Abstract: The traditional filter bank multi-carrier (FBMC) channel estimation method based on training sequence and block pilot structure spends additional spectrum resources, which leads to certain limitations of applying this method in the underwater acoustic communication environment where spectrum resources are limited. To solve this problem and considering the sparsity of underwater acoustic channel, a channel estimation method of FBMC with discrete pilot structure based on compressed sensing is proposed in this paper. Firstly, based on the idea of equivalent pilot energy maximization, the discrete pilot structure is designed reasonably to solve the problem of inherent imaginary part interference in channel estimation of FBMC system. Then with this structure, the received information at pilot is extracted and the compressed sensing gOMP algorithm with good reconstruction effect is used to reconstruct the underwater acoustic channel. This method not only ensures the accuracy of channel estimation, but also effectively improves the spectrum utilization of FBMC system, which greatly improves the performance of underwater acoustic communication. The simulation results show that the proposed method is superior to the traditional methods in estimation accuracy and spectrum utilization.