Block-sparsity-aware LMS algorithm for network echo cancellation

• Published in: Electronics letters Vol. 54; no. 15; pp. 951 - 953
• Main Authors: Wei, Ye, Zhang, Yonggang, Wang, Chengcheng
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 26.07.2018
• Subjects: acoustic signal processing; adaptive filters; block‐sparsity‐aware LMS algorithm; echo; echo suppression; existing block‐sparsity; least mean squares methods; LMS algorithms; mean squares algorithm; mixed $l_{2\comma 1}; network echo cancellation; nonzero block; norm; original mean‐square‐error cost function; path impulse response; Signal processing; single block sparsity; single‐block‐sparse estimate; transient response; uniformly partitioned filter tap‐weight vector; unknown echo path; single-block-sparse estimate; least mean squares methods; adaptive filters; transient response; nonzero block; existing block-sparsity; LMS algorithms; echo suppression; network echo cancellation; mixed $l_{2\comma 1}; echo; path impulse response; unknown echo path; norm; mean squares algorithm; original mean-square-error cost function; block-sparsity-aware LMS algorithm; single block sparsity; uniformly partitioned filter tap-weight vector; acoustic signal processing
• ISSN: 0013-5194; 1350-911X; 1350-911X
• DOI: 10.1049/el.2018.1065

Network echo path impulse response is single-block-sparse in nature. In order to obtain a single-block-sparse estimate of the unknown echo path, a new least mean squares (LMS) algorithm is proposed by introducing the penalty of single block sparsity, which is the difference between the mixed $l_{2\comma 1}$l2,1 norm and $l_2$l2 norm of the uniformly partitioned filter tap-weight vector, into the original mean-square-error cost function. This is motivated by the fact that the difference between the mixed $l_{2\comma 1}$l2,1 norm and $l_2$l2 norm of a vector is minimised only when there is at most one non-zero block in the vector. Numerical simulation results show that the proposed algorithm can effectively estimate and track the unknown echo path, outperforming existing block-sparsity-induced LMS algorithms.