP: location (x, y), Pi, ϒi
• 2 unknown variables fro each loc: (x, y)
• Need to solve a set of “over-determined” equations
– N: number of equations
– Min JEZ
• Optimization
– Gradient methods: tend to stuck at local min
– Genetic methods: OK performance
• Hybrid approach: Genetic + Gradient Decent..
– Genetic: mutation based..
• How to reduce state space? Floor size, ….
• Receiver gain problem:
Relative difference

similar documents