UAV-Assisted Offloading via Satellite Backhaul for

Post-Disaster and Crowded Cellular Networks

Allafi Omran Mohamed Cheriet Lokman Sboui
Electrical Engineering Department Systems Engineering Department Systems Engineering Department
College of Technical Sciences École de Technologie Supérieure (ÉTS), École de Technologie Supérieure (ÉTS),
Bani Waleed, Libya University of Québec, Montreal, Canada University of Québec, Montreal, Canada
[email protected] [email protected] [email protected]

Abstract—In this paper, our aim is to enhance network users in a fixed mobile base station area, to improve network
resilience and efficiency in post-disaster and crowded cellular throughput. [11] focused on minimizing user latency while
networks by integrating unmanned aerial vehicles (UAVs) and maintaining minimum QoS, and [12] proposed an ARIMA-
satellites. The proposed two-step network recovery involves de-
termining the optimal 3D placement of UAVs to maximize user based predictive approach for multi-UAV positioning to im-
coverage in disrupted or congested areas, followed by allocating prove traffic handling capacity. In [13], the authors proposed
each UAV to a satellite for reliable backhauling and efficient to deploy UAVs as a satellite relay for a cooperative mobile
offloading, while considering satellite visibility and UAV caching communication system [14]. Therefore, there is increasing
constraints. The numerical results demonstrate the superior interest in the use of UAVs as alternative terrestrial BS in
performance of the proposed 3D placement algorithm compared
to the existing methods. Additionally, we analyze and discuss post-disaster scenarios [15].
the impact of the number of satellites on backhauling delay and Given the UAV’s lack of wired connection to the core
cached demand at UAVs. network, satellite backhauling has emerged as a viable and
Index Terms—3D UAV deployment, Base station conges- efficient solution [16]. Specifically, LEO satellites, due to their
tion/damage, Network resilience, Satellites Backhauling. proximity to earth compared to the classical Geostationary
(GEO) satellites, can provide shorter transmission delays and
I. Introduction lower signal attenuation. Previous works have explored differ-
In the evolving landscape of 5G and beyond wireless ent approaches to UAV backhauling. Authors in [16] proposed
networks, unmanned aerial vehicles (UAVs) are gaining recog- a GEO satellite-based backhauling solution using a mixed-
nition as viable complements to existing infrastructure. UAVs integer optimization problem. On the other hand, in [17], the
have significant potential in serving as aerial infrastructure, authors introduced the concept of ”big data on the fly,” where
particularly when traditional infrastructures are overloaded or computation and caching are performed at the UAV Edge.
non-operational. Circumstances such as extreme weather or In [18], the authors presented a direct-to-satellite LoRaWAN
high traffic events can overwhelm the existing communication communications design for IoT without the need for a satellite
infrastructure. Therefore, the deployment of temporary UAVs ground station. Furthermore, the altitude of the satellites can
during such instances is critical to ensure continuity of service be used to calculate the revisit time [19].
and, at times, survival [1]. These previous studies have investigated the deployment of
Multiple studies have addressed 2D and 3D deployment UAVs to optimize network coverage, reduce the number of
problems, aiming to maximize coverage, minimize UAV num- deployed UAVs, and maximize system throughput. However,
bers, and increase system throughput [2]–[4]. The authors the critical challenge we address in this paper is the joint
in [2] explored a dual-UAV deployment strategy to improve deployment and backhauling of UAVs using LEO satellites for
coverage area and reduce power requirements. In [3], the a rapid coverage solution. We introduce a heuristic algorithm
authors offered a K-means clustering algorithm for offloading a to determine 3D UAV locations, optimizing user coverage by
portion of users, utilizing UAVs to supplement existing macro- considering various factors such as user density, geographical
cells. In terms of 3D deployment, [5] developed an algorithm barriers, and real-time demands. Furthermore, we present a
to determine the 3D position of UAVs and user association Min-Max algorithm to ensure that UAVs transmit data to LEO
to maximize spectrum efficiency. Similarly, [6] optimized the satellites with minimized the unfulfilled demand to reduce the
placement, allocation of bandwidth, and power of 3D UAVs overall communications latency.
to maximize throughput, while meeting the user data rate The remainder of the paper is organized as follows. In
requirements [7], [8]. In [9], the authors proposed a multi- Section II, we introduce the system model. In Section III,
UAV deployment to maximize the fairness of the achieved rate we propose our solution for UAV-Users association. Then, in
among users using a swarm optimization heuristic algorithm. Section IV, we address the UAV-Satellites association problem.
On the other hand, in [10], the authors addressed multiple UAV In Section V, we present and analyze the simulation results.
deployment algorithms using fuzzy k-means clustering to serve Finally, the conclusion is presented in Section VI.
 II. System Model by ηNLoS and ηLoS . Thus, we can calculate the average path
We consider a scenario in which multiple fixed base stations loss as follows:
!
are disabled or congested due to critical situations, such as 4π fc di, j
PLi, j = 20 log10 + ηLoS Pi,LoS
j +η Pi, j , (4)
NLoS NLoS
natural disasters or high-demand events. We propose the use c
of UAVs as temporary base stations, during a deployment
Using the path loss we calculate the signal-to-noise ratio SNR
duration D, with satellite-based backhaul to restore coverage
of the user i served by UAV j given by:
and manage capacity, as shown in Fig.1.
SNRi, j = Pti, j − Pn − PLi, j , (5)
where Pti, j is the UAV’s transmitted power in dB and Pn is the
average power of noise in dBm assumed to be the same for
all UAVs. The corresponding achievable data rate is given by:
Ratei, j = B j × log2 (1 + SNRi, j ), (6)
where B j stands for the resource bandwidth of the uav j , SNR
is the signal to noise ratio. Our objective is to find the number
of UAVs and their position to maximize the number of served
users with a minimum rate Rmin , i.e., Ratei, j ≥ Rmin .
B. UAV-to-Satellite Link
Healthy UAV as aerial
Satellite
base station base station Once the UAV placement is performed, each j-th UAV has
Backhaul link
Damaged/congested Users to be a specific communication demand, D j,t given by
base station offloaded Coverage link
Nu
X
Figure 1: System model D j,t = Ratei, j δi, j , ∀ j, t, (7)
i=1
A. UAV-to-Users Link which needs to be backhauled via the LEO satellites. The
We consider NU users, randomly distributed in a geographic deployment period D is discretized into NT time slots, each
region with fixed locations and need to be offloaded, with the of duration T . Each UAV can be associated with one or
coordinates (xi , yi ) of the ith user. We aim to deploy NUAV more satellites during T , provided that the satellite has direct
UAVs with 3D coordinates (X j , Y j , Z j ) of the jth UAV. The visibility with the UAVs during its orbital pass. This orbital
UAVs have the same service capacities and power consump- pass is assumed to be aligned with the duration of a time slot
tion. The maximum number of resource blocks available for and can be a multiple of T . We denote by x j,k,t the backhauled
each UAV is denoted RBmax . In addition, each user can be demand of the UAV j fulfilled by the satellite k in the time
associated with only one UAV. We model the channel between slot T t .
the UAV and the users by the air-to-ground channel model The total demand for associated UAVs per satellite cannot
sat
considering both the line of sight (LoS) and non-line of sight exceed the capacity of the kth satellite, Ck,t , at the time slot
PNUAV sat
(NLoS) components independently. The probabilities of their T t , i.e., j=1 x j,k,t ≤ Ck,t . In cases where visible satellite
occurrence for an ith user and a jth UAV are [20]: capacity is insufficient, UAVs have the ability to cache unfilled
1 demand during T t and backhaul it when more satellites become
PLoS = , PNLoS = 1 − PLoS , (1) visible [17] . We define U j,t as the unfulfilled demand for UAV
1 + a exp(−b( 180
π θi, j − a)) j after allocating backhaul resources during T t−1 .
where a and b are constants that depend on the system N sat
environment, while θi, j represents the elevation angle estab-
X
U j,t = D j,t − x j,k,t , ∀ j, t. (8)
lished between a user and the corresponding UAV defined as k=1
Z
tan−1 ( di,jj ), where Z j is the height of the UAV j, and di, j is the
We introduce C cache
j as the maximum cache capacity for the
distance between the user i and the UAV j. The corresponding
UAV j. Therefore, the unfulfilled demand cannot exceed the
path loss for both LoS and NLoS can be calculated by:
! maximum UAV cache capacity:
4π fc di, j
PLi,LoS
j = 20 log 10 + ηLoS , (2) U j,t ≤ C cache
j , ∀ j, t. (9)
c
4π fc di, j
! The caching process inherently introduces a delay, com-
PLi, j = 20 log10
NLoS
+ ηNLoS , (3) monly known as a queuing delay. This is due to the data being
c stored (or ”queued”) in the cache at a time slot until it is ready
where, fc is the carrier frequency and c represents the speed of to be dispatched to its destination in a future time slot. In our
light. The average extra losses to free space propagation loss, model, we calculate an average total delay, DelayAvg,T ot , as
contingent upon the surrounding environment, are represented the sum of delays, for all the time slots, associated with the
fraction of demand that has been cached at all the UAVs for can serve only up to its maximum capacity, represented by the
all the period: number of resource blocks RBmax . Equation (15) ensures that
NT N UAV
a user can only connect to a single UAV, and (16) indicates
X X U j,t the all-or-nothing nature of UAV-User associations.
DelayAvg,T ot = × T. (10)
t=1 j=1
D j,t Given the binary constraint in (16), the problem is a
mixed integer programming problem and is non-deterministic
C. Satellite Visibility polynomial-time hard (NP-hard). As the problem doesn’t have
Given the high-speed orbits of LEO satellites, their visibility an analytical solution, and an exhaustive search method can
to UAVs is limited to short durations. Hence, a satellite approximate the solution, we propose a simpler algorithm
constellation is employed to ensure continuous backhauling based on a partition technique.
capabilities. Satellites experience orbital pass periods where
they are directly visible to UAVs, and this visibility is a critical B. Proposed Solution for UAV-User Association
factor influencing backhaul efficiency [19]. Table I presents
Our UAV-User association solution comprises two main
an example of satellite visibility over various time periods.
steps: estimating the preliminary number of UAVs and finding
Each row pertains to a different satellite, while each column
their 2D locations.
corresponds to a distinct time period. The entries in the table
denote whether the kth satellite is visible to the UAVs at 1) Estimation of Preliminary Number of UAVs: The first
time T , represented by a ’1’ for visible and a ’0’ for not visible. step in our solution involves approximating the minimum num-
This table can be built in advance using the two-line element ber of UAVs, N̂UAV , required to service all mobile users. This
(TLE) data that give the approximate location of satellites. estimate is derived from the data provided by the primary base
station before any disaster or outage occurs. The estimated
Table I: Example of satellites visibility with different periods. number can be adjusted later based on the dispersion of the
user distribution. We express this mathematically as:
Duration D  N 
U
Sat. T1 T2 T3 T4 · · · T NT −2 T NT −1 T NT N̂UAV = , (17)
S1 1 1 0 0 ··· 0 0 0 RBmax
S2 0 1 1 1 ··· 0 0 0 where, ⌈·⌉ denotes the ceiling operator.
.. .. .. .. .. .. .. .. .. 2) Determining 2D Locations of UAVs:
. . . . . . . . . a) Initial 2D UAV Locations: Our approach to determine
S NS at −1 0 0 0 0 ··· 1 1 1 UAVs’ 2D locations combines the K-means method and the
S NS at 0 0 0 0 ··· 0 1 1 pattern search (PS) technique [21]. The K-means method,
known for efficiently clustering unlabeled data, provides the
III. UAV-Users Association initial UAV locations [22]. The aim is to group N̂U users
A. UAV-Users Problem formulation into N̂UAV UAVs. The K-means algorithm starts with N̂UAV
In the UAV-Users association, we aim to maximize the random UAV locations, assigns users to the nearest UAV,
number of served users by determining the number and posi- and relocates UAVs to the user centroid, repeating until UAV
tions of UAVs to be deployed. The corresponding optimization locations stabilize [23].
problem can be expressed as: b) Final 2D UAV Locations: To derive the final 2D loca-
tions, we use the pattern search algorithm, utilizing the initial
NU
X locations obtained via the K-means algorithm as a starting
max δi, j (11)
NUAV ,X j ,Y j ,Z j point. The pattern search algorithm’s objective is to fine-tune
i=1
the UAV locations such that the sum of the differences between
subject to: each UAV’s capacity and the number of users it serves, denoted
Ratei, j ≥ Rmin , ∀i, j, (12) by NU, j is minimized. We define these differences using the
NUAV ≤ NUAV max
, (13) objective function SN̂UAV as follows:
NX NU
UAV s X
N̂ UAV
δi, j ≤ RBmax , ∀ j (14)
X
SN̂UAV (x1 , y1 , · · · , XN̂UAV , YN̂UAV ) = RBmax − NU, j (18)
j=1 i=1
j=1
NX
UAV s

δi, j = 1, ∀ j, (15) The pattern search algorithm initiates by pinpointing a direc-

j=1 tion around the initial 2D plane locations where a movement
by step ∆ minimizes the objective function. The algorithm
∈ {0, 1} , (16)
transitions to this new position and repeats the process until
where δi, j is a binary variable of the assocation of the UAV j there are no further reductions in the objective function. If an
and the user i. Constraint (12) ensures that the minimal rate of increase in the objective function is noted at the new position,
each user is respected. Equation (13) shows that the number of ∆ is decremented and the pattern search algorithm restarts until
available UAVs is capped. Equation (14) means that each UAV no further optimization is achievable.
 3) Evaluating Additional UAV Deployment: Upon estab- In the above optimization problem, the objective function
lishing the final 2D locations, it is possible that some users (20) aims to minimize the difference between the total demand
may not be serviced due to their location outside a UAV’s cov- (D j,t ) of each UAV and the cumulative x j,k,t (the demand of
erage area or because they have recently entered the system. UAV j backhauled by satellite i at time t). The constraint
Consequently, it becomes necessary to assess if deploying an (21) ensures that the total demand serviced by all satellites
additional UAV would effectively serve these users. The total for each UAV during all periods does not surpass the total
number of unserved users, denoted as NUnon−served , is calculated demand of the UAVs. The constraint (22) asserts that the
as NUnon−served = NU − NUserved . Here, NUEserved
represents the amount of cached data remains within the caching capacity
number of users served after the search phase. Subsequently, of the UAV. The constraint (23) ensures that the cumulative
N̂UAV can be updated once again according to equation (17). In demand assigned to each satellite in each time period cannot
the event that N̂UAV is incremented, the process of identifying exceed the capacity of the satellite. The constraint (24) is a
new 2D locations through clustering and search is repeated non-negativity constraint, stating that the backhauled demand,
until there is no further need for additional UAVs. x j,k,t , cannot be negative.
4) Determining UAV Altitudes: The altitude for the jth UAV Due to the presence of both discrete and continuous vari-
is computed based on the distance to the furthest user with ables in the UAV-satellite association, x j,k,t , this optimization
index i∗ within its coverage area, represented as di∗ , j . The problem is a mixed-integer non-linear programming (MINLP)
altitude of the UAV, Z j , is problem, which is typically NP-hard. To address this, we
di∗ , j introduce a Min-Max allocation strategy.
Zj = , (19)
tanθi∗ , j B. UAV-Satellite Association with Min-Max Allocation
Algorithm 1 presents a summary of the proposed scheme. The proposed solution is based on the Min-Max fairness
principle [24]. This strategy favors the UAVs with the most
Algorithm 1: Proposed UAV 3D Deployment cached data, thus promoting an equitable distribution. The
algorithm for the Min-Max fairness allocation is presented
Input : (xi , yi ), a,b, ηLoS , ηNLoS , γ,NUAV max
s below:
Compute the preliminary number of UAVs, N̂UAV ;
Derive initial coordinates (X j , Y j ) utilizing K-means;
Algorithm 2: Min-Max Fairness Allocation Algorithm
Determine the distance di, j applying (19);
sat
Ascertain the 2D positions, (X j , Y j ), by segmenting Input : D j,t , NUAV , NS at , T , Ck,t , C cache
j
users into equivalently-sized clusters using PS.; Output: x j,k,t , U jt
Compute UAV altitudes Z j via (19); For each period T t , t from 1 to NT ;
Refresh user statuses (Inspect for incoming users); Compute D∗j,t = D j,t + U jt−1 for all j ;
if additional UAV is required then Find j∗ = arg min j U j,t ;
Augment N̂UAV and revert to the step to derive For each Satellite k, k from 1 to Nvisib−sat,t ;
initial coordinates; if D∗j,t ≤ Ck,tsat
then
Finalize NUAV = N̂UAV ; Allocate D∗j,t from satellite k to UAV j, i.e.
Output: NUAV , and (X j , Y j , Z j ) x j,k,t = D∗j,t ;
Update Ck,t sat
= Ck,t
sat
− D∗j,t , D∗j,t = 0 and U j,t = 0 ;
else
IV. UAV-Satellite Association Allocate Ck,t sat
from satellite k to UAV j, i.e.
A. UAV-Satellite Problem Formulation x j,k,t = Ck,t ;
sat

In this step, our aim is to effectively backhaul UAV data Update Ck,t sat
= 0, D∗j,t = D∗j,t − Ck,t sat
and
by minimizing the overall unfulfilled demand across all UAVs U j,t = D j,t − Ck,t ;
∗ sat

and time intervals. The following optimization problem is: end if ;

  end for ;
N
X NT 
UAV X NSat
X 

min D j,t − x j,k,t  (20)
x j,k,t
j=1 t=1 k=1 This approach ensures that satellite resources are fairly
NSat
NT X
X distributed among UAVs, according to the Min-Max fairness
subject to x j,k,t ≤ D j,t , ∀ j, t, (21) 2
principle. The complexity of this algorithm is O(NUAV ), since
t=1 k=1 each allocation operation requires finding the UAV with the
U j,t ≤ C cache
j , ∀ j, t, (22) maximum cached data among all UAVs.
N
X UAV V. Numerical Results
sat
x j,k,t ≤ Ck,t , ∀k, t, (23)
We consider a geographical area where the UAVs are de-
j=1
ployed to serve multiple UEs, and its connection is backhauled
x j,k,t ≥ 0, ∀ j, k, t. (24) by a satellite. We consider a square area of 1 km where users
 100
are distributed in this area according to the Poisson Point Cellular Scheme
Process (PPP). Table 1 presents a summary of the various Kmeans Scheme
90 Proposed Scheme
parameters adopted.

Users Serving Effeciciency

80
Table II: Simulation Parameters
70
Parameter Value Parameter Value
Simulation area 1 km2 User density 225 /km2 60
Umax (per UAV) 25 users ηLoS 1 dB
ηNLoS 20 dB Pmax 30 dBm 50
Bblock 180 kHz a (Path loss) 9.61
Nnoise -174 dBm/Hz b (Path loss) 0.16
40
RBmax 25 fcarrier 2 GHz
Sat. Visibility 10 min Orbital Revolution 90 min
30
Max UAV Rate 20 Mbps Satellite Capacity 100 Mbps 50 100 150 200 250 300 350 400
Total Duration 60 min Caching Period 5 min Number of offloadable users

Figure 3: Serving efficiency vs users (10 UAVs).

1000 250
Fixed Cellular Scheme
K-Means Scheme RB = 225
Proposed Scheme

Number of served users in the system

800 200
RB = 180
Area Width in (m)

600 150
RB = 135

400 100 RB = 90

200 50
Coverage of Clusters
Damaged/Congested BS
Clusters' Centroids - UAV locations
0 Locations of users 0
0 200 400 600 800 1000 90 135 180 225
Area Length in (m) Total Number of offloaded users in the system

Figure 2: UAVs placement and user association (10 UAVs). Figure 4: Offloading Efficiency vs. users number (9 UAVs).

Figure 2 depicts the 2D positioning of UAVs determined by

our proposed scheme. The figure illustrates the user associa- resources align perfectly with the user demand, maximizing
tions with respective UAVs, a step crucial in ensuring optimal the efficiency of our algorithm.
Quality of Service (QoS). We therefore show that the proposed In both Figure 5 and Figure 6, we conduct a Monte Carlo
algorithm hence achieves a balanced load distribution across simulation with 1000 iterations to calculate the average delay
all UAVs. and cached data for various scenarios. Figure 5 highlights the
Figure 3 highlights the service efficiency (measured as the dependency of communication delay on the number of satel-
ratio of serviced users to total users) of our method compared lites per constellation. We show that when only one satellite
to traditional cellular and K-means methods (without pattern is operational, the average delay exceeds 35 minutes with an
search). For example, our method serves more users when extra 2 to 15 minutes per additional UAV, which is not suitable
NU = 400 with 16 UAVs. The effectiveness of our method for real-time communications. We also show that beyond 10
comes from using UAV mobility and real-time positioning op- satellites the delay becomes relatively negligible, allowing the
timization based on user distribution. Unlike K-means and cel- use of real-time communication such as emergency warning
lular methods that focus on planned deployment, our method messages.
considers the user-to-UAV distance during deployment, which In Figure 6, we illustrate the interdependence between the
leads to better UAV distribution and service efficiency. percentage of cached demand and the number of satellites. We
Figure 4 demonstrates the user offloading performance of show that a minimum of 6 satellites are needed to avoid UAV
our proposed scheme in comparison to two benchmark meth- caching. We also show that the addition of a UAV can increase
ods: fixed-mobile cellular and K-means schemes. This figure the caching percentage by approximately 50% when we have
shows that our proposal successfully serves more users across less than 12 satellites. However, with 18 satellites, UAVs stop
varying network capacities, thereby outperforming the baseline caching entirely which highlight the crucial role of satellites
strategies. In particular, in the 180 users scenario, the system in managing UAV demand caching.
 60
NUAV=20 [3] B. Galkin, J. Kibilda, and L. A. DaSilva, “Deployment of UAV-mounted
NUAV=15 access points according to spatial user locations in two-tier cellular
50
NUAV=10
networks,” in 2016 Wireless Days (WD), March 2016, pp. 1–6.
[4] L. Sboui, H. Ghazzai, Z. Rezki, and M.-S. Alouini, “Energy-Efficient
NUAV=5
Power Allocation for UAV Cognitive Radio Systems,” pp. 1–5, 2017.
Average Delay (min)

40 [5] X. Sun, N. Ansari, and R. Fierro, “Jointly optimized 3D drone mounted

base station deployment and user association in drone assisted mobile
access networks,” IEEE Trans. on Vehic. Tech., vol. 69, no. 2, pp. 2195–
30 2203, 2020.
[6] S. Zhang and N. Ansari, “3D drone base station placement and resource
allocation with fso-based backhaul in hotspots,” IEEE Trans. on Vehic.
20
Tech., vol. 69, no. 3, pp. 3322–3329, 2020.
[7] L. Sboui, H. Ghazzai, Z. Rezki, and M.-S. Alouini, “Achievable Rates
of UAV-Relayed Cooperative Cognitive Radio MIMO Systems,” IEEE
10
Access, vol. 5, pp. 5190–5204, 2017.
[8] L. Sboui, H. Ghazzai, and Y. Massoud, “Integrating Cognitive Radio
0
MIMO UAVs in Cellular Networks for 5G and Beyond,” in 2022
2 4 6 8 10 12 International Conference on Electrical, Computer, Communications and
Number of Satellites per Constellation Mechatronics Engineering (ICECCME), 2022, pp. 1–6.
[9] A. Akarsu and T. Girici, “Fairness aware multiple drone base station
Figure 5: Average back-hauling delay vs. number of satellites. deployment,” IET Communications, vol. 12, no. 4, pp. 425–431, 2018.
[10] S. Afrin, M. S. Hossain, M. R. Iqbal, A. Refat, and A. U. Tamim,
“A fuzzy logic approach for improving throughput of the UAV-assisted
100
wireless networks,” in 2021 IEEE Int. Conf. on Signal Processing,
NUAV=20
Information, Communication Systems (SPICSCON), 2021, pp. 91–95.
90 [11] S. Zhang and N. Ansari, “Latency aware 3D placement and user
NUAV=15
association in drone-assisted heterogeneous networks with fso-based
80 N =10
backhaul,” IEEE Trans. on Vehic. Tech., vol. 70, no. 11, pp. 11 991–
Percentage of Cached Demand

UAV
N =5 12 000, 2021.
UAV
70
[12] B. Mahapatra, A. Verma, D. Gupta, P. K. Sharma, and A. K. Turuk,
60 “Traffic-aware UAV placement strategies for load balancing in 5g cellular
hotspots,” in 2021 Advanced Communication Technologies and Signal
50 Processing (ACTS), 2021, pp. 1–6.
[13] T. Akiyoshi, E. Okamoto, H. Tsuji, and A. Miura, “Performance
40 improvement of satellite/terrestrial integrated mobile communication
30
system using unmanned aerial vehicle cooperative communications,” in
2017 Int. Conf. on Information Networking (ICOIN), 2017, pp. 417–422.
20 [14] L. Sboui, H. Ghazzai, Z. Rezki, and M.-S. Alouini, “On the throughput
of cognitive radio MIMO systems assisted with UAV relays,” in 2017
10 13th International Wireless Communications and Mobile Computing
Conference (IWCMC), 2017, pp. 939–944.
0
2 4 6 8 10 12 14 16 18 20 [15] M. Matracia, N. Saeed, M. A. Kishk, and M.-S. Alouini, “Post-disaster
Number of Satellites per Constellation communications: Enabling technologies, architectures, and open chal-
lenges,” IEEE Open Jour. of the Communications Society, vol. 3, pp.
Figure 6: Cached demand vs. number of satellites. 1177–1205, 2022.
[16] Y. Zhu, W. Bai, M. Sheng, J. Li, D. Zhou, and Z. Han, “Joint UAV
access and geo satellite backhaul in iort networks: Performance analysis
and optimization,” IEEE Internet of Things Jour., vol. 8, no. 9, pp. 7126–
VI. Conclusion 7139, 2021.
In this paper, we proposed a novel scheme for optimal 3D [17] J. Xu, K. Ota, and M. Dong, “Big data on the fly: UAV-mounted mobile
edge computing for disaster management,” IEEE Trans. on Network
UAV deployment and UAV-satellite associations. The UAV Science and Engineering, vol. 7, no. 4, pp. 2620–2630, 2020.
deployment procedure used K-means clustering and a pattern [18] J. A. Fraire, S. Henn, F. Dovis, R. Garello, and G. Taricco, “Sparse
search algorithm to determine UAV positions and optimized Satellite Constellation Design for LoRa-based Direct-to-Satellite Internet
of Things,” in GLOBECOM 2020 - 2020 IEEE Global Communications
altitudes. The UAV-satellite resource allocation problem was Conference, 2020, pp. 1–6.
solved using a Min-Max fairness principle algorithm to ensure [19] P. Zong and S. Kohani, “Optimal satellite leo constellation design
an equitable resource distribution. Numerical simulations vali- based on global coverage in one revisit time,” Int. Jour. of Aerospace
Engineering, vol. 2019, pp. 1–12, 2019.
dated the effectiveness of our approach, demonstrating reduced [20] A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal lap altitude for
communication delays and improved network performance. maximum coverage,” IEEE Wireless Comm. Letters, vol. 3, no. 6, pp.
The results support the potential of UAV-satellite integrated 569–572, 2014.
[21] V. Torczon, “On the convergence of pattern search algorithms,” SIAM
networks for resilient communication systems, particularly in Jour. on optimization, vol. 7, no. 1, pp. 1–25, 1997.
challenging environments. [22] J. A. Hartigan and M. A. Wong, “Algorithm as 136: A k-means clustering
algorithm,” Jour. of the Royal Statistical Society. Series C (Applied
References Statistics), vol. 28, no. 1, pp. 100–108, 1979.
[23] A. Omran, L. Sboui, M. Kadoch, Z. Chang, J. Lu, and R. Liu, “3D
[1] N. Sharma, M. Magarini, D. N. K. Jayakody, V. Sharma, and J. Li, “On- deployment of multiple UAVs for emergent on-demand offloading,” in
demand ultra-dense cloud drone networks: Opportunities, challenges and Int. Wireless Comm. and Mobile Comp. (IWCMC), 2020, pp. 692–696.
benefits,” IEEE Communications Magazine, vol. 56, no. 8, pp. 85–91, [24] B. Radunovic and J.-Y. Le Boudec, “A unified framework for max-min
August 2018. and min-max fairness with applications,” IEEE/ACM Transactions on
[2] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Drone small cells Networking, vol. 15, no. 5, pp. 1073–1083, 2007.
in the clouds: Design, deployment and performance analysis,” in 2015
IEEE Global Communications Conf. (GLOBECOM), Dec 2015, pp. 1–6.