1

Edge Computing in IoT: A 6G Perspective

Mariam Ishtiaq, Nasir Saeed, Senior Member, IEEE, Muhammad Asif Khan, Senior Member, IEEE,

Abstract—Edge computing is one of the key driving will heavily rely on state-of-the-art artificial intelligence
forces to enable Beyond 5G (B5G) and 6G networks. (AI) and machine learning (ML) technologies for intel-
Due to the unprecedented increase in traffic volumes and ligent network operations and management. In addition
computation demands of future networks, multi-access to the infrastructural transformations, B5G/6G networks
(or mobile) edge computing (MEC) is considered as a
are also expected to support computationally intensive
arXiv:2111.08943v2 [cs.NI] 15 May 2022

promising solution to provide cloud-computing capabilities

within the radio access network (RAN) closer to the end
services and applications.
users. There has been a significant amount of research on To support the two major transformations in network
MEC and its potential applications; however, very little infrastructures and network services, edge computing has
has been said about the key factors of MEC deployment gained significant attention and is considered as a built-
to meet the diverse demands of future applications. In this in service in 6G networks. Although, a huge amount of
article, we present key considerations for edge deployments research literature has been dedicated on MEC capabili-
in B5G/6G networks including edge architecture, server ties such as caching services and computation offloading
location and capacity, user density, security etc. We further techniques, there has been very limited insights on MEC
provide state-of-the-art edge-centric services in future
deployment.
B5G/6G networks. The paper also present experimental
evaluation of edge-based services deployment including
In this article, we first present the edge computing
video transcoding and deep learning inference. Lastly, architecture, evolution, and the underlying core concepts.
we present some interesting insights and open research Then we discuss the key considerations of deploying
problems in edge computing for 6G networks. edge computing infrastructure and services. We then
discuss some interesting use-cases of edge-centric ser-
Index Terms—Cloud computing, edge computing,
B5G/6G architecture vices in different industry verticals. Lastly, We conclude
the article by presenting open research problems in this
realm and some key insights.
I. I NTRODUCTION
II. E DGE C OMPUTING IN 5G AND B5G N ETWORKS
B EYOND 5G (B5G) and 6G networks are the cat-
alyst to the digital future of humanity and the
realization of Industry 5.0. As connectivity and digital
The standard 5G network architecture is built on
cloud-native technology enablers like software defined
transformation accelerate, worldwide industry verticals networking (SDN), virtualization, and multi-access edge
will increasingly rely on 5G and B5G networks to enable computing. Conventionally, cloud computing is imple-
novel applications and services categorized as ultra-low mented using the three cloud computing services (CCS)
latency, mission-critical reliability, and a high degree of as shown in Fig. 1.
mobility. With Moore’s law, the traditional system on a 1) Software-as-a-service (SaaS): The consumer can
chip (SoC) approach is evolving to system-technology use the provider’s applications running on a cloud
co-optimization (STCO), and consequent system de- infrastructure where accessibility is possible from
velopment methodologies are adopted rapidly. In this various client devices via a thin client interface
context, B5G/6G networks are envisioned to serve a (e.g., web-based email) or a program interface. In
diverse range of applications by utilizing MEC as a built SaaS, the underlying cloud infrastructure, including
in service with cloud-native concepts like leveraging network, servers, operating systems, and storage,
operations within and across data centers, communicat- are not managed or controlled by the consumer.
ing in a micro-service environment, and simultaneously 2) Platform-as-a-service (PaaS): The consumer can
providing secure services and applications [1]. deploy applications created using programming lan-
B5G and 6G networks are envisioned to fully and guages, libraries, services, and tools supported by
seamlessly integrate various verticals such as Internet the provider onto the cloud infrastructure. Like
of things (IoT), aerial networks (aka drones), satellite SaaS, in PaaS, the consumer does not manage or
access, and under-water communication. To keep up with control the underlying cloud infrastructure, includ-
this staggering expectation, future networks (B5G/6G) ing network, servers, operating systems, or storage.
 2

However, in PaaS, the consumer can control the puting services, such as a data center. There are three
deployed applications and possibly configuration inherently distributed architectural deployment models
settings for the application-hosting environment. for B5G networks. Firstly, cloud computing provides
3) Infrastructure-as-a-service (IaaS): In IaaS, the on-demand storage and computing resources. Secondly,
consumer can deploy and run arbitrary software, edge computing places computing resources closer to the
including operating systems and applications, with end-users to reduce delay, enhance processing speed, and
the provision of processing, storage, networks, and ensure on-premise security. Also, edge computing gives
other fundamental computing resources. The un- an added advantage of scalability in terms of increasing
derlying cloud infrastructure cannot be managed users/devices and optimizing reliability and user experi-
or controlled by the consumer. However, the con- ence. Finally, MEC is an architectural standard for edge
sumer can manage the operating systems, storage, computing by ETSI, providing deployment flexibility at
deployed applications, and limited networking com- radio nodes, aggregation points, and at the edge of the
ponents (e.g., host firewalls). core network. In addition, reusability (of microservices),
extensibility (to include new use cases), and efficiency
(in terms of computation time resources and energy) are
some of the key features of the MEC standard.
Mobile network operators (MNOs) can provide dif-
ferent deployment strategies. MNOs can move the on-
premise edge infrastructure to public or private clouds to
ensure enterprise connectivity, including cloud comput-
ing, as-a-service connectivity, and MEC infrastructure.
Using such a hybrid approach, communication service
providers (CSPs) can pass over cloud platforms (infras-
tructure and system), focusing entirely on service dif-
ferentiation. Independent vendors can also give product
integration support and customer application support.
However, to gain full control of the 5G core, strict service
level agreements (SLAs) need to be in place with cloud
owners to ensure compliance.

III. C ONSIDERATIONS FOR E DGE - CENTRIC B5G/6G

N ETWORKS
Fig. 1: Cloud computing services (CCS).
From MNO’s perspective, some of the primary moti-
One of the shortcomings of traditional cloud com- vations for B5G/6G networks are to deliver high band-
puting architecture is that it doesn’t meet the ultra-low width and support a high density of users, operate on
latency as B5G/6G would commit. The higher delay in a cost and energy-efficient infrastructure, and ensure
cloud computing is mainly caused by the inherited longer reliability and security of operation. Some of the main
distance from the end-users. Edge computing overcomes considerations for B5G networks to support future ser-
this by bringing computing and storage capabilities vices are given in Table I and discussed in this section.
closer to the end-users. 1) Edge Architecture and Location: Edge comput-
There are different tiers of edge computing archi- ing enables networks to become more efficient
tecture: device tier, edge tier, and central tier. The by providing a virtualization platform, expanding
device tier includes devices with limited storage and the enterprise service base, and offering network
computing capabilities like sensors, actuators, and user congestion control protocols. From a micro opera-
interface devices. This tier often uses proximity net- tor’s perspective, edge location plays a critical role
works with limited range and bandwidth. The edge tier in ascertaining these core values. In networking
comprises intermediate nodes like edge nodes, control terms, edge location refers to its closeness to the
nodes, IoT gateways, and other nodes that need low consumer. However, this location may vary for
latency or high bandwidth to provide direct support to operators, enterprises, end-users, and application
the nodes in the device tier. The central tier offers a types, on a need basis. Some of these needs or
wide connectivity span, potentially using cloud com- metrics to decide edge location can be affordable
 3

latency, user equipment mobility, network resource accounts for powerful computational processing and
management, and computation cost. Various dif- massive data acquisition locally at edge networks
ferent edge locations can be considered to make [7]. It plays an important role in implementing
the edge availability secure and accessible. Mobile smart and efficient resource scheduling strategies
edge servers, retail centers, manufacturing stations, in a complex environment with heterogeneous re-
housing compounds, 5G cellular base stations, and sources and a massive number of devices. To this
the edge of the mobile operator’s core network can end, federated learning (FL) executes a training
account for a few edge location possibilities [2]. algorithm common to all edges to support decentral-
These possibilities can be explored by ensuring an ized learning further while keeping the data local.
uninterrupted power supply and security to reap the The result is a robust machine learning model able
benefits of edge computing. to address critical issues like data privacy, data
To meet the stringent QoS requirements of B5G security, data diversity, continual real-time learning,
services, MEC promises proximity, a high access and hardware efficiency [8].
rate, and low latency. While there is no definite 4) Security: Although the EaaS approach of 5G-MEC
blueprint for future RAN deployment, distributed improves the visibility over traditional networks and
unit (DU) and Central unit (CU) functions can be enables faster identification of potential security
combined or split and deployed at the radio site threats, the simultaneous operation of a massive
or in centralized locations. Thus, we can say that number of IoT devices, cloud-based deployments,
location can be determined by services, density, the and third-party applications expands the attack sur-
distance between sites, and transport network speed. face. Edge computing provides enhanced security
2) Edge Capacity: For 5G networks, multiple-input, compared to cloud computing by allowing filtration
multiple-output (MIMO) technology is employed of sensitive data at the source rather than sending
to deliver the capacity needs. This multi-antenna it to a central data center. Therefore, fewer inter-
transmission technology boosts network capacity device transfers of data using EaaS means better
and data throughput significantly by increasing the security. However, MEC is subject to vulnerabilities
number of transmission ports. The associated cost in virtualization technologies like virtual machine
is that a higher number of processing resources escape, virtual machine manipulation, DNS ampli-
are required for all the antenna signals. To trans- fication, and virtual network function (VNF) loca-
port all individual antenna signals to the baseband tion shift [9]. Attacks on host-level orchestration
processing function, the transport capacity of the components can collapse the MEC servers. Other
RAN network reduces. For this backhaul offload- possible security vulnerabilities and their mitigation
ing scenario, edge networks play a major role in strategies include detection of access to unautho-
compensating for the over-consumption of network rized files or applications, monitoring and blocking
resources by extending the finite network resources, any information leakage, system anomaly detection,
thus enhancing the delay and capacity performances and prevention from over-utilization of host MEC
of 5G networks [6]. Therefore, artificial intelligence resources. Using a privacy-preserving edge intelli-
(AI) and associated techniques enable smart edge gence model like FL is currently being explored as
processing and open new opportunities for powerful an option to preserve data integrity by leveraging
computational processing, massive data acquisition, the distributed nature of edge-centric 5G verticals.
and edge-caching capabilities requiring higher-order FL ensures data privacy by executing the shared
edge capacity. Hence, the vertical use case de- learning models locally and not sending the training
termines the required edge capacity. For instance, data to central servers.
video streaming requires bulks of data caching, The key edge considerations are listed in Table I.
while for weather monitoring, small data updates
over a period of time might be enough. IV. B5G/6G E DGE N ETWORKS - U SE C ASES
3) Edge Intelligence: Traditional machine learning This section provides an edge-centric overview of 5G
and deep learning techniques use higher volumes of applications and a projected view of B5G architectural
training data and a large processing time, thus im- requirements.
posing storage and delay constraints on the central-
ized server. This is not a viable learning approach A. Video Analytics
for B5G and 6G networks that promise massive con- Video analytics aims to automatically recognize spa-
nectivity with ultra-low latency. Edge intelligence tial and temporal events in videos, which is a data-
 4

TABLE I: Edge considerations for vertical industries use-cases.

Edge Considerations Video Analytics ITS Manufacturing Energy Agriculture
Architecture Single/Multi server Multi server Single server Singe/Multi server Multi server

Edge location 5G RAN RSU or RAN On-premise On-premise RAN

Storage capacity Very High Medium/High Low Low/Medium Medium/High

Computing capacity Very High High Low/Medium Medium Medium

Typical Latency Re-

<1s 10 ms - 1 s < 100 ms < 10 ms <1s
quirement

Data & ML model Data & ML model Online adversarial Online adversarial DDoS and Data
Security Issues
poisoning poisoning attacks and DDoS attacks poisoning attacks
Object detection, Anomalies
Fault isolation,
Object detection, tracking, Anomaly detection, detection,
predictive
Edge Intelligence recommendation, traffic/vehicle process automation crop treatment
maintenance
summarization etc. prediction & & control etc. recommendations
etc.
control etc. etc.
Real-world
5G-ENSURE [3] 5G-ENSURE SONATA [4] COGNET [5] 5G-ENSURE
deployments

intensive operation and gains particular advantage from screens, and AR/VR devices. Without edge, such data-
the edge in terms of less data transfer time (less delay), intensive use cases cannot promise an enhanced user
local data processing (secure), and efficient resource experience due to adversaries like delays and jitters.
utilization. Some other vital application areas of video analytics are:
The Role of Edge Computing: Massive and time- • Smart City. To enable smart cities, video analytics

critical data processing are key requirements of video at the edge can play a key role. It can find ap-
analytics. Low latency is primarily required for real-time plications in smart traffic monitoring, instant fire or
content distribution and processing. Content caching smoke detection, waste disposal, and environmental
of frequently used data at edge servers for on-request monitoring.
delivery to mobile devices is expected to reduce round- • Surveillance. Live video surveillance at the edge

trip times (RTT) and provide high bandwidth in 5G can instantly provide actionable results to avoid
wireless systems. However, the associated drawback is accidents and to enforce law and order.
the limitation in processing capability at the edge com- • Autonomous Vehicles. A self-driving car can max-

pared to the data center’s, therefore posing delays from imize the benefit of edge video analytics using
software operation in a virtual environment. MEC has instantaneous data pre-processing for decisions and
been integrated with 5G architecture to employ VNFs on post-processing for continued training and feed-
the edge servers to overcome this delay. Therefore, B5G back.
networks can envision decentralized approaches like fog
computing to reduce computing overhead further and B. Intelligent Transport System (ITS)
reduce end-to-end delay. Intelligent transportation systems (ITS) have a great
Use Cases and Applications: Some of the key potential to make transportation systems safe, efficient,
B5G/6G verticals and use cases rely on video analytics safe, and sustainable. IoT-driven intelligent transporta-
for an operation like face recognition from traffic and se- tion systems are defined by an amalgam of complex,
curity cameras, object tracking and motion detection for data-intensive, dynamic, and uncertain operations like
surveillance, AI learning of patterns in live stream video task management, resource allocation, and security that
archives, etc. For example, the real-time experience of can be effectively managed using edge computing.
event audiences such as sports and concerts can be The Role of B5G/6G and Edge Computing: Future ITS
enhanced using video analytics at the edge. Augmented applications are considered to be one of the most promis-
reality (AR) and virtual reality (VR) functions can be ing vertical sectors for mobile networks. Two radio
efficiently applied to videos captured from different cam- access technologies have been defined and are currently
era angles and displayed through smartphones, tablets, competing for market adoption: IEEE wireless access in
 5

Fig. 2: V2X system with multi-access edge computing.

vehicular environments (WAVE)/ dedicated short-range
communications (DSRC) based on the IEEE 802.11p
and 3GPP LTE cellular vehicle-to-everything (C-V2X)
[10]. Since both these technologies rely on different
channel access schemes, three major challenges arise Fig. 3: Process control for smart manufacturing.
when simultaneously accessing the spectrum in the 5.9 vices to the users of the vehicle, including Internet
GHz band: coexistence, interoperability (for wide market access, comfort services, and content sharing.
acceptance), and backward compatibility. The essence of
B5G/6G is to satisfy a wide spectrum of heterogeneous
end devices and vehicles requiring ultra-low latency C. Smart Manufacturing
and high bandwidth. This scale of computational power B5G/6G networks will bring manufacturers and net-
within proximity to the end-users can be made available works operators together to build smart factories capable
at the edge of the network, using an edge cloud or edge of innovation, process automation and flexibility. The
data center (EDC). An edge cloud brings about a signifi- current industrial networks have been designed for static
cant reduction in end-to-end latency in comparison with manufacturing processes where even a single change in
a centralized cloud. EDC networks face traffic burstiness, the manufacturing workflow requires long maintenance
which is aggravated even further by the myriad of 5G operations. With the fifth industrial revolution (Industry
access technologies. For example, even though the 5G 5.0), the evolution of programmable network, increased
mmWave offers massive peak-data-rate links, it makes security and dynamic traffic flows over heterogeneous
the available capacity fluctuate. Therefore, the design of industrial environments will redefine the manufacturing
flexible EDC networks with promising ultra-low latency vertical of B5G connectivity.
and high and scalable bandwidths is a promising research
The Role of Edge Computing: The success and sus-
direction.
tenance of Industry 5.0 lies strongly in the adoption of
Fig. 2 shows a simple heterogeneous V2X system emerging information and communications technology
using MEC where ITS applications can communicate (ICT) technologies, such as loT, artificial intelligence,
with road-side units (RSUs) and base stations. The MEC augmented reality for troubleshooting. By exploring ac-
hosts can be co-located with these radiating elements cess to real-time production information, cooperation and
(and/or with C-RAN aggregation points). advanced control features, novel value added services are
Applications and Use Cases: ITS can be broadly promised for industrial operators and customers. Edge
categorized into three service areas [11]: computing and B5G/6G systems will play a key role
• Safety services. To minimize accidents and mitigate in enabling Industry 5.0 by lending the network slicing
risks for passengers and other users (pedestrians and paradigm for industrial use cases over a wide range of
vehicles). application areas.
• Non-safety services. To improve traffic manage- Fig. 3 shows a layered architecture for process control
ment to maximize the efficiency of the existing in a smart factory or manufacturing facility. Each device
road network and enhance experience by factors like and processes coordinates with the services platform for
traffic congestion minimization. designing, creating and delivering the services. Platform
• Infotainment services. To provide a range of ser- layer is rendered dynamic by the network slicing layer
 6

which caters to 5G network resource management ac- DERs are integrated into a single edge comput-
cording to the service. Data management and storage is ing VPP network infrastructure. The objective of
handled at the edge using MEC. To reduce the total av- VPP is to dissipate power generated by individual
erage end-to-end transmission delay, efficient scheduling units during peak load, thus relieving load on the
algorithms for backhaul links are required to route real- smart grid. To achieve this, the VPP control center
time traffic. (centralized data center) collects data from edge
Applications and Use Cases: Manufacturing vertical devices like sensors, control units, and VPP appli-
includes use cases like: cations. The edge computing gateway enables rapid
• Real time data analytics finds important appli- response and real-time power dispatch. This allows
cations in safety procedures, product and process predictive maintenance of the power grid by actively
customization for output maximization and better monitoring the operation of each power plant and
resource utilization. energy storage.
• Augmented reality can be used to determine oper- • Real time data analysis To manage VPP services,
ational inefficiencies to mitigate the cost of produc- optimal scheduling control of DERs is required in
tion downtime. the VPP. In this realm, time-varying optimization
• Object tracking Real-time motion and robot track- of 5G can be employed for the real-time control
ing can overcome the tedious manual work and of a VPP that could adjust the output of the
increase the efficiency of the manufacturing process. DERs to maintain quality in the power network
and achieve the objectives of both the customer
and the utility company. This can reduce generation
D. Energy
cost and power losses while increasing VPP service
To keep up with United Nation’s Sustainable Develop- reliability.
ment Goals 2030, it is vital to ensure provision of clean
energy for energy intensive sectors. 5G is an enabler V. E XPERIMENTAL A NALYSIS OF E DGE - BASED
for connected power distribution grids promising remote S ERVICES
control and demand handling while minimizing energy This section presents some illustrative analysis of two
wastage. edge-based services, including video transcoding and
The Role of Edge Computing: 5G has a key role to deep learning inference. Video transcoding is essential in
address urgent issues like climate change and disaster real-time video streaming in content delivery and future
monitoring, since it has a massive user base and can VR/AR services. Edge servers can be used to enable
render energy usage efficient. To address the dynamic fast and efficient transcoding services to enable these
energy requirements of the 5G use cases, intelligent services. However, the limited processing capabilities
VPPs can ensure diversified resources, integrated energy of the edge servers can cause performance degradation
consumption, production and trade. MEC has the po- when the number of simultaneous requests increases.
tential to enable B5G with this capability. For instance, Fig. 4 illustrates different scenarios where simultaneous
energy harvesting is a challenging task to supply it for transcoding jobs are executed at the edge server. The
low power devices at long ranges, since large aperture amazon web services (AWS)’s C5n provides a high-
antennas are required. However, large antennas have a performance computing (HPC) platform for batch pro-
narrowing field of view, which limits their operation cessing workloads, media transcoding, scientific mod-
if they are widely dispersed from a 5G base station. eling, machine learning inference, and other compute-
The advantage of 5G edge centric cellular systems is intensive applications to transcode the videos. Videos
that they are simultaneously truly ubiquitous, reliable, of different sizes and different source bit rates are
scalable, and cost-efficient. With smart integration of transcoded to evaluate the edge computing scenarios.
beam-forming techniques like [12], B5G/6G itself can Fig. 4 depicts the edge computing time of transcoding
be an indispensable wireless power grid for IoT devices. videos of different duration, illustrating edge computing
Applications and Use Cases: Energy vertical includes performance concerning the number of jobs. This intu-
use cases like: itively motivates the use of efficient scheduling of jobs
• Virtual Power Plant (VPP) A virtual power plant in edge computing platforms.
(VPP) is a distributed cloud-based power plant that We further analyze the computation power of an
aggregates a high number of distributed energy edge computing platform in Fig. 5 which depicts the
resources (DERs), into an integrated and monitored cumulative distribution (CDF) of the number of failed
network to enhance power generation. In a VPP, requests when using multiple simultaneous transcoding
 7

1-Job 4-Jobs 16-Jobs execute the tasks using the local resources on the end
2-Jobs 8-Jobs device (mobile or IoT device) and partially offload to the
50 edge server. Fig. 6 depicts our experimental analysis of
performance inference delay while running the VGG16
deep learning model for image classification. The model
40
Edge Compute Time (s)

is divided into two parts for each inference task de-

pending upon the available local and remote computing
30 resources and the available bandwidth of the wireless
channel. The analysis is extended to use IoT devices of
20 different CPU resources (i.e., 1.0 GHz, 1.5 GHz, and
2.0 GHz). The acquired results are representative of the
VGG16 model, which indicates that for IoT devices of
10
certain processing power, the inference delay increases
when more layers are executed locally at the IoT device
0 and potentially exceed a given threshold at a certain point
0 1 2 3 4 5 6 7 8
Video Duration (depending upon the application delay requirement).
Hence, in such kind of collaborative environment, the
Fig. 4: Edge-assisted video transcoding in various set- processing tasks need to be optimally divided among
tings. IoT devices and the edge server.

jobs. We consider the request as failed if it does not 1.0 GHz 1.5 GHz 2.0 GHz
meet the time deadline (i.e., when the transcoding time
exceeds the source video duration). 0.030

1-Job 4-Jobs 16-Jobs 0.025

2-Jobs 8-Jobs
Inference delay

0.020
100
0.015
80
0.010
Failed Requests

60 0.005

40 0.000
0 5 10 15 20
Layers processed at IoT device
20
Fig. 6: Inference delay using distributed edge computing.

VI. F UTURE R ESEARCH D IRECTIONS

Fig. 5: Failed processing requests.
Edge computing is an integral part of 5G IoT networks
While efficient scheduling of simultaneous jobs in the and envisioned 6G technologies that promise to provide
multi-edge environment can potentially solve the prob- more bandwidth with reduced latency to support novel
lem to a greater extent, the problem persists when only cloud applications. In this paper, we presented key issues
one or fewer edge servers are insufficient to execute the for edge computing in B5G/6G networks in terms of
computations within the deadline. Collaboration among edge location, edge capacity, user density, local learning,
edge servers and end devices is a promising approach to and security. In the following, we present some of the
address this challenge. Although such collaborations can promising future research directions for edge computing
use different setups, the standard practice is to partially in envisioned 6G networks.
 8

A. 6G as Human Centric Networks networks should use AI-based services such as big data
6G networks shall be designed based on human- analytics for key decision-making in various real-time
centric services that require tight-coupling of the QoE scenarios. Edge computing thus plays a significant role in
metrics to human senses rather than generic require- 6G networks. The idea is that powerful edge servers can
ments. For example, 5G requires a latency requirement be deployed as an intermediate layer to provide localized
of 1 ms. In contrast, the 6G network provides latency and ultra-low latency processing for real-time applica-
based on the human reaction times, such as the 100 tions. As cloud-based processing can suffer from long
ms for auditory reaction, 10 ms for visual reaction, delays and transmission overhead, edge computing can
and 1 ms for perceptual response [13]. Edge computing complement cloud computing. The network services are
can play a vital role, thus providing guaranteed delay designed such that fast content caching and processing
performance. The edge server has a more accurate local are done at the edge to perform delay-sensitive critical
view of the network parameters and the users’ instanta- tasks, whereas cloud servers are used to achieve deeper
neous requirements; it becomes more convenient to adapt and more efficient processing at the expense of higher
to the varying conditions and requirements when the delays. However, edge resources are limited as compared
application is running at the edge rather than the remote to the cloud servers, and training AI models may be
cloud server. Nevertheless, the research on investigating infeasible or inefficient. In this case, pervasive AI can
edge computing for 6G human-centric networks is still be a promising solution that uses distributed network
in infancy and needs the researcher’s attention. resources, including edge servers, cloud servers, and user
devices, to intelligently distribute the AI computation
B. Integrated Terrestrial and Aerial Networks (i.e., training and inference tasks) across various devices,
to minimize the delay. Pervasive AI provides a further
B5G and 6G networks incorporate aerial and satellite advantage in terms of privacy (due to no sharing of data)
networks as an essential part [14]. Aerial networks and scalability (using the huge distributed resources).
consisting of drones or unmanned aerial vehicles (UAVs)
are considered as the first choice for applications such
as search-and-rescue, surveillance, and wildlife con- D. Immersive Augmented and Mixed Reality Streaming
servation. These applications usually require real-time 6G networks would be able to meet the unprecedented
video analytics, which has several challenges in terms requirements that 5G might not support. When we con-
of communication (ultra-low latency), huge processing, sider the huge volumes of data in future wireless services
and extra-large storage capacity. Although MEC in video such as 8K videos, VR, AR, and even eXtended Reality
applications brings significant improvement in latency (XR), the existing 5G support becomes insufficient. The
reduction and computation offloading, live video analyt- limitation of 5G basically arises from the design and
ics still poses challenges. Some key issues are (i) Compu- baseline KPIs that aim to reach average data rates of up
tation migration from one MEC server to another when to 100 Mbps (downlink) and 50 Mbps (uplink) [264]. In
the drones fly away in overlapping coverage areas. (ii) addition to the incapacity of 5G to deliver higher data
Rapidly varying wireless link quality requires joint com- rates for volumetric applications, 5G also does not meet
munication and computation and control (3C) services. the requirements of ultra-precision streaming tasks, such
(iii) Limited onboard energy of drones allows minimal as remote surgeries and other XR applications. There-
local computation. (iv) D2D offloading of content and fore, investigating edge computing to support immersive
computations among drones on the fly to complement in augmented and mixed reality streaming is an exciting
areas of poor coverage or increasing demands than the area of research.
current capacity of drones and/or MEC servers. MEC-
assisted drones and drone-drone cooperation in the video R EFERENCES
IoT (VIoT) services involve large-sized content sharing
[1] T. K. Rodrigues, J. Liu, and N. Kato, “Application of cybertwin
and intensive offloading, which intensify the challenges for offloading in mobile multiaccess edge computing for 6G
of normal drone cooperation. networks,” IEEE Internet of Things Journal, vol. 8, no. 22, pp.
16 231–16 242, 2021.
[2] L. Nadeem, M. A. Azam, Y. Amin, M. A. Al-Ghamdi, K. K.
C. Self Sustained Networks (SSNs) using Edge Intelli- Chai, M. F. N. Khan, and M. A. Khan, “Integration of d2d,
gence network slicing, and mec in 5g cellular networks: Survey and
challenges,” IEEE Access, vol. 9, pp. 37 590–37 612, 2021.
Another test case for 6G networks will be the use
[3] “5G ENSURE Project,” https://5g-ppp.eu/wp-content/uploads/
of artificial intelligence (AI) in the entire system archi- 2014/02/5G-PPP White-Paper Phase-1-Security-Landscape
tecture (i.e., from the network core to the edge). 6G June-2017.pdf, last accessed 16 November 2021.
 9

[4] “SONATA Project: Service Programming and Orchestration for power grid,” Scientific Reports, vol. 11, no. 1, pp. 1–9, 2021.
virtualised Software Networks,” https://www.sonata-nfv.eu/, last [13] P. Yang, Y. Xiao, M. Xiao, and S. Li, “6G wireless communica-
accessed 16 November 2021. tions: Vision and potential techniques,” IEEE Network, vol. 33,
[5] “CogNet Project: Building an Intelligent System of Insight pp. 70–75, 7 2019.
and Action for 5G Network Management,” http://www.cognet. [14] N. Saeed, H. Almorad, H. Dahrouj, T. Y. Al-Naffouri, J. S.
5g-ppp.eu/, last accessed 16 November 2021. Shamma, and M.-S. Alouini, “Point-to-Point communication
[6] S. Bolettieri, R. Bruno, and E. Mingozzi, “Joint device asso- in Integrated Satellite-Aerial 6G Networks: State-of-the-art and
ciation and resource allocation for time-critical iot applications future challenges,” IEEE Open Journal of the Communications
in mec-empowered 5g networks,” in 2021 IEEE International Society, vol. 2, pp. 1505–1525, 2021.
Conference on Smart Computing (SMARTCOMP), 2021, pp.
175–180.
[7] Y. Zhang, “Mobile edge computing for beyond 5g/6g,” in Mariam Ishtiaq received her Master’s Degree in Electronics and
Mobile Edge Computing. Springer, 2022, pp. 37–45. Communication Engineering from Dongguk University, Seoul, South
[8] D. Unal, M. Hammoudeh, M. A. Khan, A. Abuarqoub, Korea. Her current research interests include underground sensor
G. Epiphaniou, and R. Hamila, “Integration of federated ma- networks and applications of cloud computing and intelligence in
chine learning and blockchain for the provision of secure big 5G and B5G networks. Contact her at: [email protected]
data analytics for internet of things,” Computers & Security,
vol. 109, p. 102393, 2021. Nasir Saeed (S’14-M’16-SM’19) is currently an Associate Professor
[9] P. Ranaweera, A. Jurcut, and M. Liyanage, “Mec-enabled 5g at Department of Electrical Engineering, Northern Border University,
use cases: A survey on security vulnerabilities and counter- Arar, KSA. His current areas of interest include localization, cognitive
measures,” ACM Computing Surveys (CSUR), vol. 54, no. 9, radio networks, underwater and underground communications, and
pp. 1–37, 2021. aerial networks. Contact him at: [email protected]
[10] N. Saeed, T. Y. Al-Naffouri, and M.-S. Alouini, “Wireless
communication for flying cars,” Frontiers in Communications Muhammad Asif Khan is a Post-Doctoral Research Fellow at Qatar
and Networks, vol. 2, p. 16, 2021. Mobility Innovation Center (a trade name of Qatar University QSTP-
[11] A. Alalewi, I. Dayoub, and S. Cherkaoui, “On 5G-V2X Use B). He received his Ph.D. degree in Electrical Engineering from
Cases and Enabling Technologies: A Comprehensive Survey,” Qatar University. His research focuses on mobile edge computing,
IEEE Access, vol. 9, pp. 107 710–107 737, 2021. distributed machine learning, and computer vision. He is a senior
[12] A. Eid, J. G. Hester, and M. M. Tentzeris, “5g as a wireless member of IEEE. Contact him at: [email protected].