Abstract: The acoustic channel has the characteristics of sparsity, so the sparse channel estimation algorithm with high accuracy and low complexity is of great significance to the underwater acoustic communication. The problem of channel estimation based on adaptive filtering algorithm is essentially a problem of solving parameters of linear regression model. In allusion to the shortcomings of high complexity and low accuracy of traditional Least Square (LS), Least Mean Square (LMS) and Recursive Least Squares (RLS) algorithms in sparse channel estimation, which is due to neglecting multicollinearity of arguments in solving the problem of linear regression model with these algorithms, in this paper the norm of the channel tap coefficient is added to the cost function of the classical RLS algorithm to constrain itself so as to improve the accuracy of sparse channel estimation. Then, a slide window is used to process the cost function to reduce the computational burden of the algorithm. On this basis, the Dichotomous Coordinate Descent (DCD) algorithm is introduced to search for the solution that minimizes the cost function in a single iteration, the complexity of the algorithm can be further reduced. The simulation results show that the proposed algorithm is superior in the estimation accuracy and complexity compared with the classical algorithm.