Multichannel Adaptive Filtering with Sparseness Constraints

Inhalt:
The performance of adaptive filtering can be enhanced by incorporating prior system knowledge. In this paper, we systematically consider regularization strategies exploiting sparseness for the identification of acoustic room impulse responses specifically for multichannel systems. Due to the additional dimensions in the multichannel case, a structured regularization appears to be a natural choice. Based on this concept, we present a generic regularized Newtontype algorithm. This generic formulation allows us to discuss various properties specific to the multichannel case and forms a valuable basis for the future development of efficient algorithms. Index Terms — multichannel adaptive filtering, structured regularization