Network Slicing and Resource Allocation in Cloud Radio Access Networks

Abstract

Wireless networks based on 5G and beyond 5G technologies are required to support high data traffic demand and diverse service requirements, such as different delay or throughput requirements. Radio Access Network (RAN) slicing has been introduced to address these requirements by virtualizing and sharing the network resources among multiple service providers (SP), which has gained significant attention in recent years. However, most of the current studies are on traditional decentralized network architectures where distributed base stations manage their resources independently from each other, leading to energy and cost inefficiency. We investigate RAN slicing in Cloud Radio Access Networks (C-RAN) as a centralized architecture for virtualizing and sharing the baseband computing resources among several Remote Radio Heads. From the perspective of slicing granularity, the existing solutions are categorized into two groups, user-level and SP-level slicing. In the former, the resources are allocated in small granularity and directly to end-users from all SPs, resulting in higher utilization of resources. In the latter, resources are allocated in bulk to different SPs, resulting in more isolated slices. We study three problems by considering different types of resources in C-RAN and different levels of slicing. First, we consider spectrum resource slicing, where the problem is modeled as a multiple knapsacks with assignment restriction problem, and several heuristic algorithms are proposed to solve it. Then, we investigate RAN slicing in joint transmission (an advanced technique in 5G to increase user data rate) based C-RANs. Taking a user-level slicing approach, we formulate the problem as a sum-rate maximization problem, and a successive convex approximation-based algorithm is proposed to solve it. In the third problem, we study RAN slicing in C-RAN with Multi-access Edge Computing capability, where computation-intensive tasks are offloaded to edge servers. A machine learning-based and user-level slicing framework is proposed to slice the resources among SPs intelligently and efficiently. In this work, the resource allocation decisions are made in multiple stages, rather than a onetime decision-making framework in the existing studies, resulting in higher resource utilization and success rate for requests as evidenced by simulation studies.