2946 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO.

10, OCTOBER 2021

Hovering UAV-Based FSO Communications:

Channel Modelling, Performance Analysis,
and Parameter Optimization
Jin-Yuan Wang , Member, IEEE, Yang Ma, Rong-Rong Lu, Jun-Bo Wang , Member, IEEE,
Min Lin , Member, IEEE, and Julian Cheng , Senior Member, IEEE

Abstract— Relay-assisted free-space optical (FSO) communi- locations. Numerical results show that the derived theoretical
cation systems are exploited as a means to mitigate the limit- expressions are accurate to evaluate the outage performance of
ing effects of the turbulence induced atmospheric scintillation. the system. Moreover, the proposed optimization schemes are
However, conventional ground relays are stationary, and their efficient and can improve performance significantly.
optimal placement is not always feasible. Due to their mobility
and flexibility, unmanned aerial vehicles (UAVs) provide new Index Terms— FSO communications, outage probability,
opportunities for FSO relaying systems. In this paper, a hovering parameter optimization, relay, UAV.
UAV-based serial FSO decode-and-forward relaying system is
investigated. In the channel modelling for such a system, four I. I NTRODUCTION
types of impairments (i.e., atmospheric loss, atmospheric turbu-
lence, pointing error, and link interruption due to angle-of-arrival
fluctuation) are considered. Based on the proposed channel
model, a tractable expression for the probability density function
R ECENTLY, free-space optical (FSO) communications
has attracted considerable attention to overcome the
spectrum congestion problem [2], [3], because unlike the
of the total channel gain is obtained. Closed-form expressions of radio frequency (RF) wireless communications, FSO commu-
the link outage probability and end-to-end outage probability are nications are unlicensed, directional, immune to electromag-
derived. Asymptotic outage performance bounds for each link
and the overall system are also presented to reveal insights into netic interference, and not easily intercepted. Nevertheless,
the impacts of different impairments. To improve system per- it is widely acknowledged that the terrestrial FSO signal is
formance, we optimize the beam width, field-of-view and UAVs’ impaired by three important factors including atmospheric
loss, atmospheric turbulence, and pointing error, which are
Manuscript received October 15, 2020; revised March 2, 2021; accepted
April 12, 2021. Date of publication June 14, 2021; date of current ver- all distance-dependent [4]. Moreover, the transceivers in FSO
sion September 16, 2021. This work was supported in part by the Open communications are limited by the strict requirement of line-
Research Fund of Key Laboratory of Broadband Wireless Communication of-sight (LoS) alignment [5]. These disadvantages lead to the
and Sensor Network Technology under Grant JZNY202115, in part by
the Open Project of Shanghai Key Laboratory of Trustworthy Computing, development of the relay-assisted FSO communications [6],
in part by the Opening Foundation of Key Laboratory of Opto-Technology whereby relays are placed properly between the source and
and Intelligent Control, Ministry of Education under Grant KFKT2020-06, the destination to improve system performance and reliability.
in part by the National Key Research and Development Program under
Grant 2018YFB1801905, in part by the National Natural Science Foundation However, affected by the obstacles (such as lakes, mountains
of China under Grant 61960206005 and Grant 61960206006, in part by and buildings), the optimal positions to deploy relays may not
the Jiangsu Province Basic Research Project under Grant BK20192002, be always feasible, and thus novel relaying schemes should be
and in part by the Key International Cooperation Research Project under
Grant 61720106003. This article was presented at the IEEE International exploited.
Conference on Communications, Dublin, Ireland, 2020 [1]. (Corresponding To revolutionize the commonly-employed relaying net-
authors: Jin-Yuan Wang; Jun-Bo Wang.) work architectures, the unmanned aerial vehicle (UAV) based
Jin-Yuan Wang is with the Key Laboratory of Broadband Wireless Com-
munication and Sensor Network Technology, Nanjing University of Posts and relaying scheme was proposed for FSO communications [7].
Telecommunications, Nanjing 210003, China, also with the Shanghai Key Compared with the conventional terrestrial relays, UAV-based
Laboratory of Trustworthy Computing, East China Normal University, Shang- relays have the intrinsic advantages of finding better com-
hai 200062, China, and also with the Key Laboratory of Opto-Technology
and Intelligent Control, Lanzhou Jiaotong University, Lanzhou 730070, China munication environment and establishing the LoS links by
(e-mail: [email protected]). adjusting positions dynamically. With the perfect combination
Yang Ma and Jun-Bo Wang are with the National Mobile Communications of FSO communications and UAVs, the UAV-based FSO com-
Research Laboratory, Southeast University, Nanjing 211111, China (e-mail:
[email protected]; [email protected]). munication system is currently gaining significant attention,
Rong-Rong Lu and Min Lin are with the College of Telecommunications which is regarded as a promising technology in many fields,
and Information Engineering, Nanjing University of Posts and Telecom- such as emergency response and military operation [8]–[10].
munications, Nanjing 210003, China (e-mail: [email protected];
[email protected]). It is expected that employing UAVs in FSO communication
Julian Cheng is with the School of Engineering, The University of British systems will inspire promising and innovative applications for
Columbia, Kelowna, BC V1V 1V7, Canada (e-mail: [email protected]). future communication systems.
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/JSAC.2021.3088656. For UAV-based FSO relaying systems, accurate channel
Digital Object Identifier 10.1109/JSAC.2021.3088656 modeling, tractable performance indicator expressions and
0733-8716 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2947

efficient parameters optimization methods are of vital impor- How to efficiently find LoS link and obtain satisfied system
tance. Several analytical models for hovering UAV based FSO performance for UAV-based FSO communication systems with
communications were proposed [11]. In [12], the alignment obstacles? Under the obstacle scenario, the UAV’s location was
and stability analyses for inter-UAV communications were pre- optimized within a feasible region bounded by the minimum
sented. In [13], the ergodic sum rate for a UAV-based relay net- accepted elevation angles of source and destination [26]. Note
work with mixed RF/FSO channel was derived, and the joint that the AoA fluctuation in [26] is ignored, the considered log-
effects of the atmospheric loss, the atmospheric turbulence, normal distributed turbulence is only suitable for weak turbu-
and the geometric and misalignment loss on the FSO channel lence, and the feasible region limits the deployment of UAVs.
were considered. In [14], the throughput for a UAV-based Therefore, the parameter optimization problem in UAV-based
mixed FSO/RF system with a buffer constraint was analyzed. FSO communications should be further investigated.
Note that the effects of the angle-of-arrival (AoA) fluctuations Motivated by the aforementioned discussions, this paper
of due to orientation deviations of hovering UAVs were not focuses on a hovering UAV-based serial relaying FSO commu-
considered in [11]–[14]. By considering the non-orthogonality nication system, and further investigates the channel modeling,
of the laser beam and the random fluctuations of the UAV’s performance analysis and parameter optimization problems.
orientation and position, a novel channel model was proposed The contributions of this paper are summarized as follows:
for UAV-based FSO communications [15]. In [16], the FSO • Channel modeling: The considered FSO system involves
channel impaired by weak turbulence, pointing error and AoA three kinds of links, i.e., the ground-to-UAV (GU) link,
fluctuation was established. By considering the AoA fluctua- the UAV-to-UAV (UU) link, and the UAV-to-ground (UG)
tion and pointing error, the outage performance of an FSO link. For each link, the effects of atmospheric loss,
communication system operating on high-altitude airborne atmospheric turbulence, pointing error and link inter-
platforms was investigated [17]. Recently, by jointly con- ruption due to AoA fluctuation are jointly considered.
sidering the effects of atmospheric attenuation, atmospheric Different from [18], we consider the independence of
turbulence, pointing error and link interruption due to AoA four impairments, and obtain a simple and tractable
fluctuation, a novel statistical channel model for hovering expression for the probability density function (PDF) of
UAV-based FSO communications was derived [18]. In [19], the total channel gain, which can be effectively used for
by taking into account the effect of nonzero boresight point- performance analysis. The accuracy of the derived PDF
ing error, an extended model is established for UAV-based is verified in Section V.
FSO communications. However, due to the large number • Performance analysis: Based on the derived PDF of
of complex functions and operators, the derived expressions the total channel gain, the closed-form expressions of
[18], [19] are cumbersome, and intuitive insights cannot easily the link outage probability and the end-to-end outage
be obtained for system design. To facilitate the parameter probability are derived. Through asymptotic analysis,
optimization, we are motivated to develop tractable channel the asymptotic bounds of the link outage probability
models and theoretical expressions. and the end-to-end outage probability are then obtained.
For parameter optimizations, there exist two fundamen- Furthermore, insights about the derived expressions are
tal tradeoffs in UAV-based FSO communications [16], [17], also provided. Numerical results verify the accuracy of
[20], [21]. First, the beam width balances the fading of the derived theoretical expressions, and these expressions
pointing error and the strength of received signal. That is, can be directly utilized to evaluate system performance
increasing beam width will mitigate the impact of pointing rapidly without time-intensive simulations.
error, but it also reduces the strength of the received signal. • Parameter optimization: To further improve the system
Second, the selection of field-of-view (FoV) involves a tradeoff performance, the beam width, the FoV and the UAVs’
between AoA fluctuation and received noise. In other words, locations are optimized. Specifically, the minimum beam
increasing the value of FoV can reduce the impact of AoA width is derived first, and the actual beam width of
fluctuation but increase the received noise as well. For network each link should be adjusted to be larger than or equal
optimizations, the deployment or trajectory of UAVs can be to the minimum value. Then, the FoV is optimized by
designed to take full advantage of their ultra-flexibilities. Such minimizing the end-to-end outage probability with the
topics have been studied in open literature. In [22]–[24], beam width constraint, and the asymptotically optimal
the performance indicators like system throughput, network FoV is derived by solving a nonlinear equation. Finally,
coverage, energy efficiency, and task completion time were under the obstacle scenario, the locations of UAVs are
optimized to find the optimal static locations or trajectory optimized by solving a minimization problem. Then,
of UAVs. When designing the UAVs’ trajectory, most prior the problem is transformed to a min-max problem, which
solutions rely on simplified channel models that are based can be effectively solved by using the embedded function
on either the assumption that LoS link always exists or the “fmincon” in MATLAB [25]. Numerical results verify
LoS statistic model. However, in reality, the LoS link may the effectiveness of the proposed parameter optimization
be obstructed by some obstacles (such as mountains and schemes.
buildings). In other words, the LoS link may not always be The rest of the paper is organized as follows. Section II
available in practical systems. Therefore, the existence of the presents the system model. In Section III, the exact outage
LoS link should be considered in the parameter optimization probabilities and the corresponding asymptotic bounds for the
process [25]. Consequently, an important question arises: UAV-based FSO relaying system are analyzed. In Section IV,

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
2948 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 10, OCTOBER 2021

electrical signal at node i in the ith link is given by

yi = Rhi xi + ni , i = 1, · · · , N + 1, (1)
where R is the optoelectronic conversion factor of the PD
at node i; xi is the transmitted optical signal of node
i − 1; hi is the channel gain of the ith link, and ni is the
signal-independent additive white Gaussian noise having mean
zero and variance σn,i2
, i.e., ni ∼ N (0, σn,i
2
). The transmitted
optical signal is taken as symbols drawn equal-probably from
the OOK constellation such that xi ∈ {0, 2Pt }, and Pt is the
average transmitted optical power per link, which is related
Fig. 1. A UAV-assisted serial relaying FSO communication systems. to the total transmit optical power P by Pt = P/(N + 1).
At each receiver, the background noise is assumed to be the
the parameter optimization is further investigated. Numerical dominant noise source, and the noise variance is a quadratic
results are presented in Section V. Finally, Section VI con- function with respect to the FoV of the receiver [18], i.e.,
cludes the paper.
Notations: Throughout this paper, regular font indicates a σn,i
2 2
= ΛθFoV,i , (2)
scalar. X ∼ N (μ, σ 2 ) denotes that X follows a Gaussian where Λ is a coefficient related to the wavelength, bandwidth
distribution having mean μ and variance σ 2 ; Γ(·) denotes the of the optical filter, spectral radiance and lens area. θFoV,i is
Gamma function; Kn (·) denotes the modified Bessel function the receiver’s FoV at node i.
x 2
of the second kind with order n [27]; erf(x) = √2π 0 e−t dt
denotes the error function; δ(·) denotes the Dirac delta func- B. Channel Model
tion; Gm,n
p,q [·] is the Meijer’s G-function [27]; and Pr(·) denotes There are three types of links in the considered system, i.e.,
the probability of an event. the GU link, the UU links, and the UG link. Actually, the first
link is the GU link, the (N + 1)th link is the UG link, and all
II. S YSTEM M ODEL other links (i.e., from the second link to the N th link) belong
As shown in Fig. 1, we consider a UAV-based multi-hop to the UU links. For all the links, the total channel gain hi
relaying FSO communication system, which includes one can be formulated as a product of four impairments, i.e.,
source node, one destination node, and N UAV-based serial hi = hi hi hi
(l) (a) (pe) (aoa)
hi , i = 1, · · · , N + 1, (3)
relay nodes. In the system, the transmit optical signal from the (l) (a)
source node propagates through N serial relay nodes before where hi is the atmospheric loss, hi
is the atmospheric
(pe) (aoa)
detection at the destination node. Specifically, the source turbulence, hi is the pointing error, and hi is the link
node is equipped with a laser diode (LD) and transmits the interruption due to AoA fluctuation. Inspired by the ideas
optical signal to the first relay node. For each relay node, in [17], [18] and [25], three typical adjustable parameters
a photodiode (PD) is first employed to convert the received (i.e., beam width, FoV, and link distance) are considered to
optical signal to electrical signal, the decode-and-forward (DF) be different in each link for the sake of the system parameter
relaying scheme is then implemented, and the recovered signal optimization, and the other parameters are assumed to be
is finally converted to optical signal by using an LD and the same in all N + 1 links unless otherwise stated. In the
transmitted to the next node. At the detection node, a PD following, the statistical characteristic of the four impairments
is utilized to perform the photoelectric conversion. Different will be analyzed, respectively.
from the ground fixed relays, the relay nodes are assumed 1) Atmospheric Loss: Referring to [28], the atmospheric
(l)
to be mounted on the UAVs. Moreover, we assume that the loss hi is determined by the exponential Beers-Lambert law
UAVs hover at their own fixed positions, and the orientations as
of the transceiver in each link are aligned with each other. (l)
hi = exp(−Zi Φ), i = 1, · · · , N + 1, (4)
However, the instantaneous positions of the relay nodes and
the orientations of the transceiver deviate from the mean values where Zi is the distance of the ith link, and Φ is the
due to numerous random events related to UAV hovering or atmospheric attenuation coefficient related to visibility.
building swaying. 2) Atmospheric Turbulence: For the atmospheric turbu-
lence, the Gamma-Gamma fading model is used here since its
distribution is in close agreement with measurements under
A. Received Signal Model (a)
various turbulence conditions. Therefore, the PDF of hi is
Here, the intensity modulation and direct detection is given by [29]
employed and the on-off keying (OOK) is used as the modu-   αi +βi
2(αi βi ) 2 (a) αi +βi
(hi ) 2 −1
(a)
lation scheme. For simplification, the source node is denoted fh(a) hi =
by node 0, the relay nodes are denoted by nodes 1, 2, · · · , N , i Γ(αi )Γ(βi )
  
and the destination node is denoted by node N + 1. In the × Kαi −βi 2 αi βi hi
(a)
,
considered system, there are N + 1 links, and the ith link is
built by node i − 1 and node i. Mathematically, the received i = 1, · · · , N + 1, (5)

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2949

b) GU link: For the first link, θtx,1 is zero. The variances

of position deviation xt,1 is assumed to be zero-mean Gaussian
distribution with variance σp,g 2
, i.e., xt,1 ∼ N (0, σp,g
2
). More-
over, we can obtain xr,1 ∼ N (0, σp,u ). Therefore, xd,1 , yd,1 ∼
2

N (0, σp,u
2
+ σp,g
2
).
c) UG link: Referring to the analysis for GU and UU
links, we can obtain xt,N +1 ∼ N (0, σp,g 2
), xr,N +1 ∼
N (0, σp,u ), and xθt ,N +1 ∼ N (0, ZN +1 σangle,u ) for the
2 2 2

(N + 1)th link. Consequently, xd,N +1 , yd,N +1 ∼ N (0, σp,u 2

+
σp,g + ZN +1 σangle,u ).
2 2 2

Assuming that the variables in (7) are independent of each

other [18],
 the total radial displacement can be expressed as
rtr,i = x2d,i + yd,i 2 , which follows a Rayleigh distribution,

Fig. 2. The Gaussian beam footprint at receiver aperture for the ith link. i.e.,
 
where the parameters αi and βi related to large-scale and rtr,i rtr,i
2
small-scale eddies are given by frtr,i (rtr,i ) = 2 exp − 2 , rtr,i ≥ 0, (8)
σs,i 2σs,i
⎧ ⎡ ⎛ ⎞ ⎤−1
⎪
⎪
⎪
⎪ ⎢ ⎜
2
0.49σR,i ⎟ ⎥ where the total displacement variances σs,i
2
for different links
⎪
⎪ αi = ⎣exp ⎝ 
⎪
⎪ 7/6 ⎠ − 1⎦ are derived as
⎪
⎨ 12/5
1 + 1.11σR,i ⎧
⎡ ⎛ ⎞ ⎤−1 (6) ⎪
⎨σp,u + σp,g ,
2 2
i=1
⎪
⎪
⎪
⎪ 2 σs,i = 2σp,u + Zi σangle,u ,
2 2 2 2
i = 2, . . . , N (9)
⎪
⎪ ⎢ ⎜ 0.51σR,i ⎟ ⎥ ⎪
⎪
⎪ βi = ⎣exp ⎝  5/6 ⎠ − 1⎦ , ⎩ 2
⎪
⎩ 12/5 σp,u + σp,g
2
+ ZN
2
σ 2
+1 angle,u , i = N + 1.
1 + 0.69σR,i
Based on the radial displacement model in (8), the PDF of
(pe)
where σR,i 2
= 1.23Cn2 k 7/6 Zi
11/6
is the Rytov variance related hi is written as [20]
to the link distance Zi , Cn2 is the index of refraction structure  
parameter of atmosphere, and k = 2π/λ is the optical wave ζ 2 (pe) 2
= ζi 2 (hi )ζi −1 ,
(pe)
fh(pe) hi
number with λ being the wavelength. i
Ai i
3) Pointing Error: Fig. 2 shows the Gaussian beam footprint 0 ≤ hi
(pe)
≤ Ai , i = 1, · · · , N + 1, (10)
at receiver aperture, where the beam is orthogonal to the
receiver lens plane. We consider a circular detection aperture where Ai = [erf(vi )]√ 2
is the√
fraction of the collected power
of radius ra at node i, i = 1, · · · , N + 1. The radial when rtr,i = 0, vi = πra / 2wz,i is the ratio between the
displacement vector from the receiving aperture center to the aperture radius and the beam width. ζi2 = wz eq 2
,i /4σs,i is the
2
beam center is expressed as rd,i = [xd,i , yd,i ], which results ratio between the squared equivalent √ beam width and displace-
2
from three different vectors: i) the displacement vector induced ment variance, where ωz2eq ,i = ωz,i 2
πerf(vi )/(2 vi e−v ) ≈
by the transmitter’s orientation deviation rθt ,i = [xθt ,i , yθt ,i ], √
ωz,i
2
+ 3/(2 2) is the equivalent beam width [30].
ii) the displacement vector induced by the transmitter’s posi- 4) Link Interruption Due to AoA Fluctuation: As depicted
tion deviation rt,i = [xt,i , yt,i ], and iii) the displacement in Fig. 3, due to the effect of AoA fluctuation, the beam is
vector induced by the receiver’s position deviation rr,i = no longer orthogonal to the receiver plane. When an incident
[xr,i , yr,i ]. Consequently, the displacements located along the laser with θa,i arrives at the receiving plane, the airy pattern
horizontal axe and the vertical axe at the detector plane are, may be out of detector range occasionally due to the relatively
respectively, given by large orientation deviations of hovering UAV, and the case
 that received AoA exceeds the range of FoV will result in an
xd,i = xt,i + xr,i + xθt ,i
(7) outage. Here, the AoAs for different links are defined as [18]
yd,i = yt,i + yr,i + yθt ,i .
⎧
⎪
⎪θrx,i + θry,i ,
⎪ i=1
2 2
According to the central limit theorem, the position and
⎨
orientation deviations follow Gaussian distributions as they θa,i  2 2
⎪ (θ + θrx,i ) + (θty,i + θry,i ) , i = 2, . . . , N
result from numerous random events [18]. ⎪
⎪  tx,i
a) UU links: For these links, both xt,i and xr,i follow the ⎩ θ2 + θ2 , i = N + 1,
tx,i ty,i
same Gaussian distribution with mean zero and variance σp,u 2
, (11)
i.e., xt,i , xr,i ∼ N (0, σp,u ). Moreover, xθt ,i = Zi tan θtx,i 
2

Zi θtx,i , where θtx,i ∼ N (0, σangle,u

2
). This indicates that where θtx,i and θty,i are the horizontal and vertical misalign-
xθt ,i ∼ N (0, Zi σangle,u ). Similarly, the distribution of yd,i
2 2
ment orientations of the transmitter in the ith link. Similarly,
can also be analyzed. As a result, xd,i , yd,i ∼ N (0, 2σp,u 2
+ θrx,i and θry,i represent the horizontal and vertical orientation
Zi σangle,u ).
2 2
deviations of the receiver. Therefore, θa,i , i = 1, · · · , N + 1

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
2950 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 10, OCTOBER 2021

(i.e., θtx,i and θty,i ) contribute to the pointing error and the
(pe)
AoA fluctuation simultaneously, the correlation between hi
(aoa)
and hi is weak when θFoV,i σangle,u . The reason
is provided as follows. Note that σangle,u is the standard
deviation of θtx,i and θty,i . Under a small value of σangle,u
relative to θFoV,i , compared with rt,i and rr,i , the impact of
rθt, i  [Zi θtx,i , Zi θty,i ] in Fig. 2 on pointing error is weak.
This indicates that the transmitter’s orientation misalignment
is weakly correlated with pointing error when θFoV,i
σangle,u . Moreover, when θFoV,i σangle,u , θtx,i and θty,i
have little effect on θa,i , and thus the transmitter’s orienta-
tion misalignment and the AoA fluctuation are also weakly
correlated. Consequently, the four impairments in (3) can be
approximated as independent variables. Specially, for the GU
Fig. 3. A schematic diagram about the impact of AoA fluctuation for the link, the pointing error and the AoA fluctuation are no longer
ith link. relevant since orientation deviations of the source node are
approximately zero, and these four impairments are practically
is Rayleigh-distributed, i.e., independent.
  According to the above analysis, the PDF of hi is approxi-
θa,i θa,i
2
mated as
fθa,i (θa,i ) = exp − , θa,i ≥ 0,  
mi σangle,u
2 2mi σangle,u
2 +∞
1 hi
fhi (hi )   fh(aoa) fhi (hi ) dhi , (15)
(12) 0 hi i hi
where hi  hi hi hi . By solving (15), we derive the PDF
(l) (a) (pe)
where mi varies with the type of links and is given by
 of hi and state the result in the following theorem.
1, i = l or N + 1
mi = (13) Theorem 1: For the UAV-based FSO communication sys-
2, i = 2, . . . , N. tem, the PDF of the overall channel gain hi is approximated
We focus on the main lobe of the airy pattern, which has as
 
the most power. The width of the main lobe is approximately θFoV,i
2

equal to 2.4λ, which is much smaller than the typical detector fhi (hi )  exp − δ(hi )
2mi σangle,u
2
size [31]. Consequently, a reasonable consideration is the   
zero-one distribution that describes whether the incident laser θFoV,i
2
αi βi ζi2
+ 1 − exp −
is located on the receiving FoV or not [32]. Mathematically, 2mi σangle,u
2 (l)
Ai hi Γ(αi )Γ(βi )
given a fixed FoV, the link interruption (i.e., hi
(aoa)
= 0) occurs  ! 
αi βi ! ζi2
if θa,i > θFoV,i , and the maximum signal power (i.e., hi
(aoa)
= 3,0
× G1,3 h ! . (16)
(l) i ! ζ 2 − 1, αi −1, βi − 1
1) is collected otherwise. Therefore, the corresponding PDF of Ai hi i
(aoa)
hi is given by Proof: See Appendix A.
  To verify the accuracy of the approximated PDF expression
(aoa) θFoV,i
2
(aoa)
fh(aoa) (hi ) = exp − δ(hi ) in Theorem 1, we provide Fig. 4 in Section V. As can be
i 2mi σangle,u
2
seen, the approximate PDF in (16) is tight as long as θFoV,i ≥
  
θFoV,i
2 5σangle,u, which justifies the proposed approximation.
(aoa)
+ 1−exp − δ(hi −1),
2mi σangle,u
2
B. Link Outage Probability
i = 1, · · · , N + 1. (14)
1) Exact Expression: The outage probability is an important
III. O UTAGE P ERFORMANCE A NALYSIS metric to evaluate the performance of the FSO communication
system. The outage probability of the ith link pi is expressed
In this section, the overall statistical characteristic of the as the probability that the instantaneous SNR Υi falls below
FSO channel is analyzed. Then, the link outage probability the specified threshold Υth , i.e.,
and the end-to-end outage probability are derived. Finally,
the asymptotic behaviors of the outage probabilities are stud- pi = Pr(Υi < Υth ) = Pr(hi < hth,i ), (17)
ied. Some insights are provided as well.
where the corresponding threshold of the channel gain is
" #
A. Overall Channel Statistical Characteristic Υth σn,i
2
θFoV,i Υth Λ
hth,i = = . (18)
For the four impairments in (3), hi and hi
(l)
are
(a) 2R2 Pt2 RPt 2
(pe)
obviously independent and also independent of hi and The exact expression of (17) is provided in the following
(aoa)
hi . Although the orientation deviations of the transmitters theorem.

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2951

Theorem 2: For the UAV-based FSO communication sys- Proof: The proof is straightforward by submitting (19)
tem, the exact expression of the link outage probability pi , into (21).
i = 1, . . . , N + 1 is given by 2) Asymptotic Bound: Due to the impact of the intermediate
     links performance bound in (20), the corresponding system
θFoV,i
2
θFoV,i
2
pi = exp − + 1 − exp − performance is also limited by AoA fluctuation when the
2mi σangle,u
2 2mi σangle,u
2
transmitted power is large. The asymptotic bound for the end-
 # ! 
ζi2 3,1 αi βi θFoV,i Υth Λ !! 1, ζi2 + 1 to-end outage probability is given in the following theorem.
× G ,
Γ(αi )Γ(βi ) 2,4 Ai h(l) RPt 2 ! ζi2 , αi , βi , 0 Theorem 5: For the UAV-based FSO communication sys-
i tem, the asymptotic bound of the end-to-end outage probability
(19) Pbound
out
for large transmit power Pt is given by
where mi is given by (13).   
Proof: See Appendix B. θFoV,1
2
Pbound = 1 − 1 − exp − 2
out
2) Asymptotic Bound: For large transmit power Pt , 2σangle,u
the asymptotic bound of the link outage probability is given   
θFoV,N
2
+1
in the following theorem. × 1 − exp − 2
Theorem 3: For the UAV-based FSO communication sys- 2σangle,u
N
  
tem, the asymptotic bound of the link outage probability $ θFoV,i
2
pbound,i , i = 1, . . . , N + 1 for large transmit power Pt is × 1 − exp − 2 . (23)
i=2
4σangle,u
given by
 
θFoV,i
2
Proof: The proof is straightforward by substituting (20)
pbound,i = exp − . (20)
2mi σangle,u
2 into (21).
Remark 2: It is worth mentioning that the asymptotic bound
Proof: See Appendix C. Pbound
out
in (23) deteriorates with the increase of the number of
Remark 1: In Theorem 3, the asymptotic bound (20) varies relays N . Interestingly, increasing relay number worsens the
with θFoV,i
2
and mi . From (13), it is known that mi = 1 for outage performance when the system outage performance has
the UG/GU link, while mi = 2 for the UU links. Given a fixed attained the bound in (23).
θFoV,i
2
for all the links, the asymptotic bound (20) for the UU
links is larger than that for the GU/UG link, the asymptotic
IV. PARAMETER O PTIMIZATIONS
bound (20) for the GU link is the same as that for the UG link,
which indicates that the UU links achieve the worst asymptotic In this section, the key parameter optimizations for the
outage performance, and the GU link and the UG link achieve UAV-based FSO communications are investigated. Specifi-
the same asymptotic outage performance. cally, the beam width, the FoV, and the UAV locations are
optimized.
C. End-to-End Outage Probability
1) Exact Expression: For DF relaying, outage of each A. Beam Width Adjustment
intermediate link may lead to the outage of the relaying
system. Therefore, the end-to-end outage probability is given Since hovering UAVs have larger deviations of the position
by and orientation than fixed ground platforms, pointing error
in UAV-based links is more severe, hence the adjustment for
N
$ +1
beam width is especially important to improve the diversity
P out = 1 − (1 − pi ). (21)
order gain and outage performance for different communica-
i=1
tion conditions.
Then, the end-to-end outage probability is obtained in the
In (10), the ratio between the squared equivalent beam width
following theorem.
and displacement variance is ζi2 = wz eq 2
,i /(4σs,i ) ≈ [ωz,i +
2 2
Theorem 4: For the UAV-based FSO communication sys- √
3/(2 2)]/(4σs,i ). In (5), βi denotes the parameters related
2
tem, the exact expression of the end-to-end outage probability
to small-scale eddy in atmospheric turbulence. According to
P out is expressed as
[30] and [33], it can be known that the condition ζi2 < βi
P out    represents that pointing error becomes dominant in relation to
θFoV,1
2
atmospheric turbulence in the ith link. Since the turbulence
= 1 − 1 − exp − 2 parameter βi , which is related to the atmospheric condition,
2σangle,u
   N    cannot be chosen arbitrarily, adjusting beam width wz,i to
θFoV,N +1
2 $ θFoV,i
2
satisfy ζi2 ≥ βi is an appropriate measure. By letting ζi2 = βi ,
× 1−exp − 2 1−exp − 2
2σangle,u i=2
4σangle,u the minimum value of received beam width in the ith link is
N
  !  given by
$+1
ζi2 3,1 αi βi
! 1, ζ 2 + 1
× 1− G !
hth,i ! 2 i . "
Γ(αi )Γ(βi ) 2,4 Ai h(l) ζi , αi , βi , 0 2 − √ 3
i=1 i ωz,i = 4βi σs,i
min
. (24)
(22) 2 2

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
2952 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 10, OCTOBER 2021

Here, to avoid the severe impairment of pointing error, C. UAV Location Optimization
the beam width of each link in the FSO relaying system should
To improve the reliability of the end-to-end link, the UAVs’
satisfy
positions can be optimized. To avoid the undesirable bound
of AoA fluctuation (i.e. eq. (20)), the FoV of each receiver
ωz,i ≥ ωz,i
min
, i = 1, · · · , N + 1. (25)
is typically set to be a large fixed value. In this subsection,
we assume that the FoV is sufficiently large so that the AoA
It is noteworthy to mention that the power received at
fluctuation is neglectable.
fixed-size detector reduces with an increase of the beam width.
Here, we consider an FSO relaying system with N UAVs
Consequently, excessive increase of the beam width is also
and N0 obstacles. To facilitate the optimization, all UAVs are
inadvisable.
assumed to be located at the same height. For all UAVs,
the two-dimensional coordinates in the XY plane are opti-
mized. Without loss of generality, the positions of UAVs
B. FoV Optimization With the Beam Width Constraint
are set to be (xi , yi ), i = 1, · · · , N . Moreover, (x0 , y0 ) and
Due to the asymptotic bound (23), the achievable end-to-end (xN +1 , yN +1 ) are the coordinates of source node and desti-
outage probability is limited. Therefore, after determining the nation node, respectively. Due to the existence of obstacles,
minimum beam width, the FoV optimization problem should the UAVs cannot be deployed on the positions of obstacles.
be considered to minimize the end-to-end outage probability, An indicator function fn (xi , yi , xi+1 , yi+1 ) is used to indicate
i.e., whether the ith link is block by the nth obstacle. Specifically,
if the nth obstacle blocks the ith link, fn (xi , yi , xi+1 , yi+1 ) ≥
min P out 0, otherwise fn (xi , yi , xi+1 , yi+1 ) < 0. Here, the UAV loca-
θFoV,1 ,...,θFoV,N +1
tion optimization problem can be formulated as
s.t. ωz,i ≥ ωz,i
min
, i = 1, . . . , N + 1. (26)
min P out
Since FoV adjustment of one link does not affect the other x1 ,y1 ,...,xN ,yN

links, problem (26) is reduced to optimizing FoV of each link s.t. fn (xi , yi , xi+1 , yi+1 ) < 0,
individually, i.e., i ∈ 0, 1, · · · , N ; n ∈ 1, 2, · · · , N0 . (31)

min pi To solve the optimization problem, we assume that the

θFoV,i
atmospheric turbulence is dominant in relation to the pointing
s.t. ωz,i ≥ ωz,i
min
. (27) error (i.e., ζi2 > βi , i = 1, · · · , N + 1), and the FoV θFoV,i
is large. With these assumptions, the expression (C.2) can be
Considering the complexity of the exact link outage prob- further written as
ability and the demand of practical FSO communications,
we optimize the system performance for large Pt . By sub-  β i
ζi2 Γ(αi − βi ) αi βi  βi
stituting (2) and (18) into (C.2), we obtain the link outage pi ≈ hth,i
probability for large Pt with respect to θFoV,i as Γ(αi )Γ(βi )(ζi2 − βi )βi (l)
Ai hi
⎛ " ⎞βi
% & ζi2 Γ(αi − βi ) α β Υ σ 2
pi  L(θFoV,i
2
) + Θ 1 − L(θFoV,i
2
) (θFoV,i )βi , (28) = ⎝ i i th n,i ⎠
Γ(αi )Γ(βi )(ζi2 − βi )βi Ai h(l) R 2
i
where L(θFoV,i
2
) and Θ are defined as ' () *
ai
⎧   −βi
⎪ θ 2 × (Pt ) . (32)
⎪
⎪ FoV,i
⎨L(θFoV,i ) = exp − 2mi σ 2
2
⎪
angle,u
 # β i We assume that the imax th link is the longest with para-
⎪
⎪ ζ 2
Γ(α − β ) α β Υ Λ meters aimax and βimax . Since the parameter βi decreases
⎪
⎪Θ = i i i i i th
.
⎩ Γ(αi )Γ(βi )(ζi2 − βi )βi Ai h(l) RPt 2 monotonically with an increase of the ith link distance,
i
(29) we have
+N +1 ,N +1
By taking derivative of (28) with respect to θFoV,i and 1 − i=1 (1 − pi ) i=1 pi
lim = lim
setting it to be zero, we obtain the following nonlinear equation Pt →∞ pimax Pt →∞ pimax
,N +1 −βi
%  2 & i=1 ai Pt
βi −2 = lim
mi Θσangle,u
2
βi (θFoV,i ) 1 − L θFoV,i Pt →∞ a −βimax
 2   2  imax Pt
β
+ Θ (θFoV,i ) i L θFoV i
− L θFoV,i = 0. (30) = 1, (33)

In this paper, the asymptotically optimal θFoV,i is updated which indicates that the system outage probability is deter-
to match different communication conditions by numerically mined by the performance of the link having the maximum
solving (30). distance. Therefore, for a large value of Pt , problem (31) can

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2953

Fig. 4. Comparisons of the derived PDF (16) and the simulated PDF with different θFoV,i for different links when Zi = 250 m and wzi = 2 m, where
(a)-(c) are for the GU link, (d)-(f) are for the UU links, (g)-(i) are for the UG link.

be transformed into σp,u = σp,g = 10 cm, the standard deviation of UAV

orientation σangle,u = 1.2 mrad, and γth = 10 dB [18].
min max{Z1 , Z2 , · · · , ZN }
x1 ,y1 ,··· ,xN ,yN
s.t. fn (xi , yi , xi+1 , yi+1 ) < 0 A. Channel Modelling Results
i ∈ 0, 1, · · · , N ; n ∈ 1, 2, · · · , N0 , (34) The derivation of the PDF in (16) is based on the assumption
 that the four types of impairments in (3) are independent of
2 2
where Zi = (xi − xi+1 ) + (yi − yi+1 ) is the distance of each other. To verify the accuracy of (16), we compare the
the ith link. To solve problem (34) effectively, the grouping curves of (16) with the simulated PDFs for different links
optimization or multi-variate optimization can be used, which when Zi = 250 m and wzi = 2 m, as shown in Fig. 4.
can be implemented by using the embedded function “fmin- Note that Figs. 4(a)-(c) are for the GU link, Figs. 4(d)-(f)
con” in MATLAB [25]. are for the UU links, and Figs. 4(g)-(i) are for the UG link.
For the simulated PDFs, the four impairments in (3) are
set to be correlated random variables. As can be seen in
V. N UMERICAL R ESULTS
Figs. 4(a), (d) and (g), when θFoV,i = 5 mrad, the differences
In this section, some numerical results will be presented between analytical results when hi = 0 and simulation results
to support the theoretical claims made in previous sections. are large. Specifically, for the GU, UU and UG links, fhi (0)
Unless stated otherwise, the default values of the parameters in (16) are 1.3×10−2, 1.7×10−4 and 1.7×10−4, respectively.
used for simulations are given by λ = 1550 nm, Cn2 = But the simulated values at hi = 0 are 172.7, 626.6, and 75.6,
5 × 10−14 m−2/3 , R = 0.9, Φ = 1 km−1 , σn2 = 6.4 × 10−14 respectively. For other values of hi in Fig. 4(a), (d) and (g),
corresponding to θFoV,i = 8 mrad, receiver lens radius ra = the analytical results match the simulation results well. This
5 cm, the standard deviation of UAV and ground displacement indicates that small FoV in each link will result in the

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
2954 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 10, OCTOBER 2021

Fig. 5. Link outage probability for an UU link versus average transmitted Fig. 6. Link outage probability versus average transmitted optical power for
optical power under different turbulence conditions when θFoV = 8 mrad, 2
different kinds of links when θFoV,i = 5 mrad, σR,i = 1, wz,i = 2 m and
wz = 2 m and Z = 250 m. Zi = 250 m.

mismatch between the analytical results and simulation results

when hi = 0. When θFoV,i ≥ 6 mrad (i.e., θFoV,i ≥
5σangle,u ), as shown in Figs. 4(b), (c), (e), (f), (h) and (i),
the analytical curves match the simulation results well. This
indicates that the derived tractable expression (16) is accurate
when θFoV,i is large and can be directly used for performance
analysis.

B. Performance Analysis Results

In this subsection, we will examine the accuracy of the
derived link outage probability and the end-to-end outage
probability under different scenarios.
Fig. 5 plots the link outage probability for a UU link
versus the average transmitted optical power Pt under different
turbulence conditions σR 2
= 0.6, 1, 2, 3 when θFoV = 8 mrad,
wz = 2 m and Z = 250 m. For comparison, the asymptotic
Fig. 7. End-to-end outage probability versus average transmitted optical
bound (20) in Theorem 3 is also presented. For small Pt , power for different numbers of relays when θFoV,i = 6 mrad, wz,i = 2 m
the values of link outage probabilities decrease with the and the distance between source and destination nodes ZSD = 2 km.
increase of Pt or with the decrease of σR2
. For large Pt , the val-
ues of link outage probabilities tend to stable values, which
are independent of the average transmitted optical power UU link is always larger than that of the UG/GU link. These
and the turbulence condition. Moreover, the stable values insights verify the analysis in Remark 1.
approach the asymptotic bound (20). This verifies the accuracy Fig. 7 shows the end-to-end outage probability versus the
of (20) in Theorem 3. Moreover, it should be emphasized average transmitted optical power Pt for different numbers of
that all analytical results in Fig. 5 present close agreement relays N when θFoV,i = 6 mrad, wz,i = 2 m and the distance
with Monte-Carlo simulation results, and the accuracy of the between source and destination nodes ZSD = 2 km. For
derived expression (19) in Theorem 2 is verified. comparison, the asymptotic bound (23) in Theorem 5 is also
Fig. 6 compares the link outage performance for different presented. In this simulation, the obstacles are not considered,
kinds of links when θFoV,i = 5 mrad, σR,i 2
= 1, wz,i = 2 and N UAV nodes are uniformly deployed between the source
m and Zi = 250 m. In addition to obtaining the similar node and the destination node. For small Pt , the end-to-end
conclusions in Fig. 5, some other interesting insights can outage probability performance improves with the increase of
also be found in Fig. 6. As shown in Fig. 6, the UU link Pt or N . However, at large Pt , the outage performance does
achieves the maximum link outage probability, and the GU not improve with Pt , but approaches the asymptotic bound
link achieves the comparable performance to the UG link. (23), which verifies the accuracy of (23) in Theorem 5. More-
Moreover, the GU link achieves the same asymptotic outage over, the asymptotic bound (23) varies with N . Specifically,
performance as the UG link, and the asymptotic bound of the the asymptotic bound with N = 4 achieves the largest value,

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2955

Fig. 9. Link outage probability of an UU link versus normalized FoV for

Fig. 8. End-to-end outage probability versus the standard deviation σangle,u
with different numbers of relays when Pt = 20 dBm, θFoV,i = 6 mrad, different Pt when Z = 250 m and wz = 2 m.
wz,i = 2 m, and ZSD = 2 km.

which coincides with the conclusion in Remark 2. Similar

to Fig. 5, the accuracy of the derived expression of (22) in
Theorem 4 is also verified by using simulations.
Fig. 8 shows the end-to-end outage probability versus the
standard deviation σangle,u with different numbers of relays
when Pt = 20 dBm, θFoV,i = 6 mrad, wz,i = 2 m, and
ZSD = 2 km. As can be observed, with the increase of
σangle,u , the end-to-end outage probability increases sharply
and then tends to one. This indicates that the system perfor-
mance dramatically degrades with the increase of σangle,u .
Moreover, a large σangle,u (for example, ≥3.4 mrad) will
directly result in a complete interruption of the system. Fur-
thermore, when σangle,u is small, the outage probability per-
formance improves with the number of UAV relays. Therefore,
the UAVs’ orientation fluctuation and the number of relays are Fig. 10. End-to-end outage probability versus FoV with different numbers
two major concerns for practical system design. In Fig. 8, all of relays when Pt = 20 dBm, σangle,u = 1.2 mrad, θFoV,i = 6 mrad,
analytical results match well with simulation results, which wz,i = 4 m, and ZSD = 2 km.
also verifies the accuracy of the derived expression in (22).

Pt = 0, 5, 10, 15 and 20 dBm, the optimal values of φFoV

C. Parameter Optimization Results (in mrad) are 3.25, 4.25, 5.25, 6.08, and 6.92, respectively.
In this subsection, to verify the efficiency of the proposed By solving (30), we obtain the corresponding asymptotically
optimization schemes, the parameter optimization results will optimal φFoV values as 1.94, 3.67, 4.98, 6.02, and 6.91.
be shown. The differences between the optimal φFoV in Fig. 9 and
Fig. 9 shows the link outage probability of a UU link the asymptotically optimal φFoV using (30) are small when
versus the normalized FoV φFoV for different Pt values when Pt is large. This indicates that, when Pt is large, eq. (30)
Z = 250 m and wz = 2 m. Here, the normalized FoV is can be directly used to determine the optimal FoV without
defined as φFoV = θFoV /σangle,u , and the beam width wz is time-consuming simulations.
adjusted to satisfy (25). As can be seen, with the increase Fig. 10 shows the end-to-end outage probability versus
of Pt , the link outage probability performance improves. FoV with different numbers of relays when Pt = 20 dBm,
Moreover, all analytical results accurately match simulation σangle,u = 1.2 mrad, θFoV,i = 6 mrad, wz,i = 4 m, and
results in the entire normalized FoV range, which verifies ZSD = 2 km. In Fig. 10, the obstacles are not considered,
the accuracy of (19) in Theorem 2. All curves tend to the and the UAV relays are placed equidistant. The exhaustive
asymptotic bound in (20) with the decrease of φFoV , which search scheme and the asymptotically optimal scheme are
indicates the correctness of Theorem 3. Furthermore, with the provided. In the exhaustive search scheme, the FoVs for the
increase of φFoV , the link outage probability decreases first GU, UG and UU links are set to be identical. In asymptotically
and then increases. It can be observed from Fig. 9 that, when optimal scheme, the optimal FoVs for all links are respectively

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
2956 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 10, OCTOBER 2021

TABLE I
T HE S ETUP OF F OUR C ASES IN F IG . 11

Fig. 11. Outage probability versus average link transmit power for different
scenarios when ZSD = 2 km and wz,i = 4 m.

obtained by (30). It can be seen that when FoV is increased,

all curves of the exhaustive search scheme decrease first and
then increase, this indicates that an optimal FoV exists for
each curve. Specifically, for N = 1, 2, 3 and 4, the optimal
FoVs in the exhaustive search scheme are 4.7 mrad, 7.4 mrad,
9.2 mrad, and 10.8 mrad, as marked on the figure. The
corresponding minimum outage probabilities are 8.95 × 10−3,
4.26×10−4, 5.40×10−6, and 3.13×10−8, respectively. More-
over, the FoVs of different links for the asymptotically optimal
scheme by using (30) are also provided on the figure. As seen,
the asymptotically optimal scheme can achieve comparable
outage performance to the exhaustive search scheme, which
verifies the accuracy of (30).
To verify whether the link with maximum distance (i.e.,
max-distance link) dominates the end-to-end outage perfor-
mance, Fig. 11 shows the outage probability versus average
Fig. 12. Optimal relay deployment for the UAV-based FSO relaying system
transmit power Pt for different scenarios when ZSD = 2 km with one cylindrical obstacle. (a) one relay, (b) two relays, (c) three relays,
and wz,i = 4 m. In this simulation, the beam width satisfies (d) four relays.
ζi2 > βi , i = 1, . . . , N +1. Four cases are considered, as shown
in Table I. Note that the asymptotic bounds for Cases 1 and TABLE II
3 are 1.39 × 10−11 , the asymptotic bound for Case 4 is T HE O PTIMAL R ELAY L OCATIONS AND L INK D ISTANCES IN F IG . 12
2.78×10−11, which are too small and not plotted in the figure.
For cases with large FoV (i.e., Cases 1, 3 and 4) in Fig. 11,
the outage probabilities of the max-distance link converge to
that of the overall system when Pt is large (for example, when
Pt ≥ 12.5 dBm for Case 1, when Pt ≥ 27.5 dBm for Case 3,
or when Pt ≥ 20 dBm for Case 4), which verifies that the
link with the maximum distance dominates the overall system
performance, just as analyzed in (33). However, for the case
with small FoV (i.e., Case 2), the outage probability of the
max-distance link does not match with that of the overall
system at large Pt . In this case, the AoA fluctuation has a
strong effect on outage performance. Moreover, the perfor- Specifically, Fig. 12 plots optimal relay deployment for the
mance of Case 2 is better than that of Case 1 before reaching UAV-based FSO relaying system with one cylindrical obstacle.
the asymptotic bound (23) since larger FoV introduces more In the simulation, the coordinates of the source and the
background noise. destination are (0.1 km, 0.1 km) and (2 km, 2 km). The
After optimizing the UAVs’ locations, Fig. 12 and Fig. 13 obstacle is centered at (0.6 km, 1 km) with radius 0.5 km.
show the deployment of UAVs under different scenarios. From Fig. 12(a) to Fig. 12(d), the optimal locations of UAVs

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2957

model, closed-form expressions were obtained for link outage

probability and end-to-end outage probability. Subsequently,
the asymptotic outage performance bounds were also inves-
tigated. Moreover, the beam width, FoV and UAVs’ loca-
tions were optimized. Numerical results verify the accuracy
of the derived theoretical expressions and the efficiency of
the proposed optimization schemes. The obtained theoretical
expressions enable designers to evaluate outage performance
rapidly without time-intensive simulations. The derived para-
meter optimization results can help determine the optimal
available parameter choices when designing the UAV-based
FSO systems.

A PPENDIX A
P ROOF OF Theorem 1
By using (4), (5), (10) and expressing Kn (·) in terms of the
Meijer’s G-function [34], we can obtain the PDF of hi as [35]
αi βi ζi2
fhi (hi ) = (l)
Ai hi Γ(αi )Γ(βi )
 ! 
αi βi ! ζ 2
× G3,0 h ! 2 i
1,3 (l) i ! ζ − 1, αi − 1, βi − 1 .
Ai hi i
(A.1)
By substituting (14) and (A.1) into (15), we derive the PDF
Fig. 13. Optimal relay deployment for the UAV-based FSO relaying system of hi as
with two cylindrical obstacles. (a) one relay, (b) two relays, (c) three relays,    
θFoV,i
2 ∞
1 hi
(d) four relays.
fhi (hi ) = exp −  δ  fhi (hi ) dhi
2mi σangle,u
2
0 h i h i
TABLE III   
θFoV,i
2
T HE O PTIMAL R ELAY L OCATIONS AND L INK D ISTANCES IN F IG . 13 + 1 − exp −
2mi σangle,u
2

∞  
1 hi
×  δ − 1 fhi (hi ) dhi . (A.2)
0 hi hi
∞
With δ(ax) = δ(x)/|a| and −∞ f (a)δ(x − a)da = f (x)
[36], eq. (16) can be derived.

A PPENDIX B
P ROOF OF Theorem 2
By (16) and (17), pi is given by
 
θFoV,i
2 hth,i
and link distances are listed in Table II. By adding another pi = exp − δ(hi )dhi
cylindrical obstacle at (1.6 km, 1.2 km) with radius 0.2 km, 2mi σangle,u
2
'0 () *
Fig. 13 shows the optimal relay deployment for the FSO
  
=1
relaying system. In this case, the optimal relay locations and θFoV,i
2
αi βi ζi2
link distances are provided in Table III. As can be seen from + 1 − exp −
2mi σangle,u
2 (l)
Ai hi Γ(αi )Γ(βi )
the above results, through optimizing the relays’ deployment,  
!
the feasible positions of relays can always be found, and all hth,i
αi βi ! ζ 2
link distances tend to be approximately equal. × 3,0
G1,3 h ! i
(l) i ! ζ 2 − 1, αi − 1, βi − 1
dhi .
0 Ai hi i
' () *
VI. C ONCLUSION I
(B.1)
This paper investigated the channel modelling, outage prob-
ability analysis, and parameter optimization for UAV-based By (26) in [34], I in (B.1) can be further written as
FSO relaying systems. A tractable and accurate channel  ! 
αi βi ! 0, ζi2
model was established by taking into account atmospheric 3,1
I = hth,i G2,4 h ! .
(l) th,i ! ζ 2 − 1, αi − 1, βi − 1, −1
loss, atmospheric turbulence, pointing error, and link inter- Ai hi i
ruption due to AoA fluctuation. Based on the channel (B.2)

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
2958 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 10, OCTOBER 2021

Then, by (9.31.5) in [27], eq. (B.2) can be written as [10] Y. Zeng, Q. Wu, and R. Zhang, “Accessing from the sky: A tutorial on
 !  UAV communications for 5G and beyond,” Proc. IEEE, vol. 107, no. 12,
Ai hi
(l)
α β ! 1, ζ 2
+ 1 pp. 2327–2375, Dec. 2019.
hth,i !! 2 i
i i
I= G3,1 . (B.3) [11] A. Kaadan, H. Refai, and P. Lopresti, “Spherical FSO receivers for
αi βi 2,4 Ai h(l) ζi , αi , βi , 0 UAV communication: Geometric coverage models,” IEEE Trans. Aerosp.
i
Electron. Syst., vol. 52, no. 5, pp. 2157–2167, Oct. 2016.
Submitting (B.3) and (18) into (B.1), we obtain (19). [12] A. Kaadan, H. H. Refai, and P. G. LoPresti, “Multielement FSO trans-
ceivers alignment for inter-UAV communications,” J. Lightw. Technol.,
vol. 32, no. 24, pp. 4785–4795, Dec. 15, 2014.
A PPENDIX C [13] H. Ajam, M. Najafi, V. Jamali, and R. Schober, “Ergodic sum rate
P ROOF OF Theorem 3 analysis of UAV-based relay networks with mixed RF-FSO channels,”
IEEE Open J. Commun. Soc., vol. 1, pp. 164–178, Jan. 2020.
According to (18), when the transmit power Pt tends to [14] J.-H. Lee, K.-H. Park, Y.-C. Ko, and M.-S. Alouini, “Throughput
infinity, we have maximization of mixed FSO/RF UAV-aided mobile relaying with a
" buffer,” IEEE Trans. Wireless Commun., vol. 20, no. 1, pp. 683–694,
Υth σn,i
2 Jan. 2021.
lim hth,i = lim = 0. (C.1) [15] M. Najafi, H. Ajam, V. Jamali, P. D. Diamantoulakis,
Pt →∞ Pt →∞ 2R2 Pt2 G. K. Karagiannidis, and R. Schober, “Statistical modeling of the
FSO fronthaul channel for UAV-based communications,” IEEE Trans.
In (19), the Meijer’s G-function cannot provide intuitive Commun., vol. 68, no. 6, pp. 3720–3736, Jun. 2020.
insights on the behavior of hth,i . For a large transmit power Pt , [16] S. Huang and M. Safari, “Free-space optical communication impaired
by angular fluctuations,” IEEE Trans. Wireless Commun., vol. 16, no. 11,
the outage probability is dominated by the behavior of the PDF pp. 7475–7487, Nov. 2017.
near the origin. Therefore, by employing (07.34.06.0006.01) [17] V. V. Mai and H. Kim, “Beam size optimization and adaptation for
in [37] or [6], we can approximate (19) as high-altitude airborne free-space optical communication systems,” IEEE
     Photon. J., vol. 11, no. 2, pp. 1–13, Apr. 2019.
θFoV,i
2
θFoV,i
2 [18] M. T. Dabiri, S. M. S. Sadough, and M. A. Khalighi, “Channel modeling
pi ≈ exp − + 1 − exp − and parameter optimization for hovering UAV-based free-space optical
2mi σangle,u
2 2mi σangle,u
2
links,” IEEE J. Sel. Areas Commun., vol. 36, no. 9, pp. 2104–2113,
 κi Sep. 2018.
ζi2 Γ(αi − κi )bi αi βi [19] M. T. Dabiri, M. Rezaee, I. S. Ansari, and V. Yazdanian, “Channel
× (hth,i )κi , modeling for UAV-based optical wireless links with nonzero bore-
Γ(αi )Γ(βi )κi Ai h(l) sight pointing errors,” IEEE Trans. Veh. Technol., vol. 69, no. 12,
i
pp. 14238–14246, Dec. 2020.
κi = min(ζi2 , βi ), (C.2)
[20] A. A. Farid and S. Hranilovic, “Outage capacity optimization for free-
space optical links with pointing errors,” J. Lightw. Technol., vol. 25,
where bi = 1/(ζi2 − βi ) if ζi2 > βi , and bi = Γ(βi − ζi2 ) if no. 7, pp. 1702–1710, Jul. 2007.
ζi2 < βi . Because the small-scale turbulence eddies parameter [21] M. A. Amirabadi and V. T. Vakili, “A new optimization problem in
βi > 1 [38], the inequality κi > 1 always holds. FSO communication system,” IEEE Commun. Lett., vol. 22, no. 7,
pp. 1442–1445, Jul. 2018.
By (C.1), the second term in (C.2) tends to zero when Pt [22] J.-H. Lee, K.-H. Park, M.-S. Alouini, and Y.-C. Ko, “Free space optical
tends to infinity. Therefore, Theorem 3 holds. communication on UAV-assisted backhaul networks: Optimization for
service time,” in Proc. IEEE Globecom Workshops (GC Wkshps),
Waikoloa, HI, USA, Dec. 2019, pp. 1–6.
R EFERENCES [23] D. Wu, X. Sun, and N. Ansari, “An FSO-based drone assisted mobile
[1] Y. Ma, J.-Y. Wang, J.-B. Wang, M. Lin, H. Zhang, and C. Chang, access network for emergency communications,” IEEE Trans. Netw. Sci.
“Outage performance analysis and parameter optimization of hovering Eng., vol. 7, no. 3, pp. 1597–1606, Jul. 2020.
UAV-based FSO system,” in Proc. IEEE Int. Conf. Commun. (ICC), [24] O. Esrafilian, R. Gangula, and D. Gesbert, “Learning to communicate
Dublin, Ireland, Jun. 2020, pp. 1–6. in UAV-aided wireless networks: Map-based approaches,” IEEE Internet
[2] A. S. Hamza, J. S. Deogun, and D. R. Alexander, “Classification frame- Things J., vol. 6, no. 2, pp. 1791–1802, Apr. 2019.
work for free space optical communication links and systems,” IEEE [25] B. Zhu, J. Cheng, M.-S. Alouini, and L. Wu, “Relay placement for FSO
Commun. Surveys Tuts., vol. 21, no. 2, pp. 1346–1382, 2nd Quart., 2019. multihop DF systems with link obstacles and infeasible regions,” IEEE
[3] M. A. Khalighi and M. Uysal, “Survey on free space optical communi- Trans. Wireless Commun., vol. 14, no. 9, pp. 5240–5250, Sep. 2015.
cation: A communication theory perspective,” IEEE Commun. Surveys [26] M. T. Dabiri and S. M. S. Sadough, “Optimal placement of UAV-
Tuts., vol. 16, no. 4, pp. 2231–2258, 4th Quart., 2014. assisted free-space optical communication systems with DF relaying,”
[4] X. Li, X. Zhao, P. Zhang, and S. Tong, “BER performance of IEEE Commun. Lett., vol. 24, no. 1, pp. 155–158, Jan. 2020.
FSO communication system with differential signaling over corre- [27] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
lated atmospheric turbulence fading,” China Commun., vol. 17, no. 4, Products, 7th ed. New York, NY, USA: Academic, 2007.
pp. 51–65, Apr. 2020. [28] L. C. Andrews and R. L. Phillips, Laser Beam Propagation Through
[5] J.-Y. Wang, J.-B. Wang, M. Chen, Y. Tang, and Y. Zhang, “Outage Random Media, vol. 52. Bellingham, WA, USA: SPIE, 2005.
analysis for relay-aided free-space optical communications over turbu- [29] T. V. Pham, T. C. Thang, and A. T. Pham, “Average achievable
lence channels with nonzero boresight pointing errors,” IEEE Photon. rate of spatial diversity MIMO-FSO over correlated gamma–gamma
J., vol. 6, no. 4, pp. 1–15, Aug. 2014. fading channels,” IEEE/OSA J. Opt. Commun. Netw., vol. 10, no. 8,
[6] S. Huang, V. Shah-Mansouri, and M. Safari, “Game-theoretic spectrum pp. 662–674, Aug. 2018.
trading in RF relay-assisted free-space optical communications,” IEEE [30] R. Boluda-Ruiz, A. García-Zambrana, C. Castillo-Vázquez, and
Trans. Wireless Commun., vol. 18, no. 10, pp. 4803–4815, Oct. 2019. B. Castillo-Vázquez, “Novel approximation of misalignment fading
[7] W. Fawaz, C. Abou-Rjeily, and C. Assi, “UAV-aided cooperation for modeled by Beckmann distribution on free-space optical links,” Opt.
FSO communication systems,” IEEE Commun. Mag., vol. 56, no. 1, Exp., vol. 24, no. 20, pp. 22635–22649, Oct. 2016.
pp. 70–75, Jan. 2018. [31] R. Gagliardi and S. Karp, Optical Communications. Hoboken, NJ, USA:
[8] M. Alzenad, M. Z. Shakir, H. Yanikomeroglu, and M.-S. Alouini, Wiley, 1995.
“FSO-based vertical backhaul/fronthaul framework for 5G+ wireless [32] M. T. Dabiri, S. M. S. Sadough, and I. S. Ansari, “Tractable optical
networks,” IEEE Commun. Mag., vol. 56, no. 1, pp. 218–224, Jan. 2018. channel modeling between UAVs,” IEEE Trans. Veh. Technol., vol. 68,
[9] Y. Dong, M. Z. Hassan, J. Cheng, M. J. Hossain, and V. C. M. Leung, no. 12, pp. 11543–11550, Dec. 2019.
“An edge computing empowered radio access network with UAV- [33] F. Yang, J. Cheng, and T. A. Tsiftsis, “Free-space optical communication
mounted FSO fronthaul and backhaul: Key challenges and approaches,” with nonzero boresight pointing errors,” IEEE Trans. Commun., vol. 62,
IEEE Wireless Commun., vol. 25, no. 3, pp. 154–160, Jun. 2018. no. 2, pp. 713–725, Feb. 2014.

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: HOVERING UAV-BASED FSO COMMUNICATIONS 2959

[34] V. S. Adamchik and O. I. Marichev, “The algorithm for calculating inte- Jun-Bo Wang (Member, IEEE) received the B.S.
grals of hypergeometric type functions and its realization in REDUCE degree in computer science from the Hefei Uni-
system,” in Proc. Int. Symp. Symbolic Algebr. Comput. (ISSAC), Tokyo, versity of Technology, Hefei, China, in 2003, and
Japan, 1990, pp. 212–224. the Ph.D. degree in communications engineering
[35] M. Sheng, P. Jiang, Q. Hu, Q. Su, and X.-X. Xie, “End-to-end average from the National Mobile Communications Research
BER analysis for multihop free-space optical communications with Laboratory, Southeast University, Nanjing, China,
pointing errors,” J. Opt., vol. 15, no. 5, pp. 1–7, Apr. 2013. in 2008. He is currently an Associate Professor with
[36] S. C. Gupta, “Delta function,” IEEE Trans. Educ., vol. E-7, no. 1, the National Mobile Communications Research Lab-
pp. 16–22, Mar. 1964. oratory, Southeast University. From October 2008 to
[37] Wolfram. The Wolfram Functions Site. Accessed: Jun. 20, 2021. August 2013, he was with the Nanjing Univer-
[Online]. Available: http://functions.wolfram.com/ sity of Aeronautics and Astronautics, China. From
[38] N. Wang and J. Cheng, “Moment-based estimation for the shape March 2011 to February 2013, he was a Post-Doctoral Fellow with the
parameters of the gamma-gamma atmospheric turbulence model,” Opt. National Laboratory for Information Science and Technology, Tsinghua Uni-
Exp., vol. 18, no. 12, pp. 12824–12831, Jun. 2010. versity, Beijing, China. From September 2016 to September 2018, he held
the European Commission Marie Skłodowska-Curie Fellowship. He was a
Research Fellow with the University of Kent, U.K. His current research
interests include visible light communications, information theory, and coding.

Jin-Yuan Wang (Member, IEEE) received the

B.S. degree in communication engineering from Min Lin (Member, IEEE) received the B.S. degree
the Shandong University of Science and Technol- in electrical engineering from the National Uni-
ogy, Qingdao, China, in 2009, the M.S. degree in versity of Defense Technology, Changsha, China,
electronic and communication engineering from the in 1993, the M.S. degree in electrical engineering
Nanjing University of Aeronautics and Astronautics, from the Nanjing Institute of Communication Engi-
Nanjing, China, in 2012, and the Ph.D. degree in neering, Nanjing, China, in 2000, and the Ph.D.
information and communication engineering from
degree in electrical engineering from Southeast Uni-
Southeast University, Nanjing, in 2015. From Jan- versity, Nanjing, in 2008. From April 2015 to
uary 2016 to June 2019, he was a Lecturer with October 2015, he has visited the University of Cal-
the Nanjing University of Posts and Telecommuni- ifornia, Irvine, as a Senior Research Fellow. He is
cations, Nanjing. Since July 2019, he has been an Associate Professor with
currently a Professor and a Supervisor of Ph.D. and
the Nanjing University of Posts and Telecommunications. He has authored or graduate students with the Nanjing University of Posts and Telecommuni-
coauthored over 100 journal articles/conference papers. His current research cations, Nanjing, China. He has authored or coauthored over 130 articles.
interest includes visible light communications. He serves as the Workshop
His current research interests include wireless communications and array
Chair for ICFEICT 2021. He serves as a Technical Program Committee signal processing. He has served as the Track Chair for Satellite and Space
Member for many international conferences, such as IEEE ICC and WTS. Communications (SSC) of IEEE ICC 2019 and a TPC Member for many
He also serves as a reviewer for many journals. IEEE sponsored conferences, such as IEEE ICC and GLOBECOM.

Julian Cheng (Senior Member, IEEE) received the

Yang Ma received the B.S. degree in integrated B.Eng. degree (Hons.) in electrical engineering from
circuit design and integrated systems from Hangzhou the University of Victoria, Victoria, BC, Canada,
Dianzi University, Hangzhou, China, in 2018, and in 1995, the M.Sc. (Eng.) degree in mathematics
the M.S. degree in communication and information and engineering from Queens University, Kingston,
systems from the National Mobile Communications ON, Canada, in 1997, and the Ph.D. degree in
Research Laboratory, Southeast University, Nanjing, electrical engineering from the University of Alberta,
China, in 2021. His current research interests include Edmonton, AB, Canada, in 2003. He is currently a
wireless optical communications and unmanned aer- Full Professor with the School of Engineering, Fac-
ial vehicle communications. ulty of Applied Science, The University of British
Columbia, Kelowna, BC, Canada. He was with Bell
Northern Research and NORTEL Networks. His current research interests
include machine learning and deep learning for wireless communications, opti-
cal wireless technology, and quantum communications. He was the Co-Chair
of the 12th Canadian Workshop on Information Theory in 2011, the 28th Bien-
nial Symposium on Communications in 2016, and the Sixth EAI International
Rong-Rong Lu received the B.S. degree in elec- Conference on Game Theory for Networks in 2016. He currently serves as an
tronic science and technology from the Nanjing Uni- Area Editor for the IEEE T RANSACTIONS ON C OMMUNICATIONS. He was a
versity of Posts and Telecommunications, Nanjing, past Associate Editor of the IEEE T RANSACTIONS ON C OMMUNICATIONS,
China, in 2019, where she is currently pursuing the the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS, the IEEE
M.S. degree in communication and information sys- C OMMUNICATIONS L ETTERS , and IEEE A CCESS . He served as a Guest
tems. Her current research interest includes optical Editor for a Special Issue of the IEEE J OURNAL ON S ELECTED A REAS
wireless communications. IN C OMMUNICATIONS on Optical Wireless Communications. He is also a
Registered Professional Engineer with the Province of British Columbia,
Canada. He serves as the President for the Canadian Society of Information
Theory and the Secretary for the Radio Communications Technical Committee
of IEEE Communications Society.

Authorized licensed use limited to: CNR Area Ricerca Pisa. Downloaded on September 05,2023 at 15:01:49 UTC from IEEE Xplore. Restrictions apply.