Adaptive proximal forward-backward splitting for sparse system identification under impulsive noise

In this paper, we propose a robust sparsity-aware adaptive filtering algorithm under impulsive noise environment, by using the Huber loss function in the frame of adaptive proximal forward-backward splitting (APFBS). The APFBS attempts to suppress a time-varying cost function which is the sum of a smooth function and a nonsmooth function. As the smooth function, we employ the weighted sum of the Huber loss functions of the output residuals. As the nonsmooth function, we employ the weighted ℓ1 norm. The use of the Huber loss function robustifies the estimation under impulsive noise and the use of the weighted ℓ1 norm effectively exploits the sparsity of the system to be estimated. The resulting algorithm has low computational complexity with order, where is the tap length. Numerical examples in sparse system identification demonstrate that the proposed algorithm outperforms conventional algorithms by achieving robustness against impulsive noise.

[1]  Tung-Sang Ng,et al.  Fast least mean M-estimate algorithms for robust adaptive filtering in impulse noise , 2000, 2000 10th European Signal Processing Conference.

[2]  Isao Yamada,et al.  Sparsity-aware adaptive filters based on ℓp-norm inspired soft-thresholding technique , 2012, 2012 IEEE International Symposium on Circuits and Systems.

[3]  J. Moreau Fonctions convexes duales et points proximaux dans un espace hilbertien , 1962 .

[4]  J. Chambers,et al.  A robust mixed-norm adaptive filter algorithm , 1997, IEEE Signal Processing Letters.

[5]  Isao Yamada,et al.  Sparsity-aware adaptive filtering based on a Douglas-Rachford splitting , 2011, 2011 19th European Signal Processing Conference.

[6]  Kazuhiko Ozeki,et al.  An adaptive filtering algorithm using an orthogonal projection to an affine subspace and its properties , 1984 .

[7]  Paul Petrus,et al.  Robust Huber adaptive filter , 1999, IEEE Trans. Signal Process..

[8]  B. Ripley,et al.  Robust Statistics , 2018, Encyclopedia of Mathematical Geosciences.

[9]  T. Hinamoto,et al.  Extended theory of learning identification , 1975 .

[10]  A. Hero,et al.  Regularized Least-Mean-Square Algorithms , 2010, 1012.5066.

[11]  Isao Yamada,et al.  Acceleration of adaptive proximal forward-backward splitting method and its application to sparse system identification , 2011, 2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[12]  Patrick L. Combettes,et al.  Signal Recovery by Proximal Forward-Backward Splitting , 2005, Multiscale Model. Simul..

[13]  I. Yamada,et al.  Adaptive Projected Subgradient Method for Asymptotic Minimization of Sequence of Nonnegative Convex Functions , 2005 .

[14]  Isao Yamada,et al.  A sparse adaptive filtering using time-varying soft-thresholding techniques , 2010, 2010 IEEE International Conference on Acoustics, Speech and Signal Processing.

[15]  B. Mercier Lectures on Topics in Finite Element Solution of Elliptic Problems , 1980 .

[16]  Tania Stathaki,et al.  A normalized robust mixed-norm adaptive algorithm for system identification , 2004, IEEE Signal Processing Letters.

[17]  Isao Yamada,et al.  Sparse system identification by exponentially weighted adaptive parallel projection and generalized soft-thresholding , 2010 .

[18]  Jacob Benesty,et al.  A robust variable step-size affine projection algorithm , 2010, Signal Process..

[19]  J. Nagumo,et al.  A learning method for system identification , 1967, IEEE Transactions on Automatic Control.

[20]  Sergios Theodoridis,et al.  Online Sparse System Identification and Signal Reconstruction Using Projections Onto Weighted $\ell_{1}$ Balls , 2010, IEEE Transactions on Signal Processing.

[21]  Jacob Benesty,et al.  A New Robust Variable Step-Size NLMS Algorithm , 2008, IEEE Transactions on Signal Processing.

[22]  Gerhard Schmidt,et al.  Adaptive Algorithms for the Identification of Sparse Impulse Responses , 2006 .

[23]  Frederick R. Forst,et al.  On robust estimation of the location parameter , 1980 .