Na Li，Guangyi Liu，Huimin Zhang，Quan Zhao，Yun Zhao，Zhou Tong，Yingying Wang，Junshuai Sun

Future Research Lab，China Mobile Research Institute，Beijing 100053，China

*The corresponding author，email: linawx@chinamobile.com

Abstract: The convergence of information，communication，and data technologies (ICDT) has been identified as one of the developing trends of the sixth generation(6G) network.Service-based architecture(SBA) as one of the promising information technology，has been preliminarily introduced into the fifth generation (5G) core network (CN) and successfully commercialized，which verifies its feasibility and effectiveness.However，SBA mainly focuses on the control plane of CN at present and the SBA-CN user plane is being studied by the industry.In addition to further evolving the SBA-CN，SBA radio access network(RAN)should also be considered to enable an end-toend SBA，so as to meet more comprehensive and extreme requirements of future applications，as well as support fast rollout requirements of RAN devices.

Keywords:6G;micro-service;SBA-RAN

I.INTRODUCTION

To meet the requirements of diversified services on the flexibility and scalability of the network，the softwarization，virtualization and cloudification of network has become an inevitable trend[1].Cloud RAN has also been widely favored by global operators.

Cloud RAN helps operators build low-cost，resource-efficient，green，fully automated radio networks.It has two core features that lay the foundation for successful deployment of cloud RAN:

· Common hardware platforms(including accelerators):The capabilities of traditional dedicated hardware base station(BS)are limited by the capabilities of hardware.A common approach to improve the signal processing capacity of BS is to supplement the baseband processing boards.Additional hardware boards will not only increase equipment costs，but also increase power consumption which is heavily affected by the number of boards and is less related to the work load.Different from traditional dedicated hardware BSs，RAN functional software of cloud RAN is carried on the commercial-off-the-shelf(COTS)platform in cloud RAN，which is facilitated for the sharing of hardware resources and the on demand deployment of functional software.It is expected that the equipment cost will be reduced by 40%and operation cost by 30%[2]，and the power consumption of each BS is expected to be saved by about 5%in typical scenarios.

· Cloud native technology:In order to meet the user needs，equipment manufacturers have to constantly upgrade the software version of BS.Currently，the upgrade is cumbersome，timeconsuming，error-prone，and prone to introduce service interruptions.Differently，in cloud RAN，with cloud native technologies such as Kubernetes and the principle of DevOps，the RAN functions can be deployed in containers on bare servers，leading to a faster version upgrade and application online.

Equipment vendors are actively developing the cloud RAN products，and the performance differences of the products with dedicated hardware are gradually narrowing.As early as 2016，the world’s first cloud RAN system developed by Nokia is commercialized in South Korea，which has been large-scale commercial deployed in mid-2019.The second generation of the system with virtualized centralized units(CUs)and virtualized distributed units(DUs)is in trial and will be commercialized in 2022.Ericsson has also launched a series of cloud RAN products and is leveraging Intel Xeon processors and other technologies to improve cloud RAN performance for 5G and beyond.During the same period，Samsung deployed the industry’s first fully virtualized end-to-end(E2E)commercial 5G RAN，demonstrated that the virtualized RAN solutions are capable of processing massive amounts of data and its performance can achieve the same level of traditional dedicated hardware products.

Virtualized RAN and cloud RAN are not the ultimate goal of future radio networks.In our understanding，the soft network [3] or SBA-RAN would be the next step of cloud RAN.Cloud RAN has provided a flexible and scalable platform for future radio networks，but its service capabilities remain limited:From the functional point of view，the minimum development granularity of building base band unit(BBU)is CU or DU，which is relatively large and still cannot meet the requirements of rapid online and flexible deployment of specific new functions.From the perspective of interface，point-to-point dedicated interfaces are still used within BS，between BSs，and between BS and CN.Whenever the BS or CN network functions (NFs) change，the corresponding modifications have to be made on related interfaces，resulting in high standardization workload and high complexity in operation administration and maintenance(OAM).

In order to respond nimbly to more diversified function requirements，quality of service (QoS) requirements，management strategy requirements，deployment requirements and exposure requirements of the future applications and scenarios，and make the network be capable of forward compatibility，it is necessary to make efforts to improve the service capability of the next generation of wireless access network so as to better leverage the platform advantages of cloud RAN.

The study of SBA-RAN includes the definition of RAN services，the interface design between RAN services，the interface design between RAN services and CN services，the interface design between RAN services and the third parties，the construction of cloud native infrastructure，and the orchestration and management of the functionalities and resources[4].

The study of SBA-RAN has begun，but there has not much research progress.A study item(SI)on servicebased N2 has been discussed on 3GPP SA2 [5]，but the research work has been postponed.The main idea of the SI is to introduce a new service-based N2 interface between NG-RAN and CN，while retaining the traditional N2.The messages related to UE mobility and NAS signaling are transmitted through the traditional N2 interface，and other messages are transmitted through the new service-based N2 interface.Another study on UPF enhancement for exposure and SBA has been approved recently [6].There are two main objectives for the study item.One is to study UPF event exposure services registration/ deregistration，and discovery via the network repository function (NRF).Another is to study UPF event exposure services that would support.In [7]，a RAN service engine is proposed that allows services to customize and extend RAN functionalities using containerized micro-services，but the relevant RAN services are not defined.In [8]，only five general RAN functions are provided under the SBA framework，i.e.physical layer enforcement function，physical layer control function，user plane (UP) enforcement function，UP control function，and radio connection control function，which cannot meet the customization requirements of future applications.

This paper gives our preliminary views on the design of SBA-RAN，and in our understanding，it is the first paper to consider SBA-RAN from a protocol stack perspective.The contributions include:

· RAN control plane (CP) services are deduced from the protocol stack perspective under the proposed guidelines;

· System operations between the CP services are defined;

· Fundamental problems when designing SBARAN-UP are raised and potential solutions are given;

· Possible performance advantages of the proposed SBA-RAN are detailed analyzed;

· The application scenarios of SBA-RAN for vertical industries and personal users are given;

· The challenges of SBA-RAN are detailed discussed.

The rest of this paper is organized as follows.Section II will analyze the overall design principles.Based on the principles，a possible scheme of SBARAN-CP and its key technical advantages are provided in Section III.Similarly，Section IV provides a possible scheme of SBA-RAN-UP along with its key technical advantages.Further，the application scenarios of SBA-RAN are given in Section V.Section VI detailed analyzes the potential chanllenges of SBA-RAN and Section VII concludes the paper.

II.THE GENERAL DESIGN PRINCIPLE OF SBA-RAN

Before designing SBA-RAN，it is necessary to understand the concept of micro-service and determine the guiding principles of service definition，to avoid serious problems caused by unreasonable service definition.

At present，there is no unified definition of microservices in the industry.The most accessible description of micro-service is from Martin Fowler:The micro-service architectural style is an approach to develop a single application as a suite of small services，each running in its own process and communicating with lightweight mechanisms，often an hyper text transfer protocol (HTTP) resource application programming interface(API).These services are built around business capabilities and independently deployable by fully automated deployment machinery.There is a bare minimum of centralized management of these services，which may be written in different programming languages and use different data storage technologies.

When designing RAN services，high cohesion and loose coupling should be carefully considered.If a service has several responsibilities，there are several corresponding reasons to change this service.Therefore，a small and cohesive service with a single responsibility is required，which will reduce the size of the service and improve its stability.Loose coupling is a more important feature of SBA.If too much content is exposed by the service provider，the coordination between the service provider and consumer will be frequent，which will reduce the autonomy of the service.

III.CONTROL PLANE SCHEME OF SBARAN

3.1 Definition of Control Plane Services

When redefining the CP services，the design principles proposed in Section II such as high cohesive and loose coupling as well as the relationships between radio resource management(RRM)modules should be kept in mind.

Services are defined in response to external requirements.For RAN services，the demands may come from CN NFs，other RAN nodes，or UEs.The CP interface protocol between RAN and CN is NG application protocol(NGAP)[9].Therefore，we can extract the service capabilities of RAN for CN requirements from the NGAP procedures.Similarly，we can explore other RAN service capabilities from the Xn application protocol (XnAP) interface protocol between the BSs [10] and from the radio resource control (RRC)interface protocol between the BS and UE[11].

After redefining，the RAN services should include radio bearer management service (RBS)，connection mobility management service (CMS)，local location service (LLS)，multicast broadcast service (MBS)，data collection service (DCS)，signaling transmission service (STS)，paging service (PS)，RAN exposure service (RES)，and random access service (RAS)，as shown in Table 1.By separating bearer management and connection mobility management into different services，the further separation of data and control at the access network is realized，which will help to realize a more flexible end-to-end data control.Different from the traditional peer-to-peer dedicated interface，the interfaces between these services will apply a lightweight protocol such as RESTful/HTTP，which is beneficial for rapid interface development，rapid upgrade，and the network capabilities exposure.

Table 1. RAN services and service operations.

a)RBS

RBS is primarily used to manage UP operations and can therefore be an independent service.The radio bearer context that was previously in the UE context will also be managed by the RBS，for example this context should be stored through a dedicated database controlled by the RBS itself.At the same time，the exposure of the radio bearer context related information should also be controlled by the RBS.

Access and mobility management function (AMF)can request the RBS to create，modify，or release(part of) UE context，which may be security key，mobility restriction list，UE radio capability，UE security capa-bilities，etc.The CMS and neighbour RBS can also make a similar request in handover(HO)scenarios or in dual-connectivity scenarios.The status information such as the results of connection establishment can be directly notified to the relevant service consumers and no pre-subscription is required.Other services，such as RES and session management function(SMF)，can also subscribe the notification of radio bearer modification.

CreateRBContext is used to create an association between CMS and RBS.The CMS provides the necessary information to establish the radio bearers，such as UE identity(ID)，CMS ID.RBS will response with the RB context ID and possibly the QoS profile.CMS can also request to modify or release the RB context through UpdateRBContext and ReleaseeRBContext requests.ContextPush can be requested by the OAM to transfer an RBContext into another RBS.

b)CMS

Traditionally，UE context management is essentially a database that contains and manages all UE related radio access information，such as UE radio bearer context，UE radio capabilities，restricted list of access areas，and it is tightly coupled with connection and mobility management and UE radio capability management.A change in UE RRC connection or mobility will inevitably lead to a change in UE context.So these three parts of RAN capabilities are better to be formed into a single service: CMS.More specifically，the responsibility of CMS generally includes three aspects，i.e.，UE context management，mobility management and multi-connectivity management.

AMF or CMS can initiate the creation，modification，release，or the release trigger of UE context during the initial access，HO，or multi-connectivity procedure.Similar to RBContextPush，UEContextPush can also be triggered by the OAM and is used to transfer a UEContext to another CMS.The suspension or resumption of the UE context can be requested by the AMF that wishes the UE to be in RRC_inactive state.With this request，the RAN will suspend or resume the UE logical NG connection while keeping the UE context on the RAN.UEContextTransfer is primarily used in the HO procedure when the target CMS requests the UE context from the source CMS.When the AMF requests the RAN to verify whether the UE radio capabilities are compatible with the Internet protocol (IP)multimedia subsystem (IMS) voice configuration，it will use UEContextInfoCheck.AMF or other services may use UEContextInfoDiscovery to request part of UE context information，such as UE status information.HOPreparation is used when AMF or source CMS requests the target CMS to reserve resources for a UE，and HOCancel is used when they wish to cancel their HO request.Other services may use HOInfoSubscribe to subscribe HO related information.For example，the AMF or source CMS may subscribe the information whether UE has been successfully HO to the target cell.

SNaddition/SNchange/SNrelease is primarily used by the master-CMS to request the secondary-CMS to reserve/ modify/ release resources for the UE in dual connectivity.InfoNotify is used for example to notify secondary-CMS whether UE has successfully applied the requested configuration.

The source CMS can use RRCMessagetransfer service operation to request the target CMS to forward an RRC message to the UE，which is similar to the NR RRC message encapsulated in an LTE RRC message.TraceActivate/ Deactivate is used by the master CMS to activate or deactivate the trace of the secondary CMS in multi-connectivity.

c)LLS

LLS is mainly used to provide UE location information，such as geographic location，positioning method used，positioning failure cause.

EventNotify is used to inform the UE location when a particular event occurs.Events can be (implicitly)subscribed by the service consumers in advance.ProvideLocationInfo is used to provide the location of UE.The service consumer may also use CancelLocation to cancel the location request for a specific UE.

d)MBS

MBS is used to provide multicast and broadcast service to broadcast system information or multicast business data.Transfer and cancel are used for requesting to perform or cancel a multicast/broadcast data transmission.Resource and QoS requirements can be carried in the request.StatusNotify is used to inform the transmission status，such as whether a multicast/broadcast message has been successfully received by the UE.

e)DCS

DCS is used to obtain the measurement results from RAN or UE.ActivateTrace and DeactivateTrace can be requested by the service consumer to activate or deactivate a trace session for a UE.StatusNotify is used to inform the results of activation or deactivation.

f)STS

STS includes the transport of non-access stratum(NAS) messages and non-UE messages.Considering RAN is only a transparent pipeline for transmission which is different from other RAN services，so this function is defined as an independent service.

Subscribe-notify and transfer are two forms of NAS message transmission，where the NAS messages include NAS-mobility management messages，NASsession management messages，long term evolution positioning protocol messages，and short messages.NASNonDeliveryNotification is used for informing the transmission status of these messages.NASMessageRoute is mainly requested by the AMF for rerouting the initial UE messages to other AMFs.

g)PS

PS is used to ensure the reachability of the UE，or to transfer some small control signaling/data to the UE.

h)RES

RAN capabilities are exposed through this service.

Considering the interactions between BS and UE，another two RAN services may be introduced.

i)Radio connection control service

This service helps UE maintain RRC connection，including RRC establishment，RRC reconfiguration，RRC connection recovery，RRC release，RRC connection status notification，RRC connection reject notification.Since this service shares the same database with CMS，it is better to integrate it into CMS.

j)Random access service

RAS is used to help UE to access network，or access information(such as system information)，or complete uplink synchronization.

In addition to these basic functional services，RAN CP can also provide some extended services such as data storage service and intelligent data analysis service.

3.2 Key Technical Advantages of SBA-RANCP

Technical advantage 1:RAN services and CN services can directly communicate with each other，which reduces unnecessary forwarding through AMF.

In order to reduce the interface management work and interface protocol design work，dedicated interfaces and communication paths are assumed between traditional network elements，and such interfaces are not defined between any two network elements.For example，the RAN CP can directly communicate with only one CN NF (i.e.，AMF)，in other words，there is only one interface between RAN CP and CN.If RAN CP wants to communicate with other CN NFs，the messages have to be first transmitted to AMF，even if some messages do not need to be processed by the AMF.Such dedicated interface will not only increase the work of standardization，but also lead to complex procedure and long latency.

Taking the location procedure as an example.There is no direct interface between gNB and location management function (LMF)，and the relevant data needs to be transferred through AMF.Accordingly，the communication between UE and LMF also needs to go through the gNB and AMF，which in many cases are only responsible for transparent data transmission.Unnecessary data forwarding will inevitably bring transmission latency，leading to a long location estimation latency and making AMF become the bottleneck of data transmission.

In SBA-RAN CP，RAN related service，i.e.，DCS，can directly communicate with LMF through lightweight interface，without the needing of AMF forwarding like 5G.The steps of positioning procedure are reduced and thus the positioning latency is saved.

Technical advantage 2:Serial processing parallelization improves processing efficiency.

With SBA-RAN-CP，some procedures can be performed in parallel from serial，thereby processing efficiency can be improved.

Taking the multicast procedure as an example.The multicast related RAN services mainly include CMS，RBS，MBS and resource allocation service (this service may be a user plane service).During parameter configuration in Figure 2，the MBS can directly communicate with the other three services.For example，the MBS will send the multicast cell list，time/frequency resource requirements，and QoS requirements of the multicast service to the resource allocation function.It will also request RBS to establish the multicast radio bearers.Then it may request CMS to inform the UE of the multicast identity，key，radio resource allocation indication，start time of multicast reception，multicast radio bearer configuration，unicast or multicast bearer modification/suspension indication，etc.，via RRC dedicated signaling.With these modifications，the traditional serial-performed configuration operations can be converted into operations that can be performed in parallel，which improves the processing efficiency.

Figure 1. RAN services overview.

Figure 2. Illustration of the MBS procedure in SBA.

IV.SCHEME OF SBA-RAN-UP

4.1 User Plane Service Definition

Traditional mobile communication protocols follow the open system interconnect (OSI) reference model，as shown in Figure 3.Each layer receives specific services provided by its lower layer and is responsible for providing specific services to its upper layer.The interaction between upper and lower layers follows the “interface” convention，and the interaction between the same layer follows the “protocol” convention[12].

Figure 3. The“protocol”and“interface”in OSI reference model.

The same principle also applies to the design of 5G RAN，as shown in Figure 4.For example，RRC and service data adaptation protocol (SDAP) as the upper layer protocols consume the services provided by packet data convergence protocol (PDCP)，which are transfer of control plane data service，transfer of user plane data service，header compression service，encryption service，and integrity protection service.At the same time，PDCP as the upper layer of radio link control(RLC)，consumes the services provided by RLC，which are transparent mode(TM)data transfer，unacknowledged mode (UM) data transfer，acknowledged mode(AM)data transfer[13-16].

Figure 4. The“services”defined in 5G RAN.

The problem of this layered architecture is that the fixed protocol and service model restrict flexible crosslayer signaling interaction and cross-layer function combination.Therefore，we propose to use microservice-based approach to modularize UP functions and combine them flexibly to better meet diverse application needs.SBA-RAN-UP essentially aims to break through the traditional cured layered protocol architecture，so that the access relationship between functions is unlimited instead of only restricted to the upper layer and lower layer.

To break through the traditional layered protocol design，the key technical problem to be solved is the definition of UP services.Each functional module in RAN is relatively independent，which can basically be considered as a service.The SBA-RANUP services may include data service adaptation service，header compression service，header decompression service，integrity protection service，packet duplication service，retransmission service，segmentation service，AM transmission service，UM transmission service，multiplexing service，reordering service，etc.

The value of SBA-RAN is that it can enable free combination of functional services on demand，as shown in Figure 5.Therefore，another key issue to solve is to determine an optimum processing order.In our view，there are two possible ways:

Figure 5. Example of SBA-RAN-UP.

Option 1:The service consumer dynamically selects the service provider instance，in other words，the current service selects the next-hop service.

All instances of the service provider or the next-hop service and their addresses can be obtained from NRF，which is consistent with 5G mechanism.Based on the information from NRF，the service consumer or the current service will first conduct preliminary screening of service provider instances under its requirements such as required network slice.Then it will choose the most appropriate service provider instance from the remaining ones considering the instance capacity and load.

In this option，all services on the processing chain need to obtain the association between the processing chain ID and service combination in advance.This association information can be maintained and provided by a specific service，referred to as the processing chain management service (PMS).When the service consumer receives the packet with a specific processing chain ID，it will infer directly which service is the next-hop service，and then select the most appropriate instance of the next-hop service and forward the packet to this instance.The advantage of this option is that it can realize load balancing between service instances in real-time.

Option 2:The PMS is responsible for determining the combination of service instances and their association corresponding to the data processing chain.

The association between the processing chain ID and the service instance combination is determined and sent to the corresponding service instance by the PMS.Different from option one，all the next-hop service instances on the processing chain have been determined by the PMS before data transmission，so it is unnecessary for the current service to select the nexthop service instance.Compared to option one，this option provides a lower data processing latency.

4.2 Key Technical Advantages of SBA-RANUP

Technical advantage 1:Enable data preprocessing which could reduce the signaling overhead.

In the 5G and earlier networks，each packet needs to carry an Internet protocol(IP)header and a transmission control protocol (TCP)/ user datagram protocol(UDP)/ stream control transmission protocol (SCTP)header.Sometimes the header takes up more bytes than the size of the packet itself，for example，IPv4 header takes up 20 bytes，IPv6 header 40 bytes，TCP header 20 bytes，UDP header 8 bytes.

In order to reduce the signaling overhead of Uu interface (the interface between UE and base station)，IP header compression is introduced.However，some specific algorithms and operations are required for header compression，which increases the complexity of data processing and the processing delay.With SBA-RAN，a service similar to network address translation can be introduced in RAN to replace the lengthy IP address and port number with a simpler application ID.For downlink data transmission，this service converts the information such as IP address in the packet header into application ID and sends it to UE.For uplink data transmission，the IP address and port number information are no longer necessary in the packets，for they can be replaced by a simpler application ID.The application ID will then be changed by the RAN service to the traditional IP address before the packet is forwarded to the CN.The mapping relationship between application ID and IP address & port number can be configured individually for each UE in advance.This operation requires the lower layer to be capable of clipping the TCP/IP header，which is not possible in the traditional OSI model.

In addition to tailoring and modifying the packet header，the BS can also act as the agent of UE to carry out some operations，such as TCP checksum.The purpose of TCP checksum is to avoid packet corruption caused by router failure during data transmission.TCP checksum is calculated by the transmitter and verified by the receiver.If the receiver detects an error in the checksum，the TCP segment will be simply discarded.The information used for TCP checksum includes TCP headers and TCP data，as well as 12 bytes of pseudo-header(as shown in Figure 6).

Figure 6. TCP pseudo-header(in light blue)for checksum computation.

Tranditionally，the UE and application server excute the TCP checksum operations.While with SBARAN-UP，a TCP checksum service can be introduced in the RAN to perform such opertion on behalf of UE.For downlink data transmission，the BS can verify the checksum on behalf of UE.If the validation is passed，the checksum field can be removed from the TCP header，ending up saving some overhead.For uplink data transmission，the checksum calculation can be skipped by the UE and be performed by the BS，which will also reduce the overhead of the Uu interface.

Technical advantage 2:Optimize the header composition to avoid header information redundancy.

A typical cross-layer issue is congestion control，where the congestion control is performed at the TCP layer while congestion detection occurs at the IP layer.In order to solve the congestion control problem，a collaborative optimization scheme of TCP and IP which is called explicit congestion notification(ECN)is proposed.

The ECN implementation procedure is as follows.An ECN-capable TCP host sends a TCP segment with an ECN field of 10 or 01 at the IP header over an ECNcapable TCP connection.Routers that support ECN set the ECN field at the IP header to 11 when they detect congestion.When the TCP receiver receives a TCP segment with an ECN field of 11，it will set the ECE(ECN-ECHO)flag in the TCP header.When the TCP transmitter host receives an acknowledge(ACK)with the ECE flag，the transmitting window begins to reduce.On the next packet，the transmitter host sets the CWR flag to indicate the transmitting window has been reduced.Upon receiving a packet with the CWR flag，the receiver will stop setting the ECE flag in the following ACK.

To make the ECN mechanism possible，both the TCP layer and the IP layer need to introduce congestion indicator fields，namely ECE field at the TCP layer and ECN field at the IP layer，as shown in Figure 7.These two parts indicate the same information，but they are essentially required in the layered protocol system.With SBA-RAN-UP，only one field is a must and the information of that field can be exposed to all other services，thus avoiding the problem of header information redundancy.For example，the ECN field of 10 can be used to indicate that a congestion is experienced on the go path，and 11 to indicated that congestion is experienced on the return path.

Figure 7. ECN field in IP header and ECE field in TCP header.

Technical advantage 3:Maximize the value of data and make data processing operations be visible.

In the OSI model，the value of data is limited to the layer to which it belongs.For example，RLC status report will only be used at RLC layer.When the transmitting RLC entity receives a status report from the receiving RLC entity，it will know which packets have been received successfully and which packets need to be retransmitted.If the transmitting TCP layer wants to know whether the TCP layer packet has been received successfully，the receiver needs to confirm it with a TCP feedback.

In fact，however，there is an association between the RLC feedback and the TCP feedback.When receiving the RLC status report from the receiver，the transmitter can infer which TCP packets have been received successfully according to the mapping relationship between TCP packets and RLC packets，thus reducing the signaling overhead and reducing the TCP RTT time.Reducing TCP round-trip time can effectively guarantee TCP throughput performance[17].

It is difficult to realize such cross-layer information association in the traditional layered OSI model，but it can be realized well in SBA-RAN by simply adding“sequence number association” service in the transmitter side.

In addition to maximizing the value of data，SBARAN will also realize the traceability of data processing.

In the traditional OSI model，the processing of each layer is only visible to the layer itself.For example，after being processed by the RLC receiving entity on distributed unit (DU)，the RLC service data units are directly delivered to the PDCP entity of central unit(CU)，but the PDCP entity does not know what has been done by the RLC entity.Similarly，when the data is further delivered to UPF，the UPF also does not know what the BS has done to the data.From an end-to-end point of view，it is difficult for the operator to know the root cause of a data processing exception due to the lack of intermediate processing operation records.SBA-RAN-UP provides a possibility for recording intermediate processing operations.Operators can request the intermediate services to record their operations in the header or interface message so that operators can know what has happened in the process.

Technical advantage 4:Speeding up the data processing process by removing unnecessary processing steps.

The processing module functions in UPF are as follows [18].Link and IP forwarding (FWD) modules are mainly used for packet forwarding.The Security module implements security protection functions such as packet verification，distributed denial of service，anti-spoofing，which are used to check the validity of the packets and timely discard the packets that do not pass the verification.The general packet radio service tunneling protocol (GTP) is used to establish the GTP tunnel with the BS and is responsible for GTP signaling processing，GTP packet encapsulation/ de-encapsulation，etc.DPI is used to analyze the traffic based on the protocol identification library，which needs to update in time.According to the classification results of packets，QoS module makes differentiated packet processing to different UE categories or application categories，to provide faster and more reliable forwarding service for high-priority packets.Charging module makes statistical charging based on the classification results of packets.Since the statistical dimensions and granularity of charging are directly related to the applications，the logic function of charging is complicated.

The logic of the link，IP FWD，security，GTP，rule lookup，and QoS modules are relatively simple，while that of the charging and DPI modules are more complicated.With SBA-RAN-UP，some application information may be carried directly in the packet header，so the DPI module which is mainly used for traffic classification may not be needed.Of course，the DPI can still be retained for other analysis use，but it is no longer a must in data processing，thus reducing the data processing time.

Figure 8 shows the hardware acceleration model of UPF[19，20].Currently，modules with relatively simple logic are implemented by hardware acceleration，while the logic-complicated charging and DPI modules are implemented by software in the central processing unit (CPU).After receiving the data stream，the hardware acceleration card will extract N-tuple and query in the flow table to determine whether there is a matching table entry.If not，the packet will be sent to the CPU，which will process the packet in a slow mode.After the processing of security，GTP，and DPI，the CPU will decide whether the data stream needs to be offloaded to the hardware acceleration card.If needed，a flow table is generated and configured to the hardware acceleration card.The subsequent packets will match the entries in the flow table，and then be associated with the rules such as QoS and charging according to the bearer table.With these，the hardware acceleration card completes the fast processing of the packets.

Figure 8. UPF hardware acceleration.

If application information can be carried directly through a packet header，it may no longer be necessary to obtain the offloading strategy by uploading the first one or more packets to the CPU to generate a flow table.Configuration table entries are available in advance，which makes it possible to remove the offloading strategy generation step in the hardware acceleration procedure，therefore the data processing can be further accelerated.

Technical advantage 5:Provide more refined QoS management for 6G typical applications.

Holographic communication has become the representative application of 6G.The collector of holographic communication is a group of cameras，and the number of cameras and the resolution of collection directly affect the results of holographic display.In order to meet the increasing demands of users for high resolution and high fidelity display quality，the number of cameras and the resolution will continue to expand，and correspondingly the data volume will also continue to rise.It is obvious that high throughput and multi-source data composition are the two representative characteristics of holographic communication.The importance of each data source to the application is different，and the importance of different frames in each data source is also different，which is determined by the application itself.

Like conventional video streaming，the data generated by each view point of holographic communication is compressed into three types of video frames: I frame，P frame and B frame.I frame is the key frame that requires to be lossless transmitted，while P frame and B frame can be lost to some degrees which depends on the recovery ability of application.

Different E2E QoS guarantee mechanisms should be provided for different video frames.For example，the key part of the video frame，i.e.，I frame，can be extracted as a higher-priority stream，and B/P frame stream as a lower-priority stream.With SBA-RANUP，the network will easily differentiate between the data according to the application information carried in the packet header，such as type，importance and priority.Besides，various QoS requirements for different data streams can also be carried along with the packets and these QoS requirements are comprehensible for all the services on the processing chain.With this，the network could in real-time adjust its QoS polices.

V.APPLICATION SCENARIOS OF SBARAN

5.1 On Demand Providing Necessary Services for Vertical Industries

The traditional BS has comprehensive functionalities，such as allowing the interoperability between 5G and 4G，supporting a variety of frequency bands，supporting Global Navigation Satellite System(GNSS)positioning，supporting IMS voice.This constant addition makes the BS omnipotent，but at the same time，be more complex and one change makes all change.For industrial scenarios，the BS with lightweight，low cost and flexible deployment is more needed to adapt to the customized vertical requirements.

With SBA-RAN，vertical industry partners are able to choose the services what they really want and deploy the services on different types of cloud platforms in different geographic locations as needed.For example，RAN services for ultra-low latency and ultra-high reliability applications may be deployed on edge cloud platforms to reduce access latency; RAN services for high-broadband applications may be deployed on convergent cloud platforms to reduce deployment costs.In addition，vertical industry partners can deploy RAN services with CN services on the same platform for better system performance.

5.2 Providing User-specific Service Capabilities for Individual Users

Traditional networks provide services for users based on the QoS requirements of the requested service.For the same service，the network delivers almost the same configuration to different users.However，in the future，different users may have different requirements even for the same service.The service-based RAN makes it possible.Based on user needs，customized network function service instances for UE can be generated and deployed on appropriate cloud platforms，for example，some UE-specific service instances related to AI model training can be deployed at the edge cloud platform to help UE timely complete computation-intensive tasks.

VI.CHALLENGES OF SBA-RAN

In the process of studying SBA-RAN，the inherent limitations of traditional RAN must be considered on the one hand，and the new problems caused by the introduction of new technologies must be solved on the other hand.Therefore，the challenges are enormous.

6.1 Service Definition

At present，there is no unified and feasible algorithm for service splitting in the industry.If the RAN functionalities are not decomposed properly，it is likely to develop a distributed monolithic application: a distributed system containing a large number of tightly coupled services that must be deployed together.This will combine the drawbacks of both monolithic architecture and microservice architecture.The BS has been developed as a monolithic application for decades，which makes the service decomposition be more complicated.The difficulties can be observed from the tightly coupled RRM modules.

6.2 Network Performance

The telecom equipment requires always-on operation without interruption，that is why BSs have always used dedicated hardware instead of general purpose processing platform (GPPP).The main advantage of GPPP is that it can accomplish more tasks，but at the cost of scattered computing power and reduced performance.

6.3 Test and Operation

Different RAN services may be implemented by different vendors.Due to the different technical solutions of different manufacturers and their understanding on interface specifications may also be different，the interoperation testing will be full of challenges.At the same time，it may be difficult to distinguish the responsibilities of vendors at the stage of installation or maintenance，which would affect the recovery process.As the number of devices that need to be maintained increases，the complexity and cost of network management increases.

6.4 Energy Efficiency

The SBA-RAN will be implemented on the generic server.This kind of server applies general-purpose chips，which have the advantages of low cost and high flexibility and the disadvantages of low processing efficiency and high power consumption.According to industry research，the number of generalpurpose chips used to implement 5G BS functions is 18 times than dedicated chips，and the power consumption is about 30 times than dedicated chips.Similarly，the high power consumption is also tricky issue for service-base RAN and needs to be carefully considered.

6.5 Heterogeneous Hardware

SBA-RAN seems to be a software reform，but in fact it depends heavily on the support of the underlying heterogeneous hardware platform.This involves the abstraction of hardware resources，virtualization，orchestration and management.The virtualization capabilities of heterogeneous hardware vary greatly due to different application scenarios.In addition，heterogeneous hardware solutions usually involve many hardware manufacturers，cloud manufacturers and network equipment manufacturers，and the technical challenges cannot be completely solved by any one company，therefore，a close cooperation between the suppliers in different fields are required.

6.6 Network Security

The accuracy and reliability of data directly affect the performance of a network or equipment.Therefore，it is important to ensure the security of data，especially the data exchanged between RAN services from different vendors.In addition，the risks of network attacks caused by network exposure，security risks introduced by cloudification and virtualization，and vulnerability of open source code also need to be considered simultaneously.

6.7 Initial Cost

The cost advantage of service-based RAN is not significant at this initial stage.Vendors involved in BS development include software providers，hardware providers，radio remote unit providers and system integrators.Before scale，the cost advantages are not obvious.But it can be expected that the cost advantage would be significant once scaled up.

6.8 The Way of Standardization

Currently，it may take two years for a software version to go from research，standardization to test and deployment.In addition，new versions need to be backward compatibility with older versions，making the introduction of new technologies or features very complicated.The existing standardization organization structure and way of working cannot meet the rapid development and deployment requirements of SBA-RAN，therefore adjustment is highly required.

VII.CONCLUSION

This paper analyzed the driving forces and necessity of SBA-RAN.Based on the proposed design principles，the services of control plane and user plane of RAN were defined from the protocol perspective，and the advantages brought by SBA-RAN were analyzed.There are at least two advantages of SBA-RAN-CP，more specifically，one is avoiding unnecessary AMF forwarding by introducing service-based interface between RAN services and CN services，another is improving processing efficiency by optimizing the serial processing procedures.The main advantage of SBARAN-UP is accelerating the data processing process since only the most necessary steps would be carried out.Some other advantages were also detailed analyzed in the paper，including signaling overhead reduction，header structure optimization，data value maximization，and more refined QoS management.After the basic schemes，two typical usage scenarios were given to explain how SBA-RAN can be applied.Though the future of SBA-RAN is exciting，there are still some challenges and problems to solve before its commercialization.This work detailed analyzed the challenges of SBA-RAN，which are also the future research directions of our team.

ACKNOWLEDGEMENT

This work was supported by the National Key R&D Program of China(2020YFB1806800).