This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 1

Cognitive Service Architecture

for 6G Core Network
Yuanzhe Li, Jie Huang, Qibo Sun, Tao Sun and Shangguang Wang, Senior Member, IEEE

Abstract—5G communication is making much progress in

Control Panel
achieving the Internet of Things and improving the quality of
user experience in large bandwidth scenarios. By introducing a NSSF AUSF UDM NEF NRF
variety of new technologies, the performance of 5G has been
greatly improved. However, emerging applications put forward
more stringent requirements in terms of latency, reliability, peak
AMF SMF PCF AF
data rate, service continuity, etc. Communication technology
still needs to be further developed. In this paper, the next
generation of core network is conceptualize. Inspired by the Data UE (R)AN UPF
UPF DN
nervous system of octopus, we propose a new cognitive service Control
architecture. Cognitive service architecture is a new architecture User Panel
designed for 6G core network. It is proposed to enhance the NEF: Network Exposure Function NSSF: Network Slice Selection Function
NRF: Network Repository Function AF: Application Function
core network so that it is qualified for the increasingly high UDM: Unified Data Management AUSF: Authentication Server Function
requirement for quality of service and complicated scenarios. PCF: Policy Control Function AMF: Access and Mobility Management Function
We first give a short vision on the 6G core network. Then UPF: User Plane Function SMF: Session Management Function
(R)AN: (Radio) Access Network DN: Data Networks
cognitive service architecture is demonstrated in detail. A case
study is demonstrated to show how cognitive service architecture Fig. 1. 5G Service Based Architecture.
enhance the performance of system. Enabling technologies for
6G cognitive service architecture are discussed at last.
network and core network. Besides, a variety of network tech-
Index Terms—6G, Core Network, Cognitive Service.
nologies, such as network slicing, software defined network
I. I NTRODUCTION and network function virtualization, are introduced to make
A. An Overview of 5G Network the network architecture flexible [6]. For example, with the
application of network function virtualization, traditional func-
T HE past few years have witnessed the rapid development
and accelerating deployment of the 5G communication
technology. 5G is not a simple update of the 4G communica-
tions have realized the decoupling of software and hardware.
Virtual network functions running on cloud servers take the
tion technology. It stands out in terms of three communication place of dedicated devices. This reduces the equipment cost
scenarios including enhanced Mobile BroadBand (eMBB), and makes the deployment of system more flexible.
massive Machine Type Communications (mMTC) and Ultra Based on these technologies, service based architecture [1]
Reliable Low Latency Communications (URLLC) [1]. In is adopted as the standard of 5G core network architecture.
particular, 5G network provides user devices with 0.1 Gbps As is shown in Fig. 1, 5G core network is split into multiple
data rates in the uniform spatial distribution with 10-20 Gbps network functions. Each network function is achieved by
peak data rates in eMBB [2]. For mMTC, the number of means of multiple microservices. By defining a unified inter-
connected devices supported in 5G increases 10 to 100 times face specification, the network functions are decoupled from
[3]. For delay-sensitive applications, 5G provides uRLLC to each other. The decoupled network functions can be expanded
achieve low latency service with reliability. When end to end and deployed independently. As a result, the flexibility and
latency is as low as 1 ms, the reliability is guaranteed as high efficiency of core network are improved. The same network
as 99.99% [4]. function can be called by multiple network functions. This
In order to increase the user’s access bandwidth, various reduces the coupling degree of interface definition between
emerging technologies have been proposed, such as millimeter network functions and bring advantages such as better load
wave communication, massive multiple-input multiple-output, balance, better disaster tolerance, easy capacity expansion and
ultra-dense network, etc. To achieve low latency in 5G core upgrading. Finally, it realizes the customization of the whole
network, mobile edge computation is proposed [5]. Services network function on demand [7].
can be deployed at proximity to prevent the unnecessary However, 5G still remains to be further developed in the
latency contributed by the transmission between radio access following aspects.
1) Poor Coverage: Currently, 5G network is deployed
Yuanzhe Li, Jie Huang, Qibo Sun and Shangguang Wang, are with the State
Key Laboratory of Networking and Switching Technology, Beijing University mainly as terrestrial mobile communication. However, high
of Posts and Telecommunications, Beijing, China. (Corresponding author: deployment cost and technology limitations makes it impos-
Shangguang Wang) sible to cover everywhere. In fact, remote areas are hardly
E-mail:{buptlyz;huangjie;qbsun;sgwang}@bupt.edu.cn
Tao Sun is with China Mobile Research Institute, Beijing, China. covered. At present, about 80% of the land area and more than
E-mail:[email protected] 94% of the sea area on the earth are not within the coverage

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 2

TABLE I B. Vision of 6G Core Network

C OMPARISON OF POSSIBLE KEY PERFORMANCE INDICATORS BETWEEN
6G AND 5G In the 6G communication, new technologies, such as tera-
hertz communications, visible light communications, advanced
Factors 6G 5G access-backhaul integration, etc., are introduced to realize
ultra-high peak rate, ultra mass access, ultra-high reliability
Peak data rate > 1000 Gbps 10 Gbps
in communication network [11]. As is shown in Table I,
Connection density > 10 million/km2 1 million/km2 compared with 5G, the performance of 6G will be greatly
improved. However, only introducing new technologies is not
Mobility > 1000 km/h 350 km/h
enough. It is difficult to solve the aforementioned problems
Delay < 0.1 ms T ens ms within current architecture. The core network architecture
should be redesigned to achieve a powerful, flexible and
Reliability > 99.99999% 99.99%
intelligent network. In the following, we will give a brief
vision of the 6G network.
of terrestrial mobile communication networks [6]. That means 1) Multiple Types of Mobile Communication: To access
network is not available when users are in mountains or sailing the Internet from anywhere, satellite communication, un-
across the sea where base stations are not deployed. Terrestrial manned aerial vehicle (UAV) communication and maritime
mobile communication cannot provide network access for communication will be deeply integrated into 6G network.
upper air, either. They will work as a supplement of terrestrial communication
and greatly expand the coverage of 6G network. Note that,
Although non-3GPP access network has already been sup-
although satellite communication, UAV communication and
ported since 3G [8]. 5G core network has supported satellite
maritime communication are all access network technologies,
communication [9], as well. The existing network architectures
the expansion of the access network needs the support of the
are mainly designed for terrestrial mobile communication.
core network. The complex and changeable access scenarios
They cannot guarantee the quality and continuity of service
set higher requirements for 6G core network. Take satellite
when users switch between different access modes.
communication as an example (shown in TABLE II), it has
2) Not Capable of Internet of Everything: 5G has made larger coverage and is not constrained by terrain. However,
much progress in achieving the goal of Internet of Things, satellite communication also have problems such as long
especially in the three typical scenarios, i.e. eMBB, mMTC latency, higher path loss, Doppler shift and more frequent
and uRLLC. However, it still has a long way to go. It still service migration. This presents great challenge to the 6G
fails to provide service for data-rate intensive applications with core network. First, unlike base stations located at one place,
ultra low latency [10]. For example, multi-sensory XR are fast moving satellites and UAVs make network control more
among those applications which 5G network does not support complicated. The core network has to deal with the frequent
well enough. Multi-sensory XR applications indicates virtual handover. To cope with the highly dynamic scenarios, the
reality, augmented reality and mixed reality applications that ability to cognize and predict changes in the access network
provide not only immersive experience, but also strong real- environment is indispensable to core network. Second, latency
time interaction. However, current network architecture is still of different communications varies a lot, making it hard for
not capable of satisfying both low latency and high data rate at to provide integrated service with guaranteed quality. Third,
the same time. Besides, how to guarantee the quality of service handovers between different communications trigger not only
in highly dynamic environments, for example frequent service communication handovers but also service migrations. Both
migration triggered by user movement, is still a challenge. communication topology and computation provision have to
3) Lacking in Intelligence and Flexibility: Through the be rescheduled.
ubiquitous network connection, large number of computing 2) Reliable Low Latency with Mobile Broadband: As is
requests converge to the edge of the network. Edge devices mentioned, multi-sensory XR will be killer applications in the
are faced with huge demand for computing resources. The age of 6G. To satisfy the service quality requirements of these
constant emergence of new delay-sensitive and computation- applications, both large bandwidth and low latency should be
intensive applications further aggravates the scarcity of re- guaranteed to prevent deterioration of user experience such
sources. To cope with this situation, an on-demand real- as black edge and frame dropping [12]. Considering XR
time scheduling of communication resources and computing are usually wearable devices, users may move around while
resources should be achieved efficiently. However, current 5G wearing it. Communication hand-off and service migration
network architecture is not intelligent and flexible enough to be may be frequently triggered, which will increase uncertainty
qualified. It lacks real-time perception and adaptive cognition of the system. In addition, multi-sensory XR applications
of scenario changes. Although 5G introduces network slicing, are computation exhausted and latency sensitive. Its real-
it mainly provides services based on a small amount of time rendering needs a lot of computing resources and a
semantic adaptation and simple rule matching for specific quick response. Both insufficiency of computing resources and
scenarios. It is difficult to deal with changeable network network congestion will result in latency constraint violation.
scenarios with high demand on short latency, such as multi- Providing a good network for these applications poses a huge
sensory XR applications and internet of vehicles. challenge to the 6G core network. To guarantee the quality of

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 3

TABLE II applying machine learning technologies in 6G system. Rubayet

S ATELLITE COMMUNICATION AND TERRESTRIAL COMMUNICATION Shafin et al. [20] give a possible roadmap on realizing artificial
intelligence-enabled cellular network in 6G. Shuhang Zhang
Satellite mobile Terrestrial mobile et al. [21] propose a UAV-to-Everything networking in 6G
Factors
communication communication
and put forward a reinforcement learning based framework for
Service link &
Link type Service link UAV-to-Everything communication. Wen Sun et al. [22] we
feeder link
Transmission present a new vision of Digital Twin Edge Networks in 6G
Above 600 km About 1 km
distance scenario and propose a mobile offloading scheme to reduce
offloading latency. Improving the performance of wireless
Delay Hundreds ms Tens ms
communication at 6G base station is another hot topic. Shanzhi
Path loss Above 180 dB Within 140dB Chen et al. [23] study beam-space multiplexing in 6G sys-
tem. Ertugrul Basar [24] introduces reconfigurable intelligent
Up to several
Doppler shift Within kHz level surface assisted communications to index modulation. It is
hundred kHz
a potential beyond MIMO solution that can be applied in
Cell Radius hundreds km 300-500 m 6G. Rony Kumer Saha [25] study the dynamic spectrum
Inter-beams, access and reuse in millimeter-wave spectrum and propose
Mobility Inter-cells a hybrid technique involving interweave-underlay spectrum
inter-satellites
access and reuse. In Rony Kumer Saha’s another work[26], he
Service Migration Satellite mobility,
Only user mobility puts forward dynamic exclusive-use spectrum access method
Trigger user mobility
for 6G millimeter-wave communication. There is not much
work on 6G network architecture. Guan Gui et al. [27] discuss
service, 6G core network should have the ability to identify key performance indices and propose a general architecture
changes in the network environment. Then, communication which demonstrate how different technologies are integrated.
and computation resources should be rescheduled quickly and Xiuquan Qiao et al. [28] propose design principles for a dis-
fine-grained. tributed, decentralized, and intelligent application provisioning
3) Communication Integrated with Artificial Intelligence: architecture for 6G.
6G network is obliged to fulfill the vision of Internet of Every- Most existing works mentioned above focus on either giving
thing. Large number of heterogeneous devices are connected a vision of 6G network or studying a specific technology
by 6G network and empowered by Artificial Intelligence that may be adopted in 6G. Very few work studies how to
(AI) applications. As AI applications are usually computation upgrade the 6G network architecture. [27] [28] give a general
intensive, how to guarantee computation provision is a key 6G architecture, but lack specific design.
challenge. Traditional cloud driven AI poses heavy load on D. Contribution
backbone networks and suffers from long latency. Developing
In this paper, we focus on upgrading the core network
computation resources at network edge is also faced with re-
architecture of 6G. Although 5G core network architecture has
source limitation challenge. In previous generations of mobile
made much progress by adopting service based architecture, it
communication technologies, the communication network only
is still not qualified for the emerging scenarios such as space-
serve as a pipeline between user devices and cloud servers.
air-ground multiple access network, internet of everything and
As a result, communication and computation resources are
providing intelligent and flexible network service. The exist-
scheduled individually. 6G will make a difference. It is not a
ing centralized architecture of core network lacks real-time
simple pipeline system for information. Instead, it’s more like
perception and adaptability to the complex and changeable
a large field integrated with AI and computation resources. We
scenarios in 6G. In this paper, we focus on the core network
call it the network empowered AI. All the devices in this field
architecture of 6G. The contributions are as follows:
is empowered and scheduled by the network and thus form
an Internet of Everything. On the other hand, to make the 1) We propose a cognitive service architecture for 6G core
network work efficiently, the core network of 6G must have network. As far as we know, this is the first 6G core
the ability to schedule all the devices within the coverage of network architecture that have ever been put forward.
the network. In this process, the controllers have to leverage 2) Basic conceptions of cognitive service architecture are
the power of AI to deal with the fast changing heterogeneous presented. Possible enabling technologies are analyzed
environment. We call it the AI empowered network. and discussed.
3) A case study is conducted to evaluate the performance of
C. Related Work cognitive service architecture. In the process of session
Most existing works discuss the key performance indices establishment, cognitive service architecture can save
and challenges [6] [10] [11] [13] [14] [15] [16] [17]. They more than 55.11% time than non-cognitive service ar-
analyze the technological trends and give potential solutions. chitecture.
Some work focus on discussing how to apply a specific The remainder of this paper is organized as follows. Sec-
technology in 6G network. T. Hewa et al. [18] give a vision tion II demonstrates the design of cognitive service architec-
on the role of blockchain in 6G and talk about the formidable ture. In specific, Section II-A presents features of octopus’s
challenges. Nei Kato et al. [19] analyze the challenges of nervous system. Section II-B describes cognitive service and

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 4

Retina
Basic Module
20M
3M motoneurons
Optic lobe 18M sensory receptors
（~60M neurons）
Real Time
0.5M AI Reasoning
Perception

……
Central Brain
（~50M neurons）
total 8 arms
Cognitive Service
0.5M

Optic lobe Fig. 3. Cognitive Service.

（~60M neurons） 3M motoneurons
18M sensory receptors
20M
adopting special motion control strategies. For example, in a
grasping action, the octopus’s arm is divided into three parts
Retina
by pseudo-joints. The middle section is as long as the near
Peripheral nervous system
section. The far section is like a hand to grab food with a set
Fig. 2. The Nervous System of an Octopus. of suction cups. Pseudo-joints are reshaped for each grasping
other key concepts of cognitive service architecture. Section III action, and can be adjusted according to the position of the
summarizes enabling technologies for cognitive service archi- target and arm. The octopus calculates the position of the
tecture. Section V concludes the paper. pseudo-joints from the arm itself: after touching the target,
two muscle activation waves propagate toward each other. One
II. 6G C ORE N ETWORK A RCHITECTURE propagates along the contact position to the proximal end and
Diversified objectives, changeable service scenarios and one propagates along the base of the arm to the far end, and
personalized user requirements not only require 6G network to the elbow is where the two waves collide [31].
have large capacity, ultra-low delay, but also a remarkable de- Third, the structure of the octopus’s nervous system is
gree of plasticity. In the face of the ever-changing requirements quite unique. As is shown in Fig. 2, the octopus’ nervous
in the distributed scenario, 6G network architecture should system is divided into three parts: a central brain, two large
have enough flexibility and scalability and is able to make very optic lobes connected to eyes and axial nerve cords along
fine-grained adjustments to the network in the control layer. the arms [30]. Most nerve cells are located in their nerve
In this section, we dive into the design of 6G core network cords of its arms. Only part of nerve cells are localized in
and try to propose an improved 6G core network architecture. the brain. This structure is dramatically different from skeletal
animals. Besides, the number of afferent and efferent fibers
A. Inspiration are relatively few. This means most complex motor skills are
To satisfy requirements of new applications in 6G era, organized in peripheral nervous system in their arms [30][32].
6G core network must be powerful and flexible with high The arm neuron network produces neuron activation patterns
efficiency. It should meet following requirements: 1) ultra- that specify all the space-time details of the basic mode of
large network capacity, 2) computation resources provided in motion. The brain sends global commands to the arm neural
the proximity, 3) strong cognition ability to recognize state network to activate and scale program variables [33].
change of the environment, and 4) efficient and compatible Overall, octopuses achieve a good control of their soft
control layer. body by leveraging a distinguished nervous system. Massive
It is quite difficult to design and deploy a system meeting all receptors bring excellent cognitive ability. Special motion
the above requirements. To find out a proper architecture, we strategies simplify the control. And most importantly, most
try to get inspiration from animals. After a lot of investigations, decisions are made in the arms and the central nervous system
octopus stands out because of its unique nervous system. only serve as a coordinator. All these are good examples on
Octopus is known for its flexibility and agility. Its nervous how to cope with control problems in a complicated system.
system is impressive in the following three aspects.
First, it has a strong and sensitive perception ability. The
B. Cognitive Service Architecture for 6G Core Network
arms of octopus are excellent in terms of the sense of touch.
The suction cups of arms are equipped with chemical receptors Inspired by the nervous system of octopus, we put forward
by which octopus has the ability to taste what it touches. cognitive service architecture for 6G core network. Cognitive
Because the powerful cognitive ability brought by chemical service architecture is an upgrade of 5G service based archi-
receptors, octopus can recognize its skin and prevent arms tecture. In the following, we will introduce what is cognitive
from tangled or stuck to each other [29]. service, how it works and what is cognitive service architecture
Second, it has an excellent control of its soft and complex in details.
body. The octopus’s body is not limited by joints and skele- 1) Cognitive Service: Cognitive service is the key concept
tons. The long and flexible arms can be extended, shortened of cognitive service architecture, which is an upgrade of tradi-
and bent at any point in any direction and length. The nervous tional network functions. As is shown in Fig. 3, the upgrade is
system must deal with infinitely large degrees of freedom when conducted by means of introducing two abilities. The first is
it tries to control the arms [30]. To simplify the process of the real time perception ability. Traditional network function
control, the octopus reduces the arms’ degrees of freedom by interfaces are transformed into polymorphic interfaces with

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 5

cognitive ability, which endow cognitive services with multi-

Knowledge Generation Module Knowledge
dimensional real-time perception ability, including request Graph
Model Library

flow, resource and topology status, operation and maintenance

Model Matching
Knowledge Quality Knowledge
events, etc. The second is AI reasoning ability. AI operators, Reasoning Assessment Update

rule matching units and approximate reasoning units are built Entity
Alignment
Ontology
Construction
in the network functions to realize online feature matching
Third-party Data Knowledge
and local reasoning. Thus, basic module of traditional network Database Integration Representation
Different 6G
Environment
function, real time perception and AI reasoning work as a Entity Relationship Attribute
Extraction Extraction Extraction
whole to form a cognitive service. Knowledge Mining Module
However, only introducing cognitive service is not enough.
The real time perception ability of different cognitive services Structured
Data
Unstructured
Data
should coordinate with each other to conduct complex percep- Communication Computation
Resource Resource
tion. Besides, the AI reasoning ability should be updated reg-
ularly to achieve better performance. Therefore, at the system
Fig. 4. Knowledge Graph for 6G Communication.
level, cognitive scheduling and knowledge graph management
are introduced. Cognitive scheduling arranges, prunes and scenarios.
merges the core network system and dynamically adjusts the In traditional network architecture, it is difficult for the
system capability according to the cognitive information of the applications to accurately get the network performance in real
network function and the polymorphic interface. Knowledge time. In turn, the network cannot make an adjustment in real
graph management works based on the theory of knowl- time according to the needs of the applications, either. The
edge graph. It gathers information from network functions inability to adaptively match computing resources and com-
and interfaces and update the knowledge graph of cognitive munication resources leads to a decline in service experience
service. Then the cognitive ability of network functions and in highly dynamic environments, especially for delay-sensitive
polymorphic interfaces are upgraded based on the updated and resource-exhausted applications. Therefore, the cognitive
knowledge graph. ability is of great importance. This is just like an octopus
Knowledge graph is the core component of system’s cogni- needs to know where the food is and which arm is best for
tive ability. It is in charge of analyzing and modeling of service getting food. Cognitive service architecture helps a lot to solve
quality demand and scene context. As is shown in Fig. 4, the this problem. It makes the core network have the ability to
knowledge graph building system consists of two modules: perceive real-time computing demand and computing resource
Knowledge mining module and knowledge generation module. distribution. The first step is to build the unified modeling of
Knowledge mining module has three sub-modules, which are heterogeneous computing and express it by means of ontology
in charge of extracting entity, relationship and attribute from modeling and semantic description. Then, based on the age
unstructured data, respectively. Knowledge generation module of information theory[34], the status of communication and
makes use of structured data and information extracted by computation resources are jointly evaluated.
knowledge mining module to generate knowledge graph. It has 2) Intelligent Scheduling: One of the most important fea-
several sub-modules. Knowledge Representation represents tures that distinguish 6G from previous generations of com-
entities and relations as the structure of the three-tuple. Entity munication is that it does not only play a role of pipeline
alignment points synonymous entities with different name to connecting difference devices. In other words, 6G commu-
the same objective object. Ontology construction use top-down nication is not just communication itself. Instead, it is more
method to construct a hierarchical relationship between a set like a field which interact with all the devices in it, and vise
of conceptual definitions and concepts. Knowledge reasoning versa. That is, just connecting network devices together is not
further explore the hidden knowledge based on the existing enough. 6G network should provide intelligent scheduling for
knowledge base to expand the knowledge base and obtain rea- both communication resources and computation resources. For
soning results. The process of constructing knowledge graph one thing, lacking any kind of resource will inevitably lead
works as follows. First, user requirements for communication to deterioration of user experience. For another, to prevent
and computation resources are extracted from massive service resource waste and improve the overall performance of the
instances based on data mining. The structured data in it is system, communication resources and computation resources
directly integrated with data in the third-party database. The should be jointly scheduled at a fine-grained level. Although
unstructured data first enters the knowledge mining module. It network slicing makes great progress in terms of manag-
is processed through entity extraction, relationship extraction ing communication resources conveniently leveraging network
and attribute extraction. Then, the integrated and mined data function virtualization, it is still in a coarse-grained level and
goes through knowledge representation and entity alignment lacks flexibility. To fulfill intelligent scheduling, new measures
operation. Next, if the represented knowledge does not exist on should be taken and relevant technologies should be upgraded.
the ontology, the ontology is constructed first. After that, the Based on the distributed edge core network, we introduce
represented knowledge is updated to the new ontology. Finally, AI as a Function in 6G core network to act as an intelligent
all knowledge ontology forms knowledge graph after the resource scheduler. AI as a Function is different from AI as
process of quality assessment. The knowledge graph generated a Service. AI as a Service is a service provided by service
in this scenario can be imported into a model library for new providers which play a role of enhancing application perfor-

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 6

Cloud Core
Network

Edge Core
Network Edge Core
Network
Basic Module

Real Time
AI Reasoning
Perception

Cognitive Service

Basic Module

Real Time
AI Reasoning
Perception

Cognitive Service

Edge Core
Network Edge Core
Network
Basic Module

Real Time
AI Reasoning
Perception

Cognitive Service

Basic Module

Real Time
AI Reasoning
Perception

Cognitive Service

Buoy D2D

Buoy

Brain-computer
Multi-sensory XR
Interface

Basic Module

Knowledge Graph Real Time

Perception
AI Reasoning

Cognitive Service
Cognitive Service Cognitive Scheduling

Cloud Edge
AI Steering Communication Link
Collaboration

Fig. 5. Cognitive Service Architecture for 6G.

mance. It includes a large variety of intelligent services such as cope with massive factors, such as user movement and scene
voice assistant, computer vision, intelligent recommendation changes that may result in service quality degradation. Then
and so on. They are all user-oriented intelligent services. the coordination scheme is fulfilled by means of service
However, AI as a Function is a network function of 6G core rescheduling, resource redistribution and service migration.
network. It works for the network and serve as a scheduler. The key is to dynamically adjust the service chain and keep
It is transparent to users but help to improve the overall it adapted to the ever-changing user behavior.
performance of the network.
3) Cognitive Service Architecture: In 5G network, edge
Intelligent scheduling is non-trivial in resource scheduling. computing is introduced to provided computation resources
Although mobile edge computing is introduced to equip access in the proximity. Although the 5G core network adopt service
network with edge servers, the computation resource of a based architecture, it still works as a whole logically. As we
single edge node is still limited. Generally, the arrival of mentioned above, communication resources and computation
the task is non-uniform. The computation resource needed resources should be scheduled jointly to cope with different
per task is uncertain, either. Poor scheduling will lead to resource-exhausted situations. However, in previous genera-
computation resource shortage or network congestion locally. tions of communications, the network only serve as a pipeline
To guarantee the quality of experience, computation resource and has no cognitive ability to computation resources. Besides,
of different edge nodes and the status of network should be core network and edge nodes are located at different level of
jointly arranged. That is, the access control of computing the network. The division of network and computation results
tasks and network routing is decided at the same time. For in much difficulty in the joint scheduling.
example, the routing protocol used in 6G network could Inspired by the octopus’s distributed nervous system struc-
be redesigned. The routing process considers not only the ture, core network will be split into edge core network and
status information of the network, but also the distribution cloud core network in 6G cognitive service architecture. Edge
of available computation resources. core network will further sink to the edge of the network. As
Just resource scheduling is not enough. Even with ample is shown in Fig. 5, cognitive service architecture will leverage
computation resources and good network status, users may still edge computing to form a multi-center architecture to provide
suffer from service quality degradation if services coordination efficient, flexible, ultra-low delay and ultra large capacity net-
relationship is not carefully scheduled. To deal with this prob- work services. Edge core networks act like peripheral nervous
lem, multiple measures should be taken. First, it is essential system in octopus’s arms, while the cloud core network plays
to realize the intelligent monitoring of service operation status a role of central brain. Most of the cognitive services will be
and establish a service operation quality prediction model. deployed at the edge core networks. Cognitive scheduling and
Then, the service collaboration trigger mechanism should take knowledge graph management are mainly deployed at edge
user requirements, equipment failures and system iteration core network as well. Cloud core network will no longer
upgrades into consideration. To achieve the coordination of directly participate in the communication. It is in charge of
the services, the collaboration strategy has to be carefully AI steering and helps edge core networks to coordinate with
designed. Deep reinforcement learning will be leveraged to each other.

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 7

With core network sinking to the edge, the control panel B. Polymorphic Interface Supporting Cognitive Services
of communication resources and computation resources are Traditional adaptive systems are mostly used for resource
finally laid at the same tier of the system, which paves the allocation in data centers, such as load balancing, container
way for realizing the visions in Section I-B. First, the cognitive arrangement, etc. Communication resources mainly rely on
service architecture works like an octopus. The flexible archi- manual configuration, lacking in flexible adaptive mechanism.
tecture makes it adapt to various access scenarios. Besides, However, this is not suitable for distributed edge core network.
the cognitive service can be optimized for different mobile For one thing, the edge core networks are located at different
communications. Second, as control panel of communication places, which is not convenient for unified management. For
resources and computation resources are all deployed at the another, traditional measures are unable to meet the require-
proximity of users, low latency is guaranteed while bandwidth ments for real-time and rapid adjustment. Therefore, poly-
pressure is relieved. Due to the sinking deployment of edge morphic cognitive service interface technology is introduced.
core network, the cognitive service architecture will realize the By introducing rule matching, approximate reasoning and
whole network coverage from the core network to the user other technologies, polymorphic interface has the functions
devices. On this basis, service continuity between different of situational awareness, demand identification, state statistics
communication modes could be guaranteed. When service and so on. The introduction of polymorphic interface endows
scenario and requirement changes, service link and commu- 6G core network with fine-grained perception ability. The
nication link can all switch seamlessly. Third, through AI as polymorphic interfaces not only complete the function of
a function running in edge core network, edge core network traditional network interfaces, but also have the ability of
supports service adaptation, service migration, service col- network state perception. It serves as a unified semantic
laboration and service evolution in terrestrial communication, description of the whole system and acts like the chemical
satellite communication, UAV communication and maritime receptors on an octopus’s arm. Although the cognitive ability
communication. of one polymorphic interface is simple, large quantities of
polymorphic interfaces work together to form the basis of
III. E NABLING T ECHNOLOGIES FOR 6G C OGNITIVE cognitive capabilities in cognitive service architecture.
S ERVICE A RCHITECTURE Behind the polymorphic interface is a lightweight learning
agent which runs real-time AI reasoning decision-making unit.
A. Unified Semantics for Network The agent has the ability of fast reasoning and learning based
service adaptation. The function of the agent is similar to the
6G network will attach great importance to the satisfaction
peripheral nervous system in the octopus arm. Due to the large
of users’ personalized needs, especially when it comes to
number of deployments on various edge devices, the learning
multi-sensory XR. In cognitive service architecture, unified
agent must be lightweight and able to work in a resource-
semantics should be defined to describe the cognitive objects
constrained environment. At the same time, these agents do
of the system.
not handle complex tasks. They only match the extraction
First, it helps to recognize the personalized needs of users.
of information recognized by the polymorphic interface and
It is necessary to conduct quantitative modeling and analy-
perform predefined operations. Agents with the same function
sis of human subjective experience to meet the information
in the system upgrades through the knowledge graph, federal
processing and transmission needs of differences. Besides, it
learning and transfer learning.
is important to explore the service cognitive mechanism and
the coupling of business logic and cognitive function based C. Service Continuity Guarantee
on user quality of experience. Building a quantitative model The introduction of edge core network architecture and
of human subjective experience is the first step to give core other communications (satellite, UAV and maritime commu-
network ability of cognition. nication) in 6G networks has brought great challenges to the
Second, a unified quantitative model of network states continuity of services. When users move across two edge core
should be defined. As mention above, 6G core network must networks, it is necessary to solve the cross-domain hand-off
jointly schedule communication and computation resources. of the network and the seamless migration of services [35].
Therefore, the model does not only contain communication On the other hand, the movement of satellites and drones may
states of the network such as network delay, bandwidth, jitter, also trigger network hand-off and service migration [6]. To
packet loss rate, etc., it also contains computation resource guarantee service continuity, the following technologies are
state including the distribution and the usage of different kind required.
of computing resources. The seamless service migration technology makes the ser-
Third, the mapping relationship between system resource vice migration process completely transparent to users through
status and user requirements is indispensable. As edge core the unified scheduling of communication resources and com-
network is introduced to 6G cognitive service architecture, puting resources. The main purpose is to overcome the link
network state and resource provision should match with user interruption caused by network re-establishment, service state
demands. Therefore, it is not enough if 6G core network only preservation, data transmission, and state restoration during
has the ability to recognize user demands and network state. service migration. Such interruptions will greatly affect the
Finding out the gap and the effect of interaction between them user experience of applications [36] that are sensitive to delay
is the key to make a decision. and require high reliability.

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 8

By learning and matching the tracks of users, satellites and

UAVs, the migration prediction technology can predict the Session Establishment
occurrence time and destination node of service migration. Procedure Begins

As a result, the system can migrate some state data of UDM Query
services in advance to reduce the downtime in the process of 5G Service 6G Cognitive
service migration. Generally, the reliability of the migration Based Architecture Architecture Service Architecture
Selection
process is guaranteed by increasing the calculation redun-
dancy(multiple edge nodes working together) and network link Policy Selection
redundancy(setting up multiple links simultaneously). Based
on the migration prediction, the cost brought by redundancy Apply Policy
Knowledge Graph
can be reduced. The higher the accuracy of prediction, the Selects Policy & UPF
UPF Selection
higher the utilization rate of system resources.
In addition, a lightweight migration carrier technology is
needed. The existing technologies are not suitable for service N4 Session
Establishment
migration. Virtual machine is too cumbersome, which will
Setup Uplink
bring too much additional data transmission. Virtual machine Data Path
is not suitable for microservice, either. Although container Setup Downlink
technology is light enough, it is not conducive to the preser- Data Path
vation of context state and remote recovery. Therefore, a Session Establishment
lightweight migration carrier technology combining advan- Procedure Ends

tages of virtual machines and containers is needed. It should Fig. 6. Session establishment procedure
not only meet requirements of microservice deployment, but
also can quickly encapsulate the current runtime context state E. Next Generation of Network Slicing
to meet the rapid migration and recovery of services. Although network slicing in 5G realizes the flexible allo-
cation of communication resources [37], it is still a coarse-
D. Universal Platform for Computation Network Integration grained network management method. 5G network slicing
One of the main design principles of 6G core network is provides services for fixed process scenarios based on a few
to achieve the integration of communication and computation semantic adaptation and simple rule combinations. It is diffi-
resources. In order to achieve this goal, we first need to build cult to meet the low delay service requirements of 6G network
a universal platform to achieve unified dynamic scheduling. caused by the changeable scenarios and the complex service
From the perspective of the platform, the application instances combination. Therefore, 6G network slicing technology should
and network functions are all services. The only difference lies be upgraded to meet the following requirements: 1) It should
in who they serve. Therefore, the platform is conducive to a support more fine-grained and flexible resource allocation. 2)
more flexible scheduling of computation and communication It supports the cooperation of multiple distributed edge core
resources. networks. 3) Network functions and computation resources
The cross-domain orchestration management technology are jointly managed and scheduled in the form of services.
supports collaboration between different edge core networks. 4) In the 6G high mobility and high reliability scenario,
Since the edge core network is deployed on a universal service migration are guaranteed with high reliability. 5) Each
platform, the number and location of edge core network network slice is deployed and released rapidly to achieve high
deployments can be very flexible. The cross-domain orchestra- efficiency. To meet these requirements, The upgrade mainly
tion management technology flexibly adjusts the deployment comes in two folds. On the one hand, the next generation
scheme of edge core networks according to the dynamic re- of network slicing should leverage cognitive service and
quirements of communication and computing resources in the intelligent scheduling to meet the above requirements. On the
city. The edge networks can merge or split to adjust the number other hand, fine-grained and flexible network slicing is one of
of edge nodes. The coverage area size of a single edge core the key technologies that enable cognitive service architecture
network changes as well according to the distribution of user to integrate computing resources and communication resources
requests. In addition, cross-domain orchestration management in a unified architecture.
technology supports grayscale deployment of core network More specifically, the upgrade of network slicing relies on
functions and smooth upgrade of network functions. the upgrade of its enabler technologies. That is, Network Func-
In cognitive service architecture, network functions and tions Virtualization (NFV) and Software Defined Networking
computing resources are abstracted as services. To achieve the (SDN) should also be upgraded to meet the high requirement
unified management of communication resources and com- of flexibility and scalability in 6G network. The controller
puting resources, a distributed service discovery mechanism of NFV and SDN should be upgraded leveraging cognitive
is need in the distributed core network scenario to achieve service. This enables the scheduling of network and network
efficient cross domain service discovery, service registration functions to be integrated into cognitive service architecture.
and service analysis, etc. The mechanism is event driven, has Besides, the virtualization of NFV and SDN should be realized
standard control interface and supports iterative optimization with finer granularity. This can guarantee the flexibility in
of service composition. resource scheduling.

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 9

a flexible and highly efficient architecture. This requires the 6G

225 core network to have cognitive capabilities and to achieve inte-
CSA
200 Non-CSA grated scheduling of communication resources and computing
resources. Inspired by the nervous system of octopus, we
175
Total Time (ms)

propose cognitive service architecture for 6G core network. In

150
this architecture, network functions are upgraded to cognitive
125 services by adding real time perception and AI reasoning abil-
100 ities. Besides, the previous core network is divided into edge
75 core networks and cloud core network. Edge core network is in
50 charge of resource management and scheduling of a specific
25 area. Most network management and control are conducted
0 20 40 60 80 100 at edge core network. The cloud core network only serve as
Number of users a coordinator. The key idea of cognitive service architecture
Fig. 7. Total time of session establishment
is to achieve a flexible network with cognitive ability by
imitating the nervous system of an octopus. A case study is
IV. C ASE S TUDY ON C OGNITIVE S ERVICE A RCHITECTURE conducted and the result show that by applying knowledge
Cognitive service architecture improve the overall perfor- graph, cognitive service architecture can simplify the workflow
mance and flexibility of core network by means of introduc- of network function interaction and improve the performance
ing cognitive service and upgrade the network architecture. of system. Possible enabling technologies are also discussed
However, as 6G networks are still in a very early stage of in this paper.
research and standards of 6G network haven’t been released In future work, we will concentrate on enhancing the
yet, it’s very hard to test the overall performance of the system coordination between different edge core networks. Typical
architecture. To demonstrate the performance improvement real world scenarios, such as dynamic service deployment
brought by cognitive service architecture, the process of ses- among different edge core networks and the migration of
sion establishment is chosen to conduct an evaluation. computation-intensive application will be tested. Besides, we
will upgrade current 5G core network functions to make it
A. Experiment Setup
applicable in 6G cognitive service architecture.
The test environment is set up based on free5gc [38],
which is an open source project that implement the 5G ACKNOWLEDGMENT
core network. It is used to emulate the process of session This work was supported in part by the National Key R&D
establishment of 5G network. In this experiment, a knowledge Program of China (2018YFE0205503), NSFC (61922017,
graph is added to the system to optimize the procedure of 62032003, and 61921003) and BUPT Excellent Ph.D. Students
session establishment. As is shown in Fig. 6, the first step Foundation(CX2019133).
of establishing a session is querying data from Unified Data
Management (UDM). Then, if the system works in 5G Service R EFERENCES
Based Architecture, policy selection, applying policy and User [1] 3GPP, “System architecture for the 5G System,” 3rd Generation
Plane Function(UPF) selection are conducted in sequence. Partnership Project (3GPP), Technical Specification (TS) 23.501.
However, in 6G cognitive service architecture, these works [Online]. Available: ftp://ftp.3gpp.org/Specs/archive/23 series/23.501/
[2] M. Agiwal, A. Roy, and N. Saxena, “Next Generation 5G Wireless
are conducted by the knowledge graph. After finishing the Networks: A Comprehensive Survey,” IEEE Communications Surveys
selection policy and UPF, N4 session is established. Then Tutorials, vol. 18, no. 3, pp. 1617–1655, Third quarter 2016.
follows the setup of uplink and downlink path. [3] A. Gupta and R. K. Jha, “A Survey of 5G Network: Architecture and
Emerging Technologies,” IEEE Access, vol. 3, pp. 1206–1232, 2015.
B. Total Time of Session Establishment [4] I. Parvez, A. Rahmati, I. Guvenc, A. I. Sarwat, and H. Dai, “A Survey on
Low Latency Towards 5G: RAN, Core Network and Caching Solutions,”
We measure the total time of establishing sessions with IEEE Communications Surveys Tutorials, vol. 20, no. 4, pp. 3098–3130,
different number of users. In Fig. 7, CSA denotes cognitive Fourth quarter 2018.
service architecture while Non-CSA denotes Non-cognitive [5] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, “Edge Computing: Vision
and Challenges,” IEEE Internet of Things Journal, vol. 3, no. 5, pp.
service architecture. Generally, CSA uses less time than Non- 637–646, oct 2016.
CSA. When there is only one user, the total time of CSA is [6] S. Chen, Y.-C. Liang, S. Sun, S. Kang, W. Cheng, and M. Peng,
56.47 ms less than that of Non-CSA. As the user number “Vision, Requirements, and Technology Trend of 6G: How to Tackle
the Challenges of System Coverage, Capacity, User Data-Rate and
increases, the gap becomes larger. When user number is 98, Movement Speed,” IEEE Wireless Communications, vol. 27, no. 2, pp.
the gap even reaches 128.43 ms. With different user number, 218–228, apr 2020.
CSA saves more than 55.11% time than Non-CSA. This is [7] M. Condoluci, S. H. Johnson, V. Ayadurai, M. A. Lema, M. A. Cuevas,
M. Dohler, and T. Mahmoodi, “Fixed-Mobile Convergence in the 5G
because CSA leverages knowledge graph to simplify part of Era: From Hybrid Access to Converged Core,” IEEE Network, vol. 33,
session establishment workflow. no. 2, pp. 138–145, mar 2019.
[8] 3GPP, “Architecture enhancements for non-3gpp accesses,” 3rd Gener-
V. C ONCLUSION ation Partnership Project (3GPP), Technical Specification (TS) 23.402.
6G network is destined to realize full coverage, full scene [Online]. Available: ftp://ftp.3gpp.org/Specs/archive/23 series/23.402/
[9] G. Giambene, S. Kota, and P. Pillai, “Satellite-5G Integration: A
and Internet of Everything. In order to cope with the complex Network Perspective,” IEEE Network, vol. 32, no. 5, pp. 25–31, sep
and changing scenarios, the 6G core network must implement 2018.

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2021.3063697, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. **, NO. *, 2020 10

[10] W. Saad, M. Bennis, and M. Chen, “A Vision of 6G Wireless Systems: [35] S. Wang, R. Urgaonkar, M. Zafer, T. He, K. Chan, and K. K. Leung,
Applications, Trends, Technologies, and Open Research Problems,” “Dynamic Service Migration in Mobile Edge Computing Based on
arXiv:1902.10265 [cs, math], jul 2019. Markov Decision Process,” IEEE/ACM Transactions on Networking,
[11] M. Giordani, M. Polese, M. Mezzavilla, S. Rangan, and M. Zorzi, vol. 27, no. 3, pp. 1272–1288, jun 2019.
“Toward 6G Networks: Use Cases and Technologies,” IEEE Commu- [36] L. Ma, S. Yi, N. Carter, and Q. Li, “Efficient Live Migration of Edge
nications Magazine, vol. 58, no. 3, pp. 55–61, mar 2020. Services Leveraging Container Layered Storage,” IEEE Transactions on
[12] iLab, “Cloud VR Black edge and Network Delay Relationship White Mobile Computing, vol. 18, no. 9, pp. 2020–2033, sep 2019.
Paper,” Shenzhen, China, Huawei, White Paper, 2019. [37] A. A. Barakabitze, A. Ahmad, R. Mijumbi, and A. Hines, “5G network
[13] J. M. C. Brito, L. L. Mendes, and J. G. S. Gontijo, “Brazil 6G Project slicing using SDN and NFV: A survey of taxonomy, architectures and
- An Approach to Build a National-wise Framework for 6G Networks,” future challenges,” Computer Networks, vol. 167, p. 106984, feb 2020.
in Proceedings of 2nd 6G Wireless Summit (6G SUMMIT 2020), mar [38] free5gc, “free5gc/free5gc,” 2020. [Online]. Available:
2020, pp. 1–5. https://github.com/free5gc/free5gc
[14] M. H. Alsharif, A. H. Kelechi, M. A. Albreem, S. A. Chaudhry, M. S.
Zia, and S. Kim, “Sixth Generation (6G) Wireless Networks: Vision, Yuanzhe Li received a Bachelor degree of Engi-
Research Activities, Challenges and Potential Solutions,” Symmetry, neering degree in communication engineering from
vol. 12, no. 4, p. 676, apr 2020. Beijing University of Posts and Telecommunica-
[15] A. K. Yerrapragada and B. Kelley, “On the Application of K-User MIMO tions(BUPT), in 2016. Currently, he is a Ph.D. candi-
for 6G Enhanced Mobile Broadband,” Sensors, vol. 20, no. 21, p. 6252, date at the State Key Laboratory of Networking and
jan 2020. Switching Technology, BUPT. His research interests
[16] L. U. Khan, I. Yaqoob, M. Imran, Z. Han, and C. S. Hong, “6G Wireless include mobile edge computing, cloud computing
Systems: A Vision, Architectural Elements, and Future Directions,” and service computing.
IEEE Access, vol. 8, pp. 147 029–147 044, 2020.
[17] S. Nayak and R. Patgiri, “6G Communications: Envisioning the Key
Issues and Challenges,” arXiv:2004.04024 [cs, eess], apr 2020.
[18] T. Hewa, G. Gür, A. Kalla, M. Ylianttila, A. Bracken, and M. Liyanage,
Jie Huang is currently a Ph.D. candidate at the State
“The Role of Blockchain in 6G: Challenges, Opportunities and Research
Key Laboratory of Networking and Switching Tech-
Directions,” in Proceedings of 2nd 6G Wireless Summit (6G SUMMIT
nology, Beijing University of Posts and Telecom-
2020), mar 2020, pp. 1–5.
munications (BUPT). Her research interests include
[19] N. Kato, B. Mao, F. Tang, Y. Kawamoto, and J. Liu, “Ten Challenges in
mobile edge computing and service computing.
Advancing Machine Learning Technologies toward 6G,” IEEE Wireless
Communications, vol. 27, no. 3, pp. 96–103, jun 2020.
[20] R. Shafin, L. Liu, V. Chandrasekhar, H. Chen, J. Reed, and J. C. Zhang,
“Artificial Intelligence-Enabled Cellular Networks: A Critical Path to
Beyond-5G and 6G,” IEEE Wireless Communications, vol. 27, no. 2,
pp. 212–217, apr 2020.
[21] S. Zhang, H. Zhang, and L. Song, “Beyond D2D: Full Dimension UAV-
to-Everything Communications in 6G,” IEEE Transactions on Vehicular Qibo Sun received his Ph.D. degree in communica-
Technology, vol. 69, no. 6, pp. 6592–6602, jun 2020. tion and electronic system from the Beijing Univer-
[22] W. Sun, H. Zhang, R. Wang, and Y. Zhang, “Reducing Offloading sity of Posts and Telecommunication in 2002. He is
Latency for Digital Twin Edge Networks in 6G,” IEEE Transactions currently an associate professor at the Beijing Uni-
on Vehicular Technology, vol. 69, no. 10, pp. 12 240–12 251, oct 2020. versity of Posts and Telecommunication in China.
[23] S. Chen, S. Sun, G. Xu, X. Su, and Y. Cai, “Beam-Space Multiplexing: He is a member of the China computer federation.
Practice, Theory, and Trends, From 4G TD-LTE, 5G, to 6G and Beyond,” His research interests include services computing,
IEEE Wireless Communications, vol. 27, no. 2, pp. 162–172, apr 2020. Internet of things, and network security.
[24] E. Basar, “Reconfigurable Intelligent Surface-Based Index Modulation:
A New Beyond MIMO Paradigm for 6G,” IEEE Transactions on
Communications, vol. 68, no. 5, pp. 3187–3196, may 2020.
[25] R. K. Saha, “A Hybrid Interweave–Underlay Countrywide Millimeter- Tao Sun got the B.S. degree on Automation in
Wave Spectrum Access and Reuse Technique for CR Indoor Small Cells 2003, and Ph.D. degree on Control Science and
in 5G/6G Era,” Sensors, vol. 20, no. 14, p. 3979, jan 2020. Engineering in 2008, both from Tsinghua University,
[26] ——, “On Exploiting Millimeter-Wave Spectrum Trading in Country- China. He currently is leading the research and stan-
wide Mobile Network Operators for High Spectral and Energy Efficien- dardization of the mobile network architecture and
cies in 5G/6G Era,” Sensors, vol. 20, no. 12, p. 3495, jan 2020. service evolution in China Mobile Research Institute.
[27] G. Gui, M. Liu, F. Tang, N. Kato, and F. Adachi, “6G: Opening New He worked on directions such as network slicing,
Horizons for Integration of Comfort, Security, and Intelligence,” IEEE 5G capabilities enablement for verticals, introducing
Wireless Communications, vol. 27, no. 5, pp. 126–132, oct 2020. AI to network and new IP technologies. He has
[28] X. Qiao, Y. Huang, S. Dustdar, and J. Chen, “6G Vision: An AI- more than 10 years of experience on Mobile network
Driven Decentralized Network and Service Architecture,” IEEE Internet standardization work. He currently serve as Vice
Computing, vol. 24, no. 4, pp. 33–40, jul 2020. Chair of 3GPP SA2.
[29] P. Graziadei, “Receptors in the Suckers of Octopus,” Nature, vol. 195,
no. 4836, pp. 57–59, jul 1962. Shangguang Wang is a Professor at the School
[30] B. Hochner, “An Embodied View of Octopus Neurobiology,” Current of Computing, Beijing University of Posts and
Biology, vol. 22, no. 20, pp. R887–R892, oct 2012. Telecommunications, China. He is a Vice-Director
[31] G. Levy and B. Hochner, “Embodied Organization of Octopus vulgaris of the State Key Laboratory of Networking and
Morphology, Vision, and Locomotion,” Frontiers in Physiology, vol. 8, Switching Technology. He has published more than
mar 2017. 150 papers, and his research interests include ser-
[32] L. Zullo, G. Sumbre, C. Agnisola, T. Flash, and B. Hochner, “Nonsoma- vice computing, cloud computing, and mobile edge
totopic Organization of the Higher Motor Centers in Octopus,” Current computing. He served as General Chairs or TPC
Biology, vol. 19, no. 19, pp. 1632–1636, oct 2009. Chairs of IEEE EDGE 2020, IEEE CLOUD 2020,
[33] G. Sumbre, Y. Gutfreund, G. Fiorito, T. Flash, and B. Hochner, “Control IEEE SAGC 2020, IEEE EDGE 2018, and IEEE
of octopus arm extension by a peripheral motor program,” Science (New ICFCE 2017, etc., and Vice-Chair of IEEE Technical
York, N.Y.), vol. 293, no. 5536, pp. 1845–1848, sep 2001. Committee on Services Computing (2015-2018). He is currently serving as
[34] C. Li, S. Li, and Y. T. Hou, “A General Model for Minimizing Age of Executive Vice-Chair of IEEE Technical Committee on Services Computing
Information at Network Edge,” in Proceedings of IEEE Conference on (2021-), and Vice-Chair of IEEE Technical Committee on Cloud Computing
Computer Communications (INFOCOM 2019), apr 2019, pp. 118–126. (2020-). He is a senior member of the IEEE.

1551-3203 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on May 15,2021 at 10:59:22 UTC from IEEE Xplore. Restrictions apply.