Spline adaptive filter with arctangent-momentum strategy for nonlinear system identification

• Published in: Signal processing Vol. 164; pp. 99 - 109
• Main Authors: Yang, Liangdong, Liu, Jinxin, Yan, Ruqiang, Chen, Xuefeng
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2019
• Subjects: Impulsive noise; Momentum; Nonlinear system identification; Spline adaptive filter; Wiener model; Wiener model; Momentum; Impulsive noise; Spline adaptive filter; Nonlinear system identification
• ISSN: 0165-1684; 1872-7557
• DOI: 10.1016/j.sigpro.2019.06.007

•An enhanced SAF is proposed.•The arctangent cost function is adopted to improve the robustness against the impulsive noises.•A modified momentum SGD is proposed to improve the convergence speed and reduce the steady-state error. In order to mitigate the interference of impulsive noises in the identification of Wiener-type nonlinear systems using traditional spline adaptive filter (SAF) algorithm, an enhanced SAF, named SAF-ARC-MMSGD, is proposed in this paper. Two improvements have been made in the proposed algorithm. First, the arctangent (ARC) function which is insensitive to large outliers is adopted to construct the cost function, so that the robustness of SAF against impulsive noises is improved. In addition, a modified momentum stochastic gradient descent (MMSGD) method containing two modifications, which are decay technique and selective strategy, is put forward in order to further improve the convergence speed and decrease the steady-state error during iteration. The convergence property of the proposed algorithm has been theoretically analyzed. The results of the numerical simulations have confirmed that the proposed algorithm has superior performance compared with the existing SAF related algorithms.