a*robust,*decentralized*approach* to*rf9based* locaon...
TRANSCRIPT
MoteTrack A robust, decentralized approach
to RF-‐based loca:on tracking
Team Members : Amit Jain
Harpreet Singh David Thole Tengyu Wang
What is MoteTrack? • Robust, decentralized approach to RF-‐based loca6on tracking • Based on low-‐power transceiver coupled with modest amount of computa6on and storage. • Doesn’t rely upon back-‐end server or network infrastructure. • This design allows system to func6on despite of significant failures of the radio beacon infrastructure • The deployment of MoteTrack,
– consisted of 23 beacon nodes – distributed across our Computer Science building – Achieved loca6on accuracy of 0.9 and 1.6m
• MoteTrack tolerates failure of up to 60% of beacon nodes without severely degrading accuracy • MoteTrack’s performance is analyzed over varies condi6ons including
– Variance in number of obstruc6ons – Beacon node failure – Radio signature perturba6ons – Receiver sensi6vity – Beacon node density
Introduc:on • Radio signal informa:on can used to determine loca:on of a
roaming node with close meter-‐level accuracy. • RF based loca:on tracking system can have wide range of
applica:ons e.g.: firefighters • RF based loca:on tracking system are of great importance in
applica:on which demands robustness of the loca:on tracking infrastructure.
• MoteTrack needs prior calibra:on before it can be used thus makes it an open issue for research
• Exis:ng approaches to RF-‐based loca:on tracking are – Centralized – BriOle
Why makes MoteTrack different? • Robust, decentralized approach to RF-‐based localiza:on • Uses a network of baOery-‐operated wireless nodes to measure, store, and
compute loca:on informa:on. • Loca:on tracking is based on empirical measurements of radio signals
from mul:ple transmiOers, an algorithm similar to RADAR . • To achieve robustness, MoteTrack extends this approach in three
significant ways: – Uses decentralized approach to compute loca:ons that run on programmable beacon nodes, rather than back-‐end
server – Loca:on signature database is replicated across nodes to minimize per-‐node storage overhead thus achieving high
robustness to failure – Dynamic radio signature distance metric
Background and Related work • A number of indoor loca:on tracking systems have been proposed in the literature, based on RF signals,
ultrasound, infrared, or some combina:on of modali:es. • Given a model of radio signal propaga:on in a building or other environment, received signal strength can
be used to es:mate the distance from a transmiOer to a receiver, and thereby triangulate the posi:on of a mobile node.
• Above approach drawbacks:
-‐requires detailed models of RF propaga:on and does not account for varia:ons in receiver sensi:vity and orienta:on.
• MoteTrack’s basic loca:on es:ma:on uses a signature based approach that is largely similar to RADAR • Goal is not to improve upon the accuracy of the basic signature-‐based localiza:on scheme, but rather to
improve the robustness of the system through a decentralized approach
MoteTrack’s Goal and Challenges Robustness??? • Robustness with respect to loca:on tracking. • One form of robustness, then, is graceful degrada:on in loca:on
accuracy as base sta:ons fail (say, due to fire, electrical outage, or other causes)
• Another form of robustness is resiliency to informa:on loss. • A third type of robustness has to do with perturba:ons in RF signals
between the :me that the signature database was collected and the :me that the mobile node is using this informa:on to es:mate loca:on.
• Final type of robustness has to do with the loca:on es:ma:on computa:on.
Challenges • The collec:on of RF signatures and loca:on calcula:on
must be resilient to loss of informa:on and signal perturba:on. Thus requires distance metric.
• Decentralizing the loca:on tracking system. – Allow base sta:on nodes to perform loca:on es:ma:on buO here can arise problems
• An alterna:ve is to allow the mobile device to perform loca:on es:ma:on directly.
• Simplest form, the en:re RF signature database could be stored on the mobile node.
• In cases where a mobile user only carries a small RF beacon or listener (e.g., embedded into a firefighter’s equipment), this may not be feasible.
MoteTrack Overview
q Beacon Node: Berkeley Mica2 sensor
q Reference Signature: A signature combined with a known three Dimensional loca:on(x,y,z). -‐-‐-‐-‐Offline loca:on es:ma:on.
q Received Signature: Aggregates beacon messages received over some :me period into a signature.-‐-‐-‐Online loca:on es:ma:on.
q Signature Form:{sourceID, powerLevel, meanRSSI}
MoteTrack Overview
MoteTrack Overview
Loca6on es6ma6on (Centralized approach)
∑∈
−=Tt
sr tmeanRSSItmeanRSSIsrM |)()(|),(
r:------ reference signature s:------ received signature T ------ set of signature tuples in both t: ------ a tuple in T
q Centroid of reference signatures: • K nearest reference signatures
• All r with ra:o: , r* is the nearest reference signature.
csrMsrM
<)*,(),(
q ManhaWan distance
Making RF-‐based Localiza6on Robust
No single points of failure
handle incomplete data and fail nodes
Decentralized loca:on es:ma:on
Adap:ve signature distance metric
Making RF-‐based Localiza6on Robust
Decentralized loca6on es6ma6on protocol:
q K beacon Node send their reference slice.
q K beacon nodes send their loca:on es:mate
q Max-‐RSSI beacon node sends its loca:on es:mate
Distribute reference signature database to beacon nodes
q Greedy distribu:on algorithm
q Balanced distribu:on algorithm
Making RF-‐based Localiza6on Robust
∑∑−∈−∈
++=)()(
)()(),(),(srt
rrst
snalbidirectio tmeanRSSItmeanRSSIsrMsrM ββ
∑−∈
+=)(
)(),(),(rst
sonalunidirecti tmeanRSSIsrMsrM β
Adap6ve signature distance metric
q No beacon Node failure:
q A large number of beacon Node failure:
Making RF-‐based Localiza6on Robust
Adap6ve Scheme: Beacon Nodes periodically measure the neighborhood, defined as set of other beacon nodes they can hear.
If the intersec:on between the current and original neighborhood is large, use bidirec:onal distance Metric.
If failed nodes exceeds some threshold, use unidirec:onal distance
Equipment
• TinyOS Plaaorm (hOp://www.:nyos.net) • WriOen in C (NesC -‐ hOp://nescc.sourceforge.net)
– 3000 lines • Example Sensor:
TinyOS
• “Event Based” • Minimal linux distro with scheduling capabili:es
• Power management op:mized • WriOen in nesC
NesC
• Component behavior defined in interfaces • Programs built out of components – “wired” together
• Similar to the Android programming – Android split into Ac:vity, Service, ContentProvider, BroadcastReceiver
• Ac:vi:es receive Intents, produce Results – Ac:vi:es “wired” together.
Space Used
• 1742 m^2 – Computer Science Building – first floor – 1330 m^2 of in room space – 412 m^2 of hallway space – 23 beacons
Key: Blue dots = Fixed loca:ons Red Squares = Acquired Signature Locs
Tes:ng Involved
• Environmental Changes – Doors opened/closed – Time of Day
• Algorithm Changes – Number of neighbors (KNN-‐like) – Greedy/balanced Algorithm types
• Diversifying the signals
MoteTrack Accuracy
Density and Performance
K – Reference Signatures
Effects of Time of Day
• Would :me of day have an impact? • Why?
Effects of Time of Day Cont…
• Very liOle effect. • But…
– Does this check vastly different :mes of day. E.g. 10AM vs 2AM
Use of mul:ple signal frequencies
• First a ques:on…given what we know about how this works, why would we want to use mul:ple signal frequencies?
Future Work
q Offline Pre-‐installa:on and calibra:on of beacon nodes is required in circumstances like in mul:-‐car highway accident, which is not feasible.
Ø AD HOC mechanism is used in these type of
cases.
q One approach is to use GPS to automa:cally populate the signature database.
For example use of PDA by medics in medical sciences. q Greedy distribu:on technique is used for popula:ng the reference signatures database.
q Loca:on tracking accuracy increases as more and more reference signatures are acquired. q GPS loca:on es:ma:on related errors can be handled using GPS devices using WAAS(Wide Area Augmenta:on System).
Conclusion
q Basic RF approach of localiza:on is extended to: Ø A new highly ROBUST approach. Ø A DECENTRALIZED approach.
q Decentralized loca:on es:ma:on protocol relies only on local data, local communica:on and opera:onal nodes.
MoteTrack System
Implementa:on of approach
Deployment of approach
Evalua:on of approach
q MoteTrack is based on: Ø Berkeley Mica2 Ø MicaZ Ø TmoteSky
q Why MoteTrack ?
Ø Small in Size Ø Inexpensive device Ø Easily embedded in environments like walls etc. Ø High loca:on tracking accuracy Ø Can bear node failures & signal perturba:ons
without any errors.
Ques:ons?