6G Service-Oriented Space-Air-Ground Integrated Network - A Survey

CJA 2272 No.
of Pages 18
3 January 2022
Chinese Journal of Aeronautics, (2022), xxx(xx): xxx–xxx
1
Chinese Society of Aeronautics and Astronautics

& Beihang University
Chinese Journal of Aeronautics
[email protected]
www.sciencedirect.com
of
3 6G service-oriented space-air-ground integrated
4 network: A survey
o
5 Nan CHENG a, Jingchao HE a, Zhisheng YIN b, Conghao ZHOU c,
Pr
6 Huaqing WU c, Feng LYU d, Haibo ZHOU e,*, Xuemin SHEN c
a
7 School of Telecommunications Engineering, Xidian University, Xi’an 710071, China
b
8 School of Cyber Engineering, Xidian University, Xi’an 710071, China
c
9 Department of Electrical and Computer Engineering, University of Waterloo, Waterloo N2L 3G1, Canada
10
11
d
e
ed
School of Computer Science and Engineering, Central South University, Changsha 410083, China
School of Electronic Science and Engineering, Nanjing University, Nanjing 210023, China
12 Received 12 May 2021; revised 24 December 2021; accepted 24 December 2021

13
ct
15 KEYWORDS Abstract As an indispensable component of the emerging 6G networks, Space-Air-Ground Inte-
16
re
17 Mobile Edge Computing grated Networks (SAGINs) are envisioned to provide ubiquitous network connectivity and services
18 (MEC); by integrating satellite networks, aerial networks, and terrestrial networks. In 6G SAGINs, a wide
19 Network Function Virtual- variety of network services with the features of diverse requirements, complex mobility, and multi-
20 ization (NFV); dimensional resources will pose great challenges to service provisioning, which urges the develop-
21 Network slicing; ment of service-oriented SAGINs. In this paper, we conduct a comprehensive review of 6G
or
22 Service-oriented network; SAGINs from a new perspective of service-oriented network. First, we present the requirements
23 Software Defined Network- of service-oriented networks, and then propose a service-oriented SAGINs management architec-
24 ing (SDN); ture. Two categories of critical technologies are presented and discussed, i.e., heterogeneous
25 Space-Air-Ground Inte-
resource orchestration technologies and the cloud-edge synergy technologies, which facilitate the
26 grated Networks (SAGINs)
nc
interoperability of different network segments and cooperatively orchestrate heterogeneous

resources across different domains, according to the service features and requirements. In addition,
the potential future research directions are also presented and discussed.
27 Ó 2021 Production and hosting by Elsevier Ltd. on behalf of Chinese Society of Aeronautics and
Astronautics. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).
U
1. Introduction 28
* Corresponding author.
With the development of wireless networks, the information 29
E-mail address: [email protected] (H. ZHOU).
services have been evolving, satisfying the requirements of peo- 30
Peer review under responsibility of Editorial Committee of CJA.
ple’s life, vertical fields, and the society. From 2G to 4G mobile 31
network, the services mainly focus on the data rate, with the 32
typical services such as Internet surfacing, video streaming, 33
Production and hosting by Elsevier and online gaming. The 5G mobile network also considers net- 34
https://doi.org/10.1016/j.cja.2021.12.013
1000-9361 Ó 2021 Production and hosting by Elsevier Ltd. on behalf of Chinese Society of Aeronautics and Astronautics.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: CHENG N et al. 6G service-oriented space-air-ground integrated network: A survey, Chin J Aeronaut (2022), https://doi.org/
10.1016/j.cja.2021.12.013
CJA 2272 No. of Pages 18
3 January 2022
2 N. CHENG et al.
35 work applications and services for vertical areas, such as effective in service coverage,6 and can provide broadband net- 96
36 industry, medical, and transportation. There are three service work access with multi-beam antennas.7 In Iridium constella- 97
37 scenarios classified in 5G, which are enhanced Mobile Broad- tion, 66 satellites with inter-satellite links provide network 98
38 Band (eMBB), ultra-Reliable and Low-Latency Communica- access to users anywhere and anytime. In recent years, 99
39 tions (uRLLC), and massive Machine-Type Communications SpaceX’s STARLINK introduced the concept of broadband 100
40 (mMTC).1,2 The eMBB scenario is mainly for further evolve- satellite Internet. In the satellite constellation proposed in 101
41 ment of broadband services, such as high-quality video stream- 2016, the first period of STARLINK (1,584 satellites in 24 102
42 ing and big data cloud storage, and thus the technical focus of planes) can provide high-date-rate network access for users 103
43 eMBB is improving the spectrum efficiency and data rate. The in 95 percent of the area between 58 degrees north latitude 104
44 uRLLC scenario is mainly for the latency and reliability- and 58 degrees south latitude. HAPs, operating at a height 105
45 critical services such as remote surgery, industry automation, of 17–22 km above Earth’s surface in the stratospheric region, 106
46 and driving safety. As the name suggests, it focuses on guaran- can be used in several different scenarios such as broadcast/- 107
of
47 teeing transmission latency and reliability. The mMTC sce- multicast HDTV signal, high-speed network access, naviga- 108
48 nario is mainly for the dense machine-type communications, tion, disaster and emergency detection, remote sensing, and 109
49 such as body monitoring and smart city, which focuses on emergency communications.8 In addition, UAVs have gained 110
50 energy efficiency and network capacity. Such a scenario classi- attention due to their flexibility and ability of fast 111
51 fication in 5G network can help identify the services and response.9–13 For example, during a special event, such as con- 112
o
52 enabling technologies in a certain scenario, and therefore each certs and sport matches when the cellular network is not suffi- 113
53 service, as long as is classified, can be fulfilled by a set of tech- cient for serving the crowded area, UAVs can be dynamically 114
54 nologies accordingly. Nevertheless, the 3-fold classification is deployed to assist the terrestrial network to enhance service 115
Pr
55 relatively coarse, and can hardly cover all services and applica- qualities.14,15 116
56 tions accurately. Compared with terrestrial networks, on-demand service ful- 117
57 In the emerging 6G era, the network services will continue fillment is more challenging in SAGINs. In SAGINs, a net- 118
58 to evolve, with the particular features of intelligence, complex- work service may include nodes, links, and resources across 119
59 ity, dynamics, and customization. 6G is envisioned as an multiple network domains and segments. Therefore, the ser- 120
60 Information-Communication-Data technology convergence ed vice fulfillment is significantly affected by different network 121
61 network, where the big data and Artificial Intelligence (AI) protocols, dynamic link conditions, high mobilities, and 122
62 are considered as fundamental components, i.e., native AI. A heterogeneous resources, individual of which renders service 123
63 soaring number of intelligent services, such as networked fulfillment in SAGINs overwhelmingly challenging. In addi- 124
64 robots, cognitive Internet of Things (cognitive IoT),3 self- tion, as SAGINs generally cover much larger areas than con- 125
65 driving, and digital twins, will require abundant networked ventional terrestrial networks, it is required to handle much 126
66 AI capability. Meanwhile, the increasing needs of ubiquitous more services simultaneously, which further complicates the 127
ct
67 global coverage and high-speed network access have urged service fulfillment process. However, the network architecture 128
68 the developments of Space-Air-Ground Integrated Networks and key enabling technologies for on-demand service fulfill- 129
69 (SAGINs). With SAGINs, diverse services can be supported ment in SAGINs have not been studied systematically. In this 130
70 anywhere and anytime. However, end-to-end services in paper, we present the concept of 6G service-oriented SAGINs 131
re
71 SAGINs over large geographical areas could cross many net- from the perspective of network services and applications, and 132
72 work domains, and thus require the orchestration of multi- discuss the advantages and enabling technologies. Further- 133
73 dimensional resources. In addition, the complex and heteroge- more, recent research works, technical challenges, and future 134
74 neous mobility in SAGINs makes the service fulfillment more directions are reviewed and described. This paper could pro- 135
challenging. Moreover, 6G network is envisioned to provide vide insights in research and development of 6G SAGINs,
or
75 136
76 specified services with distinctive and long-tail requirements especially addressing the issues of fulfilling complicated and 137
77 for each user, further increasing the difficulty in network numerous services with dynamic network environment and 138
78 orchestration and management. Meanwhile, with the increas- multi-dimensional network resources. 139
79 ing customizing capability of the network (e.g., with The remainder of the paper is organized as follows. Sec- 140
nc
80 Software-Defined Networking (SDN) and Network Function tion 2 introduces components and applications of 6G SAGINs 141
81 Virtualization (NFV) technologies), services deployed by users and main challenges in service-oriented techniques. Section 3 142
82 themselves will emerge and should be paid more attention. describes the overall architecture of the paper. Sections 4 143
83 Therefore, network service fulfillment, i.e., supporting network and 5 discuss related works on two main research issues in 144
84 services with network capabilities, is a significant yet challeng- service-oriented techniques in 6G SAGINs, where Section 4 145
ing issue in 6G network. However, the key technologies and focuses on the match between services and heterogeneous
U
85 146
86 methods in 6G network service fulfillment are still under inves- resources, and Section 5 investigates cloud-edge synergy. Sec- 147
87 tigation of academia and industry. tion 6 presents three potential research directions in future 148
88 As an indispensable characteristic of 6G network, the service-oriented SAGINs. Finally, Section 7 concludes the 149
89 SAGINs provide seamless global network coverage, which is paper. 150
90 essential for rural area Internet access, maritime communica-
91 tions, disaster recovery, autonomous driving, etc.4,5 In future 2. Service-oriented SAGINs applications and requirements 151
92 6G SAGINs, satellites, High-Altitude Platforms (HAPs), and
93 Unmanned Aerial Vehicles (UAVs) will be integrated with ter- Different from conventional terrestrial networks, SAGINs 152
94 restrial networks. Compared to conventional terrestrial wire- have distinctive features, in terms of architecture, protocols, 153
95 less communications, satellite communications are more cost-
10.1016/j.cja.2021.12.013
3 January 2022
6G service-oriented space-air-ground integrated network 3
154 resource orchestration, and services. The service-oriented tified for efficient service fulfillment. The following describes 210
155 SAGINs are a combination of SAGINs network architecture, three specific scenarios. 211
156 service management framework, and on-demand service tech-
157 nologies. In this section, we provide a brief introduction on 2.2.1. Automated driving and smart cars 212
158 SAGIN network architecture and analyze the requirements For automated driving, networks can provide road condition 213
159 of service-oriented SAGINs. Then, we give the design of ser- information and vehicles security surveillance. However, in 214
160 vice management framework and the review of enabling tech- remote areas, vehicles (including ships) cannot access the Inter- 215
161 nologies in the following sections. net through conventional terrestrial networks due to the lack 216
of network deployment or high access cost. LEO satellites 217
162 2.1. Introduction on SAGINs network architecture can provide fast data transmission globally with relatively 218
low costs.6 In addition, a single carrier beam of the satellite 219
163 The SAGINs architecture should support the integration of has wide coverage, and accessing the satellite network signifi- 220
of
164 space networks, aerial networks, and terrestrial networks with cantly reduces handoffs compared with accessing terrestrial 221
165 significant heterogeneity.16 Compared with conventional ter- networks, and improves the service continuity. Furthermore, 222
166 restrial base stations, LEO satellites are highly mobile and high network loads introduced by automated driving HD- 223
167 one satellite may cover an area with a radius of hundreds, even map downloading can be efficiently reduced by caching and 224
thousands of kilometers. Therefore, it is likely that a massive broadcasting the contents in UAVs.17 In addition, through
o
168 225
169 number of services are required to be fulfilled simultaneously, the combination of uploading from on-board sensors and 226
170 which poses great challenges to the service capability of the satellite remote imaging, the regional road condition informa- 227
Pr
171 satellite network. UAVs are flexibly deployed and scheduled tion and the driver’s safety situation can be updated in real 228
172 to assist network services in dedicated areas, such as time. 229
173 computing-intensive services in IoT network and massive
174 access services in crowded urban areas. However, the energy 2.2.2. Smart city 230
175 of UAVs is generally limited, and the cost and policy of A smart city is a modern city with versatile sensors and appli- 231
176 deployment in large-scale should be carefully considered. ances to collect data from citizens, vehicles, buildings, and 232
177 Notice that different segments are originally designed to oper- environments, which should be managed efficiently to provide 233
178
179
180
ate independently and work with different communication and
networking protocols. Therefore, how to design an efficient
network architecture suitable for various scenarios and ser-
ed high quality services.18 The smart city includes services such as
remote treatment, intelligent traffic management, intelligent
234
235
crime detection, and intelligent power supply. As a supple- 236
181 vices is a challenging problem, and has attracted much ment, the space network and aerial network in SAGINs can 237
182 research attention. assist terrestrial networks in data collection, on-demand mon- 238
183 There are mainly four basic SAGINs architectures. (A)
ct
itoring, and data traffic offloading. For example, based on 239
184 Hybrid satellite-terrestrial relay network. Due to the influence data collected by meteorological satellites, radars, and other 240
185 of bad weather or obstacles, terrestrial relays can be deployed sensors, the water level of rivers near the city can be predicted 241
186 in this architecture to receive satellite signals and offset the automatically and instructions can be carried out intelligently, 242
187 masking effect in direct satellite-to-user communications; (B)
re
thereby reducing the occurrence of disasters. In addition, 243

188 Satellite-terrestrial backhaul network. In remote areas such UAVs and road side sensors can be utilized to collect data in 244
189 as deserts and mountain areas, the deployment of backhaul terms of road status, traffic congestions, and accidents.19 245
190 links is not feasible with unacceptable cost. Satellite networks
191 can be applied to serving as backhaul networks to provide 2.2.3. Network coverage enhancement 246
or
192 seamless services; (C) Cognitive satellite-terrestrial network.

Satellites, UAVs, and HAPs are robust to disasters such as 247
193 The satellite-to-user link is considered as the primary link that
earthquake and flood. Therefore, SAGINs can help disaster 248
194 can occupy the spectrum resources anytime. In the opposite,
rescue and recovery when the terrestrial network infrastructure 249
195 the secondary network (i.e., terrestrial wireless network) can
is damaged in disasters. For instance, UAVs or wreckers 250
196 only transmit data in unused spectrum; and (D) Cooperative
nc
equipped with communication equipment can act as backup 251

197 satellite-terrestrial network. Different from the three architec-
base stations for network accessing, which helps reduce the dis- 252
198 tures above where different network segments work separately,
aster loss as much as possible.20 Besides, analyzing data from 253
199 each segment in this cooperative architecture is integrated clo-
satellite images meteorological observatory is potential to pre- 254
200 sely. Satellite networks can not only complement the terrestrial
dict potential environmental changes and disasters which is 255
201 network coverage to achieve ubiquitous communications, but
beneficial to take precautions. Furthermore, mobile network 256
U
202 can also enable traffic offloading to relieve the terrestrial traffic
operators can dispatch UAVs and HAPs as supplement to 257
203 burden.
relieve network traffic congestion in crowded areas. 258
204 2.2. Application scenarios

2.3. Requirements of service-oriented SAGINs 259
205 Since SAGINs can provide many benefits such as seamless

Instead of focusing on network coverage, user access, and data 260
206 coverage, cost-effective broadband network access, and flexi-
transmission, the service-oriented SAGINs aim at providing 261
207 ble network management, it enables a series of novel services
guaranteed services, i.e., the goal of network design and man- 262
208 and scenarios in 6G networks. The distinctive characteristics
agement is from the perspective of network services. The net- 263
209 of the services and scenarios in SAGINs should be clearly iden-
work coverage, user access, and data transmission, etc. are 264
10.1016/j.cja.2021.12.013
3 January 2022
4 N. CHENG et al.
265 considered as components of network service capability, and 2.3.3. On-demand network reconfigurability 324
266 are scheduled on-demand. To achieve the goal, conventional The conventional network is a vertically integrated system 325
267 network-centric and user-centric architecture and technologies where the control plane and data plane are closely coupled 326
268 are required to evolve to service-centric ones. In the sequel, the in a networking equipment, such as routers. Although such 327
269 requirements of service-oriented SAGINs architecture and systems are designed to provide services to most customers, 328
270 technologies are presented and discussed. they lack flexibility and ability to innovate, since they are usu- 329
ally very expensive in expertise and time to develop and deploy 330
271 2.3.1. Diverse service requirement fulfillment new network services.22 However, service-oriented networks 331
272 SAGINs are large-scale networks which cover a variety of sce- should be capable of accommodating customized services with 332
273 narios, users, and environments. Therefore, there will be signif- diverse requirements. To achieve this goal, the network is envi- 333
274 icant diversity in service requirements, which is a critical issue sioned to reconfigure according to the service demand, which 334
275 in on-demand service fulfillment. Some initial efforts have been coincides with the concept of network as a service. SDN/ 335
of
276 made in defining new service requirements in 6G network. NFV-related technologies are extensively investigated to 336
277 New 6G demands are proposed, i.e., extreme-high-speed and improve the network reconfigurability by virtualizing network 337
278 high-capacity communications, extreme coverage extension, functions as building blocks, which can be further connected 338
279 extreme-low power consumption and cost reduction, and chained to create services. In service-oriented SAGINs, 339
extreme-low latency, extreme-reliable communication, and two critical issues in network reconfigurability are worth to
o
280 340
281 extreme-massive connectivity & sensing.21 Considering these be studied. First, due to the large network scale, it is more dif- 341
282 new demands, the three 5G service scenarios are reinforced ficult for the control plane to achieve timely and reliable global 342
Pr
283 into further-eMBB (feMBB), ultra-mMTC (umMTC), Extre- coordination, which further degrades the performance of net- 343
284 mely Reliable and Low Latency Communications (ERLLC), work reconfiguration. Second, since the UAVs are mostly con- 344
285 Mobile Broadband Reliable Low Latency Communications trollable, it is important to reconfigure the positions and 345
286 (MBRLLC), and massive-uRLLC (muRLLC).7 Besides, there resources of the UAVs when composing the service function 346
287 might be new service demands which cannot be handled by chain, which is essential to achieve higher service performance. 347
288 existing methods. For example, with the proliferation of AI,
289 the demand of service intelligence level is essential in 6G era, 2.3.4. Multi-dimensional resource and mobility management 348
290
291
292
which has not been defined currently. In future 6G SAGINs,
how to meet the diverse service requirements should be
investigated.
ed The core idea of service-oriented network is to accurately
match the network capability and the service demands, in
349
350
which efficiently allocating the network resources plays the 351
most important role. In SAGINs, the end-to-end service fulfill- 352
293 2.3.2. Dealing with customized services ment usually involves the network configuration and resource 353
ct
294 In 5G era, the network services extend from mobile user ser- orchestration over multiple domains. This introduces more 354
295 vices to vertical field services, including industry, automobile, severe challenges in service fulfillment since in SAGINs the 355
296 education, government, etc. With the development of 6G net- resources in different domain share very high heterogeneity. 356
297 work, it is envisioned that the network should support more First, the resources are dedicated to specific domain. For 357
re
298 customized and personalized services with guaranteed quality example, the cellular systems exploit licensed FDD/TDD 358
299 for any service from any user. This trend stems from multiple bands, V2X communication system owns dedicated 75 MHz 359
300 new features of the emerging 6G network. First, the network band at 5.9 GHz, and satellite communications use very 360
301 native AI has achieved ubiquitous intelligence across the whole high-frequency bands, such as Ku/Ka bands. It is envisioned 361
302 network, which not only enables flexible and automatic net- that the SAGINs system can achieve higher performance if 362
or
303 work management, but also generates more personalized ser- the resources can be merged and used as a pool, yet the result- 363
304 vices due to deeper understanding of user needs. For ing interference could be an important issue to address. Sec- 364
305 example, the human bond communications transmit the fea- ond, the resources have a variety of limitations in different 365
306 tures of a subject in the way humans perceive it in a holistic domains, which stem from the fact that the network devices 366
307 way, which involves all the five sense and thus require very per- are very distinctive, especially in large-scale SAGINs. For 367
nc
308 sonalized service guarantee. Second, with the development of example, although the UAVs can serve as flexible flying base 368
309 SDN and cloud technologies, the network services and func- stations, they are usually very limited in energy. For LEO con- 369
310 tions are separated from physical network protocols and stellations, the capacity of link with laser communication is 370
311 devices, which facilitates customized service definition and high, while the on-satellite computing and caching resources 371
312 deployment. For instance, most state-of-the-art cloud plat- are often limited due to the satellite payload. Third, in 372
U
313 forms support the deployment of customized services using SAGINs, the network mobility is more complex than the ter- 373
314 software development kit or web APIs, such as Azure, AWS, restrial network due to the controllable UAV movements 374
315 Huawei Cloud, and so forth. Last, intent-based networking and very high-speed satellites. Such a network mobility will 375
316 is capable of translating user intents into network configura- bring dynamic service demands and complicates the service 376
317 tion and management operations without clearly describing fulfillment, and thus should be carefully considered. 377
318 the service requirements by defining the performance indica-
319 tors, such as data rate and latency. In this way, the intents 3. Service-oriented SAGINs management framework 378
320 can be seen as customized service requirements with unlimited
321 small granularity, and thus, is very challenging. Worse still, in Each network segment in SAGINs has its own pros and cons 379
322 SAGINs, the number of users and user categories will be very in coverage, capacities, and flexibility, as mentioned in Sec- 380
323 large, further requiring more customized services. tion 2. Through integrating the heterogeneous resources effi- 381
10.1016/j.cja.2021.12.013
3 January 2022
382 ciently and economically, networks have the ability to support demand service slices, providing guaranteed and separated ser- 417
383 more versatile services with differentiated Quality of Service vice fulfillment. 418
384 (QoS) requirements. However, fully exploiting the SAGINs In this review paper, we focus on the on-demand service 419
385 service capability is significant. We design a novel service- layer and related critical enabling technologies. Two categories 420
386 oriented SAGINs management framework which efficiently of technologies are considered, i.e., heterogeneous resource 421
387 manages the network services and orchestrates the network orchestration technologies and cloud-edge synergy technolo- 422
388 resources to achieve on-demand service fulfillment. As shown gies, with the subcategories and organization in this paper 423
389 in Fig. 1, the framework is composed of three layers, i.e., shown in Fig. 2. 424
390 SAGINs physical network layer, service analysis layer, and
391 on-demand service layer. The SAGINs physical network layer 3.1. Heterogeneous resources orchestration 425
392 is where the physical network operates, and offers a large
393 amount of data containing information about the network sta- For the heterogeneous resource orchestration, network slicing, 426
of
394 tus and the services, which are collected by ubiquitous sensing SDN, and NFV techniques have been investigated in SAGINs 427
395 or digital twin technologies. The data is sent to the service to achieve deep integration, efficient resource utilization, and 428
396 analysis layer, which carries out the very important service flexible network operation. 429
397 analysis function. To achieve service-oriented network, it is Network slicing techniques abstract physical network 430
398 significant to first clearly describe the diverse and customized resources of SAGINs into a shared virtual resource pool, to
o
431
399 service requirements and the implicit service demands in differ- support services with multi-dimensional resource require- 432
400 ent network contexts. In addition, the SAGINs resources ments. Existing studies on SAGINs network slicing are mainly 433
401 should be sensed and monitored in a very small granularity.
Pr
concentrated on network architecture and resource allocation. 434
402 For more details of service analysis, some existing research For the network slicing-oriented architecture, the integration 435
403 could be referred, such as intent-based networking,23 digital/- of heterogeneous networks is studied to improve the network 436
404 cyber twin,24 and resource monitoring.25 However, rather than interoperability and facilitate flexible end-to-end service slices 437
405 digging into the details of this layer, this paper focuses more on creation over multiple network segments.17,26–29 Resource allo- 438
406 the third layer, i.e., the on-demand service layer. The on- cation mainly focuses on the improvement of network perfor- 439
407 demand service layer takes the extracted information about ed mance such as network pressure reduction and capacity 440
408 the network services, resources, and contexts, and orchestrates enhancement.30–33 Based on a flexible and unified architecture, 441
409 the network service capability accordingly. Specifically, the heterogenous resources can be allocated to individual services 442
410 network virtualization module considers the service requests efficiently, which guarantees the service quality, reduces the 443
411 and the associated network contexts, and composes the service network redundancy, and promotes network scalability. 444
412 by connecting individual virtual network functions. The cloud/ For the resource allocation, the most critical issue is to effi- 445
413 edge synergy module coordinates the multi-dimensional ciently orchestrate multi-dimensional heterogeneous resources. 446
ct
414 resources and mobility, and outputs the service capability. Some researchers solve this issue based on traditional opti- 447
415 Then, the network slicing module matches the service function mization algorithms, which utilizes network information to 448
416 chain with the network service capability and forms on-
re
or
nc
U
Fig. 1 SAGINs service management framework.
10.1016/j.cja.2021.12.013
3 January 2022
6 N. CHENG et al.
o of
Fig. 2 Service-oriented techniques in 6G SAGINs.
Pr
449 execute efficient decision and control instructions.34–35 How- a group of adjacent devices, which reduces the mobility man- 491
450 ever, it is revealed that the traditional optimization approaches agement costs. 492
451 are not feasible in a large-scale network since the computa-
452 tional complexity could be very high.36,37 Recent works pro- 3.2. Cloud-edge synergy in SAGINs 493
453 pose to use AI-based resource allocation methods in
454 SAGINs, e.g., Reinforcement Learning (RL)-based methods, SAGINs can significantly benefit from the cloud-edge synergy 494
which can work when information on network model is
455
456
457
unknown.36,38 Since the vanilla RL methods are not suitable
to a relatively large network scale, deep RL is proposed to
ed by cooperatively reaping the advantages of both heteroge-
neous cloud and edge resources. In SAGINs, the terrestrial
network can provide a high-speed and stable network access
495
496
497
458 cope with the problem of scalability by using neural but it is hard to cover remote areas due to the geography 498
459 networks.39,40 and economic reasons. As a complementary, the satellite net- 499
460 The existing research on SDN/NFV is mainly about three work could offer seamless coverage such that the cloud servers 500
ct
461 aspects. First, for the resource orchestration in SAGINs, there could be constantly accessed, which provides powerful com- 501
462 have been research on SDN-based global network resource puting ability, but the delay is relatively large. In addition, 502
463 orchestration.10,41–43 In order to satisfy the diversified require- edge servers can be deployed at HAPs/UAVs to form a flexible 503
464 ments in the service-oriented SAGINs, these studies take and cost-effective Mobile Edge Computing (MEC) system, 504
re
465 advantage of the global control function of the SDN architec- offering low-latency services. Confronting the various require- 505
466 ture and calculate multiple transmission paths,42 combining AI ments of the emerging services, it is indispensable to exploit the 506
43,44
467 and other methods to achieve efficient allocation of vari- cloud-edge collaborative computing techniques in SAGINs to 507
468 ous network resources including communication resources, better expose network capabilities for on-demand service ful- 508
469 caching resources, and computing resources. Besides, there fillment. Through cloud-edge synergy, the network resources 509
or
470 have also been studies on employing NFV to schedule network can be collaboratively orchestrated according to service 510
471 resources in specific scenarios,45–47 e.g., SDN/NFV is analyzed requirements and characteristics. However, existing works on 511
472 for resource scheduling in the stadium.48 Second, for the net- cloud-edge synergy in SAGINs have not been conducted well 512
473 work management in SAGINs, considering the future compared with terrestrial networks. There are two main 513
service-oriented network requirements, SDN demonstrates
nc
474
research issues in cloud-edge synergy, the task offloading and 514
475 efficiency on managing heterogeneous networks. Particularly, the service deployment. Particularly, the task offloading is to 515
476 the characteristics of the separation of SDN control plane divide a complex task into several sub-tasks, which can be pro- 516
477 and data plane is beneficial to reduce the complexity of the cessed by several servers cooperatively, and the service deploy- 517
478 overall network,49 thereby improving the management effi- ment is to decide the locations of severs or VNFs to maximize 518
479 ciency.50–52 In addition, under the SDN architecture, a NFV- network efficiency. In Section 5, some latest studies containing 519
U
480 based intelligent gateway is designed in a specific home net- the above two issues and other miscellaneous aspects will be 520
481 work to enhance the QoS of Follow Me Services (FMS).53 Last introduced from the perspective of terrestrial networks57–60 521
482 but not least, mobility management has become a critical issue and SAGINs,61–63 respectively, and the challenges of future 522
483 in SAGINs, since the services bear the complex mobility fea- cloud-edge synergy will also be presented. 523
484 tures of different network segments. Towards the goal of con-
485 tinuous and seamless services, SDN and NFV can be used for 4. Heterogeneous resource orchestration in SAGINs 524
486 mobility management, one idea is to use SDN to predict node
487 movement and link interruption globally and combine mathe-
488 matical models such as game theory to maximize terminal ben- In future 6G SAGINs, space, aerial, and terrestrial networks 525
489 efit.41,54–56 By leveraging NFV techniques, multiple functions will be deeply integrated. It is forecasted that mobile data traf- 526
490 of the network can be bundled in a single physical device or fic will reach 77.5 exabytes per month by 2022,64 and the num- 527
10.1016/j.cja.2021.12.013
3 January 2022
528 ber of IoT connected devices worldwide will be 25 billion in features of the network slicing based system are summarized 559
529 2025.65 How to utilize the limited but heterogeneous resources as follows. 560
530 efficiently to accommodate the surging data has become an
531 important research direction. Besides, future networks need (1) Automation: automation means that the network pro- 561
532 to support vehicular communications, tactile Internet, smart vides service supports or refines the local network per- 562
533 cities, and other applications mentioned in Sections 1-2, which formance automatically without manual interventions. 563
534 not only have a higher requirement in individual performance When the service request of users arrives or its QoS 564
535 metrics, such as coverage, latency, reliability, throughput, and updates, the network slicing-based SAGINs orchestra- 565
536 energy utilization, but also require a mixture of them. Conven- tor allocates network resources automatically in the 566
537 tional terrestrial networks can hardly fulfill these requirements form of Virtual Network Function (VNF) instantiation, 567
538 at a reasonable cost. In this section, we investigate two promis- scaling up/down, scaling out/in, and cancellation. 568
539 ing techniques, i.e., network slicing and SDN/NFV, which (2) Stabilization: considering the highly dynamic network 569
of
540 play an important role in matching between SAGINs services topologies, user mobility, and real-time service require- 570
541 and heterogeneous resources. ments in SAGINs, it is very likely to render the optimal 571
slice configuration to sub-optimal or even unqualified. 572
542 4.1. Network slicing For example, in latency-critical services, medical acci- 573
dents may occur due to unreliable network performance 574
o
543 4.1.1. Overview of network slicing in remote surgery and unstable transmission latency or 575
544 Network slicing is considered as the virtual, modular, and iso- rate would affect Quality of Experience (QoE) in remote 576
conference or video streaming services. 577
Pr
545 lated network technique built upon virtualized network
546 resources abstracted from separated physical infrastructures.29 (3) Customization: in future service-oriented SAGINs, net- 578
547 The network slicing-based SAGINs architecture, as shown in works should have ability to support multi-dimensional 579
548 Fig. 3, can be divided into three parts: service instance layer, requirements customized by users in latency, band- 580
549 network slice instance layer, and resource layer.66 Firstly, widths, storage, etc. Focusing on differentiated services, 581
550 resources of physical infrastructures are abstracted into a vir- networks can be realized based on the separation of con- 582
551 tual resource pool by NFV and are defined as resource blocks
ed trol and data planes, which are programmable in poli- 583
552 by the network manager. When service requests of users arrive, cies, operations, and protocols. 584
553 they are described in the form of several network indexes (e.g., (4) Isolation: to guarantee the QoS of different services and 585
554 bandwidth, caching, computing, coverage, power, etc.). Then, different users, there should be isolation between net- 586
555 the corresponding dedicated virtual resource blocks are work slices. With network slicing techniques, isolation 587
556 mapped into physical resource layer by dedicated resource is assured through the usage of orchestration mecha- 588
557 mapping mechanisms. Finally, a network slice is confirmed nisms and virtual resources mapping algorithms to 589
ct
558 and available to support services. In addition, several critical avoid mutual interference.67 590
591
re
or
nc
U
Fig. 3 Network slicing based on SAGINs architecture.
10.1016/j.cja.2021.12.013
3 January 2022
8 N. CHENG et al.
592 4.1.2. Related research drift-plus-penalty based resource method to minimize the 651
593 There have been extensive research works in SAGINs, which power consumption in providing reliable and low-latency ser- 652
594 can be categorized into the investigation of network slicing- vices.74 For the traffic offloading issue in HAPs, an interfer- 653
595 based architecture and resource allocation in network slicing. ence cancellation strategy is proposed based on spectrum 654
sharing and power allocation schemes,72,75,76 which efficiently 655
596 (1) SAGINs architecture for network slicing saves spectrum resources while achieving the same perfor- 656
597 mance. In these schemes, only unicasting is considered, and 657
598 To reduce network complexity and facilitate flexible hetero- there is much room for further performance improvement. 658
599 geneous resource allocation, the SAGINs architecture for net- For the proactive HAPs broadcasting and vehicle caching, a 659
600 work slicing has been investigated.26,29–31,33,68–71 A novel radio network sharing and slicing strategy is used to minimize the 660
601 multiservice network adaptive architecture (NORMA) based data rate of the roadside units to guarantee QoS in vehicular 661
602 on SDN/NFV and perception of QoE and QoS is presented, networks.17 In order to handle the traffic in complicated and 662
of
603 where SDN controller and coordinator are defined for intra- dynamic network efficiently, a slicing strategy based on the 663
604 slicing resource allocation and resource competition in physi- Monte Carlo tree search is proposed in the context of fog radio 664
605 cal layer.30 Based on network slicing in satellite-terrestrial net- access networks, which achieves better delay and throughput 665
606 works, the broadband data transmission of users in remote performance compared with benchmark algorithms.77 Another 666
areas is achieved by Internet access from satellite or terrestrial research focuses on the total revenue of network operators in 667
o
607
608 segments.26 Ahmed et al. proposed an on-demand satellite- Cloud-RAN (C-RAN) from the perspective of user number 668
609 based network slicing framework, named on-demand adaptive and idle time.78 An entity-oriented radio access network shar- 669
ing strategy is presented in the cellular network to achieve effi- 670
Pr
610 network slicing, to support services with highly customizing in
611 the form of an end-to-end function chain.29 Focusing on QoS- cient on-demand resource allocation, at the cost of reduction 671
612 restricted capacity maximization, a study on HAPs constella- in network flexibility since deep packet inspections and infor- 672
613 tion is proposed.33 In order to fulfill real-time processing capa- mation from Radio Access Network (RAN) are needed.68 To 673
614 bility in orbit, a flexible space-based edge computing system address this issue, a dynamic and efficient RAN resource allo- 674
615 architecture and a resource management mechanism are pro- cation mechanism, named application-oriented framework for 675
616 posed to reduce the delay and improve the throughput.31

ed RAN, is presented,79 where the resource allocation is allowed 676
617 The network backhaul bandwidth pressure can be relieved in both central controller and applications, and the network 677
618 and comprehensive benefits of the satellite network can be revenue is shown to be 40 percent higher than network reser- 678
619 improved. A novel network architecture is proposed from vation scheme68 and twice as good as Per-Base-station Reser- 679
620 the perspective of business model, where the networks are par- vation.80 The reliability of V2V links from the perspective of 680
621 titioned between infrastructure providers, mobile virtual net- Signal to Interference plus Noise Ratio (SINR) outage is ana- 681
622 work operators, service providers.69 Network slicing is lyzed.81,82 Based on this, considering the resources of roadside 682
ct
623 required to assure mobile virtual network operators to have units, an online algorithm named JRPSV is proposed,35 which 683
624 a complete virtual network. To fulfill diverse content applica- fulfills ultra-reliability and low-latency services in Vehicle-to- 684
625 tions with multi-dimensional QoS, a service-oriented hierarchi- Vehicle (V2V) communications and complies with the queuing 685
626 cal soft slicing framework is designed, where one slice rules, and thus acquires asymptotically optimal network 686
re
627 corresponds to one service and resources are reused oppor- capacity. Some other researchers focus on maximizing the 687
628 tunistically in inter-slice and intra-slice to obtain the multiplex- sum rate of users. Jiang et al. proposed a dynamic network 688
629 ing gain.72 The results show that the throughput under the slicing scheme which timely allocates resources based on the 689
630 proposed strategy is better than the hard-slicing scheme. requested Service Level Agreement (SLA) to maximize sum 690
rates of UEs considering user fairness.83 A mixed integer
or
691
631 (2) Resource allocation non-linear programing problem for allocating sub-carriers 692
and power resources in RAN is formulated to solve this opti- 693
632 Since space networks, aerial networks, and terrestrial net- mization problem, and a low complexity resource allocation 694
633 works have their own features, how to allocate the distinctive scheme is proposed, including the heuristic NS assignment 695
nc
634 resources to different network slices appropriately is worth and the many-to-one matching sub-carrier allocation algo- 696
635 investigating. Resource allocation strategies in network slicing rithm.34 To provide customized services in vehicular network, 697
636 can be divided into two main categories: traditional optimiza- the spectrum allocation between BSs to maximize network 698
637 tion algorithms and AI-based approaches. throughput is transformed into a solvable form and solved 699
638 Traditional optimization-based resource allocation schemes through an Alternative Concave Search (ACS) algorithm.84 700
639 exploit off-the-shelf optimization methods to orchestrate AI techniques have shown superiority in resource allocation
U
701
640 multi-dimensional resources given an optimization utility and in extremely complicated networks.36,37,43,44 Particularly, a 702
641 several network constraints. One important research concen- neural network-based resource allocation mechanism has been 703
642 tration is the maximization of network revenue from the per- applied to satellite networks, aerial networks, and terrestrial 704
643 ceptive of the number of users, power, latency, etc. Lyu networks, respectively,85–87 and an efficient deep learning- 705
644 et al. proposed a service-oriented resource slicing scheme and based resource allocation in the integrated network of all three 706
645 optimization for Space-Air-Ground Integrated Vehicular Net- segments is presented.43 A multi-agent RL-based resource allo- 707
646 work (SAGIVN) to maximize the system revenue in terms of cation algorithm named LESS-DS is proposed to reduce the 708
647 the average amount of admitted service requests.73 An online handoff cost and satisfy user QoS requirement in RAN slic- 709
648 control framework is designed to make decisions on UAV dis- ing.88 In addition, an RL-based slicing allocation scheme is 710
649 patching, request admission, scheduling, and resource slicing proposed from the perspective of the physical infrastructure 711
650 for different services. Kasgari et al. proposed a Lyapunov provider to maximize long-term network utility.36 Traditional 712
10.1016/j.cja.2021.12.013
3 January 2022
713 RL methods are not applicable in heterogeneous resource allo- Ground (S2G) links, and Ground-to-Space (G2S) links. Subse- 741
714 cation (especially in SAGINs) with a large action space, which quently, these VNFs are embedded into physical resources by 742
715 may result in intolerable convergence time.37,39 Based on con- network operators, and then services will be supported by the 743
716 ventional Deep Q-Learning (DQL), discrete normalized physical network. Particularly, the network controlling and 744
717 advantage functions are introduced into DQL to reduce com- forwarding can be separated into three logical planes by 745
718 putation complexity.38 SDN techniques as data plane, control plane, and application 746
plane, where the data transmission can be carried out across 747
719 4.2. SDN/NFV planes by interfaces.54 Due to the programmability of SDN, 748
the network configuration can be adaptive to different require- 749
720 4.2.1. Overview of SDN/NFV ments of services, thereby simplifying network management 750
721 To cope with more diverse services and higher QoS require- and upgrading and improving service diversity and network 751
722 ments in future service-oriented networks, the network hard- flexibility.90 With NFV, heterogeneous network resources 752
of
723 ware equipment may need to be updated frequently since the can be isolated from each other, which is convenient for the 753
724 network infrastructure is only able to provide dedicated ser- management and transplantation with on-demand services. 754
725 vices, leading to intolerable costs for network operators. Besides, network functions can be decoupled by NFV tech- 755
726 Besides, the integration of space, aerial, and ground segments niques from its dedicated hardware devices to realize the flex- 756
ible deployment of network functions.91–93 In SAGINs, SDN/ 757
o
727 is challenging due to their differentiation in protocols,
728 resources, and network environment. NFV technologies are mainly exploited in three different ways. 758
729 SDN/NFV-based network architecture is investigated, SDN/NFV in satellite networks: Considering that satellites 759
have on-board processing and switching capabilities, it is con- 760
Pr
730 which shows a promising solution for building networks with
731 on-demand service in Fig. 4 (redrew from 90).43,52,89 When venient to employ SDN-enabled switches that are responsible 761
732 the service request, such as Artificial Intelligence of Things for executing commands from the controller. An SDN-based 762
733 (AIoT), from the user is received, it will be described into a satellite network architecture, named SDSN, is proposed.43 763
734 specific sequence of VNFs, named Service Function Chain In the SDSN, the data plane is composed of satellite switches, 764
735 (SFC), which are composed of corresponding virtual resource which are responsible for simple stream-based packet forward- 765
736 blocks (containing communication, computation and storage

ed ing, the control plane is composed of controllers located in the 766
737 resources). Considering the real-time network states, the SFC ground station, and control information is forwarded through 767
738 can be mapped into physical network links, containing Geostationary Earth Orbit (GEO). With SDN techniques, bet- 768
739 Space-to-Air (S2A) links, Air-to-Space (A2S) links, Air-to- ter performance can be achieved with guaranteed diversified 769
740 Ground (A2G) links, Ground-to-Air (G2A) links, Space-to- user requirements and effective satellite resource utilization; 770
ct
re
or
nc
U
Fig. 4 SDN/NFV-based network architecture in SAGINs.
10.1016/j.cja.2021.12.013
3 January 2022
10 N. CHENG et al.
771 SDN/NFV in terrestrial networks: With SDN/NFV tech- resources including virtual computing, storage and network 829
772 niques, the programmability and reconfiguration capabilities resources, and non-virtual resources.47 830
773 of each network component can be improved in terrestrial net-
774 work. For example, SDN-based architecture in terrestrial net- (2) Network management 831
775 works can make gateway switching and stream-based data 832
776 forwarding more flexible and intelligent.42 Due to the separation of control and data planes and the 833
777 SDN/NFV in backhaul networks: In SAGINs, there might programmability, SDN networks can achieve more flexible 834
778 be a significant increase in the number of routing paths due to and efficient network management, which exactly conforms 835
779 the involvement of space networks and aerial networks in rout- to the on-demand network configurability requirement of 836
780 ing. Packet routing path selection and how to avoid congestion service-oriented SAGINs. 837
781 or idleness in path selection is a key issue to improve the over-
782 all network performance. The central SDN controller can (A) SDN-based overall network management 838
of
783 obtain information of the entire network and make a globally 839
784 optimal path selection according to the overall network status Bao et al. proposed an SDN-based satellite network archi- 840
785 to reduce unnecessary link switching. However, in SAGINs, tecture, termed as OpenSAN, which decouples the data and 841
786 the information collection itself might be extremely inefficient control planes.49 The data plane is formed by routers and 842
787 due to dynamic topology, unstable links, and long propagation multi-layer satellites around the world, which is responsible 843
o
788 delays. for data forwarding for different services. The control plane 844
is composed of GEO satellites, which reduces the number of 845
789 4.2.2. Related research required ground stations and reduces the complexity; the man- 846
Pr
790 SDN/NFV has been widely used to enhance the resource agement plane is responsible for mobility management. 847
791 orchestration, network management and mobility manage- Through this layered network, the complexity of the network 848
792 ment in the SAGINs, which can be summarized as follows. management can be reduced and the flexibility of the network 849
can be improved to support advanced technologies in future 850
793 (1) Resource orchestration network. Zhang et al. also proposed a SDN-based SAGINs 851
794 (A) SDN-based overall resource orchestration

ed architecture.86 To improve the overall network management 852
795 efficiency, a new operationally responsive space (ORS) satellite 853

networking scheme based on SDN is proposed, which adopts a 854
796 To improve the resource utilization, Wu et al. proposed an lightweight architecture consisting of a physical layer, a con- 855
797 SDN-based satellite network architecture where the data plane trol layer, and an application layer.50 Specifically, the physical 856
798 is composed of satellite switches, and the control plane is com- layer defines the standards of network equipment, the control 857
799 posed of controllers located in the ground station.43 To guar- layer maintains the network status centrally by communicating 858
ct
800 antee the flexibility, openness, interoperability, and with the physical layer through OpenFlow, and the application 859
801 evolvability of the network, a SDN-based aerial backbone net- layer provides users with network services, such as data trans- 860
802 work architecture is proposed.42 Particularly, the segmented mission protocols, satellite networking management protocols, 861
803 routing by SDN is designed to improve the transmission relia- etc. Secinti et al. proposed an aerial network management pro- 862
re
804 bility and bandwidth utilization by balancing network traffic tocol based on SDN where each drone is deployed as an SDN 863
805 load among multiple reliable transmission paths. For the switch which runs under the instructions issued by the central 864
806 transmission efficiency of both control and data planes, com- controller and calculates the drone’s trajectory.51 Results show 865
807 plex network management operations through SDN are trans- that this protocol outperforms benchmark protocols in con- 866
808 ferred to a cloud-based synchronous data network controller, trolling end-to-end interrupts, and the end-to-end delay is also
or
867
809 where the computational burden of mobile nodes and drones reduced. In addition, to address the dynamic network topology 868
810 is reduced and the utilization of wireless channel resources and stochastic real-time service requirements in satellite net- 869
811 can be improved.10 Different from the network with indepen- works, a network management based on Viterbi algorithm is 870
812 dent network segments, an SDN-based SAGIVN is investi- proposed, where the service function chain (SFC) mapping 871
nc
813 gated and a hybrid and layered control architecture is process is established as a Markov model and the VNF of 872
814 designed, where an AI-based solution is proposed to promote the overloaded node is considered.52 This work reveals that 873
815 effective network slicing, mobility management, and content better network management performance can be achieved by 874
816 caching, realizing flexible, reliable, and scalable network migrating resources to other idle nodes. 875
817 resource managements.43
(B) NFV-based network management
U
876
818 (B) NFV-based resource orchestration 877
819 For the network management, the network deployment and 878
820 In addition to the abovementioned research on the overall intelligent management based on NFV and the NFV architec- 879
821 network design based on SDN, there are also studies on the ture will be compatible with legacy networks.94 NFV can also 880
822 use of NFV to implement resource orchestration for specific be integrated into specific networks to reduce the use of special 881
823 networks. NFV is used to build a configurable and scalable equipment and facilitate network upgrades.53 882
824 network and combines bandwidth resources to design a
825 resource allocation strategy for high-density scenarios, e.g., (3) Mobility management 883
826 stadiums.46 An optimal resource allocation problem in multi- 884
827 domain NFV system is investigated with the competition of The mobility of network nodes, especially LEO satellites 885
828 multiple NFV service providers that owns different types of which have high-speed relative movement with earth, results 886
10.1016/j.cja.2021.12.013
3 January 2022
887 in a dynamic network topology in SAGINs. Due to the con- the users to the cloud servers. To address this issue, the 943
888 stant change of node positions, the connection status between MEC is proposed to decentralize cloud services to the network 944
889 the nodes and networking performance are also changing, edge to achieve lower latency,96,97 where MEC servers can be 945
890 which leads to significant issues such as performance degrada- deployed in access networks such as on base stations.98,99 Par- 946
891 tion, increasing overhead, and network outage. ticularly, the MEC can deploy services in advance, which 947
shortens the distance between resources and users.100 By inte- 948
892 (A) SDN-based mobility management grating cloud and edge resources, the energy consumption of 949
893 terminals, core network traffics, and the network security 950
894 SDN-based networks have the ability to collect information and scalability can be better addressed. In Fig. 5, a heteroge- 951
895 and regulate global resources. To achieve effective mobility neous and multi-layered cloud-edge synergy architecture is 952
896 management, a prototype system based on SDN is designed.39 demonstrated for service-oriented SAGINs. From the perspec- 953
897 The system predicts future events (e.g., link down) through the tive of on-demand services, the operations and functions of the 954
of
898 trajectory of aerial nodes, and then reduces the impact of net- cloud-edge synergy architecture is described as follows. The 955
899 work events through SDN-based optimization, thereby edge servers and controllers can be deployed at BSs, UAVs, 956
900 improving the aerial network performance and availabil- LEO satellites, etc. Computation-intensive tasks can be off- 957
901 ity.54,55 A seamless handover mechanism based on SDN is pro- loaded from user devices to the edge servers for higher energy 958
902 posed, where the ground controller is responsible for the efficiency and lower computing delay. Furthermore, service 959
o
903 collection of overall network status information and the caching schemes can be applied to cache the application ser- 960
904 LEO satellite is only responsible for the exchange in the data vices and related databases such that the user tasks which 961
905 plane.41 Particularly, the user terminal can connect to multiple require these services can be executed.101 To enable these func- 962
Pr
906 satellites at the same time, where the link can be changed tionalities, the SAGINs edge controllers should be carefully 963
907 according to the signal strength to reduce the impact of satel- deployed to facilitate efficient data collection, and the UAV 964
908 lite and user movement. Besides, a satellite resource sharing networking and trajectory should be optimized to provide bet- 965
909 game model based on the bipartite graph framework is ter link performance and network coverage. The cloud servers 966
910 designed, which selects suitable satellites and frequency bands can be deployed in terrestrial data centers, which can be 967
911 for mobile terminals to maximize the terminal revenue, and ed accessed even from rural areas via SAGINs global coverage. 968
912 designs random access algorithms based on the goal of maxi- Through cloud-edge synergy, the service performance can be 969
913 mizing user space (i.e., the feasible area of the subsequent optimized by coordinating the advantages and disadvantages 970
914 access terminals).56 of both edge and cloud.102 However, due to the large network 971
scale and high mobility, it is important to design efficient 972
915 (B) NFV-based mobility management cloud-edge synergy mechanisms for timely and performance 973
916 guaranteed service fulfillment. 974
ct
917 By NFV technologies, the life cycle of network functions
918 and its constituent resources can be managed, which is comple- 5.2. Related research 975
919 mentary to SDN. Prados-Garzon et al. proposed a theoretical
920 framework to evaluate the performance of the virtualized Cloud-edge synergy techniques are originally designed for ter- 976
re
921 Mobility Management Entity computing system (vMME) restrial networks, and are recently applied by some research- 977
922 based on NFV.95 Particularly, the virtual part hypervisor of ers, into SAGINs. In the following, we begin with some 978
923 the vMME system combines three logical components: the latest research works on terrestrial cloud-edge synergy, and 979
924 front end, management entity service logic, and State Database then review the cloud-edge synergy in SAGINs. 980
(SDB). By predicting the number of given user equipment and
or
925
926 MTC equipment, the communication network can be updated 5.2.1. Cloud-edge synergy in terrestrial network 981
927 in advance to cope with the high mobility. An Evolved
The most important function of cloud-edge synergy is to deli- 982
928 virtualization-based Packet Core (EPC) entity is presented,
ver network services in different locations according to the ser- 983
929 which is divided into groups according to the control traffic
vice capabilities, in order to achieve service fulfillment in terms 984
nc
930 where entities that exchange large amounts of traffic are

of different metrics, such as delay, traffic pressure, security, 985
931 grouped into the same group, i.e., mobility management entity
etc. Most of the work focuses on the optimization of service 986
932 and the home user server entity grouping.91 In addition, the
latency, since the most important motivation of introducing 987
933 authentication and authorization processes are executed inter-
edge computing is to reduce the overall latency. To reduce 988
934 nally without the need for network data transmission, which
the task processing latency, Kai et al. proposed a hierarchical 989
further reduces the impact of mobility.
U
935
computing scheme with pipeline offloading, where the task 990
computing and offloading execute simultaneously.58 By using 991
936 5. Cloud-edge synergy
the multi-layer servers and the cloud data center, a RL-based 992
computing strategy is designed, which achieves a lower service 993
937 5.1. Overview latency compared with the pure cloud computing and edge 994
computing strategies.103 In addition, a Federated Learning 995
938 Cloud servers can provide abundant cloud resources accu- (FL) framework is designed in MEC systems for the model 996
939 rately according to service requests to guarantee the user training of deep RL, based on which, the dynamic resources 997
940 QoS. However, a soaring number of time-sensitive services (e.g., communication, storage, and computing) optimization 998
941 require the rigorous latency, and the cloud-based technologies strategy is deployed.104 Service deployment enables VNF to 999
942 can hardly ensure it due to the long transmission paths from instantiate in separate nodes such as the edge and cloud ser- 1000
10.1016/j.cja.2021.12.013
3 January 2022
12 N. CHENG et al.
o of
Pr
Fig. 5 Cloud-edge synergy architecture in service-oriented SAGINs.
ed
1001 vers, which reduces service latency and improves network UAV networks and collaborative cloud-edge task scheduling 1034
1002 profit. Some existing research works have been conducted in via SAGINs. 1035
1003 relatively simple scenarios with only cloud servers59 or multi- Due to the flexible networking capability, UAV networks 1036
1004 access edge computing servers. 60,105–107 An integrated archi- can provide on-demand caching and computing resources by 1037
tecture, which considers both the cloud and the edge and is controlling the trajectory and orchestrating the resources of
ct
1005 1038
1006 compliant with the ETSI,108 is presented to minimize the UAVs. A UAV caching scheme is studied by Chen et al., where 1039
1007 sum of latency by balancing loads among mobile edge plat- QoE is the main concern when the deployment of UAVs is 1040
1008 forms and edge virtualization computing infrastructure, and designed.113 A ML framework named echo state networks is 1041
a new Tabu Search-based algorithm is presented to find proposed to predict the users’ content request distribution
re
1009 1042
1010 near-optimal solutions.109 Miscellaneous functions of cloud- and mobility pattern, based on which the optimal user-UAV 1043
1011 edge synergy have also been studied. To sufficiently use the full association is derived. To make an efficient computation task 1044
1012 computing capacity of BSs and cloud, a task splitting strategy offloading decision considering the constrained energy at 1045
1013 is proposed to divide the task into several subtasks, and then UAVs, the dynamic offloading is formulated as a Constrained 1046
or
1014 the divided subtasks are processed in the edge nodes and cloud Markov Decision Process (CMDP), which is solved by the lin- 1047
1015 severs seperatively.57 Conventional Machine Learning (ML) ear programming. The lowest task processing delay is achieved 1048
1016 needs user data to train models, which may result in user pri- comparing to the random scheme which executes actions ran- 1049
1017 vacy leaking. FL enables the ML model to be trained locally domly and the greedy scheme that selects BS prior to offload 1050
1018 and upload the trained model data instead of raw data, which more tasks.114 Furthermore, a comprehensive scheduling of 1051
nc
1019 protects user privacy and saves network resources.110 An asyn- association control, computation task allocation, transmission 1052
1020 chronous FL mechanism is proposed to preserve privacy in power and bandwidth allocation, UAV computation resource, 1053
1021 MEC systems, and the network node can join or leave flexibly, and deployment position is conducted to minimize the maxi- 1054
1022 which is suitable in highly dynamic SAGIN scenarios.111 Shao mum computation delay among IoT devices.115 Considering 1055
1023 et al. proposed a learning-based framework with an integra- the migration cost due to the user mobility, a joint task 1056
U
1024 tion of stochastic simulation, neural network, and genetic algo- offloading and migration scheme is proposed in RL-based 1057
1025 rithm, to execute edge server deployment in BSs to minimize mobility-aware MEC network to achieve the maximum system 1058
1026 demands-weighted distance. 112 Simulations have shown that revenue.116 1059
1027 the algorithm is better than the Random Fit algorithm, Top- Owing to the seamless coverage and abundant computing/- 1060
1028 first algorithm, and the K-means algorithm. caching resources provided by SAGINs, the cloud-edge syn- 1061
ergy can be further enhanced, even in the areas where it is 1062
1029 5.2.2. Cloud-edge synergy in SAGINs difficult or cost-ineffective to deploy terrestrial BSs. However, 1063
1030 In SAGINs, the cloud-edge synergy can be further extended the high mobility, heterogeneous resources, and constrained 1064
1031 since the user can access the edge and cloud through ubiqui- interoperability will make the collaborative cloud-edge task 1065
1032 tous coverage. The existing works mainly focus on two scheduling more challenging in SAGINs. Due to the wide 1066
1033 research directions, i.e., edge caching/computing through broadcasting and coverage of satellites, MEC services can be 1067
10.1016/j.cja.2021.12.013
3 January 2022
1068 provided by satellite networks at a low cost. To provide the tomized services with distinctive requirements. Worse still, 1124
1069 low latency ubiquitous services in SAGINs, an approach with since the network environment and contexts also significantly 1125
1070 joint request dispatching and service placement is proposed, affect the service requirements, the wide variety of network 1126
1071 which shows better performance comparing to the greedy algo- scenarios is a critical issue in on-demand service provisioning. 1127
1072 rithm in terms of the service user ratio and average hop Although some initial works have been conducted to address 1128
1073 count.63 Zhou et al. proposed an SAGIN-based bidirectional the dynamic scenario and service, e.g., a graph neural network 1129
1074 mission offloading framework, which remedies the deficiency is proposed to address the scalable radio resource management 1130
1075 of terrestrial networks in coverage and space-aerial networks that is difficult for convention deep neural networks with fixed 1131
1076 in power and other resources, to achieve substantial network number of inputs and outputs,120 a general framework and 1132
1077 gain in reliability and cost reduction.117 The latency and methodologies for service-oriented network is still missing. 1133
1078 energy optimization problem in SAT-IoT networks is formu- Therefore, the service definition and identification should be 1134
1079 lated as a dynamic mixed-integer programming problem, first addressed in service-oriented networks. Generally speak- 1135
of
1080 where a two-stage optimization is conducted, including the ing, two important functions need to be investigated, i.e., ser- 1136
1081 computing and communication resource allocation and the vice description framework and service sensing. Service 1137
1082 joint optimization of user association and offloading, which description should provide a generalized framework to repre- 1138
1083 are solved by the Lagrange multiplier method and the deep sent the features and requirements of the network services, 1139
1084 RL-based method, respectively.62 Learning-based models can which serves as the theoretical foundation for service identifi- 1140
o
1085 adjust resource orchestration and network configuration to cation and on-demand service fulfillment. Currently, the ser- 1141
1086 adapt to real-time network status. Considering the energy con- vice description mainly exploits classification-based methods, 1142
1087 straint and computation constraint of remote devices, an such as the three use cases in 5G network and the model- 1143
Pr
1088 SAGIN edge/cloud computing architecture is proposed for based service orchestration in off-the-shelf network manage- 1144
1089 offloading the computation-intensive applications, where ment systems.121 As this method lacks accuracy and flexibility 1145
1090 UAVs provide the near-user edge computing and satellites pro- to describe 6G customized services, a more general method is 1146
1091 vide the cloud computing.102 Based on this framework, a thus required. Potential solutions can borrow the ideas from 1147
1092 learning-based computational offloading method is further the field of ontology engineering. An ontology is defined as 1148
1093 proposed, which dynamically learns the optimal offloading ed ‘‘a formal representation of knowledge as a set of concepts 1149
1094 strategy to minimize the total task delay, energy consumption, within a domain, and the relationships between those con- 1150
1095 and server usage costs. To make task scheduling decisions cepts”.122 By using ontology engineering, it is feasible to 1151
1096 online, a deep risk-sensitive RL-based algorithm is proposed explicitly represent the network services as a set of concepts, 1152
1097 to minimize the long-term average delay of IoT services in for example, the environment, network context, requirement, 1153
1098 SAGINs, where the low-latency and extended service lifespan the interrelation, taxonomy, and rules between them accord- 1154
1099 of IoT services can be achieved.118 Considering a cache- ingly. Currently, the ontology-based network management 1155
ct
1100 enabled satellite-UAV-vehicle integrated network where the has been investigated,123–125 however, application of ontology 1156
1101 GEO satellite serves as the cloud server and the UAVs are in defining and fulfilling network services needs to be further 1157
1102 deployed as the edge caching servers, an energy-aware coded studied. 1158
1103 caching strategy is proposed to provide more multicast oppor- A service needs to be sensed before identified, which 1159
re
1104 tunities and significantly reduce the total energy involves the information sensing of service features and 1160
1105 consumption.119 requirements, network conditions, environments, etc., and 1161
1106 Although the cloud-edge synergy in SAGINs can flexibly the transition of the information from the physical world to 1162
1107 deploy services in multi-tier networks, considering different the digital world. There are two potential solutions, i.e., joint 1163
server capacity and the dynamic topology, it is challenging sensing-communication technique and digital/cyber twin. Sen-
or
1108 1164
1109 to perform the global task scheduling. In addition, the place- sor fusion, which combines network sensing (the detection of 1165
1110 ment of servers also affects the service performance. The loca- communication environment using radio signals) and other 1166
1111 tion, quantity, and density of cloud/edge servers should be ubiquitously deployed sensors,126 will provide complete sce- 1167
1112 optimized.103 However, both users and access nodes/servers nario data for 6G service-oriented network.127 With the net- 1168
nc
1113 in SAGINs are dynamic, which may render solution optimal work context awareness, the communication services could 1169
1114 to sub-optimal. Therefore, for cloud-edge synergy-based be fulfilled more efficiently with guaranteed performance. 1170
1115 SAGINs, how to deploy edge cloud servers, whether to offload Recently, a joint sensing-communication cooperative sensing 1171
1116 tasks to the cloud or the edge, and who manages the offloading UAV network is proposed where UAV equipment can gener- 1172
1117 of tasks are issues that must be resolved in the future. ate a sensing beam and a communication beam orthogonally 1173
to enhance the spectrum reuse.128 Cybertwin is the digital twin
U
1174
1118 6. Future research directions in the communication network systems, which is the digital 1175
representation of humans, things, and processes in the virtual 1176
1119 In this section, some potential future research directions in cyberspace.24 Cybertwin can offer abundant information 1177
1120 service-oriented SAGINs are discussed. required in identifying the network services. For example, 1178
the end user can connect its cybertwin with the required ser- 1179
1121 6.1. Service definition and identification vice, and the cybertwin will integrate the network context 1180
and abundant historical data to evaluate the accurate service 1181
requirements for the time being and in the future. However, 1182
1122 As discussed in Section 2.3, the 6G service-oriented network the research on either technique is still in its infant stage, espe- 1183
1123 faces the challenge of handling diverse, long-tail, and cus- cially when applied in service identification. 1184
10.1016/j.cja.2021.12.013
3 January 2022
14 N. CHENG et al.
1185 6.2. AI-based networking paradigms cious interference, masquerade attack, passive eavesdrop, mas- 1242
sage intercept, etc.92,93 Although the isolation is generally 1243
1186 Pervasive network intelligence will be one distinctive feature of considered between different network slices, the vulnerability 1244
1187 6G network, which not only increases the service diversity, but could penetrate through tenants with different service require- 1245
1188 also more importantly, offers higher capability to handle on- ments, due to different security isolation levels among slices. 1246
1189 demand services in a more flexible, efficient, adaptive, auto- With the limited resources at satellite and the wide coverage 1247
1190 matic, and cost-effective way. Therefore, developing advanced of satellite beams, malicious attacks are easy to hide and diffi- 1248
1191 AI-based service-fulfillment technologies is of great signifi- cult to eliminate. Recently, the physical layer security as a 1249
1192 cance. Specifically, such technologies have the potential to lightweight secure approach has shown promising solutions 1250
1193 address the following two important issues in service- for secure transmission of such massive connections in 1251
1194 oriented network, i.e., massive data exploitation by big data SAGINs, which is a key-free security approach based on 1252
1195 techniques and service identification by intent-based information-theoretic security.132 However, the security is 1253
of
1196 networking: guaranteed separately between heterogeneous multi-tier net- 1254
1197 Massive network data mining - The 6G SAGINs generate work, directing a research issue of compatible security 1255
1198 massive network data from network nodes, operations, and approach in SAGINs, e.g., symbiotic security. In addition, 1256
1199 ubiquitous deployed sensors, which if appropriately exploited, considering the centralized control mode of SDN, the SDN 1257
controller charges the control function and status information 1258
o
1200 will significantly enhance the network and service perfor-
1201 mance. First, the data contains valuable information for ser- of the overall network, while the SDN controller is easy to be 1259
1202 vice identification, such as the service type, requirements, targeted by hackers, and the inherent security is flawed to the 1260
network. Particularly, an attacker can forge many packets that 1261
Pr
1203 user preference, and network contexts. A big data and ML-
1204 based network context utilization process is described, which the switch cannot process according to the characteristics of 1262
1205 includes context acquisition, context modelling, context rea- the protocol, forcing the switch to request many processing 1263
1206 soning, and context distribution.129 Second, network data by the controller, thus the communication and computing 1264
1207 can facilitate automatic service management. The life cycle of resources are occupied. Block-chain technology has the char- 1265
1208 services includes the service monitoring, fault detection, heal, acteristics of decentralization, non-tampering and anonymity, 1266
1209 and scale functions, which can result in very high cost in time
ed which realizes information recording without a trusted third 1267
1210 and money if handled by humans. Therefore, through service party, and can guarantee the security of data in the process 1268
1211 data monitoring and analysis, the abnormal events can be of collection, transmission, storage, and calculation. Some 1269
1212 inferred and coped with in a fully automatic manner, which documents point out that blockchain technology can guaran- 1270
1213 significantly saves the cost and enhances the service perfor- tee the data and privacy security of MEC and SAGINs.133 1271
1214 mance. With the fast development of big data and ML, pow- Blockchain can provide security for MEC and SAGINs, but 1272
1215 erful big data tools have been emerging recently, such as may cause delays due to sequential validation requests and 1273
ct
1216 distributed storage and process, data analytics, data visualiza- consume large amounts of storage resources.134 The cloud ser- 1274
1217 tion, etc. However, utilizing big data techniques in service- ver happens to have computing and storage capabilities, which 1275
1218 oriented network needs further investigation. can solve the computing delay and storage problems caused by 1276
1219 Intent-based networking - Advances in neural language blockchain. Therefore, it is promising to use blockchain tech- 1277
re
1220 understanding and deep learning algorithms have facilitated nology to guarantee the security of SAGINs based on cloud- 1278
1221 the development of knowledge which can be used to covert edge synergy architecture.135 1279
1222 user queries in a given language (i.e., intents) into a structured

1223 representation that can be processed automated services.130 7. Conclusions 1280
or
1224 Through intent-based techniques, the network service, which

1225 might be ambiently expressed by users’ intent, can be explicitly In this paper, a comprehensive study of 6G SAGINs has been 1281
1226 and accurately identified. A three-layer intent-based network conducted from a service-oriented perspective. The require- 1282
1227 management system is proposed.131 Within the architecture, ments of service-oriented network have been presented based 1283
1228 the intent layer is responsible for translating the service from on what the service-oriented SAGINs architecture has been 1284
nc
1229 user intents in network layer to key performance indicators, proposed. Two categories of enabling key technologies, hetero- 1285
1230 service level agreements, processes, and targets in business geneous resource orchestration technologies and cloud-edge 1286
1231 layer, which is an intent-based automation process with less synergy technologies, have been surveyed and discussed. The 1287
1232 human workforce. However, the intent-based techniques have future research directions are finally discussed. 1288
1233 not been thoroughly studied, and their ability to identify the
U
1234 wide variety of SAGINs services remains doubtful and needs Funding 1289
1235 further investigation.
This work was supported by the National Key Research and 1290
1236 6.3. Security
Development Program of China (2020YFB1807700). 1291
1237 For the proposed service-oriented SAGINs architecture, new Declaration of Competing Interest 1292
1238 networking security issues will be exposed due to its inherent
1239 networking features. Considering the open connectivity and The authors declare that they have no known competing 1293
1240 dynamic topology, SAGINs are vulnerable to security threats financial interests or personal relationships that could have 1294
1241 for the random access of multi-mode terminals, including mali- appeared to influence the work reported in this paper. 1295
10.1016/j.cja.2021.12.013
3 January 2022
1296 References 23. Pang L, Yang C, Chen D, et al. A survey on intent-driven 1362
networks. IEEE Access 2020;8:22862–73. 1363
1297 1. 3GPP. Study on New Radio (NR) access technology physical 24. Yu Q, Ren J, Fu Y, Li Y, Zhang W. Cybertwin: An origin of next 1364
1298 layer aspects. 2017. Standard No.: TR 38.804. generation network architecture. IEEE Wirel Commun 2019;26 1365
1299 2. ITU-R M. [IMT-2020.TECH PERF REQ]—minimum require- (6):111–7. 1366
1300 ments related to technical performance for IMT-2020 radio 25. He HL, Ren PY, Du QH, et al. Traffic-aware overload control 1367
1301 interface(s). 2017. Standard No.: M.2410-0 (11/2017). scheme in 5G ultra-dense M2M networks. Trans Emerging Tel 1368
1302 3. Wu Q, Ding G, Xu Y, et al. Cognitive Internet of Things: a new Tech 2017;28(9) e3146. 1369
1303 paradigm beyond connection. IEEE Internet Things J 2014;1 26. de Cola T, Bisio I. QoS optimisation of eMBB services in 1370
1304 (2):129–43. converged 5G-satellite networks. IEEE Trans Veh Technol 1371
1305 4. Shen X, Cheng N, Zhou H, et al. Space-air-ground Integrated 2020;69(10):12098–110. 1372
1306 Networks: review and prospect. Chin J Internet Things 2021;4 27. Höyhtyä M, Ojanperä T, Mäkelä J, et al. Integrated 5G satellite- 1373
1307 (3):3–19. terrestrial systems: use cases for road safety and autonomous 1374
ships23rd Ka and Broadband Communications Conference. p. 1–9. 1375
of
1308 5. Niu Z, Shen XS, Zhang Q, et al. Space-air-ground integrated
1309 vehicular network for connected and automated vehicles: chal- 28. Hendaoui S, Mannai A, Zangar N. Cognitive CQI/5QI based 1376
1310 lenges and solutions. Intell Converged Netw 2020;1(2):142–69. scheme for software defined 5G hybrid satellite-terrestrial 1377
1311 6. Qu Z, Zhang G, Cao H, et al. LEO satellite constellation for network: Slicing for ultra reliability and video congestion 1378
1312 Internet of Things. IEEE Access 2017;5:18391–401. offloading. 2020 International Symposium on Networks, Comput- 1379
o
1313 7. You XH, Wang CX, Huang J, et al. Towards 6G wireless ers and Communications (ISNCC); 2020 Oct 20-22; Montreal, 1380
1314 communication networks: Vision, enabling technologies, and new Canada. Piscataway: IEEE Press; 2020. p.1–7. 1381
1315 paradigm shifts. Sci China Inf Sci 2020;64(1):1–74. 29. Ahmed T, Alleg A, Ferrus R, et al. On-demand network slicing 1382
using SDN/NFV-enabled satellite ground segment systems. 2018 1383
Pr
1316 8. Mohammed A, Mehmood A, Pavlidou FN, et al. The role of
1317 high-altitude platforms (HAPs) in the global wireless connectiv- 4th IEEE Conference on Network Softwarization and Workshops 1384
1318 ity. Proc IEEE 2011;99(11):1939–53. (NetSoft); 2018 Jun 25-29; Montreal, Canada. Piscataway: IEEE 1385
1319 9. Xiong F, Li A, Wang H, et al. An SDN-MQTT based Press; 2018.p.242–6. 1386
1320 communication system for battlefield UAV swarms. IEEE 30. Pavon IL, Sánchez JR, Colazzo A. 5G NORMA network 1387
1321 Commun Mag 2019;57(8):41–7. architecture—intermediate report. EU H2020 5G NORMA. 1388
1322 10. Yuan ZH, Huang XW, Sun LL, et al. Software defined mobile Tech Rep Deliverable D 2017. 1389
sensor network for micro UAV swarm. 2016 IEEE International 31. Cao SZ, Wei JY, Han H, et al. Space edge cloud enabling 1390
1323
1324
1325
Conference on Control and Robotics Engineering (ICCRE); 2016
Apr 2-4; Singapore. Piscataway: IEEE Press; 2016.p.1–4.
ed network slicing for 5G satellite network2019 15th International
Wireless Communications & Mobile Computing Conference
(IWCMC); 2019 Jun 24-28; Tangier, Morocco. Piscataway: IEEE
1391
1392
1393
1326 11. Cheng N, Quan W, Shi W, et al. A comprehensive simulation
1327 platform for space-air-ground integrated network. IEEE Wirel Press; 2019. p. 787–92. 1394
1328 Commun 2020;27(1):178–85. 32. NGMA Alliance, Description of network slicing concept, 2016. 1395
1329 12. Shi W, Li J, Cheng N, et al. Multi-drone 3-D trajectory planning 33. Dong F, Han H, Gong X, et al. A constellation design 1396
ct
1330 and scheduling in drone-assisted radio access networks. IEEE methodology based on QoS and user demand in high-altitude 1397
1331 Trans Veh Technol 2019;68(8):8145–58. platform broadband networks. IEEE Trans Multimed 2016;18 1398
1332 13. Ma T, Zhou HB, Qian B, et al. A large-scale clustering and 3D (12):2384–97. 1399
1333 trajectory optimization approach for UAV swarms. Sci China Inf 34. Zhao ZL, Bai J, Tie ZZ, et al. Network slicing technique assisted 1400
resource allocation in small-cell networks2019 IEEE 19th Inter- 1401
re
1334 Sci 2021;64(4):1–16.

1335 14. Shi W, Li J, Xu W, et al. Multiple drone-cell deployment analyses national Conference on Communication Technology (ICCT); 2019 1402
1336 and optimization in drone assisted radio access networks. IEEE Oct 16-19; Xi’an, China. Piscataway: IEEE Press; 2019. p. 843–8. 1403
1337 Access 2018;6:12518–29. 35. Chen Y, Wang Y, Liu M, et al. Network slicing enabled resource 1404
1338 15. Cheng N, Xu W, Shi W, et al. Air-ground integrated mobile edge management for service-oriented ultra-reliable and low-latency 1405
vehicular networks. IEEE Trans Veh Technol 2020;69(7):7847–62. 1406
or
1339 networks: architecture, challenges, and opportunities. IEEE

1340 Commun Mag 2018;56(8):26–32. 36. Li R, Zhao Z, Sun Q, et al. Deep reinforcement learning for 1407
1341 16. Qiu J, Grace D, Ding G, Zakaria MD, Wu Q. Air-ground resource management in network slicing. IEEE Access 1408
1342 heterogeneous networks for 5G and beyond via integrating high 2018;6:74429–41. 1409
1343 and low altitude platforms. IEEE Wirel Commun 2019;26 37. Shen X, Gao J, Wu W, et al. AI-assisted network-slicing based 1410
next-generation wireless networks. IEEE Open J Veh Technol 1411
nc
1344 (6):140–8.
1345 17. Zhang S, Quan W, Li J, et al. Air-ground integrated vehicular 2020;1:45–66. 1412
1346 network slicing with content pushing and caching. IEEE J Sel 38. Sciancalepore V, Costa-Perez X, Banchs A. RL-NSB: Reinforce- 1413
1347 Areas Commun 2018;36(9):2114–27. ment learning-based 5G network slice broker. IEEE/ACM Trans 1414
1348 18. Connected vehicles in smart cities: The future of transportation. Netw 2019;27(4):1543–57. 1415
1349 Available form: https://interestingengineering.com/. 39. Qi C, Hua Y, Li R, et al. Deep reinforcement learning with 1416
discrete normalized advantage functions for resource manage- 1417
U
1350 19. Ding G, Wu Q, Zhang L, et al. An amateur drone surveillance

1351 system based on the cognitive Internet of Things. IEEE Commun ment in network slicing. IEEE Commun Lett 2019;23(8):1337–41. 1418
1352 Mag 2018;56(1):29–35. 40. Chen X, Zhang H, Wu C, et al. Optimized computation 1419
1353 20. Casoni M, Grazia CA, Klapez M, et al. Integration of satellite offloading performance in virtual edge computing systems via 1420
1354 and LTE for disaster recovery. IEEE Commun Mag 2015;53 deep reinforcement learning. IEEE Internet Things J 2019;6 1421
1355 (3):47–53. (3):4005–18. 1422
1356 21. Nakamura T. 5G evolution and 6G. 2020 IEEE Symposium on 41. Yang B, Wu Y, Chu X, et al. Seamless handover in software- 1423
1357 VLSI Technology; 2020 Jun 16-19; Honolulu, USA. Piscataway: defined satellite networking. IEEE Commun Lett 2016;20 1424
1358 IEEE Press; 2020. p. 1–5. (9):1768–71. 1425
1359 22. Zilberman N, Watts PM, Rotsos C, et al. Reconfigurable 42. Chen K, Zhao S, Lv N, et al. Segment routing based traffic 1426
1360 network systems and software-defined networking. Proc IEEE scheduling for the software-defined airborne backbone network. 1427
1361 2015;103(7):1102–24. IEEE Access 2019;7:106162–78. 1428
10.1016/j.cja.2021.12.013
3 January 2022
16 N. CHENG et al.
1429 43. Wu H, Chen J, Zhou C, et al. Resource management in space-air- 61. Khalid O, Khan MUS, Huang Y, et al. EvacSys: A cloud-based 1496
1430 ground integrated vehicular networks: SDN control and AI service for emergency evacuation. IEEE Cloud Comput 2016;3 1497
1431 algorithm design. IEEE Wirel Commun 2020;27(6):52–60. (1):60–8. 1498
1432 44. Faraci G, Grasso C, Schembra G. Reinforcement-learning for 62. Cui G, Li X, Xu L, et al. Latency and energy optimization for 1499
1433 management of a 5G network slice extension with UAVsIEEE MEC enhanced SAT-IoT networks. IEEE Access 1500
1434 INFOCOM 2019 - IEEE Conference on Computer Communica- 2020;8:55915–26. 1501
1435 tions Workshops (INFOCOM WKSHPS); 2019 Apr 29-May 2; 63. Li C, Zhang Y, Hao X, et al. Jointly optimized request 1502
1436 Paris, France. Piscataway: IEEE Press; 2019. p. 732–7. dispatching and service placement for MEC in LEO network. 1503
1437 45. Gil Herrera J, Botero JF. Resource allocation in NFV: a China Commun 2020;17(8):199–208. 1504
1438 comprehensive survey. IEEE Trans Netw Serv Manag 2016;13 64. Cisco. Cisco visual networking index: Global mobile data traffic 1505
1439 (3):518–32. forecast update[Internet]. Available from: https://www.cisco.- 1506
1440 46. Pedreno-Manresa J-J, Khodashenas PS, Siddiqui MS, et al. On com/c/en/us/solutions/collateral/executive-perspectives/annual- 1507
1441 the need of joint bandwidth and NFV resource orchestration: a internet-report/white-paper-c11-741490.html. 1508
1442 realistic 5G access network use case. IEEE Commun Lett 2018;22 65. Statista, Number of Internet of Things (IoT) connected devices 1509
of
1443 (1):145–8. worldwide from 2019 to 2030[Internet]. Available:https://www. 1510
1444 47. Pham TM, Chu HN. Multi-provider and multi-domain resource statista.com/statistics/1183457/iot-connected-devices-worldwide/ 1511
1445 orchestration in network functions virtualization. IEEE Access . 1512
1446 2019;7:86920–31. 66. Afolabi I, Taleb T, Samdanis K, et al. Network slicing and 1513
o
1447 48. Sharma P, Tyagi S. Enhancing network security: An SDN softwarization: a survey on principles, enabling technologies, and 1514
1448 (Software Defined Networking) method. Int J Eng Technol Innov solutions. IEEE Commun Surv Tutor 2018;20(3):2429–53. 1515
1449 2020;9(5):1359–63. 67. Jawad N, Salih M, Ali K, et al. Network slicing to enable 1516
1450 49. Bao J, Zhao B, Yu W, et al. OpenSAN: A software-defined scalability and flexibility in 5G mobile networks. IEEE Commun 1517
Pr
1451 satellite network architecture. Proc ACM SIGCOMM Comput Mag 2017;55(5):72–9. 1518
1452 Commun 2015;44(4):347–8. 68. Mahindra R, Khojastepour MA, Zhang HH, et al. Radio access 1519
1453 50. Feng J, Jiang L, Shen Y, et al. A scheme for software defined network sharing in cellular networks2013 21st IEEE International 1520
1454 ORS satellite networking2014 IEEE fourth international confer- Conference on Network Protocols (ICNP); 2013 Oct 7-10; 1521
1455 ence on big data and cloud computing; 2014 Dec 3-5; Sydney, Goettingen, Germany. Piscataway: IEEE Press; 2013. p. 1–10. 1522
1456 Australia. Piscataway: IEEE Press; 2014. p. 716–21. 69. Bor-Yaliniz I, Yanikomeroglu H. The new frontier in RAN 1523
1457 51. Secinti G, Darian PB, Canberk B, et al. SDNs in the sky: robust heterogeneity: Multi-tier drone-cells. IEEE Commun Mag 1524
1458
1459
1460
end-to-end connectivity for aerial vehicular networks. IEEE
Commun Mag 2018;56(1):16–21.
ed
52. Pan CS, Shi JJ, Yang L, et al. Satellite network load balancing
2016;54(11):48–55.
70. Jou BT, Vidal O, Cahill J, et al. Architecture options for satellite
integration into 5G networks2018 European Conference on
1525
1526
1527
1461 strategy for SDN/NFV collaborative deployment.2019 IEEE Networks and Communications (EuCNC); 2018 Jun 18-21; 1528
1462 SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Ljubljana, Slovenia. Piscataway: IEEE Press; 2018. p. 398–9. 1529
1463 Trusted Computing, Scalable Computing & Communications, 71. Zhang S, Luo H, Li J, et al. Hierarchical soft slicing to meet 1530
ct
1464 Cloud & Big Data Computing, Internet of People and Smart City multi-dimensional QoS demand in cache-enabled vehicular net- 1531
1465 Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/ works. IEEE Trans Wirel Commun 2020;19(3):2150–62. 1532
1466 SCI); 2019 Aug 19-23; Leicester, UK. Piscataway: IEEE Press; 72. Foo YC, Lim WL, Tafazolli R, et al. Other-cell interference and 1533
1467 2019. p. 1406–11. reverse link capacity of high altitude platform station CDMA 1534
1468 53. Jawad N, Salih M, Ali K, et al. Smart television services using system. Electron Lett 2000;36(22):1881. 1535
re
1469 NFV/SDN network management. IEEE Trans Broadcast 2019;65 73. Lyu F, Yang P, Wu HQ, et al. Service-oriented dynamic resource 1536
1470 (2):404–13. slicing and optimization for space-air-ground integrated vehicu- 1537
1471 54. Iqbal H, Ma J, Stranc K, et al. A software-defined networking lar networks. IEEE Trans Intell Transp Syst 2021. https://doi.org/ 1538
1472 architecture for aerial network optimization2016 IEEE NetSoft 10.1109/TITS.2021.3070542. 1539
1473 Conference and Workshops (NetSoft); 2016 Jun 6-10; Seoul, 74. Kasgari ATZ, Saad W. Stochastic optimization and control 1540
or
1474 Korea (South). Piscataway: IEEE Press; 2016. p. 151–5. framework for 5G network slicing with effective isolation2018 1541
1475 55. Ding G, Wu F, Wu Q, et al. Robust online spectrum prediction 52nd Annual Conference on Information Sciences and Systems 1542
1476 with incomplete and corrupted historical observations. IEEE (CISS); 2018 Mar 21-23; Princeton, USA. Piscataway: IEEE 1543
1477 Trans Veh Technol 2017;66(9):8022–36. Press; 2018. p. 1–6. 1544
1478 56. Wu Y, Hu G, Jin F, et al. A satellite handover strategy based on 75. Grac D, Thornton J, Chen GH, et al. Improving the system 1545
nc
1479 the potential game in LEO satellite networks. IEEE Access capacity of broadband services using multiple high-altitude 1546
1480 2019;7:133641–52. platforms. IEEE Trans Wirel Commun 2005;4(2):700–9. 1547
1481 57. Ren J, Yu G, He Y, et al. Collaborative cloud and edge 76. Foo YC, Lim WL, Tafazolli R. Centralized downlink call 1548
1482 computing for latency minimization. IEEE Trans Veh Technol admission control for high altitude platform station UMTS with 1549
1483 2019;68(5):5031–44. onboard power resource sharingProceedings IEEE 56th Vehicular 1550
1484 58. Kai C, Zhou H, Yi Y, et al. Collaborative cloud-edge-end task Technology Conference; 2002 Sep 24-28; Vancouver, Canada. Pis- 1551
U
1485 offloading in mobile-edge computing networks with limited cataway: IEEE Press; 2002. p. 549–53. 1552
1486 communication capability. IEEE Trans Cogn Commun Netw 77. Xiong K, Leng S, Hu J, et al. Smart network slicing for vehicular 1553
1487 2021;7(2):624–34. fog-RANs. IEEE Trans Veh Technol 2019;68(4):3075–85. 1554
1488 59. Bari F, Chowdhury SR, Ahmed R, et al. Orchestrating virtual- 78. Tang J, Shim B, Quek TQS. Service multiplexing and revenue 1555
1489 ized network functions. IEEE Trans Netw Serv Manage 2016;13 maximization in sliced C-RAN incorporated with URLLC and 1556
1490 (4):725–39. multicast eMBB. IEEE J Sel Areas Commun 2019;37(4):881–95. 1557
1491 60. van den Bossche R, Vanmechelen K, Broeckhove J. Cost-optimal 79. He J, Song W. AppRAN: Application-oriented radio access 1558
1492 scheduling in hybrid IaaS clouds for deadline constrained network sharing in mobile networks2015 IEEE International 1559
1493 workloads2010 IEEE 3rd International Conference on Cloud Conference on Communications (ICC); 2015 Jun 8-12; London, 1560
1494 Computing; 2010 Jul 5-10; Miami, USA. Piscataway: IEEE Press; UK. Piscataway: IEEE Press; 2015. p. 3788–94. 1561
1495 2010. p. 228–35.
10.1016/j.cja.2021.12.013
3 January 2022
1562 80. Kokku R, Mahindra R, Zhang H, et al. NVS: A substrate for 2018 - IEEE Conference on Computer Communications; 2018 Apr 1630
1563 virtualizing wireless resources in cellular networks. IEEE/ACM 16-19; Honolulu, USA. Piscataway: IEEE Press; 2018. p. 207–15. 1631
1564 Trans Netw 2012;20(5):1333–46. 102. Cheng X, Lyu F, Quan W, et al. Space/aerial-assisted computing 1632
1565 81. Sun W, Strom EG, Brannstrom F, et al. Radio resource offloading for IoT applications: a learning-based approach. IEEE 1633
1566 management for D2D-based V2V communication. IEEE Trans J Sel Areas Commun 2019;37(5):1117–29. 1634
1567 Veh Technol 2016;65(8):6636–50. 103. Zhang Y, Di B, Wang P, et al. HetMEC: Heterogeneous multi- 1635
1568 82. Sun W, Yuan D, Strom EG, et al. Cluster-based radio resource layer mobile edge computing in the 6G era. IEEE Trans Veh 1636
1569 management for D2D-supported safety-critical V2X communi- Technol 2020;69(4):4388–400. 1637
1570 cations. IEEE Trans Wirel Commun 2016;15(4):2756–69. 104. Wang X, Han Y, Wang C, et al. In-edge AI: Intelligentizing 1638
1571 83. Jiang M, Condoluci M, Mahmoodi T. Network slicing manage- mobile edge computing, caching and communication by feder- 1639
1572 ment & prioritization in 5G mobile systemsProc. European ated learning. IEEE Netw 2019;33(5):156–65. 1640
1573 Wireless 2016, 22th European Wireless Conference; 2016 May 18- 105. Wang SQ, Urgaonkar R, Zafer M, et al. Dynamic service 1641
1574 20; Oulu, Finland. Piscataway: IEEE Press; 2016. p. 1–6. migration in mobile edge-clouds2015 IFIP Networking Confer- 1642
1575 84. Peng H, Ye Q, Shen X. Spectrum management for multi-access ence (IFIP Networking); 2015 May 20-22; Toulouse, France. Pis- 1643
of
1576 edge computing in autonomous vehicular networks. IEEE Trans cataway: IEEE Press; 2015. p. 1–9. 1644
1577 Intell Transp Syst 2020;21(7):3001–12. 106. Katsalis K, Papaioannou TG, Nikaein N, et al. SLA-driven VM 1645
1578 85. Deng ZL, Du Q, Li N, et al. RL-based radio resource slicing scheduling in mobile edge computing2016 IEEE 9th International 1646
1579 strategy for software-defined satellite networks2019 IEEE 19th Conference on Cloud Computing (CLOUD); 2016 Jun 27 - Jul 2; 1647
o
1580 International Conference on Communication Technology (ICCT); San Francisco, USA. Piscataway: IEEE Press; 2016. p. 750–7. 1648
1581 2019 Oct 16-19; Xi’an, China. Piscataway: IEEE Press; 2019. p. 107. Wen Z, Cala J, Watson P, et al. Cost effective, reliable and secure 1649
1582 897–901. workflow deployment over federated clouds. IEEE Trans Serv 1650
1583 86. Zhang N, Zhang S, Yang P, et al. Software defined space-air- Comput 2017;10(6):929–41. 1651
Pr
1584 ground integrated vehicular networks: challenges and solutions. 108. Mobile Edge Computing (MEC): Framework and reference 1652
1585 IEEE Commun Mag 2017;55(7):101–9. architecture. 2016. Standard No.: ETSI GS MEC 003. 1653
1586 87. Albonda HDR, Perez-Romero J. An efficient RAN slicing 109. Brik B, Frangoudis PA, Ksentini A. Service-oriented MEC 1654
1587 strategy for a heterogeneous network with eMBB and V2X applications placement in a federated edge cloud architectureICC 1655
1588 services. IEEE Access 2019;7:44771–82. 2020 - 2020 IEEE International Conference on Communications 1656
1589 88. Sun Y, Feng G, Zhang L, et al. Distributed learning based (ICC); 2020 Jun 7-11; Dublin, Ireland. Piscataway: IEEE Press; 1657
1590 handoff mechanism for radio access network slicing with data 2020. p. 1–6. 1658
1591
1592
1593
sharingICC 2019 - 2019 IEEE International Conference on
Communications (ICC); 2019 May 20-24; Shanghai, China. Pis-
cataway: IEEE Press; 2019. p. 1–6.
ed 110. Lim WYB, Luong NC, Hoang DT, et al. Federated learning in
mobile edge networks: a comprehensive survey. IEEE Commun
Surv Tutorials 2020;22(3):2031–63.
1659
1660
1661
1594 89. Wang G, Zhou S, Zhang S, et al. SFC-based service provisioning 111. Lu X, Liao Y, Lio P, et al. Privacy-preserving asynchronous 1662
1595 for reconfigurable space-air-ground integrated networks. IEEE J federated learning mechanism for edge network computing. 1663
1596 Sel Areas Commun 2020;38(7):1478–89. IEEE Access 2020;8:48970–81. 1664
ct
1597 90. Chen K, Zhao S, Lv N. Network monitoring information 112. Shao M, Liu J, Yang Q, et al. A learning based framework for 1665
1598 collection in the SDN-enabled airborne tactical network. Int J MEC server planning with uncertain BSs demands. IEEE Access 1666
1599 Aerosp Eng 2018;2018:1–20. 2020;8:198832–44. 1667
1600 91. Hawilo H, Shami A, Mirahmadi M, et al. NFV: State of the art, 113. Chen M, Mozaffari M, Saad W, et al. Caching in the sky: 1668
1601 challenges, and implementation in next generation mobile proactive deployment of cache-enabled unmanned aerial vehicles 1669
re
1602 networks (vEPC). IEEE Netw 2014;28(6):18–26. for optimized quality-of-experience. IEEE J Sel Areas Commun 1670
1603 92. Cunha VA, da Silva E, de Carvalho MB, et al. Network slicing 2017;35(5):1046–61. 1671
1604 security: challenges and directions. Internet Technol Lett 2019;2 114. Zhou CH, Wu W, He HL, et al. Delay-aware IoT task scheduling 1672
1605 (5):e125. in space-air-ground integrated network2019 IEEE Global Com- 1673
1606 93. Li X, Samaka M, Chan HA, et al. Network slicing for 5G: munications Conference (GLOBECOM); 2019 Dec 9-13; Waiko- 1674
or
1607 challenges and opportunities. IEEE Internet Comput 2017;21 loa, USA. Piscataway: IEEE Press; 2019. p. 1–6. 1675
1608 (5):20–7. 115. Mao S, He S, Wu J. Joint UAV position optimization and 1676
1609 94. Haider S, Akhunzada A, Mustafa I, et al. A deep CNN ensemble resource scheduling in space-air-ground integrated networks with 1677
1610 framework for efficient DDoS attack detection in software mixed cloud-edge computing. IEEE Syst J 2021;15(3):3992–4002. 1678
1611 defined networks. IEEE Access 2020;8:53972–83. 116. Wang D, Tian X, Cui H, et al. Reinforcement learning-based 1679
nc
1612 95. Prados-Garzon J, Ramos-Munoz JJ, Ameigeiras P, et al. Model- joint task offloading and migration schemes optimization in 1680
1613 ing and dimensioning of a virtualized MME for 5G mobile mobility-aware MEC network. China Commun 2020;17(8):31–44. 1681
1614 networks. IEEE Trans Veh Technol 2017;66(5):4383–95. 117. Zhou S, Wang G, Zhang S, et al. Bidirectional mission offloading 1682
1615 96. Chiang M, Zhang T. Fog and IoT: An overview of research for agile space-air-ground integrated networks. IEEE Wirel 1683
1616 opportunities. IEEE Internet Things J 2016;3(6):854–64. Commun 2019;26(2):38–45. 1684
1617 97. Kai K, Cong W, Tao L. Fog computing for vehicular Ad-hoc 118. Zhou C, Wu W, He H, et al. Deep reinforcement learning for 1685
U
1618 networks: paradigms, scenarios, and issues. J China Univ Posts delay-oriented IoT task scheduling in SAGIN. IEEE Trans 1686
1619 Telecommun 2016;23(2):56–96. Wireless Commun 2021;20(2):911–25. 1687
1620 98. Abbas N, Zhang Y, Taherkordi A, et al. Mobile edge computing: 119. Gu SS, Sun XY, Yang ZH, et al. Energy-aware coded caching 1688
1621 a survey. IEEE Internet Things J 2018;5(1):450–65. strategy design with resource optimization for satellite-UAV- 1689
1622 99. Wang S, Zhang X, Zhang Y, et al. A survey on mobile edge vehicle integrated networks. IEEE Internet Things J 2021. https:// 1690
1623 networks: Convergence of computing, caching and communica- doi.org/10.1109/JIOT.2021.3065664. 1691
1624 tions. IEEE Access 2017;5:6757–79. 120. Shen Y, Shi Y, Zhang J, et al. Graph neural networks for scalable 1692
1625 100. Mao Y, You C, Zhang J, et al. A survey on mobile edge radio resource management: Architecture design and theoretical 1693
1626 computing: The communication perspective. IEEE Commun Surv analysis. IEEE J Sel Areas Commun 2021;39(1):101–15. 1694
1627 Tutorials 2017;19(4):2322–58. 121. ONAP Consumption models[Internet]. Available: https://www. 1695
1628 101. Xu J, Chen LX, Zhou P. Joint service caching and task offloading onap.org/wp-content/uploads/sites/20/2020/06/ 1696
1629 for mobile edge computing in dense networksIEEE INFOCOM ONAP_EUAG_Whitepaper_061720.pdf. 1697
10.1016/j.cja.2021.12.013
3 January 2022
18 N. CHENG et al.
1698 122. Ontology (information science)[Internet]. Available: https://en. 130. GLUE’s benchmark leaderboard[Internet]. Available from: 1721
1699 wikipedia.org/wiki/Ontology(information_science). https://gluebenchmark.com/. 1722
1700 123. Lopez de Vergaro JE, Villagra VA, Asensio JI, et al. Ontologies: 131. Zeydan E, Turk Y. Recent advances in intent-based networking: 1723
1701 Giving semantics to network management models. IEEE Netw A survey2020 IEEE 91st Vehicular Technology Conference 1724
1702 2003;17(3):15–21. (VTC2020-Spring); 2020 May 25-28; Antwerp, Belgium. Piscat- 1725
1703 124. de Vergara JEL, Villagra VA, Berrocal J. Applying the Web away: IEEE Press; 2020. p. 1–5. 1726
1704 ontology language to management information definitions. IEEE 132. Bankey V, Upadhyay PK. Physical layer security of multiuser 1727
1705 Commun Mag 2004;42(7):68–74. multirelay hybrid satellite-terrestrial relay networks. IEEE Trans 1728
1706 125. Sahlmann K, Scheffler T, Schnor B. Ontology-driven device Veh Technol 2019;68(3):2488–501. 1729
1707 descriptions for IoT network management2018 Global Internet of 133. Li BH, Liang RC, Zhou W, et al. LBS meets blockchain: an 1730
1708 Things Summit (GIoTS); 2018 Jun 4-7; Bilbao, Spain. Piscat- efficient method with security preserving trust in SAGIN. IEEE 1731
1709 away: IEEE Press; 2018. p. 1–6. Internet Things J 2021. https://doi.org/10.1109/ 1732
1710 126. Ding G, Wang J, Wu Q, et al. Robust spectrum sensing with JIOT.2021.3064357. 1733
1711 crowd sensors. IEEE Trans Commun 2014;62(9):3129–43. 134. Sun W, Wang L, Wang P, et al. Collaborative blockchain for 1734
of
1712 127. Wild T, Braun V, Viswanathan H. Joint design of communica- space-air-ground integrated networks. IEEE Wirel Commun 1735
1713 tion and sensing for beyond 5G and 6G systems. IEEE Access 2020;27(6):82–9. 1736
1714 2021;9:30845–57. 135. Tselios C, Politis I, Kotsopoulos S. Enhancing SDN security for 1737
1715 128. Chen X, Feng Z, Wei Z, et al. Performance of joint sensing- IoT-related deployments through blockchain2017 IEEE Confer- 1738
o
1716 communication cooperative sensing UAV network. IEEE Trans ence on Network Function Virtualization and Software Defined 1739
1717 Veh Technol 2020;69(12):15545–56. Networks (NFV-SDN); 2017 Nov 6-8; Berlin, Germany. Piscat- 1740
1718 129. Sezer OB, Dogdu E, Ozbayoglu AM. Context-aware computing, away: IEEE Press; 2017. p. 303–8. 1741
1719 learning, and big data in Internet of Things: a survey. IEEE
Pr
1742
1720 Internet Things J 2018;5(1):1–27.
ed
ct
re
or
nc
U
10.1016/j.cja.2021.12.013

6G Service-Oriented Space-Air-Ground Integrated Network - A Survey

Uploaded by

Copyright:

Available Formats

6G Service-Oriented Space-Air-Ground Integrated Network - A Survey

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

6G Service-Oriented Space-Air-Ground Integrated Network - A Survey

Uploaded by

Copyright:

Available Formats

CJA 2272 No.

Chinese Society of Aeronautics and Astronautics

12 Received 12 May 2021; revised 24 December 2021; accepted 24 December 2021

interoperability of different network segments and cooperatively orchestrate heterogeneous

thereby reducing the occurrence of disasters. In addition, 243

192 seamless services; (C) Cognitive satellite-terrestrial network.

equipped with communication equipment can act as backup 251

204 2.2. Application scenarios

205 Since SAGINs can provide many benefits such as seamless

Fig. 1 SAGINs service management framework.

Fig. 3 Network slicing based on SAGINs architecture.

597 mance. In these schemes, only unicasting is considered, and 657

616 posed to reduce the delay and improve the throughput.31

736 blocks (containing communication, computation and storage

Fig. 4 SDN/NFV-based network architecture in SAGINs.

794 (A) SDN-based overall resource orchestration

795 efficiency, a new operationally responsive space (ORS) satellite 853

930 where entities that exchange large amounts of traffic are

1222 user queries in a given language (i.e., intents) into a structured

1224 Through intent-based techniques, the network service, which

1299 2. ITU-R M. [IMT-2020.TECH PERF REQ]—minimum require- (6):111–7. 1366

1305 4. Shen X, Cheng N, Zhou H, et al. Space-air-ground Integrated 2020;69(10):12098–110. 1372

1319 9. Xiong F, Li A, Wang H, et al. An SDN-MQTT based Press; 2018.p.242–6. 1386

1321 Commun Mag 2019;57(8):41–7. architecture—intermediate report. EU H2020 5G NORMA. 1388

1334 Sci 2021;64(4):1–16.

1339 networks: architecture, challenges, and opportunities. IEEE

1350 19. Ding G, Wu Q, Zhang L, et al. An amateur drone surveillance

1355 (3):47–53. (3):4005–18. 1422

1358 IEEE Press; 2020. p. 1–5. (9):1768–71. 1425

1361 2015;103(7):1102–24. IEEE Access 2019;7:106162–78. 1428

You might also like