A Variable Step Size LMS Based on Sparsity for System Identification

• Published in: Applied Mechanics and Materials Vol. 475-476; no. Sensors, Measurement and Intelligent Materials II; pp. 1060 - 1066
• Main Authors: Ju, Hua, Chen, X.Q., Fan, Wei, Huang, W.G., Zhu, Z.K.
• Format: Journal Article
• Language: English
• Published: Zurich Trans Tech Publications Ltd 01.12.2013
• Subjects: Adjustment; Algorithms; Attraction; Errors; Least mean squares; System identification; LMS; System Identification; Sparsity; Steady-State; V-RZALMS
• ISBN: 3037859717; 9783037859711
• ISSN: 1660-9336; 1662-7482; 1662-7482
• DOI: 10.4028/www.scientific.net/AMM.475-476.1060

In many practical applications, the impulse responses of the unknown system are sparse. However, the standard Least Mean Square (LMS) algorithm does not make full use of the sparsity, and the general sparse LMS algorithms increase steady-state error because of giving much large attraction to the small factor. In order to improve the performance of sparse system identification, we propose a new algorithm which introduces a variable step size method into the Reweighted Zero-Attracting LMS (RZALMS) algorithm. The improved algorithm, whose step size adjustment is controlled by the instantaneous error, is called Variable step size RZALMS (V-RZALMS). The variable step size leads to yielding smaller steady-state error on the premise of higher convergent speed. Moreover, the sparser the system is, the better the V-RZALMS performs. Three different experiments are implemented to validate the effectiveness of our new algorithm.