Performance Analysis and Optimization of Virtualized Cloud-RAN Systems

Cloud radio access networks (C-RAN) are a promising solution against the ossification of wireless systems. C-RANs provide a platform for rapid innovation and deployment of new wireless technologies. However, they also present a set of challenges un-encountered in traditional systems. The goal of this thesis is to identify, study and provide solutions for those challenges. The challenges studied in this thesis fall into two broad categories; the first set of challenges is about multiplexing several network slices on the same physical infrastructure. The second set of challenges stems from the cloud computing concept itself and how it affects the wireless systems architecture. For the first part, we start at the PHY-layer, and focus on the question of how multiple network slices can be accommodated on the same infrastructure. We conduct a performance analysis of the alternative multiplexing and scheduling schemes that can be used for slicing and interference coordination. Next, we show how we can integrate the effects of statistical multiplexing into PHY-layer performance indicators, and provide an algorithm for admission control combined with resource slicing using both FDMA and SDMA. For the cloud computing challenges, we start by looking at how the cloud computing model combined with the demands of wireless networks raise the need for efficient distributed scheduling schemes. We provide a completely distributed solution that achieves up to 92% efficiency and discuss the effects of the nature of the scheduler on the performance. One of the main goals of C-RAN is providing more energy-efficient systems through dynamic resource scaling. We investigate this problem from both the radio access part as well as the cloud computing part. For the radio access, we propose an optimization and control framework for the activation, association and clustering of remote radio heads (RRH). The problem is solved using the successive geometric programming approach for signomial optimization. For the cloud computing part, we propose a predictive control framework for anomaly-aware scaling of computing resources. Our proposed scheme is based on the Gaussian process model and provides 95% prediction accuracy and 90% anomaly detection accuracy.Ph.D