1
Abstract—Unmanned Aerial Vehicles (UAVs) have enormous
potential in the public and civil domains. These are particularly
useful in applications where human lives would otherwise be
endangered. Multi-UAV systems can collaboratively complete
missions more efficiently and economically as compared to single
UAV systems. However, there are many issues to be resolved
before effective use of UAVs can be made to provide stable and
reliable context-specific networks. Much of the work carried out in
the areas of Mobile Ad Hoc Networks (MANETs), and Vehicular
Ad Hoc Networks (VANETs) does not address the unique
characteristics of the UAV networks. UAV networks may vary
from slow dynamic to dynamic; have intermittent links and fluid
topology. While it is believed that ad hoc mesh network would be
most suitable for UAV networks yet the architecture of multi-UAV
networks has been an understudied area. Software Defined
Networking (SDN) could facilitate flexible deployment and
management of new services and help reduce cost, increase
security and availability in networks. Routing demands of UAV
networks go beyond the needs of MANETS and VANETS.
Protocols are required that would adapt to high mobility, dynamic
topology, intermittent links, power constraints and changing link
quality. UAVs may fail and the network may get partitioned
making delay and disruption tolerance an important design
consideration. Limited life of the node and dynamicity of the
network leads to the requirement of seamless handovers where
researchers are looking at the work done in the areas of MANETs
and VANETs, but the jury is still out. As energy supply on UAVs is
limited, protocols in various layers should contribute towards
greening of the network. This article surveys the work done
towards all of these outstanding issues, relating to this new class of
networks, so as to spur further research in these areas.
Index Terms—Unmanned Aerial Vehicle, UAV, Multi-UAV
Networks, ad hoc networks, communication networks, wireless
mesh networks, software defined network, routing, seamless
handover, energy efficiency
I. INTRODUCTION
A. The growing importance of UAV networks
Unmanned Aerial Vehicles (UAVs) are an emerging
*Formal corresponding author
The manuscript was submitted on 14
th
October 2014.
Lav Gupta is pursuing doctoral program in Computer Science & Engineering
at Washington University in St Louis, MO 63130 USA (email:
lavgupta@wustl.edu).
Raj Jain is Professor of Computer Science & Engineering at Washington
University in St Louis, MO 63130, USA (email: [email protected]l.edu).
Gabor Vaszkun is with Ericsson Hungary (email: [email protected])
technology that can be harnessed for military, public and civil
applications. Military use of UAVs is more than 25 years old
primarily consisting of border surveillance, reconnaissance and
strike. Public use is by the public agencies such as police,
public safety and transportation management. UAVs can
provide timely disaster warnings and assist in speeding up
rescue and recovery operations when the public communication
network gets crippled. They can carry medical supplies to areas
rendered inaccessible. In situations like poisonous gas
infiltration, wildfires and wild animal tracking UAVs could be
used to quickly envelope a large area without risking the safety
of the personnel involved.
UAVs come
in various sizes.
Large UAVs
may be used
singly in
missions while
small ones may
be used in
formations or
swarms. The
latter ones are
proving to be
quite useful in
civilian
applications. As
described by Daniel and Wietfeld in [1] they are likely to
become invaluable inclusions in the operations of police
departments, fire brigades and other homeland security
organizations in the near future. Besides, advances in
electronics and sensor technology have widened the scope of
UAV network applications [2] to include applications as diverse
as traffic monitoring, wind estimation and remote sensing [3].
In this context it would be relevant to mention that the
current FAA guidelines allow a government public safety
agency to operate unmanned aircraft weighing 4.4 pounds or
less, within the line of sight of the operator; less than 400 feet
above the ground; during daylight conditions; within Class G
airspace; and outside of 5 statute miles from any airport,
heliport, seaplane base, spaceport, or other location. For model
aircrafts FAA guidance does not address size of the model
aircraft. The guidelines of Federal Aviation Authority [4] say
that model aircraft flights should be kept below 400 feet above
ground level (AGL), should be flown a sufficient distance from
populated areas and full scale aircraft, and are not for business
Survey of Important Issues in UAV
Communication Networks
Lav Gupta*, Senior Member IEEE, Raj Jain, Fellow, IEEE, and Gabor Vaszkun
At a recent show in Decatur, Illinois, on farming
applications of small UAVs drew 1400 attendees
from 33 states and 6 countries. According to Stu
Ellis, an organizer, “You could spend four to five
hours walking an 80-acre soybean field,” and noted
that the same ground could be covered in half an
hour or less by a drone. However, in its June 2014
notice, FAA has made it clear that drones that are
not being used by hobbyists or any commercial use
must have prior FAA approval. The agency
specifically mentions farming – along with
photography and delivery services – as types of
businesses subject to regulation. Meanwhile,
Association for Unmanned Vehicles Systems
International has projected an $82 billion economic
impact for the period 2015-2025. (Adapted from St
Louis Post dated 07/27/2014)
Published in IEEE Communications Surveys and Tutorials, Volume PP, Issue 99, November 2015
2
purposes.
B. The challenges of UAV networks
Promising though it may be, this area is relatively new and
less explored. There are many issues to resolve before effective
use of UAVs can be made to provide stable and reliable
context-specific networks. As we shall see later, while it offers
the promise of improved capability and capacity, establishing
and maintaining efficient communications among the UAVs is
challenging.
All the constituents of the UAV communication networks
pose challenging issues that need resolution. Unlike many other
wireless networks, the topology of UAV networks remains fluid
with the number of nodes and links changing and also the
relative positions of the nodes altering. UAVs may move with
varying speeds depending on the application, this would cause
the links to be established intermittently. What challenges
would such a behavior pose? Firstly, some aspects of the
architectural design would not be intuitive. The fluid topology,
the vanishing nodes and finicky links would all challenge the
designer to go beyond the normal ad hoc mesh networks.
Second, the routing protocol cannot be a simple implementation
of a proactive or a reactive scheme. The inter-UAV backbone
has to repeatedly reorganize itself when UAVs fail. In some
cases the network may get partitioned. The challenge would
then be to route the packet from a source to a destination while
optimizing the chosen metric. The third challenge would be to
maintain users’ sessions by transferring them seamlessly from
an out of service UAV to an active UAV. Lastly, there need to
be ways of conserving energy of power starved UAVs for
increasing the life of the network. In the next section we bring
out all of these issues in more detail.
C. Motivation and key Issues
The area of UAV networks is challenging to researchers
because of the outstanding issues that provide motivation for
research. In mobile and vehicular networks the nodes join and
dissociate from the network frequently and, therefore, ad hoc
networks have been found to be suitable in most situations. In
addition, for quick and reliable communication between nodes,
mesh network topology is quite appropriate. Does this apply to
the UAV networks as well? In UAV networks, the nodes could
almost be static and hovering over the area of operation or
scouting around at a rapid pace. Nodes could die out for many
reasons and may be replaced by new ones. Some similarities
encourage researchers to explore the applicability of the work
done for Mobile Ad hoc Networks (MANETs) and Vehicular
Ad hoc Networks (VANETs), but works in these areas do not
fully address the unique characteristics of the UAV networks.
Table I gives important characteristics of MANETs, VANETs
and UAV networks which bring out similarities and
dissimilarities among them. We shall characterize UAV
networks in more depth in section II(C).
TABLE I
COMPARATIVE DESCRIPTION OF DIFFERENT AD-HOC NETWORKS
MANET
VANET
UAV Networks
Mobile wireless nodes
connect with other
nodes within
communication range
in an ad-hoc manner
(No centralized infra-
structure required)
Ad-hoc networks in
which vehicles are
the mobile nodes.
Communication is
among vehicles and
between vehicles
and road side units
Ad-hoc or
infrastructure based
networks of
airborne nodes.
Communication
among UAVs and
with the control
station
Slow. Typical speeds
2 m/sec. Random
movement. Varying
density, higher at
some popular places
High-speed,
typically 20-30 m/s
on highways, 6-10
m/s in urban areas.
Predictable, limited
by road layout,
traffic and traffic
rules
Speeds from 0 to
typically as high as
100 m/s. Movement
could be in 2 or 3
dimensions, usually
controlled
according to
mission.
Random, ad-hoc
Star with roadside
infrastructure and
ad-hoc among
vehicles
Star with control
center, ad-hoc/mesh
among UAVs.
Dynamic - nodes join
and leave
unpredictably.
Network prone to
partitioning.
More dynamic than
MANETs.
Movement linear.
Partitioning
common.
Stationary, slow or
fast. May be flown
in controlled
swarms. Network
prone to partitioning
Most nodes are battery
powered so energy
needs to be conserved.
Devices may be car
battery powered or
own battery
powered.
Small UAVs are
energy constrained.
Batteries affect
weight and flying
time
• Information
distribution
(emergencies,
advertising,
shopping, events)
Internet hot spots
• Traffic & weather
info, emergency
warnings,
location'based'
services
• Infotainment
• Rescue operations
• Agriculture-crop
survey
• Wildlife search
• Oil rig surveillance
'
It can be observed from Table 1 and some of the references
[3] [123] that there are many aspects of UAV networks that set
them apart from mobile ad-hoc and vehicular ad-hoc networks.
The important differences are described here:
The mobility models, like random walk, that have been used
to describe the behavior of nodes in mobile ad hoc networks
and street random walk or Manhattan models for vehicular ad
hoc networks are not quite suitable for UAV networks. The
UAVs could move randomly or in organized swarms not only
in two but also in three dimensions with rapid change in
position. The vehicular nodes are constrained to travel on roads
and only in two dimensions.
Changes in topology could be much more frequent in UAV
networks. Relative positions of UAVs may change; some
UAVs may lose all their power and may need to be brought
down for recharging; UAVs may malfunction and be out of the
network; the links may form and vanish because of the
changing positions of the nodes. In many applications density
of nodes may not be high and the network may partition
frequently. Vehicular networks have roadside infrastructure to
support communication among vehicles. The network being
fluid, the mobile stations that are locked on to a particular UAV
for communication access would need to be transferred to
3
another UAV seamlessly, without disrupting user sessions. The
research community is still trying to figure out the most
effective routing protocol as well as the seamless handover
procedure [7].
Energy constraints are much greater in small UAV networks.
While in vehicular networks power could be drawn from the car
battery which gets charged while the car is in motion. Even the
mobile ad hoc networks would typically have nodes
(smartphones, laptops) with power sources that last a few hours.
Small UAVs may typically have enough power for a flight of
thirty minutes! On one hand the transmitted signal would be of
lower power and on the other the links would be intermittent
owing to power-drained UAVs drifting away or dying.
Dynamicity of nodes would force the network to organize and
re-organize frequently. This gives rise to unique routing
requirements. The routing protocols need to use energy
efficiently so as to prolong the stability of the UAV network.
The UAV networks would usually be deployed in dire
circumstances and the network may get frequently partitioned,
sometimes for long durations. In such cases traditional solutions
do not guarantee connectivity. It has been suggested that these
networks be made delay and disruption tolerant by
incorporating store-carry-forward capabilities. If we presume
that each node has a global knowledge of the network we could
apply deterministic protocols but this may not always be a
correct assumption. On the other hand if we presume random
behavior of nodes we have to face scalability issues.
A multi-UAV network, which is fully autonomous, requires a
robust inter-UAV network with UAVs to cooperate in keeping
the network organized even in the event of link or node failure.
The UAV networks would require changes at the MAC and
network layers, have self-organizing capabilities, be tolerant to
delays, have a more flexible and automated control through
SDN and employ energy saving mechanisms at various layers
[5], [6].
That these issues need thorough attention is corroborated in a
well-detailed survey on UAV based flying ad hoc network [3].
Authors of this survey have tried to put together many of the
issues and developments in this area. However, there are issues
of self-organization, disruption tolerance, SDN control,
seamless handover and energy efficiency that we believe,
would be extremely important for building successful UAV
networks, have not been the foci of the study. Our survey
attempts to primarily take up these issues to bring out the
current status and possible directions.
The organization of rest of the paper is as follows: In Section
II, we attempt to categorize UAV networks and examine
important characteristics like topology, control, and client
server behavior. We also see the important aspects of self-
organization and automated operations through software
defined networking (SDN) that will help in identifying the work
required in this area. In Section III, we discuss requirements
from the routing protocols peculiar to UAV networks and the
need for disruption tolerant networking. In Section IV, we see
the importance of seamless handover and the need for new
research in this area. Finally, in Section V, the protocols used at
various layers for energy conservation are discussed.
II. CHARACTERIZING THE UAV NETWORK
It is important to characterize a network to understand its
nature, constraints and possibilities. How fast does the topology
change with time? How frequently does the network get
partitioned as the nodes fail or move away? How can the
network life be increased? What type of architecture would be
more suitable? Does it require self-organizing, self-healing
capabilities? Which protocols can be run at different layers?
Does it support addition and removal of nodes dynamically?
Are the links intermittent and what is their quality? In this
section we look at the characteristics that forms the common
thread in various works and the direction in which the research
is headed. Subsection A contains characteristics of multi-UAV
systems and their advantages over single UAV systems.
Subsection B discusses important features that set them apart
from other ad hoc networks and also distinguishes them from
each other. Subsection C provides categorization of UAV
networks based on some of the important characteristics
discussed. Subsection D brings forth the self-organizing
behavior of UAV networks. Lastly, Subsection E deals with
less uncovered aspect of use of SDN for centralizing and
automating control of UAV networks.
A. Multi-UAV network
Early uses of UAV were characterized by use of a single
large UAV for a task. In these systems the UAV based
communication network, therefore, consisted of just one aerial
node and one or more ground nodes. Today most public and
civil applications can be carried out more efficiently with multi
UAV systems. In a multi-UAV system, the UAVs are smaller
and less expensive and work in a coordinated manner. In most
multi-UAV systems, the communication network, proving
communication among UAVs and between the UAVs and the
ground nodes, becomes an important constituent. These UAVs
can be configured to provide services co-operatively and extend
the network coverage by acting as relays. The degree of
mobility of UAVs depends on the application. For instance, in
providing communication over an earthquake struck area the
UAVs would hover over the area of operation and the links
would be slow dynamic. As opposed to this, applications like
agriculture or forest surveillance require the UAVs to move
across a large area and links frequently break and reestablish.
The dynamic nature of the network configuration and links is
apparent from the fact that the UAVs may go out of service
periodically due to malfunction or battery drainage. This is true
also for UAVs that need to hover over an area for relatively
long periods. New UAVs have to be launched to take their
places. Sometimes some of the UAVs may be taken out of
service to conserve power for a more appropriate time. It
would, therefore, be a requirement that in all such cases the
links should automatically reconfigure themselves. Though
advantageous in many respects, multi UAV systems, add
complexities to the UAV communication network.
Some of the key advantages of multi-UAV systems are
reliability and survivability through redundancy. In a multi-
UAV system, failure of a single UAV causes the network to re-
4
organize and maintain communication through other nodes.
This would not be possible in a single UAV system. However,
to reap the real benefits of the multiple UAVs working in
collaboration, the protocols deployed need to take care of the
issues like changing topology, mobility and power constraints.
In terms of communication needs, single UAV systems would
have to maintain links with the control station(s), base stations,
servers and also provide access functionality. This puts a heavy
constraint on the limited battery power and bandwidth. In a
multi-UAV system, only one or two UAVs may connect to
control and servers and feed the other UAVs. This way most
UAVs just have to sustain the mesh structure and can easily
offer access functions for calls, video and data. Multi-UAV
systems also turn out to be less expensive to acquire, maintain
and operate than their larger counterparts. As shown through
their experiments by Mergenthaler et al. in [8], adding more
UAVs to the network can relatively easily extend
communication umbrella provided by a multi-UAV system.
Missions are generally completed more speedily, efficiently and
at lower cost with small UAV systems as compared to a single
UAV system [9]. In their work on PRoPHET-based routing
protocol [10] the authors explain how in opportunistic networks
multi-UAVs find a path even if two end points are not directly
connected leading to completion of missions. In their work on
Multi-UAV cooperative search in [11] the authors describe how
multi-UAV systems complete searches faster and are robust to
loss of some UAVs. Advantages of multi-UAV networks are
leading to their increasing use in civilian applications [1]. In
this paper we focus on the UAV networks that use multiple
small UAVs to form an unmanned aerial system (UAS). Table
II gives a comparison of single and multi-UAV systems.
TABLE II
COMPARISON OF SINGLE AND MULTI UAV SYSTEMS
FEATURE
SINGLE UAV
SYSTEM
MULTI UAV
SYSTEM
Impact of failure
High, mission
fails
Low, system
reconfigures
Scalability
Limited
High
Survivability
Poor
High
Speed of mission
Slow
Fast
Cost
Medium
Low
Bandwidth
required
High
Medium
Antenna
Omni-directional
Directional
Complexity of
control
Low
High
Failure to
coordinate
Low
Present
An issue that is important in the context of multi-UAV
systems, but not a subject of this study is coordination and
control for effective task planning with multiple UAVs. An
efficient algorithm is necessary to maneuver each UAV so that
the whole system can produce complex, adaptable and flexible
team behavior. The task-planning problem for UAV networks
with connectivity constraint involves a number of parameters
and interaction of dynamic variables. An algorithm has been
proposed for such a situation [11]. There is also an algorithm
for distributed intelligent agent systems in which agents
autonomously coordinate, cooperate, negotiate, make decisions
and take actions to meet the objectives of a particular task. The
connectivity constrained problem is NP-hard and a polynomial
time heuristic has been proposed in the literature [13]. This is a
challenging field but we would restrict our attention to UAV
based communication networks.
B. Features of the UAV networks
Developing a fully autonomous and cooperative multi-UAV
system requires robust inter-UAV communication. We do not
have enough research to say with conviction what design would
work best. There are a number of aspects of the UAV networks
that are not precisely defined and a clarification of these would
help in characterizing the UAV networks:
1) Infrastructure-based or ad hoc?
Most of the available literature treats UAV networks as ad
hoc networks. Research on MANETs and VANETs are often
cited with reference to UAV networks but they do not
completely address the unique characteristics of the UAV
networks. Depending on the application, the UAV network
could have stationary, slow moving or highly mobile nodes.
Many applications require UAV nodes to act as base stations in
the sky to provide communication coverage to an area. Thus,
unlike MANET and VANET ad hoc networks, the UAV
networks could behave more like infrastructure-based networks
for these applications. These would have UAVs communicating
with each other and also with the control center. Such a network
would resemble the fixed wireless network with UAVs as base
stations except that they are aerial. There is, however, a class of
applications where the nodes would be highly mobile and
would communicate, cooperate and establish the network
dynamically in an ad hoc manner. In such a case the topology
may be determined, and the nodes involved in forwarding data
decided, dynamically. There are many issues that affect both
UAV infrastructure based and UAV ad hoc networks. For
example, replacing nodes by new nodes when they fail or their
power gets exhausted.
2) Server or client?
Another point of distinction is whether the node acts as a
server or a client. In vehicular networks they are usually clients,
in mobile ad hoc networks most of the time they would be
clients and may also provide forwarding services to other
clients’ data. In UAV networks, the UAV nodes are usually
servers, either routing packets for clients or relaying sensor data
to control centers.
3) Star or Mesh?
Architecture of UAV networks for communication
applications is an understudied area. The simplest configuration
is a single UAV connected to a ground based command and
control center. In a multi-UAV setting, the common topologies
that can be realized are star, multi-star, mesh and hierarchical
mesh. In the case of star topology all UAVs would be
connected directly to one or more ground nodes and all
5
communication among UAVs would be routed through the
ground nodes. This may result in blockage of links, higher
latency and requirement of more expensive high bandwidth
downlinks. In addition, as the nodes are mobile, steerable
antennas may be required to keep oriented towards ground node
[15]. The multi-star topology is quite similar except the UAVs
would form multiple stars and one node from each group
connects to the ground station. Figures 1a and 1b shows star
and multi-star configurations.
'
Fig. 1a) Star Configuration b) Multi-star Configuration c) Flat Mesh Network
d) Hierarchical Mesh Network'
Star configurations suffer from high latency as the downlink
length is longer than inter-UAV distance and all communication
must pass through the ground control center. Also, if the ground
center fails there is no inter-UAV communication. In most
civilian applications, however, normal operation does not
require communication among UAVs to be routed through the
ground node. An architecture that supports this would result in
reduced downlink bandwidth requirement and improved latency
because of shorter links among UAVs. In case of mesh
networks the UAVs are interconnected and a small number of
UAVs may connect to the control center [16]. Figures 1c and 1d
show flat and hierarchical mesh networks.
Some authors believe that conventional network technologies
cannot meet the needs of UAV networks. Related literature
points to the applicability of mesh networks for civilian
applications [15, [9]. There are usually multiple links on one or
more radios, interference between channels, changes in
transmitted power due to power constrains, changes in number
of nodes, changes in topology, terrain and weather effects. In ad
hoc networks the nodes may move away, formations may
break, and therefore, the links may be intermittent. Wireless
mesh networks, adapted suitably, may take care of some of
these problems. To tackle these issues the network has to be
self-healing with continuous connection and reconfiguration
around a broken path.
As compared to star networks, mesh networks are flexible,
reliable and offer better performance characteristics. In a
wireless mesh network, nodes are interconnected and can
usually communicate directly on more than one link. A packet
can pass through intermediate nodes and find its way from any
source to any destination in multiple hops. Fully connected
wireless networks have the advantages of security and
reliability. Such a network can use routing or flooding
technique to send messages. The routing protocol should ensure
delivery of packets from source to destination through
intermediate nodes. There are multiple routes and the routing
protocol should select the one that meet the given objectives.
The routing devices can organize themselves to create an ad hoc
backbone mesh infrastructure that can carry users’ messages
over the coverage area through multiple hops. Moreover, they
can also route packets originated from the command and
control center and directed to emergency operators or people
and vice-versa. The control center can process data to extract
information for the support of decisions during emergency [17].
Due to unique features of UAV nodes described above,
sometimes the existing networks routing algorithms, which
have been designed for mobile ad hoc networks (MANETs),
such as BABEL or the Optimized Link-State Routing (OLSR)
protocol, fail to provide reliable communications [8], [14]. We
shall provide more details about routing in the next section. The
major differences between star and mesh networks are given in
Table III.
TABLE III
COMPARISON OF STAR AND MESH NETWORKS PROPERTIES'
'
Star Network
Mesh Network
Point-to-point
Multi-point to multi-point
Central control point present
Infrastructure based may have a control
center, Ad hoc has no central control
center
Infrastructure based
Infrastructure based or Ad hoc
Not self configuring
Self configuring
Single hop from node to central
point
Multi-hop communication
Devices cannot move freely
In ad hoc devices are autonomous and
free to move. In infrastructure based
movement is restricted around the
control center
Links between nodes and central
points are configured
Inter node links are intermittent
Nodes communicated through
central controller
Nodes relay traffic for other nodes
Scalable
Not scalable
4) Delay and Disruptions prone networks
All wireless mobile networks are prone to link disruptions.
The UAV networks are no exception. The extent of disruption
depends on how mobile the UAVs are, the power transmitted,
inter-UAV distances and extraneous noise. In the applications
where UAVs provide communication coverage to an area, the
UAVs are hovering and, therefore, probability of disruptions
would be low. On the other hand, in applications requiring fast
UAV mobility, there is a higher likelihood of disruptions.
Delays in transmitting data could be because of poor link
quality or because one or more UAV nodes storing the data
because of end-to-end path not being available. We will see
more details of this in Section III.
C. Categorization of UAV Networks
We are now ready to categorize the UAV networks. In
Subsection 2.2 we have seen a number of features that could
form the basis of this categorization. So how do we categorize
UAV networks, the applications of which require varying
UAVs
Ground Control Center
(a)
(b)
UAVs
Ground!Control!Center!
(c)!
(d)!
6
degree of node mobility, different network architectures,
routing and control? In the ultimate analysis it is the
applications that would be important. Therefore, considering
distinct applications, and then deciding the properties that
would needed to make these applications possible, could lead us
to a generic categorization as given in Table IV.
Like Internet delivery there are many applications in which
UAV assisted wireless infrastructure needs to be established for
providing area wide coverage. Disaster struck areas, remote
villages or deep sea oil rigs would all require quick deployment
of a UAV based network that could provide voice, video and
data services. In these applications the UAVs hover over an
area and are virtually stationary. These can be considered to be
cellular towers or wireless access points in the sky. In contrast,
sensing applications like detection of forest fires or survey of
crops would require mobile UAVs. There are other
applications, especially military, where fast moving UAVs
would be required to make forays into enemy territory. We
would focus here largely on the first category of applications
and to some extent on the sensing applications. When UAVs
are used to create wireless communication infrastructure then
depending on the application, all UAVs could be directly under
control of the ground control center or they could form a
wireless mesh network with one or two UAVs communicating
with the control center. In these applications the UAVs act as
servers for routing users’ communication and control
information. This is distinct, for example, from the cases where
UAVs carrying sensors are used to collect information or those
sent out for carrying out attack. In these cases, UAVs act as
clients. Delay or disruption probability in the Internet delivery
class of applications is much less as compared to other
applications because the UAVs are stationary and formations
are easily coordinated. When a UAV fails, the network would
be expected to reconfigure and the ongoing sessions would
have to be seamlessly transferred to other UAVs. In other
applications where nodes are more dynamic, the links may
function intermittently and special routing, reconfiguration, and
disruption handling features may be required. In the following
sections we shall see in more detail the state of research in these
areas.
TABLE IV
UAV CATEGORIZATION!
!
Property Internet Delivery Sensing Attack
Other general applications in
the category
Disaster communication,
Oil exploration
Remote health
Reconnaissance, search,
detecting forest fires, tracking
wild animals
War: Multi-UAV attack
UAV Position during
communication
Fixed position Slow change, coordinated
movement
Frequent change
Mobility speed of UAV during
communication
~0 miles/hour <10 miles/hour >10 miles/hour
Network Infrastructure based, base
station in the sky
Infrastructure based Infrastructure based/Infrastructure
less, Ad-hoc
Topology Star/Mesh Mesh Mesh
Control (communication) Centralized (position control
based)
Centralized ( task control based) Distributed (task control based)
Individuals controlling each UAV
UAV as Client or Server Server (routes
communication and control)
Server (when receiving from
sensors) / client(when carrying
sensors
Server (delivering info to
formations) /client (for attack)
Routing Through server (control from
central, data through central
or among UAVs)
Central or mesh (control from
central data to central, also data
among UAVs)
Mesh routing (control from central,
data among UAVs)
Delay/disruption (because of
node and link failures)
Low probability (p<0.1) Medium probability
(0.1<=p<0.5)
High probability (p>=0.5)
Type of Communication
C=Client, I=Infrastructure
U2U, U2I, C2U U2U (if buffer node), U2I, U2C U2I (commands), U2U
Control( path control, position) Remote – position control Remote – position control, path
control
Remote (less likely)
Auto (observe, orient, decide, act)
Path Control – real-time multi UAV
coordination, collision avoidance
7
D. Self-organization in networks
One reason why mesh networks are considered suitable for
UAV based networks is because of their self-forming and
reorganization features. Once the nodes have been configured
and activated they form mesh structure automatically or with
guidance from the control center. When this happens, the
network becomes resilient to failure of one or more nodes.
There is inherent fault tolerance in mesh networks. When a
node fails, the rest of the nodes reconfigure the network among
them. By the same token a new node can be easily introduced.
Support for ad hoc networking, self-forming, self-healing and
self-organization enhance the performance of wireless mesh
networks, make them easily deployable and fault tolerant [18]
[19].
Studies on self-organization in the context of sensor networks
and wireless ad hoc networks can help in understanding the
requirements of UAV networks. Self-organization consists of
the following steps: When a node fails or a new node appears,
its neighbor(s) finds out about the available nodes through the
neighbor discovery process. Changes in the network, in the
form of removal or addition of devices in the network, cause a
number of nodes to exchange messages for re-organization.
This could cause collisions while accessing the medium and
impact network performance. Medium access control deals with
control of access to the medium and minimizes errors due to
collisions. The next step involves establishment of connectivity
between nodes during self-organization through local
connectivity and path establishment. Once the connectivity has
been established, the service recovery management process
carries out network disruption avoidance and recovery from
local failures. Finally, energy management carries out load
balancing of data forwarding responsibilities in the self-
organized network and also the processes involved in reducing
energy consumption in the battery operated devices. The goal in
UAV networks would then be to ensure connectivity of all
active network nodes so that the mesh network is maintained
through multi-hop communication to provide best access to the
users [20].
Wireless mesh networks are prone to link failure due to
interference, mobility or bandwidth demand. This will lead to
degradation of network performance but can be effectively
tackled by making the network reconfigurable. The nodes
monitor their links and any failures trigger the reconfiguration
process. The autonomous network reconfiguration system
requires computational overhead and reasonable bandwidth.
There are times when some of the UAVs could go out of
service because of battery drain or communication failure. In
such cases the remaining nodes in the network re-organize and
re-establish communication. While the benefits of self-
organization are huge and encouraging, challenges in self-
organization are bigger making this an exciting research
problem [21].
E. SDN-automating UAV network control
The UAV networks are limited in communication resources.
Nodes are non-permanent, connectivity is intermittent and
channels may be impaired. This translates into utilization
challenges in planning and allocation of resources. Different
networks utilize different routing protocols and consequently
even the nodes that use the same access technology may not
work in another network because of the differences in higher
layers of the protocol stack. For instance vehicular networks use
IEEE 802.11p Wireless Access Vehicular Environment
(WAVE) to support Intelligent Transportation Systems (ITS)
applications. For wireless mesh networks, the IEEE 802.11s
amendment has been standardized for self-configuring multi-
hop technologies. But in both environments there is no
consensus on the routing protocols to be used, as well as on
most of the network management operations to be performed.
As a consequence, nodes using a particular access technology in
a network may not operate in another network with same access
but different higher layer protocols.
The above problems could be tackled by building in the
capability of defining the protocol stack in software. This way
the UAVs could be programmed to flexibly work in different
environments. However, this is not the only reason why
software control of the network is desirable. There are a number
of other requirements arising in case of networks like MANET,
VANET and UAV networks. They need to support dynamic
nodes and frequent changes in the topology. Nodes may fail, for
example, because of battery drainage and need to be replaced
by new nodes. The links are intermittent and need to be
accordingly dealt with. SDN provides a way to
programmatically control networks, making it easier to deploy
and manage new applications and services, as well as tune
network policy and performance [22], [23].
Deployment of SDN has been extensive in fixed
infrastructure-based networks. However, much of this
deployment has been in datacenters, as it was believed that
SDN was suitable for networks with centralized control and
wireless ad hoc mesh networks were decentralized. Separation
of forwarding devices and the controller also raised the concern
for security, balance of control and flexibility. There are several
policy issues relating to balance of control and cooperation
between controllers that need to be addressed. Because of the
benefits it is expected to provide, increasing interest is being
shown by the academia and the industry in application of SDN
in dynamic mobile wireless environments. In networks like
VANETs, the use of SDN can help in path selection and
channel selection. This helps in reducing interference,
improving usage of wireless resources including channels and
routing in multi-hop mesh networks. Despite increasing interest
there is no clear and comprehensive understanding of what are
the advantages of SDN in infrastructure-less wireless
networking scenarios and how the SDN concept should be
expanded to suit the characteristics of wireless and mobile
communications [22], [24].
One of the commonly used protocols for implementing SDN
in wireless networks is OpenFlow. OpenFlow is claimed to
deliver substantive advantages for mobile and wireless
networks. It helps in optimizing resource usage in a dynamic
environment, provides a way to automate operations, allows
finer level of control and easier implementation of the global
8
policies and faster introduction of new services [25]. OpenFlow
protocol separates forwarding, and control functionalities. The
OpenFlow switches are programmable and contain flow tables
and protocol for communication with controller(s) [26]. Figure
2 shows separation of control and forwarding functionalities
with OpenFlow interface between control and data planes.
!
Fig. 2: Software Defined Networking elements
Such a network may be formed by mounting the data plane
or the OpenFlow switches on the UAVs and the control facility
on a centralized ground controller or distributed control on
UAVs. The forwarding elements, which are simple OpenFlow
switches, rather than IP routers, contain the flow tables that are
manipulated by the controller to set the rules. The actions in the
flow tables define processing that would be applied to the
identified set of packets. Actions could be filtering, forwarding
a particular port, header rewriting etc. Figure 3a shows how the
centralized control plane base SDN can be implemented in a
UAV network.
'
'
Fig. 3: a) SDN with centralized control b) SDN with distributed control'
The control plane of the SDN network could be centralized,
distributed or hybrid. The controller has a global view of the
network and can effectively route traffic. It updates flow tables
of the OpenFlow switches with matching criteria and
processing actions [27]. In the centralized mode, that we saw
above, the controller, that defines all the actions that wireless
switches in the UAV take, has been shown to be on the ground.
It could be air-borne as well. Figure 3b shows distributed
control of the SDN network. In the distributed control mode the
control is distributed on all the UAVs and each node controls its
behavior. In the hybrid mode the controller delegates control of
packet processing to the local agent and there is control traffic
between all SDN elements.
Researchers at Karlstad University, Sweden have
demonstrated the feasibility of integration of OpenFlow with
wireless mesh networks on their KAUMesh testbed on a small
scale [28]. Simulation by authors in [23] shows the comparison
of SDN-based VANET routing to other traditional
MANET/VANET routing protocols, including GPSR, OLSR,
AODV, and DSDV. They observe that SDN-based routing
outperforms the other traditional Ad hoc routing protocols in
terms of packet delivery ratio at various speeds. The major
reason for this is the aggregated knowledge that the SDN
controller has. In [29] the authors summarize the current
situation by saying that SDN offers enhanced configuration,
improved performance and encourages innovation but it is still
in infancy and many fundamental issues still remain not fully
solved. Capabilities of SDN to fulfill various requirements of a
UAV mesh network are given in Table V.
TABLE V
UAV COMMUNICATION REQUIREMENTS AND SDN CAPABILITIES
Feature required in UAV
SDN capability
Support for node mobility
Reconfiguration through orchestration
mechanism
Flexible switching and routing
strategies
Flexible definition of rules based on
header or payload for routing data.
Dealing with unreliable wireless
links
Selection of paths and channels
Greening of network
Supports switching off devices when
not in use. Supports data aggregation
in the network
Reduce interference
Can be done through path/channel
selection
!
III. ROUTING
The UAV networks constituted for various applications may
vary from slow dynamic to the ones that fly at considerable
speeds. The nodes may go out of service due to failure or power
constraints and get replaced by new ones. In green networks,
radios in the nodes may be automatically switched off for
power conservations when the load is low. Link disruption may
frequently occur because of the positions of UAVs and ground
stations. Additionally, the links could have high bit error rates
due to interference or natural conditions. The reliability
requirements from the UAV networks are also diverse. For
example, while sending earthquake data may require a 100%
reliable transport protocol, sending pictures and video of the
earthquake may be done with lower reliability but stricter delay
and jitter requirements. Bandwidth requirements for voice, data
and video are different. The UAV networks, therefore, have all
the requirements of mobile wireless networks and more. Node
mobility, network partitioning, intermittent links, limited
resources and varying QoS requirements make routing in UAV
a challenging research task. Several existing routing protocols
have been in the zone of consideration of researchers and quite
a few variations have been proposed. While newer protocols
attempt to remove some shortcoming of the traditional ones, the
field is still open. Subsection A discusses the important issues
that routing protocols need to deal with. Applicability of
Control'Plane'
OpenFlow'
Packet'
forwarding'
elements''
SDN Controller
Control plane communication
(flow rules)
Data plane communication
Ground Control Center
Control plane communication
(flow rules)
Data plane communication
Ground Control Center
9
various traditional protocols to UAV networks has been
discussed in Subsection B. Also included in this subsection are
new variants of these traditional protocols as well as some
recently proposed UAV specific routing protocols. UAV
networks are prone to delays and disruptions and Subsection C
discusses the routing protocols that can be used in such
situations.
A. Routing issues to be resolved
In addition to the requirements present in the generic wireless
mesh networks, e.g., finding the most efficient route, allowing
the network to scale, controlling latency, ensuring reliability,
taking care of mobility and ensuring required QoS; routing in
airborne networks requires location-awareness, energy-
awareness, and increased robustness to intermittent links and
changing topology. Designing the network layer for UAV
networks is still one of the most challenging tasks [3].
There are some papers that have looked into the use of
existing routing protocols for possible use in airborne networks.
Although conventional ad hoc routing protocols are designed
for mobile nodes, they are not necessarily suitable for airborne
nodes because of varying requirements of dynamicity and link
interruptions. Therefore, there still exists a need for a routing
protocol tailored to the particular needs of airborne networks
that adapts to high mobility, dynamic topology and different
routing capabilities [30]. Routing protocols try to increase
delivery ratio, reduce delays and resource consumption.
Additionally, one has to consider issues related to scalability,
Loop freedom, energy conservation, and efficient use of
resources also needs to be resolved [31].
B. Applicability of existing routing protocols
A number of routing protocols that have been proposed for
MANETs try to adapt the proactive, table based, protocols of
the wired era to ad hoc wireless networks with mobile nodes.
Many of these and also on-demand or reactive protocols suffer
from routing overhead problems and consequently have
scalability and bandwidth issues. Conditional update based
protocols reduce overheads but location management remains
an issue in dynamic networks like UAV networks. Some
protocols that introduce the concept of cluster heads introduce
performance issues and single point of failure [32]. A survey on
WMN stationary or mobile nodes, points out that available
MAC and routing protocols do not have enough scalability and
the throughput drops significantly as the number of nodes or
hops increase. It goes on to add that protocol improvement in a
single layer will not solve all the problems so all existing
protocols need to be enhanced or replaced by new ones for
UAV Networks [33].
Due to apparent similarity of UAV networks with MANETs
and VANETs researchers have studied protocols used in those
environments for possible application in aerial networks. Even
in these environments the search for more improved protocols
is on. However, multi-UAV networks may have a number of
different requirements to take care of e.g. mobility patterns and
node localization, frequent node removal and addition,
intermittent link management, power constraints, application
areas and their QoS requirements. Due to many of these issues
peculiar to UAV networks, while modifications have been
proposed to MANET protocols, there is a need to develop new
routing algorithms to have reliable communication among
UAVs [7] and from UAVs to the control center(s). We shall
discuss a number of networking protocols using the following
well-known classification with a view to see their usefulness for
UAV Networks: 1) Static protocols 2) Proactive protocols 3)
On-demand or Reactive protocols, and 4) Hybrid protocols.
1) Static Routing Protocols
Static protocols have static routing tables, which are
computed and loaded when the task starts. These tables cannot
be updated during an operation. Because of this constraint,
these systems are not fault tolerant or suitable for dynamically
changing environment. Their applicability to UAV networks is
limited and is only presented here for academic interest.
• One such protocol is Load Carry and Deliver Routing
(LCAD) [34] in which the ground node passes the data to a
UAV, which carries it to the destination. It aims to
maximize the throughput while increasing security.
Because of use of a single UAV, data delivery delays are
longer in LCAD but it achieves higher throughput. LCAD
can scale its throughput by using multiple UAVs to relay
the information to multiple destinations. It is possible to
use this for delay tolerant and latency insensitive bulk data
transfer.
• Multi Level Hierarchical Routing [MLHR] [7] solves the
scalability problems faced in large-scale Vehicular
networks. These networks are normally organized as flat
structures because of which performance degrades when
the size increases. Organizing the network as hierarchical
structure increases size and operation area. Similarly UAV
networks can be grouped into clusters where only the
cluster head has connections outside cluster. The cluster
head disseminates data by broadcasting to other nodes in
the cluster. In UAV networks frequent change of cluster
head would impose large over head on the network. Figure
4 depicts this pictorially.
!
Fig. 4: Multi-Level Hierarchical Routing in UAV networks
• Data Centric Routing: Routing is done based on the
content of data. This can be used for one-to-many
Affected Area
Affected'Area'
CH
CH
CH: Cluster Head
10
transmission in UAV networks when the data is
requested by a number of nodes, It works well with
cluster topologies where cluster head is responsible for
disseminating information to other nodes in the cluster
[3].
2) Proactive routing protocols
Proactive routing protocols (PRP) use tables in their nodes to
store all the routing information of other nodes or nodes of a
specific region in the network. The tables need to be updated
when topology changes. The main advantage of proactive
routing is that it contains the latest information of the routes.
But to keep the tables up-to-date a number of messages need to
be exchanged between nodes. This makes them unsuitable for
UAV network because of bandwidth constraints. Another issue
that makes them unsuitable for UAV networks is their slow
reaction to topology changes causing delays [7]. Two main
PRPs that have been used in MANETS/VANETS are
Optimized Link State Routing (OLSR) and Destination-
Sequenced Distance Vector (DSDV) protocols.
• Optimized Link State Routing (OLSR) is currently one of
the most employed routing algorithms for ad hoc networks.
Routes to all destinations are determined at startup and then
maintained by an update process. Nodes exchange topology
information with other nodes of the network regularly by
broadcasting the link-state costs of its neighboring nodes to
other nodes using a flooding strategy. Other nodes update
their view of the network by choosing the next hop by
applying shortest path algorithm to all destinations. OLSR
[124], therefore, tracks network topology. In UAV the node
location and interconnecting links change rapidly and this
would cause increased number of control messages to be
exchanged. Increased overhead of topology control
messages would lead not only to contention and packet loss
but would also put a strain on already constrained
bandwidth of UAV networks. Optimization involves
selecting some nodes as multipoint relays (MPR), which
alone forward the control traffic reducing the transmission
required. MPRs declare link state information to all the
nodes that selected them as an MPR. MPRs also
periodically announce to the network that it has
reachability to all the nodes that selected it as an MPR.
OLSR link-quality extension can be used to take into
account the quality of links. The details are given in RFC
3636 [35] and an evaluation on several metrics in [36].
While flooding of control signals is avoided with addition
of MPRs, recalculation of routes is still an issue with
limited computing power of UAV nodes. GSR, global state
routing, is another variation of link state routing that
provides improvement by restricting update messages only
between intermediate nodes. The size of update message is
large and intermediate nodes keep changing in aerial
networks. This increases both overhead and bandwidth
required. FSR, fisheye state routing, attempts to reduce
overhead by more frequent small updates to the closer
“within fisheye scope” nodes. In networks where the nodes
are mobile this may introduce inaccuracy. Another
variation called STAR, source-tree adaptive routing,
reduces the overheads by making the update dissemination
conditional rather than periodic. However, it increases
memory and processing overheads in large and mobile
networks. DREAM, distance routing effect algorithm for
mobility, has a different approach, as each node knows its
geographical coordinates using a GPS. These coordinates
are periodically exchanged between each node and stored
in a routing table (called a location table). This consumes
lesser bandwidth than link state. It is more scalable.
Frequency of updates can be made proportional to velocity
of nodes to reduce overhead. The maximum velocity is
used to calculate the possible distance that the destination
could move. This process is repeated at each hop with an
undefined recovery mechanism if there are no one-hop
neighbors within the wedge [125]. However, the authors in
[126] have found that the complexity of DREAM does not
appear to provide benefits over a simple flood.
• Destination Sequenced Distance Vector (DSDV) is a table-
driven proactive routing protocol, which mainly uses the
Bellman–Ford algorithm with small adjustments for ad hoc
networks. It uses two types of update packets-“full-dump”
and “incremental”. Whenever the topology of the network
changes, “incremental-dump” packets are sent. This
reduces overhead but overhead is still large due to periodic
updates. DSDV uses sequence numbers to determine
freshness of the routes and avoid loops. True to the
characteristics of proactive protocols, this protocol needs
large network bandwidth for update procedures. Along
with this the computing and storage burden of a proactive
protocol puts DSDV at a disadvantage for aerial networks.
In [127] authors have compared performance of DSDV
with other protocols.
• BABEL is based on distance-vector routing protocol. It is
suitable for unstable networks as it limits frequency and
duration of events such as routing loops and black holes
during re-convergence. Even when mobility is detected it
quickly converges to a loop free, not necessarily optimal,
configuration. BABEL, explained in RFC 6126 [37] has
provisions for link quality estimation. It can implement
shortest-path routing or use a metric. One of its downside is
that it relies on periodic routing table updates and generates
more traffic than protocols that send updates when the
network topology changes [32]. In [128] Rosati et al. have
shown on that based on datagram loss rate and average
outage time, BABEL fails to deliver in the UAV
environment.
• Better Approach to Mobile Ad Hoc Network
(B.A.T.M.A.N.) [38] is a relatively new proactive routing
protocol for Wireless Ad hoc Mesh Networks that could be
used in environments like Mobile Ad hoc Networks
(MANETs). The protocol proactively maintains
information about the existence of all nodes in the mesh
that are accessible via single-hop or multi-hop
communication links. For each destination the next hop
neighbor is identified which is used to communicate with
11
this destination. B.A.T.M.A.N algorithm only cares about
the best next-hop for each destination. It does not calculate
the complete route, which makes a very fast and efficient
implementation possible. Thus communication links may
have varying quality in terms of packet loss, data rates, and
interference. New links may appear and known links
disappear frequently. B.A.T.M.A.N. considers these
challenges by doing statistical analysis of protocol packet
loss and propagation speed and does not depend on state or
topology information from other nodes. Routing decisions
are based on the knowledge about the existence or lack of
information. As nodes continuously broadcast origin
messages (OGMs); without packet loss, these messages
would overwhelm the network. The scalability of
B.A.T.M.A.N. depends on packet loss and thus it is unable
to operate in reliable wired networks. B.A.T.M.A.N.
protocol packets contain only a very limited amount of
information and are therefore very small. If some protocol
packets are lost, the information about them helps in better
routing decisions. B.A.T.M.A.N.'s crucial point is the
decentralization of the knowledge about the best route
through the network — no single node has all the data. A
network of collective intelligence is created. This approach
has shown in practice that it is reliable and loop-free [39].
The self-interference caused by data traffic leads to
oscillations in the network. The batman-adv performs
better than B.A.T.M.A.N [129]. Comparative studies of
B.A.T.M.A.N. have shown varying results. Some studies
show performance of B.A.T.M.A.N. to be same as OLSR.
When maximum number of nodes with maximum packet
length scenario is taken, OLSR performed better than
B.A.T.M.A.N. With mobility factor also OLSR performed
better than B.A.T.M.A.N.. In bandwidth-limited networks
all protocols gave similar throughputs but without
bandwidth restrictions BABEL and B.A.T.M.A.N. perform
better than OLSR for large number of users (15 or more).
For small network sizes (5 mesh nodes), the three routing
protocols behave similarly [40], [41]. In another study with
a static wireless network of 7x7 grid of nodes,
B.A.T.M.A.N. outperforms OLSR on almost all
performance metrics. In another study B.A.T.M.A.N.
performed 15% better than OLSR [42].
3) Reactive routing protocols
Reactive Routing Protocol (RRP) is an on-demand routing
protocol in which a route between a pair of nodes is stored
when there is communication between them. RRP is designed
to overcome the overhead problem of proactive routing
protocols. Because of on-demand nature, there is no periodic
messaging making RRP bandwidth efficient. On the other hand,
the procedure of finding routes can take a long time; therefore,
high latency may appear during the route finding process.
Reactive protocols can be of two types: source routing and hop-
by-hop routing. In source routing each data packet carries the
complete source to destination address so intermediate nodes
can forward packets based on this information. No periodic
beaconing is required to maintain connectivity. This does not
scale well as route failure probability increases with the
network size and also the header size grows increasing the
overhead. In hop-by-hop routing, each data packet only carries
the destination address and the next hop address. Intermediate
nodes maintain routing table to forward data. The advantage of
this strategy is that routes are adaptable to the dynamically
changing environment. The disadvantage of this strategy is that
each intermediate node must store and maintain routing
information for each active route and each node may require
being aware of their surrounding neighbors through the use of
beaconing messages [8], [32]. Two commonly used RRPs are
Dynamic Source Routing (DSR) and AODV.
• Dynamic Source Routing (DSR) has been designed mainly
for multi-hop wireless mesh ad hoc networks of mobile
nodes. It allows networks to be self-organizing and self-
configuring without the need for any existing network
infrastructure. DSR works entirely on demand and it scales
automatically to what is needed to react to changes in the
routes currently in use. It’s “route discovery” and “route
maintenance” mechanisms allow nodes to discover and
maintain routes to arbitrary destinations. It allows selection
from multiple routes to any destination and this feature can
be used for applications like load balancing. It guarantees
loop-free routing [43]. When applied to UAV networks,
finding a new route with DSR can be cumbersome [44].
Each packet has to carry addresses of all nodes from source
to destination making it unsuitable for large networks and
also for networks where topology is fluid.
• Ad hoc On Demand Distance Vector (AODV) is a hop-by-
hop reactive routing protocol for mobile ad hoc networks.
It adjusts well to dynamic link conditions, has low
processing and memory overhead, low network utilization,
and determines unicast routes to destinations within the ad
hoc network [45]. It is similar to DSR but unlike DSR each
packet has only destination address so overheads are lower.
The route reply in DSR carries address of every node while
in AODV only the destination IP address. In AODV, the
source node (and also other relay nodes) stores the next-
hop information corresponding to each data transmission.
There is minimal routing traffic in the network since routes
are built on demand. It has been studied for possible
adaptation in UAV networks. However, there are delays
during route construction and link or node failures may
trigger route discovery that introduces extra delays and
consumes more bandwidth as the size of the network
increases. The throughput drops dramatically as
intermittent links become more pervasive. The number of
RREQ messages increases at first, as the route discovery is
triggered more frequently and after some point starts to
decrease as network performance is so impacted that even
RREQ messages can no longer be carried over the network
[30]. Shirani et al. [139] have proposed a combined routing
protocol, called the Reactive-Greedy-Reactive (RGR), for
aerial applications. This protocol combines Greedy
Geographic Forwarding (GGF) and reactive routing
mechanisms. The proposed RGR employs location
information of UAVs as well as reactive end-to-end paths
12
in the routing process. Simulation results show that RGR
outperforms existing protocols such as AODV in search
missions in terms of delay and packet delivery ratio.
A number of researchers have compared various protocols in
different settings. In one investigation, it has been found that for
TCP traffic and mobile nodes, in terms of routing overheads,
DSR outperforms AODV, TORA (discussed later) and OLSR
as it sends the least amount of routing traffic into the network.
In terms of throughput, OLSR outperformed other protocols in
all the scenarios considered [46]. In another investigation effect
of some modifications in AODV has been studied. AODV has a
flat structure and has the potential to support dynamic networks.
To take care of the significant overhead due to frequent
topology changes and intermittent links, the researchers utilized
some stable links and limited duplicate flooding during route
construction. This scenario is possible in UAV networks
because UAV can hover over a location and form relatively
stable links with each other. The flat structure of AODV is then
converted into hierarchical routing structure by segmenting the
network into clusters. The advantage here is that duplicate route
requests are limited to certain clusters instead of flooding the
entire network [30].
4) Hybrid routing protocols
By using Hybrid Routing Protocols (HRP), the large latency
of the initial route discovery process in reactive routing
protocols can be decreased and the overhead of control
messages in proactive routing protocols can be reduced. It is
especially suitable for large networks, and a network is divided
into a number of zones where intra-zone routing is performed
with the proactive approach while inter-zone routing is done
using the reactive approach. Hybrid routing adjusts strategy
according to network characteristics and is useful for MANETs.
However, in MANETs and UAV networks dynamic node and
link behavior makes it difficult to obtain and maintain
information. This makes adjusting routing strategies hard to
implement.
• Zone Routing Protocol (ZRP) [47] is based on the concept
of zones. Zones are formed by sets of nodes within a
predefined area. In MANETs, the largest part of the whole
traffic is directed to nearby nodes. Intra-zone routing uses
proactive approach to maintain routes. The inter-zone
routing is responsible for sending data packets to outside of
the zone. It uses reactive approach to maintain routes.
Knowledge of the routing zone topology is leveraged by
the ZRP to improve the efficiency of a globally reactive
route query/reply mechanism. The proactive maintenance
of routing zones also helps improve the quality of
discovered routes, by making them more robust to changes
in network topology.
Nodes belonging to different subnets must send their
communication to a subnet that is common to both nodes.
This may congest parts of the network. The most dominant
parameter influencing the efficiency of ZRP is the zone
radius. However, the cost of ZRP is increasing complexity,
and in the cases where ZRP performs only slightly better
than the pure protocol components, one can speculate
whether the cost of added complexity outweighs the
performance improvement
• Temporarily Ordered Routing Algorithm (TORA) [48] is a
hybrid distributed routing protocol for multi-hop networks,
in which routers only maintain information about adjacent
routers. Its aim is to limit the propagation of control
message in the highly dynamic mobile computing
environment, by minimizing the reactions to topological
changes. It builds and maintains a Directed Acyclic Graph
(DAG) from the source node to the destination. TORA
does not use a shortest path solution, and longer routes are
often used to reduce network overhead. It is preferred for
quickly finding new routes in case of broken links and for
increasing adaptability [7]. The basic, underlying algorithm
is neither distance-vector nor link-state; it is a type of a
link-reversal algorithm. The protocol builds a loop-free,
multipath routing structure that is used for forwarding
traffic to a given destination. Allows a mix of source and
the destination initiated routing simultaneously for
different destinations. TORA may produce temporary
invalid routes.
5) Geographic 2-dimension and 3-dimension protocols
A number of routing schemes have been proposed for 2-
dimensional networks that can be modeled using planar
geometry. Many prominent ones have been described in this
survey and some have been shown to have good performance
on metrics like delivery, latency and throughput in simulations.
Geographic schemes assume knowledge of geographic position
of the nodes. They assume that the source knows the
geographic position of the node and sends message to the
destination co-ordinates without route discovery. The most
common technique used in geographic routing is greedy
forwarding in which each node forwards the message to a node
closest to the destination based only on the local information. A
situation may arise when the message gets trapped in the local
minimum and progress stops. The recovery mechanism is
usually face routing which finds a path to another node, where
greedy forwarding can be resumed [130].
In many applications, it may be more appropriate to model
UAV networks in 3 dimensions. Some protocols have been
proposed which, like their 2-dimensional counterparts, use
greedy routing that attempts to deliver packet to the node
closest to the destination. But the recovery from local minima
becomes more challenging, as the faces surrounding the
network hole now expand in two dimensions, and are much
harder to capture. They differ mainly in the recovery methods
when packets get stuck in the local minima [131].
Greedy-Hull-Greedy (GHG) protocol proposed in [132]
involves routing on the hull to escape local minima. This is a 3
dimensional equivalent of routing on the face for 2-dimensional
protocols. The authors have proposed PUDT (Partial Unit
Delaunay Triangulation) protocol to divide the network space
into a number of closed sub-spaces to limit the local recovery
process.
In [133] the authors propose a Greedy-Random-Greedy
(GRG) protocol using which the message is forwarded greedily
13
until a local minimum is encountered. To come out of local
minima a randomized recovery algorithm like region-limited
random walk or random walk on the surface is used. The
authors present simulation results for following five recovery
algorithms: random walk on the dual, random walk on the
surface, random walk on the graph, bounded DFS on a spanning
tree, and bounded flooding. DFS on the spanning tree has
shown good performance for sparse networks while random
walk approaches perform well for denser networks.
GDSTR (Greedy Distributed Spanning Tree Routing)-3D
routing scheme described in [134] uses 2-hop neighbor
information during greedy forwarding to reduce the likelihood
of local minima, and aggregates 3D node coordinates using two
2D convex hulls. They have shown through simulation and
testbed experiments that GDSTR-3D is able to guarantee packet
delivery and achieve hop stretch close to 1.
Lam and Qian contend that GHG assumes a unit-ball graph
and requires accurate location information, which are both
unrealistic assumptions [131]. GRG uses random recovery,
which is inefficient and does not guarantee delivery. The
authors propose MDT (multi-hop Delaunay triangulation)
protocol, which provides guaranteed delivery for general
connectivity graphs in 3 dimensions, efficient forwarding of
packets from local minima and low routing stretch. The authors
also provide simulation results that show that MDT provides
better routing stretch compared to a number of 2-dimensional
and 3-dimensional protocols. They also show its suitability for
dynamic topologies with changes in the number of nodes and
links.
Table VI summarizes many of the routing protocols
discussed above and their applicability to the UAV
environment.
C. Routing in networks prone to delays and disruptions
According to the Delay Tolerant Networking Research
Group, the term Delay Tolerant Networking relates to extreme
and performance-challenged environments where continuous
end-to-end connectivity cannot be assumed. Initially defined for
interplanetary communication, it is now concerned with
interconnecting highly heterogeneous networks together. When
disaster strikes, and normal communication breaks down, lack
of information flow could cause delay in rescue and recovery
operations. In such situations a UAV network with delay
tolerant features is considered to be one of the most effective
communication methods [49]. In the absence of these features,
situations like intermittent links, dynamic network, node
failures and node replacements could result in breakdown of
communication. Applications in such cases must tolerate delays
beyond those introduced by conventional IP forwarding, and
these networks are referred to as delay/disruption tolerant
networks.
In many harsh situations connectivity is intermittent and
networks get partitioned for long durations. This causes
transmission delays longer than threshold limits described by
TCP protocol and packets whose destinations cannot be reached
are usually dropped, making TCP inefficient.
TABLE VI
APPLICABILITY OF PROTOCOLS TO UAV NETWORKS
Protocol type
Problems in application to UAV networks
Static
Fixed tables, not suitable for dynamic topology, does
not handle changes well, not scalable, higher possibility
of human errors
LCAD
Delivery delays
MLHR
CH becomes single point of failure, capacity issues at
CH
Data Centric
Network overload due to query-response. Problems as
in cluster based.
Proactive
Large overhead for maintaining tables up-to-date,
bandwidth constrained networks cannot use them, slow
reaction to topology changes results in delays
OLSR, GSR,
FSR
Higher overheads, routing loops
DSDV
Consumes large network bandwidth, higher overheads,
periodic updates
BABEL
Higher overheads and more bandwidth requirement due
to periodic updates
B.A.T.M.A.N.
Depends on packet loss, does not perform well if
network is reliable.
On-demand or
reactive
High latency in route finding. Source routing does not
scale well, for large network overhead may increase
because of large header size. For hop-by-hop
intermediate node must have the routing table.
DSR
Complete route address from source to destination,
scaling is a problem, dynamic network is a problem
AODV
Lower overhead at the cost of delays during route
construction. Link failure may trigger route discovery-
more delays and higher bandwidth as the size of the
network increases
Hybrid
Hard to implement for dynamic networks
ZRP
Inter zone traffic may congest. Radius is an important
factor may be difficult to maintain in UAV networks.
Complexity higher
TORA
May produce temporarily invalid results
Geographic 3D
GHG
Requires location information which may become
unrealistic in many applications
GRG
Uses random walk recovery which is inefficient and
does not guarantee delivery of messages
GDSTR-3D
Assumes static topology
MDT
None documented
'
UDP provides no reliable service and cannot “hold and
forward.” The traditional routing protocols such AODV and
OLSR also do not work properly during disruptions. In these
cases, when packets arrive and there are no end-to-end paths for
their destinations, the packets are simply dropped [50]. While
traditional solutions do not guarantee connectivity, the protocol
used in UAV networks should in some way allow buffering and
forwarding of packets. The existing protocols developed for
infrastructure based Internet are not able to handle data
transmission in such networks and new routing protocols and
algorithms should be developed to handle transmission
efficiently [57]. In terrestrial networks, with mobile nodes,
experiments have shown 80% delivery rate in the Disaster
Information System when delay tolerance features were used
[49], [52].
For UAV networks prone to intermittent links and
partitioning we would need to build in tolerance to delays and
disruptions. End-to-end reliability methods like multi-step
14
request-response, acknowledgements and timed-out
transmission will not work due to the long delays and
disconnections in these networks. To achieve tolerance to
delays and disruptions the architecture has to be based on the
“store-carry-forward” (SCF) protocol in which a node stores
and carries the data (often for extended periods) till it can
duplicate it in one or more nearby node(s). In such a case, the
node must have adequate buffering capacity to store the data
until it gets an opportunity to forward it. To be able to deliver
messages efficiently to their destinations, for each message the
best node and time to forward data should be known. If a
message cannot be delivered immediately due to network
partition, then the nodes chosen to carry the message are those
that have highest probability of successfully delivering the
message. In SCF routing the message is moved from source to
the destination one hop at a time. Selection of the path from
source to destination depends on whether the topology that
evolves over time is deterministic or probabilistic. If the nodes
know nothing about the network states then the best they can do
is to randomly forward packets to their neighbors [51], [53].
Since end-to-end paths are not available in UAV networks,
traditional routing protocols that assume end-to-end path before
communication starts fail to deliver. Consequently, many
researchers have advanced new routing algorithms such as
Direct Delivery [135], which proposes single copy method for
intermittently connected networks. Nodes are assigned
probability of delivery of a packet to the destination. They
incorporate a hybrid scheme consisting of random and utility-
based strategies. The authors claim that as compared to
Epidemic method, which essentially involves multiple copies,
the bandwidth required is less. Epidemic routing involves
multiple copies to be sent [136]. A controlled replication
algorithm called Spray and Wait has been proposed in [137].
Since UAVs are location aware due to navigational
requirements, this could be used to perform geographical
routing if the position of the destination is known. To conserve
bandwidth and to make it possible to use routing algorithms in
energy-constrained systems, viability of beacon-less
geographical routing in opportunistic intermittently connected
mobile networks has been explored. Beacons are regularly
transmitted special messages that are used by many routing
algorithms to determine a node’s neighbors. They consume
bandwidth and energy and the information may not be accurate.
To apply this to UAV networks, it must be remembered that
resources at the UAV-nodes such as storage, power and
bandwidth are limited. The delivery rate is affected by delivery
distance, number of encounters, network condition and the
node’s movements. Delivery time is also important and in this
regard it may be noticed that this architecture is not suitable for
real-time contents as delays could extend to hours and days.
Also, a route that is suitable currently may not remain stable for
long. How long it will remain stable will depend on how fast
the topology changes. If it is determined that some routes are
stable then route caching can be employed to avoid unnecessary
routing protocol exchanges [54]. Some of the schemes use
knowledge of location of the nodes, others may flood the
network with packets and yet others may use ‘carrier’ nodes to
carry data. Let us consider briefly three types of routing
methods that can be examined for UAV networks:
1. Deterministic routing
2. Stochastic routing
a. Epidemic routing based approach
b. Estimation based approach
c. Node movement control based approach
d. Coding based approach
3. Social networks based approach
According to Cardei, Liu and Wu [138], routes in networks
with sporadic node contacts consist of a sequence of time-
dependent communication opportunities, called contacts, during
which messages are transferred from the source towards the
destination. Contacts are described by capacity, direction, the
two endpoints, and temporal properties such as begin/end time,
and latency. Graphs at different points of time need to be
overlapped to find an end-to-end path. In UAV networks, the
contacts of a node keep changing over time, as their duration
over which they can communicate is limited. With this in view,
let us now see the suitability of various types of protocols for
UAV networks.
1) Deterministic Routing
In deterministic routing methods, we assume that the future
movement and links are completely known. In the context of
UAV networks this would be possible in applications where
UAVs fly in controlled formations or in applications where they
have to hover over an area. If all the hosts have global
knowledge of the availability and motion of the other hosts then
a tree approach can be used for selecting paths. A tree is built
taking source node as the root and adding child nodes and the
time associated with these nodes. A final path can be selected
from the tree by choosing the earliest time to reach the
destination node. If the characteristic profiles are initially
unknown to the hosts then they learn these by exchanging the
available profiles with the neighbors. Paths are selected based
on this partial knowledge. This method can be improved by
requiring the hosts to record the paths that past messages have
taken. The deterministic algorithms presume global knowledge
of nodes and links in space and time. Handorean et al. present
the Global Oracular Algorithm [56] where a host has full
knowledge of characteristic profile of all hosts. The authors in
[57] provide a survey of such protocols. These methods would,
therefore, not be appropriate for cases where topology of the
network changes frequently or availability of nodes and links is
not certain. The mobility of the nodes does mean that the
network topology will constantly change and that nodes
constantly come in contact with new nodes and leave the
communication range of others. If nodes can estimate likely
meeting times or meeting frequencies, we have a network with
predicted contacts [138]. In general, deterministic techniques
are based on formulating models for time-dependent graphs and
finding a space-time shortest path in delay prone networks by
converting the routing problem to classic graph theory or by
using optimization techniques for end-to-end delivery metrics.
15
Deterministic routing protocols use single-copy unicast for
messages in transit and provide good performance with less
resource usage than stochastic routing techniques. This is true
in applications where node trajectory is coordinated or can be
predicted with accuracy, as in interplanetary networking. A
multigraph is defined where vertices represent the delay tolerant
nodes and edges describe the time-varying link capacity
between nodes. Deterministic routing mechanisms are
appropriate only for scenarios where networks exhibit
predictable topologies.
2) Stochastic routing
This is the case when the network behavior is random and
unknown. In this situation it becomes important to decide where
and when to forward the packets. One possible method to route
messages in such a case is to forward them to any contacts
within range. The decision could also be based on historical
data, mobility patterns and other information. The protocols in
this category maintain a time varying network topology that is
updated whenever nodes encounter each other. For example, in
Shortest Expected Path Routing (SEPR) nodes maintain a
stochastic model of the network. Each node constructs a time
varying graph comprising nodes they have encountered and
links that reflect the connection probability between nodes. The
key limitation of SEPR is that it has poor scalability due to its
reliance on the Dijkstra’s algorithm which runs in time of the
order of O(n
2
), where n is the number of nodes. Moreover, it is
not suitable for networks with delay tolerant features and high
node mobility [58]. A routing protocol called Resource
Allocation Protocol for Intentional Delay Tolerant Networks
(RAPID) has been proposed that considers network resources
such as bandwidth and storage when optimizing a given route
metric. This is especially critical when nodes have resource
constraints [59].
In UAV networks with intermittent links and opportunistic
contacts, routing is challenging since the time when the nodes
will come in contact and for how long are not known. When the
contact does happen, it needs to be determined whether the peer
in contact is likely to take the packet closer to the destination.
The decision to hand over the packet to the contact depends on,
besides the probability of the contact taking the packet closer to
the destination, available buffer spaces in the two nodes and
relative priority to forward this packet compared to other
packets the node holds. Additionally, if the nodes have
information about their location this information can be used to
advantage. According to [138] in the situation described above,
the main objective of routing is to maximize the probability of
delivery at the destination while minimizing the end-to-end
delay. We will discuss below some of the stochastic protocols
viz. Epidemic Routing, Spray and Wait, Node movement
control and Coding based.
a) Epidemic Routing-Based Approach
This method is used in networks of mobile nodes that are
mostly disconnected [138]. This is a stochastic, flooding
protocol. Nodes make a number of copies of messages and
forward them to other nodes called relays. The relays transfer
the messages to other nodes when they come in contact with
them. No prediction is made regarding the link or path
forwarding probability. In this way, messages are quickly
distributed through the connected portions of the network.
However, upon contact with each other nodes exchange only
the data they do not have in their memory buffer. Hence it is
robust to node or network failure, guaranteeing that upon
sufficient number of random exchanges, all nodes will
eventually receive all messages in minimum amount of time
[54]. The main issue with epidemic routing is that messages are
flooded in the whole network to reach just one destination. This
creates contention for buffer space and transmission time
Epidemic routing uses node movement to spread messages
during contacts. With large buffers, long contacts or a low
network load, epidemic routing is very effective and provides
minimum delay and high success rate, as messages reach the
destination on multiple paths. When no information is available
about the movement of nodes, packets received by a node are
forwarded to all or some of the node’s neighbors except from
where it came.
This approach does not require information about the
network connectivity. However, it requires large amounts of
buffer space, bandwidth and power. Spray and wait is a
variation in which each message has a fixed number of copies
[137]. With this approach, the source node begins with L copies
for each message. When a source or relay node with more than
one copies comes in contact with another node with no copies it
may transfer all n copies in normal mode or n/2 copies in binary
mode (Spray phase). So after the contact both nodes have n/2
copies. With one copy a direct contact with destination is
required (Wait phase). A message will be physically stored and
transmitted just once even when a transfer may virtually
involve multiple copies.
Approaches using indiscriminate or controlled flooding use
varying amount of buffer space, bandwidth and power. There
are variations like Prophet and MaxProp protocols, in which
data is arranged in order of priority based on some criteria.
However, these methods deliver good performance in a regular
movement case. MaxProp protocol puts a priority order on the
queue of packets. It determines the messages to be transmitted
or dropped first. The priorities are based on the path likelihoods
to peers [60] [61]. However, our airborne network topology
may change without any known pattern. Also, even the random
walk model may not describe movement of nodes due to
horizontal as well as vertical motions. In such cases a three-
dimensional architecture may better define the randomly violent
topological changes and a usable path can be generated in an
unpredictable moment.
In ant colony based routing strategy [140], the authors have
proposed an exploration-exploitation model for UAV networks
that is based on ant foraging behavior. Exploration is defined as
the ability to explore diversified routes while Exploitation is the
process to focus on a promising group of solutions. In networks
with frequent partitioning new routes need to be explored and
every contact opportunity is used to forward the messages. The
authors have shown that performance comes close to Epidemic
Routing (which provides an upper bound) in terms of delivery.
16
b) Estimation Based Approach
Instead of forwarding messages to neighbors
indiscriminately, intermediate nodes estimate the probability of
each outgoing link eventually reaching the destination. Based
on this estimation, the intermediate nodes decide whether to
store the packet and wait for a better chance, or decide when
and to which node to forward. Variations in this could be
decisions based on estimation of the next hop forwarding
probability or decision based on average end-to-end metrics
such as shortest path or delay [57].
A representative routing protocol for networks with delay
tolerant feature is one that uses delivery estimation. PROPHET,
(Probabilistic Routing Protocol using History of Encounters and
Transitivity), in [141] is such a protocol. PROPHET works on
the realistic premise that node mobility is not truly random. The
authors assume that nodes in a DTN tend to visit some locations
more often than others and that node pairs that have had
repeated contacts in the past are more likely to have contacts in
the future.
c) Node-Movement-Control Based Approaches
Nodes could either wait for reconnection opportunity with
another node passively or seek another node proactively. In the
reactive case, where the node waits for reconnection, there may
be long unacceptable transmission delays for some applications.
In the proactive case, a number of approaches have been
proposed to control node mobility for reducing delays. In one
class of methods, trajectories of some nodes are altered to
improve some system performance metrics like delay [62]. In
message ferrying approach special ferry nodes carry data either
on preplanned routes with other nodes coming close to a ferry
and communicating with it or ferry nodes move randomly and
other nodes send a service request as a response to which the
chosen ferry node will come close to the requesting node [63].
Using DataMules, nodes that move randomly, have large
storage capacity and renewable energy source, is another
method in this category. According to [138], controlled ferrying
can be used in UAV networks in disaster recovery where UAVs
can be equipped with communication devices capable of storing
a large number of messages and can be commanded to follow a
trajectory that interconnects disconnected partitions.
d) Coding Based Approaches
The concept of network coding comes from information
theory and can be applied in routing to further improve system
throughput. Instead of simply forwarding packets, intermediate
nodes combine some of the packets received so far and send
them out as a new single packet to maximize the information
flow [57]. Erasure coding involves more processing and hence
requires more power but improve the worst-case delay [142].
The basic idea of erasure coding is to encode an original
message into a large number of coding blocks. Suppose the
original message contains k blocks. Using erasure coding, the
message is encoded into n (n > k) blocks such that if k or more
of the n blocks are received, the original message can be
successfully decoded. In [143], the authors have shown that
network coding performs better than probabilistic forwarding in
disruption prone networks. Erasure coding includes redundant
data and is useful where retransmission is impractical. One
aspect to be kept in mind is that in UAV retransmission would
be a different path with possibly different set of nodes and also
storage needed for redundant information is limited.
Hybrid Erasure Coding combines erasure coding and
aggressive forwarding mechanism to send all packets in
sequence during node contact. If the node battery has been
drained or node loses mobility due to fault, it cannot deliver
data to the destination creating a ‘black hole’ information loss
problem. HEC solves the black hole problem by using the
nodes’ contact period efficiently.
3) Social Networks Based Routing
When mobility of nodes is not totally random and the nodes
are likely to be present with greater probability in some known
places then the random mobility model, used in a large number
of protocols, is not realistic. The nodes that visit these places
have a contact probability and will have more success in
delivering bundles. Notion of groups, communities and
popularity of nodes can also be exploited [31]. However,
protocols may overuse the popular nodes and performance may
degrade because of limited capacity of these nodes. The
expected delay may also increase because of contention.
Integration of some level of randomness may benefit
performance. This is relatively a new area and more work needs
to be done before it can be appropriately exploited [64]. Some
applications of UAV networks, e.g., providing communication
over a disaster prone area or communication over an oil drilling
platform, may provide an environment where this type of
routing can be applied.
Table VII gives DTN routing protocols and considerations
for their application in UAV networks.
IV. SEAMLESS HANDOVER
The UAV mesh nodes may be stationed over a disaster struck
area to provide communication services across the target area
and form a network with slow dynamic links. On the other
hand, in applications like crop survey, which require a
sweeping coverage of an area, UAVs may move around at
required speeds. During a prolonged mission UAVs may
periodically go out of service as they go out of power or
develop faults. Their communication interfaces may also be
shut down to conserve power, or one or more of the UAVs may
be withdrawn, when less dense network is required. In all these
cases the network needs to reconfigure and the ongoing voice,
video or data sessions are required to be handed over to one of
the working UAVs according to some predefined criteria.
Handover allows for total continuity of network communication
with only a minor increase of message latency during the
handover process [65]. Subsection A elaborates the types of
handoffs in UAV networks. Applicability of existing handoff
schemes and new developments that can be used in UAV
networks are in Subsection B. Subsection C discusses the IEEE
standard media independent handover.
17
TABLE VII
DELAY/DISRUPTION PRONE UAV NETWORKS
DTN Routing
Algorithm
Considerations for application in UAV Networks
Deterministic
Useful when future availability and location of the
nodes is known. In some UAV networks topology
may change frequently and availability of nodes and
links may not be certain.
Stochastic
Epidemic based
Requires large buffer space per node, bandwidth and
power. For UAV networks it must be noted that
message delivery time depends on the buffer size.
Comparatively delay is lower but with high-energy
expenditure.
Estimation/Probability/
Statistical
Random methods work well for small networks but
for large networks estimation result in large
overheads. Maintains encounter but no location
information. Changes in topology, like in UAV
networks, affect convergence time. Delays are
moderate at moderate energy consumption.
Complexity is also moderate. Changes in topology
affect convergence time.
Node movement based
UAVs can be made to follow a given trajectory that
will connect source and destination nodes in
partitioned networks.
Message ferrying
Location information is maintained. Large storage
space required in ferries. Delays are high but energy
expenditure is low. Can work with heterogeneous
nodes.
Coding-based
Builds in redundant information so that retransmission
is avoided. This aspect may be exploited in UAV
networks. In UAVs retransmission would require
finding a new path as disruptions are the norm.
However, maintaining redundancy and aggressive
forwarding takes additional bandwidth and buffer
space. May provide better delivery rates than
probabilistic in some settings but is inefficient if
connectivity is good.
Social Networks
Applicable when some UAV nodes have ‘more likely’
locations. Such nodes must have higher buffer size
and higher bandwidth links to avoid contention delays
and losses
A. Handoffs in UAV networks
The real advantage of wireless mesh networks become
apparent when self-organization is coupled with seamless
handover to provide continuity of service to the users.
Handover, or handoff as it is commonly called, is common in
cellular networks, where mobile stations frequently move out of
the coverage area of one cell tower and into that of a
neighboring tower. Handovers can be hard or soft. In a hard or
standard handover, the connection from the old network is
broken before it is made with the new network. This would
interrupt all the user sessions currently in progress at the mobile
node (Figure 5a).
In case of soft or seamless handover, connection is made
with the new network before breaking the connections from the
old network. The original user sessions at the mobile station are
maintained till the new link is up and handover action migrates
the session to the new link as shown in Figure 5b.
'
'
Fig. 5: a) Hard Handover b) Soft Handover
Besides being hard or soft, handovers can also be classified
as horizontal and vertical handovers. Horizontal handovers are
intra-system where the mobile access device moves from one
access point to another in the same network. In the case of
vertical handover, the transfer of connection is between two
networks of different technologies. Figure 6 depicts these two
types of handovers pictorially.
Fig. 6: Horizontal and Vertical Handovers
The handovers can also be classified based on whether the
mobile access device is controlling or assisting in the handover.
If both the mobile device and the network are involved then we
have a hybrid handover. These have been studied in the context
of VANETs and mobile IP networks but not much work is
available for UAV networks [66].
B. Applicability of Existing Handover Schemes
Lack of methods for seamless handover in UAVs, make
people look at the work done in the areas of MANETS and
VANETs. However, despite the need to provide seamless
handover in VANETs, there are few studies regarding mobility
protocols and there are no practical studies on mobility
protocols using IEEE Wireless Access in Vehicular
Environments (WAVE) communication [67]. WAVE consists
of IEEE 802.11p and IEEE 1609.x to provide connectivity
under the hostile operation conditions of VANET. However,
Affected Area
Affected Area
Out of service
Old links to out
of service plane
broken
(a)
Affected Area
Affected Area
Out of service
New links
made before
breaking old
ones
Old links to out
of service UAV
removed
Out of service
(b)
Horizontal Handover
Vertical Handover
18
these standards do not address the network mobility issues,
either inter- or intra-technology. VANETs are characterized by
the high mobility and speed of nodes, resulting in short
communication time, frequent changes in network topology and
network partitioning. As VANETs are a special type of
MANETs, routing Protocols and IEEE standards used in
MANETs have also been considered for the VANET
environment. The literature survey available on vehicular
communication is very limited and adaptation of the work on
mobile ad hoc networks [68].
The random waypoint (RWP) model, in which mobile node
movement is considered random, is commonly employed in
study of MANETs. Due to the highly dynamic characteristics of
VANETs, network partition or merging can occur frequently,
which results in the unavailability of existing route or
availability of better routes. This causes the networks to
reconfigure and may trigger handovers. The handover latency
and the packet loss during handover process may cause serious
degradation of system performance and QoS perceived by the
users. RWP model is, therefore, considered to be a very poor
approximation of vehicular mobility. A more accurate and
realistic vehicular mobility description at both macroscopic and
microscopic levels is required [69]. The selected model should
take care of temporal and spatial dependencies and
geographical restrictions that the nodes would be subjected to.
Two models that have been used are street random waypoint
(STRAW) and Manhattan. Furthermore, the scale of VANET
may vary in a large range and the node density in the network
may vary dynamically depending on various application
scenarios. These are important issues and more studies are
required on designing handover schemes for VANET
application with a view to decrease handovers and improve
packet delivery and latency.
The above-mentioned considerations apply to mobile UAV
networks as well. The problem is that there are only limited
studies on efficient seamless handover even in IEEE 802.11-
based WMNs or VANETs [67]. Many traditional mobility
management protocols, such as Mobile IP version 4 (MIPv4)
and Mobile IP version 6 (MIPv6) have been applied to provide
handover support for VANETs and schemes for performance
enhancement have been proposed. In [150] the authors propose
Vehicular IP in WAVE framework along with a mobility
management scheme supported by Proxy Mobile IPv6 (PIMP)
over WAVE for seamless communication. They propose and
analytical model that demonstrates better handover performance
of VIP-WAVE as compared to standard WAVE. In order to
improve MIPv6 to support the real time handover, Hierarchical
Mobile IPv6 (HMIPv6) and Fast Mobile IPv6 (FMIPv6) were
proposed. Inter base-station (BS) transfers of mobile devices
trigger excessive signaling and reduce throughput. To counter
this the Internet Engineering Task Force (IETF) proposed
NEMO (network mobility). Although NEMO provides good
mobility management, it does not work well in vehicular
network environment [70]. As handover layer 2 and layer 3
intervention, cross layer designs that reduce packet loss (L2
function) as well as delay (L3 function) have been proposed.
In an evaluation of MIPv6 and Proxy Mobile IPv6 (PMIPv6),
extended to enable enhanced support for multiple technologies,
and integrated with a proposed mobility manager that monitors
and selects the best access technology and network, the results
show that overall the approach based on PMIPv6 performs
better than the one based on MIPv6, especially when using
IEEE 802.11p. Also, the former is capable of performing
handover between two IEEE 802.11p networks without any
data loss, even at moderate speeds [73]. The major problem in
MIPv6 is the handover latency, which is the time required by
vehicular node to become ready for sending or receiving
packets post handover [74]. Another issue that is important is
that MIPv4, MIPv6 and HMIPv6 are designed to handle
terminal mobility, not for network mobility. NEMO protocol
was proposed in 2005. The goal of the network mobility
(NEMO) management is to effectively reduce the complexity of
handover procedure and keep mobile devices connected to the
Internet. The handover procedure is executed on the mobile
device and mobility may cause real time services like mobile
TV or voice over IP to get disconnected. Many mobility
management protocols have been proposed but high degree of
mobility in VANETs forces frequent handover causing
problems in communication. A more rigorous evaluation of
performance under various network topology and applications
is required [66], [70].
In case of UAVs seamless handover procedure would involve
exchange of a sequence of messages between the user mobile
device and the UAV node that is being taken off service and the
UAV node to which this user’s session would be transferred.
The handover procedure results in a transfer of physical layer
connectivity and state information from one UAV to another
with respect to the mobile unit in consideration. The theoretical
framework, similar to that in mobile wireless networks,
involves network discovery, trigger and execution. In the
network discovery phase, the mobile station discovers several
wireless networks on assigned channels and creates a list of
APs prioritized by the received signal strength. The station may
listen passively for broadcast service advertisements or actively
send probes. The decision to trigger transfer is taken based on
multiple criteria like signal to noise ratio. In the execution
phase the actual transfer of the sessions to the new access point
takes place. All contextual information related to the mobile
user is transferred to the new UAV. Table VIII gives important
considerations for various protocols.
C. Media Independent Handover
IEEE 802 initially did not support handover between different
types of networks. They also did not provide triggers to
accelerate mobile IP based handovers. IEEE has now
standardized Media independent handover (MIH) services
through their standard IEEE 802.21 [73]. The key function of
MIH, known as the Media Independent Handover Function
(MIHF) is between the layer 2 wireless technologies and IP at
layer 3 [75]. These services can be used for handovers and
interoperability between IEEE-802 and non-IEEE-802
networks, e.g., cellular, 3GPP, 4G. The networks could be of
19
TABLE VIII
PROTOCOLS FOR SEAMLESS HANDOVER
Protocol
Issues to be considered for UAV networks
MIPv4
IP address shortage, weak security mechanism, auto
configuration of IP [67]
MIPv6
Has high handover latency due to signaling overhead,
packet loss, not scalable, not efficient [71].
PMIPv4
Improves latency but does not ensure seamless handover.
Performs better than MIPv6, when using IEEE 802.11p.
As proxy handles signaling the mobile nodes do not need
any modifications, signaling overheads lesser than
MIPv4 [67].
HMIPv6
Introduced to reduce signaling between nodes and other
equipment. It can also reduce handover latency due to
smaller signaling and shorter path [71]
FMIPv6
Improvement over MIPv6. Relies on predictive method
that is difficult to make accurate for fast and randomly
moving mobile nodes. It is not suitable for real time
services in fast moving vehicles [71].
NEMO
IETF NEMO BS based on MIPv6. It is more effective
than terminal mobility. It is by itself not sufficient for
seamless handover, optimization is necessary. Route
optimization may not be done due to security and
incompatibility issues. Latency of link layer handover
and NEMO signaling overhead affect the overall
performance of mobility management significantly [72].
the same or different media type, wired or wireless. This
standard provides link-layer intelligence and other related
network information to upper layers to optimize handovers
between heterogeneous networks. It consists of signaling and
triggers and makes available information from MAC/PHY to
network and application layers. The standard is a hybrid
implementation as it supports cooperative use of information
available with the mobile station and the network. The mobile
station has information about signal to noise ratios of available
networks and the network knows about the information about
access points, mobile stations and the higher layer service
availability. The handover process can be initiated by
measurement reports and triggers supplied by the link layers on
the mobile station. Intra-technology handover, handover policy,
security, enhancement specific to particular link layer
technologies, higher layer (3 and above) enhancements are not
part of this standard. MIH, however, does not provide intra-
technology handover, handover policies, security and
enhancements to link layer technologies [76]. Those are
handled by the respective link layer technologies themselves.
In case of emergency response systems there are many
jurisdictions involved. This causes inter-operability issues
among jurisdictions and brings in inefficiencies in the
communications. What is required is an architecture that
provides a common networking platform for heterogeneous
multi-operator networks, for interoperation in case of cross
jurisdiction emergencies. Media Independent Handover offers a
single unified interface to higher layers through technology
independent primitives. It obtains information from lower
layers through media specific interfaces. This allows UAV
networks to interact with other technologies making it possible
to use them in an integrated manner [144].
In applications like search and rescue or real-time assistance
to troops in the field, using MIH, UAV networks can bring live
video feeds in conjunction with other networks, viz., WiFi,
WiMAX, LTE, etc. However, MIH is a nascent technology that
has not been widely deployed and evaluated. In order for this to
work, both mobile devices and the network must implement the
standard [145].
SDN based mechanisms using OpenFlow protocols have
been proposed for configuring network nodes and establishing
communication paths. IEEE 802.21 Media Independent
Handover procedures have been used to optimize handovers in
heterogeneous wireless environment. It has also been shown
how SDN can be used to support wireless communication paths
with dynamic mobility management capabilities provided by
standards compliant nodes. The combination allows enhanced
connectivity scenarios; allowing ‘always best’ connectivity
when multiple handover candidates are available and OpenFlow
procedures are triggered preemptively in order to avoid traffic
disruption [77].
V. ENERGY EFFICIENCY IN UAV NETWORKS
There could be two scenarios in UAV networks. First, energy
for communication equipment as well as for powering the UAV
comes from the same source or, alternatively, they could be
from different sources. In either case, energy consumption by
the communication equipment is substantial and can limit the
useful flying time and, possibly, the life of the network. The
important point to note here is that there is a large consumption
even when there is no transmission or reception, i.e., when the
wireless interface is idle. Power rating of a typical Wifi 802.11n
interface, in non-MIMO single antennal mode, is
1280mA/940mA/820mA/100mA under
Transmission/Reception/idle/sleep modes, respectively [151]. A
typical small drone may have battery capacity of 5200mAh,
11.1V. Such a drone draws about 12.5A and gives a flying time
of about 25 minutes. Communication equipment of the UAV in
a mesh network normally receives and transmits continuously.
A quick calculation would show that the flight time would be
reduced by 16%, of the rated value, if communication
equipment uses the same battery as the UAV. Together with
GPS and a couple of sensors, it can easily go beyond 20%. In
practice, however, the net effect of this could be more severe as
the battery voltage drops down below l1.1V even before the
power is fully drained, inhibiting normal function of the UAV.
Let us now consider the scenario where the communication
equipment has its own battery, separate from the UAV battery.
In this case, the battery weight needs to be taken into account in
the limited payload capacity of the UAV. In a series of
experiments, the authors of this paper used separate AAA
batteries to power up the airborne OpenMesh router. The router
required 8 AAA batteries to provide sufficient voltage to
operate. Typical alkaline batteries weigh about 11.5g each and
have capacity of 860mAh. [146]. If the flight is of 25 minute
duration, with continuous transmission and reception, the
consumption would be 740mAh. Eight cells would theoretically
work for about 9 hours but, as the voltage drops, the router
would stop functioning and has to be brought down for change
or recharge. In our experiments, fully charged alkaline cells
20
provided good enough voltage for about 8 hours, enough for as
many as 19 sorties. Considering the weight of the battery was
another matter. Eight cells weigh 92g. The dead weight of the
UAV used is about 1kg and it could carry about 300 grams of
payload. The batteries constituted about 30% of the payload
that could be put on the UAV! Reducing the energy
consumption, therefore, would result in increase in network
lifetime or increase in useful payload that can be carried.
Controlling transmission power of nodes, or making some
nodes go to sleep, when network can operate without them can
reduce power consumption. Nodes, that are being actively used,
are carrying traffic or helping proliferate updates required by
the routing protocol. The nodes that are powered up but not
carrying traffic, or carrying low traffic, are still consuming
energy. It is possible to avoid this waste of energy by switching
off unused network elements dynamically. We shall see a
number of ways this can be achieved. It must, however be
mentioned that it is not just at the physical layer that the devices
can be designed to save energy but also at the data link and
network layers [78]. At the physical layer the built in power
save modes of the devices may be exploited. At the data link
layer avoiding collisions and extending sleep time in power
save mode (PSM) would be important. At the network layer
routing for minimizing power consumption may be the criteria.
The concerns regarding power saving in the UAV networks are
in some ways similar to that in mobile ad-hoc networks and
wireless sensor networks. There have been a few research
efforts to adapt some of the existing energy-aware protocols to
the UAV networks. There are others that can potentially be
used in UAV applications but are still to be tried. While there
could be different ways to classify them, a commonly followed
approach is to broadly classify them on the basis of the protocol
layer they operate in. Within that broad classification, we also
classify them on the basis of energy saving strategy used. The
area is still developing therefore, rather than being exhaustive,
we would aim at including the representative protocols that
have been tried in some for in UAV networks or are potentially
good candidates. To this end we also present some cross-layer
considerations for improving energy efficiency.
It would be pertinent to mention here that the energy
harvesting or scavenging alternative to power saving has been
studied in recent times. Energy can be harvested from kinetic,
thermal and electromagnetic sources. Creation of two queues,
one for random energy arrivals and the other for data arrivals
requires redesigning of transmission algorithms for wireless
systems including UAV networks. This is in itself an evolving,
specialized area and has not been made part of this survey.
The remaining section is divided as follows: Subsection A
discusses energy efficient network layer protocols that can be
used in UAV networks. Subsection B deals with data link layer
protocols for conserving energy while the physical layer
mechanisms are given in Subsection C.
A. Energy Conservation in the Network Layer
Energy efficient routing is important for networks with
energy constraints. Energy efficient protocols typically use one
of the standard network layer protocols as the underlying
routing protocol. Many of the MANET protocols have been
tried but overall the simulation results generalized in favor of
one or the other. It is not clear which class of protocols would
work best. Applicability of most of these to different UAV
network scenarios is still to be proven. We will look here at
many of the protocols that have been tried in UAV networks, as
well as the potential candidates.
There are some well-known ways of building in energy
conservation functionality into UAV networks. Measuring and
controlling power utilization or distributing load fairly based on
some energy metric are the two important ones. Additionally,
some routing protocols may make some of the nodes to sleep,
or navigate a low energy path, to avoid waste of energy. The
protocols discussed are further classified into the following four
categories: 1) Path selection based 2) Node selection based 3)
Coordinator based 4) Sleep based. Various proposals for these
four categories are described below and summarized at the end
of this subsection.
1) Path selection based
These protocols aim to select paths that minimize total
source to destination energy requirement. Four protocols in this
category are discussed below.
a) EMM-DSR (Extended Max-Min Dynamic Source Routing)
Protocol: This mechanism maximizes energy efficiency by
finding the shortest path based on energy. It maintains a good
end-to-end delay and throughput performance. It extends the
Max-Min algorithm to maximize throughput, minimize delay
and maximize energy efficiency. This extension has been
applied to the existing on-demand dynamic source routing
protocol (DSR) in the context of mobile ad hoc networks, and
the resultant version takes the name of EMM-DSR [80].
However, in [55], the performance based on end-to-end latency
of DSR was worst for UAV networks as compared to AODV
and directional OLSR.
b) FAR (Flow Augmentation Routing): FAR is a transmission
power optimization protocol. It assumes a static network and
finds the optimal routing path for a given source–destination
pair that minimizes the sum of link costs along the path. The
cost depends on cost of a unit flow transmission over the link,
initial and residual energy at the transmitting node. The flow
augmentation algorithm requires frequent route computations
and transitions but it selects the shortest cost route each time.
As pseudo-stability of the topology can only be assumed in a
small subset of UAV network applications (e.g. communication
coverage of a remote area), there will be heavy penalty in terms
of computation of minimum cost link paths when the nodes
change their relative positions frequently [81].
c) The Minimum-energy Routing: Minimum-energy routing
saves power by choosing paths through a multi-hop ad hoc
network that minimize the total transmit energy. Distributing
energy consumption fairly maximizes the network lifetime. In
this protocol, nodes adjust their transmission power levels and
select routes to optimize performance. This takes many forms in
the UAV networks. Topology may be selected by adjusting the
power such that only immediate neighbors communicate. It is
possible to set this up in UAV networks except that the
neighbors may change more frequently. The work on multi-hop
21
communication is a swarm of UAVs [147] has reported use of
minimum-energy expenditure multi-hop routing similar to
ExOR (Extremely Opportunistic Routing). High power hops are
split into smaller low power hops. Edge weights represent
attenuation in the network and Dijkstra’s algorithm is used to
calculate the shortest (lowest energy) path. ExOR broadcasts
each packet, choosing a receiver to forward only after learning
the set of nodes that actually received the packet. ExOR
operates on batches of packets in order to reduce the
communication cost of agreement. The source includes in each
packet a list of candidate forwarders prioritized by closeness to
the destination. Highest priority node forwards the batch.
Remaining forwarders forward in order but only packets that
have not been acknowledged.
d) The Pulse protocol: A pulse, referred to as flood, is
periodically sent at fixed interval originating from infrastructure
access nodes and propagating through entire component of the
network. This pulse updates each node about the nearest pulse
source and each node tracks best route to the nearest pulse
source based on some metric. The propagation of the flood
forms a loop free routing tree rooted at the pulse source. If a
node needs to send a packet it responds to the pulse with a
reservation packet. This protocol could suffer from flood
overlap delays and can result in significant consumption of
energy [82]. This has been proposed in the context of multi-hop
fixed infrastructure wireless networks. By analogy it can be
examined for the infrastructure category of UAV applications.
2) Node Selection based
These protocols aim to select nodes that preserve battery life
of each node or exclude nodes with low energy. Three such
protocols are discussed below.
a) Power-Aware Routing: These protocols may minimize
total power requirement for end-to-end communication or
preserve battery life of each node. MTPR (Minimum Total
Transmission Power Routing) and MBCR (Minimum Battery
Cost Routing) are respectively based on these principles. In
MTPR, adaptation of transmission power could result in a new
hidden terminal problem. If this happens then higher number of
collisions would result in more transmissions resulting in higher
energy consumption. MBCR only considers the total cost
function and may include a node with very less energy if all
other nodes included have high energy. The variation, MMBCR
(Min-Max Battery Cost Routing) takes into account the
remaining energy level of individual nodes instead of the total
energy. The downside is that in the quest to minimize total
energy it can choose a long path and result in excessive delays
[90], [118]. A variation has been proposed for wireless
networks with renewable sources of energy where the node has
full knowledge of the energy it will have until the next renewal
point. The proposed algorithm is shown to be asymptotically
optimal. The proposed routing algorithm uses a composite cost
metric that includes power for transmission and reception,
replenishment rate, and residual energy [83].
b) LEAR (Localized Energy-Aware Routing): LEAR uses
DSR but allows a node to determine whether to forward the
RREQ or not depending on the residual battery power being
above a decided threshold value. If the residual power is below
the threshold value, the node drops the message and does not
participate in relaying packets. The destination node would
receive a message only when all intermediate nodes along a
route have good battery levels. If the source node does not
receive a reply then it resends the message. Intermediate nodes
lower the threshold to allow forwarding to continue [84].
c) DEAR (Distributed Energy-Efficient Ad Hoc Routing)
protocol: Conventional power aware routing algorithms may
require additional control packets for gathering information
such as network topology and residual power. While it is easy
to obtain such information in proactive routing, in case of on-
demand protocols separate control packets may be required.
DEAR uses already available RREQ packets to acquire
necessary information and no extra packets would be required.
DEAR only requires the average residual battery level of the
entire network. Nodes with relatively larger battery energy will
re-broadcast RREQ packets earlier. On-demand routing
protocols drop duplicate RREQ without re-broadcasting them.
DEAR can set up route composed of the nodes with relatively
high battery power. Simulation shows that DEAR prolongs the
network lifetime and also improves the delivery ratio by
selecting a more reliable path [85].
3) Cluster head or coordinator based
In these protocols the predominant property is selection of a
cluster head or a coordinator that will remain awake while the
other nodes can sleep to conserve power. These protocols are
quite suitable for sensor networks. They could be a natural
choice in multi-level hierarchical UAV networks and in
situations where multiple UAV networks need to communicate.
Some examples of such protocols follow:
a) CBRP (Cluster Based Routing Protocol): It is an on-
demand routing protocol in which nodes are divided into a
number of 2-hop diameter clusters. Each cluster has a cluster
head that acts as a coordinator that communicates with other
cluster heads. All nodes of a cluster communicate through the
coordinator. The protocol efficiently minimizes the flooding
traffic during route discovery and speeds up this process as well
resulting in significant energy savings [86].
b) Distributed Gateway Selection [122]: In this method, some
superior nodes in the UAV network are selected as gateways
whereas other nodes connect to the command center through
these gateways. As against selection of cluster heads, selection
of gateways takes into account movement of nodes. They
propose an adaptive gateway selection algorithm based on
dynamic network partition. The selection process goes through
several iterations. The selection is based on stability of two-hop
neighbors.
c) GAF (Geographic Adaptive Fidelity) protocol: In GAF
complete network topology is divided into a grid of fixed sized
squares. Nodes in two adjacent squares can communicate with
each other. In each square the node with the highest residual
energy is selected as the coordinator or master. Only one node
in each square needs to be aware of the geographical location of
other nodes in the area and this node may become the
coordinator. The coordinator keeps on changing within the area.
Nodes other than the coordinator can sleep without affecting
22
routing fidelity. The protocol increases the lifetime of the
network. It is a location based hierarchical protocol in which
the coordinator does no aggregation [87]. In UAV networks
nodes would not be stable in a grid. The connected set made out
of the high energy UAVs in each square cannot persist as nodes
move in an out. Frequent change of candidates for connected
set in each square and presence of low energy nodes in some of
the squares may lead to partitioning of the network.
d) Span protocol: Span reduces energy consumption without
reducing connectivity or capacity. It works between the
network and MAC layers and exploits the power saving feature
of the MAC layer, i.e., make devices sleep when no data is to
be transmitted. Each node takes a local decision whether or not
to be a coordinator for when to wake up or sleep. Coordinators
stay awake all the time and perform multi-hop routing while
other nodes can be in power save mode. Each node gets the
capability of selecting neighbors and assists in altering the
network topology. The number of links is reduced and the
nodes still get the advantage of using good QoS links through
other nodes. In UAV networks the change of coordinators
forming the backbone network would depend on the changing
position of UAVs. If a number of UAVs try to become
coordinators then all but one have to back off. During backoff
their position may change and some may not remain candidates
for forming the backbone. The amount of energy saved depends
on the node density. System lifetime of an 802.11 network in
power saving mode with Span is a factor of 2 better than
without it [88].
4) Sleep based: These protocols conserve energy mainly by
making as many nodes sleep for as long as possible. Three such
protocols are described below.
a) CDS (Connected Dominating Set) protocol: A connected
dominated node is a set of nodes such that every other node in
the network is a neighbor of one of these nodes. If the members
of the CDS set are connected then all other nodes will receive
the packets. While dominating set is always active, other nodes
can sleep to save energy [89]. With the help of CDS routing is
easy and can adapt to topology changes. This protocol allows
saving energy by selecting CDS nodes on energy metrics and
also allowing non-CDS nodes to sleep.
b) CaDet (Clustering and Decision-Tree Based) protocol:
This protocol uses data-mining techniques and analysis of real
wireless data by probabilistic location estimation and the
multiple-decision tree-based approach. These would help in
identifying the most used access points that will in turn give
information about the location of the users. These decisions
help in selecting minimum number of access points to be used,
thus reducing the power consumption and wake-up time of the
client [76]. Shi and Luo [148] propose a cluster based routing
protocol for UAV fleet networks. One of the criteria for
choosing cluster head is residual energy. The protocol forms
stable cluster architecture of UAV fleet as the basis and then
performs route discovery and route maintenance by using the
geographic location of UAVs.
c) EAR (Energy Aware Routing) protocol: In the least hops
methods, with no energy considerations, the shortest path is
always used and nodes on this path get depleted leading to
partitioning of the network. In EAR routing and traffic
engineering decisions are made taking the energy consumption
of equipment into account even at the cost of sub-optimal paths.
It may actually lead to multiple sub-optimal low energy paths.
Energy cost can be used as the only objective or in combination
with other constraints like QoS. The aim is to completely
switch-off components that will help to reduce energy
consumption of the network [90]. A way to achieve this by
synchronizing transmit and receive times at nodes or by
minimizing the necessary total received and transmit power
over a route. In such a scenario, a route with more, but shorter,
hops can be more energy efficient. In many UAV networks the
nodes are continuously transmitting and receiving and
achieving this synchronization would not be difficult [149].
Important features and energy saving strategy of network
layer protocols are summarized in Table IX
TABLE IX
ENERGY CONSERVATION TECHNIQUES IN NETWORK LAYER
Protocol
Features
Energy Saving
Strategy
1. Path selection based
EMM-DSR
Maximizes energy efficiency and
minimizes path length. Maintains
good end-to-end delay and throughput
performance.
Least energy path
FAR
Finds minimum cost path based on
initial, residual and flow of energy.
Transmit power control
Min Energy
Selects routes that minimize the total
transmitted energy. Distributing
energy consumption fairly can
maximize the network lifetime.
Transmit power control
Pulse
Nodes connected to infrastructure
send pulse flood to all the nodes.
Nodes only remember best route
pulse source with the lowest metric.
Least energy path
2. Node selection based
Power
Aware
Composite cost metric includes
transmit and receive power, residual
energy and replenishment rate. Good
for renewable energy sources.
Optimize end-to-end
power
LEAR
A route will be selected if nodes with
high power available from source to
destination for sending RREQ
Conserve energy of
low residual energy
nodes
DEAR
Only requires the average residual
battery level of the entire network.
Nodes with relatively larger battery
energy will re-broadcast RREQ
packets earlier. Prolongs network life.
Use high residual
battery life routes
3. Cluster-head or Coordinator based
CBRP
Each cluster head communicates with
other cluster heads. Nodes maintain
neighbor table. The protocol
efficiently minimizes the flooding
traffic and speeds up this process,
reduces energy consumption.
Cluster-based,
minimize flooding
traffic
Distributed
Gateway
Selection
Gateways are like cluster heads.
Selected based on stability
Only gateways
communicate with
control centers.
SPAN
High-energy nodes and those that add
to network connectivity become
coordinators. Depends on node
density for energy savings.
Use only high
energy nodes
GAF
Zoning of network. One coordinator
per zone. Disadvantage is that all
Coordinator stays
awake
23
nodes should know their geographical
positions.
4. Sleep based
CDS
Important nodes form connected set
and stay active. All nodes are either
part of CDS or one hop away.
Selected nodes stay
awake
CaDet
It is probabilistic location estimation
and multiple-decision tree based
approach. Selects minimum number
of APs to be used.
Keep nodes awake
based on client
density
EAR
Energy efficient routing decisions
even at the cost of sub-optimal paths.
Other constraints like QoS may be
used.
Low energy paths
are chosen
B. Energy Conservation in the Data Link Layer
We now discuss the approaches that help conserve energy in
the data link layer. It is the responsibility of the MAC sublayer
of the data link Layer to ensure a fair mechanism to share
access to the medium among nodes. In this process, it can play
a key role in the maximization of node’s energy efficiency.
Power-saving mode (PSM) has been defined in 802.11, which
reduces energy consumption of mobile devices by putting them
in sleep mode when idle. However, a device in PSM has to be
woken up to send or receive any packets. Power saving thus
comes at the cost of delivery delay. IEEE 802.11e defines
Automatic Power Save Delivery (APSD) [91]. IEEE 802.11n
has introduced further enhancements to the APSD protocols,
referred to as power save multi-poll (PSMP) [92]. As in its
predecessors, there are scheduled (i.e., S-PSMP) and
unscheduled (i.e., U-PSMP) versions. S-PSMP provides tighter
control over the AP/station timeline by having the AP define a
PSMP sequence that includes scheduled times for downlink and
uplink transmissions. Wastage of energy in the MAC layer is
mainly because of collisions, overheads, listening for potential
traffic and overhearing traffic meant for other nodes. Protocols,
therefore, attempt to remove one or more or these problems.
Most of the available data link energy conservation approaches
fall into the following broad categories 1) Duty cycle with
single radio 2) Duty cycle with dual radio 3) Topology control
4) Cluster based. Some of the representative protocols in each
of these categories are described below:
1) Duty cycle with single radio: These protocols involve use
of sleep-activity schedules. However, only a single radio is used
for signaling and data traffic:
a) S-MAC (Sensor-Medium Access Control): It is one of the
simple protocols proposed in this category. The node follows
fixed cycle of sleep and active times. This will consume lesser
energy than in cases where the node listen all the time when
idle. The nodes make their own schedules and broadcast them
to the neighbors. Cluster of nodes that know each other’s
schedule stay synchronized. It wastes energy handling low
traffic and this will result in low throughput [93].
b) T-MAC (Timeout-MAC): [42] As opposed to fixed
schedules of S-MAC, this protocol uses adaptable schedules
based on hearing no activity for a certain period of time. This
results in improved efficiency over S-MAC [94]
c) ECR-MAC (Efficient Cognitive Radio): This protocol uses
multiple forwarders for each node so that multiple paths are
available to the destination. The first available node is normally
used, resulting in reduced wait time involved in waiting for a
particular node to be free [95].
d) SOFA (A Sleep-Optimal Fair-Attention scheduler): this
protocol uses a downlink traffic scheduler on the AP of a
WLAN, called SOFA, which help its PSM clients to save
energy by allowing them to sleep more, hence to reduce energy
consumption. If a client still has pending packets from the
beacon period but the AP is servicing other clients then this
client has to remain awake till the last packet scheduled for it in
the beacon period has been delivered. Therefore, a large portion
of the client’s energy wastage comes from the access points
transmitting other clients’ packets before it finishes transmitting
the client’s last packet to it. SOFA manages to reduce such
energy wastage by maximizing the total sleep time of all clients
[96], [78].
e) MT-MAC (Multi-hop TDMA Energy-efficient Sleeping
MAC) Protocol: In order to improve the performance of energy
efficiency, throughput and delay in WMSN, Multi-hop TDMA
Energy-efficient Sleeping MAC (MT-MAC) protocol can be
used. The main idea of MT-MAC is to divide one frame into
many slots for nodes (primarily sensor) to forward data packets,
with TDMA scheduling method. As collision and hidden
terminal problems are considered in the scheduling algorithm,
so the probability of collision in the network will be reduced
and the throughput will be increased [97].
2) Duty cycle with dual radio: These protocols involve sleep-
activity schedules. They use separate radios for signaling and
data traffic.
a) STEM (Sparse Topology and Energy Management):
STEM has been one of the early CSMA-based MAC protocols
for sensor networks that scheduled sleep and active periods.
One of the radios is used to wake up neighboring nodes. This
radio uses duty cycles. The second radio is used to send data to
the forwarding node and is always asleep until woken up. As
waking up involves some delay there is a tradeoff in which
higher efficiency comes with higher delay [98].
b) PTW(Pipeline Tone Wakeup) scheme: This protocol aims
to take care of the tradeoff between energy efficiency and delay
present in STEM. It uses techniques to shorten the time taken to
wake up the nodes for data transfer [99].
c) LEEM (The Latency minimized Energy Efficient MAC)
protocol: This protocol also aims to reduce latency while
maintaining energy efficiency. The nodes along the path to
destination are synchronized so that they can be woken up
sequentially [100].
d) PAMAS (Power-aware Multi Access Protocol with
Signaling): This protocol uses signaling on a separate radio to
get good throughput and latency for the data. It overhears
signaling for other nodes and turns off a node’s radio for the
duration for which the medium will not be available. The
signaling interface could be left on so that the node listens to
the signal exchange and can respond fast to the traffic directed
towards it [101].
3) Topology/power control based: These protocols adjust
transmit power levels to control the connectivity to other nodes
or select topology by deciding which nodes to keep on:
a) XTC (eXtreme Topology Control): Neighbors are ranked
24
for link quality by each node broadcasting at maximum power.
Each node transmits its ranking results to neighboring nodes.
Finally, the nodes select the neighbors to be directly connected
to. The protocol does not need location information [102].
b) LFTC(Location Free Topology Control): LFTC protocol
constructs power-efficient network topology. It also avoids any
potential collision due to hidden terminal problem. Initially
each node broadcasts hello message with vicinity table. The
nodes then adjust their transmission power to communicate
with direct neighbors. Data collision and hidden terminal
problems are avoided by choosing appropriate power for data
and control. This protocol improves over XTC as it has smaller
hop count and better reliability [103].
c) PEM (Power-Efficient MAC) Protocol: PEM enhances the
network utilization and reduces the energy consumption. The
stations estimate their distances from the transmitter and obtain
the interference relations among transmission pairs through a
three-way handshaking. Transmission is scheduled based on the
interference relation [104].
d) Virtualization of NICs: An obvious way to save energy is
to switch off some of the wireless nodes during the off-peak
hours. Of course, some of the locations in the area of interest
would not be covered. These can be provided connectivity by
network coverage extension/relaying capability. Not only
shutting down nodes saves energy, NIC virtualization also
reduces energy consumption [105].
4) Cluster based MAC protocols: These protocols divide
nodes into groups and select a cluster head or a coordinator for
inter-area communication and data aggregation. Non-cluster
nodes can then have power saving sleep schedules:
a) LEACH (Low-Energy Adaptive Clustering Hierarchy)
protocol: LEACH is a scheduled MAC protocol with clustered
topology. It is a hierarchical routing algorithm that combines
reduction in energy consumption with quality of media access.
It divides the network topology into clusters and a cluster head
represents each cluster. The cluster head aggregates data for all
the nodes in a cluster and provides better performance in terms
of the lifetime [106]. The independent decision of whether or
not to become a cluster head may lead to accelerate energy
drainage for some of the nodes in the network. LEACH-C is a
centralized scheme, which can improve some of the issues
related to LEACH.
b) PEGASIS (Power-Efficient Gathering in Sensor
Information Systems): This protocol makes use of cluster heads
in a different way from LEACH and is an improvement over it.
Instead of each cluster head communicating with the
destination, a chain of cluster heads is formed. The chain is
constructed starting with the node farthest away from the
destination and finally ending up closest to the destination.
Then a cluster head is chosen that will perform data
aggregation. In case of failure of any of the nodes, the chain is
re-constructed. A new leader is selected randomly during each
round of data gathering [107].
Important features and energy saving strategy of the data link
layer protocols are described in Table X.
C. Energy conservation in the physical layer
Physical Layer (PHY) is concerned with the hardware
implementation, connectivity to neighbors, implements
encoding and signaling and moves the bits over the physical
medium. For designing energy efficient protocols it is important
to consider the physical characteristics of the electronics
involved. Understanding of physical layer is important for
successful implementation of data-link and network layers.
High mobility of nodes in UAV networks creates issues for
physical layer. We will consider these protocols in the
following four groups: 1) Dynamic voltage control 2) Node
level power scheduling 3) Choosing minimum subset, and 4)
Sleep schedules with buffering:
TABLE X
ENERGY CONSERVATION TECHNIQUES IN DATA LINK LAYER
Protocol
Features
Energy Saving
Strategy
1. Duty cycle with single radio
S-MAC
Nodes have duty cycles. Nodes
broadcast their schedules to neighbors.
Consumes less energy than protocols
that listen all the time.
Scheduled sleep
times for nodes
T-MAC
Schedules based on activity. Has better
efficiency than S-MAC
Adaptable
schedules of sleep
times
ECR-MAC
Each node has many forwarders. The
first available is used to reduce wait
time.
Reduced wait time
for forwarding
data
SOFA
Increases sleep time by allowing nodes
not to wake up when other nodes packets
are transmitted.
Sleep more
MT-MAC
Divides one frame into many slots for
sensor nodes to forward data packets.
Collisions are reduced and network
throughput will be increased.
Increased sleep
time
2. Duty cycle with dual radio
STEM
It has scheduled sleep and active period.
Separate radio for signaling and data.
The signaling radio has scheduled times
while data radio sleeps till woken up.
Higher delay.
Data radio sleeps
longer
PTW
Takes care of the tradeoff between
energy efficiency and delay of STEM.
Shortens the time taken to wake up the
nodes for data transfer.
Data radio sleeps
longer, pipelining
used to reduce
delay
LEEM
Reduce latency while maintaining
energy efficiency. Nodes along the path
to destination are synchronized so that
they can be woken up sequentially
Nodes
synchronized for
fast wakeup
PAMAS
Signaling radio is left on for listening.
Data radio of a node is switched on
when it receives packet meant for it.
Data radio sleeps
when medium is
not available
3. Topology/power control based
XTC
Each node connects to neighbors based
on link quality.
Good quality link
selection
LFTC
Nodes change topology by adjusting
power. Data collision and hidden
terminal problems are avoided. Improves
over XTC on hop count and data
delivery.
Transmission
power control for
data and control
PEM
Stations can estimate their distances
from the transmitter and obtain the
interference relation among transmission
pairs. Stations schedule their
transmission according to interference
relation.
Transmission
scheduling based
on interference
Virtualization
Switch off some nodes and provide
Shutting node and
25
of NICs
connectivity to cut off areas by coverage
extension, relaying. NIC virtualization
also reduces energy consumption.
virtualization
4. Cluster
based
LEACH
Data aggregation by cluster head. Nodes
have duty cycles. It is a hierarchical
routing algorithm that combines
reduction in energy consumption with
quality of media access.
Cluster based,
nodes
communicate with
cluster heads and
have sleep-activity
schedules
PEGASIS
Divides network into clusters. A path of
cluster heads is formed and one of them
is chosen to perform data aggregation.
Reducing cluster
heads that perform
forwarding
!
The physical layer deals with modulation and signal coding
and related signal transmission technologies. The extremely
high mobility nodes of aerial networks make design of the
physical layer more involved. In case of UAV networks the
movement is in 3D and in order that data is not lost in data
communication architectures the physical layer conditions have
to be well understood and well defined. The 3D networks have
several accentuated concerns, such as, variations in
communication distance, direction of the communicating pairs,
antenna radiation pattern, shad- owing from the UAV and
onboard electronic equipment, environmental conditions,
interferences, and jamming [150]. The authors suggest
mitigating high link loss and variations by spatial and temporal
diversity. They feel that more research is required in the
physical layer and antenna propagation for 3D environments.
1) Dynamic voltage control
These protocols adjust circuit bias voltage levels, depending
on the traffic, to optimize energy usage. In this regard as the
UAV consumes battery power, the battery voltage may drop
and the UAV may become dysfunctional. However, if the
router, and other circuits in the payload can fallback to lower
voltage levels then the flying time of the UAV can be increased.
The following protocol is an example.
a) Dynamic Voltage Scaling (DVS): DVS exploits variability
in processor workload and latency constraints and realizes this
energy-quality trade-off at the circuit level. When the traffic
load is low, the bias voltage can be reduced to get
proportionally quadratic saving in energy [108]
2) Node level power scheduling
These protocols do node level optimizations to achieve
energy efficiency and enhance network lifetime. In UAV
controlling the transmission power can control the topology of
the networks. In the extreme case connectivity can only be
maintained with single hop neighbors. Communication can then
take place in a multi-hop manner. The following protocols are
in this category.
a) LM-SPT (Local Minimum Shortest Path-Tree): This
method improves power efficiency by a localized distributed
power-efficient topology control algorithm. The main idea in
this approach is to construct an overlay graph topology over the
wireless mesh such that this topology has the required features
like increased throughput, increased network lifetime and
maintained links by varying transmission power at each node.
The algorithm balances energy efficiency and throughput in
wireless mesh network without the loss of connectivity. The
concept of this approach is based on information of the local
neighborhood that is confined to one hop for calculating the
minimum power transmission [109].
b) Minimum-energy topology: This method generates a
minimum-energy topology graph G = (V, E) where V represents
nodes and E represents the links. It has a localized distributed
topology control which calculates the optimal transmission
power to maintain connectivity and reduces power to cover
only nearest neighbors. Energy is saved and lifetime of the
network also increases [110].
c) CPLD (Complex Programmable Logic Devices): CPLDs
are used with popular application chipsets to reduce standby
power and minimize the time that key processors need to be
powered on to detect system events. To minimize power
consumption in the network, CPLD may be used in the nodes of
the mesh network, which cuts off the power from the processor
of the station while it is being idle for some specific time. It can
detect any beacon frame arriving from the base station and
switches the power on immediately to the processor without
wasting any time. Implementation of CPLDs in devices, which
are part of wireless mesh networks instead of microcontrollers,
coupled with smart electronic switches may drastically reduce
the power consumption [115].
3) Choosing Minimum Subset
These protocols work on the principle of choosing a subset
of nodes to switch on and keep the other nodes switched off in
order to save energy. Three examples of protocols in this
category are as follows.
a) CNN (Critical Number of Neighbors): CNN refers to the
minimum number of neighbors that should be maintained by
each node in the network to be asymptotically connected. This
approach to maintain connectivity only requires knowledge of
the network size is required to determine the CNN. This
information can be easily obtained from a proactive routing
protocol such as OLSR. Power is controlled for individual
nodes to maximize power savings and controlling interference
[111].
b) Virtual WLAN: This method proposes a switching scheme
that aims to power on the minimum number of devices or the
combination of devices that consume the least energy that can
jointly provide full coverage and enough capacity. By
consolidating hardware, some hardware can be put in low-
power mode and energy consumption can be reduced.
Depending on the type of device, different amounts can be
saved [112].
c) Energy Savings in Wireless Mesh Networks in a Time-
Variable Context: This is an approach to minimize the energy
in a time varying context by selecting dynamically a subset of
mesh BSs to switch on considering coverage issues of the
service area, traffic routing, as well as capacity limitations both
on the access segment and the wireless backhaul links. To
achieve the desired objective, the algorithm considers traffic
demands for a set of time intervals and manages the energy
consumption of the network with the goal of making it
proportional to the load [113].
4) Sleeping Schedules with Buffering
26
These protocols aim to prolong sleep times by buffering
packets that arrive while a node is inactive. An example is
discussed here:
a) SRA (Sleeping and Rate-Adaptation): There are two forms
of power management schemes that reduce the energy
consumption of networks. The first puts network interfaces to
sleep during short idle periods. To make it effective small
amounts of buffering is introduced for sleeping clients that will
store the packets to create a long enough gap for the client to
sleep and save energy. Potential concerns are that buffering will
add too much delay across the network and that bursts will
exacerbate loss. The algorithms arrange for routers and
switches to sleep in a manner that ensures the buffering delay
penalty is paid only once (not per link) and that routers clear
bursts so as to not amplify loss noticeably [114].
Important features and the energy saving strategy of the
physical layer protocols are given in Table XI
TABLE XI
ENERGY CONSERVATION IN PHYSICAL LAYER
Protocol
Features
Energy Saving
Strategy
1. Dynamic voltage control
DVS
Circuit voltage adjusted according to
processor load and traffic latency
requirements.
Low voltage (low
power) for low
loads
2. Node level power scheduling
LM-SPT
Transmission power is varied at each
node to get a reduced topology meeting
the objectives like throughput, network
lifetime or link quality.
Control power to
control topology
Minimum
energy
topology
Optimal power to maintain connectivity
with the nearest neighbors used. Energy
is save and network lifetime increased
Node level
transmit power
control
CPLD
Adaptive power-on time control for
electronics. CPLD cuts off power when
node is idle for a specific time. Smart
electronics
Device level power
management
3. Choosing minimum subset
CNN
Selects minimum number of neighbors
for each node for complete connectivity.
Power is controlled for individual nodes
to maximize power savings and
controlling interference. Requires
knowledge of network size
Neighbor selection
and power control
for each selected
node.
Virtual
WLAN
Devices that will consume least energy
and provide the required coverage and
capacity are powered on. Remaining
hardware is in low-power mode
Active node
selection based on
coverage and
capacity.
Energy
saving
WMN
Energy consumption is proportional to
the traffic load. Enough nodes are
selected in each time interval to meet the
coverage, capacity and traffic routing
requirements.
Traffic demand
based active node
selection
4. Sleep schedules with buffering
SRA
Nodes are allowed to sleep longer by
storing packets. The packets are then
forwarded in bursts during active
windows. Burst losses are minimized.
Increase node sleep
time with buffers
and managing
delays
D. Energy Conservation through Cross-Layer Protocols
Protocols considered so far work predominantly on solving
issues relating to one individual protocol layer. The network
layer protocols deal with path or node selection for maintaining
connectivity of the entire network. The data link layer protocols
(specifically the MAC sub-layer protocols) are more involved
in avoiding collisions and designing duty cycles. The physical
protocol concerns with device characteristics for voltage and
power control. Maximizing energy conservation in one layer
may still give overall inefficient energy conservation across all
layers. Very little work is available in the area of cross-layer
protocols in mobile ad hoc network and application of these to
UAV networks is still an open issue. We discuss here a couple
of representative work in this area.
1) MTEC (Minimum Transmission Energy Consumption)
Routing protocol with ACW (Adaptive Contention Window):
To design an energy efficient protocol a number of factors like
proportion of successful data transmissions, residual energy on
the nodes and traffic condition on the nodes need to be
considered. This cross-layer design can decrease the energy
consumption of data transmissions, but it also prolongs network
time. MTEC routing protocol works at the network layer and
takes into account the proportion of successful data
transmissions, the traffic load of nodes and the number of nodes
contending for a channel, to find a suitable path. It reduces
energy consumption and prolong network lifetime. For this it
finds nodes with sufficient residual energy for successfully
transmitting all data packets. ACW works at the MAC layer to
reduce energy consumption and throughput. Based on the
proportion of successful data transmissions, a node uses an
ACW to dynamically adjust the back-off time between different
nodes. It reduces re-transmissions by allowing links with higher
proportion of successful transmissions to use channel more
[116].
2) CLEEP (Cross Layer Energy Efficient Protocol)
The protocol works across physical, MAC and network layers.
At the physical layer level, this protocol first obtains the
transmission power required to keep two neighboring nodes
connected. It maintains a connected neighbor table for each
node in the network. This information can be utilized by the
network layer to choose a better routing path for data. The
protocol then utilizes the routing information to decide the
sleep-activity pattern of the nodes in the MAC layer for
maximizing sleep times [117].
3) CLE
2
aR
2
(cross-layer energy-efficient and reliable routing
protocol)
This has been proposed to find an energy efficient and reliable
route from the source to the destination. It counters channel
quality variation and reduces the retransmission cost for
wireless ad hoc net- works. For each node, cost of relaying data
from the source to destination is calculated if this node falls in
the route. This also takes into account retransmission cost in
terms of energy consumption. The node receiving an RREQ
also estimates the channel quality based on the received signal
strength. The node also takes its interference pattern into
account. With all these calculations CLE2aR2can find a route
with less energy consumption and high reliability [119].
VI. CONCLUSIONS
UAV networks are growing in importance and general
27
interest for civil applications. Providing good inter-UAV
connectivity and links to the users and any ground station is
quite challenging. Research relating to mobile ad hoc mesh
networks is being applied to the UAV networks, but even the
former is an evolving area. Additionally, a number of features
like dynamicity of nodes, fluid topology, intermittent links,
power and bandwidth constraints set UAV networks apart from
any other that have been researched before. Some researchers
believe that there is need to re-build everything ground up. This
includes features in the physical layer, data link layer, network
layer and the transport layer. Some issues like energy
conservation and ensuring adequate quality of service require
cross-layer design.
In this survey we attempt to focus on research in the areas of
Routing, seamless handover and energy efficiency. The reason
to undertake this survey was lack of a survey focusing on these
issues. In order to effectively process and present the available
information in correct perspective, it was considered necessary
to categorize the UAV networks based on a number of
characteristics. It is important to distinguish between
infrastructure and ad-hoc UAV networks, applications areas in
which UAVs act as servers or as clients, star or mesh UAV
networks and whether the deployment is hardened against
delays and disruptions. Through this discussion we see how
despite sharing some characteristics with mobile and vehicular
ad-hoc networks, UAV networks have their own unique
properties. Having done this classification, we focus on the
main issues of routing, seamless handover and energy
efficiency in UAV networks
Routing has unique requirements - finding the most efficient
route, allowing the network to scale, controlling latency,
ensuring reliability, taking care of mobility and ensuring the
required quality of service. In UAV networks, additional
requirements of dynamic topology (with node mobility in 2-D
and 3-D), frequent node addition and removal, robustness to
intermittent links, bandwidth and energy constraints make the
design of a suitable protocol one of the most challenging tasks
[3]. This area is still evolving and we are still to see native
UAV network protocols. Researchers have proposed
modifications to existing protocols to make them workable for
UAV scenarios. We discuss four categories of protocols and see
the extent to which they are suitable for UAV networks. Static
protocols have limited applicability as routing tables are
manually configured and cannot be changed once the network
has been launched. Proactive protocols keep up-to-date tables
but need to exchange a number of messages between the nodes.
This makes them unsuitable for UAV networks on two counts –
bandwidth constraints and slow reaction to topology changes
causing delays. In this category, a well known protocol called
OLSR would track fast changes in the UAV network at the cost
of increased overhead of control messages leading to
contention, packet loss and bandwidth consumption even with
Multi Point Relays (MPR). Another traditional protocol,
DSDV, when used in aerial networks, puts a high computing
and storage burden for maintaining freshness of routes. Newer
protocols like BABEL and B.A.T.M.A.N have also not found to
be distinctly superior. In reactive protocols like DSR, finding
new routes is cumbersome. AODV has been studied for
possible adaptation in UAV networks but delays in route
construction triggers route discovery and compounds delays.
Throughput suffers due to intermittent links. Some adaptations
like reactive-greedy-reactive (RGR) have been shown to
perform better. Hybrid protocols present a compromise between
higher latency of reactive protocols and higher overhead of
proactive protocols. In some cases like the ZRP protocol, the
added complexity in the UAV networks may outweigh the
slight improvement in performance.
In applications where UAV networks are delay and
disruption prone and partitioning is the norm rather than
exception, continuous end-to-end connectivity cannot be
assumed. The transmission delays increase beyond TCP
threshold limits and packets are dropped. In such situations a
UAV network with delay tolerant features is considered to be
one of the most effective. The architecture would be based on
store-carry-forward. If the data cannot be delivered immediately
then the nodes chosen to carry the message are those that have
highest probability of delivering the message. Selection of the
path from source to destination depends on whether the
topology that evolves over time is deterministic or probabilistic.
In case of probabilistic topology, messages may be sent through
flooding (epidemic routing) with large requirement of buffer
space, bandwidth and power. Variation like Spray and Wait use
fixed number of copies and reduce resource requirement.
Controlled ferrying can be used in UAV networks in disaster
recovery where UAVs can be equipped with communication
devices capable of storing a large number of messages and can
be commanded to follow a trajectory that interconnects
disconnected partitions. Some applications like communication
service in a remote area or over an oil rig could benefit from
social network model where nodes remain in some known
locations.
Seamless handover allows for total continuity of network
communication with only a minor increase of message latency
during the handover process. The handover latency and the
packet loss during handover process may cause serious
degradation of system performance and QoS perceived by the
users. There has been hardly any study on seamless handover in
the UAV environment and more so using IEEE Wireless
Access in Vehicular Environments (WAVE) suite of protocols.
Many mobility management protocols have been proposed but
high degree of mobility forces frequent handover and problems
in communication. IEEE has standardized Media Independent
Handover (MIH) services through their standard IEEE 802.21.
These services can be used for handovers and interoperability
between IEEE-802 and non-IEEE-802 networks, e.g., cellular,
3GPP, 4G. MIH, however, does not provide intra-technology
handover, handover policies, security and enhancements to link
layer technologies. However, MIH is a nascent technology that
has not been widely deployed and evaluated.
Energy efficiency is a very important requirement in UAV
networks. Reducing the energy consumption helps in increase
in network lifetime and useful payload that can be carried.
Energy consumption can be reduced through transmission
power control, load distribution or making nodes sleep. At the
physical layer transmission power can be reduced to the
28
minimum required for connectivity. Network layer can use the
information about connected nodes to route packets. The data
link layer schedule on/off times of signaling and data carrying
radios. Cross layer protocols will offer schemes working at two
or more layers.
There have been some good studies that have focused on a
number of issues like applications, protocols and mobility [3],
[12], [15]. In the present survey we focus on issues relating to
characterization of UAV networks, routing under constraining
circumstances, automating control with SDN, seamless
handovers and greening of UAV networks that have not been
the focus of the earlier surveys. To the best of knowledge this is
the first survey that bring forth the current research in these
areas and their importance in building successful multi UAV
networks. We expect that this survey would spur more research
work in these important but understudied areas with open
research issues.
REFERENCES
[1] K. Daniel, C. Wietfeld, “Using Public Network Infrastructures for UAV
Remote Sensing in Civilian Security Operations,” Homeland Security
Affairs, Best Papers from the IEEE Conference on Technologies for
Homeland Security, Jan 2011, http://www.hsaj.org/?article=0.3.3 Access
date: Oct 2014
[2] P. B. Bök, Y. Tuchelmann, “Context-Aware QoS Control for Wireless
Mesh Networks of UAVs,” Proceedings of the International Conference
on Computer Communications and Networks (ICCCN), Aug 2011, pp. 1–
6
[3] I. Bekmezci, O. K. Sahingoz, S. Temel, “Flying Ad-Hoc Networks
(FANETs): A survey,” Ad Hoc Networks, 2013, pp. 1254–1270
[4] US Department of Transportation Federal Aviation Authority,
“Integration of Civil Unmanned Aircraft Systems (UAS) in the National
Airspace System (NAS) Roadmap,” First Edition, 2013,
http://www.faa.gov/uas/media/uas_roadmap_2013.pdf Access date: Oct
2014
[5] N. Nandiraju et al., “Wireless Mesh Networks: Current Challenges and
Future Directions of Web-In-The-Sky,” IEEE Wireless Communications,
2007, vol. 14, no. 4, pp. 79–89
[6] L. Song Lu, T. Huang, “A summary of key Technologies of Ad hoc
Networks with UAV Node,” IEEE International Conference on Intelligent
Computing and Integrated Systems (ICISS), 2010, pp. 944–949
[7] O. K. Sahingoz, “Networking Models in Flying Ad-Hoc Networks
(FANETs): Concepts and Challenges,” Journal of Intelligent Robotic
Systems, 2014, vol. 74, no. 1–2, pp. 513–527
[8] S. Morgenthaler, T. Braun, Zhongliang Zhao, T. Staub, M. Anwander,
“UAVNet: A mobile wireless mesh network using Unmanned Aerial
Vehicles,” IEEE Globecom Workshops (GC Wkshps), 2012, pp. 1603–
1608
[9] O. K. Sahingoz, “Mobile networking with UAVs: Opportunities and
challenges,” International Conference on Unmanned Aircraft Systems
(ICUAS), 2013, pp. 933–941
[10] T. K. Huang, C. K. Lee, and L. J. Chen, “PRoPHET+: An adaptive
PRoPHET-based routing protocol for opportunistic network,” 24
th
IEEE
International Conference on Advanced Information Networking and
Applications (AINA), April 2010, pp. 112–119
[11] Yanli Yang, M. M. Polycarpou, A. A. Minai, “Multi-UAV Cooperative
Search Using an Opportunistic Learning Method,” Journal of Dynamic
Systems, Measurement, and Control, Jan 2007, vol. 129, no. 5, pp. 716–
728
[12] Z. Zhao, Torstten Braun, “Topology Control and Mobility Strategy for
UAV Ad hoc Networks,” Joint ERCIM eMobility and Mobisense
Workshop, June 2012.
[13] I. Bekmezci, M. Ermis, S. Kaplan, “Connected Multi UAV Task Planning
for Flying Ad Hoc Networks,” IEEE International Black Sea Conference
on Communications and Networking (BlackSeaComm), 2014, pp. 28–32
[14] S. Rosati, K. Kruzelecki, L. Traynard, B. Rimoldi, “Speed-Aware Routing
for UAV Ad-Hoc Networks,” IEEE Globecom Workshops (GC Wkshps),
2013, pp. 1367–1373
[15] E. W. Frew, T. X. Brown, “Airborne Communication Networks for Small
Unmanned Aircraft Systems,” Proceedings of the IEEE, 2008, vol. 96, no.
12, pp. 2008–2027
[16] Jun Li, Yifeng Zhou, L. Lamont, “Communication Architectures and
Protocols for Networking Unmanned Aerial Vehicles,” IEEE Globecom
Workshops, 2013, pp. 1415–1420
[17] I. Dalmasso, I. Galletti, R. Giuliano, F. Mazzenga, “WiMAX networks for
emergency management based on UAVs,” IEEE First AESS European
Conference on Satellite Telecommunications (ESTEL), 2012, pp. 1–6
[18] F. J. Watkins, R. A. Hinojosa, A. M. Oddershede, “Alternative Wireless
Network Technology Implementation for Rural Zones,” International
Journal of Computers, Communications & Control, ISSN 1841–9844 Feb
2013, vol. 8, no. 1, pp.161–165.
[19] Y. Wang, Y. J. Zhao, "Fundamental issues in systematic design of
airborne networks for aviation," IEEE Aerospace Conference, 2006.
[20] A. P. Athreya, P. Tague, “Network Self-Organization in the Internet of
Things,” IEEE International Workshop of Internet-of-Things Networking
and Control (IoT-NC), Jun 2013, pp. 25–33
[21] P. Sharnya, J. S. Raj, “Self Organizing Wireless Mesh Network,”
International Journal of Innovation and Applied Studies, ISSN 2028–9324
Jun 2013, vol. 3, no. 2, pp. 486–492
[22] M. Mendonca, B. N. Astuto, K. Obraczka and T. Turletti, “Software
Defined Networking for Heterogeneous Networks,” IEEE MMTC E-
Letters, 2013, vol. 8, no. 3, pp. 36–39
[23] I. Ku et al., “Towards Software-Defined VANET: Architecture and
Services,” 13
th
Annual Mediterranean Ad Hoc Networking Workshop
(MED-HOC-NET), 2014, pp. 103–110
[24] S. Costanzo, L. Galluccio, G. Morabito, S. Palazzo, “Software Defined
Wireless Networks: Unbridling SDNs,” European Workshop on Software
Defined Networking (EWSDN), 2012, pp. 1–6
[25] Open Networking Foundation (ONF), “OpenFlow™-Enabled Mobile and
Wireless Networks,” ONF Solution Brief, Sept 2013,
https://www.opennetworking.org/images/stories/downloads/sdn-
resources/solution-briefs/sb-wireless-mobile.pdf Access date: Oct 2014
[26] N. McKeown et al., “OpenFlow: enabling innovation in campus
networks,” ACM SIGCOMM Computer Communication Review, 2008,
vol. 38, no. 2
[27] A. Detti, C. Pisa, S. Salsano, N. Blefari-Melazzi, “Wireless Mesh
Software Defined Networks (wmSDN),” IEEE 9
th
International
Conference on Wireless and Mobile Computing, Networking and
Communications (WiMob), Oct 2013, pp. 89–95
[28] P. Dely, A. Kassler, N. Bayer, “Openflow for wireless mesh networks,”
Proceedings of 20
th
International Conference on Computer
Communications and Networks (ICCCN), July–Aug 2011, pp. 1–6
[29] W. Xia, Y. Wen, C. H. Foh, D. Niyato and H. Xie, “A Survey on
Software-Defined Networking,” IEEE Communications Surveys &
Tutorials, 2014(to be published)
[30] Bo Fu, L. A. DaSilva, “A Mesh in the Sky: A Routing Protocol for
Airborne Networks,” IEEE Military Communications Conference
(MILCOM 2007), Oct 2007 , pp. 1 –7
[31] J. Shen, S. Moh , I. Chung, “Routing Protocols in Delay Tolerant
Networks: A Comparative Survey,” The 23
rd
International Technical
Conference on Circuits/Systems, Computers and Communications (ITC-
CSCC 2008), 2008
[32] M. Abolhasan, T. Wysocki, E. Dutkiewicz, "A review of routing protocols
for mobile ad hoc networks," Ad Hoc Networks, Jan 2004, vol. 2, no. 1,
pp. 1 –22
[33] F. Akyildiz, X. Wang, W. Wang, “Wireless mesh networks: a survey,”
Computer Networks, Mar 2005, vol. 47, no. 4, pp. 445–487,
[34] Chen-Mou Cheng, Pai-Hsiang Hsiao, H. T. Kung, D. Vlah, “Maximizing
Throughput of UAV-Relaying Networks with the Load-Carry-and-Deliver
Paradigm,” IEEE Wireless Communications and Networking Conference
(WCNC), Mar 2007, pp. 4417–4424
[35] T. Clausen, P. Jacquet, “Optimized Link State Routing,” IETF RFC 3626,
Oct 2003
[36] Bow-Nan Cheng, S. Moore, "A comparison of MANET routing protocols
on airborne tactical networks," Military Communications Conference
(MILCOM), Oct–Nov 2012, pp.1–6
29
[37] J. Chroboczek, “The Babel Routing Protocol,” IETF RFC 6126, Apr 2011
[38] A. Neumann, C. Aichele, M. Lindner, S. Wunderlich, “IETF Better
Approach To Mobile Ad hoc Networking (B.A.T.M.A.N.),” IETF Internet
Draft draft-wunderlich-openmesh-manet-routing-00, Apr 2008
[39] D. S. Sandhu, S. Sharma, “Performance Evaluation of BATMAN, DSR,
OLSR Routing Protocols - A Review,” International Journal of Emerging
Technology and Advanced Engineering, Jan 2012, vol. 2, no. 1, pp. 184–
188
[40] D. Murray, M. Dixon, T. Koziniec, “An Experimental Comparison of
Routing Protocols in Multi Hop Ad Hoc Networks,” Australasian
Telecommunication Networks and Applications Conference, Oct–Nov
2010, pp. 159–164
[41] A. Sharma, N. Rajagopalan, “A Comparative Study of B.A.T.M.A.N. and
OLSR Routing Protocols for MANETs,” International Journal of
Advanced Trends in Computer Science and Engineering (IJATCSE), 2013,
vol. 2, no. 5, pp. 13–17,
[42] D. Johnson, N. Ntlatlapa, C. Aichel, “Simple pragmatic approach to mesh
routing using BATMAN,” 2nd IFIP International Symposium on Wireless
Communications and Information Technology in Developing Countries,
Oct 2008, pp. 1–10
[43] D. Johnson, Y. Hu, D. Maltz, “The Dynamic Source Routing Protocol
(DSR) for Mobile Ad Hoc Networks for IPv4,” IETF RFC 4728, Feb
2007
[44] T. X. Brown, B. Argrow, C. Dixon, S. Doshi, R. G. Thekkekunel, D.
Henkel, “Ad hoc UAV ground network (AUGNet),” AIAA 3
rd
Unmanned
Unlimited Technical Conference, Sept 2004, pp. 29–39
[45] C. Perkins, E. Belding-Royer, S. Das, “Ad hoc On-Demand Distance
Vector (AODV) Routing,” IETF RFC 3561, July 2003
[46] B. Nyirenda, J. Mwanza, “Performance Evaluation of Routing Protocols
in Mobile Ad hoc Networks (MANETs),” Blekinge Institute of
Technology, Jan 2009,
http://www.bth.se/fou/cuppsats.nsf/all/247e26e7f470cf9fc1257543005b1b
fc/$file/manets_thesis_report.pdf Access date: Oct 2014
[47] Z. J. Haas, M. R. Pearlman, P. Samar, “The Zone Routing Protocol (ZRP)
for Ad Hoc Networks,” IETF Internet Draft draft-ietf-manet-zone-zrp-04,
July 2002
[48] V. Park, S. Corson, “Temporally-Ordered Routing Algorithm (TORA),”
IETF Internet Draft draft-ietf-manet-tora-spec-04, July 2001
[49] N. Uchida, N. Kawamura, N. Williams, K. Takahata, Y. Shibata,
“Proposal of Delay Tolerant Network with Cognitive Wireless Network
for Disaster Information Network System,” 27
th
International Conference
on Advanced Information Networking and Applications Workshops
(WAINA), Mar 2013, pp. 249–254
[50] Ashraf E. Al-Fagih, Hossam S. Hassanein, “Routing Schemes for Delay-
Tolerant Networks-An Applications Perspective,” Technical Report 2012-
588, Ph.D. Depth Paper, Dec 2012,
http://research.cs.queensu.ca/TechReports/Reports/2012-588.pdf Access
date: Oct 2014
[51] Jiajia Liu, Xiaohong Jiang, H. Nishiyama, N. Kato, ”A General Model for
Store-carry-forward Routing Schemes with Multicast in Delay Tolerant
Networks,” 6
th
International ICST Conference on Communications and
Networking in China (CHINACOM), Aug 2011, pp. 494–500
[52] M. Le, Joon-Sang Park, and M. Gerla, “UAV Assisted Disruption
Tolerant Routing,” IEEE Military Communications Conference
(MILCOM), Oct 2006, pp. 1–5
[53] Z. Zhang, “Chapter 8 - Routing protocols in Intermittently Connected
Mobile Ad Hoc Networks and Delay-Tolerant Networks,” Algorithms and
Protocols for Wireless, Mobile Ad Hoc Networks, Wiley, 2009
[54] S. Jain, K. Fall, R. Patra, “Routing in Delay Tolerant Network,” ACM
SIGCOM Computer Communication Review, Oct 2004, vol. 34, no. 4, pp.
145–158
[55] A.J. Alshabtat, L. Dong, “Low Latency Routing Algorithm for
Unmanned Aerial Vehicles Ad-Hoc Networks ,” World Academy of
Science, Engineering and Technology Vol:5, 2011
[56] R. Handorean, C. Gill, G. C. Roman, “Accommodating Transient
Connectivity in Ad Hoc and Mobile Settings,” Pervasive Computing, Apr
2004, vol. 3001, pp. 305–22.
[57] Zhensheng Zhang, “Routing in intermittently connected mobile ad hoc
networks and delay tolerant networks: overview and challenges,” IEEE
communications Surveys & Tutorials, 2006, vol. 8, no. 1, pp. 24–37
[58] Zhenxin Feng, Kwan-Wu Chin, “State-of-the-Art Routing Protocols for
Delay Tolerant Networks,” Technical Report, Oct 2012,
http://arxiv.org/pdf/1210.0965.pdf, Access date: Oct 2014
[59] A. Balasubramanian, B. Levine, A. Venkataramani, “DTN Routing as a
Resource Allocation Problem,” ACM SIGCOMM Computer
Communication Review, Oct 2007, vol. 37, no. 4, pp. 373–384
[60] R. Ramanathan, R. Hansen, P. Basu, R. Rosales-Hain, R. Krishnan,
“Prioritized epidemic Routing for Opportunistic Networks,” ACM
Proceedings of the 1
st
International MobiSys Workshop on Mobile
Opportunistic Networking (MobiOpp), Jun 2007, pp. 62–66
[61] J. Burgess, B. Gallagher, D. Jensen, B. N. Levine, “MaxProp: Routing for
Vehicle-Based Disruption-tolerant Networks,” Proceedings of the 25
th
IEEE International Conference on Computer Communications
(INFOCOM), Apr 2006, pp. 1–11
[62] Q. Li, D. Rus, “Communication in Disconnected Ad Hoc Networks Using
Message Relay,” Journal of Parallel and Distributed Computing, 2003,
vol. 63, no. 1, pp. 75–86.
[63] W. Zhao, M. Ammar, E. Zegura, “A Message Ferrying Approach for Data
Delivery in Sparse Mobile Ad Hoc Networks,” Proceedings Of the 5th
ACM International Symposium on Mobile Ad Hoc Networking and
Computing (MobiHoc), 2004, pp. 187–98
[64] M. R. Schurgot, C. Comaniciu, K. Jaffres-Runser, “Beyond Traditional
DTN Routing: Social Networks for Opportunistic Communication,“, IEEE
Communications Magazine, July 2012, vol. 50, no. 7, pp. 155–162
[65] M. Ohleger, G. G. Xie, J. H. Gibson, “Extending UAV Video
Dissemination via Seamless Handover: A Proof of Concept Evaluation of
the IEEE 802.21 Standard,” 46
th
Hawaii International Conference on
System Sciences (HICSS), Jan 2013, pp. 5106–5114
[66] M. Saini, S. Mann, “Handoff Schemes for Vehicular Ad-Hoc Networks: A
Survey,” 1
st
International Conference on Advancement in Electronics &
Communication Engineering (ICAECE), Mar 2013
[67] J. Dias, A. Cardote, F. Neves, S. Sargento, A. Oliveira, “Seamless
Horizontal and Vertical Mobility in VANET,” IEEE Vehicular
Networking Conference, Nov 2012, pp. 226–233
[68] H. Tyagi, A. K. Vatsa, “Seamless Handoff through Information Retrieval
in VANET Using Mobile Agent,” International Journal of Computer
Science Issues (IJCSI), Jul 2011, vol. 8, no. 4, pp. 634–640
[69] Yibo Yang, Hongling Li, Qiong Huang, “Mobility Management in
VANET,” 22nd Wireless and Optical Communication Conference
(WOCC), May 2013, pp. 298–303
[70] YS. Chen, CS. Hsu, CH. Cheng, “Network Mobility Protocol for
Vehicular Ad Hoc Networks,” International Journal of Communication
Systems, Mar 2013
[71] K. Zhu, D. Niyato, P. Wang, E. Hossain, D. I. Kim, “Mobility and hand
off management in vehicular networks: a survey,” Wireless
Communications and Mobile Computing, 2011, vol. 11, no. 4, pp. 459–
476.
[72] H. Petander, E. Perera, KC. Lan, A. Seneviratne, “Measuring and
improving the performance of network mobility management in IPv6
networks,” IEEE Journal on Selected Areas in Communications, 2006,
vol. 24, no. 9, pp. 1671–1681
[73] IEEE Computer Society, “Part 21: Media Independent Handover
Services,” 802.21c-2014 IEEE Standard for Local and metropolitan area
networks, IEEE Press, 2014
[74] R. S. Shukla, N. Tyagi, “Global Mobility And Handover Management For
Heterogeneous Network In VANET,” International Journal of Next-
Generation Networks (IJNGN), Sep 2013, vol. 5, no. 3, pp. 23–34
[75] J. Stein, “Survey of IEEE 802.21 media independent handover services,”
Technical Report, 2006, http://www.cs.wustl.edu/~jain/cse574-
06/handover.htm Access date: Oct 2014
[76] K. Taniuchi et al., “IEEE 802.21: Media Independent Handover: Features,
Applicability, and Realization,” IEEE Communications Magazine, Feb
2009, vol. 47, no. 1, pp. 112–120
[77] C. Guimaraes, D. Corujo, R. L. Aguiar, F. Silva, and P. Frosi,
“Empowering Software Defined Wireless Networks Through Media
Independent Handover Management IEEE Global Communications
Conference (GLOBECOM), Dec 2013, pp. 2204–2209
[78] S. Mamechaoui, F. Didi, G. Pujolle “A Survey On Energy Efficiency For
Wireless Mesh Network,” International Journal of Computer Networks &
Communications (IJCNC), Mar 2013, vol. 5, no. 2, pp. 105–124
[79] C. Yu, B. Lee and H. Yong Youn, “Energy efficient routing protocols for
mobile ad hoc networks,” Wireless Communications and Mobile
30
Computing, 2003, vol. 3, no. 8, pp. 959–973
[80] B. Bouyedou , M. Feham, F. Didi, H. Labiod, “Improvement of DSR
Performances in Mobile Ad hoc Networks with Trade-off between Energy
Efficiency and Path Shortness,” IEEE International Workshop on ITS for
an Ubiquitous Roads (UBIROADS), Jun 2009
[81] J-H. Chang, L. Tassiulas, “Energy conserving routing in wireless ad-hoc
networks,” Proceedings of the 19
th
Annual Joint Conference of the IEEE
Computer and Communications Societies (INFOCOM), 2000, vol. 1, pp.
22–31.
[82] B. Awerbuch, D. Holmer, H. Rubens, “The Pulse Protocol: Energy
Efficient Infrastructure Access,” Proceedings of the 23
rd
Annual Joint
Conference of the IEEE Computer and Communications Societies
(INFOCOM), 2004, vol.2, pp. 1467–1478
[83] L. Lin, N.B. Shroff, R. Srikant, “Asymptotically Optimal Power-Aware
Routing for Multihop Wireless Networks with Renewable Energy
Sources,” Proceedings of the 24
th
Annual Joint Conference of the IEEE
Computer and Communications Societies (INFOCOM), 2005, vol. 2, pp.
1262–1272
[84] K. Woo, C. Yu, H. Y. Youn, B. Lee, “Non-blocking, localized routing
algorithm for balanced energy consumption in mobile ad hoc networks,”
Proceedings of the 9
th
International Symposium on Modeling, Analysis
and Simulation of Computer and Telecommunication Systems
(MASCOTS), 2001, pp. 117–124.
[85] H. Gil, J. Yoo, J. Lee, “An On-Demand Energy-Efficient Routing
Algorithm for Wireless Ad Hoc Networks,“ Proceedings of the 2
nd
International Conference on Human Society@Internet (HSI), June 2003,
pp. 302–311
[86] C. Bemmoussat, F. Didi, M. Feham, “Efficient Routing Protocol To
Support QoS In Wireless Mesh Network,” International Journal of
Wireless & Mobile Networks (IJWMN), Oct 2012, vol. 4, no. 5, pp. 89–
104
[87] Y. Xu, J. Heidemann, D. Estrin, “Geography-informed energy
conservation for ad hoc routing,” Proceedings of the 7
th
Annual
International Conference on Mobile Computing and Networking
(MobiCom), 2001, pp. 70–84
[88] B. Chen, K. Jamieson, H. Balakrishnan, R. Morris, “Span: An Energy-
Efficient Coordination Algorithm for Topology Maintenance in Ad Hoc
Wireless Networks,” Journal of Wireless Networks, Sep 2002, vol. 8, no.
5 pp. 481–494
[89] M. Cardei, M. X. Cheng, X. Cheng, D. Z. Du, “Connected Domination in
Multihop Ad Hoc Wireless Networks,” Proceedings of the 6
th
Joint
Conference on Computer Science and Informatics, Mar 2002, pp. 1–5
[90] J. C. C. Restrepo, C. G. Gruber, and C. M. Machuca, “Energy profile
aware routing,” IEEE International Conference on Communications
Workshops (ICC Workshops), Jun 2009, pp. 1–5
[91] IEEE Computer Society, “Part 11: Wireless Medium Access Control
(MAC) and Physical Layer (PHY) Specifications: Medium Access
Control (MAC) Quality of Service (QoS) Enhancements,” 802.11e-2005
IEEE Standard, IEEE Press, 2005.
[92] IEEE Computer Society, “Part 11: Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) specifications: Amendment 5:
Enhancements for Higher Throughput,” 802.11n-2009 IEEE Standard,
IEEE Press, 2009
[93] Wei Ye, J. Heidemann, D. Estrin, “Medium access control with
coordinated adaptive sleeping for wireless sensor networks,” IEEE/ACM
Transactions on Networking, June 2004, Vol. 12, no. 3, pp. 493–506
[94] Tijs van Dam, Koen Langendoen, “An adaptive energy-efficient MAC
protocol for wireless sensor networks,” Proceedings of the 1
st
international conference on Embedded networked sensor systems
(SenSys), 2003, pp. 171–180
[95] S. M. Kamruzzaman “An Energy Efficient Multichannel MAC Protocol
for Cognitive Radio Ad Hoc Networks,” International Journal of
Communication Networks and Information Security (IJCNIS), Vol. 2, No.
2, August 2010, pp. 112-119
[96] Y. Zhou, “Energy-efficient protocols and topologies for sensor and
personal-area networks,” PhD thesis, Washington State University,
Pullman, WA, USA, 2007, ISBN: 978-0-549-19359-3
[97] Q. Fan, J. Fan, J. Li, X. Wang, “A Multi-hop Energy-efficient Sleeping
MAC Protocol based on TDMA Scheduling for Wireless Mesh Sensor
Networks,” Journal of Networks, Sep 2012, vol. 7, no. 9, pp. 1355–1361
[98] Xue Yang, N. F. Vaidya, “A Wakeup Scheme for Sensor Networks:
Achieving Balance between Energy Saving and End-to-end Delay,“
Proceedings of the 10
th
IEEE Real-Time and Embedded Technology and
Applications Symposium (RTAS), May 2004, pp. 19–26
[99] C. Schurgers, V. Tsiatsis, S. Ganeriwal, M. Srivastava, “Optimizing
Sensor Networks in the Energy-Latency-Density Design Space,” IEEE
Transactions on Mobile Computing, 2002, vol. 1, no. 1, pp. 70–80
[100] M. Dhanaraj, B. S. Manoj, C. S. R. Murthy, “A New Energy Efficient
Protocol for Minimizing Multi-Hop Latency in Wireless Sensor
Networks,” Proceedings of the 3
rd
IEEE International Conference on
Pervasive Computing and Communications (PerCom), Mar 2005, pp.
117–126
[101] S. Singh, C. S. Raghavendra, “PAMAS – Power aware multi-access
protocol with signaling for ad hoc networks,” ACM SIGCOMM Computer
Communication Review, July 1998, vol. 28, no. 3, pp. 5–26
[102] R. Wattenhofer, A. Zollinger, “XTC: A Practical Topology Control
Algorithm for Ad-Hoc Networks,” Proceedings of the 18
th
International
Parallel and Distributed Processing Symposium, Apr 2004
[103] Jang-Ping Sheu, Shin-Chih Tu, Chi-Hung Hsu, “Location-Free Topology
Control Protocol in Wireless Ad Hoc Networks,” IEEE Wireless
Computer Communications and Networking Conference, Mar 2007, pp.
66–71
[104] Kuei-Ping Shih, Chih-Yung Chang, Chien-Min Chou, Szu-Min Chen, “A
Power Saving MAC Protocol by Increasing Spatial Reuse for IEEE
802.11 Ad Hoc WLANs”, Proceedings of the 19
th
International
Conference on Advanced Information Networking and Applications
(AINA), Mar 2005, vol. 1, pp. 420–425
[105] Y. Al-Hazmi, H. de Meer, “Virtualization of 802.11 Interfaces for
Wireless Mesh Networks,” 8
th
International Conference on Wireless On-
Demand Network Systems and Services (WONS), Jan 2011. pp. 44–51
[106] W. R. Heinzelman, A. Chandrakasan, H. Balakrishnan, “Energy-Efficient
Communication Protocol for Wireless Microsensor Networks,“
Proceedings of the 33
rd
Annual Hawaii international Conference on
System Sciences (HICSS), Jan 2000
[107] H. O. Tan, I. Korpeoglu, “Power efficient data gathering and aggregation
in wireless sensor networks,” ACM SIGMOD Record, Dec 2003, vol. 32,
no. 4, pp. 66–71
[108] Lap-Fai Leung, Chi-ying Tsui , Hu, X.S., “Exploiting dynamic workload
variation in low energy preemptive task scheduling,” Proceedings of
Design, Automation and Test in Europe, ISBN: 0-7695-2288-2, 2005
Vol1 pp. 634-639
[109] F. O. Aron, A. Kurien, Y. Hamam, “Topology Control Algorithm for
Effective Power Efficiency and Throughput for Wireless Mesh
Networks,” 3
rd
International Conference on Broadband Communications,
Information Technology & Biomedical Applications, Nov 2008, pp. 89–96
[110] F. O. Aron, T. O. Olwal, A. Kurien, M. O. Odhiambo, “A Distributed
Topology Control Algorithm to Conserve Energy in Heterogeneous
Wireless Mesh Networks,” Proceedings of World Academy of Science,
Engineering and Technology, July 2008, vol. 30, pp. 530–536
[111] P. Santi, “Topology Control in Wireless Ad Hoc and Sensor Networks,”
ACM Computing Surveys (CSUR), June 2005, vol. 37, no. 2, pp. 164–194
[112] H. Coskun, I. Schieferdecker, Y. Al-Hazmi, “Virtual WLAN: Going
beyond Virtual Access Points,” Journal of Electronic Communications of
the EASST (ECEASST), Mar 2009, vol. 17, pp. 1–12
[113] A. Capone, F.Malandra, B. Sansò, “Energy Savings in Wireless Mesh
Networks in a Time-Variable Context,” Journal of Mobile Networks and
Applications, Apr 2012, vol. 17, no. 2, pp. 298–311
[114] S. Nedevschi, L. Popa, G. Iannaccone, S. Ratnasamy, D. Wetherall,
“Reducing Network Energy Consumption via Sleeping and Rate-
Adaptation,” Proceedings of the 5
th
USENIX Symposium on Networked
Systems Design and Implementation (NSDI), 2008, pp. 323–336
[115] S. Dey, H. Sarmah, S. Samantray, D. Divakar, S. S. Pathak, “Energy
Efficiency in Wireless Mesh Networks,” IEEE International Conference
on Computational Intelligence and Computing Research (ICCIC), Dec
2010, pp. 1–4
[116] Chuan-Chi Weng, Ching-Wen Chen, Po-Yueh Chen, Kuei-Chung Chang,
“Design of an energy-efficient cross-layer protocol for mobile ad hoc
networks,” IET Communications, May 2013, vol. 7, no. 3, pp. 217–228
[117] Shujuan Liu, Yuebin Bai, Mo Sha, Qingyuan Deng, Depei Qian, “CLEEP:
A Novel Cross-Layer Energy-Efficient Protocol for Wireless Sensor
Networks,” 4
th
International Conference on Wireless Communications,
Networking and Mobile Computing (WiCOM), Oct 2008, pp. 1–4
31
[118] S. Sarkar, K. Majumder, “A Survey on Power Aware Routing Protocols
for Mobile Ad-Hoc Network,” Proceedings of the International
Conference on Frontiers of Intelligent Computing: Theory and
Applications (FICTA), 2013
[119] CLE2aR2: Kuei-Ping Shih, Hung-Chang Chen, Chun-Chih Li and
Hsiang-I Chen, “A Cross-Layer Energy-Efficient and Reliable Routing
Protocol for Wireless Ad Hoc Networks,” Computers and
Communications, 2008. ISCC 2008. IEEE Symposium on, 436 – 441
[120] The carbon footprint of ICTs, Hopeton S Dunn,
http://akgul.bilkent.edu.tr/GISW/gisw2010_en.pdf, GISwatch, APC and
Hivos, 2010, ISBN 92-95049-96-9
[121] B. Shah, Kim Ki-Il, “A Survey on Three-Dimensional Wireless Ad Hoc
and Sensor Networks,” International Journal of Distributed Sensor
Networks, 2014
[122] L. Feng, “A Distributed Gateway Selection Algorithm for UAV
Networks,” IEEE Transactions On Emerging Topics In Computing,
December, 2014(Current Version March, 2015),pp. 22-33
[123] Y. Li, M. St-Hilaire, T. Kunz, “Enhancing the RGR Routing Protocol for
Unmanned Aeronautical Ad-hoc Networks,” Carleton University, Systems
and Computer Engineering, Technical Report SCE-12-01, August 2012
[124] S. Balndin et al (Eds.): New2AN/ruSMART 2014, LNCS 8638, Springer
International Publishing, 2014, pp. 245-252].
[125] K. Peters, A. Jabbar, K. Egemen, E.K. Cetinkaya. J.P.G. Sterbenz, “A
Geographical Routing Protocol for Highly-Dynamic Aeronautical
Networks,” IEEE Wireless Communications and Networking Conference
(WCNC), 2011, pp. 492-497.
[126] T. Camp, J. Bolen, B. Williams, L. Wilcox, W. Navidi, “Performance
comparison of two location based routing protocols for ad hoc networks,”
Proceedings of IEEE INFOCOM, 2002, pp. 1678–1687
[127] I. Turnadereli; I. Bekmezci, “Performance Comparison Of MANET
Routing Protocols For UAV Ad Hoc Networks,” International Symposium
on Computing in Science & Engineering, Oct 2013, p169
[128] S. Rosati S., K. Kru˙zelecki, L. Traynard, B. Rimoldi, “Speed-Aware
Routing for UAV Ad-Hoc Networks,” Workshop on Wireless and Control
for Unmanned Autonomous Vehicles, Globecom 2013
[129] J. Pojda, A. Wolff, M. Sbeiti, C. Wietfeld, “Performance Analysis of
Mesh Routing Protocols for UAV Swarming Applications,” 8th
International Symposium on Wireless Communication Systems, 2011
[130] Bose, P.; Morin, P.; Stojmenovic, I.; Urrutia, J. (1999). "Routing with
guaranteed delivery in ad hoc wireless networks". Proc. of the 3rd
international workshop on discrete algorithms and methods for mobile
computing and communications (DIALM '99). pp. 48–55]
[131] S.S. Lam, C. Qian, “Geographic Routing in d-dimensional Spaces with
Guaranteed Delivery and Low Stretch,” SIGMETRICS’11, 2011
[132] C. Liu, J. Wu, “Efficient Geometric Routing in Three Dimensional Ad
Hoc Networks,” IEEE INFOCOM, 2751 – 2755
[133] R. Flurry, R. Wattenhofer, “Randomized 3D Geographic Routing,” The
27
th
Conference on Computer Communications, INFOCOM 2008.
[134] J. Zhou, X. Jiaotong, Y.C.B. Leong, P.S. Sundaramoorthy, “Practical 3D
Geographic Routing for Wireless Sensor Networks,” Proceedings of the
8th ACM Conference on Embedded Networked Sensor Systems(SenSys
’10), 2010 pp. 337-350
[135] T. Spyropoulos, K. Psounis, C.S. Raghavendra, “Single-copy routing in
intermittently connected mobile networks,” Proceedings of the IEEE
Conference on Sensor and Ad Hoc Communications and Networks,”
SECON, 2004
[136] A. Vahdat, D. Becker, “Epidemic Routing for Partially-Connected Ad
Hoc Net- works,” Technical Report CS-2000-06, Duke University (July
2000).
[137] T. Spyropoulos, K. Psounis, C.S. Raghavendra, “Spray and wait: Efficient
routing in intermittently connected mobile networks,” Proceedings of
ACM SIGCOMM Workshop on Delay Tolerant Networking, WDTN,
2005
[138] I. Cardei, C. Liu, J. Wu, “Routing In Wireless Networks With Intermittent
Connectivity” Encyclopaedia of Wireless and Mobile Communications,
2013
[139] R. Shirani, M. St-Hilaire, T. Kunz, Y. Zhou, J. Li, L. Lamont, “Combined
Reactive-Geographic routing for Unmanned Aeronautical Ad-hoc
Networks,” 8th International Conference on Wireless Communications
and Mobile Computing (IWCMC), 2012
[140] X. Li, L. Ci, B. Cheng, C. Tian, M. Yang, R. Wang, F. Xiao (Eds) , “Ant
Colony Based Routing Strategy in UAV Delay Tolerant Networks,”:
CWSN 2012, CCIS 334, pp. 191–203
[141] A. Lindgren, A. Doria, O. Chelen, “Probabilistic routing in intermittently
connected networks.” SIGMOBILE: Mobile Computing Communications
Review, volume 7, pages 19–20. ACM Press, July 2003
[142] E. Ashraf, A.F. Hossam, S. Hassanein,“Routing schemes for DTN-An
Applications Perspective,” Telecommunications Research Lab (TRL)
School of Computing Queen's University Kingston, Ontario, Canada
Technical Report 2012
[143] J. Widmer, J.Y. LeBoudec, “Network Coding for Efficient
Communication in Extreme Networks,” WDTN, 2005
[144] V. Visvan, A. Beevi, N. Nasseema, “Efficient Media Independent
Handover Scheme for Mission-Critical Management,” IOSR Journal of
Computer Engineering, Volume 7, Issue 5 (Nov-Dec. 2012), PP 10-14
[145] M. Ohlege, G. Xie, J.H. Gibson, “Extending UAV Video
Dissemination via Seamless Handover: A Proof of Concept Evaluation of
the IEEE 802.21 Standard,” 46th Hawaii International Conference on
System Sciences, 2013
[146] http://en.wikipedia.org/wiki/AAA_battery
[147] R. Purta, S. Nagrecha, G. Madey, “Multi-hop Communications in a
Swarm of UAVs,” Proceedings of the Symposium on Agent Directed
Simulation (ADS'13/SpringSim), 2013).
[148] N. Shi, X. Luo, “A Novel Cluster-Based Location-Aided Routing
Protocol for UAV Fleet Networks,” International Journal of Digital
Content Technology and its Applications (JDCTA) Volume 6, Number18,
October 2012
[149] H.C. Christmann, E.N. Johnson, “Design and Implementation of a Self-
configuring Ad-hoc Network for Unmanned Aerial Systems,” AIAA
Infotech@Aerospace Conference and Exhibit, Rohnert Park, CA, October,
2007.
[150] S. Cespedes, N. Lu, X. Shen, “VIP-WAVE: On the Feasibility of IP
Communications in 802.11p Vehicular Networks,” IEEE Transactions On
Intelligent Transportation Systems, Vol 14, Issue 1, 2013, PP 82-97
[151] D. Halperin, B. Greenstein, A. Sheth, D. Wetherall “Demystifying
802.11n Power Consumption,” Proceedings of the international
conference on Power aware computing and systems (HotPower’10), 2010
Lav Gupta is a senior member of
IEEE. He received BS degree from Indian
Institute of Technology, Roorkee, India in
1978 and MS degree from Indian Institute
of Technology, Kanpur, India in 1980. He
is currently pursuing PhD degree in
Computer Science & Engineering at
Washington University in St Louis,
Missouri, USA.
He has worked for about 15 years in the area of
telecommunications planning, deployment and regulation. With
the sector regulatory authority he worked on technology and
regulation of next generation networks. He has also worked as
senior teaching faculty of Computer Science and Access
Network Planning for a number of years in telecommunications
academies. He is the author of one book, 5 articles and has been
a speaker at many international seminars.
He was recipient of best software award from Computer
Society of India in 1982 and best faculty award at Etisalat
Academy, UAE in 1998.
Raj Jain is a Fellow of IEEE, a Fellow
of ACM, a Fellow of AAAS. He received
BS degree in Electrical Engineering from
APS University in Rewa, India in 1972 and
MS in Computer Science & Controls from
32
IISc, Bangalore, India in 1974 and the Ph.D. degree in Applied
Math/Computer Science from Harvard University in 1978.
Dr. Jain is currently a Professor of Computer Science &
Engineering at Washington University in St. Louis. Previously,
he was one of the Co-founders of Nayna Networks, Inc - a next
generation telecommunications systems company in San Jose,
CA. He was a Senior Consulting Engineer at Digital Equipment
Corporation in Littleton, Mass and then a professor of
Computer and Information Sciences at Ohio State University in
Columbus, Ohio. He has 14 patents and has written or edited 12
books, 16 book chapters, 65+ journal and magazine papers, and
10e5+ conference papers.
He is a winner of ACM SIGCOMM Test of Time award,
CDAC-ACCS Foundation Award 2009, and ranks among the
top 100 in CiteseerX's list of Most Cited Authors in Computer
Science.
Gabor Vaszkun received his BS and
MS degrees in Computer Science from
Budapest University of Technology &
Economics, Budapest, Hungary in 2012
and 2014 respectively.
Mr. Vaszkun is currently working with
Ericsson Hungary. During 2014 he was a
Research Scholar in the Department of Computer Science &
Engineering at Washington University in St. Louis, Missouri,
USA. He has interned with Ericsson Research for 3 months
during his Masters program. He is the author of two published
conference papers. He has also participated in a number of
Poster Sessions and Demonstrations in national and
International Conferences.
He received Rosztoczy scholarship for the year 2014 to
pursue his research in USA. He is also a recipient of the
WorldQuant scholarship awarded to outstanding students to
pursue higher education in the fields of science and quantitative
studies.