In order to quickenthe convergence rateand enhanceperformanceforanti-noise when identify the unknown coefficients of a sparse system. A variable step-size l 1 -LMS algorithm is proposed. In this paper, using current error and correlation value of errorto adjust step-size and zero-attracting part. Moreover, when the current error satisfies different conditions, the two step-size formulas will switch dynamically. Some l 1 -LMS algorithms availableadapted fixed step-size. However, fixed step-size cannot balance the contradictions convergence rate and the steady-state mean square error (MSE), efficiently.In addition, how to improve performance for noise resistance may be still a problem even with correlated inputs.All are the problem needs to be studied urgently. In order to solve theabovedifficulties,an improvedvariable step-size l 1 -LMS algorithm is proposed. Performances of the proposed l 1 -LMS algorithm are analyzed.After theoretical analysis,the algorithm, experiments with different signal-to-noiseratio (SNR) are given to show the efficiency of the proposed l 1 -LMS algorithm. The variable step-size l 1 -LMS algorithm can achieve faster convergence rate, and implacable anti-noise, and identify the unknown coefficients efficiently.