Professional Documents
Culture Documents
net/publication/322514466
CITATIONS READS
0 194
5 authors, including:
Some of the authors of this publication are also working on these related projects:
SBSeg´17 - XVII Brazilian Symposium on Information and Computer System Security View project
All content following this page was uploaded by Petros S. Bithas on 19 January 2018.
Abstract—In this paper, important statistical properties of the The dGG distribution is a generic and versatile fading
double-generalized Gamma (dGG) distribution, a generic model channel model, which has recently been used for modeling
that has been used for modeling the double-scattering radio prop- the turbulence-induced fading in free-space optical commu-
agation fading conditions, are obtained and investigated. In this
context, useful statistical metrics, such as the probability density nication systems, e.g., [11], [12]. In the context of wireless
function, cumulative distribution function, and the moments, are communications, many efforts have been devoted to analyti-
derived. Moreover, the second-order statistics of this distribution cally describe important statistical metrics of this distribution,
are also analytically studied for the first time. Based on them, which, in general, have led to approximated solutions, e.g.,
exact expressions for the outage probability, the average symbol [13], [14]. However, important statistical characteristics of
error probability, and the capacity of a transmit antenna selection
system operating in a vehicle-to-vehicle (V2V) communication this generic distribution have not been studied yet, despite
environment are obtained. In addition, the level crossing rate its effectiveness to describe a variety of V2V communication
and the average fade duration are studied. Simplified asymptotic conditions. Such an analytical framework, can be used to study
closed-form expressions have been also obtained and used to the performance of emerging and future wireless communi-
investigate the diversity and coding gains. cation systems and this is the subject of the current paper.
Furthermore, an attractive low complexity communication
I. I NTRODUCTION scheme that has recently employed in V2V communication
studies is the transmit antenna selection (TAS), e.g., [15]–
With the rapid growth of the information and communica- [17]. Based on this technique, the number of expensive radio
tion technologies (ICT), vehicles have been equipped with a frequency (RF) chains is considerably reduced, while an
plethora of sensors, onboard computers, and communication important performance improvement is achieved. However, to
units, resulting to what is known as connected vehicles. The the best of the authors’ knowledge, the performance of TAS
application of ICT to the road transport sector is expected in dGG fading channel has not been investigated in the past.
to provide improvements regarding the environmental perfor- In this paper, important statistical characteristics for the
mance, efficiency, safety, and security [1]. In this context, dGG fading distribution are presented. Moreover, the second-
different types of communications have appeared, including order statistics of this generic distribution have also been
vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I). studied for the first time. These novel expressions are applied
An important factor that plays a critical role on these systems’ to the performance analysis of TAS scheme operating in dGG
performance is the behavior of the wireless medium. In partic- fading environment. For this scenario, the performance has
ular, in scenarios where both the source and the destination, been studied in terms of outage probability (OP), symbol
of the transmitted signal, are moving, the so-called double- error probability (SEP), capacity, average output signal-to-
scattering fading model is observed [2]. Widely adopted noise ratio (SNR), level crossing rates (LCR), and average fade
distributions for modeling the double-bouncing scattering are duration (AFD). This paper differs also from previous ones,
the double-Rayleigh, -Nakagami, and -Weibull [3]–[6]. These since it employs the dGG fading distribution for the first time
distributions, provide good fit to experimental data for mobile- in V2V communication channels, exploiting thus its versatility
to-mobile communications, as it has been verified by many to more accurately describe vehicular channel models.
experimental campaigns, e.g., [7]. These are the reasons why The remainder of the paper is organized as follows. In
many authors in the past have employed them for modeling Section II, the marginal and second-order statistics of the
this kind of environments, e.g., [8]–[10]. Nevertheless, all dGG distribution are presented. In Section III, the system
the aforementioned double-bouncing distributions represent and channel models of the TAS scheme are presented along
special cases of the double-generalized Gamma (dGG), which with a stochastic analysis for the received SNR statistics.
by adjusting its three parameters can accurately model a In Section IV, the derived statistics are employed for the
plethora of different propagation conditions. performance evaluation, while in Section V, representative
numerical examples are presented and discussed. In Section B. Second-Order Statistics
VI, some concluding remarks are provided. Let 𝑅(𝑡) be a dGG process with marginal PDF given by
II. D OUBLE -G ENERALIZED G AMMA S TATISTICS (1). Moreover, let 𝑅˙ denote the time derivative of 𝑅, i.e.,
˙
𝑅(𝑡) = 𝑑𝑅(𝑡)/𝑑𝑡, at time 𝑡. Since the dGG random process is
A. Marginal Statistics
a product of two independent GG random processes, the joint
Let 𝑅 = 𝐺1 × 𝐺2 denote a random variable (RV) following ˙
PDF of 𝑅(𝑡) and 𝑅(𝑡) can be expressed as [20]
the dGG distribution, in which 𝐺𝑖 (with 𝑖 ∈ {1, 2}) are
𝑓𝑅𝑅˙ (𝑥, 𝑥)
˙ =
generalized-Gamma (GG) distributed RVs with probability ∫ ∞∫ ∞ ( )
density function (PDF) given by [18, eq. (1)]. Thus, the PDF 1 𝑥 𝑥˙ 𝑦𝑥
˙ (6)
2
𝑓 ˙
𝐺1 𝐺1 , − 2
𝑓𝐺2 𝐺˙ 2 (𝑦, 𝑦)
˙ 𝑑𝑦𝑑𝑦,
˙
of 𝑅 can be written as 0 −∞ 𝑦 𝑦 𝑦 𝑦
𝑚1 +𝑚2
2𝛽 Ω̂ 2 𝛽
( √ ) in which 𝑓𝐺𝑖 𝐺˙ 𝑖 (⋅, ⋅) represents the joint PDF of the GG
𝑓𝑅 (𝑥) = 𝑥 2 (𝑚1 +𝑚2 )−1 𝐾𝑚2 −𝑚1 2 Ω̂𝑥𝛽/2 , random process 𝐺𝑖 (𝑡) and its derivative 𝐺˙ 𝑖 (𝑡), which has been
Γ (𝑚1 ) Γ (𝑚2 )
(1) derived in [18, eq. (24)]. Assuming 𝑚1 = 𝑚2 = 𝑚, Ω1 =
Ω2 = Ω, substituting [18, eq. (24)] in (6), using [19, eq.
in which 𝛽 and 𝑚1 , 𝑚2 are distribution’s shaping parameters,
( 〈 2〉 )𝛽/2 (3.323/2)], and performing some mathematical manipulations,
Ω̂ = 𝑚 1 𝑚2
Ω1 Ω2 , Ω𝑖 = 𝔼 𝐺𝑖 Γ(𝑚𝑖 )/Γ(𝑚𝑖 + 2/𝛽) 𝑚𝑖, 𝔼 ⟨⋅⟩ the following expression for the joint PDF of 𝑅(𝑡) and 𝑅(𝑡) ˙
denotes expectation, Γ(⋅) is the Gamma function [19, eq. is deduced
(8.310/1)], and 𝐾𝑣 (⋅) is the modified Bessel function of the
second kind and order 𝑣 [19, eq. (8.446)]. Based on its 3 shap- 𝛽 2 𝑚2𝑚+0.5 𝑥𝛽(𝑚+0.5)−2
˙ = √
𝑓𝑅𝑅˙ (𝑥, 𝑥)
ing parameters, the dGG distribution can be efficiently utilized 2𝜋𝜔Ω2𝑚+0.5 Γ(𝑚)2
∫ ∞ −1 ( 𝑚 ) [ ]
to provide excellent fit in different V2V channel measurements 𝑦 𝑚𝛽 2 𝑥˙ 2 𝑥2𝛽−2
× √ exp − 𝑦 exp
data, while its mathematical tractability can be exploited for 0 𝑥𝛽 𝑦 −1 + 𝑦 Ω (𝑥𝛽 + 𝑦 2 ) 2𝜔 2 Ω𝑦
analytical investigations. The cumulative distribution function [ ( ) ]
𝑚𝑥𝛽 𝛽 2 𝑥˙ 2
(CDF) of 𝑅 is given by × exp − + 1 𝑦 −1 𝑑𝑦,
Ω 2𝜔 2 𝑥2
Ω̂𝑝1 ( 1−𝑝 )
𝛽 (7)
𝑥𝛽𝑝1 𝒢2,1
1
𝐹𝑅 (𝑥) = 1,3 Ω̂𝑥 𝑝2 ,−𝑝2 ,−𝑝1 , (2)
Γ(𝑚1 )Γ(𝑚2 ) with 𝜔 denoting the frequency in radians per second. The
with 𝑝1 = 𝑚1 +𝑚2
2 , 𝑝2
= 𝑚2 −𝑚1
and
2 , 𝒢 𝑚,𝑛
𝑝,𝑞 [⋅∣⋅]
denoting integral in (7) can be easily evaluated using well-known math-
the Meijer’s G-function [19, eq. (9.301)]. It is noted that ematical software packages such as Mathematica or Matlab.
Meijer G-function is a built-in function in many mathematical 1) Level Crossing Rates: LCR is defined as the number
software packages, e.g., Mathematica, Maple, and thus can be of times per unit duration that a random process crosses
directly evaluated. Assuming integer values for 𝑚2 , (2) can a predefined threshold 𝑅th in the negative direction, and
be simplified to mathematically is given as
( ∫ ∞
𝑚∑2 −1
𝑘1 +𝑚1
1 2Ω̂ 2 𝛽 𝑁𝑅 (𝑅th ) = 𝑥𝑓
˙ 𝑅𝑅˙ (𝑅th , 𝑥)
˙ 𝑑𝑥.
˙ (8)
𝐹𝑅 (𝑥) = 1− 𝑥 2 (𝑘1 +𝑚1 ) 0
Γ(𝑚2 ) 𝑘1 !Γ (𝑚1 )
𝑘1 =0 Substituting (7) in (8) and using [19, eq. (3.310)] yield
) (3)
( ) ∫ ∞
× 𝐾𝑘1 −𝑚1 2Ω̂ 𝑥 1/2 𝛽/2
. 𝑚2𝑚−0.5 𝜔 𝛽(𝑚−0.5) 𝑦 −3/2
𝑁𝑅 (𝑅th ) = √ 𝑅th √
2𝜋Ω2𝑚−0.5 Γ(𝑚)2 0 𝛽
𝑅th + 𝑦2
Furthermore, for 𝑚1 = 𝑛 + 1/2, with 𝑛 ∈ ℕ, employing [19, ( )
( 𝑚 ) 𝑚𝑅th𝛽 ( )
𝛽
eq. (8.468)], and performing some mathematical manipulations × exp − 𝑦 exp − 𝑦 −1 𝑦 2 + 𝑅th 𝑑𝑦.
yield the following convenient expression for the CDF of 𝑅 Ω Ω
[ 𝑓 (𝑚1 ,𝑚2 )
(9)
1 ∑ 𝑘1 +𝑚1 −𝑘2 −1/2
𝐹𝑅 (𝑥) = 1− 𝐷1 Ω̂ 2 Assuming 𝛽 = 2, (9) simplifies to the LCR of the double-
Γ(𝑚2 ) Nakagami distribution [8, eq. (9)].
𝑘1 ,𝑘2 =0
] (4)
𝛽
( √ ) 2) Average Time Below Level: The average time below
× 𝑥 2 (𝑘1 +𝑚1 −𝑘2 −1/2) exp −2 Ω̂𝑥𝛽/2 , level (ATBL) is defined as the average length of the intervals
for which 𝑅(𝑡) is below the threshold 𝑅th divided by the LCR,
∑𝑓 (𝑚1 ,𝑚2 ) ∑𝑚2 −1 ∑∣𝑘1 −𝑚1 ∣−1/2 which mathematical is expressed as
in which 𝑘1 ,𝑘2 =0 = 𝑘1 =0 𝑘2 =0 , 𝐷1 =
√
𝜋/𝑘1 ! (∣𝑘1 −𝑚1 ∣− 12 +𝑘2 )! 𝐹𝑅 (𝑅th )
. Finally, the 𝑛1 th order moment 𝛼(𝑅th ) = , (10)
(∣𝑘1 −𝑚1 ∣− 12 −𝑘2 )!
Γ(𝑚1 )𝑘2 !22𝑘2 𝑁𝑅 (𝑅th )
of 𝑅 is given by
in which 𝐹𝑅 (𝑅th ) and 𝑁𝑅 (𝑅th ) are given by (3) and (9),
𝑛1 ( ) ( )
Ω̂− 𝛽 𝑛1 𝑛1 respectively. It should be noted that most of the previous
𝜇𝑅 (𝑛1 ) = Γ 𝑚1 + Γ 𝑚2 + . derived expressions have never been reported in the open
Γ(𝑚1 )Γ(𝑚2 ) 𝛽 𝛽
(5) technical literature.
[ √ ]𝑞
(−1)𝑞 𝑞! 𝜋
III. TAS S TATISTICS 𝐷2 =
𝑡1 ! ⋅ ⋅ ⋅ 𝑡𝑚2 ! Γ(𝑚1 )
A. System Model 𝑡 1 ! ⋅ ⋅ ⋅ 𝑡𝑚 2 !
×
Considering a low complexity communication system with 𝑣1,1 ! ⋅ ⋅ ⋅ 𝑣1,𝑚1 + 12 ! ⋅ ⋅ ⋅ 𝑣𝑚2 ,1 ! ⋅ ⋅ ⋅ 𝑣𝑚2 ,∣𝑚2 −1−𝑚2 ∣+ 12 !
𝐿 transmit antennas and one receive antenna, operating in a 𝑚∏2 −1 ∣𝑘−𝑚 ∏ 1 ∣− 2 [
1
]𝑣𝑘+1,𝑝+1
(∣𝑘 − 𝑚1 ∣ − 12 + 𝑝)!
V2V communication environment that is modeled by the dGG × ,
fading distribution. The receiver selects the transmit antenna 𝑘=0 𝑝=0
𝑘!𝑝!(∣𝑘 − 𝑚1 ∣ − 12 − 𝑝)!22𝑝
providing the highest received SNR value, via an error-free 1
( )
𝑚
∑ 2 −1 ∣𝑘−𝑚
∑ 1 ∣− 2
and zero-delay feedback link. Assuming identical parameters 𝑘 + 𝑚1 − 𝑝 − 1/2
𝑑2 = 𝑣𝑘+1,𝑝+1 .
for all diversity links, the instantaneous output SNR can be
𝑝=0
2
𝑘=0
expressed as
The corresponding PDF expression is given by 𝑓𝛾out (𝛾) =
𝛾out = max{𝛾1 , 𝛾2 , . . . , 𝛾𝐿 }, (11) 𝐿𝑓𝛾ℓ (𝛾)𝐹𝛾ℓ (𝛾)𝐿−1 . Substituting (15) in this expression and
doing some mathematical manipulations yield the following
𝐸𝑠
in which 𝛾ℓ = 𝑅ℓ2 𝑁 0
is associated with the ℓth transmitting expression for the PDF of 𝛾out
antenna (ℓ ∈ {1, 2, . . . , 𝐿}), 𝐸𝑠 is the transmitted symbol 𝐿−1 ( )
∑ 1 𝐿−1
energy, 𝑁0 is the noise variance, and 𝑅ℓ is a dGG RV with 𝑓𝛾out (𝛾) = 𝐿𝑓𝛾𝑖 (𝛾)
marginal PDF given by (1). The corresponding PDF and CDF 𝑞,𝑡𝑖 ,𝑣𝑖,𝑗
Γ(𝑚2 )𝐿−1 𝑞 (16)
of 𝛾ℓ are, respectively, given by 𝛽𝑑2
( )
× 𝐷2 𝛾ˆ 𝑑2 𝛾 2 exp −2𝑞ˆ 𝛾 1/2 𝛾 𝛽/4 .
𝑚1 +𝑚2
𝛽ˆ
𝛾 2 B. Asymptotic Analysis
𝑓𝛾ℓ (𝛾) =
Γ (𝑚1 ) Γ (𝑚2 ) (12)
𝛽
( √ ) The exact expression for 𝐹𝛾out (𝛾) does not provide a direct
× 𝛾 4 (𝑚1 +𝑚2 )−1 𝐾𝑚2 −𝑚1 2 𝛾ˆ 𝛾 𝛽/4 , physical insight of the system’s performance. In order to
provide a simplified expression, the main concern is to derive
⎡
𝑓 (𝑚1 ,𝑚2 ) an asymptotic closed-form expression for 𝐹𝛾out (𝛾). Therefore,
1 ⎣ ∑
assuming higher values of 𝛾, using 𝛾(𝑎, 𝑧1 ) ≃ 𝑧1𝑎 [21, eq.
𝑘1 +𝑚1 −𝑘2 −1/2
𝐹𝛾ℓ (𝛾) = 1− 𝐷1 𝛾ˆ 2
Γ(𝑚2 ) (13) (06.07.06.0004.01)], a closed-form asymptotic expression for
𝑘1 ,𝑘2 =0
𝛽
( √ )] the CDF of 𝛾out can be obtained as
(𝑘 +𝑚 −𝑘 2 −1/2)
×𝛾 4 1 1
exp −2 𝛾ˆ 𝛾 𝛽/4 , ( 𝛽
)𝐿
Γ(𝑚1 − 𝑚2 )𝛾 2 𝑚2 𝛾ˆ 𝑚2
[ Γ(𝑚 )Γ(𝑚 )𝛾 𝛾 ]− 𝛽2 𝐹𝛾out (𝛾) ≃ . (17)
〈 〉𝐸 Γ(𝑚1 )Γ(𝑚2 )𝑚2
and 𝛾 = 𝔼 𝑅ℓ2 𝑁
1 2
with 𝛾ˆ = Γ(𝑚1 +2/𝛽)Γ(𝑚𝑏21+2/𝛽)𝑏2 𝑠
0
=
𝛾 𝑏1 𝛾 𝑏2 . As far as channel modeling is concerned, capitalizing In the next section, easy-to-compute closed-form expressions
on the generic form of the dGG distribution, the accurate for various performance indicators are obtained. It is worth
modeling of several double-bouncing fading conditions is remarking that these expressions can only be derived using
allowed. More specifically, in (12), 𝑚𝑖 and 𝛽 are related with the mathematical framework provided above.
the severity of the fading, i.e., higher values for 𝑚𝑖 , 𝛽 denote IV. P ERFORMANCE A NALYSIS
lighter fading conditions. For the system under consideration, In this section, using the previously derived results, ana-
the CDF of 𝛾out is lytical expressions for important performance metrics of the
𝐿 quality of service (QoS) are derived, such as the OP, the SEP,
𝐹𝛾out (𝛾) = [𝐹𝛾ℓ (𝛾)] . (14)
the average capacity, the average output SNR, the LCR, and
Using (13) in (14), employing first the binomial identity, then the AFD.
multinomial identity, and after some mathematical procedure, A. Outage Probability
the following convenient closed-form expression for 𝐹𝛾out (𝛾) OP is defined as the probability that the output SNR falls
is attained below a predetermined threshold 𝛾T and is given by 𝑃out =
∑ 𝐿 ( ) 𝐹𝛾out (𝛾T ), with 𝐹𝛾out (𝛾T ) given by (15).
1 𝐿
𝐹𝛾out (𝛾) = 𝐿
𝐷2 𝛾ˆ 𝑑2 Diversity and Coding Gains: At the high SNR regime, the
𝑞,𝑡𝑖 ,𝑣𝑖,𝑗
Γ(𝑚 2 ) 𝑞 (15) OP can be characterised by two parameters: diversity
( gain,
)−𝐺
( ) ∞ 𝑑
𝛽𝑑2
× 𝛾 2 exp −2𝑞ˆ 1/2
𝛾 𝛾 𝛽/4
, 𝐺𝑑 , and coding gain, 𝐺𝑐 , i.e., 𝑃out ≃ 𝐺𝑐 𝛾 𝑏𝑖 .
Thus, based on (17), it can be concluded 1 that 𝐺𝐶 =
[ 𝛽
]− 𝑚 𝛽
𝛽 2
in which Γ(𝑚1 −𝑚2 )[Γ(𝑚1 +2/𝛽)Γ(𝑚2 +2/𝛽)]𝑚2 2 𝛾𝑇2
𝑚2
𝛽 and 𝐺𝑑 =
𝐿 𝐿 𝑞 𝑡𝑚 2 [Γ(𝑚1 )Γ(𝑚2 )]𝑚2 2 +1 𝑚2
∑ ∑ ∑ ∑
= ⋅⋅⋅ , 𝑚2 𝛽𝐿. Therefore, the diversity gain depends on the fading
𝑞,𝑡𝑖 ,𝑣𝑖,𝑗 𝑞=0 𝑡1 ,⋅⋅⋅ ,𝑡𝑚2 =0 𝑣1,1 ,⋅⋅⋅ ,𝑣𝑚 1 =0 severity, modeled by the shaping parameters 𝑚2 , 𝛽, as well as
2 ,∣𝑚2 −1−𝑚1 ∣+ 2
𝑡1 +...+𝑡𝑚2 =𝑞 𝑣1,1 +⋅⋅⋅+𝑣𝑚 ,∣𝑚 −1−𝑚 ∣+ 1 =𝑡𝑚2 the number of transmit antennas 𝐿.
2 2 1 2
B. Symbol Error Probability D. Average Output SNR
The SEP is one of the most important performance mea- The average output SNR is a performance metric providing
sures, the minimization of which is the main objective in excellent indication of the overall system’s fidelity. For the
designing wireless communication systems. For evaluating the scheme under consideration, it is defined as
SEP, the CDF-based approach is employed as follows [22]
∫ ∞ ∫ ∞
𝑃¯𝑒 =
′
(−𝑃𝑒 )𝐹𝛾out (𝛾)𝑑𝛾, (18) 𝔼 ⟨𝛾out ⟩ = [1 − 𝐹𝛾out (𝛾)] 𝑑𝛾. (24)
0 0
′
in which (−𝑃𝑒 ) denotes the negative derivative of the condi- Substituting (15) in (24), and using [19, eq. (3.351/3)], we
tional error probability. In particular, for binary phase shift arrive at the following expression
keying (BPSK)
√ or binary frequency shift keying (BFSK)
( )
𝑃𝑒 = 𝛼𝑄( 𝑏𝛾), with 𝑄(⋅) being the area under the tail of 𝐿 ( ) 22− 𝛽4 −2𝑑2 Γ 4 + 2𝑑
∑ (−1)−1 𝐿 𝛽 2
the Gaussian PDF [22, eq. (2-1-97)], while for differentially 𝔼 ⟨𝛾out ⟩ = 𝐷2 .
BPSK (DBPSK) 𝑃𝑒 = 𝛼 exp(−𝑏𝛾). For the BPSK case, the 𝑞,𝑡 ,𝑣
Γ(𝑚2 )𝐿 𝑞 𝛽ˆ
2
𝛾𝛽𝑞 𝛽
2(2+𝛽𝑑2 )
𝑖 𝑖,𝑗
following closed-form expression for SEP has been obtained (25)
𝐿 ( ) √ ( )
∑ 1 𝐿 𝛼 𝑏 𝑏 1
𝑃¯𝑒 = 𝐷 2 𝛾
ˆ 𝑑 2
√ 𝐷 3 , , (19) E. Level Crossing Rate and Average Fade Duration
𝑞,𝑡 ,𝑣
Γ(𝑚2 )𝐿 𝑞 8𝜋 2 2
𝑖 𝑖,𝑗
(23)
𝐹𝑅 (𝑅th )𝐿 𝐹𝑅 (𝑅th )
Due to space constraints, the proof has been omitted here. 𝑎𝑇 (𝑅th ) = = . (28)
𝐿𝑁𝑅 (𝑅th )𝐹𝑅 (𝑅th )𝐿−1 𝐿𝑁𝑅 (𝑅th )
Fig. 1. OP vs 𝛾𝑇 for different values of 𝛽. Fig. 3. Normalized channel capacity vs 𝐿 for different values of 𝛾 𝑏𝑖 .