Air ground integration approach

IEEE Communications Magazine • August 201840 0163-6804/18/$25.00 © 2018 IEEE
Abstract
Mobile edge computing (MEC), which enables
delay-sensitive computing tasks to be executed
at network edges, has been proposed to accom-
modate the explosive growth of the Internet
of Things. However, the growing demands for
massive connectivity, ultra-low latency, and high
reliability in various emerging resource-hungry
and computation-intensive applications pose new
challenges in network access capacity. This moti-
vates us to conceive a new MEC framework from
an air-ground integration perspective. First, we
present a review of the state-of-the-art literature.
Then the architecture and technological benefits
of the proposed framework are elaborated. Next,
four typical use cases are introduced, and a case
study is conducted to demonstrate the significant
performance improvements in computation capa-
bility and communication connectivity based on
real-world road topology. Finally, we present chal-
lenges and research directions, and conclude this
article.
Introduction
The Internet of Things (IoT), which connects a
great variety of physical things such as smart-
phones, sensors, home appliances, vehicles, and
wearable devices through a network, enables a
new paradigm shift for ubiquitous information
exchange and communication [1, 2]. However,
the long transmission distance between IoT devic-
es and remote data centers and the capacity-con-
strained backhaul links in the conventional cloud
computing paradigm impose new challenges
on reliable quality of service (QoS) and quality
of experience (QoE) provisioning. To this end,
mobile edge computing (MEC), in which comput-
ing and storage resources are placed at network
edges, has emerged as a promising solution. In
MEC, delay-sensitive computing tasks can be exe-
cuted in close proximity to IoT devices to reduce
response time and alleviate traffic congestion at
core networks [3].
However, the explosive demands for massive
connectivity, ultra-low latency, and high reliability
in delay-sensitive and multimedia-rich IoT appli-
cations pose new challenges in network access
capacity (i.e., the number of connections that
can be simultaneously accommodated by MEC
nodes). Since physical infrastructures of MEC are
generally deployed in a fixed fashion, this lack of
mobility results in inability to meet the compu-
tation and connectivity demands with the char-
acteristics of fast temporal, spatial, and spectral
variations. Considering a hot zone where a large
number of IoT devices try to connect to MEC
nodes simultaneously, this immense number of
connection requests and computing demands
will severely overwhelm the network access
and computing capacities of the MEC nodes,
which eventually incurs acute performance deg-
radation. In the opposite circumstances, a large
quantity of resources in the MEC node remain
underutilized when the loads within coverage
shift from peak to valley. Therefore, the mis-
match between irregular user demands and rigid
network capacity will seriously impede the QoS
and QoE guarantees and decrease resource uti-
lization efficiency.
To address the above challenges, we propose
an air-ground integrated MEC framework, which
actively explores the systematic and complemen-
tary integration of the communication, comput-
ing, and storage resources from both air and
ground segments to provide on-demand deploy-
ment densification of edge servers. The redundant
resources in unmanned aerial vehicles (UAVs) and
ground vehicles can be envisaged as supplemen-
tary edge computing servers for efficient resource
utilization and reliable service provisioning [4].
On one hand, it is illustrated that UAVs not only
provide vast coverage over wide geographical
areas, but also possess unique characteristics like
fast deployment, easy programmability, and high
scalability. For instance, the outage probability
of cell edge users can be effectively reduced by
the UAV-enabled light of sight (LoS) connectivity
and wide ground coverage. On the other hand,
vehicular edge computing, which distinguishes
itself from static MEC with its high geographical
distribution density and proximity to users, can
be deployed with UAVs in a complementary way
to provide service differentiation or to enhance
computing capability and communication connec-
tivity in urban areas. Specifically, the large volume
of vehicles in hotspots can provide additional
computing capability and multihop data deliv-
ery capacity. Furthermore, users who suffer from
intermittent connection and service interruption
when UAVs are running out of battery power can
be served by nearby vehicles in order to enable
seamless connection. Therefore, instead of rely-
ing on a single technology, this new paradigm
Zhenyu Zhou, Junhao Feng, Lu Tan, Yejun He, and Jie Gong
MULTIPLE ACCESS MOBILE EDGE COMPUTING FOR HETEROGENEOUS IOT
The growing demands
for massive connectivity,
ultra-low latency, and
high reliability in various
emerging resource-hun-
gry and computation-in-
tensive applications
pose new challenges in
network access capacity.
This has motivated the
authors to conceive a new
MEC framework from an
air-ground integration
perspective.
Zhenyu Zhou, Junhao Feng, and Lu Tan are with North China Electric Power University; Yejun He is with Shenzhen University;
Jie Gong is with Sun Yat-sen University. Jie Gong is the corresponding author.
Digital Object Identifier:
10.1109/MCOM.2018.1701111
An Air-Ground Integration Approach for
Mobile Edge Computing in IoT

IEEE Communications Magazine • August 2018 41
combines the advantages of different segments to
improve connection quality and capacity between
MEC nodes and IoT devices.
In this article, first, we present a literature
review. Then we narrate the proposed architec-
ture from three aspects, the cloud computing
layer, edge computing layer, and device layer,
with a particular emphasis on its technological
advancements. Next, four typical use cases are
elaborated to illustrate potential values of the air-
ground integrated MEC framework. A case study
is presented to demonstrate the significant perfor-
mance improvements in computing capability and
communication connectivity based on real-world
road topology. Finally, we discuss several open
research challenges and draw conclusions for this
article.
Related Works
In this part, we present an exhaustive commen-
tary on the development of MEC, vehicle or UAV
assisted MEC, and air-ground integration. Table 1
provides a brief classification of the related liter-
ature.
MEC has attracted a growing amount of atten-
tion in the research community. In [5], Bhaskar
et al. employed conventional cloud computing
and MEC to achieve performance gains in the
fiber-wireless (FiWi) broadband access network.
An enhanced architecture of MEC was proposed
in [6] to facilitate distributed application develop-
ment and deployment. In [7], Xu et al. provided
an efficient resource management algorithm and
addressed the challenges of integrating renewable
energy into MEC. An online joint computing and
radio resource management algorithm was devel-
oped for multi-user MEC systems in [8], where
the power consumption and long-term average
weight of MEC servers and devices were mini-
mized.
There are other works that have already pro-
vided some effective strategies to shift the func-
tionalities of MEC to vehicles or UAVs. In [9],
Hou et al. proposed a vehicular fog computing
(VFC) architecture to enhance the quality of appli-
cations and services. In [10], Sookhak et al. pre-
sented a fog vehicular computing (FVC) concept
that expanded fog computing capacity by utilizing
idle resources of vehicles and alleviated the acute
performance degradation during peak hours. In
[11], Motlagh et al. compared the performance
of local processing and UAV-assisted MEC pro-
cessing, demonstrating that the performance of
system responsiveness and energy harvesting can
be greatly improved by the latter. However, these
works ignore the specific research problems in
air-ground integrated MEC such as the interop-
eration and integration of air-ground resources,
the respective network reconfiguration, and the
mobility management.
A few works considered the research prob-
lems in vehicular networks based on space-air-
ground integration. In [12], Zhou et al. developed
an architecture of air-ground integrated coop-
erative vehicular networks to facilitate vehicu-
lar applications. In [13], Zhang et al. proposed
a space-air-ground integrated vehicular network
architecture, which exploits resources from both
space and air segments to enhance the connec-
tivity capacity for vehicular networks. However,
these works mainly target the research problems
in vehicular networks, while the use cases and
research challenges in the air-ground integrated
MEC have been neglected.
Table 1. A comprehensive summary of related works.
Category Literature Feature Merit Limit
MEC
• [5]
• Provide MEC services for FiWi networks
• Consider both MEC and conventional
cloud scenarios
• Achieve the performance gains in
FiWi broadband access network
• UAV or vehicle enabled MEC is not considered
• The air-ground integrated resources are neglected
• [6]
• Provide MEC services for distributed
applications
• Develop an effective middleware
• Enhance current MEC architecture by
deploying and developing distributed
applications more flexibly
• [7]
• Provide MEC services for renewable
energy
• Study energy harvesting MEC systems
• Increase the energy efficiency
• [8]
• Provide MEC services for multi-user systems
• Develop an online joint computing and
radio resource management algorithm
• Minimize the power consumption
and long-term average weight of
devices
UAV or vehicle
enabled MEC
• [9]
• Communication and computing-based on
vehicular resources
• Enhance the quality of applications
and services
• [10]
• Service distribution among vehicular fog
nodes
• Promote the performance of
computing and storage for fog
computing
• [11] • UAV-based MEC platform for IoT • Increase the task offloading efficiency • The air-ground integrated resources are neglected
Air-ground
integration
• [12] • Multi-UAV-aided vehicular networks
• Improve the performance of vehicular
networks
• The unique characteristics of MEC are ignored
• [13]
• Exploit the advantages of space, air, and
ground segments to promote vehicular
services
• Support diverse vehicular services
efficiently and cost-effectively
• The unique characteristics of MEC are ignored

IEEE Communications Magazine • August 201842
The Air-Ground
Integrated MEC Framework
In this section, we introduce the conceptual archi-
tecture of the proposed air-ground integrated
MEC framework and present the corresponding
technological benefits.
Architecture Overview
The air-ground integrated MEC framework is
conceived by exploring the advantages of soft-
ware-defined networking (SDN), which provides
centralized network control and flexible resource
management through the separation of data and
control functions and the advanced abstraction
of underlying infrastructures. Figure 1 illustrates
the conceptual architecture of the proposed air-
ground integrated MEC framework, which relies
on the deep convergence of three layers: the
cloud computing layer, edge computing layer,
and device layer.
The device layer is mainly composed of
different programmable and reconfigurable
equipments and elements spanning both the
air and ground network segments, including
UAVs, vehicles, base stations, routers, gate-
ways, and various IoT devices. The device layer
resources as well as the upper-layer comput-
ing and storage resources are extracted and
virtualized by a hypervisor, and are eventually
consolidated into a virtual resource pool. The
hypervisor takes over the driving of underlying
infrastructures, and provides an abstraction to
the controllers [14]. Through a proper mapping
between physical infrastructures and virtualized
network resources, multiple virtual MEC nodes
can be established and operated on the same
underlying infrastructures. Each MEC node
only occupies a slice of the overall virtualized
resources, and is independently controlled by
the upper-layer controllers.
The cloud and edge computing layers together
provide basic computing functionalities for IoT
applications. A hierarchical computing architec-
ture is adopted to handle computing task alloca-
tion and resource management at different scales.
The edge computing layer, which acts as a bene-
ficial complement to the cloud computing layer,
is more suitable for local processing of small-scale
delay-sensitive computing tasks in a distributed
fashion. In the edge computing layer, the com-
munication, computing, and storage resources are
deployed close to users and distributed across the
network edges. Lower-layer edge controllers can
only see and manage their own virtual networks,
and are controlled by the upper-layer cloud con-
trollers.
Compared to the edge computing layer, the
scale of resources in the cloud computing layer is
several orders of magnitudes larger; the resourc-
es are uniformly controlled and managed by the
cloud controller with centralized intelligence and
global network knowledge. Vendor-indepen-
dent and fine-grained control of the virtualized
resources becomes feasible via the data-control
decoupling and the high-level abstraction of phys-
ical infrastructures. To accommodate the differ-
ent QoS requirements of IoT applications, the
upper-layer controllers can reassemble the virtual
resources and allocate them to each lower-lay-
er controller, which then dynamically adjusts the
resource assignment based on traffic demands.
Benefits of Air-Ground Integrated Architecture
Connectivity Enhancement: The redundant
resources in both air and ground segments are
integrated in a complementary way to enhance
the connectivity capability between MEC nodes
Figure 1. The architecture of the air-ground integrated MEC.
Centralized
controller
Computing
resource
Storage
resource
Communicating
resource
Lower-layer
controller 1
Cloud computing
layer
Edge computing
layer
Virtualized resource
pool
Hypervisor
••••••
Device layer
UAV
Data link
Control link
Lower-layer
controller 2
Lower-layer
controller 3
Vehicle
UAV
Vehicle
Microcell
Gateway
Small cell
The air-ground integrat-
ed MEC framework is
conceived by exploring
the advantages of
software-defined net-
working (SDN), which
provides centralized
network control and
flexible resource man-
agement through the
separation of data and
control functions and
the advanced abstrac-
tion of underlying infra-
structures.

and IoT devices. Considering the scenario of a hot
pedestrian zone, the number of connections that
can be simultaneously afforded is severely limited
by the inferior non-LoS (NLoS) channel conditions
due to surrounding blockages. In comparison,
the utilization of UAVs can provide on-demand
augmentation of connectivity capacity due to the
wide coverage areas, fast deployment, and avail-
ability of LoS connections. Furthermore, vehicles
that are close to pedestrians can act as relays to
provide multihop data delivery, which expands
the effective coverage of MEC services.
Adaptive Resource Allocation: To match
resource provision with the time and space vary-
ing computing demands, the scale of resource
virtualization and the granularity of resource allo-
cation can be adjusted dynamically at different
levels. For instance, a number of interconnected
UAVs and vehicles, which contain both ground-
to-ground and ground-to-air links, and involve
a number of resources including bandwidth,
power, buffer, queue, and so on, can be mapped
to a single virtual link from a higher degree of vir-
tualization. The virtualized link resources are uni-
formly managed by the centralized controller for
load balancing and connectivity improvement.
The hierarchical control architecture also enables
flexible adjustment of resource allocation granu-
larity. Particularly, the micro-scale management
of load distribution, power control, spectrum
allocation, and user association is taken care
by the lower-layer edge controllers based on
instantaneous computing demands, while the
macro-scale management of edge server place-
ment, UAV/vehicle dispatch, and inter-segment
resource coordination are controlled by the
upper-layer cloud controllers based on long-term
observations and predictions.
Differentiated QoS Guarantees: In MEC, a
number of IoT applications with diverse require-
ments on throughput, latency, reliability, and con-
nectivity coexist in the same network. To cope
with these diverse QoS requirements and pro-
vide differentiated services, fine-grained resource
allocation is performed under the guidance of
application requests, which are first translated
into explicit instructions and delivered to the
respective controllers via standardized applica-
tion programming interfaces (APIs). With effective
data-control separation and an advanced level of
virtualization, the high-level QoS requirements
can be enforced by the hierarchical controllers
that dynamically adapt resource provisioning in
accordance with service demands. In addition
to adaptive resource orchestration, the outage
probability of users located far away from MEC
nodes can be significantly reduced by exploring
the connectivity enhancement capability provid-
ed by air-ground integration. UAVs and vehicles
that enable on-demand mobile connectivity can
provide effective data offloading for heavy-load
cells, which significantly enhances network access
capacity and QoS provisioning reliability.
Effective Mobility Management: The proposed
air-ground integrated MEC framework allows
effective mobility management via software-de-
fined air interface separation. UAVs with high alti-
tudes, large-scale coverage, and LoS connections,
can effectively eliminate unnecessary handover
and provide uninterrupted computing service
for high-mobility users. Furthermore, based on
the knowledge gained from the historical data
(e.g., the distribution of computing demands), the
macro-scale UAV/vehicle dispatch and the micro-
scale resource allocation can be jointly optimized
by the controllers. Since the virtualization of
Figure 2. Four typical use cases of the air-ground integrated MEC.
Move
Content caching
Scalable massive and mobile connectivity managementContent caching and mobile delivery
Distributed big data processing and analysisUltra-low latency support
AR applications
Crash
The proposed air-
ground integrated
MEC framework allows
effective mobility
management via soft-
ware-defined air inter-
face separation. UAVs
with high altitudes,
large-scale coverage,
and LoS connections,
can effectively eliminate
the unnecessary hando-
ver and provide unin-
terrupted computing
service for high-mobility
users.

underlying air and ground infrastructures is imple-
mented by the hypervisor in a transparent way,
the edge controllers can manage the respective
virtual resources in a highly abstracted manner
instead of directly interacting with hardware/soft-
ware heterogeneities, which significantly increases
the efficiency of inter-segment resource manage-
ment and coordination. Hence, the controllers are
able to reconfigure the network structures and
reassemble the virtualized resources with strict
timeliness in response to dynamically varying net-
work topologies and computing demands.
Use Cases
The three basic application scenarios of the
incoming fifth generation (5G) era are enhanced
mobile broadband (eMBB), massive machine-
type communications (mMTC), and ultra-reliable
low-latency communications (URLLC). In this sec-
tion, we present four use cases to demonstrate
how to support a grand set of 5G applications
with high throughput, low latency, and high mobil-
ity requirements with the proposed air-ground
integrated MEC framework. Particularly, the use
case “ultra-low latency support” corresponds to
the URLLC scenario, the use cases “distributed
big data processing and analysis” and “content
caching and mobile delivery” correspond to the
eMBB scenario, and the use case “scalable mas-
sive and mobile connectivity management” corre-
sponds to the mMTC scenario. The use cases are
shown in Fig. 2, and are detailed as follows.
Ultra-Low Latency Support: In the proposed
framework, the infrastructures of air-ground inte-
gration rely on underlying MEC devices such as
vehicular networks or UAVs instead of remote
cloud computing servers; thus, latency issues can
be improved significantly and user experience
can be enhanced effectively. In particular, the LoS
connections provided by UAVs or vehicles can
dramatically improve connection quality, which
is essential to support delay-sensitive applica-
tions. Furthermore, with the implementation of
air-ground integration, timely reactions and adap-
tive solutions can be achieved in disaster relief
and emergency management. For instance, when
a traffic accident happens, the accident vehicle
directly reports to the surrounding vehicles and
UAVs, which then provide real-time audio/video
monitoring for further inspection and manage-
ment. In practice, a secret Google project, Sky-
Bender, has already demonstrated that data can
be delivered 40 times faster compared to existing
4G LTE systems.
Distributed Big Data Processing and Analy-
sis: Different from conventional cloud comput-
ing or static edge computing with fixed location
and computing capability, the computing capac-
ity in the air-ground integration framework can
be augmented dynamically in accordance with
traffic demands. For example, a group of UAVs
that are interconnected with each other via LoS
air-to-air links can jointly process a large amount
of traffic data by exchanging some key data with
each other. In other words, the onboard data
processing capacity can be shared among UAVs
through the air-to-air connections that can be
approximated as free space communication. With
such distributed data processing capability, the
complicated network behaviors that involve large
volumes of data can be analyzed in real time to
flexibly adjust resource allocation and reconfigure
network structures according to dynamic traffic
profiles.
Scalable Massive and Mobile Connectivity
Management: By combining the technological
advantages of each segment, static base stations
with constant power supply, fixed backhaul con-
nection, and large transmission power can be
utilized to provide access service, while UAVs
and vehicles provide on-demand content deliv-
ery and computing service by exploiting their
mobility capability. In this sense, scalable massive
connectivity can be supported via opportunistic
scheduling and dynamic connectivity manage-
ment. Taking vehicular networks as an example,
the MEC based on mobile multihop connections
can be regarded as a supportive mechanism for
realizing emerging intelligent transport system
(ITS) applications such as self-driving, which not
only collects real-time information from roadside
sensors and vehicular applications directly, but
also enables local message analysis and warning
(e.g., accidents) broadcasting to nearby vehicles.
In addition, UAVs can also provide on-demand
mobile connectivity for public safety surveillance.
For instance, when a fire occurs, a group of UAVs
can be dispatched immediately to the fire site
to provide live video footage and static images,
allowing proactive reconnaissance and decision
making during the search and rescue operation.
Content Caching and Mobile Delivery: The
proposed framework can reduce the content dis-
tribution latency and eliminate network overload
via mobile content caching and delivery from the
following three aspects. First, the data that are
frequently requested can be selectively cached
in the air-ground integrated edge servers with-
out any further replicated transmission in the
backhaul. Second, vehicles that are proximate to
pedestrians and UAVs with flexible deployment
and reconfigurability are combined to optimize
network access and wireless coverage, which
in turn allows more users to fetch the contents
cached in the mobile edge servers previously, and
hence, further minimizes the duplicate content
Figure 3. The real-world road topology of the Xidan area.
Bank of China
PICC Building
Xuanwumen Inner ST.
Vehicle
Xidan Leisure Park
West Chang’an Avenue
China Minsheng Bank
UAV

transmission. Last but not least, with the mobility
support of UAVs and ground vehicles, the dupli-
cate content caching in different locations can
be greatly reduced, as they can roam across the
network to move the popular contents along with
user trajectories.
Case Study
In this section, a case study is conducted to eval-
uate the performance of the proposed air-ground
integrated MEC.
Experimental Setup
The simulation is implemented based on two
main platforms: SUMO and MATLAB. SUMO is
adopted to generate vehicular traffic in real-world
road topologies. As a representative business and
financial district in Beijing, China, the Xidan area
is selected as the evaluation scenario, which is
characterized with large-scale office buildings
and commercial districts as well as pedestrian
zones. Figure 3 shows a snapshot of the Xidan
area obtained from Google Maps. The digital map
information is available in OpenStreetMap, and
the corresponding data are imported to SUMO
by using the NETCONVERT tool. Then, based on
the road topologies including nodes, edges, con-
nection links, traffic signs, and so on, vehicular
traffic is generated by using the RANDOMTRIPS
tool with a step of 10–1 s. The total duration of
the simulation is set as 3.6  103 s. The coordi-
nates and velocity of each vehicle can be saved
into an Excel file.
The flying trajectory of UAVs is developed
based on [15], with an altitude around 60~150
m. The battery capacity of a small-type UAV is
20,000 mAh. The flight duration is around 30
min [11]. The computing capability of any UAV
is set as 0.7~1.0 GHz. The transmission power
of any UAV or vehicle is set as 23 dBm. We
assume that there are 1000~2000 mobile users
who are randomly distributed along the road-
side based on the real-road topology. Each user
might carry a couple of IoT devices, such as
smartphones, smart watches, and smart glasses.
The computing tasks generated for each user
are assumed to follow a Poisson process, which
are offloaded to MEC nodes (UAVs or vehicles).
The total delay of the offloaded portion arises
from task uploading, queuing, processing, and
result feedback. The traffic model of any MEC
node is considered as an M/M/c queue by sup-
posing that the service time obeys an exponen-
tial distribution, where c denotes the number of
co-located edge servers.
Afterward, the simulation of task offloading is
conducted in MATLAB. Some simulation param-
eters such as user location, number of computing
tasks, and channel models are generated based
on theoretical models rather than real-world mea-
surements. The reason is that these theoretical
models (e.g., the free space propagation path loss
model) have been intensively adopted in many
previous works and provide good tractability. The
real-world implementation and experiment will be
left as future work.
Numerical Results
The proposed framework is compared to three
heuristic schemes: the baseline scheme, the
vehicle-assisted scheme, and the UAV-assisted
scheme. For the purpose of fair comparison, the
numbers of static edge servers in all of the four
schemes are set the same to remove the impacts
of static edge servers and fairly compare the per-
formance gain achieved by integrating vehicles or
UAVs. Meanwhile, the total numbers of vehicles
and UAVs are kept as the same for the vehicle-as-
sisted scheme, the UAV-assisted scheme, and the
proposed scheme.
Figure 4a shows the relationship between
the number of users and the average delay per-
formance. It is observed that the average delay
performances of all four schemes increase mono-
tonically with the number of users. The reason is
that the time spent on task uploading, queuing,
and processing will increase accordingly as more
users try to connect to the same edge nodes and
offload computing tasks. Nevertheless, numerical
results demonstrate that the average delay can
be reduced by a significant margin via efficient
utilization of the air-ground integrated resourc-
Figure 4. Performance of the air-ground integrated MEC framework: a) average delay vs. the number of users.; b) CDF of successful
service connections vs. the number of users.
The number of users
(a)
11001000
1500
1000
Averagedelay(ms)
2000
2500
3000
3500
4000
4500
5000
1200 1300 1400 1500
The number of users
(b)
12001100
0.1
0
CDFofsuccessfulserviceconnections
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1300 1400 1500 1600 1700 1800 1900 2000
Baseline
Vehicle-assisted scheme
UAV-assisted scheme
Proposed scheme
Proposed scheme
UAV-assisted scheme
Vehicle-assisted scheme
Baseline

es. The average delay achieved by the proposed
scheme is 48 percent lower than that of the base-
line scheme when the number of users reach-
es 1400. The reason is that UAVs and vehicles
with LoS communication links provide effective
countermeasures to relieve the negative impacts
of multi-path fading. It is also observed that the
proposed scheme outperforms the UAV-assisted
scheme. The reason is that vehicles with flexible
deployment and reconfigurability are conceived
in a mutually beneficial way to enhance network
access capacity. For example, the large volume of
vehicles in hotspots that are close to pedestrians
can significantly reduce transmission delay due to
the shorter transmission distance.
Figure 4b shows the cumulative distribution
function (CDF) of successful service connec-
tions vs. the number of users. Here, the suc-
cessful service connections represent the ratio
of users whose computing requests are success-
fully accepted by MEC nodes. It is noted that
when the number of users increases to 1600,
only 12 percent of service requests can be
successfully connected in the baseline scheme
due to the access and backhaul related limita-
tions. In comparison, the proposed scheme out-
performs the baseline scheme by 88 percent
due to the fact that the integrated air-ground
resources with LoS connectivity and large-scale
coverage can effectively enhance QoS perfor-
mances of cell edge users and provide more
reliable communication links than the traditional
static connections. Furthermore, from both Fig.
1 and Fig. 2, we observe that the performance
of the UAV-assisted scheme always outperforms
the vehicle-assisted scheme. The reason is that
connection interruption occurs when the asso-
ciated vehicle moves away and there is no alter-
nate vehicle in proximity to continue service
provisioning, which is less likely to happen for
UAVs with larger coverage.
Challenges and Research Directions
Despite the tremendous benefits brought by the
concept and architecture for providing lower
latency and high reliability services to IoT devices,
there are still several challenges.
Coordination among UAVs and Ground Vehicles
To establish a network by UAVs and ground
vehicles, inter-UAV-vehicle communication and
coordination are required. Due to the mobility
of UAVs and ground vehicles, the coordination
among them is not as simple as the fixed infra-
structures with fiber-based backhaul. These
mobile edge servers have to be interconnected
via wireless links. One way is that each individual
UAV coordinates with its neighbors distributive-
ly, which requires little controlling overhead but
leads to higher inaccuracy. Such a strategy might
also result in an error spreading problem. The
other way is to nominate a cluster head to collect
the information of all the UAVs to make a central-
ized decision. In this case, the cluster head may
be overwhelmed by the immense overheads so
that additional computing cannot be afforded. In
this case, the round-robin protocol in which each
UAV with fewer computing tasks acts as a cluster
head for a short duration may be considered as a
candidate solution.
Joint Trajectory Design and MEC Optimization
One of the key design challenges is to adaptive-
ly adjust UAV trajectory to meet the computing
task requirements. From a single-user perspective,
research efforts should focus on how to predict
the user movement and track the trajectory so
that the computing tasks can be offloaded imme-
diately, and the computation results can be fed
back to users on time. For multi-user networks,
the design of UAV trajectory becomes even more
challenging, where the distribution and mobili-
ty of users, task priority and delay requirements,
user locations and distance to UAVs, as well as
energy consumption issues need to be jointly con-
sidered. Intuitively, users can be scheduled in a
time-division multiplexing way, that is, only one
user is served per channel by the UAV when they
are close enough to each other. Then the serving
order of users becomes critical, where not only
the transmission delay, but also the queuing and
computation delays should be taken into account.
In fact, computing and communication can be
scheduled in different time slots to fully utilize the
resources of UAVs.
Energy-Efficient Management
The limited onboard battery capacity is another
key constraint of UAVs as edge servers. Besides
the energy consumed in hovering, and acceler-
ating/de-accelerating for climbing up/down and
speed up/slow down, additional communication
and computing energy consumption, as well as
the additional weight gained by mounting the
communication and computing modules, have
to be taken into consideration. There have been
some efforts to model the propulsion energy con-
sumption of UAVs, while the onboard energy
consumption of communication and computing
requires further investigation. On the other hand,
UAV station replacement, recharging schedul-
ing, as well as energy harvesting can be consid-
ered as alternate directions to compensate the
battery capacity limitation. Compared to existing
works that solely rely on UAV-assisted commu-
nications, the proposed air-ground integration
approach provides a promising way to address
the energy consumption problem. For example,
users in hotspots with high vehicle density can be
served by nearby vehicles to reduce the energy
consumption of UAVs. Furthermore, since the
transmission distance from vehicles to pedestrians
is much shorter compared to that of UAVs, the
system energy efficiency can also be increased
significantly.
Conclusions
In this article, we have proposed an air-ground
integrated MEC framework by exploiting the
benefits of high mobility and flexible computing
resource allocation of UAVs and vehicles. By
smartly managing the aviation trajectory of UAVs
and seeking the LoS air-ground channels, the pro-
posed framework can support numerous IoT use
cases including content caching and mobile deliv-
ery, distributed big data processing and analysis,
ultra-low latency support, and so on. Numerical
results based on the case study for a real-world
road environment show that the overall delay can
be greatly reduced with the proposed air-ground
integrated MEC framework.
The limited onboard
battery capacity is
another key constraint
of UAVs as edge serv-
ers. Besides the energy
consumed in hovering,
and accelerating/de-ac-
celerating for climbing
up/down and speed up/
slow down, additional
communication and
computing energy
consumption, as well
as the additional weight
gained by mounting
the communication and
computing modules,
have to be taken into
consideration.

Acknowledgments
This work was partially supported by the Beijing
Natural Science Foundation under Grant Num-
ber 4174104, Beijing Outstanding Young Talent
under Grant Number 2016000020124G081, the
National Natural Science Foundation of China
(NSFC) under Grant Numbers 61601181 and
61771495, and the Fundamental Research Funds
for the Central Universities under Grant Number
2017MS001.
References
[1] Y. Zhang et al., “Cognitive Machine-to-Machine Commu-
nications: Visions and Potentials for the Smart Grid ,” IEEE
Network, vol. 26, no. 3, May 2012, pp. 6–13.
[2] Y. Wu et al., “Energy-Aware Cooperative Traffic Offload-
ing via Device-to-Device Cooperations: An Analytical
Approach,” IEEE Trans. Mobile Comp., vol. 16, no. 1, Jan.
2017, pp. 97–114 .
[3] X. Huang et al., “Exploring Mobile Edge Computing for
5G-Enabled Software Defined Vehicular Networks,” IEEE
Wireless Commun., vol. 24, no. 6, Dec. 2017, pp. 55–63.
[4] C. Wu et al., “A Reinforcement Learning-Based Data Storage
Scheme for Vehicular Ad Hoc Networks,” IEEE Trans. Vehic.
Tech.,vol. 66, no. 7, July 2017, pp. 6336–48.
[5] B. P. Rimal et al., “Mobile-Edge Computing Versus Central-
ized Cloud Computing Over a Converged FiWi Access Net-
work,” IEEE Trans. Net. Service Mgmt., vol. 14, no. 3, Sept.
2017, pp. 498–513.
[6] A. Carrega et al., “A Middleware for Mobile Edge Comput-
ing,” IEEE Cloud Comp., vol. 4, no. 4, July/Aug. 2017, pp.
26–37.
[7] J. Xu et al., “Online Learning for Offloading and Autoscaling
in Energy Harvesting Mobile Edge Computing,” IEEE Trans.
Cogn. Commun. Net., vol. 3, no. 3, Sept. 2017, pp. 361–73.
[8] Y. Mao et al., “Stochastic Joint Radio and Computational
Resource Management for Multi-User Mobile-Edge Comput-
ing Systems,” IEEE Trans. Wireless Commun., vol. 16, no. 9,
Sept. 2017, pp. 5994–6009.
[9] X. Hou et al., “Vehicular Fog Computing: A Viewpoint of
Vehicles as the Infrastructures,” IEEE Trans. Vehic. Tech., vol.
65, no. 6, June 2016, pp. 3860–73.
[10] M. Sookhak et al., “Fog Vehicular Computing: Augmenta-
tion of Fog Computing Using Vehicular Cloud Computing,”
IEEE Vehic. Tech. Mag., vol. 12, no. 3, July 2017, pp. 55–64.
[11] N. H. Motlagh et al., “UAV-Based IoT Platform: A Crowd
Surveillance Use Case,” IEEE Commun. Mag., vol. 55, no. 2,
Feb. 2017, pp. 128–34.
[12] Y. Zhou et al., “Multi-UAV-Aided Networks: Aerial-Ground
Cooperative Vehicular Networking Architecture,” IEEE Vehic.
Tech. Mag., vol. 10, no. 4, Dec. 2015, pp. 36–44.
[13] N. Zhang et al., “Software Defined Space-Air-Ground Inte-
grated Vehicular Networks: Challenges and Solutions,” IEEE
Commun. Mag., vol. 55, no. 7, July 2017, pp. 101–09.
[14] Z. Zhou et al., “Software Defined Machine-to-Machine
Communication for Smart Energy Management,” IEEE Com-
mun. Mag., vol. 55, no. 10, Oct. 2017, pp. 52–60.
[15] V. Sharma et al., “UAV-Assisted Heterogeneous Networks
for Capacity Enhancement,” IEEE Commun. Letters, vol. 20,
no. 6, Apr. 2016, pp. 1207–10.
Biographies
Zhenyu Zhou [M’11, SM’17] (zhenyu_zhou@ncepu.edu.cn)
received his Ph.D degree from Waseda University, Japan, in
2011. Since March 2013, he has been an associate professor
at North China Electric Power University. He is an Editor for
IEEE Access and IEEE Communications Magazine, and Workshop
Co-Chair for IEEE ISADS ’15, GLOBECOM ’18, and others. He
received the 2017 IET Premium Award, and the IEEE ComSoc
GCCTC 2017 Best Paper Award. His research interests include
green communications and vehicular communications.
Junhao Feng (fengjh@163.com) is currently working toward
an M.S. degree at North China Electric Power University.
His research interests include resource allocation, interfer-
ence management, and energy management in D2D com-
munications.
Lu Tan (13717568897@163.com) is currently working
toward an M.S. degree at North China Electric Power Uni-
versity. Her research interests include green communications
and smart grid.
Yejun He (heyejun@126.com) received his Ph.D.degree in
information and communication engineering from Huazhong
University of Science and Technology in 2005. He is a full pro-
fessor with the College of Information Engineering, Shenzhen
University, China, where he is the director of the Guangdong
Engineering Research Center of Base Station Antennas and
Propagation, and the director of the Shenzhen Key Laboratory
of Antennas and Propagation. His research interests include
wireless mobile communication, antennas, and radio frequency.
He is a Fellow of IET.
Jie Gong [S’09, M’13] (gongj26@mail.sysu.edu.cn) received
his B.S. and Ph.D. degrees from the Department of Elec-
tronic Engineering at Tsinghua University in 2008 and 2013,
respectively. He visited the University of Edinburgh in 2012.
During 2013– 2015, he worked as a postdoctorial scholar at
Tsinghua University. He is currently an associate professor
in the School of Data and Computer Science, Sun Yat-sen
University, Guangzhou, China. His research interests include
cloud RAN, energy harvesting technology, and green wireless
communications.
By smartly managing
the aviation trajectory
of UAVs and seeking
the LoS air-ground
channels, the proposed
framework can support
numerous IoT use cases
including content cach-
ing and mobile delivery,
distributed big data
processing and analysis,
ultra-low latency sup-
port, and so on.

Air ground integration approach

More Related Content

What's hot (18)

Similar to Air ground integration approach (20)

Recently uploaded (20)

Air ground integration approach