ISBN 9783843944939

978-3-8439-4493-9, Reihe Ingenieurwissenschaften

Christian Hofmann
Efficient System Identification using Parallel Connections and Cascades

274 Seiten, Dissertation Universität Erlangen-Nürnberg (2020), Softcover, A5

Zusammenfassung / Abstract

This thesis is dedicated to efficient System Identification (SysId) and Acoustic Echo Cancellation (AEC) for two classes of applications of outstanding practical relevance. Class 1 is SysId for single-channel systems with nonlinearly distorting playback equipment (e.g., miniaturized amplifiers and loudspeakers of smartphones in hands-free mode). Class 2 is SysId for the linear Multiple-Input/Multiple-Output (MIMO) systems emerging as Loudspeaker-Enclosure-Microphone Systems (LEMS) between the loudspeakers of a premium sound reproduction system (an object-based rendering system like a Wave Field Synthesis (WFS) system) and a microphone array for voice control.

The concept of Significance-Aware (SA) filtering is introduced as efficient solution for class 1. SA filtering gains efficiency by decomposing the identification of the nonlinear system into synergetically interacting adaptive subsystems: a long but efficient linear system capturing the dispersive component and a structurally more complex nonlinear system, which nevertheless causes low computational effort due to its short memory.

For class 2, the concept of Source-Specific System Identification (SSSysId) is proposed. SSSysId employs reference information from an object-based rendering system (virtual source signals and the driving filters distributing the virtual source signals to the loudspeakers). SSSysId gains computational efficiency by identifying the serial connection of the driving filters with the LEMS instead of the LEMS itself and using these identified source-specific systems to track the actual LEMS in a second stage. Unlike in SA filtering, efficiency is not caused by shortening the temporal support but by reducing the number of channels.

The presented computational complexity analyses and experimental results confirm both efficiency and efficacy of SA filtering and SSSysId.

Finally, a brief outlook on the generalization of SA filtering and SSSysId to nonlinear MIMO systems is provided.