Information Sciences and Systems Lab

1
Information Sciences and Systems Lab
Energy-Efficient Distributed Detection
Introduction Our research lies in the interface
of communications, signal processing (computing)
and networking. Sponsored by National Science
Foundation, our recent work focuses on
distributed and collaborative information
processing and crosslayer design (with a physical
layer emphasis) in wireless ad hoc and sensor
networks, and associated information-theoretic
and computation-theoretic analysis.
Virtual MIMO Communications in Wireless Networks

• Distributed Detection Via Multihop Fusion
• Significant energy reduction compared with
  direct transmission
• Multihop forwarding
• Histogram Fusion
• Multihop Log-likelihood Ratio (LLR) Fusion
• Joint optimization of fusion rules and
  transmission structure
• LLR fusion performs best.

• Nodes in close proximity can cooperate in
  transmission to form a virtual MIMO system
• Virtual MIMO reduces the energy consumption for
  the same throughput and BER target

Communications - Relay/Cooperative diversity -
Virtual MIMO - Distributed source coding --
Information spreading

• Determination of the optimal transmission
  strategy depends on many interacting factors

• Distributed Detection With A Multiple Access
  Channel
• MAC fusion achieves centralized performance
  asymptotically with a properly designed local
  mapping rule.
• Better bandwidth efficiency and detection
  performance than PAC (Parallel Access Channel)
  fusion

Signal Processing - Detection and estimation -
Localization and tracking - Statistical
inference - Distributed computation
Networking - Belief propagation - Network
coding - Clustering - Geographic routing
Error rate for detection of a sinusoid signal in
correlated Gaussian noise
Accelerating Distributed Consensus
Optimal Throughput and Energy Efficiency for
Sensor Networks A Crosslayer Study
Dynamic Self-Calibration in Wireless Networks A
Belief Propagation Approach

• We investigate distributed algorithms leading to
  faster computation of aggregate functions of node
  values
• Cluster-based Solutions
• A cluster acts as an entity and information
  exchange among clusters
• Faster convergence due to better connectivity
• Less communication burden

• Belief propagation is a general class of
  distributed message-passing algorithms, intended
  to solve the NP-hard probabilistic inference
  problems

• Its application in wireless networks requires a
  reverse thinking, intended to serve as a general
  framework for collaborative information
  processing and dissemination
• Particle filtering can be exploited for
  efficient message composition, processing and
  transmission in practice

• For given MAC schemes and detectors, we optimize
  average number of transmissions per slot and the
  transmission power for
• Throughput Maximization
• Throughput-constrained Energy Minimization

For more information, contact Dr. Huaiyu Dai
2108 Engr Bldg II Phone [phone]
Email Huaiyu_Dai AT ncsu.edu

• Nonreversible Random Walk Based Solutions
• Known schemes are based on reversible chains
• We propose a Location-aided Distributed
  Averaging (LADA) algorithm based on nonreversible
  chains
• Node classifies neighbors by their relative
  locations
• Converges substantially faster than known
  schemes