1

On the Transmission Probabilities in Quantum

Key Distribution Systems over FSO Links
Hui Zhao, and Mohamed-Slim Alouini

Abstract
arXiv:2010.04457v1 [cs.IT] 9 Oct 2020

In this paper, we investigate the transmission probabilities in three cases (depending only on the
legitimate receiver, depending only the eavesdropper, and depending on both legitimate receiver and
eavesdropper) in quantum key distribution (QKD) systems over free-space optical links. To be more
realistic, we consider a generalized pointing error scenario, where the azimuth and elevation pointing error
angles caused by stochastic jitters and vibrations in the legitimate receiver platform are independently
distributed according to a non-identical normal distribution. Taking these assumptions into account, we
derive approximate expressions of transmission probabilities by using the Gaussian quadrature method.
To simplify the expressions and get some physical insights, some asymptotic analysis on the transmission
probabilities is presented based on asymptotic expression for the generalized Marcum Q-function when
the telescope gain at the legitimate receiver approaches to infinity. Moreover, from the asymptotic
expression for the generalized Marcum Q-function, the asymptotic outage probability over Beckmann
fading channels (a general channel model including Rayleigh, Rice, and Hoyt fading channels) can be
also easily derived when the average signal-to-noise ratio is sufficiently large, which shows the diversity
order and array gain.

Index Terms

Beckmann distribution, free-space optics, generalized pointing errors, quantum key distribution, and
transmission probability.

This paper was accepted for publication in IEEE Transactions on Communications on Oct. 4, 2020.
This work was funded by the office of sponsored research (OSR) at KAUST, and the European Research Council under the
EU Horizon 2020 research and innovation program/ERC grant agreement no. 725929 (ERC project DUALITY).
H. Zhao was with the Computer, Electrical, and Mathematical Science and Engineering Division, King Abdullah University
of Science and Technology, Thuwal 23955-6900, Saudi Arabia, and he is now with the Communication Systems Department,
EURECOM, Sophia Antipolis 06410, France (email: hui.zhao@kaust.edu.sa).
M.-S. Alouini is with the Computer, Electrical, and Mathematical Science and Engineering Division, King Abdullah University
of Science and Technology, Thuwal 23955-6900, Saudi Arabia (email: slim.alouini@kaust.edu.sa).
Digital Object Identifier 10.1109/TCOMM.2020.3030250

October 12, 2020 DRAFT

 2

I. I NTRODUCTION

Quantum communication provides a promising solution to break the Shannon channel capacity limit
[1] and achieve an unprecedented level of security [2] simultaneously, two competing tasks which cannot
be realized in conventional technologies [3]. In this context, quantum key distribution (QKD) or quantum
cryptography is a method for sharing the secret cryptographic keys between two legitimate parties to
achieve the secure communications by taking advantage of the laws of quantum mechanics and quantum
non-cloning theorem [4], [5]. However, further investigation and real application of QKD did not attract
much attention until it was proved that the quantum computer was able to break public-key cryptosystems,
which are commonly used in the modern cryptography [6], [7].
The connection implementation of QKD includes two main medium, i.e., fiber cable and free-space
optics (FSO). Compared to the fiber cable, the implementation of QKD over FSO links is more convenient
and easier due to the flexibility of the free space connection and satellite support for distributing
quantum keys worldwide [8], [9]. Moreover, FSO is an alternative transport technology to interconnect
high capacity networking segments in current and future communication systems, because of its cost-
effectiveness, high-bandwidth availability, and interference-immunity [10]. The authors in [11]–[18] have
presented some basic performance analysis works over classical FSO links or hybrid RF-FSO links, but
those previous works do not consider the QKD mechanism. Actually, the research work about QKD over
FSO links in the communications field is considerably limited.
In practical systems, the security of QKD strongly depends on the device implementation [19]–[22].
That is, a third party may have a side channel by making use of any deviation of a QKD device from
the theoretical model. For example, two zero-error attacks on commercial QKD systems were reported
where the defects in quantum signal encoding and detection were exploited [19], [20]. Besides, some
imperfections in QKD designs can be also exploited by a plethora of quantum hacking attacks using
current technologies [23].
In real commercial QKD implementations, a single-photon mechanism is typically used to convey
the information, and the corresponding common detection scheme is called single photon avalanche
photodiodes (SPADs) where the SPAD diode is operated in Geiger mode (reverse-biased above the
breakdown voltage to create an avalanche) to count single-photons [24]. However, this detection scheme
can lead to the information leakage, because the avalanche created by the incoming photon can emit a
secondary photon which may be intercepted by a third party, namely the eavesdropper. This secondary
photon emission (the photon emission in the sender is the first emission) is called backflash, which
is quenched along with the avalanche, i.e., the backflash is quenched if the detection bias is lowered

October 12, 2020 DRAFT

 3

below the breakdown voltage [22], [24]. Previous measurements show that the probability of detecting
backflash is greater than 0.4%, and more than 0.04 photons emerging from the devices are contained
in the backflash given the 10% nominal detection efficiency of SPADs [22], [25]. These measurements
provide a reference for the backflash resulting in the information leakage, although the measurement
results may change in different detector types and optical components.
An unevadable vulnerability in the FSO QKD systems is the random pointing error due to stochastic
jitters and vibrations which can be caused by building sway, thermal expansion, and week earthquakes, in
the urban FSO systems [26]–[28]. Similarly, for satellite communications, there are internal and external
reasons for stochastic vibrations [29]. For example, the structure deformations caused by temperature
gradients, and the gravitational force inhomogeneity over the satellite orbit, are two main external
reasons for stochastic vibrations in satellite systems. The internal sources include electronic noise, antenna
pointing operation, and solar array driver [30].
The authors in [31] first investigated the performance of received powers at both the legitimate receiver
and eavesdropper in FSO QKD systems with taking random pointing errors into account, and derived
the closed-form expressions for the corresponding average received powers. However, the authors in [31]
assumed that the azimuth and elevation pointing errors are identically independently distributed, and more
specifically, these two pointing errors are modeled by Gaussian distribution with zero-mean and the same
variance which may be a little ideal. In the practical systems, the mean and variance of these two pointing
errors are typically different. Moreover, the authors in [31] did not consider the transmission probability
depending on the received power threshold, which is very important and useful for the system evaluation
and design. This is because we need to know the transmission probability depending on some conditions
in the average level, apart from the average received powers, when evaluating and designing FSO QKD
systems.
Actually, the pointing error angle, divided into the azimuth and elevation pointing errors, can be
modeled by the Beckmann distribution [32]–[35], a generalized model including the Rayleigh, Hoyt
and Rice models. Specifically, the Beckmann model is reduced into the Hoyt model for zero-mean and
different variance of two sub-part pointing errors, and the different non-zero mean and equal variance case
refers to the Rice case in the Beckmann distribution. The zero-mean and equal variance case considered
in [31] is the most simple scenario in the Beckmann distribution, denoted by the Rayleigh case. The
authors in [36] expanded the pointing error model in [31] to the Beckmann distribution. In [36], exact
closed-form expressions for the average received powers at both the legitimate receiver and eavesdropper
were derived, as well as finding the maximum points of the telescope gain at the legitimate receiver in
some special cases analytically. However, the authors in [36] still did not investigate the transmission

October 12, 2020 DRAFT

 4

probabilities depending on a variety practical conditions, which is also important for the FSO QKD system
evaluation. Similar to [37]–[39], the transmission probability in this paper is defined as the probability
that the received power is satisfied one or more pre-set thresholds in the FSO QKD system, which is
obviously a natural variant of the outage probability in traditional communications [35].
Motivated by observing those facts outlined above, we investigate the performance of QKD systems
over FSO links in terms of transmission probabilities depending on three different conditions. The main
contributions of this paper are summarized as follows:

1) Closed-form expressions with a high accuracy for transmission probabilities depending on three dif-
ferent conditions, i.e., legitimate receiver, eavesdropper and both legitimate receiver and eavesdrop-
per, are derived based on the Gaussian quadrature rule, where the accuracy grows with increasing
the summation terms in the Gaussian quadrature.
2) Asymptotic expression for the generalized Marcum Q-function is derived after some mathematical
manipulations, which can be used to derive the asymptotic outage probability over Beckmann fading
channels when the average signal-to-noise ratio (SNR) approaches to infinity, showing the diversity
order and array gain, since those two metrics govern the outage probability behaviour in high SNRs.
3) By using the asymptotic result for the generalized Marcum Q-function, the asymptotic expressions
for three transmission probabilities are easily derived, which are valid in the high value region
of the telescope gain at the legitimate receiver. Besides providing some insights, these asymptotic
expressions are significantly concise, resulting in a much faster calculation than the analytical
expressions that need to be computed based on the Gaussian quadrature rule.
4) We also present some specific expressions for those three transmission probabilities in some
simplified cases which result in exact expressions or more concise forms. More specifically, exact
closed-form expressions in Rayleigh, Hoyt and Rice cases (three special cases of the Beckmann
distribution) for the transmission probability depending only on the legitimate receiver are given. In
the Rayleigh case, we present a more concise expression for the transmission probability depending
on both the legitimate receiver and eavesdropper.

The remainder of this paper is organized as follows. The system model is presented in Section II. The
transmission probabilities depending on three different conditions are analyzed in Sections III, IV and V,
respectively. In Section VI, some numerical results are generated and used to validate the correctness of
derived closed-form expressions, as well as presenting some interesting comparisons. Section VII finally
concludes the paper.

October 12, 2020 DRAFT

 5

Fig. 1: Secure QKD System Over FSO links

II. S YSTEM M ODEL

As depicted in Fig. 1, there is a sender (Alice) located on a absolutely static platform1 communicating
a legitimate receiver (Bob) located on a platform suffering from stochastic jitters and vibrations, such as
a laser satellite system, over a FSO link in open areas. This vibrating platform in the legitimate receiver
results in a random pointing error, where the stochastic deviation angle (θ ) is divided into two parts, i.e.,
the azimuth pointing error (θH ) and the elevation pointing error (θV ), and therefore, θ can be written as
q
θ = θH 2 + θ2 ,
V

where θH and θV are normally assumed to be independent Gaussian random variables, i.e., θH ∼
2 ) and θ ∼ N (µ , σ 2 ), where µ and σ 2 (or µ and σ 2 ) represent the mean and variance
N (µH , σH V V V H H V V

of θH (or θV ), respectively.
The SPADs detection scheme adopted by the legitimate receiver is assumed in this system setting. The
received power at Bob is2 [26], [29], [31]

PD (θ) = K1 L (θ) GD ,

where L(θ) = exp(−GD θ 2 ) is the pointing loss factor, GD = (πdD /λ1 )2 is the telescope gain at the
legitimate receiver, dD is the unobscured circular aperture diameter of the telescope, λ1 is the wavelength,

1
We can also consider a non-static platform for the sender. In a relative motion aspect, if the sender is assumed to be relatively
static to the receiver, this will induce the same analysis.
2
This power can be also regarded as an average received power over instantaneous received photon counts influenced by both
the shot noise and the dead time of the SPAD receiver. Here, we focus on the transmission probability depending on the average
power, rather than the instantaneous performance.

October 12, 2020 DRAFT

 6

and K1 is a constant depending only on the system design, given by

 
LA (Z1 ) λ1 2
K1 = ηq PS GS ηS ηD ,
Z12 4π
in which PS is the optical transmitter power, ηq is the quantum efficiency, GS is the telescope gain of the
sender, LA (·) is the atmospheric loss, and Z1 is the distance between the sender and legitimate receiver.
As discussed in the introduction section, the SPADs detection scheme leads to the backflash due to
the secondary photon emission caused by the avalanche. A third party (eavesdropper, Eve) can make use
of this backflash to intercept the secondary photon, and thereby wiretapping the conveyed information
from the sender to the legitimate receiver. The received power at the eavesdropper is given by [31], [36]

PE (θE , α) = K2 PD (θ) L (θE ) GD ,

q
where θE = (θV + α)2 + θH
2 , α is the pointing direction error angle in the wiretap FSO link, and K
2

is a system constant, given by

 
LA (Z2 ) λ2 2
K2 = ηB ηq ηD ηE GE ,
Z22 4π
in which ηB is the probability of backflash, Z2 is the distance between the legitimate receiver and
eavesdropper, ηE and GE are the optical efficiency and telescope gain of the eavesdropper respectively,
and λ2 is the backflash wavelength.

III. T RANSMISSION P ROBABILITY D EPENDING ONLY ON L EGITIMATE R ECEIVER

In this section, we want to evaluate the transmission probability performance given a received power
threshold (λD ) at the legitimate receiver. In this context, the transmission probability depending only on
the legitimate receiver (TPLR) is defined as
n o n o
TPLR = Pr PD (θ) ≥ λD = Pr K1 GD L (θ) ≥ λD , (1)

which can be further written by substituting the expression for L(θ), given by
 
2 −1 λD
TPLR = Pr θ ≤ ln . (2)
GD K1 GD
Let ΘD = −1
GD ln Kλ1 G
D
D
. By substituting the probability density functions (PDFs) of θV and θH into (2),
the TPLR for arbitrary µV , µH , σV and σH can be derived as
n o q p 
2 2 2
TPLR = Pr θ ≤ ΘD = Pr θV + θH ≤ ΘD
 
−µV )2 (θH −µH )2
ZZ ZZ exp − (θV2σ 2 − 2
2σH
V
= fθV (θV )fθH (θH )dθV dθH = dθV dθH , (3)
√ √ 2πσV σH
√ √
θV2 +θH
2
≤ ΘD θV2 +θH
2
≤ ΘD

October 12, 2020 DRAFT

 7

where the double integral, unfortunately, cannot be solved in a closed-form, and therefore, there are some
approximation methods for this double integral, such as [32] and [33]. However, those approximation
methods proposed by [32] and [33] are still complicated for calculation.
Here, we provide another approximation method based on the Gaussian quadrature rule [40, Ch. 9],
shown in Theorem 1.

Theorem 1. An approximate result for the TPLR based on the Gaussian quadrature rule is
1 XN
TPLR ≈ 1 − √ ωi fTPLR (xi ), (4)
π i=1

where N , ωi , and xi are the summation terms, weights, and selected points of the Gauss-Hermite
quadrature (GHQ, a special case of Gaussian quadrature), and fTPLR (·) is given by
r  !
√ 1 √ 2 √ 2 
fTPLR (x) = Q 1 λ, ΘD − 2σV x + µV × I ΘD ≥ 2σV x + µV , (5)
2 σH

in which λ = µ2H /σH2 represents the noncentrality parameter, Q (·, ·) denotes the generalized Marcum
· 

1, if A is true;
Q-function [42]. and I{·} denotes the indicator function, i.e., I {A} =

0, otherwise.

Proof. See Appendix A.

Remark 1. Although a high accuracy requires many terms in (4), resulting in a much slower calculation,
especially when Q 1 (·) cannot be directly calculated in some softwares, such as Matlab, Theorem 1
2

provides an analytical tool to investigate the TPLR. We will present an asymptotic expression presented
in the III-A subsection, rather than (4), for getting a high accuracy result in a special case.

As θV follows the Gaussian distribution, for σV2 ≪ σH

2 , according to [44, Eq. (4)], the TPLR in (54)

can be robustly approximated by

( s !)
√ ΘD − θV2 
TPLR = EθV 1 − Q1 λ, 2 I ΘD ≥ θV2
2 σH
σV2 ≪σH
2
2 1  √  1  √ 
≈ φTPLR (µV ) + φTPLR µV + 3σV + φTPLR µV − 3σV , (6)
3 6 6
where
s !
√ Θ D − x2
φTPLR (x) = 1 − Q 1 λ, 2 I {ΘD ≥ x2 } . (7)
2 σH

This robust approximation for σV2 ≪ σH

2 was proposed by [44]–[46]. This robust result becomes more

tighter along with the ratio of σV2 /σH

2 approaching to zero.

October 12, 2020 DRAFT

 8

A. Asymptotic Result for TPLR as ΘD → 0

Before presenting the asymptotic analysis for TPLR, we first give the following proposition.

Proposition 1. The asymptotic expression for QM (a, b), the generalized Marcum Q-function, as b → 0,
is given by
 2
b→0 exp − a2

2M 2M +1
QM (a, b) ≃ 1 − b + o b , (8)
Γ (M + 1) 2M
where M , a, b are non-negative, o(·) and Γ(·) denote the higher order term and Gamma function [42],
respectively.

Proof. See Appendix B.

We present some numerical results in Fig. 2 to validate the correctness of the derived asymptotic
expression for the generalized Marcum Q-function. It is obvious that the asymptotic results match the
exact results very well when b is sufficiently small.

10 0

-1
10
1 − QM (a, b)

-2
10

-3
10
M = 3, 2, 1, 0.5
-4
10 Exact
Asymptotic

-15 -10 -5 0 5 10
b [dB]
Fig. 2: 1 − QM (a, b) versus b for a = 1.

For GD → +∞, we have

−1 λD
ΘD = ln → 0+ , as GD → +∞. (9)
GD K1 GD
In the following, we will analyze the asymptotic behaviour of TPLR when ΘD → 0+ (or equivalently
GD → +∞).

Lemma 1. For ΘD → 0 (or equivalently GD → +∞), by using Proposition 1, the asymptotic expression
for TPLR can be derived as
√  
ΘD →0 π λ µ2V
TPLR ≃ exp − − 2 ΘD . (10)
4Γ (1.5) σV σH 2 2σV

October 12, 2020 DRAFT

 9

Proof. See Appendix C.

Remark 2. In the GD saturation case, a linear mapping from ΘD to the TPLR is derived, shown in
Lemma 1, which is interesting in the performance analysis aspect. This asymptotic expression can not
only simplify the TPLR calculation significantly, but also reveal the relationship between the TPLR and
ΘD in the GD saturation case.

Remark 3. In fact, the asymptotic expression presented in Lemma 1 can be also viewed as the asymptotic
result for the outage probability over Beckmann fading channels (including Rayleigh, Hoyt and Rice fading
channels), where ΘD represents the received SNR threshold, and the instantaneous SNR at the receiver
is γ = θV2 + θH
2 .

Although the atmospheric turbulence is not a main investigation in this paper (we will consider this
issue in our future work), we can simply analyze this impact on the TPLR in the GD saturation case based
on Lemma 1, shown in Corollary 1 where the turbulence is modeled by a Gamma-Gamma distribution,
a widely adopted turbulence model [34].

Corollary 1. In the GD saturation case, if the atmospheric turbulence modeled by a Gamma-Gamma

distribution is considered over the FSO link, the asymptotic result for the TPLR is
√  
π λ µ2V
TPLR = exp − − 2 ΘD
4Γ (1.5) σV σH 2 2σV
√  
π (ψ(αD ) + ψ(βD ) − ln(αD βD )) λ µ2V
+ exp − − 2 , (11)
4GD Γ (1.5) σV σH 2 2σV
| {z }
Atmospheric Turbulence

where ψ(·) denotes the digamma function [42], αD and βD are the fading parameters of large-scale and
small-scale fluctuations, respectively.

Proof. It is obvious that the TPLR in (2) becomes

n o  
−1 λD
TPLR = Pr ID K1 GD L(θ) ≥ λD = Pr θ 2 ≤ ln
GD ID K1 GD
 
1
= Pr θ 2 ≤ ΘD + ln ID , (12)
GD
where ID represents the atmospheric turbulence following a Gamma-Gamma distribution with the PDF
[34, Eq. (11)],
Dα D+β
2(αD βD ) 2 αD +βD
 p 
fID (x) = x 2 −1 KαD −βD 2 αD βD x , (13)
Γ(αD )Γ(βD )
where K· (·) denotes the modified Bessel function of the second kind [42].

October 12, 2020 DRAFT

 10

1
Define Θ′D = ΘD + GD ln ID . In view of Lemma 1, it is easy to derive the asymptotic result for the
TPLR as
√  
ΘD →0 π λ µ2V 
TPLR ≃ exp − − 2 · EID Θ′D
4Γ (1.5) σV σH 2 2σV
√  2  √  
π λ µV π λ µ2V
= exp − − 2 ΘD + exp − − 2 · EID {ln ID } .
4Γ (1.5) σV σH 2 2σV 4GD Γ (1.5) σV σH 2 2σV
| {z }
Atmospheric Turbulence

(14)
By using the PDF of ID , the expectation of ln ID with respect to ID can be obtained as [34]
αD +βD Z ∞
2(αD βD ) 2 αD +βD
 p 
E {ln ID } = × ln(x) · x 2 −1 KαD −βD 2 αD βD x dx
Γ(αD )Γ(βD ) 0

= ψ(αD ) + ψ(βD ) − ln(αD βD ), (15)

Combining (14) and (15) yields Corollary 1.

Remark 4. Considering ψ(x) − ln x ≤ 0 for x > 0 in Corollary 1, we can conclude that the impact
of atmospheric turbulence is always negative to TPLR in the GD saturation case. Further, the negative
impact is quantified by the second part in (11), compared to the TPLR without atmospheric turbulence
in Lemma 1.

B. Special Case for µV = µH = 0 and σV 6= σH

To get the exact closed-form expression for TPLR, we relax the conditions for the statistical charac-
teristics of θV and θH , i.e., µV = µH = 0 and σV 6= σH . In this simplified case, θ 2 follows the Hoyt
distribution, and the corresponding TPLR is given by [43]

2 !
2q 1 − q2 1 + q2
TPLR = Ie ,
ΘD , (16)
1 + q2 1 + q 2 4q 2 σV2 + σH 2

where Ie(·, ·) denotes the Rice Ie-function defined in [43, Eq. (3)], and q ∈ [0, 1] is given by


σV /σH , for σV ≤ σH ;
q= (17)

σH /σV , for σV > σH .

From the asymptotic analysis for the general parameter settings, the asymptotic expression in the Hoyt
distribution case (µV = µH = 0) can be easily derived as
√
ΘD →0 π
TPLR ≃ ΘD . (18)
4σV σH Γ (1.5)

October 12, 2020 DRAFT

 11

The asymptotic expression can be also easily derived by using the asymptotic result for the generalized
Marcum Q-function in Proposition 1, because the Rice Ie-function can be written in the Marcum Q-
function form, given by [43]
Z x
Ie (k, x) = exp (−t) I0 (kt) dt
0
" r !
1 p  r p 
=√ Q1 1 + 1 − k2 x, 1 − 1 − k2 x
1 − k2
r !#
p  r p 
−Q1 1 − 1 − k2 x, 1 + 1 − k2 x . (19)

When q = 1, i.e., σV = σH = σ , the distribution of θ is reduced into the Rayleigh distribution, and
the corresponding TPLR becomes
 
ΘD
TPLR = 1 − exp − 2 . (20)
2σ

C. Special Case for µV 6= µH 6= 0 and σV = σH = σ

For µV 6= µH and σV = σH = σ , i.e., the Rice case, we can rewrite the TPLR as

TPLR = Pr θV2 + θH 2
≤ ΘD
 2  2 
 θ θ Θ 
V  + q H  ≤ D
= Pr  q
 µ2 + µ2 µ2 + µ2 µ2V + µ2H 
V H V H
v 2  2 s 
u 
u θV θH ΘD 
u
= Pr t q  + q
  ≤ . (21)

 µ2V + µ2H µ2V + µ2H µ2V + µ2H 

µV µH
Let sin β = √ and cos β = √ It is easy to see that
µ2V +µ2H µ2V +µ2H
 
θV σ2
q ∼ N sin β, 2 ,
µ2V + µ2H µV + µ2H
 
θH σ2
q ∼ N cos β, 2 .
µ2V + µ2H µV + µ2H

From the definition of the Rice distribution, we can conclude that

v 2  2  
u
u
u θV  + q H θ  ∼ Rice 1, q σ ,
t q
2
µV + µH2 µ2V + µ2H µ2V + µ2H
and the corresponding TPLR can be easily derived by using the well-known standard cumulative distri-
bution function (CDF) of the Rice distribution, given by
q 
µ2V + µ2H √ΘD
TPLR = 1 − Q1  , . (22)
σ σ

October 12, 2020 DRAFT

 12

From the derivation in Proposition 1, when ΘD → 0, the asymptotic TPLR is

 2 
ΘD →0 µV + µ2H 1

TPLR ≃ exp − 2 2
ΘD + o Θ2D . (23)
2σ 2σ

IV. T RANSMISSION P ROBABILITY D EPENDING ONLY ON E AVESDROPPER

The transmission probability depending only on eavesdropper (TPE) is defined as the probability that
the received power at the eavesdropper is less than a threshold (λE ), i.e.,
n o n

o
TPE = Pr K2 PD (θ) L (θE ) GD ≤ λE = Pr K1 K2 G2D exp −GD 2θ 2 + 2αθV + α2 ≤ λE
( )
2 −1 λE α2
= Pr θ + αθV ≥ ln − . (24)
2GD K1 K2 G2D 2
| {z }
ΘE

An approximation for the TPE will be given in Theorem 2 based on the GHQ.

Theorem 2. The TPE can be approximated as

1 XN
TPE ≈ √ ωi fTPE (xi ), (25)
π i=1

where N , ωi , and xi are the same as those in (4), and fTPE (x) is given in (26),
 v
u Θ − √2σ x + µ
2 − α √2σ x + µ

u
 √ E V V V V
fTPE (x) = Q 1  λ, t 2
2 σH
 √ 2 √ 
×I ΘE ≥ 2σV x + µV + α 2σV x + µV . (26)

Proof. See Appendix D.

From [44]–[46], for σV2 ≪ σH

2 , the TPE can be robustly approximated by

(a) 2 1  √  1  √ 
TPE ≈ φTPE (µV ) + φTPE µV + 3σV + φTPE µV − 3σV , (27)
3 6 6
where (a) follows [44, Eq. (4)], and φTPE (·) is given by
s !
√ ΘE − x2 − αx
φTPE (x) = Q 1 λ, 2 I {ΘE ≥ x2 + αx} . (28)
2 σH

A. Analysis on ΘE versus GD

ΘE is given by
−1 λE α2
ΘE = ln − . (29)
2GD K1 K2 G2D 2

October 12, 2020 DRAFT

 13

The first derivative of ΘE with respect to GD is

λE
∂ΘE ln K1 K 2 + 2
2 GD
= 2 . (30)
∂GD 2GD
∂ΘE
Let ∂GD = 0. The root for GD > 0 is
r
λE
G∗D =e . (31)
K1 K2
We can conclude that ΘE is an increasing function over GD ∈ (0, G∗D ), and a decreasing function over
GD ∈ (G∗D , ∞). The maximum value of ΘE is
√
K1 K2 α2
Θ∗E = ΘE |GD =G∗D = √ − . (32)
e λE 2
For GD → ∞, we have
−1 λE α2 α2
lim ln − = − . (33)
GD →∞ 2GD K1 K2 G2D 2 2
For GD → 0+ , we have the limit
−1 λE α2
lim + ln − → −∞. (34)
GD →0 2GD K1 K2 G2D 2

Remark 5. Combining the analysis on ΘE and (24), we can conclude that the TPE is decreasing over
GD ∈ (0, G∗D ) due to the increase in ΘE , while the TPE is increasing over GD ∈ (G∗D , ∞) due to the
decrease in ΘE . When GD is sufficiently large, the TPE will increase to an upper bound, because ΘE
goes to a constant (−α2 /2). If GD → 0+ , ΘE → −∞, resulting in the TPE going to 1.

B. Asymptotic Analysis for TPE

−α2
Lemma 2. When ΘE → 4 from the right side in the real number axis, the asymptotic result for TPE
is given by
2 √
!  
ΘE → −α
4 π α2 + 4ΘE λ (α/2 + µV )2 −α2
TPE ≃ 1− exp − − × I ΘE ≥ . (35)
16σH σV Γ (1.5) 2 2σV2 4
−α2
For ΘE ≤ 4 , the TPE is exactly equal to 1.

Proof. See Appendix E.

Remark 6. α reflects the distance between the sender and the eavesdropper. Specifically, α → 0 means
that the eavesdropper is close to the sender. Lemma 2 presents a quantitative relationship between ΘE
−α2
(or GD ) and α. When ΘE ≤ 4 , the event of received power at the eavesdropper below the pre-set
threshold (λE ) happens with probability one. Further, Lemma 2 reveals the linear trend of TPE with
 2 +
respect to ΘE when ΘE → −α 4 .

October 12, 2020 DRAFT

 14

V. TRANSMISSION PROBABILITY DEPENDING ON BOTH LEGITIMATE RECEIVER AND EAVESDROPPER

The transmission probability depending on both the legitimate receiver and the eavesdropper (TPRE)
is defined as
n o n o
TPRE = Pr PD (θ) ≥ λD , PE (θE ) ≤ λE = Pr K1 GD L (θ) ≥ λD , K1 K2 G2D L (θ) L (θE ) ≤ λE ,
(36)
which can be further written by using the definition of ΘD and ΘE ,
 
n
2 2 2 2
o ΘE − θV2 − αθV 2
θH ΘD − θV2
TPRE = Pr θH ≤ ΘD − θV , θH ≥ ΘE − θV − αθV = Pr 2 ≤ 2 ≤ 2 .
σH σH σH
(37)
By applying the GHQ method, an approximate result for the TPRE is derived in Theorem 3.

Theorem 3. A closed-form expression for an approximate TPRE is

1 XN
TPRE ≈ √ ωi fTPRE (xi ), (38)
π i=1

where N , ωi , xi are the same as those in (4), and fTPRE (·) is given in (39),
 " r
ΘE − ΘD µV √ 1 √ 2 √ 
fTPRE (x) = I x ≥ √ −√ Q1 λ, ΘE − 2σV x + µV − α 2σV x + µV
α 2σV 2σV 2 σH
 !
√ 2 √ 
× I ΘE ≥ 2σV x + µV + α 2σV x + µV
s !#
√ 1 √ 2   √ 2 
−Q 1 λ, ΘD − 2σV x + µV I ΘD ≥ 2σV x + µV . (39)
2 σH

Proof. See Appendix F.

For σV2 ≪ σH
2 , according to [44]–[46], the TPRE in (73) can be robustly approximated by

σV2 ≪σH
2
2 1  √  1  √ 
TPRE ≈ φTPRE (µV ) + φTPRE µV + 3σV + φTPRE µV − 3σV , (40)
3 6 6
where φTPRE (x) is given in (41),
 " s  !
ΘE ΘD √ ΘE x2 − αx
φTPRE (x) =I x ≥ − Q1 λ, 2 − 2 I {ΘE ≥ 2 (x2 + αx)}
2α α 2 2σH σH
s !#
√ Θ D − x2
−Q 1 λ, 2 I {ΘD ≥ x2 } . (41)
2 σH

October 12, 2020 DRAFT

 15

A. Asymptotic Analysis

Lemma 3. The asymptotic expression for TPRE as GD → +∞ is the same as that for TPLR, given by
√  
GD →+∞ πΘD λ µ2V
TPRE ≃ exp − − 2 . (42)
4σH σV Γ (1.5) 2 2σV

Proof. The proof by solid mathematical manipulations is removed due to the page limitation.
−α2
Lemma 3 can be explained by the fact that for GD → +∞, ΘE → 2 , and ΘD → 0+ . From the
−α2
previous analysis, we know that for ΘE < 4 , there is no real root for ΘE − θV2 − αθV = 0, i.e.,
−α2
ΘE − θV2 − αθV is always less than zero for any θV . From the second equal sign in (37), for ΘE < 4 ,
n 2 2
o
the TPRE becomes TPRE = Pr σθH2 ≤ ΘDσ−θ 2
V
, which is exactly the definition of TPLR.
H H

Remark 7. The asymptotic result for the TPRE in Lemma 3 reveals that although the increase in GD
also induces an increase in the received power at the eavesdropper, the positive impact of increasing
GD on the TPRE will domain the overall performance when GD → ∞, and more exactly, there is a
convergence trend of TPLR and TPRE (i.e., the eavesdropping impact vanishes) in the GD saturation
case.

B. Special Case for µV = µH = 0 and σV = σH = σ

As the TPRE cannot be solved in an exact closed-form, even for the simplest case, i.e., Rayleigh
distribution. Here, we only analyze the special case for the TPRE in the Rayleigh case (µV = µH = 0
and σV = σH = σ ).
Let X = θV2 + θH
2 and Y = 2θ 2 + 2θ 2 + 2αθ = 2X + 2αθ . The TPRE can be rewritten as
V H V V
n

o
TPRE = Pr K1 GD exp (−GD X) ≥ λD , K1 K2 G2D exp −GD Y + α2 ≤ λE
(
)
−1 λD −1 λE exp GD α2
= Pr X ≤ ln , Y ≥ ln . (43)
GD K1 GD GD K1 K2 G2D
Let


−1 λD −1 λE exp GD α2
ΘD = ln , ΘE = ln .
GD K1 GD GD K1 K2 G2D
The TPRE is further written as
Z ΘD Z ∞
Pr {X ≤ ΘD , Y ≥ ΘE } = fX,Y (x, y)dydx, (44)
0 ΘE
where fX,Y (·, ·) represents the joint PDF of X and Y . From the derivation of fX,Y (·, ·) in Appendix G,
we can write the TPRE in an integral form as
 x  Z 2x+2α√x  #− 21
" 
y − 2x 2
Z ΘD
1
TPRE = exp − 2 √
x− dydx. (45)
4πσ 2 α 0 2σ max{ΘE ,−2α x+2x} 2α

October 12, 2020 DRAFT

 16

By considering the derivation in Appendix H, the TPRE can be easily approximated by

 
Θ D XN ′ ′ ΘD ′ ΘD
TPRE ≈ ωf x + , (46)
8πασ 2 i=0 i TPRE 2 i 2
where ωi′ and x′i are the weights and selected points over the standard integral interval (i.e., [-1,1]) in
′
the Gaussian quadrature respectively, and fTPRE (x) is given by
 x    √ 
′ max {ΘE , −2α x + 2x} − 2x
fTPRE (x) = exp − 2 απ − 2α arcsin √ . (47)
2σ 2α x

Remark 8. There is no special function involved in (46), resulting in a much faster calculation for TPRE.
Moreover, the built-in function for calculating the non-integer order generalized Marcum Q-function is
not available in some softwares, such as Matlab, which brings more difficulty to implement Theorem 3.

C. Further Simplification for the Rayleigh Case

√ √
If we let max {ΘE , −2α x + 2x} be always −2α x + 2x. The integral form of the TPRE in (45)
becomes
Z Z √ "  2 #− 12
ΘD  2x+2α x
1 x  y − 2x
TPRE = exp − 2 √
x− dydx. (48)
4πσ 2 α 0 2σ −2α x+2x 2α
By using the following integral identity,
Z √ "  2 #− 12
2x+2α x
y − 2x
√
x− dy = 2απ, (49)
−2α x+2x 2α
the closed-form expression for the TPRE can be finally derived as
 
ΘD
TPRE = 1 − exp − 2 . (50)
2σ
If GD → +∞, we have ΘD = 1
GD ln Kλ1 G
D
D
→ 0+ . In this case (ΘD → 0+ ), by using exp (−x) ≃ 1−x
for x → 0, the asymptotic result for TPRE is
ΘD 1 K1 GD
TPRE ≃ 2
= 2 ln . (51)
2σ 2σ GD λD
From the previous analysis, we know that the condition for using this very simple expression (50) is
−α2
ΘE < 4 . In this condition, the impact on the TPRE from the eavesdropper vanishes.

VI. N UMERICAL R ESULTS

A. TPLR Simulations

In this subsection, we run some Monte-Carlo simulations to validate the derived closed-form expres-
sions for the TPLR. By referring to the parameter settings in [31], [36], some selected simulation results

October 12, 2020 DRAFT

 17

are shown in Figs. 3–6, where 107 realizations are generated to get each average result according to the
statistical properties, i.e., µV , µH , σV and σH .
2 is obvious due to more weaker vibrations.
In Fig. 3, the growing trend of TPLR with decreasing σH
As expected, the TPLR is decreasing as ΘD approaches to zero (or equivalently GD → +∞).

0
10

-1
10
TPLR

2
10 -2 σH = 1 × 10−9 , Simulation
2
σH = 1 × 10−10 , Simulation
2
σH = 1 × 10−11 , Simulation
Gaussian Quadrature
Asymptotic
10 -3
-130 -120 -110 -100 -90 -80
ΘD [dB]

Fig. 3: TPLR versus ΘD for µV = 1 × 10−8 , µH = 5 × 10−8 , σV2 = 1 × 10−12 and N = 300 for Gaussian
quadrature.

Figs. 4–5 plot the TPLR versus ΘD for different variance cases in the Hoyt and Rice models respec-
tively, where the trends with respect to ΘD (or variance) are the same as those in Fig. 3. The proposed
asymptotic results match the exact results very well especially in the low ΘD region in Figs. 3–5. It
is worth noting that the asymptotic results are calculated much faster than the results by the Gaussian
quadrature rule, which provides an alternative and efficient method, especially when we need a very high
accuracy (several thousand terms may be needed in Gaussian quadrature method).
The results shown in Fig. 6 validate the high accuracy of the robust approximation proposed in [44]–
[46], especially for σV2 ≪ σH
2 . We can easily see that the gap between the exact and robust results almost

vanishes as the ratio of σV2 /σH

2 approaches to zero.

B. TPE Simulations

In this subsection, we run the Monte-Carlo simulation to validate the correctness of the derived closed-
form expressions for the TPE. To facilitate the simulation setting, we assume that PS = 0 dB, GS =
GE = 109 , ηS = ηD = ηE = 0.9, ηq = 0.1, ηB = 0.04, λ1 = λ2 = 780 nm, Z1 = Z2 = 900 km,
and LA (D1 ) = LA (D2 ) = 0.5, by considering Table I in [31]. As the robust approximation for the

October 12, 2020 DRAFT

 18

0
10
Simulation
Analysis
-1
10 Asymptotic

-2
10

TPLR
10 -3

-4
10
σV2 = 1 × {10−9 , 10−10 , 10−11 , 10−12 }
-5
10

-150 -140 -130 -120 -110 -100 -90 -80

ΘD [dB]
2 = 5 × 10−11 .
Fig. 4: TPLR in the Hoyt case versus ΘD for µV = µH = 0, and σH

10 0

-1
10

-2
10

Simulation
TPLR

10 -3 Analysis
Asymptotic
-4
10

-5
10 σ 2 = 1 × {10−9 , 10−10 , 10−11 , 10−12 }

-6
10

-150 -140 -130 -120 -110 -100 -90 -80

ΘD [dB]

Fig. 5: TPLR in the Rice case versus ΘD for σV = σH = σ , µV = 1 × 10−7 and µH = 5 × 10−7 .

Gaussian distribution has been investigated very well in [44]–[46], we do not validate the high accuracy
for σV2 ≪ σH
2 in the following simulations.

As shown in Fig. 7, the TPE remains 1 before a sharp decrease up to the lowest bound, and after
this lowest bound, the TPE grows rapidly to 1 as GD increases, which is exactly as the TPE changing
analysis in Remark 5. To show the lowest bound, Fig. 8 uses the log-scale to plot TPE in Fig. 7, where
this lowest bound grows with increasing µH . From Figs. 7–8, it is obvious that a large µH results in
a large TPE, which can be explained by the fact that the received power performance at the legitimate
receiver becomes worse, thereby accordingly decreasing the received power at the eavesdropper.
2 2
Fig. 9 plots the TPE versus ΘE from − α4 to 15α2 . As analyzed in the IV-B subsection, for ΘE → − α4
from the right side in the real number axis, the TPE can be approximated by a linear function. Moreover,

October 12, 2020 DRAFT

 19

10 0

10 -1

TPLR
-2
10

-3 σV2 = 1 × 10−12
10
σV2 = 1 × 10−14
σV2 = 1 × 10−18
Robust Approximation
-4
10
-150 -140 -130 -120 -110 -100 -90
ΘD [dB]
2 = 1 × 10−10 , µ = 1 × 10−8 and µ = 5 × 10−8 .
Fig. 6: TPLR versus ΘD for σH V H

there is a decreasing trend in the TPE with respect to α.

0.9

0.8

0.7

0.6
TPE

0.5

0.4

0.3

0.2 µH = 1 × 10−8 , Simulation

µH = 1 × 10−6 , Simulation
µH = 1 × 10−5 , Simulation
0.1
Gauss-Hermite Quadrature
0
8 10 12 14 16 18
10 10 10 10 10 10
GD
Fig. 7: TPE versus GD for µV = 1 × 10−7 , σV2 = 1 × 10−12 , σH
2 = 1 × 10−13 , λ
E = 1 × 10
−20 ,

α = 1 × 10−9 , and N = 30 for the Gaussian quadrature.

C. TPRE Simulations

In this subsection, the same system parameter settings as those in the first paragraph in the VI-B
subsection are assumed for convenience.
In Fig. 10, the TPRE versus GD is plotted, where a fluctuation is obvious in the medium GD region,
while a monotonic decreasing is presented in the high GD region, and this decreasing trend can be

October 12, 2020 DRAFT

 20

10 0

10 -1

10 -2

-3
10

TPE
-4
10

10 -5
µH = 1 × 10−8 , Simulation
µH = 1 × 10−6 , Simulation
10 -6 µH = 1 × 10−5 , Simulation
Gauss-Hermite Quadrature

10 -7
8 10 12 14 16 18
10 10 10 10 10 10
GD
Fig. 8: TPE versus GD in the log-scale of Fig. 7.

0.9

0.8

0.7
TPE

0.6

α = 1 × 10−7 , Simulation
0.5 α = 1.5 × 10−7 , Simulation
α = 2 × 10−7 , Simulation
0.4 α = 2.5 × 10−7 , Simulation
Asymptotic
0.3
0 5 10 15
ΘE ×α2

Fig. 9: TPE versus ΘE for µV = 1 × 10−7 , µH = 1 × 10−8 , σV2 = 1 × 10−12 , σH

2 = 1 × 10−13 , and

λE = 1 × 10−20 .

approximated by a linear function (independent on α, which has been proved in the asymptotic analysis
for TPRE) in the log-scale. A smaller α means a more close eavesdropper around the sender, which
results in the decrease in the TPRE.
There also exists a fluctuation in (λD , λE ) = (10−15 , 10−20 ), (10−20 , 10−20 ) cases in Fig. 11. In
contrast, the TPRE for (λD , λE ) = (10−15 , 10−15 ) remains constant (equal to 1) before a monotonic
decrease. From Fig. 11, we can also observe that the TPRE is an increasing function with respect to λE ,
and a decreasing function with respect to λD , which is easily explained by the joint probability properties
in (36).

October 12, 2020 DRAFT

 21

10 0

-1
10

TPRE
-2
10

α = 1 × 10−7 , Simulation
-3
α = 1 × 10−6 , Simulation
10
α = 1 × 10−5 , Simulation
Gaussian Quadrature
Asymptotic
-4
10
10 8 10 10 10 12 10 14 10 16
GD

Fig. 10: TPRE versus GD for µV = 1 × 10−8 , µH = 5 × 10−8 , σV2 = 1 × 10−12 , and σH
2 = 1 × 10−10 ,

λD = 1 × 10−15 , λE = 1 × 10−20 , and N = 300 for Gaussian quadrature.

0
10

-1
10

-2
10
TPRE

10 -3 (λD , λE ) = (10−15 , 10−20 ), Simulation

(λD , λE ) = (10−20 , 10−20 ), Simulation
(λD , λE ) = (10−15 , 10−15 ), Simulation
10
-4 Asymptotic
Gaussian Quadrature

10 8 10 10 10 12 10 14 10 16 10 18
GD

Fig. 11: TPRE versus GD for µV = µH = 0, σV2 = σH

2 = 1×10−11 (Rayleigh distribution), α = 1×10−6 ,

and N = 110.

VII. C ONCLUSION

In this paper, closed-form expressions for TPLR, TPE, and TPRE were derived based on the Gaussian
quadrature rule, as well as the corresponding asymptotic formulas valid in the high telescope gain at
the legitimate receiver region by using the asymptotic result for the generalized Marcum Q-function
which could be also employed to derive the asymptotic expression for outage probability (showing the
diversity order and array gain) over Beckmann fading channels. The approximate expressions bases on
the Gaussian quadrature rule needs many terms for a high accuracy and the computation becomes more

October 12, 2020 DRAFT

 22

slower with increasing the terms. Alternatively, we can use the asymptotic expressions to get results with
a very high accuracy in the high region of the telescope gain at the legitimate receiver. Moreover, some
closed-form expressions or more concise expressions in some special cases (like Rayleigh, Hoyt and Rice
pointing error model cases) were also derived for much faster computations.

A PPENDIX A: PROOF OF T HEOREM 1

We can rewrite (2) as

 2 
n
2 2
o n
2 2
o θH ΘD − θV2
TPLR = Pr θV + θH ≤ ΘD = Pr θH ≤ ΘD − θV = Pr 2 ≤ 2 , (52)
σH σH
2 /σ 2 follows the non-central Chi-squared distribution with the unit degree, and the corresponding
where θH H

cumulative distribution function (CDF) is given by [41]

√ p 
FθH2 /σH2 (x) = 1 − Q 1 λ, xI {x ≥ 0} . (53)
2

By using the CDF of θH2 /σ 2 and iterative expectation operation, the TPLR can be derived as
H
( s !)
√ ΘD − θV2 
TPLR = EθV 1 − Q 1 λ, 2 I ΘD ≥ θV2
2 σH
Z +∞ ! s !
2 √
(a) 1 (θV − µV ) ΘD − θV2  2

= 1− q exp − 2 Q1 λ, 2 I ΘD ≥ θV dθV , (54)
2πσ 2 −∞ 2σ V
2 σ H
V

where E{·} denotes the expectation operator, and (a) follows the PDF of θV , i.e., PDF of Gaussian
θ√
V −µV
distribution. Let x = in (54), and then, we can get the integral form as (55),
2σV
 v 
+∞
Z u √
2  
√ t Θ D −
1
u 2σV x + µV √ 2
exp −x2 Q 1  λ,

TPLR = 1 − √ 2 I Θ D ≥ 2σ V x + µ V  dx,
π 2 σH
−∞

(55)
which can be approximated by using the Gaussian quadrature rule, shown in Theorem 1.

A PPENDIX B: P ROOF OF P ROPOSITION 1

The Marcum Q-function is defined in the integral form,

Z ∞  M −1  2 
x x + a2
QM (a, b) = x exp − IM −1 (ax) dx, (56)
b a 2
where I· (·) denotes the modified Bessel function of the first kind [42]. To derive the asymptotic result
for the Marcum Q-function, we first express I· (·) into the infinite series form,
∞  ax 2n+M −1
X 1
IM −1 (ax) = . (57)
n!Γ (n + M ) 2
n=0

October 12, 2020 DRAFT

 23

When x → 0, it is easy to see that the leading term (i.e., the lowest order term of x) for approximating
I· (·) in (57) is
1  ax 2n+M −1  1  ax M −1
IM −1 (ax) ≃  = . (58)
n!Γ (n + M ) 2 
n=0 Γ (M ) 2
By using the asymptotic result for I· (·), the asymptotic result for the Marcum Q-function, as b → 0, can
be derived as
Z ∞  2 
 x M −1 x + a2
QM (a, b) = x exp − IM −1 (ax) dx
a 2
|0 {z }
=1
Z b  2 
 x M −1 x + a2
− x exp − IM −1 (ax) dx
0 a 2
 2
b→0 exp − a2 Z b  2
x
2M −1
≃ 1− x exp − dx, (59)
Γ (M ) 2M −1 0 2
where the integral in the last step can be easily derived in a closed-form by using the definition of the
Rb  2  2

lower incomplete Gamma function [42], given by 0 x2M −1 exp − x2 dx = 2M −1 Υ M, b2 , where
Υ(·, ·) denotes the lower incomplete Gamma function. By using the asymptotic result for Υ(·, ·), i.e.,
xs
Υ (s, x) → s as x → 0, we can finally derive the asymptotic result for the Marcum Q-function as (8).

A PPENDIX C: P ROOF OF L EMMA 1

From the derivation in Proposition 1, the CDF of the non-central Chi-squared distribution can be
approximated by
√ p  x→0 exp − λ

FθH2 /σH2 (x) = 1 − Q 1 λ, xI {x ≥ 0} ≃ 2
√ x0.5 I {x ≥ 0} . (60)
2
Γ (1.5) 2
Substituting the asymptotic expression for FθH2 /σH2 (·) into (52) yeilds

( 1 )
exp − λ2 ΘD − θV2 2  2

TPLR ≃ √ E θV 2 I ΘD ≥ θV
Γ (1.5) 2 σH

Z +∞ ! 1
exp − λ2 1 (θV − µV )2 ΘD − θV2 2 
= √ q exp − 2 2 I ΘD ≥ θV2 dθV
Γ (1.5) 2 2πσ 2 −∞ 2σV σH
V

Z √ ΘD ! 1
exp − λ2 1 (θV − µV )2 ΘD − θV2 2
= √ q exp − dθV . (61)
Γ (1.5) 2 2πσ 2 −√ΘD 2σV2 σH2
V

As ΘD → 0, the exponential function in the last step in (61) can be approximated by

!  
(θV − µV )2 θV →0 µ2V
exp − ≃ exp − 2 + o (θV ) , (62)
2σV2 2σV

October 12, 2020 DRAFT

 24

Therefore, the asymptotic result for TPLR can be further written as

√  1
λ

 2  R √ΘD ΘD −θ V
2 2
dθV
exp − 2 µV − ΘD 2
σH
TPLR ≃ √ exp − 2 q , (63)
Γ (1.5) 2 2σV 2πσ 2 V

where the integral can be easily solved in a closed-form,

Z √ ΘD  1
ΘD − θV2 2 π
√ 2 dθV = ΘD . (64)
− ΘD σH 2σH
Finally, the asymptotic expression for TPLR is derived as (10).

A PPENDIX D: P ROOF OF T HEOREM 2

The TPE can be written as

 2 
n
2 2
o θH ΘE − θV2 − αθV
TPE = Pr θV + αθV + θH ≥ ΘE = Pr 2 ≥ 2 , (65)
σH σH
2 /σ 2 follows the non-central Chi-squared distribution with the unit degree and non-centrality
where θH H

parameter λ. By using the CDF of θH 2 /σ 2 , we can get

H
( s !)
√ ΘE − θV2 − αθV 
TPE = EθV Q 1 λ, 2 I ΘE ≥ θV2 + αθV
2 σH
Z +∞ ! s !
1 (θV − µV )2 √ ΘE − θV2 − αθV 
=q exp − 2 Q1 λ, 2 I ΘE ≥ θV2 + αθV dθV .
2πσ 2 −∞ 2σ V
2 σ H
V

(66)
θ√
V −µV
Let x = 2σV
in (66), and then we can get the following integral form,
Z +∞
1

TPE = √ exp −x2 fTPE (x)dx, (67)
π −∞
which can be easily approximated by using the GHQ method, shown in Theorem 2.

A PPENDIX E: P ROOF OF L EMMA 2

By using the asymptotic expression for FθH2 /σH2 (·), the TPE can be written as

( 1 )
exp − λ2 ΘE − θV2 − αθV 2 
TPE ≃ 1 − √ E θV 2 I ΘE ≥ θV2 + αθV , (68)
2Γ (1.5) σH
where
( 1 )
ΘE − θV2 − αθV 2 
E θV 2 I ΘE ≥ θV2 + αθV
σH
Z +∞ ! 1
1 (θV − µV )2 ΘE − θV2 − αθV 2 
=q exp − 2 2 I ΘE ≥ θV2 + αθV dθV . (69)
2πσV2 −∞ 2σV σH

October 12, 2020 DRAFT

 25

To have a possible integral interval, the indicator function shows that we must have

θV2 + αθV − ΘE ≤ 0. (70)

To have real solutions for the above inequality, we have

−α2
α2 + 4ΘE ≥ 0 =⇒ ΘE ≥ . (71)
4
2
If ΘE < − −α
4 , the indicator function in (69) is always equal to zero. Therefore, the TPE is always 1,
2
which is straightforward. In the following, we consider the case of ΘE ≥ − −α
4 . Let
p p
−α + α2 + 4ΘE −α − α2 + 4ΘE
θ1 = , θ2 = .
2 2
The integral in (69) can be derived as (72),
  Zθ1 ! 1
−α2 (θV − µV )2 ΘE − θV2 − αθV 2
I ΘE ≥ exp − dθV
4 2σV2 σH2
θ2
2   ! Zθ1  1
ΘE → −α
4 −α2 (α/2 + µV )2 ΘE − θV2 − αθV 2
≃ I ΘE ≥ exp − dθV
4 2σV2 σH2
θ2
  !

−α2 (α/2 + µV )2 π α2 + 4ΘE
= I ΘE ≥ exp − . (72)
4 2σV2 8σH
Finally, combining (68) and (72) yields (35).

T HEOREM 3
A PPENDIX F: P ROOF OF

Using the iterative expectation method, i.e., E {XY } = EY EX|Y (X |Y ) in (37), we have
  
ΘE − θV2 − αθV ΘD − θV2
TPRE = EθV I 2 ≤ 2
σH σH
    
ΘD − θV2 ΘE − θV2 − αθV
× FθH2 /σH2 2 −FθH2 /σH2 2 , (73)
σH σH
where
   
ΘD − θV2 ΘE − θV2 − αθV
FθH2 /σH2 2 − F 2 2
θH /σH 2
σH σH
s ! s !
√ ΘE − θV2 − αθV  √ ΘD − θV2 
= Q1 λ, 2 I ΘE ≥ θV2 + αθV − Q1 λ, 2 I ΘD ≥ θV2 ,
2 σH 2 σH
(74)

October 12, 2020 DRAFT

 26

n  o
ΘE −θV2 −αθV ΘD −θV2
and I σH 2 ≤= I θV ≥ ΘE −Θ
σH2
α
D
. The TPRE is further written as (75),
Z +∞ " s  !
1 √ ΘE − θV2 − αθV 
TPRE = q Q1 λ, 2 I ΘE ≥ θV2 + αθV
2πσV2 −∞
2 σ H
s !# !  
√ 2
ΘD − θV2  2
(θ V − µ V ) ΘE − ΘD
−Q 1 λ, 2 I ΘD ≥ θV exp − I θV ≥ dθV .
2 σH 2σV2 α
(75)
θ√
V −µV
Let x = , and then, the integral form becomes (76),
2σV
Z +∞  
1 2

ΘE − ΘD µV
TPRE = √ exp −x I x ≥ √ − √
π −∞ α 2σV 2σV
" r
√ 1  √  2 √ 
× Q1 λ, ΘE − 2σV x + µV − α 2σV x + µV
2 σH
 !
√ 2 √ 
× I ΘE ≥ 2σV x + µV + α 2σV x + µV
s !#
√ 1 √ 2   √ 2 
−Q 1 λ, ΘD − 2σV x + µV I ΘD ≥ 2σV x + µV dx. (76)
2 σH
By using the GHQ method, an approximate result is shown in Theorem 3.

A PPENDIX G: D ERIVATION OF THE J OINT PDF fX,Y (·, ·)

2 . We have X = θ 2 + Z and Y = 2θ 2 + 2Z + 2αθ . The PDF of Z is
Let Z = θH V V V

1 1
 z 
fZ (z) = √ z − 2 exp − 2 , z ≥ 0. (77)
2πσ 2 2σ
The Jacobian matrix is
   
 ∂X ∂X   
 ∂θV ∂Z
  2θV 1 
 =  = 4θV − 4θV − 2α = −2α. (78)
 ∂Y ∂Y   
 ∂θ ∂Z
  4θV + 2α 2 
V

The joint PDF of X and Y is finally derived as

     2 !
 ∂ (X, Y ) −1 1 y − 2x y − 2x
fX,Y (x, y) = fθV (θV ) fZ (z)   = fθ fZ x −
∂ (θV , Z)  2α V 2α 2α
" # 1
  −2 n
1  x  y − 2x 2 √ √ on o
= exp − x − I −2α x + 2x ≤ y ≤ 2x + 2α x I x ≥ 0 . (79)
4πσ 2 α 2σ 2 2α

October 12, 2020 DRAFT

 27

A PPENDIX H: D ERIVATION OF THE D OUBLE I NTEGRAL IN TPRE

In order to simplify the notations, we define the double integral in (45) as I1 . By using the integral
identity
√
2x+2α
Z x"  2 #− 21  
y − 2x ΘE − 2x
x− dy = απ − 2α arcsin √ , (80)
2α 2α x
ΘE

I1 can be written as
Z ΘD  x   √ 
max {ΘE , −2α x + 2x} − 2x
I1 = exp − 2 απ − 2α arcsin √ dx, (81)
0 2σ 2α x
N
P

ΘD ΘD ′ ΘD
which can be easily calculated by using Gaussian quadrature method, I1 ≈ 2 ωi′ fTPRE
′
2 xi + 2 .
i=0

R EFERENCES

[1] C. E. Shannon, “Communication theory of secrecy systems," Bell Syst. Tech. J., vol. 28, no. 4, pp. 656–715, 1949.
[2] H.-K. Lo, and H. F. Chau, “Unconditional security of quantum key distribution over arbitrarily long distances," Science,
vol. 283, no. 5410, pp. 2050–2056, Mar. 1999.
[3] M. Sasaki, M. Fujiwara, R.-B. Jin, M. Takeoka, T. S. Han, H. Endo, K.-I. Yoshino, T. Ochi, S. Asami, and A. Tajima,
“Quantum photonic network: Concept, basic tools, and future issues," IEEE J. Sel. Topics in Quantum Electron., vol. 21,
no. 3, pp. 6400313, May, 2015.
[4] H.-K. Lo, M. Curty, and K. Tamaki, “Secure quantum key distribution," Nat. Photonics, vol. 8, pp. 595–604, 2014.
[5] W. Tittel, “Quantum key distribution breaking limits," Nat. Photonics, vol. 13, pp. 310–311, 2019.
[6] K. Inoue, “Quantum key distribution technologies," IEEE J. Sel. Topics in Quantum Electron., vol. 12, no. 4, pp. 888–896,
Aug. 2006.
[7] C. Cheng, R. Lu, A. Petzoldt, and T. Takagi, “Securing the internet of things in a quantum world," IEEE Commun. Mag.,
vol. 55, no. 2, pp. 116–120, Feb. 2017.
[8] A. Husagic-Selman, W. Al-Khateeb, and S. Saharudin, “Feasibility of QKD over FSO link," in Proc. International
Conference on Computer and Communication Engineering (ICCCE 2012), July 2012, Kuala Lumpur, Malaysia, pp. 362–368.
[9] P. V. Trinh, A. T. Pham, A. Carrasco-Casado, and M. Toyoshima, “Quantum key distribution over FSO: Current development
and future perspectives," in Proc. Electromagnetics Research Symposium (PIERS-Toyama), Aug. 2018, pp. 1672–1679.
[10] I. S. Ansari, F. Yilmaz, and M.-S. Alouini, “Performance analysis of free-space optical links over Malaga (M) turbulence
channels with pointing errors," IEEE Trans. Wireless Commun., vol. 15, no. 1, pp. 91–102, Jan. 2016.
[11] H. Lei, H. Luo, K.-H. Park, I. S. Ansari, W. Lei, G. Pan and M.-S. Alouini, “On secure mixed RF-FSO systems with TAS
and imperfect CSI," IEEE Trans. Commun., vol. 68, no. 7, pp. 4461–4475, Jul. 2020.
[12] K.-J. Jung, S. S. Nam, M.-S. Alouini, and Y.-C. Ko “Unified statistical channel model of ship (or shore)-to-ship FSO
communications with pointing errors," in Proc. 2019 IEEE Conference on Standards for Communications and Networking
(CSCN), Oct. 2019, pp. 1–4.
[13] K.-J. Jung, S. S. Nam, Y.-C. Ko, and M.-S. Alouini, “BER Performance of FSO Links over Unified Channel Model for
Pointing Error Models," in Proc. 2018 IEEE International Conference on Communications Workshops (ICC Workshops),
May 2018, pp. 1–6.

October 12, 2020 DRAFT

 28

[14] H. Lei, H. Luo, K.-H. Park, Z. Ren, G. Pan, and M.-S. Alouini, “Secrecy outage analysis of mixed RF-FSO systems with
channel imperfection," IEEE Photon. J., vol. 10, no. 3, pp. 7904113, Jun. 2018.
[15] H. Lei, Z. Dai, K.-H. Park, W. Lei, G. Pan, and M.-S. Alouini, “Secrecy outage analysis of mixed RF-FSO downlink
SWIPT systems," IEEE Trans. Commun., vol. 66, no. 12, pp. 6384–6395, Dec. 2018.
[16] H. Lei, Z. Dai, I. S. Ansari, K.-H. Park, G. Pan, and M.-S. Alouini, “On secrecy performance of mixed RF-FSO systems,"
IEEE Photon. J., vol. 9, no. 4, pp. 7904814, Aug. 2017.
[17] I. S. Ansari, F. Yilmaz, and M.-S. Alouini, “Impact of pointing errors on the performance of mixed RF/FSO dual-hop
transmission systems," IEEE Wireless Commun. Lett., vol. 2, no. 3, pp. 351–354, May 2013.
[18] M. A. Esmail, H. Fathallah, and M.-S. Alouini, “Outage probability analysis of FSO links over Foggy channel," IEEE
Photon. J., vol. 9, no. 2, pp. 7902312, Feb. 2017.
[19] F. Xu, B. Qi, and H.-K. Lo, “Experimental demonstration of phase-remapping attack in a practical quantum key distribution
system," New J. Phys., vol. 12, pp. 113026, 2010.
[20] L. Lydersen, C. Wiechers, C. Wittmann, D. Elser, J. Skaar, and V. Makarov. “Hacking commercial quantum cryptography
systems by tailored bright illumination," Nat. Photonics, vol. 4, pp. 686–689, Aug. 2010.
[21] A. Huang, S. Sajeed, P. Chaiwongkhot, M. Soucarros, M. Legre, and V. Makarov, “Testing random-detector-efficiency
countermeasure in a commercial system reveals a breakable unrealistic assumption," IEEE J. Sel. in Quantum Electron.,
vol. 52, no. 11, pp. 8000211, Nov. 2016.
[22] A. Meda, I. P. Degiovanni, A. Tosi, Z. Yuan, G. Brida, and M. Genovese, “Quantifying backflash radiation to prevent
zero-error attacks in quantum key distribution," Light Sci. Appl., vol. 6, no. pp. e16261, 2017.
[23] H. Weier, H. Krauss, M. Rau, M. Furst, S. Nauerth, and H. Weinfurter, “Quantum eavesdropping without interception: an
attack exploiting the dead time of single-photon detectors," New J. Phys., vol. 13, pp. 073024, Jul. 2011.
[24] Y. Shi, J. Z. J. Lim, H. S. Poh, P. K. Tan, P. A. Tan, A. Ling, and C. Kurtsiefer, “Breakdown flash at telecom wavelengths
in InGaAs avalanche photodiodes," Opt. Express, vol. 25, no. 24, pp. 30388–30394, Nov. 2017.
[25] A. Lacaita, S. Cova, A. Spinelli, and F. Zappa, “Photon-assisted avalanche spreading in reach-through photodiodes," Appl.
Phys. Lett., vol. 62, no. 6, pp. 606–608, Feb. 1993.
[26] S. Arnon, “Effects of atmospheric turbulence and building sway on optical wireless-communication systems," Opt. Lett.,
vol. 28, no. 2, pp. 129–131, Jan. 2003.
[27] I. S. Ansari, M.-S. Alouini, and J. Cheng, “Ergodic capacity analysis of free-space optical links with nonzeros boresight
pointing errors," IEEE Trans. Wireless Commun., vol. 14, no. 8, pp. 4248–4264, Aug. 2015.
[28] L. Yang, M.-S. Alouini, and I. S. Ansari, “Asymptotic performance analysis of two-way relaying FSO networks with
nonzero boresight pointing errors over double-generalized gamma fading channels," IEEE Trans Veh. Technol., vol. 67, no.
8, pp. 7800–7805, Aug. 2018.
[29] S. Arnon and N. S. Kopeika, “Laser satellite communication network–vibration effect and possible solutions," Proc. IEEE,
vol. 85, no. 10, pp. 1646–1661, Oct. 1997.
[30] A. Abdi, W. C. Lau, M.-S. Alouini, and M. Kaveh, “A new simple model for land mobile satellite channels: first- and
second-order statistics," IEEE Trans. Wireless Commun., vol. 2, no. 3, pp. 519–528, May 2003.
[31] J. Kupferman, and S. Arnon, “Zero-error attacks on a quantum key distribution FSO system," OSA Continuum, vol. 1, no.
3, pp. 1079–1086, Nov. 2018.
[32] J. P. Pena-Martin, J. M. Romero-Jerez, and F. J. Lopez-Martinez, “Generalized MGF of Beckmann fading with applications
to wireless communications performance analysis," IEEE Trans. Commun., vol. 65, no. 9, pp. 3933–3943, Sep. 2017.

October 12, 2020 DRAFT

 29

[33] B. Zhu, Z. Zeng, J. Cheng, and N. C. Beaulieu, “On the distribution function of the generalized Beckmann random variable
and its applications in communications," IEEE Trans. Commun., vol. 66, no. 5, pp. 2235–2250, May 2018.
[34] H. AlQuwaiee, H.-C. Yang, and M.-S. Alouini, “On the asymptotic capacity of dual-aperture FSO systems with generalzied
pointing error model," IEEE Trans. Wireless Commun., vol. 15, no. 9, pp. 6502–6512, Sep. 2016.
[35] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels. New York, NY, USA: Wiley, 2000.
[36] H. Zhao and M.-S. Alouini, “On the performance of quantum key distribution FSO systems under a generalized pointing
error model," IEEE Commun. Lett., vol. 23, no. 10, pp. 1801–1805, Oct. 2019.
[37] H.-S. Nguyen, T.-S. Nguyen, and M. Voznak, “Successful transmission probability of cognitive device-to-device communi-
cations underlaying cellular networks in the presence of hardware impairments," J. Wireless Com. Network, no. 208 (2017),
Dec. 2017.
[38] J. Gao, “On the successful transmission probability of cooperative cognitive radio ad hoc networks," Ad Hoc Netw., vol.
58, pp. 99–104, Apr. 2017.
[39] Y. Jiang et al., “Analysis and optimization of cache-enabled Fog radio access networks: Successful transmission probability,
fractional offloaded traffic and delay," IEEE Trans. Veh. Technol., vol. 69, no. 5, pp. 5219–5231, May 2020.
[40] S. Venkateshan and P. Swaminathan, Computational Methods in Engineering. Academic Press, 2014.
[41] D. J. Maširević, “On new formulas for the cumulative distribution function of the noncentral Chi-square distribution,"
Mediterr. J. Math., vol. 14, no. 2, pp. 66 (2017), Apr. 2017.
[42] I. S. Gradshteyn, I. M. Ryzhik, Table of Integrals, Series, and Products, 7th edition. Academic Press, 2007.
[43] J. M. Romero-Jerez, and F. J. Lopez-Martinez, “A new framework for the performance analysis of wireless communications
under Hoyt (Nakagami-q) fading," IEEE Trans. Inf. Theory, vol. 63, no. 3, pp. 1693–1702, Mar. 2017.
[44] G. Pan, C. Tang, X. Zhang, T. Li, Y. Weng, and Y. Chen, “Physical-layer security over non-small-scale fading channels,"
IEEE Trans. Veh. Technol., vol. 65, no. 3, pp. 1326–1339, Mar. 2016.
[45] J. M. Holtzman, “A simple, accurate method to calculate spread multiple access error probabilities," IEEE Trans. Commun.,
vol. 40, no. 3, pp. 461–464, Mar. 1992.
[46] H. Zhao, Y. Liu, A. Sultan-Salem, and M.-S. Alouini, “A simple evaluation for the secrecy outage probability over
generalized-K fading channels," IEEE Commun. Lett., vol. 23, no. 9, pp. 1479–1483, Sep. 2019.

October 12, 2020 DRAFT