Holistic Service-Based Architecture for Space-Air-Ground Integrated Network for 5G-Advanced and Beyond

Xiaoyun Wang,Tao Sun,Xiaodong Duan,Dan Wang,Yongjing Li,Ming Zhao,Zhigang Tian

1 China Mobile Communications Group Cor.,Ltd.,Beijing 100053,China

2 China Mobile Research Institute,Beijing 100053,China

3 Tsinghua University,Beijing 100084,China

Abstract: The Service-based Architecture (SBA) is one of the key innovations of 5G architecture that leverage modularized,self-contained and independent services to provide flexible and cloud-native 5G network.In this paper, SBA for Space-Air-Ground Integrated Network (SAGIN) is investigated to enable the 5G integration deployment.This paper proposes a novel Holistic Service-based Architecture (H-SBA)for SAGIN of 5G-Advanced and beyond,i.e.,6G.The H-SBA introduces the concept of end-to-end servicebased architecture design.The “Network Function Service”, introduced in 5G SBA, is extended from Control Plane to User Plane,from core network to access network.Based on H-SBA, the new generation of protocol design is proposed,which proposes to use IETF QUIC and SRv6 to substitute 5G HTTP/2.0 and GTP-U.Testing results show that new protocols can achieve low latency and high throughput,making them promising candidate for H-SBA.

Keywords: holistic service-based architecture; SBA;5G-advanced evolution; space-air-ground integrated architecture;6G architecture

I.INTRODUCTION

5G deployment is fast in recent years.As a new generation of mobile communication technology,5G provides users with a higher bandwidth and lower latency service experience, which are key features enabling variant vertical applications.By the end of April 2021,428 operators in 132 countries/territories are investing in 5G networks in the form of tests, trials, pilots, planned and actual deployments.Among those,153 operators in 64 countries/territories have launched commercial 3GPP-compatible 5G services[1].Global 5G mobile users are expected to grow to more than 3 billion by the end of 2025[2].

5G network modes are divided into Non-Stand Alone (NSA) and Stand Alone (SA).In 5G NSA mode, by leveraging 4G network, 5G network introduction can be quickly achieved.The 5G SA mode adopts a brand-new Service-based Architecture and independently builds the standard alone 5G core network, which well enables the three major technical advantages of enhanced Mobile Broadband (eMBB),massive Machine Type Communications (mMTC)and Ultra-reliable and Low Latency Communications(uRLLC), and support versertile vertical scenarios.Until the end of April 2021,68 operators are identified as investing in 5G SA(including those evaluating/testing,piloting,planning,deploying as well as those that have launched 5G SA networks)[1].

Satellite network has developed rapidly in recent years due to two reasons,1)the delay of satellite transmission has been reduced;2)the cost of satellite manufacturing and launching have been reduced.The large-scale adoption of medium and low-orbit satellites can greatly reduce the round-trip transmission delay.For example,the delay of low-orbit satellites can reach within 40ms [3], making the delay of satellite transmission comparable to terrestrial broadband network[4].The large-scale production of satellites and the advancement of vehicle technology such as recoverable rockets [5] have also significantly reduced the cost of satellite manufacturing and launching,laying a solid foundation for its entry into the civilian market.

At present,many companies such as OneWeb,O3b,SpaceX, Telesat, etc.have proposed satellite Internet plans [6].By the end of April 2021, Starlink has put more than 1,445 satellites into orbit and will provide satellite internet services in 35 states.The data transmission speed of the Starlink network is around 50Mbps to 150Mbps,and the network delay is around 20ms to 40ms[7].

An SAGIN may have different deployment options according to the type of the platforms involved.The platforms are grouped into three main categories:spaceborne platform, airborne platform and groundborne platform.The classification of spaceborne platforms typically depends on three main parameters,such as altitude,beam footprint size,and orbit.Spaceborne platforms can be differentiated as:

•Geostationary Earth Orbiting(GEO)has a circular and equatorial orbit around Earth at 35786 km altitude and the orbital period is equal to the Earth rotation period.The GEO appears fixed in the sky to the ground observers.GEO beam footprint size ranges from 200 to 3500 km.

• Medium Earth Orbiting (MEO) has a circular orbit around Earth, at an altitude varying from 8000 to 25000km.MEO beam footprint size ranges from 100 to 1000km.

• Low Earth Orbiting (LEO) has a circular orbit around Earth,at an altitude between 300 to 2000 km.LEO beam footprint size ranges from 100 to 1000 km[8].

LEO and MEO are also known as Non-GEO(NGSO) satellites for their motion around Earth with a lower period than the Earth rotation time; in fact, it varies from 1.5 to 12 hours.

From the perspective of the relatively mature constellation configurations at present, they are basically composed of single-layer satellites network,and there is no mixed constellation composed of multi-layer satellites at present.For example, iridium constellation, O3b constellation, INMARSAT constellation,etc., they are LEO constellation, MEO constellation and GEO satellite constellation respectively.

Facing 5G-Advanced and 6G,the integration of cellular network and satellite network has become a trend[9], which is called SAGIN in Figure 1 The SAGIN connects the multi-dimensional space of air,sky,earth and sea, providing users a more reliable and consistent service experience.Some research institutions and standardization organizations such as ITU,ETSI,and 3GPP are doing research and standardization on satellite and 5G integration.Since 2017, 3GPP has formulated some scenarios, basic functions and performance requirements for the integration of satellite and 5G[10-12], as well as some procedures and network management content [13, 14].Combined with the actual needs of the operator’s network,we believe 5G and satellite integration has four main benefits:

Figure 1. A diagram of SAGIN.

1) Seamless global integration.As a supplement to the ground-based cellular network, satellites can solve the needs of wide-area scenarios such as aviation, ocean, desert, and unmanned land.In the country where the terrestrial cellular network coverage rate is relatively high, it can also be used as a low-cost coverage solution to solve the needs of base station bandwidth relay in remote mountainous areas,islands and other scenarios such as emergency communications,offshore drilling platforms and other industries.In countries where terrestrial cellular network coverage is not complete,it can be used as a supplement to terrestrial cellular network to provide broadband mobile communication services for individuals,vehicles,and families.

2)Rapid expansion of hotspot areas.When hot spots such as venues need to be temporarily expanded, the general approach is to open a new base station or deploy an emergency communication vehicle,which requires a relatively long time and the cost is relatively high.The directional beam of the satellite can be adjusted by software to quickly align the area that needs expansion to achieve low-cost and rapid expansion.

3) New services such as multicast and broadcast.The satellite network has natural broadcasting characteristics, the link loss and delay in the coverage area are relatively consistent, which avoids the ”near and far”effect in the ground mobile cellular network,and users have a similar experience [2, 15].Using the satellite broadcast link to realize the message broadcast of the nodes on the chain can significantly reduce network bandwidth consumption and reduce the delay jitter of data synchronization between nodes.

4) Highly reliable communication backup.Using the satellite network to carry the base station transmission backup link,or deploying part of the wireless network and core network equipment on the satellite as a disaster recovery backup node, can effectively improve the base station’s ability to withstand various natural disasters and enhance the stability of the mobile network.

The integration of 5G networks and satellites will stimulate the evolution of the network.However, the research on the integrated architecture and key technologies are still immature.This paper analyzes the development trend of the integration of cellular and satellite network from the perspective of the evolution of 5G, specifically, the evolution of 5G SBA for SAGIN.The structure of this paper is as follows: Section I introduces the development of the 5G SA network,the status of the satellite network and the background of the intelligent integration of the two networks in 5G Advanced era and the realization of SAGIN.Section II introduces the concept, framework and message mechanism of 5G Service-based Architecture.Section III describes the design principles and implementation methods of the SAGIN architecture,with 5G deployment options and solutions.Section IV proposes novel Holistic Service-based Architecture and protocols for SGAIN towards 5G Advance and 6G era.The 5G network are enhanced from architecture, capabilities and protocols aspects.In Section V, a comparative test was carried out on the existing GTP protocol and the new SRv6 protocol.The testing results were presented, providing insights for the protocol design of the H-SBA.The paper is summarized in Section VI and pointed out future research directions such as intelligent multiple access coordination,deep space communication etc.

II.SBA EVOLUTION BACKGROUND

2.1 The 5G SBA Architecture

To define a future proof architecture,the 5G architecture design leveraged technologies such as Serviceoriented Architecture (SOA) and Micro-Service philosophy.The design also considers the cellular network intrinsic mechanism.All of these trigger the novel architecture design of 5G, i.e., Service-based Architecture[16].This architecture shown in Figure 2 is designed also considering the technical trends of cloud-native network function, as well as the underlining cloud infrastructure.

Figure 2. 5G Service-based Architecture.

In SBA, the elementary module is the “Service”which defines a Network Function.This breaks the black-box design approach of traditional cellular network architecture.The finer granular service enables the network customization based on those Service and do the corresponding orchestration.

For SBA, the corresponding interface is Servicebased.Such protocol stack employs Internet protocols which are evaluated,designed as Table 1.Those protocols are selected among variant candidates considering aspects such as performance,readability,etc.The Service-based interface can be invoked flexibly as re-quired,as compared with traditional Point-to-Point interface where the link between two points,established beforehand using telecom specific protocol such as Diameter,SCTP or GTP.

Table 1. Service-based interface(SBI)protocol.

2.2 SBA Framework and the Enhancement

Service registration, discovery and authentication are the three key elements of service framework.The Network Repository Function (NRF) is introduced to realize the service framework [16].A new service will register to NRF with its information such as service type,network address related parameters.In this way,the service plays a role of service provider so that this service can be discovered whenever necessary.Those who invoke a certain service are called as service consumers.A service consumer discovers proper service provider through service discovery mechanism.The authentication mechanism is used to make sure a service can only be invoked by those who are allowed to access the service.

In 3GPP Release 16 specification, the SBA framework is enhanced.One aspect to enhance the framework is to introduce a Service Communication Proxy(SCP).This makes the service communication more easy.The invoking service message can be sent to SCP and SCP delivers the corresponding message to the proper service provider.

2.3 Challenges Applying SBA to Space-Air-Ground Integrated Network

There are some main differences that Space-Air-Ground Integrated Network are considered as compared routine 5G network.Such differences are characterized as long distances, large number of network elements,complicated networking,dynamic nature of the topology.All of these bring great challenges to existing 5G SBA architecture.

1)Dynamic topology.In the territorial network,the network elements are static.Even though there may be some dynamic deployment cases where network functions can be scaled in or out in virtualized or Cloud deployment situation.In the Space-Air deployment cases, either the access network or the core network can be deployed in separate nodes (e.g., different orbits of satellites).The low orbits satellites can maintain its coverage of certain areas with several or tens of minutes, while the middle or high orbits satellites are about tens or several hours.Speed between satellites,and the satellites with the ground are high.This leads to unstable topology.Large-scale network topology exhibits complicated time and space variant characteristics.Therefore, it is necessary to do accurate modeling of the space and looking for proper topology organization to make the network stable and reliable.There are three categories on the dynamic topology(Refer to 3.2 for more detailed explore),i.e.,

(1)Base station and core network control plane are on the ground.The user plane is on the satellite.The user plane interfaces with base station and core network(i.e.,N3 N4 interfaces)will change.

(2) Base station is on the ground, and some of the core network functions such as user plane,part of the control plane (AMF and SMF) are on the satellite.This leads to topology changes over N1,N2,N3,N6,Namf and Nsmf interfaces.

(3) All base stations are on the ground.All core network functions are on the satellite.This leads to N1,N2,N3 interfaces not stable.

2) Huge amount terminal handover.As compared with existing model of static base station and moving terminals,the satellite enabled network contains moving base stations.The regular movement of a low-orbit satellite makes the terminals move out of the overage of the satellite oftentimes.Thus tens of thousands terminals may change their link several or tens of minutes for the interface N1 between UE and AMF.Therefore, the Radio Resource Management (RRM) which is used as the decision criteria for handover need to be evaluated.The link switch overhead and accuracy need to be improved.

3)Computing resource,storage and energy are limited in the Space-Air-Ground Integrated Network.The satellite payload and bandwidth resource need to be increased and packet processing capability needs to be enhanced.Since the satellite running in space, therefore, the size is limited leading to limited payload resource.For satellite network, the network related resources are quite limited.Such key constrains require to define lightweight,stable,easy for maintenance and must-to-have functions for SAGIN.

4) The long distance makes the long latency and long time to rescue.The link between satellites or between satellite and ground, both are long distance.The distance is related with the orbit type, i.e., GEO,MEO or LEO.The long distance makes the network topology or the routing failure cannot be discovered timely, impacts user experience [8] and is difficult to guarantee end to end reliability.This dramatically impacts latency constraint services.The opened channel expose to users thus also to attackers.Therefore, effective secured communication mechanism is needed.The encryption and the secured link mechanism also impact the latency of the satellite communication.

III.5G DEPLOYMENT OPTIONS FOR SAGIN

3.1 Overview

As described in previous sections, SBA provides a well basis for SAGIN.The Service-based Architecture is software oriented thus leads to decouple with hardware, which is flexible for 5G network deployed in the servers on the ground,as well as in the infrastructure of the satellite.Micro-service manner can be easily used for function customization on-demand.The Open API interface is also used for the satellite to be integrated with the ground network.

This section proposes intelligent SAGIN design based on SBA and analyzes the on-demand customization deployment of the core network in low, middle and high orbit cases.

3.2 Design Principles of SAGIN

In view of the limitation on the size, weight, power consumption, the design principles of SAGIN architecture are listed as below:

1)Optimize user experience.A major feature of SAGIN is the wide-area coverage.How to ensure that users can realize seamless and continuous service in complex network configuration is very important.

2) Minimization of network costs.Since the production and launch costs of satellites are relatively high and the resource in satellite is very limited, how to meet the needs of converged networks with fewest satellites and the lightest payload is also an important principle.

3) Extensible network logic architecture.A general and extensible logic network architecture can construct different converged networks according to variant needs of different industry networks.

Based on the above principles,this paper systematically proposes a series of solutions for 5G and satellite convergence networking.

3.3 On-demand 5G Deployment for SAGIN

The current 3GPP standard addresses the scenario where satellites are used as the backhaul(between 5G base stations and core networks [11]) or used as 5G base stations[14]shown in Figure 3.The architecture of the satellite as a backhaul is similar to 4G era.In order to adapt to the backhaul delay,the relevant timers in the 5G network are adjusted.The architecture of a satellite as a 5G base station is unique in the 5G stage.The air interface between the satellite and the terminal adopts an enhanced 5G air interface protocol.In addition, part or all of the base station’s functions are deployed on the satellite, and the core network is deployed on the ground, combined with satellite access.The mobility management,session management,multi-connection management,etc.are also enhanced.

Figure 3. Satellite as a backhaul and satellite as a base station architecture.

Figure 4. Converged architecture of core network deployed on satellites.

Another idea of SAGIN is to deploy all or part of the 5G core network on satellites to reduce the satelliteto-ground interaction process and further reduce the end-to-end delay.The network functions of the core network in SAGIN can be built according to the needs of the user plane delay,the control plane signaling load and delay.The base station can be on the ground or on the satellite.There are three types of deployment:

1)Type A:The user plane on the satellite to achieve low latency experience.The user plane network function(UPF)and application function(AF)are deployed on the satellite,while the control plane is deployed on the ground so as to save the satellite resource.The satellite is used to implement edge computing scenarios while other data flow will route in the network on ground.This type of network function is responsible for the processing of user plane data, and its location will directly affect the user’s experience.UPF and AF which deployed on the satellite are closest to the end user,and the transmission delay becomes smaller.

2) Type B: Part of the core network control function on the satellite.The user plane Network Function(NF) UPF, AF and the control plane Access and Mobility Management Function(AMF)and Session Management Function(SMF)are deployed on the satellite to handle frequent signaling exchanges.This will reduce user handover delays and session establishment delays.This deployment type is suitable for scenarios that require low control plane delay.

3) Type C: The whole network is on the satellite.All NFs in core networks such as Unified Data Management (UDM) are deployed on satellites.Network functions such as UDM are responsible for the user’s subscription and authentication.Such network functions do not directly interact with base stations and terminals, and do not interact frequently with other network functions,which will not significantly reduce the user’s business experience.In most cases, such network elements do not need to be deployed on satellites except in some specific cases such as closed industry private networks.

3.4 SAGIN Architecture of High,Medium and Low Orbit Satellite Hybrid Networking

The large numbers of applications of high, medium,and low orbit satellite hybrid networking and intersatellite links will make satellite constellations a cloud infrastructure.The “Service Set” [16] mechanism in 5G networks can make full use of cloud-based infrastructure to improve the robustness of the system.This paper proposes a SAGIN architecture leveraging“Service Set” mechanism illustrated by Figure 5, which distributes the network functions of the 5G core network on multiple satellites in different orbits.

Figure 5. SAGIN architecture of high,medium and low orbit satellite hybrid networking.

Figure 6. Holistic Service-based Architecture.

Figure 7. A schematic for UPF service.

The Service Set mechanism means that multiple NF instances form a set, and each NF in the set shares a set of context.The user can randomly access to any instance, and the instance obtains the user’s context from the shared set database, thereby achieving stateless network access and random access of users.When the network function of the core network is deployed on the satellite nebula in a distributed manner,the user context shared in the set can be stored on the medium orbit satellites.These satellites together form a set.When a satellite cannot provide services to a user due to mobility, other satellites in the set can quickly obtain the user’s context from the shared database of medium or high orbit satellites.The user plane network functions such as UPF and AF, as well as the control plan functions such as AMF and SMF, which are more relevant to the user service experience mentioned in the above section, can be deployed on loworbit satellites closer to users,while,the NF like UDM can be deployed on medium and high orbit satellites.

IV.HOLISTIC SBA FOR SAGIN

4.1 Holistic Service-based Architecture Design

Section III focuses on the realization of SAGIN based on the existing 5G Service-based Architecture.However,there are limitations with legacy architecture and protocols and may not suitable for 5G-Advanced and 6G era.

This section proposed a new enhancement of Service-based Architecture for 5G-Advanced and beyond, i.e., 6G.The idea is to leverage“Service”concept and extend from Control Plane to the User Plane and to the access network.

In order to support SAGIN network, 5G Servicebased Architecture needs to expand to support terminals access to 5G network through satellite.A typical mode is to deploy the Radio Access Network (RAN) on the ground, and to deploy UPF and part of 5GC on the satellite.The difference between the satellite network and the routine of 5G network are discussed in Section 2.3, i.e., dynamic topology between RAN and 5GC/UPF,the network mobility instead of terminal mobility,limited resources and bandwidth.Therefore, SBA architecture needs to be enhanced to provide more flexible and efficient architecture and protocols for SAGIN.

It is proposed to extend Service-based Architecture including RAN and UPF and the Holistic Servicebased Architecture will be designed and discussed in this section.The unified service-based interface will replace the N4 interface(between SMF and UPF)and the N2 interface(between gNB and AMF),thereby reducing the types of interfaces, making the design of the whole network interface more simplified.

4.2 Extend Service-based Concept to Access Network

Cloud oriented base station has become a recognized development trend in the industry.

C-RAN of 4G supports BBU pooling and X86 server environment can carry the upper layer of RAN.5G deploys the lower layer of the original BBU into AAU, puts the MAC and RLC layer with high realtime processing requirements in Distributed Unite(DU), while puts PDCP and RRC layer with low real-time processing requirements in Centralized Unit(CU).The pooling and centralization of CU can realize the unified scheduling of base station resources[18].In the future,for the base station on the satellite,there is high requirement of low power consumption and high processing capacity, and the cloud technology will become the key aspect to solve this problem.

In order to leverage cloud deployment of base station, this paper proposes the idea of enhancing base station to support 5G SBA.It is suggested the highlevel processing functionality of base stations is designed as services,and can be invoked by 5GC control plane NF.Therefore, the N2 interface(between RAN and 5GC control plane) should be enhanced to support SBI, and N3 interface (between RAN and 5GC user plane)should be enhanced to support unified user plane protocol.

According to the functionalities of the base station,the services provided by the base station can be consisted of access control service,mobility management service,session management service,and event exposure service,etc.

- Access control service is used to stipulate broadcast information,help terminal access,and be responsible for the initial authentication of wireless access.

- Mobility management service supports switching between Xn(interface between base stations)and N2 interfaces.The service can invoke another mobility management service of the adjacent base station to transfer the context of Xn switching,and it can also invoke the mobility management service in AMF to realize switching in N2 interfaces.

- Session management service is responsible for transmitting session related information between the base station and SMF.In this way,the interaction between RAN and 5GC can be carried out by invoking SMF services directly instead of through AMF to transfer session information.

- Event exposure service supports wireless information, such as terminal location and radio processing capacity exposed to the 5G core network.

The N2 interface is suggested to be redefined with supports service-based protocol.The N2 interface is conceptualized as an interface between RAN and AMF, but under the Service based Architecture, the N2 interface will be split up to support direct interactions between RAN and AMF and between RAN and SMF in the form of service invocation.In addition,considering the topological time-varying characteristics of satellite network, the corresponding servicebased protocol should support the rapid establishment and deletion of network connections on demand.

The N3 interface needs to support a unified userplane protocol.Considering the time-varying characteristics of satellite network topology and the requirement of limited resources,it is suggested to adopt native IP transmission protocol without GTP tunneling mechanism.

4.3 Extend Service-based Concept to User Plane

User Plane Function(UPF)supports service-based architecture can help realize full cloud deployment of the 5G core network.5G SBA currently only supports the service mechanism of the control plane of the 5G core network.However,as the main functionality for 5G data processing, UPF needs to enhance its support for the service design, including defining the typical services of UPF,and supporting the service based interface instead of current N4 interface using Packet Forwarding Control Protocol(PFCP)[17].

According to the functions of UPF, the services of UPF can be designed as: path control service, policy control service,event exposure service,etc.

- Path control service:responsible for the establishment of transmission path and tunnel processing.This service can be used to update tunnel information,or to insert/delete a UPF from an existing data path.

- Policy control service: this service can configure the corresponding PCF policy and/or policy trigger information on UPF to carry out appropriate data traffic processing and realize Packet Filter Description (PFD) management.The SMF can invoke the policy control service to transmit policies such as Quality of Service(QoS),legal interception,charging,etc.,to UPF.After the configuration is finished, the policies are executed when the corresponding uplink/downlink data starts to transmit.

- Event exposure service: Some information in the local UPF is expected to be exposed to the local NEF(possibly deployed in the Mobile Edge Computing(MEC)platform)and then sent to the local AF.This information may include specific terminal locations, UPF loading status, latency from UPF to terminal, and QoS profiles for specific applications.This information is relevant to the performance and capability of the local 5G network,and the vertical industry is looking forward this information to confirm whether the 5G System(5GS)SLA can be guaranteed.

4.4 Protocol Evolution-the Design for Holistic Service Based Interface

In order to meet the characteristics of SAGIN with limited computing power and limited space, the dynamic topology, as well as the end-to-end cloud oriented design, this section proposes new control plane and user plane protocols for the system.

1) Control plane protocol evolution, introducing QUIC protocol to replace existing 5G control plane transport layer protocol i.e.HTTP/2.0.

For the control plane, it is suggested to replace the HTTP/2.0/TCP for current 5G network with QUIC/UDP.HTTP/2.0 over QUIC is an internet transport protocol that is quite similar to TCP + TLS +HTTP/2.0,but based on UDP.UDP provides a way for applications to send encapsulated IP packets without establishing a connection[12].UDP does not require a handshake to establish a connection, and it moves a lot of work from the transport layer to the application layer.In this way,the future upgrade of the QUIC protocol is completely independent of the underlying operating system,and only the terminal and server upgrade to the specified version are needed.There are five key features for the QUIC protocol: shorter link establishment time, Multiplexing with no queue head blocking, improved congestion control, automatic error correction,and connection migration.

For the satellite network,the adoption of QUIC protocol has the following advantages:

(1) The relationship between nodes in 5G network when carried on satellite presents a time-varying state.If a long connection is maintained between nodes,frequent connection establishment and release will happen.If the QUIC protocol is adopted,0-RTT connection can be established between two nodes, and data packets can be sent directly without tedious connection establishment process.Data communication can be achieved in a shorter time, and there is no need to maintain a long connection, which saves network resource.

(2)The QUIC protocol supports connection migration:TCP uses an IP 5-tuple to represent a unique connection.When switching from one satellite access to another satellite access, the UE IP address changes,and a new TCP connection must be created to continue the data transmission.The UDP-based QUIC protocol abandons the concept of using IP 5-tuple to identify the connection stream, but uses 64-bit random numbers as the ID of the connection.Under the QUIC protocol, there will be no reconnection when switching between satellite accesses, or switching between 5G and satellite access,so as to improve the experience of the service layer.

(3) Both the satellite-ground link and the intersatellite link are long-distance transmission.If the TCP connection continues, packet loss or out of order will result in packet retransmission, which will greatly affect the system performance.While QUIC is designed with multiplexing properties,the loss of a packet usually affects only one certain stream.There is no dependency between the data streams in the QUIC.Moreover, QUIC is applied in the application layer and can support to change the congestion control algorithm more flexibly, such as the introduction of a wider range of message marking or the introduction of a fast retransmission mechanism.

2) User plane protocol evolution, exploring protocol based on SRv6.SRv6 refers to the application of Segment Routing to the IPv6 data plane,and Segment Routing Header(SRH)is inserted into an IPv6 packet with containing the Segment List represented by an IPv6 address list.The destination address of the message will be updated segment-by-segment to complete segment forwarding.There are functional instructions included in SRH in order to conduct corresponding operations on data packets,such as adding QoS identification, measurement indication, timestamp information,etc.,so as to facilitate better network design and customized encapsulation to meet the diversified requirements of new services.

The advantages of using SRv6 transport protocol as following:

(1)Path arrangement in advance to reduce the computational complexity and time of routing

In case of satellite access,the position between UE and the network,as well as the position between nodes in the network,is changing due to the rapid change of the satellite link.In current 5G network, this causes frequent tunnel establishment and release.If the routing and forwarding method is based on native IP without tunnel, the complexity of network maintenance and management of the tunnel can be avoided.In addition,in the case of multiple network nodes deployed in satellites,the relative positions between satellites will run in accordance with the predetermined orbit,so the use of SRv6 source address routing with planning the path in advance,is conducive to reduce the complexity of the system.

(2)Cross-layer unified control to achieve better SLA guarantee.

One of the key features of SRv6 is its compatibility for IPv6.Any network based on IPv6 transport can accept SRv6 messages seamlessly.When the carrying network, the transport network and the mobile network all use SRv6 to transmit the user plane data,it can realize the seamless integration of the transmission protocols in the three domains,which lays a good foundation for the unified data control.

The network can add QoS indication in SRH of SRv6 and be recognized by the underlying router,so that the carrying network, transport network, and mobile network can achieve cross-domain end-to-end QoS guarantee.

(3) Various new functions can be defined flexibly by using SRH functional instructions.For example,by adding time stamps and identifying time stamps,it can assist the network to realize end-to-end delay and jitter measurement,so as to get the network real-time state.

V.EVALUATION OF CANDIDATE PROTOCOLS FOR FULL-SERVICE ARCHITECTURE NETWORKS

5.1 Protocol Comparison Between GTP-U and SRv6

This section will describe the comparative testing of SRv6 and GTP-U so as to justify the feasibility of protocol replacement.The testing environment is described as below.

A bare-metal server is used, on which the Linux namespace mechanism (ip netns) is utilized to construct different network namespaces as different nodes in the user plane transmission process.And then these isolated namespaces are connected with vethpair.The server is equipped with a 24-core Intel Xeon X5675@3.07GHz CPU, 64GiB RAM, and the Linux kernel of 4.15.0-38-generic is installed.The establishment of GTP-U tunnel is realized based on libgtpnl(https://github.com/osmocom/libgtpnl), and the configuration of SRv6 decapsulation and forwarding strategy is realized with iproute tools(iproute-4.20).Two test topologies are simulated: the single-hop case is shown in Figure 8 with only one UPF as a forwarding node and the multi-hop case is shown in Figure 9 with 2 UPFs as forwarding nodes.

Figure 8. Single-hop case.

Figure 9. Multi-hop case.

Two important metrics, delay and throughput are evaluated on above test setup.

1)Delay comparison between GTP-U and SRv6

Iperf2 (https://sourceforge.net/projects /iperf2/) is used as client and server to stimulate the forwarding nodes.Packets of 500 Byte (the usual data length of mobile terminal) length are transmitted.The packet transmission interval is 1ms, and total 100,000 packets are transmitted.The distribution of one-way delay from the UE to the DN is shown in Figure 10.

Figure 10.One-way Delay distribution of GTP-U and SRv6 in single hop and multiple hop scenarios.

As shown in the Figure 10, for single-hop SRv6 transmission, 99% of the data varies between 7-9us,with an average value of 7.85us.For single-hop GTPU transmission, 99% of data varies between 7-10us,with an average value of 8.10us.For multi-hop SRv6 transmission,99%of data varies between 8-11us,with an average value of 9.64us; for multi-hop GTP-U transmission,99%of the data varies between 10-13us,with an average value of 10.95us.

SRv6 has lower delay than GTP-U, because GTPU messages have to be decapsulated and encapsulated during UPF processing,while SRv6 messages in UPF only need to update the destination IP address and decrease the Segment Left (SL) by 1.Since no decapsulation and encapsulation operation are needed, the processing time is saved.

2) Throughput comparison between GTP-U and SRv6

Iperf (https://software.es.net/iperf/) is used to test the throughput of the UDP stream between UE and DN.During the test,the bandwidth of the whole channel is fixed at 1Gbps, and different byte length are tested.The test results are shown in Figure 11.

Figure 11. Throughput test for GTP-U and SRv6 in single hope and multiple hop scenarios.

As shown in Figure 11, in the case of a single hop, the throughput of GTP-U is roughly equivalent to SRv6 with different packet length.When the packet length is 500 Byte, the throughput is 372 Mbps, and the processing rate is 93Kpps.In the case of multihop,the throughput of both GTP-U and SRv6 increase as the packet length increases, but the throughput of SRv6 is significantly better than GTP-U.For example, when the packet length is 500Byte, the throughput and the processing rate of GTP-U is 297Mbps and 74.25Kpps respectively, while the corresponding value of SRv6 is 329Mbps and 82.25Kpps.In multihop conditions,SRv6 uses less computing power than GTP-U for the same data rate, because encapsulation and decapsulation of GTP-U packets consume more power, while SRv6 packets process only need update the destination IP address and reduce the SL by 1.

Through the delay and throughput test of GTP-U and SRv6, it can be seen that GTP-U packets need to be decapsulated and encapsulated at each hop of UPF.In contrast, the SRv6 protocol does not require complex encapsulation and decapsulation operations,which need smaller processing delay and less CPU processing resources, so it is more conducive to the fast processing of user plane packets.

5.2 Protocol Comparison Between HTTP/2.0 and QUIC

This section will describe the comparative testing of HTTP/2.0 and QUIC/HTTP/3 so as to verify the feasibility of protocol replacement.The testing environment is described as below.HTTP/2.0 over QUIC adopts a lot of advanced schemes from HTTP/2.0,however a brand-new diagram mapping mode is designed, since QUIC is based on UDP.For historical reason, the term ‘HTTP/2.0 over QUIC’ means actually HTTP/3.

This testing is conducted on two Alibaba Cloud ECS(Elastic Compute Service).The experimental environment and equipment are as follows: 2vCPU of 2.5 GHz clock speed, 3.2 GHz turbo frequency Intel® Xeon ® Platinum 8269 (Cascade), 8 GiB RAM.QUIC/HTTP/3 and HTTP/2.0 Server and Client are deployed on these two ECSs, and they are connected through the vSwitch of the Alibaba Cloud Virtual Private Cloud.Two tests are designed to inspect the request-response latency in single flow case and multiple flow case.

1)Load transmission test with single flow

To compare the connection establishment time, total 100,000 times handshaking are fired in ideal channel environment without packet loss.For QUIC, the 0-RTT feature is enabled, and HTTP/2.0 uses 3-RTT TCP handshaking mechanism.It’s observed that the average connection establishment delay of HTTP/2.0 is 4.873ms,and the average connection establishment delay of QUIC is 3.242ms.It can be seen from Figure 12 that QUIC with the 0-RTT feature enabled can use a shorter time to establish a connection, and the delay of connection establishment is reduced by 33%compared to HTTP/2.0.

Figure 12. HTTP/2.0 and QUIC connection establishment delay comparison.

Tc (Linux command line tool to control network packet loss rate) can be used to control the packet loss rate and transmission delay on both the Client and Server side to simulate the gradual deterioration of the network environment.The transmission performance results of HTTP/3 and HTTP/2.0 are shown in Figure 13.

Figure 13. HTTP2.0 and HTTP/3 transmission performance comparison.

It can be seen from Figure 13 that HTTP/3 has better network performance than HTTP/2.0.In a weak network environment (high packet loss rate), due to large network delay, TCP will be blocked and a large number of retransmissions will occur.In a weak network scenario, the performance of HTTP/3 is significantly better than HTTP/2.0.Based on the comparison of two kinds of packet length of 500 byte and 100 byte,we also observe that the impact of packet length is almost the same for both HTTP/2.0 and HTTP/3,because it’s congestion control and retransmission not transmission time introduce more latency.

It can be seen from Figure 13 that the performance of HTTP2.0 deteriorates faster as the packet loss rate becomes larger or the transmission delay becomes larger.

Many factors would impact the HTTP/2.0 performance, including TCP handshaking, congestion control, TLS handshaking and resuming, HTTP multiplexing,etc.Because QUIC and HTTP/3 standards are in draft phase, their implementations are also in fast evolution.Therefore, above comparison is only used to explain the different design methodology leads to different performance characteristics.Since the comparison in this paper are in system level,more analyzation is needed for the contribution of different mechanisms to the overall QUIC and HTTP/3 latency.

2)Load transmission test with Multiple flows

Concurrent 10 connections are simulated to test the relationship between the packet loss rate and transmission delay of HTTP/3 and HTTP/2.0 under heavy load,and the result is shown in Figure 14.

Figure 14. Comparison of HTTP2.0 and HTTP/3 transmission performance in multiple flow case.

As can be seen from the Figure 14,under heavy load condition,the performance deterioration of HTTP/2.0 is more obvious than HTTP/3.As explained above,the performance may be changed with different implementation and parameters,but the different characteristics of these two protocols can be obviously observed.

Through the test of HTTP/2.0 and HTTP/3 in terms of transmission delay, it can be seen that QUIC has a smaller connection establishment delay and better performance in a weak network environment.This is mainly due to the 0-RTT connection establishment feature of QUIC and the UDP-based transmission,which avoids performance degradation caused by TCP congestion control.Especially for the weak network environment of satellite communication,this test proves that the use of QUIC and HTTP/3 protocol is more advantageous than the use of HTTP/2.0.

VI.CONCLUSION

6.1 Summary of the Paper

This paper analyzed the 5G architecture evolution towards 5G-Advanced and 6G era, considering the intelligent integration of Air-Space-Ground network.Three main contributions are made in this paper.

1.Proposed Holistic Service-based Architecture.As one of the key innovations of 5G,SBA is extended to meet the 5G-Advanced and 6G.The extension not only about the service used on the user plane,radio access network,as well as applies to the Air-Sky-Ground integrated cellular network.

2.Gave intelligent architecture integration options.The paper analyzed the deployment architecture in which the network can intelligent assign network functions per requirements.In particular,the control plane functions are investigated how to best meet with the integrated architecture.

3.New protocol evolution analysis and simulation.The protocols used for 5G SBA evolution to 6G are analyzed, considering the latest protocols developed in IETF and potential deployment in industry.The simulation results show that the new protocols such as QUIC and SRv6 are promising.

6.2 Future Work

In the 5G-Advanced era and 6G,the converged architecture is one important trends option for the industry.There are other directions need to investigate further such as:

Intelligent multiple access control: assign use traffic flows between ground communication and satellite communications.The traffic shall be smart routed among those two category accesses so that the user experiences can be best guaranteed.The traffic can be parallel or switched between those accesses.

The architecture investigation on Planet Internet will also be considered.Where, protocols such as “Delay tolerance network”will be used to meet with the long distance of the deep space communications.How the protocols apply to the integrated network architecture need to be investigated.

With the development and more Satellite network can be considered as new infrastructure.The communication capability and the computing capabilities also need to be enhanced to support Satellite network.

ACKNOWLEDGEMENT

This work has been funded by Tsinghua University-China Mobile Communications Group Co.,Ltd.Joint Institute.