Signal Processing and Speech Communication Laboratory
homephd theses › Nonlinear System Identification for Mixed Signal Processing

Nonlinear System Identification for Mixed Signal Processing

Status
Finished
Student
Heinz Koeppl
Mentors
Research Areas

The thesis considers methods for the identification of weakly nonlinear systems, met in mixed analog-digital systems for data-transmission. Depending on the available knowledge about the system to be identified different algorithms and model structures can be applied. Thus, one distinguishes between glass-box, gray-box and black-box methods. The contribution of the thesis to the glass-box methods is a scheme for the automatic determination of the Volterra kernels of a weakly nonlinear circuit utilizing Kronecker products. In the field of gray-box methods a model structure and its parameter estimation is presented that allow to incorporate the available knowledge about the linearization of the weakly nonlinear system efficiently into the identification. Black-box methods are extended through the application of model-complexity regulating algorithms from the area of machine learning. Furthermore the relation between the accuracy of the identification and properties of the excitation signal for the identification is investigated and a signal optimization method is proposed. The developed methods are presented using an exemplary circuit and are also applied to the identification of a VDSL (very-high data rate digital subscriber line) line driver circuit.