Based on analysis of the LS-series channel estimation algorithms, the matching pursuit (MP) algorithm with small computation amount was improved by importing both a threshold adjusting mechanism which could effectively avoid false paths and an iterative correction method to increase the estimation precision for important coefficients of the sparse underwater acoustic channel. In order to overcome the dependence on the information of channel’s length, a novel channel estimation algorithm which could adapt to real channel’s length was proposed. Decision feedback equalizer (DFE) coefficients were computed recurring to the real-time channel estimation results while the evaluation of channel estimation’s quality which could be fed back to adjust channel estimation algorithm was given using DFE’s output signals. The novel algorithm could effectively implement estimations for unknown underwater acoustic environments which also benefited reliable underwater acoustic communications. The proposed method was validated by simulations and experiments.