Distributed parallel processing state estimation algorithms
Konstantinos N. Plataniotis · 1994
In this dissertation parallel processing algorithms for state estimation are developed and analyzed. The state estimation problem is addressed from two different perspectives. First, the statistical filters for state estimation are extended to obtain decentralized algorithms for multisensor state estimation. The developed algorithms provide optimal solution for the linear estimation and detection problems. Further, new decentralized adaptive estimation algorithms are derived for adaptive multisensor systems. Secondly, a new class of estimators called neural estimators are developed by combining elements of emerging neural technology with the results existing in filtering and adaptive signal processing. The developed neural estimators are capable of providing efficient solution for nonlinear and adaptive estimation. These neural estimators are suitable for modern multisensor systems. Because of their massive, parallel structure these estimators can accommodate vast quantities of data providing a powerful counterpart to the decentralized statistical estimators. In the sequence, the developed estimators are applied to a series of problems. The neural estimators are applied to generic adaptive/nonlinear estimation problems and compared with the statistical estimators. The statistical and the corresponding neural estimators are also applied to ship positioning and heave compensation in ocean engineering. Finally, the neural algorithms are used to detect actual inputs in an adaptive channel equalization scenario.