Distributed detection and decision fusion using particle filter concepts

In this thesis, a distributed detection and decision fusion system that operates under non Gaussian noise with unknown parameters is developed. The main objective is to find decision rules for the local detectors and the fusion center without making unrealistic assumptions about statistics of the observed data. The proposed scheme is based on concepts of using particles. In order to form a dynamical model of the problem, observed data is modeled as an AR process which is driven by Gaussian mixture noise. The proposed system consists of a particle filter, used for estimating the unknown noise parameters, followed by particle swarm optimization (PSO) which achieves distributed detection and decision fusion of local decisions. The fusion rule is designed, without assuming independence of the decisions of the local sensors, by using copula functions to relate the marginal densities of the sensor observations to the statistical dependency between the sensor decisions. The parameter of the copula function used is estimated using PSO. The probability of error values obtained by using the proposed method are compared with theoretical values and promising results are obtained.