End-to-End Network Slicing in Virtualized OFDMA-Based Cloud Radio Access Networks

We consider the resource allocation for the virtualized OFDMA uplink cloud radio access network (C-RAN), where multiple wireless operators (OPs) share the C-RAN infrastructure and resources owned by an infrastructure provider (InP). The resource allocation is designed through studying tightly coupled problems at two different levels. The upper-level problem aims at slicing the fronthaul capacity and cloud computing resources for all OPs to maximize the weighted profits of OPs and InP considering practical constraints on the fronthaul capacity and cloud computation resources. Moreover, the lower-level problems maximize individual OPs' sum rates by optimizing users' transmission rates and quantization bit allocation for the compressed I/Q baseband signals. We develop a two-stage algorithmic framework to address this two-level resource allocation design. In the first stage, we transform both upper-level and lowerlevel problems into corresponding problems by relaxing underlying discrete variables to the continuous ones. We show that these relaxed problems are convex and we develop fast algorithms to attain their optimal solutions. In the second stage, we propose two methods to round the optimal solution of the relaxed problems and achieve a final feasible solution for the original problem. Numerical studies confirm that the proposed algorithms outperform two greedy resource allocation algorithms and their achieved sum rates are very close to sum rate upper-bound obtained by solving relaxed problems. Moreover, we study the impacts of different parameters on the system sum rate, performance tradeoffs, and illustrate insights on a potential system operating point and resource provisioning issues.