In order to solve the problem that a single LMS filter is restricted by mutual influence at convergence speed and stable state error to lead to the performance decrease of the recognition system,the convex combination of leastmeansquare algorithm is employed in this paper by paralleling use of a fast and a slow LMS filter. To further improve the algorithm's capability, a new lowcomplexity convex combination of CLMS algorithm is proposed by improving the traditional CLMS. The proposed algorithm simplifies and improves the renewal iterative formula of parameters' sum by using modified arc tangent function and sign function respectively. Meanwhile the paper employs an instantaneous transfer scheme combined with the window length of to accelerate the convergence rate. Theoretical analysis and simulation results suggest that under the conditions of the influence of noise, relative signal input and unstable environment, the proposed algorithm can not only maintain a superior capability of tracking and mean square, but also possess a higher convergence rate.