Revisiting Downlink Scheduling in a Multi-Cell OFDMA Network: From Full Base Station Coordination to Practical Schemes

We revisit the user scheduling problem on the downlink of OFDMA cellular networks. Our aim is to find the optimal system-wide schedule of a multi-cell system to understand how much a fully coordinated scheduling could improve performance. It is a highly non-convex integer problem. We propose a method to upper bound this problem by a signo-mial programming problem that can be solved. The solution to the signomial problem can be used to derive a feasible solution to the original global scheduling problem. We show numerically that the gap between the upper bound and that feasible solution is very small. We then provide results both for homogeneous and heterogeneous networks. In the homogeneous case, we show that the coordinated scheduling significantly outperforms a simple benchmark that uses a local scheduling based on equal power. However, the centralized scheduler is very complex and it requires all the channel state information in the system. Therefore, we use the feasible solutions to the system-wide problem to derive a practical scheme based on a well-parameterized soft frequency reuse and a simple local scheduler. We show that the coordinated scheduling performs only 20% better than this practical scheme. The method we study can also be applied to obtain an upper bound for the case of heterogeneous networks. We compare the solution obtained by fully coordinating the scheduling of all the base stations with practical schemes inspired by the one obtained for the homogeneous case and show that the difference between the upper bound and the performance of the best of these schemes is 21%, which again questions the need for coordination.