Abstract:
This thesis establishes algorithms and analytical methods for implementing and
evaluating distributed information fusion in networks with non-deterministic
communications connectivity. The methods developed in this thesis encompass any
sequence of fusion events within an arbitrary network resulting from random channel
characteristics, network delays, and ad hoc networking.
A distributed fusion approach is developed and proposed as a general solution for
distributed fusion agents, enabling each agent to operate autonomously and
collaboratively as network conditions allow. The resulting Local Fusion Graph method
enables fusion agents to exchange data on an ad hoc or opportunistic basis for distributed
fusion. The method’s decentralized approach is inherently able to overcome difficulties
experienced by existing distributed fusion methods resulting from messages that are
dropped, delayed, or received out of order. The method provides an algorithm that can be
implemented into distributed fusion agents without a priori knowledge of network
architecture, membership, or communications patterns. It also provides a graphical
approach that can be used for analysis as well as simulation model development.
A stochastic-based fusion formulation is similarly developed and proposed as a
general solution for distributed estimation and trend analysis. The method encapsulates
the effects of non-deterministic behaviors and characteristics into probabilistic factors
that are integrated into the fusion equations. The resulting stochastic fusion method
enables average estimation performance of distributed fusion networks having non-deterministic
characteristics and ad hoc connectivity. The method also greatly simplifies
the simulation and analysis performance approximations for distributed fusion by
providing significant reductions in simulation complexity and computational
requirements while using non-idealized communications characteristics.
The two methods are implemented in computer-based models for distributed
tracking in networks with non-deterministic connectivity. The results of the computer
models are used to assess general trends in estimation capabilities with respect to average
network connectivity. The analyses also address communications requirements relative
to other distributed fusion approaches in the context of estimation accuracy.
The results of this thesis are the ability to implement, model, and assess
distributed information fusion in arbitrary communications networks with realistic
communications characterizations. The solutions proposed in this thesis demonstrate
increased estimation capabilities under non-ideal networking conditions that are
characteristic of wireless mobile networking environments.
