Sparse least mean fourth filter with zero-attracting ℓ1-norm constraint

Guan Gui, Fumiyuki Adachi

Research output: Contribution to conferencePaperpeer-review

16 Citations (Scopus)


Traditional stable adaptive filter was used normalized least-mean square (NLMS) algorithm. However, identification performance of the traditional filter was especially vulnerable to degradation in low signal-noise-ratio (SRN) regime. Recently, adaptive filter using normalized least-mean fourth (NLMF) is attracting attention in adaptive system identifications (ASI) due to its high identification performance and stability. In the case of sparse system, however, the NLMF filter cannot identify effectively due to the fact that its algorithm neglects the inherent sparse structure. In this paper, we proposed a sparse NLMF filter using zero-attracting ℓ1-norm constraint to exploit the sparsity and to improve the identification performance. Effectiveness of the proposed filter is confirmed from two aspects: 1) stability is derived equivalent to well-known stable NLMS filter; 2) identification performance of the proposed is verified by mean square deviation (MSD) standard in computer simulations. When comparing with conventional adaptive filter, the proposed one can achieve much better identification performance especially in low SNR regime.

Original languageEnglish
Publication statusPublished - 2013
Event9th International Conference on Information, Communications and Signal Processing, ICICS 2013 - Tainan, Taiwan, Province of China
Duration: 2013 Dec 102013 Dec 13


Other9th International Conference on Information, Communications and Signal Processing, ICICS 2013
Country/TerritoryTaiwan, Province of China


  • Normalized least-mean fourth (NLMF)
  • Normalized least-mean square (NLMS)
  • Sparse system identification
  • adaptive filter

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Information Systems
  • Signal Processing


Dive into the research topics of 'Sparse least mean fourth filter with zero-attracting ℓ<sub>1</sub>-norm constraint'. Together they form a unique fingerprint.

Cite this