WaveScope

1
WaveScope An Adaptive Wireless Sensor Network
System for High Data-Rate Applications
Students Staff (MIT) Kyle Jamieson Stanislav
Rost Arvind Thiagarajan Mei Yuan

• PIs
• Hari Balakrishan (MIT)
• Sam Madden (MIT)
• Kevin Amaratunga (Metis Design)

NSF NETS/NOSS Informational Meeting
10/18/05 http//wavescope.csail.mit.edu
2
Outline

• Trends, requirements, architecture
• The Wavescope System
• Broadcast state aware networking
• Wavescope QP Declarative queries with
• Signal-oriented operations
• Statistical models

3
Yesterdays WSN Monitoring Applications

• Periodic monitoring
• repeat
• wake up and sense
• transmit data
• sleep for minutes
• Event-based monitoring
• Transmit data on external event
• Low data rates duty cycles

4
Next-generation WSN Apps High-Rate
Low-Latency

• High sensing rates O(102 105) Hz
• Non-trivial analysis of gathered data
• Correlations, aggregates, signal processing
• Closed-loop control
• Many domains
• Industrial monitoring, civil infrastructure,
  medical diagnosis, automotive,

5
Example Industrial Monitoring
Aka condition-based monitoring

• Preventive maintenance of fabrication plant
  equipment (Intel)
• Done manually today, offline processing
• Sense vibration (acceleration)
• 100 machines, gt10 observation points per machine
• 10-40 kHz frequency band
• Aggregate data rate about 10 100 Mbps
• Real time monitoring -gt in-net. signal processing
• E.g., freq. xform to capture relevant freq. bands

6
Three Testbeds

• Automotive monitoring (CarTel)
• Vibration, microphone signals
• Small scale, in-lab deployment with microphones
• 10 cars by 2006
• http//cartel.csail.mit.edu
• Pipeline Monitoring (Ivan Stoianov)
• Airplane wing monitoring (Metis Design)
• Vibration signatures for structural weakness

7
Pipeline Monitoring
Source Ivan Stoianov
8
WaveScope Research Thrust
General-purpose, reusable, end-to-end
systeminfrastructure for monitoring and control
in high-rate, low-latency WSNs

• Network architecture
• Congestion management quality aware routing
• Broadcast-based architecture
• Generalized state management
• Information processing
• In-the-net processing operators
• Data fusion, probabilistic models, signal
  processing

9
WaveScope Architecture
10
Broadcast-based Architecture

• With wires, links are shielded from one another
• Sharing starts only at network layer
• Wireless networks have no such shielding
• Radios are not wires!
• Unnatural and inefficient to think in terms of
  links
• Need a new abstraction that embraces broadcast
• Many new techniques frame combining,
  opportunistic routing, multi-radio diversity,
  network coding, etc.
• Open question Can we build a broadcast-based
  wireless network architecture?

11
In-the-net processing State semantics

• Internet architecture soft state, fate sharing
• Does not accommodate in-the-net processing
• Open question What are the right principles for
  dealing with state upon failure, churn, topology
  reconfiguration, etc?
• Example In-network database computing aggregate
  over last ten minutes of data from several
  sensors.

12
WaveScope Architecture
13
Information Processing in WSNs

• TinyDB Sensornets meets relational databases
• Streaming data aggregation, filtering, joins
• WaveScope QP
• High-rate, signal-oriented data processing
• Statistical models and inference
• To deal with noisy and missing data

14
WaveScope QP Challenges

• Support high rate sensing (gt a few Hz)
• Provide signal oriented operations
• Information intelligence (models)
• Detect failures outliers
• Detect correlations
• Predict missing values

15
Goal 1 Generalizing to Signals

• Want signal level processing
• Maintain generality, application-independence
• Include e.g., wavelet, time-series operators
• Workflow style programming
• Connect up processing operators
• Specify high-level sampling rate
• Specify energy/lifetime constraints
• Specify signal-level filters

16
Goal 2 Statistical Models

• Idea Build a model of the data, use to answer
  queries
• Sensor readings update the model as needed
• Example models probability distribution
• Benefits
• Transmit less data
• Report correlations, detect anomalies
• Smart interpolation for missing data
• Answer complex probabilistic queries

17
Interface Challenge

• How do users pose queries?
• Query language
• Boxes and arrows
• How do users specify rates and priorities?
• How do users select and specify models?

18
Status and Wrap-up

• High-rate and low-latency will be a defining
  feature of next-generation WSNs
• Requires signal oriented thinking
• Techniques to model data, detect outliers,
  predict missing values
• In-network intelligence
• Current status
• Several signal-oriented testbeds
• Audio, automotive, pipelines
• Converging on common set of SP primitives
• Broadcast-based, state-aware networking
• See poster