<p>Ocean acoustic tomography (OAT) is frequently used for the estimation of sound speed variations in the shallow ocean waveguide. As the first step of OAT, acoustic wideband rays along the different paths are exactly separated by two particular parameters of direction of arrival (DOA) and time of arrival (TOA). However, the multipath propagation of ray paths induces interferences between different rays, which makes it impossible to identify these raypaths. In this paper, a two-dimensional wideband grid-free algorithm is proposed for separating acoustic raypaths by using their DOAs and TOAs. Firstly, a continuous formulation of the raypath separation problem is presented. Then, an atomic norm minimization (ANM) problem is designed by exploiting an atomic norm that promotes signal sparsity in the continuous domain. An equivalent maximization problem is introduced to solve such an ANM problem, which can be solved efficiently with semidefinite programming. Finally, certain parameters of DOA and TOA are achieved through an optimization variable of the maximization problem. The experimental results show that the proposed method obtains more accurate separation performance compared to conventional compressive sensing-based algorithms.</p>