Radio Access Network Slicing Based on C/U Plane Separation

Hui Zhao, Liqiang Zhao,＊, Kai Liang, Chengkang Pan

1 State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an 710071, Shaanxi Province, China

2 China Mobile Research Institute, Beijing 100053, China

I. INTRODUCTION

Driven by the rapid development of user equipment (UE) functions and applications,data traffic is expected to surge to 30.6eb by 2020 [1]. Meanwhile, the rise of Internet of Things (IoT) [2] broadens the scope of mobile communication services. In the face of more application services and scenarios, as well as more emergency performance requirements,a innovative network is envisioned. Network slicing is considered as one of the most promising technologies to enable customized services for users. Network slicing [3] is a concept of multiple logical networks that provides specific network capabilities and network characteristics on common physical infrastructures. Each network slice can be seen as an independent virtualized end-to-end network,which includes a set of network function instances with related computing, storage, and network resources.

There are many efforts towards network slicing. 3GPP SA2 made some definitions of the architecture for next generation system and related issues of network slicing in [4]. Software-defined network (SDN) [5] and network function virtualization (NFV) [6] are two important means of network slicing. [7] indicated that virtualization, orchestration and isolation are the major design principles of network slicing which can ensure the resource sharing and parallel operation among network slices.Also, [8]-[9] implemented network slicing based on SDN/NFV. Nakao et al. [10] studied the end-to-end network slicing for 5G mobile networks and discussed some application use cases based on OpenAirInterface (OAI).

As one of the important parts of network slicing, RAN slicing attracts the attention of many parties, which is still full of challenges.[11] presented the Radio Visor to share and manage radio resource in the 3D grid (time,frequency and radio element). This ensured isolation of each slice and the main component, controller, could provide customized services flexibly based on specific radio resources. [12] proposed a framework featuring RAN abstraction. A two-level MAC scheduler was used to abstract and share the radio resources, and enabled to enforce network slicing in the RAN flexibly and dynamically. They also proofed the validity about the proposed framework based on OAI. Although these works considered the isolation and resources allocation among RAN slices, they ignored the vast signaling overhead due to the tightly coupled C/U plane of traditional RAN architecture at air interface. When designing a RAN slice,an independent logical RAN, it is necessary to consider the control plane and user plane jointly, which will trigger vast control signaling overhead. To tackle the insufficient, it is natural to decouple control plane from data plane. Aided by C/U plane separation, the deployment and management of slices can be more concise. As far as we know, there are few researches on the combination of the two technologies.

This paper presents a novel slicing scheme for radio access networks (RANs) based on control/user (C/U)plane separation.

In this paper, we present a novel scheme for RAN slicing based on C/U plane separation,and two main contributions are as follows:

· We propose a practical scheme of C/U plane separation for air interface by joint consideration of network functions, logical channels, physical channels, physical signals and user status. Meanwhile, we divide the conventional eNBs into two sub-eNBs,namely CeNB and UeNB respectively.

· We get the CeNBs and UeNBs virtualized,and abstract the computing, storage and radio resources. Specifically, we develop two groups of RAN slices, namely control-plane slices and user-plane slices, which are in charge of ubiquitous network coverage and data transmission at low and high frequency respectively. We implement a centralized controller to manage and allocate resources to the set of RAN slices.

The rest of this paper is organized as follows. We begin with an overview of the system model. Then we give a practical C/U plane separation scheme, and elaborate the design of two RAN slices. We show the experimental results in next section. Finally, the conclusion is given.

Fig. 1 The system model of RAN slicing based on C/U plane separation

II. SYSTEM MODEL

Figure 1 outlines the system model of RAN slicing based on C/U plane separation. With SDN/NFV and C-RAN, it consists of three layers: the virtual controller, the virtual CeNBs/UeNBs and the remote radio heads(RRHs) pool. We describe the key features of the system model as follows.

2.1 C/U Plane Separation

As the basis for implementing RAN slices,the air interface is designed to be C/U plane separated. We separate the control plane from the data plane so that a conventional eNB is divided into CeNB and UeNB. The CeNB is in charge of ubiquitous network access with large coverage, whilst the UeNB is responsible for supporting data transmission in a small range.

RRHs handle the transition between baseband signals and RF signals, and its specific functions are defined by the upper layer.Therefore, we can dynamically schedule RRHs to play the role of CeNBs or UeNBs by equipped low or high frequency antennas. In this way, we can ensure the large coverage for CeNBs and small range for UeNBs.

2.2 RAN Slicing

Based on virtualization, the virtual CeNBs and UeNBs could be realized by software defined running in VMs, and network/radio resources can be abstracted and shared among multiple applications. As shown in figure 1, we can allocate corresponding resources to different virtual CeNBs and UeNBs, to develop a set of logical RANs, referring to control-plane slices and user-plane slices. It is more convenient to update or add new functions and services.Note that, the control-plane slices are merely in charge of network access for users. We can develop various user-plane slices with different service types according to different user requirements, only for data transmission.Therefore, we can provide users with customized services.

In our design, the user completes the initial access under the coverage of control-plane slices at low frequency, and we can turn off the other user-plane slices apart from the control-plane slices to save resources and energy if there is no service request. When the user requests for service, the virtual controller will sense the request and then dispatch the user-plane slice to satisfy the user demand at high frequency.

2.3 Centralized control

Inspired by SDN, we consider the controller in the upper layer. The controller has a global view of the RANs so that it could choose appropriate slices and allocate required resources including computing, storage, spectrum, etc.according to different QoS requirements. In addition, it can adjust the allocation of resources among different slices dynamically.

The features mentioned above enable the deployment of RAN slices to be more flexible,scalable and easy to manage. We can formulate a set of isolated logical networks, RAN slices, on common physical infrastructure. Besides, based on C/U plane separation, we can allocate resources for each user-plane slice by the controller to provide customized services to users, without considering the redundant control signaling interaction. This makes the scheduling among the slices more simple and efficient thanks to centralized management of the virtual controller.

The combination of RAN slicing and C/U plane separation provides a promising direction towards 5G RANs. The specific designs are described as follows in detail.

III. C/U PLANE SEPARATION SCHEME FOR AIR INTERFACE

Due to the complicated coupling relationship among the physical layer channels and the characteristics of the frame structure, it is difficult to realize the extreme separation between control and user signals in physical layer. We should consider how to separate to ensure the transparency to users. Therefore, we jointly consider network functions, logical channels,physical channels, physical signals and user status, and then propose a practical C/U plane separation scheme of LTE at air interface,as shown in figure 2, also we could allocate CeNB at low frequency for large control coverage and allocate UeNB at high frequency for high-speed transmission in a small range . The specific separation scheme is as follows.

Firstly, from the aspect of user status, we can categorize the user status into two groups,inactive and active. In our design, the inactive users are only served by CeNBs, while the active users are severed by both CeNBs and UeNBs. Referring to [13], we consider the network functions in five aspects, synchronization, broadcast, paging, multicast and unicast.The primary task of inactive users, who are detached from the network, is to perform time/frequency synchronization and then acquire the system configurations from the target cell by receiving the system broadcast information. Inactive users will also receive multicast or paging message in low-power mode. Thus,CeNBs should offer inactive users synchronization, broadcast, paging and multicast.Relatively synchronization and unicast are necessary for UeNBs to provide active users with data transmission.

After network functions separation, logical channels can be separated according to the mapping among them. In accordance with LTE Release 8, we reserve Broadcast Control Channel (BCCH), Common Control Channel(CCCH), Paging Control Channel (PCCH),Multicast Traffic Channel (MTCH) and Multicast Control Channel (MCCH) at CeNBs,which are in charge of network access. While the Dedicated Control Channel (DCCH) and Dedicated Traffic Channel (DTCH) are left at UeNBs to cope with data transmission.

For physical channels, it should be noted those Synchronization Channel (SCH), Downlink Reference Signal Channel (DL-RS),Physical Control Format Indicator Channel(PCFICH) and Physical Downlink Control Channel (PDCCH) are owned by CeNBs and UeNBs simultaneously due to the frame structure of LTE FDD. They determine whether radio frames can be received or decoded successfully. In addition, Physical Multicast Channel (PMCH) and Physical Broadcast Channel (PBCH) are reserved by CeNBs,whilst Physical Downlink Shared Channel(PDSCH) is left at UeNBs. To guarantee the separation at eNBs is transparent to users, the frame structure should not change regardless of control signaling or data information.

Fig. 2 The C/U plane separation scheme

At last, the separation of physical channels naturally leads to the wireless signals separation. Among all wireless signals, synchronization, frame control and pilot signals are essential for both CeNBs and UeNBs to estimate the wireless channels and assist decoding.Besides, CeNBs also hold broadcast and multicast signals, whilst UeNBs hold data signals.

In general, compared with the two-layer network functionality separation scheme in[13], and the improved separation scheme in [14], we jointly consider the separation scheme from five aspects, which is more practical by separating in this fine-grained way.

IV. RAN SLICING BASED ON C/U PLANE SEPARATION

As two drivers for RAN slicing, SDN and NFV can realize the more flexible and scalable network. Therefore, we combine SDN and NFV to design the RAN slices. Meanwhile,based on the C/U plane separation, we implement two groups of slices, that control-plane slices and user-plane slices.

4.1 Control/user-plane slices

In our design, we employ NFV technique to create virtual network functions (VNFs), such as virtual CeNBs and UeNBs. The VNFs work on the general physical platform independently and provide various APIs for different network operators. Besides, the network and radio resources can be abstracted by virtualization and allocated to any VNF according to the requirements. Combining different VNFs with corresponding resources, we can get a set of logical RANs, referring to RAN slices.

Based on the proposed C/U plane separation scheme, we design the control-plane slices and user-plane slices respectively. The control-plane slices have the characteristic of continuous larger coverage and are responsible for interaction of control signaling. Relatively,the user-plane slices are similar to the small cell for increasing network capacity, which can be scheduled or turn on/off dynamically to offer data services merely instead of considering coverage continuity.

There can be multiple user-plane slices under the coverage of one control-plane slice.Taking into account the shortage of spectrum resources and the interference between control-plane slices and user-plane slices under the same frequency networking, we arrange control-plane slices in low frequency band,which concern continuous coverage, while allocate user-plane slices in high frequency band for facilitating capacity expansion. Compared with high frequency, low frequency suffers less severe transmission loss, thus a wide range of coverage can be achieved. On the other hand, high frequency band has more available spectrum resources. These features are consistent with our design principles for control/user-plane slices.

4.2 Virtual controller

To implement RAN slices, a centralized controller is requested to abstract and manage the overall network resources and functions.A controller can manage multiple CeNBs and UeNBs, while a CeNB or UeNB can only connect to one controller. Referring to SDN,the controller ought to own global view of the network and schedule or allocate appreciate resources to diverse slices. Therefore, the control strategy mainly includes the following points:

· Real-time monitoring of network status.

· For the creation and deletion of slices.

· According to the status information and users requests to optimize and schedule network/radio resources from a global perspective for each slice.

· Ability to support multiple wireless protocols in heterogeneous network.

V. EXPERIMENTS

A hardware testbed assisted by SDR is set up for feasibility analysis of the proposed RAN slicing scheme. We adopt srsLTE, an open source software, to implement physical layer transceiver functionality aligned with LTE Release 8 and Universal Software Radio Peripheral (USRP) as the hardware platform.The experimental scenario is shown in figure 3, which consists of software defined virtual controller, virtual eNBs (CeNBs or UeNBs),RRHs an UE.

Specifically, the virtual controller, running on VMware, communicates with the virtual CeNBs or UeNBs via virtual switch with socket. The virtual eNBs include one CeNB for control signaling interaction and multiple UeNBs undertaking different service types,which run on different VMs respectively. They connect with USRP N210, which act as the RRH, via Gigabit Ethernet cable. In addition,UE is also software defined and equipped with USRP N210 in the same way.

Fig. 3 Testbed

For the sake of universality, we design two user-plane slices over the testbed which bearing audio and video service by two virtual UeNBs respectively, namely audio slice and video slice. The virtual controller can allocate corresponding computing, storage and radio resources for different slices depending on their service types. It is worth mentioning that the virtual CeNB can also be seen as a slice,which is responsible to users for accessing to network with corresponding resources. In our test, we allocate the control-plane slices at 1.8GHz with 5M bandwidth, while the user-plane slices at 3.6GHz with 5M bandwidth.

Figure 4a and figure 4c dedicate the CPU share, whilst figure 4b and figure 4d show the rate of video slice and audio slice. We can see that before being dispatched to serve users,both user-plane slices are in the closed state and do not occupy any resource, while the control-plane slice is always on to maintain the network coverage. When there are highrate video service requests from users, the video slice is turned on by the virtual controller. The results show the rate of video slice is about 6Mbps, which is consistent with the user demands for smooth viewing experience.The audio service requests are handled in a similar manner. The rate of audio slice reaches 3Mbps, and the CPU share is lower than the video slice, which is confirming to our desgn.

The results show that it is feasibility to deploy different slices according to different service requirements based on the proposed scheme. The RAN slices we designed can guaranteed the user demands well. Furthermore, the deployment of slices becomes more simple thanks to the C/U plane separation, that is user-plane slices can be created and managed by the centralized virtual controller and without considering the complicated interaction of control signaling.

VI. CONCLUSION

In this paper, taking into account the correlation between the network slicing and C/U plane separation at air interface, we have presented a novel RAN slicing scheme based on C/U plane separation. Further, we have proposed C/U plane separation by dividing eNBs into two sub-eNBs called CeNB and UeNB,for transmitting control data and user data at the low and high frequency respectively. In addition, we have developed two groups of RAN slices, namely control-plane slices and user-plane slices respectively. Finally, the experimental results have shown the feasibility that the proposed scheme can satisfy the user demands appropriately.

ACKNOWLEDGEMENTS

This work was supported in part by National Natural Science Foundation of China(61372070), Hong Kong, Macao and Taiwan Science & Technology Cooperation Program of China (2014DFT10320), and the 111 Project (B08038).

[1] CISCO VISUAL NETWORKING INDEX: Global mobile data traffic forecast update, 2015–2020.White Paper, 2016.

[2] A. Al-Fuqaha, M. Guizani, M. Mohammadi et al., “Internet of things: a survey on enabling technologies, protocols, and applications,” IEEE Communications Surveys & Tutorials, vol. 17,no. 4, 2015, pp. 2347-2376.

[3] NGMN Alliance, “Description of Network Slicing Concept,” Available: https://www.ngmn.org/uploads/media/160113_Network_Slicing_v1_0.pdf,accessed Jan. 2017.

[4] 3GPP, 3GPP TR 23.799, Study on architecture for next generation system, release 14, version 7.0.Available: http://www.3gpp.org/DynaReport/23799.htm, 2016.

[5] B. A. A. Nunes et al., “A Survey of Software-Defined Networking: Past, Present, and Future of Programmable Networks,” IEEE Commun.Surveys & Tutorials, vol. 16, no. 3, 2014, pp.1617–1634.

[6] C. Liang and F. R. Yu, “Wireless Network Virtualization: A Survey, Some Research Issues and Challenges,” IEEE Commun. Surveys & Tutorials,vol. 17, no. 1, 2015, pp. 358–380.

[7] J. Ordonez-Lucena, P. Ameigeiras, D. Lopez et al., “Network Slicing for 5G with SDN/NFV:Concepts, Architectures, and Challenges,” IEEE Communications Magazine, vol. 55, no. 5, 2017,pp. 80-87.

[8] X. Zhou, R. Li, T. Chen et al., “Network slicing as a service: enabling enterprises’ own software-defined cellular networks,” IEEE Communications Magazine, vol. 54, no. 7, 2016, pp. 146-153.

[9] Open Networking Foundation, Tr-526: Applying SDN architecture to 5G slicing. Available:https://www.opennetworking.org/images/stories/downloads/sdn-resources/technical-eports/Applying SDN Architecture to 5G Slicing TR-526.pdf, 2016.

[10] A. Nakao, P. Du, Y. Kiriha et al., “End-to-end network slicing for 5G mobile networks,” Journal of Information Processing, vol. 25, 2017, pp. 153-163.

[11] S. Katti and L. Li, “RadioVisor: A Slicing Plane for Radio Access Networks,” Proc. ACM 3rd Wksp.Hot Topics in Software Defned Networking, 2014,pp. 237–238.

[12] A. Ksentini, N. Nikaein, “Toward Enforcing Network Slicing on RAN: Flexibility and Resources Abstraction,” IEEE Communications Magazine,vol. 55, no. 6, 2017, pp. 102-108.

[13] X. Xu, G. He, S. Zhang et al., “On functionality separation for green mobile networks: concept study over LTE,” IEEE Communications Magazine, vol. 51, no. 5, 2013, pp. 82-90.

[14] T. Zhao, L. Wang, X. Zheng et al., “Hycell: Enabling green base station operations in software-defined radio access networks,” Proc. IEEE International Conference on Communication Workshop, 2015, pp. 2868-2873.