Distributed Data Fusion in Sensor Networks

1
Distributed Data Fusion in Sensor Networks

• PAMI Research Group
• ECE Department
• Bahador Khaleghi

2
Outline

• Sensor Networks
• Characteristics
• Applications
• Challenges
• Distributed Data Fusion
• Distributed Kalman Filtering
• Whats Next
• References

3
Sensor Networks

• Definition
• A network of large number of sensing,
  computation, and communication enabled devices
  performing distributed data gathering
  collaboratively
• Originally developed for military applications
• Multi-disciplinary field
• Wireless communication, computer networks, MEMS,
  system and control, computer science
• Sensor node (mote) components
• Sensing, computing, communication, and energy
  source units

EPIC Mote (UC Berkeley)
Mote architecture
4
Sample Architecture
5
Requirements
WSN
Size, cost, computational power, bandwidth, and
energy constrained
Prone to failure (e.g. obstruction, loss of motes)
Distributed preferably self-organized
Large number densely deployed motes
Wireless communication media
6
WSN Applications

• Distributed sensing and monitoring
• Military (reconnaissance and detection)
• Environment (fire/flood detection, bio-complexity
  mapping)
• Industry and business (process control and
  inventory management)
• Civilian (home automation)
• Space exploration
• Target tracking
• Military (surveillance, targeting)
• Public (traffic control)
• Healthcare and rescue (tracking elderly, drug
  administration)
• Business (human tracking)

7
PermaSense Project

• Long-lived deployment of WSN in environmental
  monitoring (since 2006)
• Goals
• Develop a set of wireless measurement units for
  use in remote areas with harsh environmental
  monitoring conditions
• Gathering of environmental data that helps to
  understand the processes that connect climate
  change and rock fall in permafrost area
• Specs
• Two field sites in Swiss Alps
• 25 sensor nodes
• Ultra low power (148 uA)
• Ruggedized for durability (3 years unattended
  lifetime)
• Modular architecture (4 tiers)

8
WSN Challenges

• Communication network
• Architecture and protocol stack (mostly network
  and DL layer)
• Topology
• Positioning of the sensors (could be random)
• Homogeneous vs. Heterogeneous
• Dynamic or static
• Clustering
• Sensor Management
• Efficient resource allocation
• Security (DOS attack and sink/black/worm/jamming
  holes)
• Fault tolerance (wrt link or node failure)
• Hardware platform design
• Realize low cost and tiny sensor nodes using MEMS
  and NEMS technologies
• Evaluation framework
• Measure performance quantitatively (accuracy,
  latency, scalability, stability, fault tolerance)
• Sensing and Data Fusion
• How to fuse data from many sensors using local
  communication

9
Distributed Data Fusion

• Solve detection and estimation problems using
• Centralized algorithms data is relayed to a
  central sink
• Issues data congestion, scalability, reliability
• Distributed algorithms data is used to compute
  local estimates forwarded to nearby nodes
  receiving nodes fuse data and update local
  estimates
• DDF design objectives
• Scalability deployable in large networks
• Efficiency (limited resources) less
  transmissions and computing
• Robustness and reliability no centralized weak
  spot, handle network imprecations (e.g. delayed
  information)
• Autonomy (self-adaptability)

10
Early Work

• Uncorrelated errors across quantities to be
  fused
• Time-invariant states
• Linear system dynamics
• Linear sensor models

Rao et al. 1991 11 fully decentralized Kalman
filtering assuming perfect instantaneous
communication among all nodes
Uhllmann 1996 12 Covariance Intersection (CI)
permits the optimal fusion of estimates that are
correlated to an unknown degree
1970
2000
Shalom and Tse 1975 9 tracking in a cluttered
environment with probabilistic data association
Mutambara 1998 13 Distributed and
Decentralized Extended Information Filter (DDEIF)
estimates information about nonlinear state
parameters, observations, and system dynamics
(time-varying states)
Chong et al. 1983 10 how to optimally account
for correlations due to common information
(static states)
11
Recent Work
Boyd et al. 2005 16 gossip-based methods for
distributed averaging problem (each node
communicates with no more than one neighbor in
each time slot)

• Li et al. 2003 15 first general and systematic
  approach to development of distributed fusion
  rules (optimal fusion with time-invariant states)

2000
Present

• Kumar et al 2003 14 DFuse architectural
  framework for dynamic application-specified data
  fusion in future sensor networks
• Fusion API facilitating fusion function
  implementation
• Distributed dynamic fusion function assignment
  and relocation (accommodating dynamic nature of
  WSN)

• Olfati-Saber et al. 2006 4 Distributed Kalman
  Filter (DKF)

12
Distributed Kalman Filtering

• Distributed algorithm for Kalman filtering
• Applicable in large-scale sensor networks with
  limited capabilities (e.g. local communication,
  routing)
• Analyzable performance in terms of properties of
  the network
• Excellent robustness properties regarding various
  network imperfections, including delay, link
  loss, network fragmentation, and asynchronous
  operation
• Assumes identical sensing models across WSN
• Discrete-time approach
• Decomposes KF into n collaborative mirco-KFs with
  local communication
• Estimating inputs for each micro-KF involves two
  dynamic consensus problems solved using two
  consensus filters
• Low-pass CF fusion (average) of measurements
• Band-pass CF fusion (average) of
  inverse-covariance matrices

13
Consensus Filters

• CFs are distributed algorithms that allow
  calculation of average-consensus of time-varying
  signals
• Tracking uncertainty principle

Sensing model
Collective dynamics
14
Extensions to DKF

• Revised DKF (2007) 5
• Recently proposed by R. Olfati-Saber
• Three types of DKF
• 1st Applicable to sensor networks with different
  observation matrices (sensing models)
• 2nd and 3rd Continuous-time distributed Kalman
  filters with different consensus strategies
• Diffusion DKF (2008) 7
• Proposed by Cattivelli et al.
• Assumes linear system dynamics and sensing model
• Replaces consensus with diffusion process and
  outperforms DKF
• Multi-scale DKF (2008) 8
• Proposed by Kim et al.
• Based on newly introduced multi-scale consensus
  algorithm
• Faster convergence and order-of-magnitude
  reduction of the communication cost

15
What Could Be Done Further

• Extension of Diffusion DKF to
• Heterogeneous networks
• Nonlinear systems
• Multi-scale diffusion scheme
• An in-depth comparison between the DKF and other
  existing decentralized fusion algorithms
• Deployment of DKF (and its variants) in practical
  applications (e.g. surveillance, monitoring,
  etc.)

16
References

• 1 I.F. Akyildiz, W. Su, Y. Sankarasubramaniam,
  E. Cayirci, Wireless sensor networks a survey,
  Computer Networks 38 (2002) 393422
• 2 C. F. García-Hernández, P. H.
  Ibargüengoytia-González, J. García-Hernández,
  and J. A. Pérez-Díaz, Wireless Sensor Networks
  and Applications a Survey, IJCSNS, VOL.7 No.3,
  March 2007
• 3 C. CHONG, AND S. P. KUMAR, Sensor Networks
  Evolution, Opportunities,
• and Challenges, PROCEEDINGS OF THE IEEE, VOL.
  91, NO. 8, AUGUST 2003
• 4 R Olfati-Saber, Distributed Kalman Filtering
  and Sensor Fusion in Sensor Networks, Lecture
  notes in control and information sciences, 2006 -
  Springer
• 5 R. Olfati-Saber, Distributed Kalman
  Filtering for Sensor Networks, Proc. of the 46th
  IEEE Conference on Decision and Control, 2007
• 6 R. Olfati-Saber, J. S. Shamma, Consensus
  Filters for Sensor Networks and Distributed
  Sensor Fusion, Proceedings of IEEE Conference on
  Decision and Control, 2005
• 7 F. S. Cattivelli, C. G. Lopes, A. H. Sayed,
  DIFFUSION STRATEGIES FOR DISTRIBUTED KALMAN
  FILTERING FORMULATION AND PERFORMANCE ANALYSIS,
  Proc. Cognitive Information Processing,
  Santorini, Greece, 2008

17
References

• 8 J. Kim, M. West, E. Scholte, and S.
  Narayanan, Multiscale Consensus for
  Decentralized Estimation and Its Application to
  Building Systems, 2008 American Control
  Conference, 2008
• 9 Y. Bar-Shalom and E. Tse, Tracking in a
  cluttered environment with probabilistic data
  association, Automatica, 11(5)451460, Sept.
  1975.
• 10 C. Y. Chong, E. Tse, and S. Mori,
  Distributed estimation in networks, In
  Proceedings of the 1983 American Control
  Conference, volume 1, pages 294300, San
  Francisco, CA, Sept. 1983.
• 11 B.S. Rao, and H.F. Durrant-Whyte, Fully
  decentralized algorithm for multisensor Kalman
  filtering, IEE PROCEEDINGS-D, Vol. 138, NO. 5,
  SEPTEMBER 1991
• 12 J. K. Uhlmann, General Data Fusion for
  Estimates With Unknown Cross Covariances,
  Proceedings of SPIE, 1996
• 13 A. Mutambara, Decentralized estimation and
  control for multisensor systems, CRC Press, 1998

18
References

• 14 R. Kumar, M. Wolenetz, B. Agarwalla, J.
  Shin, P. Hutto, A. Paul, and U. Ramachandran,
  DFuse A Framework for Distributed Data Fusion,
  Proceedings of the 1st international conference
  on Embedded networked sensor systems, pp.
  114-125, 2003
• 15 X. R. Li, Y. Zhu, J. Wang, and C. Han,
  Optimal Linear Estimation FusionPart I Unified
  Fusion Rules, IEEE TRANSACTIONS ON INFORMATION
  THEORY, VOL. 49, NO. 9, SEPTEMBER 2003
• 16 S. Boyd, A. Ghosh, S. Prabhakar, D. Shah,
  Gossip Algorithms Design, Analysis and
  Applications, Proceedings IEEE INFOCOM, 2005
• 17 http//www.permasense.ch/

19

• Thank YOU!