Device-to-Device (D2D) communication is a promising technology toward the fifth generation mobile communication. However, when D2D communication is incorporated into heterogeneous cloud radio access network (H-CRAN), the interference management between the D2D users and current users is a challenge. In this paper, we study how to assign the sub-channels of different bandwidth to multiple D2D pairs and the remote radio head users. In such a way, the sub-channels that have been pre-allocated to macro-cell users can be reused, the system performance can be maximized while the quality of service of all users can be guaranteed. Such a resource allocation problem is formulated as a mixed integer nonlinear programming (MINLP) problem which is NP-hard. To obtain the solution, the proposed problem is reformulated into a many-to-one matching sub-game with externality followed by a coalition sub-game. Then, a constrained deferred acceptance algorithm and a coalition formation algorithm are proposed to find solutions to these two sequential sub-games, respectively. Finally, we prove theoretically that both the proposed algorithms can convergent with a low computational complexity. Numerical results demonstrate that compared with existing resource allocation schemes, our proposed scheme can significantly improve the system performance in terms of overall throughput, the total number of admitted users and fairness.