Distributed data fusion: Problems and strategies

The capabilities of fast real-time computing on multi-processor systems have lead to the investigation of decentralized implementations of many signal processing algorithms. This work focusses on problems of decentralized estimation of static and dynamic systems in a hierarchical framework. In a hierarchically interconnected network with a recognized hierarchy among the sensors, the local stations produce a local estimate based solely on their own observations. These estimates (and not the raw data) are then transmitted to a central processor that combines them using a linear fusion rule to produce the final output of the entire hierarchy.; The first chapter is introductory in nature and sets the perspective for the contributions to follow. Chapter 2 provides a new look at the traditional problem of estimation of the state of a dynamic system evolving in a Gauss-Markov manner when noisy observations are only available.; Chapter 3 looks at some fundamental performance issues of such a hierarchical structure. The concept of global reconstructability is introduced, which implies that the final output of the hierarchy is identically equal to that of a centralized processor with access to all the data collected by the local processors in the hierarchical network. Some necessary and sufficient conditions are derived for global reconstructability to be achieved for the Least-Squares and the Minimum Variance problems, and it is shown that these are not expected to hold in general.; Chapter 4 reintroduces the problem of state estimation of a linear dynamic system, but with two new innovations: (1) The state dynamics incorporates some unknown inputs. (2) The measured return from the object of interest is immeresed in measurements from a competing, independent mechanism.; These two features effectively model the problem of tracking a target given to sudden accelerations or maneuvers, that is observed in a background of land or sea clutter.; Chapter 5 seeks to extend the algorithm developed in the previous chapter towards implementation in hierarchical systems of Chapter 2. The simple sub-optimal combining rule of Chapter 3 is employed for data fusion at the central processor. (Abstract shortened with permission of author.)...