Datenbestand vom 20. Mai 2019
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
aktualisiert am 20. Mai 2019
978-3-8439-0890-0, Reihe Ingenieurwissenschaften
Generalized Nonlinear System Identification using Adaptive Volterra Filters with Evolutionary Kernels
262 Seiten, Dissertation Universität Erlangen-Nürnberg (2012), Softcover, A5
This thesis presents a novel adaptive filtering framework for a unified identification of linear and nonlinear systems. The main idea is motivated by the observation that various transversal nonlinear models can be equivalently expressed by the broad class of Volterra filters in multichannel diagonal-coordinate representation.
In order to obtain a concurrent estimation of the models' coefficients and their optimum structural parameters, the proposed evolutionary Volterra estimation (EVOLVE) performs an automatic configuration of the utilized filtering memory. To this end, competing kernels of different size are employed in convex combinations, such that conclusions about further growth or shrinkage operations can be drawn and applied at every stage of an hierarchical framework. Since this approach can be efficiently realized by resorting to “virtualized” competitors and DFT-domain implementations, it bridges the gap between purely linear and different nonlinear filter structures in a flexible way.