A Frequency-Domain Adaptive Filter (FDAF) Prediction Error Method (PEM) Framework for Double-Talk-Robust Acoustic Echo Cancellation

Jose M. Gil-Cacho, Toon van Waterschoot, Marc Moonen, Søren Holdt Jensen

Research output: Contribution to journalJournal articleResearchpeer-review

8 Citations (Scopus)

Abstract

In this paper, we propose a new framework to tackle the double-talk (DT) problem in acoustic echo cancellation (AEC). It is based on a frequency-domain adaptive filter (FDAF) implementation of the so-called prediction error method adaptive filtering using row operations (PEM-AFROW) leading to the FDAF-PEM-AFROW algorithm. We show that FDAF-PEM-AFROW is by construction related to the best linear unbiased estimate (BLUE) of the echo path. We depart from this framework to show an improvement in performance with respect to other adaptive filters minimizing the BLUE criterion, namely the PEM-AFROW and the FDAF-NLMS with near-end signal normalization. One of the contributions is to propose the instantaneous pseudo-correlation (IPC) measure between the near-end signal and the loudspeaker signal. The IPC measure serves as an indication of the effect of a DT situation occurring during adaptation. We motivate the choice of FDAF-PEM-AFROW over PEM-AFROW and FDAF-NLMS with near-end signal normalization, based on performance, computational complexity and related IPC measure values. Moreover, we use the FDAF-PEM-AFROW framework to improve several state-of-the-art variable step-size (VSS) and variable regularization (VR) algorithms. The FDAF-PEM-AFROW versions significantly outperform the original versions in every simulation. In terms of computational complexity, the FDAF-PEM-AFROW versions are themselves about two orders of magnitude cheaper than the original versions.
Original languageEnglish
JournalI E E E Transactions on Audio, Speech and Language Processing
Volume22
Issue number12
Pages (from-to)2074-2086
ISSN1558-7916
DOIs
Publication statusPublished - 2014

Fingerprint

Echo suppression
adaptive filters
Adaptive filters
cancellation
echoes
Acoustics
acoustics
predictions
Computational complexity
Loudspeakers
Adaptive filtering
loudspeakers
estimates
indication

Cite this

@article{2e0b7c00ffd9413487ce7a3150a805e4,
title = "A Frequency-Domain Adaptive Filter (FDAF) Prediction Error Method (PEM) Framework for Double-Talk-Robust Acoustic Echo Cancellation",
abstract = "In this paper, we propose a new framework to tackle the double-talk (DT) problem in acoustic echo cancellation (AEC). It is based on a frequency-domain adaptive filter (FDAF) implementation of the so-called prediction error method adaptive filtering using row operations (PEM-AFROW) leading to the FDAF-PEM-AFROW algorithm. We show that FDAF-PEM-AFROW is by construction related to the best linear unbiased estimate (BLUE) of the echo path. We depart from this framework to show an improvement in performance with respect to other adaptive filters minimizing the BLUE criterion, namely the PEM-AFROW and the FDAF-NLMS with near-end signal normalization. One of the contributions is to propose the instantaneous pseudo-correlation (IPC) measure between the near-end signal and the loudspeaker signal. The IPC measure serves as an indication of the effect of a DT situation occurring during adaptation. We motivate the choice of FDAF-PEM-AFROW over PEM-AFROW and FDAF-NLMS with near-end signal normalization, based on performance, computational complexity and related IPC measure values. Moreover, we use the FDAF-PEM-AFROW framework to improve several state-of-the-art variable step-size (VSS) and variable regularization (VR) algorithms. The FDAF-PEM-AFROW versions significantly outperform the original versions in every simulation. In terms of computational complexity, the FDAF-PEM-AFROW versions are themselves about two orders of magnitude cheaper than the original versions.",
author = "Gil-Cacho, {Jose M.} and {van Waterschoot}, Toon and Marc Moonen and Jensen, {S{\o}ren Holdt}",
year = "2014",
doi = "10.1109/TASLP.2014.2351614",
language = "English",
volume = "22",
pages = "2074--2086",
journal = "IEEE/ACM Transactions on Audio, Speech, and Language Processing",
issn = "2329-9290",
publisher = "IEEE Signal Processing Society",
number = "12",

}

A Frequency-Domain Adaptive Filter (FDAF) Prediction Error Method (PEM) Framework for Double-Talk-Robust Acoustic Echo Cancellation. / Gil-Cacho, Jose M.; van Waterschoot, Toon; Moonen, Marc; Jensen, Søren Holdt.

In: I E E E Transactions on Audio, Speech and Language Processing, Vol. 22, No. 12, 2014, p. 2074-2086.

Research output: Contribution to journalJournal articleResearchpeer-review

TY - JOUR

T1 - A Frequency-Domain Adaptive Filter (FDAF) Prediction Error Method (PEM) Framework for Double-Talk-Robust Acoustic Echo Cancellation

AU - Gil-Cacho, Jose M.

AU - van Waterschoot, Toon

AU - Moonen, Marc

AU - Jensen, Søren Holdt

PY - 2014

Y1 - 2014

N2 - In this paper, we propose a new framework to tackle the double-talk (DT) problem in acoustic echo cancellation (AEC). It is based on a frequency-domain adaptive filter (FDAF) implementation of the so-called prediction error method adaptive filtering using row operations (PEM-AFROW) leading to the FDAF-PEM-AFROW algorithm. We show that FDAF-PEM-AFROW is by construction related to the best linear unbiased estimate (BLUE) of the echo path. We depart from this framework to show an improvement in performance with respect to other adaptive filters minimizing the BLUE criterion, namely the PEM-AFROW and the FDAF-NLMS with near-end signal normalization. One of the contributions is to propose the instantaneous pseudo-correlation (IPC) measure between the near-end signal and the loudspeaker signal. The IPC measure serves as an indication of the effect of a DT situation occurring during adaptation. We motivate the choice of FDAF-PEM-AFROW over PEM-AFROW and FDAF-NLMS with near-end signal normalization, based on performance, computational complexity and related IPC measure values. Moreover, we use the FDAF-PEM-AFROW framework to improve several state-of-the-art variable step-size (VSS) and variable regularization (VR) algorithms. The FDAF-PEM-AFROW versions significantly outperform the original versions in every simulation. In terms of computational complexity, the FDAF-PEM-AFROW versions are themselves about two orders of magnitude cheaper than the original versions.

AB - In this paper, we propose a new framework to tackle the double-talk (DT) problem in acoustic echo cancellation (AEC). It is based on a frequency-domain adaptive filter (FDAF) implementation of the so-called prediction error method adaptive filtering using row operations (PEM-AFROW) leading to the FDAF-PEM-AFROW algorithm. We show that FDAF-PEM-AFROW is by construction related to the best linear unbiased estimate (BLUE) of the echo path. We depart from this framework to show an improvement in performance with respect to other adaptive filters minimizing the BLUE criterion, namely the PEM-AFROW and the FDAF-NLMS with near-end signal normalization. One of the contributions is to propose the instantaneous pseudo-correlation (IPC) measure between the near-end signal and the loudspeaker signal. The IPC measure serves as an indication of the effect of a DT situation occurring during adaptation. We motivate the choice of FDAF-PEM-AFROW over PEM-AFROW and FDAF-NLMS with near-end signal normalization, based on performance, computational complexity and related IPC measure values. Moreover, we use the FDAF-PEM-AFROW framework to improve several state-of-the-art variable step-size (VSS) and variable regularization (VR) algorithms. The FDAF-PEM-AFROW versions significantly outperform the original versions in every simulation. In terms of computational complexity, the FDAF-PEM-AFROW versions are themselves about two orders of magnitude cheaper than the original versions.

U2 - 10.1109/TASLP.2014.2351614

DO - 10.1109/TASLP.2014.2351614

M3 - Journal article

VL - 22

SP - 2074

EP - 2086

JO - IEEE/ACM Transactions on Audio, Speech, and Language Processing

JF - IEEE/ACM Transactions on Audio, Speech, and Language Processing

SN - 2329-9290

IS - 12

ER -