Innetwork Surface Simplification for Sensor Fields

1
In-network Surface Simplification for Sensor
Fields

• Brian Harrington and Yan Huang
• University of North Texas
• brh,huangyan_at_cs.unt.edu

2
In-network Surface Simplification for Sensor
Fields
Self forming wireless network
gateway
Backbone network
Local Monitoring
detection
detection
cameras
Monitor human and structural health Measure
environment variables Track inventories Detect
ground vibrations Identify toxic chemical spills.
Satellites
Remote monitoring
3


In-network Surface Simplification for Sensor
Fields
Example Ecology Sensor Networks
Issues in Sensor Networks

• Interoperability
• Standardization efforts
• Heterogeneity
• Video cameras connected to wide band network
• High ends nodes, e.g. Intel XScale
• Motes with small storage and low processing
  ability
• Scalability
• Large scale deployment of small motes

• that take the Earths pulse
• National Ecology Observatory Network (NEON)
• RiverNet
• EarthScope
• GEOSS

4
In-network Surface Simplification for Sensor
Fields
Sensor Mote Peculiarities

• Small storage, low processing ability
• A typical sensor is equipped with a processor of
  a few MHz and a few kilobytes of RAM.
• transmitting 1 bit of data to a distance of 10
  meters consumes as much power as 220 to 2,900
  instructions
• Battery powered, short wireless communication
  range
• A Berkeley Mica Mote operates on two AA batteries
• communication range between a few to a few
  hundred feet depending on transmission power and
  environmental conditions.
• Lasting for a few days in full duty cycle mode,
  months to years if energy is budgeted

CrossBow MICA2/DOT Professional Kit (MOTE-KIT
5x4x)
5


In-network Surface Simplification for Sensor
Fields
Sensor Databases

• Sensors form a fine grained distributed database
• Use declarative language, e.g. to interact with
  sensor database
• SELECT attributes, aggregates
• FROM Sensordata S, EnvironmentalData E,
  HistoricalSensorData H
• WHERE predicate
• GROUP BY attributes
• HAVING predicate
• DURATION time interval
• EVERY time span e

6


In-network Surface Simplification for Sensor
Fields
Related Work

• Sensor database prototypes (YaoCIDR03,
  MaddenSIGMOD03)
• In-network aggregation (SharifzadehGIS04,
  KrishnamachariICDCS02 , FangTRP02,
  SamuelOSDI02 , VuranCN04, ConsidineICDE04 )
• Surface Simplification (HeckbertTRP97)

7
In-network Surface Simplification for Sensor
Fields
Field Model
In-network Surface Simplification

• Many phenomena in natural science are continuous
  and thus best represented as fields
• temperature, precipitation, hydraulic head, soil
  moisture, and ocean current velocity,

• Requiring all sensors to send back readings are
  too expensive
• Flat surfaces need less readings to represent
• Reduce communication cost
• By reducing the number of sensors to report
• Rationale
• Use simple in-network calculation to save more
  expensive messaging cost

8
In-network Surface Simplification for Sensor
Fields
Surface Simplification Example
Dots represent sensors. Dot in (0,0) may be the
gateway sensor with long-haul communication
capacity
9
In-network Surface Simplification for Sensor
Fields
Proposed Approach
Problem Definition

• A hierarchical quad tree based simplification
  algorithm
• A triangulation based decimation algorithm

• Given
• A set of randomly deployed resource and
  communication constrained sensors S in a physical
  field.
• Find
• Algorithms to select a subset SD of all the
  sensors in S to report to the central site so
  that the central site can reconstruct the
  surface using SD
• Objective
• Reducing the message cost
• Bounding the error.

10
In-network Surface Simplification for Sensor
Fields
Hierarchical Approach
Actual Surface

• Parents send average value for its children with
  homogeneous readings
• Readings far off from average are sent
  individually
• An incremental top-down refinement process during
  reconstruction
• using increasingly finer levels of detail sent by
  selected sensors
• Guarantees the reading received by the central
  site is within e of the real sensor readings

Reconstructed Surface
Level 0
Level 1
Level 2
Level 3
11
In-network Surface Simplification for Sensor
Fields
Analysis on Energy Consumption

• Hierarchical approach is useful if the following
  inequality holds
• P N F L N lt L N F lt 1 - P/L
• P average number of hops from a sensor to its
  parent
• N total number of sensors
• F fractional of sensors that need to report
  individually
• L average number of hops to the query
  origination
• If P is 10 and L 1000, then for F lt 99 we save!
  P must be less than L for this technique to be
  beneficial.

12
In-network Surface Simplification for Sensor
Fields
Decimation Approach

• A localized Vonoroi cell construction is used to
  create the initial triangulation
• The initial surface is incrementally refined
• We propose a probabilistic approach to select
  sensors not to report
• Concurrent error calculation and deletion by all
  sensors may result in error accumulation
• No guarantees the reading received by the central
  site is within e of the real sensor readings

13
In-network Surface Simplification for Sensor
Fields
Localized Vonoroi Cell Calculation

• ps voronoi cell convex polygon that contains
  all of the points that are closer to p than any
  other sensor
• Theorem all sensors which may clip the initial
  voronoi cell must be in c(p)
• We propose an acquisitional approach
• Build a broadcasting tree rooted with radius c(p)
  routed at p
• Collect information from the tree and refine the
  voronoi cell

14
In-network Surface Simplification for Sensor
Fields
A Probabilistic Node Deletion Scheme

• Error Estimation
• Error Accumulation

• Propose a probabilistic node deletion scheme
• p(s_i) min(e_i/ e,1)
• where e is the error threshold

15
In-network Surface Simplification for Sensor
Fields
Analysis on Energy Consumption

• To outperform the naïve algorithm, the following
  in-equation must hold
• L N gt N nh L F x N F lt
  1-1/L
• L average number of hops to the query
  origination
• N number of sensors
• nh average number of hops to reach a neighbor
  (typically nh1)
• F fractional of sensors that need to report
• For L 100, if F lt 99, we will save!
• L is approximately sqrt(N)
• For L 10, if F lt 90, we will save!

16
In-network Surface Simplification for Sensor
Fields
Experiment Setup and Results

• University of Delaware global surface monthly
  grids (http//www.jisao.washington.edu/data_sets/w
  illmott)
• Temperature and precipitation readings for 85794
  points once a month for 50 years from 1950
  through 1999
• Randomly selected 2 - 10 data
• Pretty sparse data
• Three approaches
• Naive algorithm of having all sensors report
  individually
• Hierarchical approach
• Decimation approach
• Results
• Messaging saving up-to 4 times (denser -gt more
  saving)
• Decimation method less than 4 above error
  thresholds

17
In-network Surface Simplification for Sensor
Fields
of messages w.r.t. density
Density increases savings increase
18
In-network Surface Simplification for Sensor
Fields
of messages w.r.t. e
Error thresholds increase savings increase

19
In-network Surface Simplification for Sensor
Fields
of points outside e w.r.t. density
Decimation has low error rate of less than 4 for
density between 2 and 10
20
In-network Surface Simplification for Sensor
Fields
of points outside threshold w.r.t. e
Decimation has low error rate of less than 4 for
density between 2 and 10
21
In-network Surface Simplification for Sensor
Fields
Conclusion

• In-network surface simplification is useful
• Proposed two approaches
• Hierarchical approach
• Decimation approach
• Results
• The proposed two approaches have significant
  messaging saving
• Messaging saving up-to 4 times (denser -gt more
  saving)
• Hierarchical approach has error bound
• Decimation method less than 4 above error
  thresholds

22
In-network Surface Simplification for Sensor
Fields
Future Work

• Incorporating temporal auto-correlations into our
  model
• Systematically investigate other surface
  simplification approaches
• Grid based, feature based, refinement approaches,
  and hybrid approach (decimation and refinement)
• Implement a prototype system on tinyOS/tinyDB
• Working with domain scientists
• Fault tolerance

23
Sensor Database and Data Mining Sensor Network
as a Field
References

• YaoCIDR03 Y. Yao and J. E. Gehrke. Query
  Processing in Sensor Networks. In Proceedings of
  the First Biennial Conference on Innovative Data
  Systems Research (CIDR), 2003.
• SharifzadehGIS04 M. Sharifzadeh and C. Shahabi.
  Supporting spatial aggregation in sensor network
  databases. In GIS 04 Proceedings of the 12th
  annual ACM international Syposium on Geographic
  information systems, 2004.
• HeckbertTRP97 P. S. Heckbert and M. Garland.
  Survey of polygonal surface simplification
  algorithms. Technical report,1997.
• MaddenSIGMOD03 Samuel R. Madden, Michael J.
  Franklin, Joseph M. Hellerstein, and Wei Hong.
  Design of an acquisitional query processor for
  sensor networks. In SIGMOD, 2003.
• HeckbertTRP97 Paul S. Heckbert and Michael
  Garland. Survey of polygonal surface
  simplification algorithms. Technical report, 1997.

24
Sensor Database and Data Mining Sensor Network
as a Field
References

• KrishnamachariICDCS02 Bhaskar Krishnamachari,
  Deborah Estrin, and Stephen B. Wicker. The impact
  of data aggregation in wireless sensor networks.
  In Proceedings of the 22nd International
  Conference on Distributed Computing Systems,
  pages 575578, 2002.
• FangTRP02 Q. Fang, F. Zhao, and L. Guibas.
  Counting targets Building and managing
  aggregates in wireless sensor networks. Technical
  Report P2002-10298, Palo Alto Research Center,
  2002.
• SamuelOSDI02 Samuel R. Madden, Michael J.
  Franklin, Joseph M. Hellerstein, and Wei Hong.
  Tag a tiny aggregation service for ad-hoc sensor
  networks, 2002. OSDI.
• VuranCN04 Mehmet C. Vuran, B. Akan, and Ian F.
  Akyildiz. Spatio-temporal correlation theory and
  applications for wireless sensor networks.
  Comput. Networks, 45(3)245259, 2004.
• ConsidineICDE04 Jerey Considine, Feifei Li,
  George Kollios, and John Byers. Approximate
  aggregation techniques for sensor databases. In
  Proceedings of the 20th International Conference
  on Data Engineering, 2004.