Convergence factor and steady-state error are a pair of inherent contradictions in fixed step size LMS algorithm. For a long time, how to balance the two has been the focus of research. This paper analyzed the fixed step LMS algorithm, proposed a variable step LMS algorithm, and combined them with the multiscale wavelet transform algorithm to propose a new algorithm to solve this problem. From the Maltab simulation results, it can be seen that the new algorithm solved the problem between them to a certain extent, and had stronger noise suppression ability and faster convergence speed.