CMStatistics 2015: Start Registration
View Submission - CMStatistics
B1522
Title: Multi-sensor fusion filtering problems with random measurement matrices, cross-correlated noises and packet dropouts Authors:  Aurora Hermoso-Carazo - Universidad de Granada (Spain) [presenting]
Josefa Linares-Perez - Universidad de Granada (Spain)
Raquel Caballero-Aguila - Universidad de Jaen (Spain)
Abstract: In recent years, the use of random matrices in research on fusion estimation problems of networked systems has gained great interest since they arise in many situations involving stochastic sensor gain degradation, measurement multiplicative noises or missing measurements. Furthermore, sensor networks usually produce communication random packet losses which could degrade the network performance. We address the centralized and distributed fusion filtering problems of discrete-time signals using measurements perturbed by random parameter matrices, which are transmitted by different sensors subject to random packet dropouts. Different white sequences of Bernoulli random variables with known probabilities are used to model the potential packet dropouts. Moreover, the fairly conservative assumption that the measurement noises are uncorrelated is weakened and it is assumed that all the sensor noises are one-step autocorrelated and different sensor noises are one-step cross-correlated. Using covariance information, a recursive algorithm for the centralized least-squares linear filter is derived by an innovation approach. Also, local least-squares linear filters based on the measured data of each sensor are obtained and the distributed fusion method is then used to obtain a fusion filter as the matrix-weighted sum of such local estimators that minimizes the mean squared estimation error.