Extending Manifold-Based MIMO System Identification to Adaptive Crosstalk Cancellation

Authors:

Hahn, J. ,  Kabzinski, T. ,  Jax, P.

Book Title:

ITG-Fachtagung Sprachkommunikation

Organization:

VDE

Publisher:

VDE

Pages:

p.p. 26-30

Date:

Sep. 2025

ISBN:

978-3-80076-617-8

Language:

English

Abstract

The reproduction of binaural signals using loudspeakers requires a Crosstalk Cancellation (CTC) system for the suppression of the crosstalk and equalization of the direct paths. CTC systems rely on accurate models of the transfer paths between the loudspeakers and the listener’s ears. Such models can be obtained using microphones at the ears and adaptive filtering methods for online system identification. However, adaptive filtering methods can suffer from the non-uniqueness problem and struggle with increasing movement velocity. This contribution extends a manifold-based MIMO system identification algorithm to adaptive CTC by modifying the manifold model. First, we introduce a loss function that emphasizes cross-path errors. Second, we adapt the network architecture to the specific application needs. Our simulations reveal that the proposed system yields a decreased sensitivity of the CTC performance towards listener movements.

Download

BibTeX

---