The 5th Generation cellular network may have the key feature of smaller cell size and denser resource employment, resulted from diminishing resource and increasing communication demands. However, small cell may result in high interference between cells. Moreover, the random geographic patterns of small cell networks make them hard to analyze, at least excluding schemes in the well-accepted hexagonal grid model. In this paper, a new model---the matrix graph is proposed which takes advantage of the small cell size and high inter-cell interference to reduce computation complexity. This model can simulate real world networks accurately and offers convenience in frequency allocation problems which are usually NP-complete. An algorithm dealing with this model is also given, which asymptotically achieves the theoretical limit of frequency allocation, and has a complexity which decreases with cell size and grows linearly with the network size. This new model is specifically proposed to characterize the next-generation cellular networks.