You are on page 1of 9

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
1

Energy-Efficient Drone Trajectory Planning for


the Localization of 6G-enabled IoT Devices
Sahar Kouroshnezhad1, Ali Peiravi2,
Mohammad Sayad Haghighi3, Senior Member, IEEE, Alireza Jolfaei4, Senior Member, IEEE

increase the number of stations. In remote sensing applications,


Abstract—6G will be an enabler for massive Internet of this becomes a problem, since either we have to increase the
Things (IoT) in which millions of devices communicate at number of sensors and lower their transmission ranges which
high data rates and low latencies. One key area among 6G increases the deployment cost or increase their transmission
applications is advanced sensing. However, higher speed ranges and accept faster draining of their limited energy supply
implies moving to higher frequencies, which generally [3]. Therefore, energy efficiency is a crucial matter in 6G
require more transmission power. In remote sensing, this sensing applications. To cut the cost in massively produced
causes problems, since either we have to increase the sensors, localization modules like GPS (Geographical
number of sensors and lower their communications ranges, Positioning System) are not usually included. This also helps in
or increase their ranges and accept faster battery depletion. saving energy to some extent. However, in many applications,
To cut the cost, even localization modules are not usually including transmission power tuning or topology control, an
included in sensors. However, in many applications, IoT IoT sensor must know its location.
sensors must know their locations. Recent advances in the Recent advances in Unmanned Arial Vehicles (UAVs),
field of drones have led to promising solutions for including drones, have led to promising solutions for localizing
localization. In this paper, we propose a novel approach future 6G-enabled IoT devices [4],[5]. The drone flies over the
called SEmi-Dynamic Mobile Anchor Guiding (SEDMAG)
region of interest where the objects are deployed and broadcasts
for drones which aims at energy-conservative trajectory
planning and localization of massive IoT devices. In this beacons that include coordinates and transmission power,
approach, the drone tracks the shortest path over a somewhat similar to what is done in vehicular networks [6]. The
connected graph. This path determines the visiting order of covered sensors calculate their distances from the drone with
devices. But we show that the complexity of this approach is each beacon, which is referred to as slant distance. The
high, thus, a graph reduction approach is proposed. It projection of drone beacon points and slant distances on the
reduces the complexity and decreases the drones´ energy terrestrial deployment area are called waypoints and ground
consumption and positioning delay. The drone then follows
distances, respectively [4-5,7]. Each sensor can estimate its
a Weighted Search Algorithm (WSA) to dynamically visit
the devices. Simulation results are used to verify the location after collecting at least three beacons from non-
superiority of the proposed approach. collinear waypoints. The main challenge then becomes how to
find an energy-efficient path for the drone to localize all the IoT
Index Terms— Internet of Things, Energy Efficiency, 6G, sensors, since energy efficiency will be at the core of any
Advanced Sensors, Drone, Dynamic Path Planning, Localization.
protocol designed for 6G [1].
Considerable research attention has been drawn to mobile
I. INTRODUCTION anchor path planning in recent years including both static and
dynamic trajectories. In a dynamic planning, drone trajectory is
6G will be an enabler for massive IoT in which millions of
connected devices and applications can operate
seamlessly along with each other and communicate at very high
found on the fly and progressively according to the environment
observations, deployment situation, etc. Therefore, they yield
data rates with low latency. One key area of 6G applications is higher performances. However, they also introduce more
advanced sensing [1]. However, higher speed means more
localization latency and communication overhead.
bandwidth and this implies moving to higher frequencies with
shorter wavelengths, which generally requires more power to In this paper, we present SEDMAG, a novel approach for
maintain the transmission ranges we used to have before [2]. drones, which aims at energy conservative trajectory planning
The general solution is to get devices closer to each other and and precise localization of a massive number of IoT devices. In

1 3
Sahar Kouroshnezhad is with the Department of Electrical Engineering, Mohammad Sayad Haghighi (corresponding author) is with the School of
Bojnourd Branch, Islamic Azad University, Bojnourd, Iran Electrical and Computer Engineering, College of Engineering, University of
(kouroshnezhad@bojnourdiau.ac.ir). Tehran, Iran (sayad@ut.ac.ir, sayad@ieee.org).
2 4
Ali Peiravi is with the Electrical Engineering Department of Ferdowsi Alireza Jolfaei is with the Department of Computing, Macquarie University,
University of Mashhad, Iran (peiravi@um.ac.ir). Sydney, NSW 2109, Australia (alireza.jolfaei@mq.edu.au).

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
2

this approach, the drone examines all possible Depth First Trees sensors lifetime. Z-power is able to control the transmission
(DFTs) of the network and uses the shortest path algorithm to power of the mobile anchor dynamically as it follows the pre-
determine the optimal visiting order. To overcome the problem determined Zcurve [12] path.
The OPTEC algorithm, which has been proposed in [15] by
of complexity, a graph reduction approach is proposed. After
Kouroshnezhad et al., is a static anchor guiding approach that
the order is determined, the drone dynamically follows a utilizes initial region estimates of sensors for optimal path
Weighted Search Algorithm (WSA) to visit the nodes/devices. planning. This study has also considered mobile anchor energy
Furthermore, a smart version of WSA is proposed to reduce the limitations in the planning. It uses integer linear programming
number of required beacons for location estimation. SEDMAG to determine the optimal beacon points based on the defined
is also extended to multiple-drone scenarios where the load is priorities and to construct the shortest path among them.
balanced among them. We evaluate the performance of WETAR is another static trajectory that has been proposed
by Kouroshnezhad et al. [5]. It prioritizes the candidate
SEDMAG and its improved versions in terms of localization
waypoints based on quality of the beacons that the sensors
coverage, localization accuracy, latency and energy would receive and their coverage ratio. Then the drone
consumption and compare them to those of well-known static trajectory is built using linear programming in the presence of
and dynamic trajectory planning methods. the sensors with given estimative regions which are acquired in
In summary, the contributions of this paper are as follows: a range-free pre-localization phase. WETAR determines both
• First, we present a novel semi-dynamic trajectory optimum trajectory and optimum transmission power of the
drone at each beacon point.
planner (SEDMAG) which works in two phases:
The presented trajectory in [16] determines the clusters first and
determination of optimal visiting order of IoT sensors then a genetic algorithm is employed to determine the global
and searching for target sensor location at each step. trajectory between cluster heads. The anchor traverses along
• Second, we introduce a smart version of the search regular hexagonal trajectory in each cluster with the cluster
approach to improve its performance. Then, we head located at the center of the hexagon.
introduce a node removal technique to reduce the The mobile anchor, in the DeteRministic dynamic bEAcon
number of visited sensors and shorten the trajectory. Mobility Scheduling (DREAMS) algorithm [17], first visits a
sensor node by moving randomly and then performs a DFT on
• At last, we propose a load balancing concept to reduce the network graph based on the instruction of the visited sensor
localization latency in multi-drone scenarios. node. During DFT, the anchor performs distance-based
The rest of this paper is organized as follows. Section II movements from node to node following RSS, and sensor nodes
describes the existing related studies. Section III details the run the built-in localization algorithms to self-localize using the
proposed trajectory planning algorithm. Evaluations of the received beacons. Simulation results of DREAMS show that it
proposed algorithm are reported in Section IV. We explain the produces accurate location estimations.
Another dynamic algorithm [18] involves search-and-decide
simulation setup and compare the performance of the proposed
phases. In the search-phase, mobile anchor follows the SCAN
method with the existing ones through simulations. Finally, a trajectory to collect neighborhood information. In the decision-
conclusion is made in Section V. phase, mobile anchor determines its movement path
dynamically according to the acquired neighborhood
II. RELATED WORK information and the Perpendicular Bisector Strategy [19].
Localization by using a drone is more effective than utilizing This paper focuses on a semi-dynamic path planning scheme
terrestrial mobile anchors for two primary reasons; first, mobile in which a mobile anchor searches unknown sensors
anchors on the ground are more probable to be at risk of being dynamically in a predetermined order. It aims to overcome
damaged by environmental and human-made threats and intrinsic issues of dynamic approaches (i.e., localization latency
second, the probability of having LOS (Line of Sight) links in and communication overhead).
air to ground channels is higher, which in turn helps to improve
positioning precision [7,8].
Since location of broadcasted beacons by the drone can III. THE PROPOSED PATH PLANNING SCHEME
strongly affect the location estimation accuracy of terrestrial A. Semi-dynamic mobile anchor guiding algorithm
IoT devices, finding an efficient drone trajectory is considered
We propose a novel path planning algorithm called
as the main task in drone-assisted localizations. In this section,
we review the works on mobile-anchor path planning covering SEDMAG by which the drone visits every sensor following the
both static [5], [9-15] and dynamic [16-19] trajectories. search instructions. This algorithm is static in the sense that
Han et al. [12] considered the path planning issue as seeking sensors’ visiting order is determined beforehand; however, the
LMAT-based beacon points [11] while traversing the SCAN local movements are dynamically specified. Multiple drones
trajectory. This method, which is called SLMAT, tries to can also be used to make the algorithm more time-efficient. We
conserve energy by reducing the number of corners along the assume that the drone or drones fly at a fixed altitude (ℎ) above
planned path.
the terrestrial region of interest. Because of this, the drone
Rezazadeh et al. [13] proposed an energy-preserving
trajectory named Z-power. It aims to increase the beacon and trajectory planning problem is converted to the path planning of

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
3

Fig. 2. Smart Weighted Search Algorithm (SWSA).


Fig. 1. Demonstration of Weighted Search Algorithm (WSA).

its projection on the ground. In this paper, whenever we refer to scenarios, a graph reduction technique is introduced which
drone, we mean its projection on the ground. significantly reduces the number of sensors to be visited. This
For simplicity, we describe the proposed algorithm with one procedure effectively reduces 𝑁 in the equation of time
drone initially. It will be modified to support multiple drones complexity. Simulation results on an Intel Core i7 processor
later. SEDMAG consists of two main steps: determining running at 2.7GHz shows that it takes 77.8s to determine the
sensors’ visiting order and the Weighted Search Algorithm optimum visiting order for 480 deployed sensors.
(WSA). We also propose three approaches to improve the
localization performance, including: 1. Smart WSA, 2. Graph 2) Weighted Search Algorithm (WSA)
reduction, and 3. Localization time balancing. The search algorithm is implemented through a sequence of
instructions and consists of a few movement steps that guides
1) Sensors’ Visiting Order Determination the drone to the target sensor in a noisy environment. Suppose
Here, we elaborate on how the drone specifies the visiting that the drone, which is a mobile anchor denoted by MA, is
order of sensors for a given undirected weighted sub-graph. The located at 𝑝𝑡 ; the center of the circle 𝐶 with radius 𝑑0 where
drone with its initial position at the center of the sub-region 𝑑0 is the estimated distance of 𝑝𝑡 to 𝑆𝑇 (the target sensor). The
collects neighborhood information of the sensors to construct WSA algorithm aims to move the 𝑀𝐴 to 𝑆𝑇 . This search
the graph 𝐺(𝑉, 𝐸) which is a weighted graph with 𝑉 = approach is discussed in three steps in accordance with Fig. 1
{𝐼𝐷𝑠1 , 𝐼𝐷𝑠2 , … , 𝐼𝐷𝑠𝑘 } (sensors’ IDs) as its vertices. Also, as follows:
𝐸 denotes the links between neighboring sensors. The 1. The 𝑀𝐴 moves to 𝑝𝑡+1 which is selected randomly on
Euclidean distances of connected vertices, which are estimated the circle C and broadcasts beacon at 𝑝𝑡+1 with a
by the RSS method, specify edges weights. The drone selects transmission range of 2𝑑0 (to guarantee 𝑆𝑇 coverage).
its closest neighbor sensor as the root vertex to construct all 2. 𝑆𝑇 and 𝑆𝑇′ in (1) will be calculated as possible
DFTs of the network graph. It also calculates the total path coordinates of the target sensor. 𝑑 shows the measured
length of all the constructed DFTs and selects the one with distance from 𝑝𝑡+1 to the target sensor.
minimum path length as the optimal vising order of sensors.
−d0 sin(α+θ) d0 sin(α−θ) (1)
This procedure consists of three steps as described below: 𝑆𝑇′ = (−d ) 𝑆𝑇 = (−α )
0 cos(α+θ) 0 cos(α−θ)
𝑑
1. A set, 𝑉𝑂 , which includes all DFTs of the network 𝛼 = 2𝑠𝑖𝑛−1
2𝑑0
(sub)graph 𝐺, with the closest neighbor sensor of the drone
3. MA moves to one of these coordinates ( 𝑆𝑇 𝑜𝑟 𝑆𝑇′ )
as the root vertex is constructed.
randomly. If MA selects the correct position, it will
2. The shortest path to traverse is determined for each subset
of 𝑉𝑂 (e.g by Dijkstra). reach the target sensor after traversing the length of
𝛼
3. The optimal sensors’ visiting order (called SP) which 𝑑0 + 𝑑 (where 𝑑 = 2𝑑0 sin ( )) and broadcasting
2
results in the shortest path length (𝐿𝑆𝑃 ) is chosen. two beacons. Otherwise, it will travel the longer
Following SP, the target sensor is specified at each step and distance of 𝑑0 + 𝑑 + 𝑑 ′ (𝑑 ′ = 2𝑑0 sin(𝛼)) and
the drone moves towards it, as explained in the next section. broadcast three beacons to visit the target sensor.
The time complexity of the shortest path construction algorithm
This algorithm guaranties an upper bound of 4.52𝑑0 for the
equals to (𝜋𝑟 2 × 𝛿)𝑁 for a network of 𝑁 sensors distributed
MA path length to reach the target sensor.
randomly with a density of 𝛿 and a communication radius of 𝑟.
Note that πr 2 × δ is the mean number of 1-hop neighbors for B. Improvement Methods
each sensor. Therefore (πr 2 × δ)𝑁 DFTs should be examined In this section we propose three improvement techniques,
to find the optimal visiting order of sensors. In order to reduce namely Smart WSA, graph reduction and topology control (in
the time complexity and make it practical in real-world

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
4
14
13

2
31
32 28 12
12 1
29 30 26
4 3
11
27 12
10
15
13
11 2 10
24 4
25 1
9
5 6
8
22 14 10 3 5
20
9 6
16 8
8 7 16
21 12
23 7 13
6
19 7 11 15

4
17 6

8
9
5
14

10
2
18 4
-2 0 2 4 6 8 10 12

0 2 (a) drone path on original graph


4 6 8 10 12 (b) drone path on reduced graph

Fig. 3. Example of drone trajrctoty on (a) original gragh (b) reduced graph

multiple drone scenarios), to achieve higher effectiveness in a) Preliminaries and Problem Formulation
terms of delay and energy. They can be combined too. Hinge-sensor: If failure of a sensor node causes the graph to
be partitioned into discrete segments, this node is called a
1) Smart Weighted Search Algorithm (SWSA) “hinge sensor”, otherwise a “non-hinge” one.
In this section, two solutions are suggested to curtail the Algebraic connectivity: It is the second smallest
drone trajectory length and improve WSA performance. This eigenvalue (𝜆2 ) of the Laplacian matrix of graph 𝐺 . This
new improved algorithm is called SWSA. eigenvalue is greater than zero if and only if, 𝐺 is connected.
While the drone moves toward its target sensor, nearby IoT
sensors may overhear its beacons and after receiving at least b) Hinge Sensor Detection
three non-collinear beacons, their location can be ascertained Here, we identify non-hinge sensors with confidence via only
by trilateration. This process which we call indirect localization local neighborhood information of each sensor; that is if a
is very likely to happen in dense sensors. sensor is identified as a non-hinge sensor, it is not definitely a
Furthermore, in some cases, an unknown sensor can cut-vertex, whereas if a sensor is identified as a hinge one, it
determine its coordinates using two beacons and its may still be a global non-hinge sensor in the whole network. To
neighborhood graph information [20]. Assume that the sensor reach this goal, an algorithm is presented as below:
network in a sub-region is depicted by the graph 𝐺(𝑉, 𝜉) in Fig. 1. Sensors exchange “Hello” messages with their 1-hop
2 where 𝑉 = {𝐴, 𝐵, 𝐶, 𝐷, 𝐸} and 𝜉= neighbors. They share their local information such as ID
{(𝐴, 𝐵)(𝐵, 𝐶), (𝐶, 𝐷), (𝐷, 𝐸), (𝐴, 𝐶)}. Let us assume that 𝑆𝑃 =
and 1-hop neighbor set. Hence, each sensor can have
{(𝐴, 𝐵)(𝐵, 𝐶)(𝐶, 𝐷)(𝐷, 𝐸)} is the optimal visiting order.
complete connectivity information.
Suppose that sensors 𝐴, 𝐵, 𝐶 and 𝐷 have already been visited by
the drone and now 𝐸 is the target sensor. According to the 2. Based on the local information gained, a subgraph of
WSA, 𝐷′ is selected as the next waypoint. After using original graph 𝐺 can be constructed for each sensor (𝐺𝑖 ). Now,
radiolocation, 𝐷 and 𝐷′ specify that the target sensor must be each sensor can determine whether it is a hinge sensor or non-
located at one of the intersection points (𝐸 or 𝐸 ′ ). Assume that hinge sensor following the algebraic connectivity theorem.
𝐸 ′ is within communication range of sensors 𝐴, 𝐶 and 𝐷 . The Laplacian matrix of 𝐺𝑖 is calculated as 𝐿(𝐺𝑖 ) = 𝐷(𝐺𝑖 ) −
However, the target sensor is within communication range of D 𝐴(𝐺𝑖 ). If its algebraic connectivity , 𝜆2 (𝐺𝑖 ) > 0 , the sensor is
only. Therefore, using the graph information, 𝐸 will be the only a non-hinge sensor; otherwise, it is a hinge one.
possible position for the target sensor.
c) Graph Reduction by Non-Hinge Sensor Removal
2) Graph Reduction Removing some of the non-hinge sensors (removable non-
To minimize localization latency and shorten the trajectory, hinge sensors) and keeping the rest (fixed sensors), will shorten
the drone should travel the shortest path that leads to full the drone path length and reduce the localization latency. The
localization coverage instead of visiting all the IoT sensors. graph reduction algorithm is discussed as follows:
Graph reduction by elimination of a non-cut-vertex will not
jeopardize its connectivity [21]. This fact motivated us to run 1. Hinge sensors are considered as initial fixed sensors.
SEDMAG on a reduced sub-graph instead of the original one. Non-hinge sensors are investigated one by one according
The main objectives of this section are as follows: (1) Locally to their degree (in an increasing order) to find removable
determine the non-cut-vertices (or non-hinge sensors) of the sensors. A non-hinge sensor will be added to the
network based on 2-hop information, (2) Curtail the drone path removable set if and only if it has at least two 1-hop
length by elimination of the removable non-hinge sensors. neighbors and it is not a fixed sensor.
2. If a non-hinge sensor finds itself a removable one, the
fixed sensors set will be updated if :

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
5

Fig. 4. Mobile anchor trajectory length calculation (TS stands for Target Sensor).

3-1) The removable sensor does not have any fixed 1-hop 1. Average localization time ( 𝑇𝐿𝑖 ) of each sector is
neighbor, three of its neighbors with highest degree will be calculated by computing the expected value of
added to the fixed set. traversed distance while considering the given speed.
3-2) The removable sensor has only one or two fixed 1-hop 2. The sector with the maximum localization latency is
neighbors, two or one of its non-fixed neighbors with selected as the Source Region (SR).
highest degree will be added to the fixed set, respectively. 3. The adjacent sector of SR with minimum localization
latency, which has links with SR, is selected as End
In Fig. 3, a network of 15 randomly-deployed nodes is
Region (ER).
depicted. The two sub-figures show the complete drone
4. In order to balance the latencies, we consider flexible
trajectory over the original network graph and the reduced
boundaries for the sectors, meaning that, boundary
graph, respectively. The sensors are shown as green circles and
sensors of SR are transferred to ER under the following
the waypoints as small red circles. Drone movement steps are
conditions:
shown with blue lines. The numbers beside the waypoints show
• The sub-graph of SR should remain
their order of broadcast. They indicate that graph reduction-
connected after removal of its boundary
based trajectory involves 16 waypoints while the original
sensors.
trajectory has 32 waypoints. From the figure, it is clear that the
length of path is shortened significantly when it is guided over • Transfer of these sensors from SR to ER result
the reduced graph rather than the original one. in average trajectory length reduction in SR.
5. The localization time of both SR and ER are updated
C. Multiple Drones and Localization Time Balancing (step1). Then steps 2 to 4 are repeated.
Employing a group of drones can reduce localization latency 6. Run steps 1 to 4 until no improvement is made.
and also prolong their lifetime. With multiple drones working
simultaneously, the resultant localization latency is calculated a) Expected Value of the Drone Trajectory Length
by (2). As discussed previously, SWSA shortens the trajectory
length by utilizing the network graph information. Therefore,
localization of sensors is possible with three or sometimes even
𝑇𝐿 = 𝑚𝑎𝑥{𝑇𝐿1 , 𝑇𝐿2 , … , 𝑇𝐿𝑀 } (2) two beacons. In this intelligent method, while the drone moves
towards the target sensor, seven different cases can happen.
where 𝑇𝐿𝑖 is the localization time of 𝑖 𝑡ℎ drone. 𝑇𝐿𝑖 can be These cases have different occurrence probabilities and lead to
calculated as follows: different traveling distances. Fig. 4 illustrates these cases.
𝐿𝑡 In Fig. 4(a), the target sensor ( 𝑆𝑇 ) has been localized by
(3)
𝑇𝐿𝑖 = (𝑁𝐵 × 𝑡𝑏 ) + receiving beacons from 𝐴, 𝐵 and 𝐶 . Therefore, the traversed
𝜈
distance to the target sensor by the drone, which is located at 𝐶,
𝑁𝐵 is the number of waypoints and 𝑡𝑏 is the time needed for is 𝑑0 . In Fig. 4(b), target sensor 𝑆 has received two beacons
broadcasting a beacon. The drone travels with a constant speed from 𝐵1 and 𝐵2 already. Thus, the two possible locations are 𝑆
of 𝜈 over a distance of 𝐿𝑡 . and 𝑆 ′ . According to SWSA, 𝑆 is the only possible location. In
In SEDMAG, the monitoring area is partitioned off equally this case, the traversed distance to the target sensor by the
and each sector is allocated to one drone. The difference in the drone, which is located at 𝐵2 , is 𝑑0 . This case is likely to
number of sensors in sectors as well as their different
happen if there exists at least one sensor with known location
distributions results in different values for 𝑇𝐿𝑖 . To better (like 𝐶) in the gray area. The gray area is in the communication
equalize these timings, the localization time-balancing scheme radius of 𝑆 ′ and is not in that of 𝑆. If localization is not possible
aims to determine the task of each drone, which can be with two beacons, the drone may select the correct
described in the following steps: location 𝑆 randomly, which leads to a traversed distance of 𝑑0 ,
but if the drone first selects the incorrect location (𝑆 ′ ) and then

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
6

moves to the correct coordinates (𝑆), the traversed distance will 𝐾−𝑚
4 𝑁 𝑚 𝑁
𝑘
be d0 + d0. Fig. 4(c) shows that 𝑆𝑇 received one beacon from
π
𝑃(𝑁𝐵 ≥ 𝑀)=∑𝑘𝑚=𝑀(𝑚 ). ( 0) . (1 − ( 0)) (6)
𝑁 𝑁
𝐵 at first. Therefore, drone has followed the search algorithm
and has moved to C. Now, the target sensor has received two In which 𝑘 is the number of sensors visited by drone so far and
beacons from 𝐵 and 𝐶 . If localization is possible with two 𝑁 is the total number of sensors. 𝛿 is the network density and 𝑟
beacons (there exists (at least) one visited sensor like A in the is the communication radius of the sensors. Accordingly, the
gray area which is located only in the communication radius of number of 1-hop neighboring sensors for the target sensor is
4
𝑆𝑇′ ), the drone will travel the distance of d0 + d0 to reach the 𝑁0 = 𝜋𝑟 2 × 𝛿.
π
target sensor. Otherwise, two cases are probable: the drone 𝑌 in (5), which is the localization probability of the target
selects the correct location which results in trajectory length of sensor with two received beacons from the localized sensors
4
d0 + d0 ; or it first moves to the incorrect location and then (discussed in Fig. 4(b)), is calculated by (7). 𝑍 in (5), which
π indicates the localization probability of the target sensor with
8
moves to the correct one with a trajectory length of d0 + d0 . one beacon is estimated according to Fig. 4(c) through (8).
π
The drone trajectory lengths to the target sensor and their
probabilities of occurrence are given in (4) and (5). E= {𝑡ℎ𝑒𝑟𝑒 𝑒𝑥𝑖𝑠𝑡 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 1 𝑣𝑖𝑠𝑖𝑡𝑒𝑑 𝑠𝑒𝑛𝑠𝑜𝑟 𝑙𝑖𝑘𝑒 𝐶 𝑖𝑛 𝑔𝑟𝑎𝑦 𝑎𝑟𝑒𝑎}
F= {2 𝑏𝑒𝑎𝑐𝑜𝑛𝑠 ℎ𝑎𝑣𝑒 𝑏𝑒𝑎𝑛 𝑟𝑒𝑐𝑒𝑖𝑣𝑒𝑑 𝑓𝑟𝑜𝑚 𝑡ℎ𝑒 𝑣𝑖𝑠𝑖𝑡𝑒𝑑 𝑠𝑒𝑛𝑠𝑜𝑟𝑠} (7)
𝑘 ′ −𝑚
d0 ′ ′ 𝑁 𝑚 𝑁1
More than 2 beacons are received. Y = p(E|F) = ∑𝑘𝑚=1 (𝑘𝑚 ) . ( 1) . (1 − ( ))
𝑁 𝑁
d0 Localization is done with 2 beacons. ′
𝑘 =𝑘−2
d0 2 beacons are received and the correct location is
selected first. Q= {𝑡ℎ𝑒𝑟𝑒 𝑒𝑥𝑖𝑠𝑡𝑠 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 1 𝑣𝑖𝑠𝑖𝑡𝑒𝑑 𝑠𝑒𝑛𝑠𝑜𝑟 𝑙𝑖𝑘𝑒 𝐴 𝑖𝑛 𝑔𝑟𝑎𝑦 𝑎𝑟𝑒𝑎}
4 R= {1 𝑏𝑒𝑎𝑐𝑜𝑛 ℎ𝑎𝑠 𝑏𝑒𝑎𝑛 𝑟𝑒𝑐𝑒𝑖𝑣𝑒𝑑 𝑓𝑟𝑜𝑚 𝑡ℎ𝑒 𝑣𝑖𝑠𝑖𝑡𝑒𝑑 𝑠𝑒𝑛𝑠𝑜𝑟𝑠}
d0 + d0 2 beacons are received and the incorrect location ′′ 𝑚 𝑘 ′′ −𝑚
π ′′ 𝑁 𝑁
is selected first. Z = p(Q|R) = ∑𝑘𝑚=1 (𝑘𝑚 ) . ( 1) . (1 − ( 1))
𝑁 𝑁 (8)
𝐷= 4 (4)
d0 + d0 1 beacon has been received and the localization is 𝑘 ′′ = 𝑘 − 1
π
done after receiving the 2nd beacon at C.
4
d0 + d0 1 beacon has been received and the correct
π location is selected randomly after receiving the
2nd beacon broadcasted at C. Table 1. Simulation Parameters
8 parameter value
d0 + d0 1 beacon has been received and the incorrect
π location is selected first. Number of sensors 80,160,240,320,400,480
{
Transmission power 𝑃𝑡 = −25𝑑𝐵𝑚
Receiver sensitivity 𝑃𝑟 = −90𝑑𝐵𝑚
𝑃(𝑁𝐵 ≥ 3) More than 2 beacons are
received. Beacon packet size 20Byte
Path loss exponent(γ) 3.3
𝑃(𝑁𝐵 = 2) × 𝑌 Localization is done with 2 Data rate 250Kbps
beacons.
Drone energy consumption 0.1j
to move by one meter
𝑃(𝑁𝐵 = 2) × (1 − 𝑌) × 0.5 2 beacons are received and the
correct location is selected first. Drone energy consumption 0.001j
to broadcast a beacon
2 beacons are received and the Drone speed 2 m/s
𝑃(𝑁𝐵 = 2) × (1 − 𝑌) × 0.5
incorrect location is selected
first.
𝑃(𝐷) = (5) 0.03
1 beacon has been received &
𝑃(𝑁𝐵 = 1) × 𝑍
Proportion of unlocalized sensors

the localization is done after 0.025


receiving the 2nd beacon at C.

1 beacon has been received and 0.02


𝑃(𝑁𝐵 = 1) × (1 − 𝑍) × 0.5 the correct location is selected
Zcurve
randomly after receiving the 2nd 0.015
SSEDMAG-reduced
beacon broadcasted at C. SSEDMAG
0.01 SEDMAG
𝑃(𝑁𝐵 = 1) × (1 − 𝑍) × 0.5 1 beacon has been received and
the incorrect location is 0.005
{ selected first.
0
d0 shows the distance between the current waypoint and the 0.016 0.024 0.032 0.04 0.048
Network density(sensors/m2)
target sensor and 𝑁𝐵 is the number of received beacons by the
target sensor. Probability of receiving at least M beacons can be Fig. 5. Localization failure.
calculated by (6).

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
7

0.7 SSEDMAG: This algorithm differs from SEDMAG in the


way that it searches for the next target. It utilizes SWSA which
0.6 is a smart searching approach to shorten the drone trajectory.
Zcurve
SSEDMAG-reduced: It aims to curtail the drone trajectory
Mean location error(m)

SSEDMAG-reduced
0.5 length by applying the SSEDMAG on the reduced graph instead
SSEDMAG
SEDMAG of the original one.
0.4
SSEDMAG-reduced-balanced: This approach is applicable
0.3 in multiple drone scenarios. It is used to balance the localization
task among drones and shorten the localization time.
0.2 In the scenario investigated here, 𝑁 stationary sensors are
deployed non-uniformly over a two-dimensional monitoring
0.1
area of 100m×100m. Note that SEDMAG trajectory planner
0 can be applied to any deployment area with any shape. We use
0.008 0.016 0.024 0.032 0.040 0.048 a square-like monitoring region just as an example. The IoT
Network density(sensors /m2)
sensors graph is assumed to be connected. Sensors have the
Fig. 6. Mean location error in multi-drone scenario.
same transmission range i.e., 𝑟𝑐 ; drones have transmission
80 ranges greater than or equal to 𝑟𝑐 . We change 𝑁 in a way that
Average number of received beacons by each sensor

SSEDMAG-reduced the density goes from 0.008 to 0.048. RSS reading is the
70
Zcurve summation of the true value and a zero-mean gaussian noise
60 SSEDMAG with a variance equal to 3 dBm. Here, we consider TElosB
SEDMAG
50 motes equipped with an IEEE 802.15.4-compliant chipcon
CC2420 radio [22]. Therefore, we use (12) for pass loss 𝑃𝐿 as
40
suggested in Annex E of [23], in order to calculate path loss for
30 802.15.4-like networks.
20 𝑑 (12)
𝑃𝐿 = 33 log + 58.5
10 8
The parameters used in the simulations are listed in Table 1.
0
0.008 0.016 0.024 0.032 0.040 0.048
Network density(sensors / m2) B. Simulation Result
Fig. 7. Average number of received beacons by each sensor Extensive simulations are carried out using Matlab. In order
to make fair comparisons, Monte Carlo method is used with
N1 or its expected value which is the number of sensors in the simulations running for 100 times and in each run, a new
gray area of Fig. 4(b) can be calculated by (9). random deployment is used.
𝜋
(9) 1) Multi-drone Scenario
𝐸[𝑁1 ] = ∫ 𝑔(𝜃)𝑓(𝜃)𝑑𝜃 Here, SEDMAG variants and a static path (Zcurve) [13] are
0 compared.
𝜋2
= 𝛿(0.9 𝑟 7 − 0.74 𝑟 6 )( + 2)
2 a) Localization Coverage Ratio
Fig. 5 plots localization failure for SEDMAG variants as well
where, g(θ) is a random variable showing the number of
as static path planner. As expected, the failure rate for all
sensors in the gray area and f(θ) is its density function. we
SEDMAG variants, which provide all sensors with sufficient
have:
number of beacons, is 0%. However, in the static path case,
𝑔(𝜃) = δ 𝑟 2 (𝜋 − 𝜃 +𝑠𝑖𝑛 𝜃) (10) 2.5% to 2.7% of sensors do not receive enough beacons to be
localized. The reason behind full localization coverage of
𝑓(𝜃) = 0.9 𝑟 5 − 0.74 𝑟 4 (11)
SEDMAG and its variants is that they choose the beacon points
in the communication circle and vicinity of sensors, while in
Zcurve, drone broadcasts beacons at pre-determined locations
IV. PERFORMANCE EVALUATION regardless of the sensors distribution. As a result, SEDMAG
In this section, the simulation setup, the wireless channel and provides full localization coverage.
other related parameters as well as the results will be described.
b) Location Error
Fig. 6 shows the mean location error which is calculated as
A. Simulation Setup the Root Mean Square Error (RMSE). It is observed that
In this paper, we implemented DREAMS[17], Zcurve [13] sensors in the static path scheme experience more errors
and SEDMAG with its four variants.
SEDMAG: SEDMAG guides the drone to visit all deployed
sensors. It finds the shortest path to decide on the sensors’
visiting order and employs WSA to find the target sensor.

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
8

800 2 18
SEDMAG SSEDMAG-reduced(4beacons) SSEDMAG-reduced(3 beacons)
700 SSEDMAG Dreams 16 SSEDMAG-reduced(4 beacons)
Energy consumption(joule)

Energy consumption(Joule)
SSEDMAG-reduced SSEDMAG-reduced(3beacons) Dreams

Mean location error(m)


600 1.5
Zcurve 14
500
12
400 1
10
300
8
200 0.5

100 6

0 0 4
0.008 0.016 0.024 0.032 0.040 0.048 0.008 0.016 0.024 0.032 0.040 0.048 0.008 0.016 0.024 0.032 0.040 0.048
Network density(sensors/m2) Network density(sensors/m2) Sensor density(sensors/m2)

Fig. 8. Total energy consumption in multi-drone Fig. 9. Mean location error in single-mobile Fig. 10. Total energy consumption in the
scenario. anchor scenario. single-MA scenario

60
compared to SEDMAG variants, and they increase with the 50 40

density of sensors too. However, it is shown that the error is 50


40 30

approximately independent of the network density and remains 40


6
very low (under 0.3) for all the SEDMAG variants. SEDMAG 30
6 30 20
2
20 10
and its variants make drone broadcast beacon mostly near 3
20

10 0
sensors. This creates high quality beacons and increases 10

localization accuracy. This figure shows that location error for 0 2 3 0 2 -10

the reduced scheme is higher than SEDMAG and SSEDMAG. 10 20 30 40 50 60 70 20 30 40 50 60


-20
30 40 50 60 70

This is due to the fact that the number of beacons each sensor
receives in the SSEDMAG-reduced scenario is almost constant Fig. 11. Graph partitioning and time balancing algorithm..(a) initial graph. (b)
(ranging from 7.49 to 8.54), and independent of the density, SR is sub-region 2 and ER is sub-region 6, but it is helpless. (c). SR is sub-
which is much smaller than what is received in SEDMAG and region 2 and ER is sub-region 3 and it is helpful.
SSEDMAG, as shown in Fig. 7. The higher the network density
70
is, the more likely the removal of sensors will be. Therefore, the SSEDMAG-reduced-balanced
majority are localized without being targeted by a drone. This 60 SSEDMAG-reduced
Localization Latency(sec)

7.2% 12.5%
is not the case for SEDMAG and SSEDMAG, for which the 50 13.5%
21.7%
error is smaller and decreases slowly with the number of the 11.9%
40 12.3%
nodes. This can be explained by the fact that in these two
schemes, the number of the beacon points is proportional to the 30

number of deployed sensors. Therefore, higher network 20


densities lead to an increase in the number of received beacons,
10
and also reduction in the average distance of the beacon points
from sensors. 0
0.008 0.016 0.024 0.032 0.040 0.048
Network density(sensors/m2)

c) Energy Consumption Fig. 12. The Impact of time balancing on localization latency.
The energy consumption of drones in SEDMAG variants as
well as Zcurve are shown in Fig. 8 for different densities. 2) Single drone Scenario
Assume that 𝑒𝑡𝑝 and 𝑒𝑏 denote the energy consumed by a drone In this section, SSEDMAG-reduced, which outperforms both
for moving along a unit of distance and broadcasting a beacon, SEDMAG and SSEDMAG in terms of latency and energy
respectively. Let 𝐿𝑡 denote the drone trajectory length and 𝑁𝐵 consumption at the same error level, is compared with
show the number of beacon points. Then, 𝐸𝑡 can be calculated DREAMS [17]. As shown in Fig. 9 and Fig. 10, SSEDMAG-
by (13), as in [15]. reduced surpasses DREAMS in terms of error and energy
consumption. Furthermore, guiding drone over the reduced
𝐸𝑡 = 𝐿𝑡 × 𝑒𝑡𝑝 + 𝑁𝐵 × 𝑒𝑏 (13) graph and visiting only critical sensors in SSEDMAG-reduced
decreases the energy consumption in comparison to DREAMS,
According to (13), the traversed path and the number of which aims to visit all sensors on the original graph.
beacons sent determine the energy consumed. As expected, an According to similar behavior of the localization latency and
increase in energy consumption is observed when the density the energy consumption in (3) and (13), respectively, energy
rises, in both SEDMAG and SSEDMAG. This can be justified consumption is only studied in this paper.
by the increase in both the path length and the number of beacon
points. However, the SSEDMAG-reduced curve tends to 3) Evaluation of the Localization Time-balancing Scheme
remain flat. It outweighs the other algorithms in terms of energy The load balancing algorithm is simulated to investigate its
consumption. Zcurve energy consumption does not vary since impact on latency. Fig. 11 shows how flexible boundaries can
the drones move along pre-assigned paths which have no help in load balancing and reduction of the latency. In Fig.
dependence on sensor distribution or density. 11(a), localization latencies of drones in sub-regions 2, 3 and 6
are 51.5s, 42s and 32.2s, respectively. According to the load

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2020.3032347, IEEE Internet of
Things Journal
9

balancing algorithm, sub-region 2 with the maximum latency is networks," Wirel. Commun. Mob. Comput., vol. 13, no. 14, pp. 1324–
selected as SR and sub-region 6 will be ER. Fig. 11(b) shows 1336, 2011.
the new trajectories after transferring some boundary sensors [10] O. Chia-Ho and W.-L. He, "Path planning algorithm for mobile
anchor-based localization in wireless sensor networks," IEEE Sens. J.
from SR to ER. However, these new paths are helpless, since
, vol. 13, no. 2, pp. 466-475, 2013.
the localization latency of sub-regions 2 and 6 increases to 57.8s
[11] J. Jiang, G. Han, H. Xu, L. Shu and M. Guizani, "LMAT: Localization
and 51.45s. Therefore, sub-region 3 is selected as ER. This way, with a mobile anchor node based on trilateration in wireless sensor
the new trajectories which are shown in Fig. 11(c) result in a networks," in IEEE GLOBECOM Conf., 2011.
latency of 44.5s for sub-region 2 and 45.51s for sub-region 3. [12] G. Han, X.Yang, L.Liu, W. Zhang and M. Guizani, “A disaster
Fig. 12 depicts the impact of load balancing on the latency for management-oriented path planning for mobile anchor node-based
different network densities. It reveals that the balancing idea localization in wireless sensor networks,” IEEE Trans. Emerg. Topic.
Comput., 2017, DOI: 10.1109/TETC.2017.2687319.
reduces the latency by 7.2% to 21.7%.
[13] J. Rezazadeh, M. Moradi, K. Sandrasegaran and R. Farahbakhsh,
"Transmission Power Adjustment Scheme for Mobile Beacon-
V. CONCLUSION & FUTURE WORK Assisted Sensor Localization," IEEE Trans. Industr. Inform, vol. 15,
no. 5, pp. 2859-2869, 2019,
We proposed a semi-dynamic drone trajectory planning
algorithm, called SEDMAG, for localizing 6G IoT devices. We [14] J. Rezazadeh, M. Moradi, A. S. Ismail and E. Dutkiewicz, "Superior
path planning mechanism for mobile beacon-assisted localization in
further improved it to reduce localization latency and energy wireless sensor networks," IEEE Sens. J., vol. 14, no. 9, pp. 3052-
overhead at the cost of some localization error increase using a 3064, 2014.
smart search algorithm and a graph reduction approach. Our [15] S. Kouroshnezhad, A. Peiravi, M. S. Haghighi and Q. Zhang, “A
simulations reveal that SEDMAG and its variants provide lower mixed-integer linear programming approach for energy-constrained
errors compared to static path planners such as Zcurve and mobile anchor path planning in wireless sensor networks localization,”
Ad Hoc Networks, vol. 87, no. 2019, pp. 188-199, 2019.
outperform the existing dynamic algorithms like DREAMS in
[16] G. Han, Ch. Zhang, J.Jiang, X.Yang and M. Guizani, "Mobile anchor
error and energy consumption. Furthermore, we proposed a nodes path planning algorithms using network-density-based
load balancing approach which balances drones load in multi- clustering in wireless sensor networks ," J. Netw. Comput. Appl., vol.
drone scenarios and results in reduction of the localization 85, pp. 64–75, May 2017.
latency. Currently, SEDMAG and its variants divide the [17] X. Li, N. Mitton, I. Simplot-Ryl and D. Simplot-Ryl, "Dynamic
monitoring area into equal size sub-regions and dedicate one beacon mobility scheduling for sensor localization,"
drone to each sector. A better approach, which is planned for IEEE Trans. Parallel. Distrib. Syst., vol. 23, no. 8, pp. 1439 - 1452,
2012.
future work, is to partition the graph using the modularity
[18] E. Erdemir and T. E. Tuncer, “Path planning for mobile-anchor based
criterion [24]. This graph partitioning algorithm makes dense wireless sensor network localization: static and dynamic Schemes,”
sub-regions which may result in lower localization latency. Ad Hoc Netw., vol. 77, no. 2018, pp. 1-10, August 2018.
[19] Z. Zhong, D.-Y. Luo, S.-Q. Liu, X.-P. Fan and Z.-H. Qu, “An adaptive
References localization approach for wireless sensor networks based on gauss-
[1] M. Giordani, M. Polese, M. Mezzavilla, S. Rangan and M. Zorzi, markov mobility model,” Acta Automatica Sinica vol. 36, no. 11, pp.
"Toward 6G Networks: Use Cases and Technologies," in IEEE 1557 – 1568, November 2010.
Commun Mag, vol. 58, no. 3, pp. 55-61, 2020. [20] M. Sayad Haghighi, K. Mohamedpour, “Neighbor discovery: security
[2] S. Jacob et al., "A Novel Spectrum Sharing Scheme using Dynamic challenges in wireless ad hoc and sensor networks”, Trends in
Long Short-Term Memory with CP-OFDMA in 5G Networks," Telecommunications Technologies, Intech, 2010, pp. 693-714.
in IEEE Trans. Cogn. Commun. Netw, doi: [21] J. A. Bondy and U. Murthy, “Graph theory with applications,”
10.1109/TCCN.2020.2970697. Elsevier, New York, 1976.
[3] M. S. Omar et al., "Multiobjective Optimization in 5G Hybrid [22] Chipcon, Dallas, TX, USA. CC2420 2.4 GHz IEEE 802.15.4 / ZigBee-
Networks," in IEEE Internet Things J., vol. 5, no. 3, pp. 1588-1597, ready RF Transceiver [Online]. Available: http://www.chipcon.com
2018. [23] IEEE Std 802.15.4-2003, Part 15.4, Wireless medium access control
[4] L. Gupta, R. Jain, G. Vaszkun, " Survey of important issues in uav (MAC) and physical layer (PHY) specifications for low-rate wireless
communication networks," IEEE Commun. Surv. Tutor., vol. 18, no. personal area networks (LR-WPANs), 2003.
2, pp. 1123-1152, 2016. [24] M. E. J. Newman, “Modularity and community structure in networks,”
[5] S. Kouroshnezhad, A. Peiravi, M. S. Haghighi and A.Jolfaei, " An in Proc. National Academy of Sciences of the United States of
energy-aware drone trajectory planning scheme for terrestrial sensors America, vol. 103, no. 23, pp. 8577–8696, 2006.
Localization," Computer Communications, vol. 154, no. 2020, pp.
542-550, 2020.
[6] N. Toorchi, M.A Attari, M.S. Haghighi, Y. Xiang, “A Markov Model
of Safety Message Broadcasting for Vehicular Networks”, IEEE
Wireless Communications and Networking Conf., 2013
[7] E. Vinogradov, H. Sallouha, S. De Bast, M. M. Azari, S. Pollin,
"Tutorial on UAVs: a blue sky view on wireless communication,"
Journal of Mobile Multimedia, 2019, arXiv:1901.02306. [Online].
Available: http://arxiv.org/abs/1901.02306.
[8] M. Mozaffari, W. Saad, M. Bennis, Y. Nam and M. Debbah, "A
Tutorial on UAVs for Wireless Networks: Applications, Challenges,
and Open Problems," in IEEE Commun. Surv. Tutor, vol. 21, no. 3,pp.
2334-2360, 2019.
[9] G. Han, H. Xu, J. Jiang, L. Shu, T. Hara and S. Nishio, "Path planning
using a mobile anchor node based on trilateration in wireless sensor

2327-4662 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Warwick. Downloaded on November 09,2020 at 08:59:32 UTC from IEEE Xplore. Restrictions apply.

You might also like