Topics in Distributed Inference

This thesis covers three research topics in the broad field of distributed Inference. The first topic is tracking for passive radar in a Digital Audio Broadcasting (DAB)/Digital Video Broadcasting (DVB) network. Passive radar is a bistatic or multistatic radar using illuminators of opportunity. Passive radar using a DAB/DVB network with Orthogonal Frequency Division Multiplexing (OFDM) is the focus of this research. In this system, bistatic range and range-rate are available as measurements, while angular information is assumed unavailable due to the poor quality/lack of the information. Due to the use of the same carrier frequency, there is additional association ambiguity between the signals and illuminators, in addition to the well-known ambiguity between the signals and targets, which leads to a significant data association problem. This research presents two track initiation algorithms and five trackers, which provide tracks directly in the geographic space (3-D Cartesian domain) without dealing with the numerous targets-measurements-illuminators associations. The second topic is testing with communication constraints for fault diagnosis. Most approaches to fault diagnosis have been focused on detecting and isolating a possible fault under cost constraints such as economic factors and computational time. In some systems, however, such as remote monitoring (e.g., satellite, sensor field) systems, there is a communication constraint between the system being monitored and the monitoring facility. In such circumstances it is desirable to isolate the faulty component with as few interactions as possible. The key feature is that multiple tests are chosen at each stage in such a way that the tests within the chosen group complement each other. We propose two algorithms for fault diagnosis under communications constraints. Their performances are analyzed in terms of the average number of testing stages as well as in terms of the required computational complexity. The third topic is sensor localization using channel energy measurements by distributed sensors. Most approaches usually assume that the target does not move significantly during the time needed to collect and process the data from the sensors. The goal of this research is to estimate the trajectory of a moving target using a network of distributed sensors that measure only the received signal strength (RSS), sampled as a function of time, without knowledge of the target amplitude/source level. To reduce the communication load, sensors communicate a reduced data set (which is described in detail later) to the fusion center (FC), generated through local processing. To generate the reduced data sets, each sensor calculates a local maximum likelihood (ML) estimate of its parameters. The FC combines the data transmitted by the sensors using a ML-like formulation based on the local FIMs. This approach has a very low communication load, performs comparably to a centralized estimator, and, due to the modularized setup, any measurement model at the sensors can be considered