Improved acoustic echo cancellation for low SNR based on blockwise combination of filters

Conference contribution
(Conference Contribution)


Publication Details

Author(s): Azpicueta-Ruiz LA, Zeller M, Figueiras-Vidal AR, Arenas-García J, Kellermann W
Publication year: 2010
Volume: 4
Pages range: 2883-2889
ISBN: 9781617827457


Abstract


Acoustic echo cancellers (AEC) are becoming increasingly important because of the widespread use of hands-free devices. Due to their simplicity, most of the cancellers rely on NLMS-type adaptive filters to model and track the time-varying echo path. Recently, adaptive combinations of filters are gaining increasing popularity as a flexible and versatile approach to overcome compromises inherent to adaptive filters, thus enhancing the overall performance. Regarding AEC scenarios, such filter combinations have already been proposed for, e.g., improving the trade-off between convergence speed and steady-state error or for reducing the dependency on varying ratios of linear and nonlinear distortions. In this paper, we present a new AEC approach, showing improved performance for unknown or time-varying signalto- noise ratios (SNR). The proposed scheme exploits the fact that the coefficient energy of a typical echo path is not uniformly distributed, but decays exponentially. Under this condition, an NLMS filter will introduce significant estimation errors for less significant filter taps due to gradient noise. Since the number of affected coefficients strongly depends on the SNR and hence on the implied noise floor, the cancellation performance may degrade considerably for low SNRs. In order to relieve the coefficient noise, the adaptive impulse response is split into a number of nonoverlapping blocks, each of which is combined with a virtual 'zero-block', having fixed zero coefficients, with timevarying relative weights for both the nonzero and the zero-block. In practice, this results in a possibly biased estimation of some of the filter coefficients. However, it has been shown that such estimates can yield advantages in terms of mean-square error, especially for low SNRs. The combination of each block is implemented by a convex mixing, where the control parameter is updated according to a stochastic gradient descent method so as to minimize the global error of the AEC. For moderate block numbers, the increment in computational cost over a conventional NLMS canceller is negligible. In particular, this contribution investigates the operation of the blockwise combined filter for low SNR conditions, comparing its performance with standard NLMS- and PNLMS-type filters. The robustness and benefits of the proposed approach are thereby experimentally verified for noise and speech inputs. Moreover, the influence of the number of blocks and the mixing parameters is also studied and indications on future work (as e.g. accounting for impulsive noise or an extension to nonlinear filters) are given. Copyright © (2010) by the International Congress on Acoustics.



FAU Authors / FAU Editors

Kellermann, Walter Prof. Dr.-Ing.
Professur für Nachrichtentechnik
Zeller, Marcus
Lehrstuhl für Multimediakommunikation und Signalverarbeitung


External institutions
Universidad Carlos III de Madrid (UC3M)


How to cite

APA:
Azpicueta-Ruiz, L.A., Zeller, M., Figueiras-Vidal, A.R., Arenas-García, J., & Kellermann, W. (2010). Improved acoustic echo cancellation for low SNR based on blockwise combination of filters. (pp. 2883-2889). Sydney, NSW, AU.

MLA:
Azpicueta-Ruiz, Luis A., et al. "Improved acoustic echo cancellation for low SNR based on blockwise combination of filters." Proceedings of the 20th International Congress on Acoustics 2010, ICA 2010 - Incorporating the 2010 Annual Conference of the Australian Acoustical Society, Sydney, NSW 2010. 2883-2889.

BibTeX: 

Last updated on 2018-06-08 at 11:53