Ad Hoc Networks 106 (2020) 102242

Contents lists available at ScienceDirect

Ad Hoc Networks
journal homepage: www.elsevier.com/locate/adhoc

Drone assisted Flying Ad-Hoc Networks: Mobility and Service oriented

modeling using Neuro-fuzzy
Kirshna Kumar a, Sushil Kumar a,∗, Omprakash Kaiwartya b, Pankaj Kumar Kashyap a,
Jaime Lloret c, Houbing Song d
a
School of Computer and Systems Sciences, Jawaharlal Nehru University, New Delhi, India
b
School of Science and Technology, Nottingham Trent University, Clifton Campus, NG11 8NS, Nottingham, UK
c
Integrated Management Coastal Research Institue, Universitat Politecnica de Valencia, Spain
d
Embry-Riddle Aeronautical University, USA

a r t i c l e i n f o a b s t r a c t

Article history: Flying ad-hoc networks enable vast of IoT services while maintaining communication among the ground
Received 9 February 2020 systems and flying drones. The domain research is focusing on flying networks assisted data centric IoT
Revised 29 May 2020
applications while integrating the benefits and services of aerial objects such as unmanned aerial vehicle
Accepted 2 June 2020
and drones. Considering the growing market significance of drone centric flying networks, quality of ser-
Available online 6 June 2020
vice provisioning is one of the most leading research themes in flying ad-hoc networks. The related liter-
Keywords: ature majorly relies on centralized base station monitored communications. Towards this end, this paper
Flying Ad-Hoc networks proposes a drone assisted distributed routing framework focusing on quality of service provision in IoT
Internet of Things environments (D-IoT). The aerial drone mobility and parameters are modeled probabilistically focusing on
Quality of Service highly dynamic flying ad-hoc networks environments. These drone centric models are utilized to develop
Routing a complete distributed routing framework. Neuro-fuzzy interference system has been employed to as-
Aerial drone
sist in reliable and efficient route selection. A comparative performance evaluation attests the benefits of
the proposed drone assisted routing framework. It is evident that D-IoT outperforms the state-of-the-art
techniques in terms of number of network performance metrics in flying ad-hoc networks environments.
© 2020 Elsevier B.V. All rights reserved.

1. Introduction role in smart agriculture [8]. The application of Drone assisted Fly-
ing Ad-Hoc networks makes current agriculture smarter by over-
Internet of Things (IoT) is one of the leading research do- coming the various challenges of farmers such as sudden climate
mains in recent years due to the growing applicability in differ- changes, disease and pest detection of crops and the presence of
ent new areas [1,2]. The novel domains such as smart healthcare, parasites [9,10].
smart home/city, intelligent transportation, Unmanned Aerial Vehi- Fig. 1 illustrates the different aerial communication systems:
cles (UAVs) simulation [3], pollution monitoring, disaster manage- satellite communication, air-ground and air-air communication. It
ment, industrial IoT, smart agriculture have emerged as prominent is highlighted that drone controller is a specific drone with higher
themes to revolutionize IoT in day to day life [4-6]. In aeronautical computing and communication capability which works as coor-
applications, for transmitting and viewing data immediately, Inter- dinator in aerial adhoc network environment. In satellite-based
net is utilized to connect actuators and sensors inside the aerial communication system, aerial object such as drone accesses In-
objects. After the trip completion, data related to flight would be ternet and remains in contact with ground while utilizing satel-
tracked in real-time with the usage of IoT devices and technolo- lite as a relay node to cover remote, ocean or polar areas [11].
gies in place for downloading data from sensors [7]. Currently, us- But these satellite-based centralized aerial communication systems
age of sensors and actuators through novel technologies such as have higher operation and maintenance cost. While comprising
drone assisted Flying Ad-Hoc networks (FANET) plays a prominent multi hop ad hoc networking among the drones, extension of net-
work architecture is known as Flying ad hoc network [12]. While
permitting and maintaining communication between drone and
∗
Corresponding author.
ground, over the region without communication infrastructure, fly-
E-mail addresses: kirshn44_scs@jnu.ac.in (K. Kumar), skdohare@mail.jnu.ac.in (S. ing ad hoc network can be utilized as a complementary communi-
Kumar), omprakash.kaiwartya@ntu.ac.uk (O. Kaiwartya), pankaj76_scs@jnu.ac (P.K. cation system. Flying ad hoc network facilitates Internet reachabil-
Kashyap), jlloret@dcom.upv.es (J. Lloret), SONGH4@erau.edu (H. Song).

https://doi.org/10.1016/j.adhoc.2020.102242
1570-8705/© 2020 Elsevier B.V. All rights reserved.
2 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Fig. 1. Aerial communication systems.

ity to these drones while traversing via these areas, with no usage pected link availability period, residual route load capacity
of costly satellite links and high delay [13]. Therefore, in this paper, and route delay.
we focus only on flying ad hoc network and ground communica- (3) Neuro-fuzzy interference system has been employed to
tion without involvement of satellite system. However, providing jointly combine three important QoS provisioning parame-
reliable and stable communication among drone and ground sta- ters to assist in reliable and efficient route selection.
tions in flying ad hoc network assisted flight communication sys- (4) A drone assisted distributed routing framework is developed
tem is a great challenge [14]. based on the drone mobility model and QoS parameters.
In high speed mobility of drones in flying ad hoc networks is (5) The proposed communication framework is tested to com-
the key challenge in ad-hoc aerial communication. It causes vari- paratively evaluate the performance with the state-of-the-
ation in network topology in frequent and unpredictable manner, art protocols considering metrics related to flying ad-hoc
which results in link breakage in multi hop communication [15]. networks environments.
Consequently, the performance of flying ad hoc network degrades The remainder of this paper is structured as follows. Section II
in terms of reliability. The requirement of drone networks having introduces related literatures of recent QoS- non provisioning and
efficient quality of service (QoS) parameters motivates to design QoS- aware routing techniques for flying ad hoc networks. In Sec-
a drone assisted distributed routing framework focusing on qual- tion III, the details of the proposed drone assisted distributed rout-
ity of service provision in drone assisted IoT environments (D-IoT). ing framework (D-IoT) is presented. Section IV discusses the im-
Towards this end, in this paper we propose a drone assisted dis- plementation and analysis of simulation results. Conclusion is pre-
tributed routing framework focusing on quality of service provi- sented in Section V.
sion to maximize network performance inside D-IoT environment.
Additionally, we compute route availability factor, residual route 2. Related work
load capacity and route delay as the route selection metric for the
purpose of QoS provisioning while selecting the optimized route. In this section, related literature on routing in flying ad-hoc net-
Neuro-fuzzy inference system has been implemented to find the works has been reviewed while focusing on QoS non provisioning
aggregated output based on QoS metrics route availability factor, routing and QoS aware routing.
residual route load capacity and route delay. The advantage of us-
ing hybrid structure; neural network with fuzzy logic counterbal- 2.1. QoS non- provisioning routing
ance each other’s such as generalization to environment is done
by neural network learning procedure (change the inbuilt rule of Two novel stability driven clustering schemes have been pro-
fuzzy logic) and error is minimized. Whereas simplification of out- posed while establishing stable clusters for highly mobile ad hoc
put generated by fuzzy inference system in quick time [43]. Hence, networks comprising ships, aircraft, cars and trains as mobile
a route having stability and properly balance traffic can be selected nodes [18]. For the scenarios with unknown position information
between drone and ground stations. Further, a best advertisement of mobile nodes, first scheme is utilized and for the scenarios with
forwarding (BADF) technique is utilized to reduce the overhead re- known position information (via GPS), second scheme is utilized.
lated to advertisement flooding generated during route selection This scheme lacks data reliability. An automatic dependent surveil-
process. Then, D-IoT and state-of-the-art protocols are evaluated lance broadcast system based geographical routing has been sug-
and compared. The key contributions of the paper can be summa- gested while utilizing aircraft position and velocity to remove bea-
rized as follows: coning of traditional routing [19]. In this scheme, next hop has
been selected based on aircraft velocity metric while adaptively
coping with highly dynamic aircraft and network topology. This
(1) A network model for drone assisted IoT environment is pre- scheme does not focus on providing optimized load capacity. Re-
sented focusing the topological aspects of aerial drones and active greedy reactive routing has been proposed for highly mobile
its mobility in flying adhoc networks. and density variable unmanned aerial vehicle communication sys-
(2) To optimize Drone network centric QoS provisioning param- tems while combining the characteristics of reactive routing tech-
eters are derived focusing on relative velocity of drones, ex- niques with geographical routing techniques [20]. In this scheme,
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 3

velocity vector-based mobility prediction technique has been uti- A routing and scheduling technique based on hybrid genetic ap-
lized to predict the aircraft location and two various scoped flood- proach has been proposed while supporting the communication
ing techniques have been used while reducing message overhead. among ground vehicle and multiple drones for efficient delivery
This scheme lacks the Quality of service metrics such as delay and of parcels [29]. Further hybrid genetic approach consists of pop-
link lifetime. ulation initialization, low visit cost crossover algorithm and three
An unmanned aerial vehicle-based communication system hierarchical education algorithms for fair distribution inside pop-
while providing connectivity and deployment modules for emer- ulation while avoiding premature convergence and minimizing the
gency disaster recovery has been proposed [21]. This system com- total delay. But this technique does not provide optimized link life-
prises three prominent sub system such as navigation system, time. Two multi-trip vehicle routing problems have been suggested
communication sub system and schemes for formation manage- for drone delivery to minimize the delivery time related to bud-
ment. The parameters such as link availability, jitter, throughput get constraint [30]. A model for energy consumption has been de-
and packet loss have not been considered in the communication rived and validated while considering payload and battery weight.
system. A two-echelon ground vehicle and its mounted drone co- The other QoS parameters such as link availability, jitter etc. have
operative routing technique (2E-GUCRP) has been proposed for in- not been considered. A vehicle assisted multi-drone scheduling
telligence, surveillance, and reconnaissance (ISR) missions while and routing technique has been proposed while optimizing anchor
minimizing the overall mission time to meet the operational con- point selection, tour assignment and route planning in each itera-
straints [22]. QoS constraints are not utilized in this routing tech- tion [31]. This technique minimizes total finish time but does not
nique. A glowworm swarm optimization and dragonfly approach- consider residual route load capacity and route availability. A traf-
based hybrid self-organized clustering protocol has been proposed fic load balancing technique has been suggested to minimize la-
for drone assisted cognitive IoT networks [23]. After introducing tency for drone-based fog network inside IoT [32]. Two algorithms:
cluster formation, management and maintenance algorithm, route heuristic and user association have been utilized sequentially, to
selection function-based routing technique has been suggested for solve the traffic load balancing problem.
optimized route selection in drone assisted IoT. This technique re- A motion driven packet forwarding scheme has been suggested
duces energy consumption but does not focus on latency and route in micro aerial vehicle networks while utilizing two predictive
connectivity. heuristics to integrate delay tolerant routing and location aware
An Unmanned Aerial Vehicles based emergency rescue frame- end-to-end routing [33]. This technique focuses on link connectiv-
work has been designed for scanning large areas searching through ity and route delay but does not consider load balancing and en-
Smartphone signal communication of missing or injured persons ergy efficiency. A multi-UAV routing technique has been proposed
[24]. This framework lacks mostly QoS parameters and perfor- to solve the multi-UAV coverage task to launch the UAVs while uti-
mance in highly dynamic aerial environment. A new protocol for lizing minimum number of vehicles with minimum delay [34]. This
providing smart communication has been suggested while analyz- technique reduces the mission time but lacks the other QoS matri-
ing sequential patterns of messages with the definition of various ces. A jamming-resilient multipath (JaRM) routing technique has
fold messages inside UAVs [25]. This protocol reduces energy con- been suggested while considering the three major routing matri-
sumption but does not consider QoS parameters such as delay and ces: link quality, traffic load and spatial distance in drone based
route stability. A swarm intelligence-based localization and clus- flying ad hoc networks [35]. Enhanced link quality and light traffic
tering techniques have been proposed to facilitate communication load are positive factors of this technique, but this technique does
in emergency inside UAV enabled IoT networks [26]. Firstly pro- not minimize the route latency. A deep reinforcement learning
posed, swarm intelligence-based localization is a particle swarm based solution for 3D continuous movement control of multiple
optimization (PSO) based three-dimensional technique while utiliz- drones has been suggested to maximize the energy efficiency and
ing bounding box algorithm to exploit in 3D search space. Secondly connectivity of drone network [36]. Furthermore, based on cover-
proposed, a swarm intelligence-based clustering is PSO based en- age fairness, QoS requirements and energy utilization inside drone
ergy efficient technique which derive fitness function for residual networks, a reward function has been formulated. The penalty for
energy, geographic location, inter cluster and intra cluster distance. disconnected drone networks has been introduced while reducing
This technique minimizes computational cost, energy consumption, the reward function value drastically. This technique only works in
but does not improve link stability. centralized ad-hoc network not in decentralized network [37-40].
A bio-inspired technique based on swarm intelligence has been
2.2. QoS- aware routing proposed to control the network topology and to support multi-
media traffic for emergency inside FANET [41]. This technique does
Link availability estimation-based routing has been proposed minimize the route delay.
while utilizing the link availability parameter for the selection and
updates of a route [27]. Firstly, semi- Markov mobility model has 3. Drone qa assisted distributed routing framework for IoT
been presented to imitate the behavior of airliners, then link avail- environment (D-IoT)
ability period, pdf of relative speed between two aircrafts and ex-
pected link lifetime have been used to select the reliable route. In In this section, the proposed drone assisted distributed routing
this scheme, relative speed of the derivation for link availability framework focusing on quality of service provision is presented
metric and pdf of the link lifetime have been utilized to select re- in detail. Firstly, a network model comprising mobility model of
liable route. Still metric of load balance has not been comprised. drone in flying adhoc networks is discussed. Secondly, QoS met-
A joint internet gateway allocation, scheduling and routing scheme rics: route availability factor, residual route load capacity and route
has been suggested to minimize the average packet delay in mo- delay are formulated. Hence, route selection approach based on
bile aeronautical ad-hoc networks [28]. Inside it, a mathematical QoS metrics, and broadcast optimization technique have been de-
programming scheme has been proposed, while comprising two scribed. It is highlighted that the mathematical modeling pre-
steps: weighted hop count minimization for scheduling and aver- sented as network model and metrics derivations for drone centric
age delay reduction for routing. Further a genetic algorithm has network environment is further realized as a complete information
also been formulated to reduce computational complexity in large routing framework utilizing neuro fuzzy technique. Moreover, it is
mobile network. But this scheme does not provide optimized link highlighted that this proposal focuses on drone centric dynamic
lifetime. network environment where parameters such as link quality, route
4 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Fig. 2. Analysis for the relative velocity of two drones.

availability, delay, residual energy have significant impact on net-

work performance modelling. Therefore, we have focused in-depth
mathematical modelling of the parameters utilizing highly scien-
tific probabilistic modelling approach. Fig. 3. Link availability period between drone nodes M and N.

3.1. Drone network mobility model

where fvm (vm ), fvn (vn ), and fα (α ) are the probability density func-
The network model consists of three components: drone, tions of vm , vn and α respectively and vmin max and vmin and
m , vm n
max
vn are minimum and maximum velocities of two drone nodes
ground stations and drone controller station. Here ground sta-
tions work as Internet gateways (IGs). This scenario concerns only M and N. For simplicity, we assume vm = vn =v. The joint pdf
about the communication between drone and IGs, not satellite fvm ,vn ,α (vm , vn , α ) can be written as
based communication. For simplification, some assumptions have fvr (vr ) = fv2 (v ) fα (α ) (3)
been considered. The distribution of all drones is done in a plane.
Physical layer, transmission power and transmission range re- Now probability density function fα (α ) can be calculated as
lated to all drones are uniform. Automatic dependent surveillance- √ 
broadcast(ADS-B) system is being equipped for all drones to ac- fα (α ) = k1 2v 1 − cos α (4)
quire real-time state vector such as position, velocity, ID and other √ √
Where k1 2v ∫π 0 1 − cos α dα = 1, on solving, we get k1 = 1/4v.
information.
Similarly, the probability density function fv (v) can be calculated as
On the basis of airliner’s mobile trace in the sky, we can cat-
egorize a drone node movement in to five phases: acceleration √ 
fv (v ) = k2 2 1 − cos α v (5)
phase, steady climb, middle smooth, steady down and decelera-
tion. In the acceleration phase, velocity of drone increases until √ √ max 2
Where k2 2 1 − cos α ∫(v )2 vdv = 1, we get k2 =
the target velocity v∝ . Drone selects targeted horizontal direction (vmin )
√
∅∝ in the range [0, 2π ]. In the steady climb phase, drone climbs 2/√ 2
1 − cos α ( (vmax ) − (vmin ) ).
2
in the target vertical direction ∅∝ in the range [0, π /2] and moves
Therefore, the probability density functionfv,α (v, α ) can be ex-
with constant velocity v∝ . Here v∝ is randomly considered in the
pressed as
range 20-50 km/hr and is considered as standard velocity in vari-
√
ous drone models. During the middle smooth phase, movement of v2 2 − 2 cos α
fvr (vr ) = fv,α (v, α ) = 
drone is steady and smooth according to Gauss Markov model [42].  2  2 (6)
Further in steady down phase, velocity of drone is equal to v∝ . The (vmax )2 − vmin
drone selects horizontal direction equals to ∅∝ and vertical direc-
tion in the range [π /2, π ]. In the end, in deceleration phase, drone It is highlighted that the mathematical derivations presented
uniformly decreases the velocity in one direction until it stops. In are highly significant for in-depth scientific basis for each of the
the starting, in acceleration phase, drone takes 5 min. for takeoff conceptual idea in the manuscript. This mathematical basis has
and in the end in deceleration phase, it also takes 5 min. for land- more value in highly dynamic drone framework implementation
ing. environment, where network communication probability depends
on metrics such as link availability, residual energy, delay, etc.
3.2. Route availability factor Therefore, the modeling derivation of these parameters mathemat-
ically verifies the validity of our conceptual framework for drones.
Route availability factor between the non-neighboring drones, is Further, to calculate the probability density function fv,α (v, α ) of
defined as the minimum link availability factor between intermedi- relative velocity (vr ) between two drone nodes, we need to cal-
ate nodes in the present route. Link availability factor is the mea- culate the probability density function fα (α ) and fv (v) separately.
sure of link reliability based on the expected link availability pe- Therefore, in Eq. (4), probability density function fα (α ) has been
riod. Assuming, drone nodes M and N are two intermediate nodes calculated. Here, α is the angle between two drone nodes and uni-
and these drones lie in the transmission range of each other. formly distributed between [0, π ]. Therefore, following probability
density function fα (α ) is calculated as given in Eq. (4) by following
3.2.1. Probability density function of relative velocity uniformly distributed variable formulation.
Let vm and vn are velocity vectors of two drone nodes M and N
and vr is relative velocity between them. According to Fig. 2, α is 3.2.2. Expected link availability period
the angle between two nodes and uniformly distributed between Here first derivation for cumulative distribution function for
[0, π ]. Let vm , vn and vr are modulus of vectors vm , vn and vr . link availability period between neighboring drone nodes is esti-
According to cosine theorem, mated. Then expected link availability period is formulated. As-

vr = v2m + v2n − 2v2m v2n cos α (1) suming starting distance vector between drone nodes M and N lies
in y-axis direction and d0 is starting distance value. The starting
Since vm , vn and α are independent, therefore joint probability relative velocity vector between nodes M and N is shown in Fig. 3
density function fvm ,vn ,α (vm , vn , α ) can be expressed as We assume angle between initial distance vector and initial rel-
f vr ( vr ) = f vm ( vm ) f vn ( vn ) f α ( α ) (2) ative velocity vector is ∅ distributed in uniform manner ranging
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 5

from 0 to π . Link distance of two neighboring drone nodes, M and

N is formulated as follows

dm = d0 cos ∅ + K 2 − d02 sin2 ∅ (7)
Then PDF of link distance of two neighboring drone nodes, M
and N is formulated as follows
f dm ( dm ) = f d0 ( d0 ) f ∅ ( ∅ ) (8)
PDF of ∅ is expressed as
Fig. 4. Route availability factor calculation.
1
f ∅ (∅ ) = (9)
π
PDF of d0 can be calculated as
  
fd0 (d0 ) = h d0 cos ∅ + K 2 − d02 sin2 ∅ (10)

Where, ∫ddmax h(d0 cos ∅ + K 2 − d02 sin2 ∅)dd0 = 1
min
6dmax sin2 ∅
On solving, we get h= 3 −
2 cos∅−2 (k2 −d 2 sin2 ∅ ) /2
3dmax max Fig. 5. Residual route load capacity calculation.
6dmin sin2 ∅
3
2 cos∅−2 (k2 −d 2 sin2 ∅ ) /2
3dmin min
Then PDF fdm (dm ) can be expressed as

d0 cos ∅ + K 2 − d02 sin2 ∅
f dm ( dm ) =
π
6dmax sin2 ∅
×(  3/2
2 cos∅ − 2 k2 − d 2 sin2 ∅
3dmax max
Fig. 6. Leaky bucket strategy.
6dmin sin2 ∅
−  3/2 (11)
2 cos∅ − 2 k2 − d 2 sin2 ∅
3dmin min 3.3. Residual route load capacity
Link availability period t between nodes M and N is expressed
In In this section, residual load capacity of route between two
as
nodes is formulated. Residual route load capacity is defined as the
dm minimum residual load capacity amongst all node’s residual load
t= (12)
vr capacity along the route. Assuming, ck is the residual load capacity
Using Eqs. (7) to (12), PDF of link availability period between for drone node k. Hence ck is expressed as
nodes M and N is is expressed as 
m
 vmax ck = δ − ωk lk (18)
f (t ) =
T
0 vr fdm vr (vr t, vr )dvr k=1
 vmax  √
v2 2−2 cos α (13) Where, δ is the maximum load capacity for drone node k, while
= 0 vr f dm ( dm )  2 d vr
dm=v t r (vmax )2 − (vmin )
2
lk and ωk are the average packet size related to traffic and average
packet arrival rate of m sources, respectively. Let Ci be the residual
Then expected link availability period is estimated as load capacity for route i, the Ci is formulated as
∞  
E (T ) = t f T (t )dt (14) C i = min ck (19)
0
In this protocol, route having minimum residual load capacity
3.2.3. Route availability factor calculation is prefer in route finding process. Fig. 5 shows the calculation pro-
Link availability period between nodes M and N is expressed as cess for residual load capacity for route AD, where residual load
follows from Fig. 3 capacity of node B having minimum residual load capacity equals
 to 3 is the residual load capacity of route AD.
MN
d0 cos ∅ + K 2 − d02 sin2 ∅
T = (15)
v 3.4. Route delay
Link availability factor between nodes M and N is formulated as
 Route delay is defined as the time required to send a data
MN T MN packet from source drone node to destination drone node. Usu-
L = min ,1 (16)
E (T ) ally route delay mainly comprises queuing delay, propagation delay
and transmission delay based on scheduling techniques, traffic con-
Let Li is the route availability factor of route i, then Li is esti- trol schemes of nodes, residual link bandwidth, processing power
mated as of ports and traffic characteristics. Here leaky bucket control strat-
  egy as illustrates in Fig. 6 is utilized to control communication vol-
Li = min LMN (17)
ume of drones. Assuming λ is the bucket capacity, μin is the input
According to Fig. 4, link availability factor which is minimum flow rate and μout is the service rate. As service rate at each drone
in this route has been considered as the route availability factor of varies, so highest data flow of link is dependent on the drone node
the route AD. That is to say which has least service rate, therefore
 
LAD = LBC = 0.7. μout = min μ1out , μ2out , μ3out , . . . . . . , μnout
6 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Fig. 7. NFIS with 3-input, 27 rules and one output.

If d is the queuing delay then according to the leaky bucket Each of these three input parameters have three member-
strategy, ship functions, according to Takagi- Sugeno fuzzy inference model
that contains 27 rules. NFIS consists of five layers architecture;
λ + μin d < μout d (20)
Fuzzy layer, T-norm layer, normalized layer, de-fuzzy layer and
hence aggregated layer. The first fuzzy layer (as known as member-
λ ship/antecedent layer) and fourth de-fuzzy layer (consequent layer)
d= (21)
μout − μin are adaptive in nature because they are updated according to re-
sults obtained and rest of the layers are non-adaptive in nature.
For the links, λ = ρ − nSmax where ρ is sudden traffic depended
The linguistic variables for three input parameters are given
on the network and Smax represents maximum packet size, then
as follows: route availability factor (L) = {below, good, top}
queuing delay is expressed as
and is denoted by {L1 ,L2 ,L3 }, residual route load capacity
ρ − nSmax (C) = {min, avg, max} that is denoted by {C1 ,C2 ,C3 }, route delay
d= (22)
μout − μin (D) = {low, medium, high} as {D1 ,D2 ,D3 }and output single metric
Let Di is the total delay of route i, and Bj and pj are bandwidth (Pk ) = {weakest, weaker, weak, medium, strong, stronger, strongest}
and propagation delay of link j respectively. The route delay can be as {P1 ,P2 ,P3 ,P4 ,P5 ,P6 ,P7 }. The first layer’s membership nodes follow
expressed as the rules influenced by If-Then rules as shown in table -2. The
antecedent parts of rules in the table-1 represent the input fuzzy
ρ − nSmax  n
Smax 
n
subspace and consequent part of rule in the table shows the out-
Di = + + pj (23)
μout − μin j=1 B j j=1
put inside the fuzzy subspace. We developed for three input pa-
rameters with three linguistic variables (33 ) 27 If-Then rules for the
3.5. Single metric proposed NFIS architecture governed by Takagi-Sugeno fuzzy infer-
ence system. The rules can be expressed as
Initially, In D-IoT, all the QoS metrics: route availability factor, Rule 1 = If L is L1 , C is C1 and D is D1 Then P1 = q1 L + r1 C + s1 D
residual route load capacity, and route delay are jointly considered + t1
for the purpose of finding the optimized route. Let Pk is consid- Rule 2 = If L is L1 , C is C1 and D is D2 Then P2 = q2 L + r2 C +
ered as the single metric for route k. The metric Pk is evaluated s2 D + t2
by employing a neuro-fuzzy inference system (NFIS). It is far bet- Rule 3 = If L is L1 , C is C1 and D is D3 Then P3 = q3 L + r3 C +
ter than fuzzy logic inference system because of unlike another s3 D + t3
artificial neural network, NFIS have higher capability to adapt an Rule 4 = If L is L1 , C is C2 and D is D1 Then P4 = q4 L + r4 C +
environment’s requirement in the learning process and adjust the s4 D + t4
weight of membership function of fuzzy logic inference system and .
reduces the error rate in determining the rules in fuzzy logic [43]. .
It is a feed-forward adaptive neural network which uses supervised Rule 25 = If L is L3 ,C is C3 and D is D1 Then P25 = q25 L + r25 C
learning algorithm for learning process. NFIS follows the learning + s25 D + t25
process of Takagi-Sugeno fuzzy inference system [44]. The basic Rule 26 = If L is L3 ,C is C3 and D is D2 Then P26 = q26 L + r26 C
architecture of NFIS with three input parameter route availability + s26 D + t26
factor (L), residual route load capacity (C), route delay (D) and one Rule 27 = If L is L3 ,C is C3 and D is D3 Then P27 = q27 L + r27 C
output single metric (P) are shown in Fig. 7. + s27 D + t27
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 7

Where L1 , C1 , D1 are membership function of input parameter (5) Aggregated Output layer- Non-adaptive nature of single
antecedent (If) part, while q1 , r1 , s1 and t1 are linear parameters of node is used to estimate the output, which measures the
consequent (then) part of Takagi-Sugeno model. The operation of overall system performance. The output is the summation of
NFIS to select single metric output Pk describe by layer wise as all the incoming signals to this layer, labeled as  inside the
follows. circle to represent the aggregated output.


(1) Fuzzy Layer- the nodes in this layer are represented by O5α = P k = Tnα Pα = α Tα f α (29)
square, which are adaptable in nature during backward pass. α α Tα
Each node resembles to membership function of input pa-
A neuro-fuzzy selection algorithm (Algorithm 1) is presented
rameters. The output of this layer is degree of membership
to describe the process of NFIS. NFIS uses hybrid learning algo-
govern by input membership function in the range of 0 and
rithm based on gradient descent and least mean square to train the
1. The membership function can be triangular, trapezoidal,
membership function of input parameter worked on two passes:
Gaussian, and generalized bell membership function. In this
forward pass and backward pass. The first layer and fourth layer
work, we considered Gaussian (Eq. (24)) and generalized bell
node are updated over time. In forward pass (training dataset),
membership function (Eq. 25).
the input signals (premise parameter {L, D, C}) are fixed in na-
  
L − zα 2 ture and propagated from first layer (as step-6) till to fourth layer
μLα (L ) = exp − (24) (step 11) of proposed Algorithm 1. In the step -7 Gaussian and
2xα
bell-shaped membership function for each input parameter {L, D,
1 C} are obtained using Eq. (24) and Eq. (25). After that in step -8
μLα (L ) =  2y (25)
1 +  m−zα 
xα
using if-then rules and minimum operator AND is applied to cus-
tomize the firing strength (it defined how much powerful the sig-
The output of first layer is given by nal) of each input parameter. Further, firing strength of each in-
put node regarding membership function is normalized with re-
O1,α = μLα (L ), α = 1, 2, 3
spect to the total firing strength. The output of fourth layer is ob-
tained as aggregated output Pk using Eq. (29) in step-11 term as
O1,α = μCα (C ), α = 1, 2, 3 consequent parameter. The obtained output is compared with ac-
tual output and error is recorded. The primary goal of the ANFIS
algorithm is to minimized mean square error (|obtained output −
O1,α = μDα (D ), α = 1, 2, 3 −actual ouput|2 ) recursively. While in the backward pass, error oc-
Where μMα , μNα and μOα are membership functions of curred in forward pass is sent back to input (first) layer and at the
adaptive node L, C and D respectively and xα , yα and zα are same time membership function of input premises are updated us-
premises parameters of membership functions that are re- ing learning process of gradient descent method. The hybrid learn-
sponsible for customize the shape of membership func- ing process (combination of forward pass and backward pass) of
tions. The membership function O1,α represents the degree one level is known as epoch. The algorithm runs for until conver-
to which L satisfies the input parameter Lα . gence (error between actual output and calculated output is in-
(2) T-Norm Layer- this layer determines the firing strength of finitesimal small) or maximum number of epochs (Emax ).
each rule associated with input signals. All the nodes in this
layer are non-adaptive in nature and are depicted by circle Algorithm 1 Neuro-fuzzy Selection Algorithm (NFSA).
with labeled π . The output of T-norm (rule) layer evaluated Input: L, D, C and Emax ;
as multiplying all the incoming signals to node and deliv- Process:
ered output to the next layer nodes. The T-Norm layer ap- 1. For E=1 to Emax .
plies generic AND operator to multiply all the input signals 2. Input the nonlinear premises into first layer of Takagi- Sugeno
inference engine.
to evaluate the firing strength of rules and generates output
3. Produce the parameter Membership function μLα (L) for each node
O2α (Tα ) as follows. according to Eq. (24) and (25) in adaptive Fuzzy layer.

O2α = Tα = μLα (L )∗μCα (C )∗μDα (D ), α = 1, 2, 3

4. Customize the firing strength of each node (Tα ) according to
(26) Eq. (26) in T-norm layer
O2α = Tα = μLα (L ) ∧ μCα (C ) ∧ μDα (D ), α = 1, 2, 3 5. Normalize the firing strength of each node (Tnα ) according to
Eq. (27) in normalized layer.
(3) Normalized layer- The firing strength of each rule is nor- 6. Update the consequent parameter of each node using Eq. (28) in
malized corresponding to summation of all rules firing Adaptive Defuzzification layer.
strength. The nature of node in this layer is also non- 7. Produce the aggregated output Pk for overall system according to
adaptive and labeled with N within circle. The normalized Eq. (29).
8. Output: Pk
firing strength of rule can be expressed the output O3α as
follows.
Tα
O3α = Tnα =  , α = 1, 2, 3 (27) 3.6. Best advertisement-based forwarding (BADF) technique
α Tα
(4) Defuzzy Layer- Nodes in this layer are adaptive in nature In this section, best advertisement-based forwarding (BADF)
and labeled with R within square. The output of adaptive technique is discussed, with the involvement of three aspects for
node is the multiplication of normalized firing strength of controlling the overhead related to advertisement flooding. Firstly,
rule and premises parameter of input parameter. The out- the IP address of originator, and broadcast ID of previously re-
put is also known as consequent parameter and can be ex- ceived IG advertisements (IGADs), and newly received IGADs are
pressed as checked by a drone. If drone finds duplicate IGADs, having sim-
ilar IP address of originator, broadcast ID of previously received
O4α = Tnα Pα = Tnα (qα L + rα C + sα D + tα ) (28)
and newly received IGAD, then the duplicate IGADs are discarded
Where Tnα normalized firing strength of rule is obtained from by this drone. Hence, the congestion caused by duplicate IGADs is
previous (third layer) and Pα is premises parameter of the node. avoided based on the advertisement flooding in the network.
8 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Fig. 8. Structure of IG advertisements (IGAD) message.

Secondly, the drones not yet taken off or already landed (hav- In this way, based on updated route parameters, source drone
ing zero velocity) discard all the received IG advertisements. These node decides a potential route to transmit packet. The drone node
drones are not involved during routing table computation. There- preserves the records of QoS parameters for each route to IGs in
fore, this results in the form of limited broadcasting and reduction the routing table. The optimal route and IG selection process of D-
in network congestion. IoT is also presented in Fig. 9.
In the last, a drone rebroadcasts the IGADs with route availabil-
ity factor and residual route load capacity, higher as compared to 4. Experimental results and discussion
threshold value and route delay lower than threshold value. Hop
count between drone node and IG should be lower as compared to In this section, simulation experiments are performed to carried
maximum hop count. Hence, this minimizes the traffic overhead out performance analysis of the of the proposed Drone assisted
caused by broadcasting advertisement. distributed routing framework focusing on QoS in IoT environment
(D-IoT). Simulations carried out to assess the performance related
3.7. QoS provisioning drone assisted routing to the proposed D-IoT framework in drone assisted IoT environ-
ment is presented focusing on simulation settings, parameters, and
The In this protocol, IGs broadcast IGADs for advertising their comparative analysis. The implementation of proposal and some
QoS metrics (L.C, D) periodically inside the network. Further, drone state-of-the-art techniques is done using network simulator (ns-
node knows their information on the basis of IGADs. In this proto- 2) environment [45]. The major simulation setting includes 200
col, we assume all IGs have the same IGAD interval. Let IGADs for drones wireless nodes enabled by 802.11b version of Wi-Fi in the
L, C and D are LGAD , CGAD and DGAD . Ts is timestamp or time at simulation area of 2km × 2km. The wireless transmission range
which packet is sent. The format for IGAD message is illustrated in of 200m was considered for Drone-to-Drone and Drone-to-Ground
Fig. 8. communication with most of computing performed in the ground
After receiving IGAD packet, drone node estimates the values of station server and drones active as service enables in drone as-
Lk , Ck and Dk on the basis of Eq. (16), (18) and (23). If the val- sisted IoT environment. The communication link bandwidth be-
ues of Lk or/and Ck are lower than the values of LGAD or/ and CGAD , tween Drone-to-Drone was considered 5Mbps and between Drone-
then QoS metrics (L.C, D)are updated in drone’s routing table and to-Ground server was considered 10Mbps. A total simulation time
location (X, Y), velocity V and Ts are updated in IGAD packet. Oth- of 50 minutes was considered for each experiment performed in
erwise current LGAD or/and CGAD are utilized in the routing table the simulator. Each data point considered in the experimental re-
and IGAD also. Further, the value of DGAD is updated by adding sults is an average of 10 simulation experiments performed under
the value of Dk . Then, on the basis of best advertisement-based similar parameter setting to avoid objectivity and bring normaliza-
forwarding technique IGAD packet is further forwarded inside the tion in result analysis.
network. The basic procedure for route selection is presented by Further, it is also clarified that in the simulation 802.11b Wi-Fi
Algorithm 2. The message for route updates is sent to the source was utilized considering its suitability and availability in ns2 net-
drone node by the intermediate drone node, if there is possibil- work simulation environment. It supports approximately 11Mbps
ity of novel link establishment or current link breakage along the network speed which was enough for the considered network set-
route. tings in our implementation. For comparative performance anal-
ysis, few traditional ad hoc networking framework and a recent
drone framework were considered including AODV [16], GPSR [17],
Algorithm 2 QoS provisioning Drone assisted routing. and JaRM [35]. Further, range of metrics considered in the per-
Input: Loc, V, LGAD = 0, CGAD , DGAD = 0, Bid ; formance analysis incudes packet delivery ratio, number of hand-
Process: offs among Drones assisted Flying ad hoc networks, overhead and
1. IGAD (Loc, V, LGAD = 0, CGAD , DGAD = 0, Bid ) route delay. It is clarified that mathematical modeling derivation in
2. IG sends IGADs periodically.
Section 3 are the basis of considering these metrics in the analysis
3. Drone node k receives IGAD packet.
4. if received packet based on BADF scheme condition then of results. Similar metrics were also considered in recent devel-
5. Node k computes Lk , Ck and Dk according to Eq. (16), (18), and opment in literature on drone assisted FANET environments. How-
(23). ever, we have carried out the simulation in ns-2 considering var-
6. if(LGAD = 0 || Lk < LGAD )then ious benefits of the simulation such as basic adhoc protocol sup-
7. LGAD = Lk .
8. end if
port, signal level setting in nodes, freedom of using range of wire-
9. if(Ck < CGAD )then less access techniques. We do agree that new simulator support
10. CGAD = Ck . much realistic network environment, and therefore we will plan to
11. end if work in future in these new simulators. The different color of the
12. DGAD = DGAD + Dk .
protocols represents the level of value of the measure metric. For
13. Update IGAD packet while replacing, Loc and V with Lock and Vk ,
and updating Ts . example, in Fig. 10, the color means the different values of packet
14. Update route QoS metrics (L.C, D) in routing table of node k. deliver ratio. Similarly, in Fig. 12, the color means the different lev-
15. Forward IGAD packet based on BADF scheme els of network overhead. A summarized list of major simulation
16. else discard IGAD packet; configuration settings is in Table 3.
17. end if
18. Compute Pk for each route k using NFSA
19. Pr = max{Pk } 4.1. Analysis of results
20. Select the route with Pr
21. Select the IG with Pr
22. Output: Optimized route and IG
Two scenarios: experimental results with same weight factors
and experimental results with varying weight factors are consid-
K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 9

Fig. 9. Optimal route and IG selection process of D-IoT.

Fig. 10. Behavior of PDR with Number of drones and Departure gap.

Fig. 11. Variation in PDR with Number of drones and Traffic load.
10 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Table 2 D-IoT is higher than AODV and GPSR because of consideration of

Fuzzy rules database.
route load balancing factor in the proposed D-IoT but not in the
IF THEN IF THEN state of the arts protocols. In case of D-IoT, packet delivery ratio is
Rule Rule higher than packet delivery ratio in case of GPSR, AODV and JaRM,
L C D P L C D P
when traffic load varies from 100 kb/s to 500 kb/s.
1. L1 C1 D1 P3 15. L2 C2 D3 P3
Overhead is defined as the amount of excess packets generated
2. L1 C1 D2 P3 16. L2 C3 D1 P2
3. L1 C1 D3 P1 17. L2 C3 D2 P1 for the successful delivery of actual number of packets between
4. L1 C2 D1 P2 18. L2 C3 D3 P1 the source and the destination. Fig. 12 illustrates the variation in
5. L1 C2 D2 P3 19. L3 C1 D1 P7 the overhead as departure gap and quantity of drones increase.
6. L1 C2 D3 P1 20. L3 C1 D2 P6 As departure gap increases overhead decreases, but when number
7. L1 C3 D1 P2 21. L3 C1 D3 P5
8. L1 C3 D2 P2 22. L3 C2 D1 P5
of drones increase then overhead also increases. Due to utilization
9. L1 C3 D3 P1 23. L3 C2 D2 P3 of best advertisement-based forwarding scheme to minimize over-
10. L2 C1 D1 P4 24. L3 C2 D3 P3 head, in case of D-IoT is far lower than AODV and JaRM. But GPSR
11. L2 C1 D2 P3 25. L3 C3 D1 P2 results less overhead as compared to both D-IoT, JaRM and AODV,
12. L2 C1 D3 P2 26. L3 C3 D2 P1
because GPSR utilizes smaller periodic hello packet for neighbor
13. L2 C2 D1 P4 27. L3 C3 D3 P1
14. L2 C2 D2 P3 discovery as compared to IGAD packet of D-IoT.
Stability is defined in terms of number of the handoffs, and
is inversely proportional to number of handoffs. Fig. 13 illustrates
the impact of departure gap and number of drones on the average
ered to analyze the performance of proposed D-IoT with the state-
handoffs per hour for all four routing protocols. The performance
of-the-art protocols.
of D-IoT is better than the state-of-the-art protocols in terms of
number of handoffs (stability) because D-IoT comprises better path
4.1.1. Analysis of results (with same weight factors) duration for new route selection. Whereas no path stability metric
In this section, experimental results have been described while is considered in case of AODV and GPSR. In terms of average hand-
assigning equal weightage for all the metrics: route availability fac- offs per hour, JaRM performs better than AODV and GPSR, because
tor, residual route load capacity, and route delay. path stability metric is utilized in case of JaRM. Still handoff per
Packet delivery ratio (PDR) is described as the ratio of the num- hour in D-IoT is less than JaRM. The results in Fig. 13 also show the
ber of successfully transmitted packets to the number of total variation in stability in terms of average number of handoffs when
transmitted packets. Fig. 10 shows the variation in packet delivery number of drones is considered as 40, 80, 120, 160 and 200. It is
ratio as departure gap of drones and number of drones increase for clearly enunciated that stability decreases as number of drones in-
all the compared protocol: AODV, GPSR, JaRM and D-IoT. Accord- creases in case of all the compared protocols. Further with the in-
ing to Fig. 10, in the starting till threshold value of departure gap crement in number of drones, there is slow and less increment in
(40 min), packet delivery ratio enhances gradually in case of all the number of handoffs in case of D-IoT as compared to state-of-
the compared protocols. But when departure gap increases more the-art protocol. This is because, the more the drones, the larger
than 40 min, then packet delivery ratio reduces for each proto- air communication traffic, and then, D-IoT is more likely to find
col. D-IoT performs better than the state-of-the-art protocols, for the next hop more stable with less node delay. As AODV utilizes
higher departure gap (having > 30 min). In case of AODV and the hop count as the only metric, and AODV and GPSR both al-
GPSR, overhead increases rapidly, therefore packet delivery ratio ways find the shortest path, which do not comprise link stability.
reduces quickly after a threshold. While for D-IoT, unstable routes Consequently, drones reach their maximum range in more frequent
are discarded, therefore packet delivery ratio is higher as compared manner, hence, it is prone to cause handoff.
to AODV, GPSR and JaRM. In case of AODV, GPSR and JaRM, packet Route delay is defined as the time required to transmit a data
delivery ratio decreases gradually when number of drones are 40 packet from source node to destination node. Fig. 14 shows the im-
or more than 40, but for D-IoT, packet delivery ratio starts de- pact on route delay with the variation in number of drone and CBR
creasing when number of drones are 100. From the Fig. 10, it is traffic load. Route delay in case of D-IoT is lower than GPSR and
clearly illustrated that packet delivery ratio for D-IoT is far higher JaRM because route delay is considered as one of the route selec-
as compared to state-of-the-art protocols, because D-IoT could se- tion metrics in D-IoT, but not in GPSR and JaRM. D-IoT also consid-
lect next hop with the shortest queue of packets among all the ers local dynamic queue delay for node which avoids the conges-
possible nodes. Thus, this results the avoidance of local blocking. tion. Whereas, D-IoT has slightly higher route delay as compared
Fig. 11 illustrates the impact on packet delivery ratio as traf- to AODV because packets are forwarded through shortest path in
fic load and number of drones vary, while keeping departure gap case of AODV, but is unstable.
equals to 25 min. As shown in Fig. 11, packet delivery ratio de-
creases as traffic load increases for all the compared protocols. It
is clearly enunciated that packet delivery ratio in case of proposed

Table 1
Notations.

Notation Description Notation Description

∝ k
v Target velocity c Load capacity of node k
∅∝ Horizontal direction Ci Load capacity of route i
v  Relative velocity Cmax Max route load capacity
vm Velocity vector of node Di Total delay of route i
E(T) Expected link availability period L Availability factor
Li Availability factor of route i C Residual load capacity
X Sender’s X coordinate D Delay
Y Sender’s Y coordinate Ts Time stamp
V Sender’s velocity Bid Broadcast ID
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 11

Table 3
Simulation setup.

Parameters Values Parameters Values

Simulation area 2km × 2km MAC protocol TDMA

Simulation time 50 min CBR packet size 512 bytes
Trans/Receiv antenna Omnidirectional CBR interval 0.01 sec
IGAD interval Uniform(3.5,4.5) s Drone-Drone link bandwidth 5 mbps
Drone-Drone trans range 200m Drone-Ground link bandwidth 10 mbps
Drone-Ground trans range 200m Packet Type UDP
Number of drone 200 Channel Type Wireless
Propagation model Free space

Fig. 12. Variation in Overhead with Number of drone and Departure gap.

Fig. 13. Variation in Handoffs with Number of drones and Departure gap.
12 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Fig. 14. Variation in Route delay with Number of drones and CBR trafficload.

4.1.2. Analysis of results (with varying weight factors) Fig. 16 shows the behavior of D-IoT in all scenarios in terms of
The performance of the proposed D-IoT is optimized while per- number of handoffs (stability) while varying the departure gap of
forming the simulation with the consideration of varying weigh- drones and number of drones. In Fig. 16(a)–(d) performance met-
tage corresponding to route metrics according to Neuro-fuzzy sys- rics: number of handoffs and departure gap between drones are
tem. Three scenarios are presented in simulation for this pur- compared, when number of drones are considered as 40, 80, 120
pose. In first scenario, the highest priority is given to route avail- and 200 respectively. As shown in all scenarios of Fig. 16, average
ability factor metric, while assigning weight factor: 0.6 for route handoffs per hour increase (stability decreases) with the increment
availability factor metric and weight factor: 0.2 for other met- in number of drones in case of all protocols. It is also illustrated
rics equally. The protocol in this scenario is expressed as D- from Fig. 16 that handoffs/hour is inversely proportion to departure
IoT1. In second scenario, the metric: residual route load capac- gap because handoffs/hour reduce when departure gap increases
ity is prioritized while assigning weight factor: 0.6 for residual in case of all protocols. In case of D-IoT1, number of handoffs are
route load capacity metric and weight factor: 0.2 for other met- slightly lesser as compared to D-IoT0, but in case of D-IoT2 and
rics equally and this scenario is denoted as D-IoT2. In third sce- D-IoT3, are slightly higher than D-IoT0. But D-IoT in all scenarios
nario, the metric: route delay is preferred while assigning weight performs much better than AODV and GPSR in terms of number of
factor: 0.6 for route delay metric and weight factor: 0.2 for other handoffs because no path stability metric is considered in case of
metrics equally The protocol in third scenario is expressed as D- GPSR and AODV. But JaRM performs better than AODV and GPSR
IoT3. Further the performance of D-IoT1, D-IoT2 and D-IoT3 are in terms of average handoffs per hour, because path stability met-
compared with D-IoT0 (without weights), AODV, GPSR and JaRM ric is utilized in case of JaRM. Still D-IoT in all scenarios performs
protocols. better than JaRM.
A comparison of packet delivery ratio between D-IoT in all sce- The results in Fig. 16 illustrates the variation in route delay with
narios, and the state-of-the-art protocols with varying traffic load the increment in number of drones and CBR traffic load for all the
and quantity of drones is shown in Fig. 15. Fig. 15(a)–(d) shows the scenarios of D-IoT. In Fig. 17–(d) performance metrics: route delay
comparison between packet delivery ratio and traffic load, when and traffic load are compared, when number of drones are con-
number of drones are considered as 40,80,120 and 200. As shown sidered as 40, 80, 120 and 200 respectively. As shown in all sce-
in all scenarios of Fig. 15, packet delivery ratio is increasing un- narios of Fig. 16, route delay is proportional to number of drones
til number of drones are 80 and then decreasing when number because route delay increases when number of drones increase in
of drones are increased in case of all protocols. It is also illustrated case of all protocols. It is also illustrated from Fig. 17 that for par-
that packet delivery ratio is lesser and decreasing rapidly in case of ticular number of drones, route delay increases with the increment
AODV and GPSR, as compared to packet delivery ratio in D-IoT0, D- in traffic load. From all scenarios of Fig. 17, it is clearly enun-
IoT1, D-IoT2 and D-I0T3. But in case of JaRM, packet delivery ratio ciated that D-IoT3 outperforms D-IoT0, D-IoT1, D-IoT2, JaRM and
is higher and decreasing slowly as compared to packet delivery ra- GPSR, because in D-IoT3 highest priority is given to route delay.
tio for AODV and GPSR, but lesser than packet delivery ratio in case But D-IoT3 has slightly higher route delay as compared to AODV
of all scenarios of D-IoT. packet delivery ratio in case of D-IoT3 is because packets are forwarded through shortest path in case of
lesser than rest scenarios of D-IoT, because of more consideration AODV, but is unstable. Still, D-IoT0, D-IoT1, D-IoT2 has better per-
of route delay as compared to route availability factor and residual formance than GPSR and JaRM in terms of route delay because of
route load capacity. D-IoT0, D-IoT1, D-IoT2 and D-IoT3 perform al- consideration of local dynamic queue delay for node which avoids
most same in terms of packet delivery ratio, while perform better the congestion.
as compared to the state-of-the-art protocols because of consid- The results considering different weight factors show the en-
eration of route load balancing factor. It is clearly illustrated that hancement in route stability, route delay and packet delivery ratio
stability and load balancing must be prioritize for heavy traffic in considering various scenarios inside the network.
order to improve packet delivery ratio.
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 13

Fig. 15. Variation in PDR and Traffic load when Number of drones are (a) 40, (b) 80, (c) 120 and (d) 200.

4.1.2. Summary of observations varying weight assignment for QoS metrics. As departure gap in-
The following lessons are learned from the design, implementa- creases packet delivery ratio increases till departure gap threshold,
tion and analysis of the results. To provide efficient communication after that packet delivery ratio starts decreasing. D-IoT outperforms
between drones and ground stations while having higher Quality the state-of-the-art protocols in terms of packet delivery ratio, be-
of service parameters in highly mobile drone assisted Flying ad-hoc cause D-IoT could select next hop with the shortest queue of pack-
networks is a great challenge. Hence, the proposed drone assisted ets among all the possible nodes. Thus, this results the avoidance
distributed routing framework focusing on quality of service provi- of local blocking.
sion in drone assisted IoT environments (D-IoT) enhances the net- The impact of traffic load is negative on packet delivery ra-
work performance. Neuro-fuzzy interference system provides the tio. Packet delivery ratio with varying traffic load and number of
reliable, balanced and efficient route selection by combining three drones is higher in case of proposed D-IoT than state-of-the-art
important QoS provisioning parameters: route availability factor, protocols because of consideration of route load balancing factor
residual route load capacity and route delay. in the proposed D-IoT. As departure gap increases overhead de-
For comparative performance analysis, few traditional ad hoc creases, but when number of drones increase then overhead also
networking framework and a recent drone framework were consid- increases. Best advertisement-based forwarding scheme used in D-
ered including AODV, GPSR, and JaRM. Further, metrics considered IoT reduces overhead in case of D-IoT as compared to AODV and
in the experimental results analysis incudes packet delivery ra- JaRM. But GPSR results less overhead as compared to D-IoT, be-
tio, number of handoffs among Drones assisted Flying ad hoc net- cause of utilization of smaller periodic hello packet for neighbor
works, overhead and route delay. It is clarified that mathematical discovery in GPSRas compared to IGAD packet of D-IoT. Average
modeling derivation in Section 3 are the basis of considering these number of handoffs increases as number of drones increases. Fur-
metrics in the analysis of results. Similar metrics were also consid- ther with the increment in number of drones, there is slow and
ered in recent development in literature on drone assisted FANET less increment in the number of handoffs in case of D-IoT as com-
environments. The analysis of experimental results is done firstly, pared to state-of-the-art protocol. This is because, the more the
based on equal weight assignment for QoS metrics and secondly, drones, the larger air communication traffic, and then, D-IoT is
14 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Fig. 16. Variation in Handoffs and Departure gap when Number of drones are (a) 40, (b) 80, (c) 120 and (d) 200.

more likely to find the next hop more stable with less node de- assist in reliable and efficient route selection. A drone assisted dis-
lay. tributed routing framework is developed based on the drone mo-
Route delay is proportional to both number of drones and traf- bility model and QoS parameters. The proposed communication
fic load in all compared protocols. Route delay in case of D-IoT is framework is tested to comparatively evaluate the performance
lower than GPSR and JaRM because route delay is considered as with the state-of-the-art protocols considering metrics related to
one of the route selection metrics in D-IoT, but not in GPSR and flying ad-hoc networks environments. The simulation results show
JaRM. D-IoT also considers local dynamic queue delay for node that D-IoT outperforms the state-of-the-arts protocols. It is high-
which avoids the congestion. Whereas, D-IoT has slightly higher lighted that the proposal can be utilized for any applications of
route delay as compared to AODV because packets are forwarded drone assisted adhoc networking such as enabling agriculture and
through shortest path in case of AODV, but is unstable. traffic related services using drones. Further, the framework can be
also utilized in drone assisted border monitoring. However, relia-
5. Conclusion bility centric development needs to be added before using it for
security-oriented drone monitoring applications. In the future re-
In this paper, a drone assisted distributed routing framework search, the authors will focus on consideration of energy utiliza-
focusing on QoS provision is presented for the enhancement of tion as performance metric including the design modification. The
the network performance in IoT environment. A network model authors will also explore the work in diverse scenarios, and appli-
for drone assisted IoT environment is presented focusing the topo- cations.
logical aspects of aerial drones and its mobility in flying ad hoc
networks. To optimize Drone network centric QoS provisioning pa- Declaration of Competing Interest
rameters are derived focusing on relative velocity of drones, ex-
pected link availability period, residual route load capacity and It is declared that there is no competing interest among authors
route delay. Neuro-fuzzy interference system has been employed regarding the publishing the research presented in this paper. All
to jointly combine three important QoS provisioning parameters to authors have read and approved the final version of the paper.
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 15

Fig. 17. Variation in Route delay and Traffic load when Number of drones are (a) 40, (b) 80, (c) 120 and (d) 200.

Supplementary materials [7] M. Schnell, S; Scalise, NEWSKY-concept for networking the SKY for civil aero-
nautical communications, IEEE Aerosp. Electron. Syst. Mag. 22 (5) (2007)
25–29.
Supplementary material associated with this article can be [8] F De Rango, G Potrino, M Tropea, AF Santamaria, P Fazio, Scalable and ligthway
found, in the online version, at doi:10.1016/j.adhoc.2020.102242. bio-inspired coordination protocol for FANET in precision agriculture applica-
tions, Comput. Electric. Eng. 74 (2019) 305–318.
[9] M. Tropea, A.F. Santamaria, F. De Rango, G. Potrino, Reactive flooding versus
link state routing for FANET in precision agriculture, in: 2019 16th IEEE Annual
References Consumer Communications & Networking Conference (CCNC), 2019, January,
pp. 1–6. IEEE.
[10] O. Kaiwartya, A.H. Abdullah, Y. Cao, R.S. Raw, S. Kumar, D.K. Lobiyal, I.F. Isnin,
[1] O. Kaiwartya, A.H. Abdullah, Y. Cao, A. Altameem, M. Prasad, C.T. Lin, X. Liu, In-
X. Liu, R.R. Shah, T-MQM: Testbed-based multi-metric quality measurement of
ternet of vehicles: motivation, layered architecture, network model, challenges,
sensor deployment for precision agriculture—A case study, IEEE Sensor. J. 16
and future aspects, IEEE Access 4 (2016) 5356–5373.
(23) (2016) 8649–8664.
[2] M. Asif-Ur-Rahman, Afsana, F., Mahmud, M., Kaiser, M.S., Ahmed, M.R., Kai-
[11] M. De Sanctis, E. Cianca, G. Araniti, I. Bisio, R. Prasad, Satellite Communications
wartya, O. and James-Taylor, A.., "Towards a heterogeneous mist, fog, and cloud
Supporting Internet of Remote Things, IEEE Internet Things J. 3 (1) (Feb. 2016)
based framework for the internet of healthcare things," in IEEE Internet of
113–123.
Things Journal.
[12] A Bujari, CE Palazzi, D Ronzan, A comparison of stateless position-based packet
[3] C.C. Baseca, J.R. Díaz, J. Lloret, Communication Ad Hoc protocol for intelligent
routing algorithms for FANETs, IEEE Trans. Mobile Comput. 17 (11) (2018)
video sensing using AR drones, in: 2013 IEEE 9th International Conference on
2468–2482.
Mobile Ad-hoc and Sensor Networks, 2013, pp. 449–453. Dalian.
[13] F. Garcia, A. Pirovano, M. Royer, Q. Vey, Aeronautical air-ground data communi-
[4] B. Yu, J. Wright, S. Nepal, L. Zhu, J. Liu, R. Ranjan, IoTChain: establishing trust
cation: current and future trends, in: Clean Mobility and Intelligent Transport
in the internet of things ecosystem using Blockchain, IEEE Cloud Comput. 5 (4)
Systems, IET Digit. Libr., London, U.K., 2015, pp. 40–416.
(Jul./Aug. 2018) 12–23.
[14] Vey, Q.; Pirovano, A.; Radzik, J. and Garcia, F; “Aeronautical ad hoc network
[5] M. Fazio, R. Ranjan, M. Girolami, J. Taheri, S. Dustdar, M. Villari, A note on
for civil aviation,” in Communication Technologies for Vehicles. Springer, pp.
the convergence of IoT, edge, and cloud computing in smart cities, IEEE Cloud
81–93, 2014.
Comput. 5 (5) (Sep./Oct. 2018) 22–24.
[15] E. Sakhaee, A. Jamalipour, The global In-flight internet, IEEE J. Sel. Areas Com-
[6] T. Rausch, S. Dustdar, R. Ranjan, Osmotic message-oriented middleware for the
mun. 24 (9) (2006) 1748–1757.
internet of things, IEEE Cloud Comput. 5 (2) (Mar./Apr. 2018) 17–25.
16 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242

Kirshna Kumar is currently a Ph.D. research scholar at

[16] Perkins, C.; Belding-Royer, E. and Das, S; “Ad Hoc on demand distance vector
School of Computer and Systems Sciences, Jawaharlal
(AODV) routing,” (RFC3561), Available online: http://www.ietf.org/rfc/rfc3561.
Nehru University, New Delhi, India. His research interests
txt (accessed on 30 January 2013).
include Internet of Things and Wireless Sensor Networks.
[17] B. Karp, H.T; Kung, GPSR: greedy perimeter stateless routing for wireless net-
He received his M. Tech degree in Computer Science and
works, in: Proc. 6th ACM Int. Conf. Mobile Comput. Netw., 20 0 0, pp. 243–254.
Technology from School of Computer and Systems Sci-
Boston, MA, USA.
ences, Jawaharlal Nehru University, New Delhi, India in
[18] E. Sakhaee, A. Jamalipour, Stable clustering and communications in pseudo-
2015, and B.Tech degree in Computer Science and Engi-
linear highly mobile Ad Hoc networks, IEEE Trans. Veh. Technol. 57 (6) (Nov.
neering from Uttar Pradesh Technical University, India in
2008) 3769–3777.
2011.
[19] D. Medina, F. Hoffmann, F. Rossetto, C.H; Rokitansky, A geographic routing
strategy for North Atlantic in-flight Internet access via airborne mesh network-
ing, ACM Trans. Netw. 20 (4) (2012) 1231–1244.
[20] Y. Li, R. Shirani, M. Hilaire, T. Kunz, Improving routing in networks of un-
manned Aeria vehicles: reactive-greedy-reactive, Wirel. Commun. Mob. Com- Sushil Kumar is currently working as Assistant Profes-
put 12 (18) (2012) 1608–1619. sor at School of Computer and Systems Sciences, Jawa-
[21] G. Tuna, B. Nefzi, G. Conte, Unmanned aerial vehicle-aided communications harlal Nehru University, New Delhi, India. Prior to that
system for disaster recovery, J. Netw. Comput. Appl. 41 (2014) 27–36. he spent three years as Lecturer in Jamia Millia Islamia
[22] Y. Liu, Z. Luo, Z. Liu, J. Shi, G. Cheng, Cooperative routing problem (Central University), Delhi, India and as Lecturer (Com-
for ground vehicle and unmanned aerial vehicle: the application on in- puter Science) in Shri Lal Bahadur Shastri R. S. Vidyapeeth
telligence, surveillance, and reconnaissance missions, IEEE Access 7 (2019) (Deemed University), New Delhi. He received his Ph.D.
63504–63518. degree in Computer Science from School of Computer and
[23] F. Aftab, A. Khan, Z. Zhang, Hybrid self-organized clustering scheme for drone Systems Sciences, Jawaharlal Nehru University, New Delhi,
based cognitive internet of things, IEEE Access 7 (2019) 56217–56227. India in 2014. His research interest includes the area of
[24] C. Cambra, S. Sendra, J. Lloret, L. Parra, Ad hoc network for emergency rescue vehicular cyber physical systems, Internet of things, Inter-
system based on unmanned aerial vehicles, Netw. Protocol. Algorithm. 7 (4) net of unmanned aerial vehicles, cybersecuirty and wire-
(2015) 72–89. less sensor networks. He is supervised eleven doctral the-
[25] I. Garcia-Magarino, G. Gray, R. Lacuesta, J. Lloret, Smart green communication ses and currently supervising eight doctoral theses in vehicular communication, en-
protocols based on several-fold messages extracted from common sequential ergy efficiency of terrestrial sensor networks, blockchain , and green and secure
patterns in UAVs, IEEE Netw. (2020), doi:10.1109/MNET.0 01.190 0417. computing in Internet of Things. Dr. Kumar has authored and coauthored over 80
[26] M. Y. Arafat and S. Moh, "Localization and clustering based on swarm intel- technical papers in international journals and conferences. Dr. Kumar served as ses-
ligence in UAV networks for emergency communications," in IEEE Internet sion chair in many international conferences and workshops. He is a reviewer in
Things J. doi: 10.1109/JIOT.2019.2925567. many IEEE/IET and other international journals. He is a reviewer for projects for
[27] L. Lei, D. Wang, L. Zhou, X.; Chen, S; Cai, Link availability estimation based various research funding organizations. He is served as guest editor in Journal of
reliable routing for aeronautical ad hoc networks, Ad Hoc Netw.. 20 (2014) Computer Networks and Communications, and Transaction on emerging technolo-
53–63. gies.
[28] F. Hoffmann, D. Medina, A. Wolisz, Joint routing and scheduling in mobile
aeronautical Ad Hoc networks, IEEE Trans. Veh. Technol. 62 (6) (July 2013) Dr. Omprakash Kaiwartya is currently working as a Lec-
2700–2712. turer at the School of Science & Technology, Notting-
[29] K. Peng, et al., A hybrid genetic algorithm on routing and scheduling for vehi- ham Trent University (NTU), UK. Previously, He was a Re-
cle-assisted multi-drone parcel delivery, IEEE Access 7 (2019) 49191–49200. search Associate at the Northumbria University, Newcas-
[30] K. Dorling, J. Heinrichs, G.G. Messier, S. Magierowski, Vehicle routing problems tle, UK, and a Postdoctoral Research Fellow at the Univer-
for drone delivery, IEEE Trans. Syst. Man Cybernet.: Syst. 47 (1) (Jan. 2017) siti Teknologi Malaysia (UTM). He received his Ph.D. de-
70–85. gree in Computer Science from Jawaharlal Nehru Univer-
[31] M. Hu, et al., Joint routing and scheduling for vehicle-assisted Multidrone sity, New Delhi, India. His research interest focuses on IoT
surveillance, IEEE Internet Things J. 6 (2) (April 2019) 1781–1790. centric future technologies for diverse domain areas in-
[32] Q. Fan, N. Ansari, Towards traffic load balancing in drone-assisted communica- cluding Transport, Healthcare, and Industrial Production.
tions for IoT, IEEE Internet Things J. 6 (2) (April 2019) 3633–3640. His recent scientific contributions are in Internet of con-
[33] M. Asadpour, K.A. Hummel, D. Giustiniano, S. Draskovic, Route or carry: mo- nected Vehicles (IoV), Electronic Vehicles Charging Man-
tion-driven packet forwarding in micro aerial vehicle networks, IEEE Trans. agement (EV), Internet of Healthcare Things (IoHT), and
Mobile Comput. 16 (3) (1 March 2017) 843–856. Smart use case implementations of Sensor Networks. He is Associate Editor of re-
[34] G.S.C. Avellar, G.A.S. Pereira, L.C.A. Pimenta, P. Iscold, Multi-UAV routing for puted SCI Journals including IET Intelligent Transport Systems, EURASIP Journal on
area coverage and remote sensing with minimum time, Sensors 15 (2015) Wireless Communication and Networking, and Ad-Hoc & Sensor Wireless Networks.
27783–27803. He is also Guest Editor of many recent special issues in reputed journals including
[35] C. Pu, Jamming-resilient multipath routing protocol for flying Ad Hoc net- IEEE Internet of Things Journal, IEEE Access, MDPI Sensors, and MDPI Electronics.
works, IEEE Access 6 (2018) 68472–68486.
[36] P. Yang, X. Cao, X. Xi, W. Du, Z. Xiao, D. Wu, Three-dimensional continuous Pankaj Kumar Kashyap recevied the M.tech degree in
movement control of drone cells for energy-efficient communication coverage, Computer science and technology from Jawaharlal Nehru
IEEE Trans. Veh. Technol. 68 (7) (July 2019) 6535–6546. University, new delhi, in 2014. He is currently pursuing
[37] M.N. Ahmed, A.H. Abdullah, H. Chizari, O. Kaiwartya, F3TM: flooding factor Ph.D from the same university. His research area inter-
based trust management framework for secure data transmission in MANETs, est is load balancing and energy optimization in wirless
J. King Saud Univ.-Comput. Inf. Sci. 29 (3) (2017) 269–280. sensor networks, Internet of Things or vehicles using ma-
[38] O. Kaiwartya, S. Kumar, Enhanced caching for geocast routing in vehicular chine learning approaches. H have published paper on
Ad Hoc network, in: Intelligent Computing, Networking, and Informatics , load balancing using neuro-fuzzy techinque in green com-
Springer, New Delhi, 2014, pp. 213–220. puting and other is genetic-fuzzy algorithm basd load bal-
[39] K.N. Qureshi, A.H. Abdullah, O. Kaiwartya, F. Ullah, S. Iqbal, A. Altameem, ncing in wireless sensor networks.
Weighted link quality and forward progress coupled with modified RTS/CTS for
beaconless packet forwarding protocol (B-PFP) in VANETs, Telecommun. Syst.
(2016) 1–16.
[40] Y. Cao, O. Kaiwartya, N. Aslam, C. Han, X. Zhang, Y. Zhuang, M. Dianati, A tra- Jaime Lloret (M’07–SM’10) received the B.Sc. + M.Sc. de-
jectory-driven opportunistic routing protocol for VCPS, IEEE Trans. Electron. gree in physics and the B.Sc. + M.Sc. degree in electronic
Syst. 54 (6) (2018) 2628–2642. engineering from the University of Valencia, Valencia,
[41] F De Rango, M Tropea, P Fazio, Bio-inspired routing over FANET in emergency Spain, in 1997 and 2003, respectively, and the Ph.D. de-
situations to support multimedia traffic, in: Proc. of the ACM MobiHoc Work- gree in telecommunication engineering (Dr.Ing.) from the
shop on innovative aerial communication solutions for FIrst Responders net- Polytechnic University of Valencia, Valencia, in 2006. He
work in emergency scenarios, 2019, pp. 12–17. is currently an Associate Professor with the Polytechnic
[42] B. Liang, Z.J. Haas, Predictive distance-based mobility management for multi- University of Valencia, Valencia, Spain. He is the Chair
dimensional PCS networks, IEEE/ACM Trans. Netw. 11 (5) (2003) 718–732. of the Integrated Management Coastal Research Insti-
[43] J.S.R. Jang, ANFIS: adaptive-network-based fuzzy inference system, IEEE Trans. tute and the Head of the Active and Collaborative Tech-
Syst. Man Cybernet. 23 (3) (1993) 665–685. niques and Use of Technologic Resources in the Educa-
[44] J. Li, L. Yang, Y. Qu, G. Sexton, An extended Takagi–Sugeno–Kang inference sys- tion (EITACURTE) Innovation Group. He is the Director of
tem (TSK+) with fuzzy interpolation and its rule base generation, Soft Comput. the University Diploma Redes y Comunicaciones de Orde-
22 (2018) 3155–3170. nadores and the University Master Digital Post Production. He has authored 22 book
[45] M. Nekrasov, R. Allen, I. Artamonova, E. Belding, Optimizing 802.15.4 outdoor chapters and has had over 360 research papers published in national and interna-
IoT sensor networks for aerial data collection, Sensors 19 (2019) 3479 2. tional conferences and international journals (over 140 with ISI Thomson JCR). Dr.
Lloret was the Internet Technical Committee Chair (the IEEE Communications So-
ciety and Internet Society) for the term 2013–2015. He has been the Co-Editor of
 K. Kumar, S. Kumar and O. Kaiwartya et al. / Ad Hoc Networks 106 (2020) 102242 17

40 conference proceedings and the Guest Editor of several international books and Informatics and IEEE Network. He is the editor of six books, including Big Data
journals. He is the Editor-in-Chief of Ad Hoc and Sensor Wireless Networks (with Analytics for Cyber-Physical Systems: Machine Learning for the Internet of Things,
ISI Thomson Impact Factor), the international journal Networks Protocols and Al- Elsevier, 2019, Smart Cities: Foundations, Principles and Applications, Hoboken, NJ:
gorithms, and the International Journal of Multimedia Communications, the IARIA Wiley, 2017, Security and Privacy in Cyber-Physical Systems: Foundations, Princi-
Journals Board Chair (eight journals), and he is (or has been) an Associate Editor of ples and Applications, Chichester, UK: Wiley-IEEE Press, 2017, Cyber-Physical Sys-
46 international journals (16 with ISI Thomson Impact Factor). He has been involved tems: Foundations, Principles and Applications, Boston, MA: Academic Press, 2016,
in over 320 program committees of international conferences, and over 130 organi- and Industrial Internet of Things: Cybermanufacturing Systems, Cham, Switzerland:
zation and steering committees. He leads many national and international projects. Springer, 2016. He is the author of more than 100 articles. His research interests
He is currently the Chair of the Working Group of the Standard IEEE 1907.1. He include cyber-physical systems, cybersecurity and privacy, internet of things, edge
has been the General Chair (or Co-Chair) of 39 international workshops and confer- computing, big data analytics, unmanned aircraft systems, connected vehicle, smart
ences. System to Recommend the Best Place to Live Based on Wellness State of the and connected health, and wireless communications and networking. His research
User Employing the Heart Rate Variability. He is an IARIA Fellow. has been featured by popular news media outlets, including USA Today, U.S. News &
World Report, Fox News, Forbes, WFTV, and New Atlas. Dr. Song is a senior member
Houbing Song (M’12–SM’14) received the Ph.D. degree of ACM. Dr. Song was a recipient of the Best Paper Award from the 12th IEEE In-
in electrical engineering from the University of Virginia, ternational Conference on Cyber, Physical and Social Computing (CPSCom-2019), the
Charlottesville, VA, in August 2012, and the M.S. degree in Best Paper Award from the 19th Integrated Communication, Navigation and Surveil-
civil engineering from the University of Texas, El Paso, TX, lance technologies (ICNS 2019) Conference, the prestigious Air Force Research Lab-
in December 2006. In August 2017, he joined the Depart- oratory’s Information Directorate (AFRL/RI) Visiting Faculty Research Fellowship in
ment of Electrical, Computer, Software, and Systems En- 2018, and the very first recipient of the Golden Bear Scholar Award, the highest
gineering, Embry-Riddle Aeronautical University, Daytona campus-wide recognition for research excellence at West Virginia University Insti-
Beach, FL, where he is currently an Assistant Professor tute of Technology (WVU Tech), in 2016.
and the Director of the Security and Optimization for Net-
worked Globe Laboratory (SONG Lab, www.SONGLab.us).
He served on the faculty of West Virginia University from
August 2012 to August 2017. In 2007 he was an Engineer-
ing Research Associate with the Texas A&M Transporta-
tion Institute. He has served as an Associate Technical Editor for IEEE Communi-
cations Magazine and a Guest Editor for IEEE Journal on Selected Areas in Com-
munications (J-SAC), IEEE Internet of Things Journal, IEEE Transactions on Industrial