IET Communications

Research Article

Comparison of adaptive algorithms for free ISSN 1751-8628

Received on 11th July 2018
Revised 28th February 2019
space optical transmission in Málaga Accepted on 25th March 2019
E-First on 17th May 2019
atmospheric turbulence channel with pointing doi: 10.1049/iet-com.2018.5666
www.ietdl.org

errors
Marko Smilić1 , Zorica Nikolić2, Dejan Milić2, Petar Spalević3, Stefan Panić4
1University of Priština, Faculty of Natural Sciences and Mathematics, Lole Ribara 29, Kosovska Mitrovica, Serbia
2University of Niš, Faculty of Electronic Engineering, Aleksandra Medvedeva 14, Niš, Serbia
3University of Priština, Faculty of Technical Sciences, Kneza Miloša 7, Kosovska Mitrovica, Serbia
4National Research Tomsk Polytechnic University, Sovetskaya 84/3, Tomsk, Russia

E-mail: marko.smilic@pr.ac.rs

Abstract: In this study, the authors investigate average capacity of free space optics communication over Málaga atmospheric
turbulence channel with pointing errors and path loss, for intensity modulated/direct detection (IM/DD) and heterodyne
detection. Various algorithms which use adaptive transmission with both types of detection are considered, such as: optimal rate
adaption (ORA), optimal power and rate adaption (OPRA), channel inversion with fixed rate (CIFR) and truncated channel
inversion with fixed rate (TIFR). Analytical closed-form expressions for channel capacities of ORA, OPRA and TIFR adaptive
transmission are presented, and the authors prove that CIFR transmission is not feasible in the strict sense for the conditions
considered. Obtained analytical results are numerically evaluated and graphically presented for different strengths of
atmospheric turbulence (in weak, moderate and strong turbulence regime) for both types of detection (IM/DD and heterodyne),
and for considered algorithms of adaptive transmission (ORA, OPRA and TIFR). The authors have developed expressions
suitable for approximating high signal-to-noise ratio channel capacity, and they graphically present and compare the asymptotic
approximations with the obtained analytical results for different strengths of turbulence for both types of detection. Also,
obtained analytical results were confirmed by Monte-Carlo simulations, and graphically compared for different strengths of
turbulence regimes.

1 Introduction Unlike RF communication, FSO requires a direct line of sight

between the transmitter and the receiver. Atmospheric effects, such
The study of wireless optical systems is generally a as rain, snow and fog, will affect the FSO link performance by
multidisciplinary undertaking involving a wide range of areas means of scattering and absorption of light, effectively attenuating
including: optical design, optoelectronics, channel modelling, the received signal. During transmission, the transmitted FSO
communications and information theory, modulation and signal is also exposed to various effects such as atmospheric
equalisation, wireless optical network architectures among many turbulence and misalignment between the transmitter and receiver
others [1]. The transport capabilities of optical communication (pointing error). The impact of these effects on free space optical
systems have increased tremendously in the past two decades, system performance are explained in [6–8]. One of the more
primarily due to advances in optical devices and technologies, and effective ways to reduce the impact of these detrimental effects is
have enabled the Internet as we know it today with all its impacts the use of adaptive transmission. Adaptive transmission is based on
on the modern society [2]. the receiver's estimation of the channel and feedback of the channel
Free space optics (FSO) links are considered as a viable state information to the transmitter. The transmitter then adapts the
solution for various applications because of the following transmit power level, symbol/bit rate, constellation size, coding
properties [2–4]: rate/algorithms or any combination of these parameters in response
to the changing channel conditions [9].
• the high-directivity of the optical beam provides high power Málaga model represents a general model of atmospheric
efficiency and spatial isolation from other potential interferers, turbulence [10], covering other less general models such as K
which is not inherent in RF/microwave communications, turbulence model, HK turbulence model, gamma and gamma-
• FSO transmission is unlicensed, gamma turbulence models, exponential-Weibull turbulence models
• the large fractional-bandwidth coupled with high optical gain and so on. System performance for outage probability (Pout),
using moderate powers permits very high data rate transmission, symbol error rate, bit error rate (BER) over Málaga atmospheric
• FSO links are relatively easy to install and easily accessible for turbulence channel are given in [11–13]. Also, BER for different
repositioning when necessary. modulation schemes over Málaga atmospheric turbulence model is
presented in [14]. Diversity is a widely used and efficient technique
Channel capacity is a fundamental measure of the maximum for improving performance of communication systems. System
amount of information which can be conveyed through a channel performance for outage probability and BER when using different
reliably. Shannon's discovery that it was possible to have arbitrarily diversity techniques are given in [15–18].
reliable communication at non-zero rates revolutionised Channel capacity represents one of the most important system
communication system design practices and established the areas performance measures. Ergodic capacity for lognormal, Rician-
of information theory and error control coding [1, 5]. The capacity lognormal, Málaga and gamma-gamma atmospheric turbulence
depends on the specific channel model and on the given input channel are given in [13, 19, 20]. Average capacity of optical
constraints [2]. wireless communication systems over I-K atmospheric turbulence
channels without pointing error is given in [21]. Channel capacity

IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585 1578

© The Institution of Engineering and Technology 2019
 k k
β − 1 γβ + Ω′ 1 − 2 Ω′ k−1
α 2
ak ≜ (3)
k − 1 (k − 1)! γ β

There is also the case of Málaga atmospheric turbulence model

with real β [12], but it will not be considered here. Parameters α
and β represent the effective numbers of large-scale and small-
scale cells, respectively, and can be related to the atmospheric
conditions. The parameters are expressed as
−1
Fig. 1  Model of FSO communication system with adaptive transmission 0.49σR2
α = exp 7/6 −1 (4)
for different atmospheric turbulence strengths is investigated in 1 + 1.11σR12/5
[22]. Capacity analysis of atmospheric turbulence channel with −1
pointing errors is presented in [23]. Algorithms for adaptive 0.51σR2
transmission over gamma-gamma atmospheric turbulence channel β = exp 12/5 5/6
−1 (5)
1 + 0.69σR
are explained in [24–26].
It is worthy to note that intensity modulation/direct detection
(IM/DD) is the main mode of detection in FSO systems, but where plane wave propagation and zero inner scale is assumed
coherent detection has also been proposed as an alternative [28]. γ = 2b0(1 − ρ) denotes the average power of the scattering
detection mode. Among the two detection types, heterodyne component received by off-axis eddies, 2b0 is the average power of
coherent detection represents a method that is more complicated, the total scatter components, parameter 0 ≤ ρ ≤ 1 represents the
but has the ability to better overcome the receiver thermal noise amount of scattering power coupled to the line-of-sight (LOS)
effects [13, 27], and therefore achieve better performance. component, Ω′ = Ω + 2b0 ρ + 2 2b0 ρΩcos(θA − θB) symbolises the
Motivated by this short review of existing results, we proceed to average power through the coherent advantages, Ω is the regular
investigate average capacity of FSO communication over Málaga power of the LOS aspect, θA − θB are the deterministic levels of
atmospheric turbulence channel with pointing error and path loss the LOS and also the coupled-to-LOS spread terms, respectively
for both proposed types of detection, IM/DD and heterodyne. Also, [11]. σR2 represents the Rytov variance and is used as a metric of
various algorithms for adaptive transmission, such as: constant turbulence strength. It is expressed by
power rate adaption (ORA), optimal power rate adaption (OPRA),
channel inversion with fixed rate (CIFR) and truncated channel σR2 = 1.23Cn2 k7/6L11/6 (6)
inversion with fixed rate (TIFR) for both types of detection will be
considered. Analytical closed-form expressions for ORA, OPRA
where k = 2π/λ is the wave number, λ is the wavelength, L is the
and TIFR adaptive transmission algorithms will be derived.
Obtained results are confirmed by Monte-Carlo simulations, and propagation distance and Cn2 is the refractive index structure
are further discussed for different turbulence strengths and types of parameter, which typically varies from 10−17m−2/3 to 10−13m−2/3 as
detection. Also, we derive asymptotic high signal-to-noise ratio turbulence strength varies from weak to strong [29].
(SNR) approximations for channel capacity expressions and
compare them graphically with analytical results. 2.2 Path loss
Path loss, Il, can be described by the exponential Beers-Lambert
2 System model laws as
We consider an FSO system with constant transmitted power than
can be adjusted according to adaptive transmission requirements. Il(L) = e− σL (7)
During transmission, the transmitted signal is exposed to various
effects such as: atmospheric turbulence, scintillation, misalignment where L denotes the propagation distance and σ is the attenuation
between the transmitter T x and receiver Rx , i.e. pointing error, coefficient.
and path loss, as shown in Fig. 1. All of these effects affect the
system performance. To maintain a target data rate, the received 2.3 Pointing error
SNR should not be less than a predetermined cut-off level denoted
by g0. When the current SNR at the receiver of the FSO link, Pointing errors also affect the transmission performance. In case of
denoted by g, falls below the cut-off level g0, the FSO link cannot zero boresight pointing error, I p, the following model is used [1]:
support the high data rate transmission. In this case, the receiver
sends a feedback signal to the transmitter so that the transmitter ξ2 2
f I p(I p) = I ξp −1
, 0 ≤ I p ≤ A0 (8)
increases its power and thus maintains the reliability of the link. ξ2
A0
2.1 Atmospheric turbulence model where ξ = ωLeq /2σs is the ratio between the equivalent beam radius
Málaga model represents a general model of atmospheric at the receiver ωLeq and the pointing error displacement standard
turbulence [10]. In this paper, we consider the case of Málaga deviation at the receiver σs. A0 = erf(v) 2 is the fraction of the
atmospheric turbulence model with integer β [11], which can be
collected power where v = πa/ 2ωLeq with erf( ⋅ ) denoting the
expressed as
error function, where as the square of the equivalent beam width is
β given by
α+k αβIa
f a(Ia) = A ∑ akIa 2
−1
Kα − k 2 (1)
k=1
γβ + Ω′ π erf(v)
ωL2 eq = ωL2 2 (9)
2ve− v
where
α
β+
α 3 Probability density function (PDF)
2α 2 γβ 2
A≜ (2) Combining the atmospheric turbulence model, pointing error and
γ 1 + 2 Γ(α) γβ + Ω′
α

path loss, we get PDF of instantaneous SNR at the receiver for

IM/DD and heterodyne detection. Closed-form expression for PDF

IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585 1579

© The Institution of Engineering and Technology 2019
for Málaga atmospheric turbulence model with pointing error and β
∫
⟨C⟩ora ∞ 1, 1
1
path loss is given in [13] for integer β = r ∑ bk g−1G2,1, 22 g
B 2 k=1 0 1, 0
(16)
ξA 2 β
γβ + Ω′
α+k
3, 0
1 ξ2 + 1
f g g = r ∑ ak ×G ηg dg
2 r
1, 3
2 gk=1 αβ ξ2, α, k
(10)
1 ξ2 + 1 For IM/DD, the integral from previous expression is solved by
× G1,3, 30 η g r
ξ2, α, k using equation (2.24.1/1) from [32], while for heterodyne
detection, the integral from expression (16) is solved by using
where equation (07.34.21.0013.01) from [33]. By applying these solutions
we obtain the closed-form expressions for ORA adaptive
αβκ γ + Ω′ transmission algorithm
η= 1 (11)
cμ r γβ + Ω′ β
⟨C⟩ora 1 0, 1, ξ2 + 1
B
= ∑
2k=1
bkG3,5, 51 η 2 (17)
Parameter r determines the type of detection technique (r = 1 for Het ξ , α, k, 0, 0
heterodyne detection and r = 2 for IM/DD) and c denotes a
constant term such that c = 1 for heterodyne detection and ⟨C⟩ora 2α
β
η2 Ξora
c = e/2π for IM/DD [13]. The received instantaneous SNR is B
= ∑
16π k = 1
2kbkG4,8, 81
16 Θora
(18)
DD
defined as
ξ2 ξ2 + 1 α α + 1 k k + 1
2Pt RI 2
where Θora = { 2 , 2
, 2 , 2 , 2 , 2 , 0, 0}, and
g= (12)
2σn2 Ξora = {0, 1,
ξ2 + 1 ξ2 + 2
, 2 }, and
2
2
where σn denotes additive white Gaussian noise, and I represents α+k
the irradiance I = IaIlI p. Average electrical SNR can be defined as ξ2 A γβ + Ω′ 2
(19)
bk = a
ln 2 k αβ
2Pt R 2 2
μ= E[I ] (13) Equations (17) and (18) can be approximated for high SNR
2σn2 values by using equation (46) as explained in Appendix 3. Results
of approximation of these expressions are presented in Figs. 2–7.
where E[ . ] is the statistical expectation. Average electrical SNR
from equation (13) can be determined as [30] 4.2 Optimal power and rate adaption
2
2Pt R 2 2 2 2
In the OPRA algorithm, the power level and rate parameters vary
μ= A0 Il κ γ + Ω′ (14) in response to the changing channel conditions. With this adaptive
2σn2
transmission policy, more power and higher data rates are allocated
when the channel condition is good, and the transmission is
4 Channel capacity terminated when the received SNR falls below a cut-off level g0.
Channel capacity is one of the most important concerns in the OPRA is not suitable for all applications because there are some
design of wireless systems, as it determines the maximum applications that require fixed rate. Channel capacity for OPRA
attainable throughput of the system. It can be defined as the adaptive algorithm is given as [31]
average transmitted data rate per unit bandwidth, for a specified
∫
∞
average transmit power, and specified level of received outage or g
⟨C⟩opra = B log2 f g dg (20)
BER [31]. g0 g0 g
Adaptive transmission techniques represent a viable solution for
reducing the influence of atmospheric turbulence, misalignment Cut-off level must satisfy condition
and path loss to propagation of transmitted signal. This is achieved
by adapting basic parameters of the transmitted signal. The main
∫
∞
1 1
advantage of the adaptive transmission is that it provides − f g dg = 1 (21)
g0 g0 g g
opportunity to attain higher spectral efficiency within the channel
bandwidth.
In this section, capacity for various adaptive transmission
algorithms (ORA, OPRA, CIFR and TIFR) will be derived.
Analytical expressions in closed-form will be given for proposed
algorithms with IM/DD and heterodyne detection.

4.1 Optimal rate adaption

With ORA algorithm, the transmitter adapts its rate only, while
maintaining a fixed power level. Thus, this algorithm can be
implemented at reduced complexity. Channel capacity ⟨C⟩ora
[bits/s] with constant transmission power policy is given by [31]

∫
∞
⟨C⟩ora = B log2 1 + g f g g dg (15)
0

where B denotes bandwidth required for transmission in [Hz].

1, 1
By rewriting ln 1 + g as a Meijer's G-function G2,1, 22 g
1, 0 Fig. 2  Capacity for adaptive transmission algorithms over weak
using equation (8.4.6/5) from [32], and substituting the identity and turbulence for IM/DD
(10) into (15), we get the following expression:
1580 IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585
© The Institution of Engineering and Technology 2019
Fig. 3  Capacity for adaptive transmission algorithms over weak Fig. 6  Capacity and simulation results for adaptive transmission
turbulence for heterodyne detection algorithms over strong turbulence for IM/DD

Fig. 4  Capacity and capacity approximation for adaptive transmission Fig. 7  Capacity and simulation results for adaptive transmission
algorithms over moderate turbulence for IM/DD algorithms over strong turbulence for heterodyne detection

substituting into (22), we obtain ⟨C⟩opra for both types of detection,

heterodyne and IM/DD
β
⟨C⟩opra 1, ξ2 + 1, 1
B
= − ∑ bk G 4, 1
3, 5 ηg0
Het k=1 0, ξ2, α, k, 0
1, ξ2 + 1
+ln g0 G2,3, 41 ηg0 (23)
ξ2, α, k, 0
ξ2 + 1, 1
+ln g0 G2,4, 40 ηg0
0, ξ2, α, k

β
⟨C⟩opra 1, ξ2 + 1, 1
B
= − ∑ bk G3,4, 51 ηg0
DD k=1 0, ξ2, α, k, 0
2
1, ξ + 1
+ln g0 G2,3, 41 ηg0 (24)
ξ2, α, k, 0
Fig. 5  Capacity and capacity approximation for adaptive transmission ln g0 α + k 7, 0 η2 Ξopra
+ 2 G3, 7 g ,
algorithms over moderate turbulence for heterodyne detection 16π 16 0 Θopra

Expression (20) can be represented as where

⟨C⟩opra 1 ξ2 ξ2 + 1 α α + 1 k k + 1
= (I − I ) (22) Θopra = , , , , , ,0 and Ξopra =
B ln 2 1 2 2 2 2 2 2 2
2 2
ξ +1 ξ +2
∞ ∞ , ,1 .
where I1 = ∫g0 ln g f g g dg and I2 = ∫g0 ln g0 f g g dg. After 2 2
solving integrals I1 and I2, as derived in Appendix 1, and

IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585 1581

© The Institution of Engineering and Technology 2019
Table 1 Turbulence conditions and parameters where
Parameter Weak Moderate Strong
β
Cn2 [m−2/3] 7.97 ⋅ 10−15 3.33 ⋅ 10−14 9.2 ⋅ 10−13 2αln 2 η2 ΞItifr
I(DD) =
16π ∑ 2kbkG
7, 0
3, 7 g
16 0 ΘItifr
(31)
σR2 0.35 1.47 40.5 k=1

α 7.52 4.06 9.765

where we have denoted
β 6 2 1
ξ2 ξ2 + 1 α α + 1 k k + 1
ΘItifr = 1, , , , , , and ΞItifr =
2 2 2 2 2 2
Equations (23) and (24) can be approximated for high SNR 2 2
ξ +1 ξ +2
values using Appendix 3. Results of approximation of these , ,2 .
expressions are presented in Figs. 2–7. 2 2

We also have
4.3 Channel inversion with fixed rate
β
Under this adaptive transmission algorithm, the transmitter adapts 2αln 2 η2 ΞPtifr
the transmit power according to the channel effects state in order to P(DD) =
16π ∑ 2kbkG
6, 1
3, 7 g
16 0 ΘPtifr
(32)
k=1
maintain a constant SNR at the receiver, i.e. inverts the channel
effects while maintaining a constant transmission rate. CIFR
algorithm generally achieves what is known as the outage capacity where
of the system; that is the maximum constant data rate that can be
supported for all channel conditions with some probability of ξ2 ξ2 + 1 α α + 1 k k + 1
ΘPtifr = , , , , , ,0 and ΞPtifr =
outage [24, 31] 2 2 2 2 2 2
2 2
ξ +1 ξ +2
1, , .
1 2 2
⟨C⟩cifr = Blog2 1 + ∞ (25)
∫0 g−1 f g g dg Equations (27) and (30) can also be approximated for high SNR
values by using the procedure outlined in Appendix 3. Results of
After detailed inspection and calculation it was determined that approximation of these expressions are presented in Figs. 2–7.
the integral from expression (25) diverges and CIFR → 0.
Regardless of the strength of atmospheric turbulence, the CIFR
5 Numerical results and discussion
adaptive transmission algorithm cannot be applied when we have a
Málaga atmospheric turbulence model with pointing error. This In order to discuss the use of adaptation techniques, numerically
result was confirmed by the Monte-Carlo simulation (shown in obtained results are also presented graphically. Results of
Figs. 2–7). Detailed explanation is given in Appendix 2. approximate high-SNR expressions for ORA, OPRA and TIFR
adaptive transmission algorithms in case of moderate turbulence
4.4 Truncated channel inversion with fixed rate scenario for IM/DD and heterodyne detection are also graphically
presented. Furthermore, the obtained results are confirmed by
Since the CIFR algorithm may exhibit a large channel capacity Monte-Carlo simulations. The simulation results are presented
penalty, or is not feasible at all, a modified channel inversion graphically in the case of strong turbulence scenario for IM/DD
algorithm is proposed where only the transmitted power is adapted and heterodyne detection.
according to the channel state provided that the received SNR is In this section, we compare average capacity for FSO
above a certain cut-off SNR g0 [24], resulting in communication over Málaga atmospheric turbulence channel with
pointing error and path loss for IM/DD and heterodyne detection
1 by using ORA, OPRA, CIFR and TIFR algorithms for adaptive
⟨C⟩tifr = Blog2 1 + ∞
1 − Pout (26) transmission. For the analysed adaptive transmission algorithms,
∫g0 g−1 f g g dg we compare the average capacity for different strengths of
atmospheric turbulence. Following parameters are assumed for
where Pout represents outage probability, defined as: FSO link: link length L = 1 km, wavelength λ = 785 nm,
g0 Ω = 1.3265, b0 = 0.1079, ρ = 0.596. Parameters used for weak,
Pout = ∫0 f g g dg. Integrals and Pout from equation (26) can be
solved by using equations (07.34.21.0085.01) and moderate and strong turbulence regimes are listed in Table 1.
(07.34.21.0084.01) from [33]. Numerical results represent ensemble averages obtained by
For heterodyne detection we obtain evaluating derived analytical forms, while the simulation uses
direct averaging of N independent channel realizations. Simulation
⟨C⟩tifr 2g0 procedure uses inverse transform sampling to generate the
= log2 1 + 1 − P(Het) (27) realisations of Málaga channel with pointing errors corresponding
B Het I(Het)
to (10).
Fig. 2 represents capacity for adaptive transmission algorithms
where over weak turbulence for IM/DD. Fig. 3 represents capacity for
adaptive transmission algorithm over weak turbulence for
β
ξ2 + 1, 2 heterodyne detection. Both figures show that there is a higher
I(Het) = ln(2) ∑ bkG2,4, 40 ηg0 (28) capacity available when using ORA and OPRA algorithms.
k=1 1, ξ2, α, k
Capacity of channel inversion is less than that of ORA and OPRA
β algorithms. Same applications may require constant rate
ln 2 1, ξ2 + 1
P(Het) =
2 ∑ bkG 3, 1
2, 4 ηg0 (29) transmission and reduced complexity at a cost of decreased
k=1 ξ2, α, k, 0 capacity. TIFR algorithm has less capacity than ORA and OPRA.
From the presented figures it can also be seen that higher channel
For IM/DD we obtain capacity is achieved for heterodyne detection. For example, at SNR
of 40 dB, the ORA capacity is 11.4 bits/s/Hz for IM/DD. For same
⟨C⟩tifr g0 value of SNR and heterodyne detection, ORA capacity is 12.9 
= log2 1 + 1 − P(DD) (30) bits/s/Hz.
B DD I(DD)
Fig. 4 represents capacities and their high-SNR approximations
for adaptive transmission algorithms over moderate turbulence for

1582 IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585

© The Institution of Engineering and Technology 2019
IM/DD. Fig. 5 represents capacities and their approximate values graphically presented for the case of moderate turbulence scenario
for adaptive transmission algorithms with heterodyne detection and both detection types. Also, the obtained results are confirmed
over moderate turbulence. For both types of detection, shown in by Monte-Carlo simulations. The simulation results are graphically
Figs. 4 and 5, capacity decreases with increased turbulence presented for the case of strong turbulence scenario, for IM/DD
strength. Also, higher capacity is achieved for heterodyne and heterodyne detection.
detection. Increasing turbulence strength has more influence on We have shown that heterodyne detection is better than IM/DD
TIFR algorithm than on ORA and OPRA algorithms. Also, Figs. 4 and provides greater reliability and capacity for all proposed
and 5 show that the results of the approximations for higher SNR turbulence scenarios. Although heterodyne detection has better
values are in good agreement with the obtained analytical results. performance, IM/DD is more often used in commercial systems
Fig. 6 represents capacities and corresponding Monte-Carlo due to its simple design.
simulations results for adaptive transmission algorithm over strong ORA and OPRA adaptive transmission algorithms allow higher
turbulence for IM/DD. The analogous results for heterodyne channel capacities when compared to TIFR algorithm. The
detection are shown in Fig. 7. Strong turbulence affects both types advantage of ORA and OPRA algorithms is in their flexibility that
of detection for all considered adaptive transmission algorithms, allows control of data rate by adjusting the power of the
resulting in lower channel capacities. In all presented figures we transmitter. By increasing or decreasing the transmitted power, we
observe that the performance gap between different adaptive have an additional degree of freedom to increase or decrease the
transmission algorithms decreases with decreasing turbulence data rate. On the other hand, TIFR algorithm is applied in systems
strength. For example, we consider TIFR algorithm with that require constant data transfer rate so it is not possible to
heterodyne detection for SNR of 40 dB, resulting in capacity of perform rate adaptation as in ORA and OPRA algorithms. The
12.54 bits/s/Hz for weak turbulence, 11.19 bits/s/Hz for moderate main advantage of fixed-rate algorithms is their simple
turbulence and 10.048 bits/s/Hz for strong turbulence. Similarly, if implementation, regardless of the fact that they exhibit lower
we consider strong turbulence and IM/DD, we can achieve values of channel capacity. Also, we note that turbulence strength
capacity of 4 bits/s/Hz for 18.7 dB of SNR when using ORA, 20 dB has more significant effect on TIFR algorithm than it has on ORA
when using OPRA, and 27.5 dB when using TIFR. When using and OPRA.
heterodyne detection in the same turbulence conditions, the same
capacity level is achieved for 13.7, 14 and 17.5 dB when 7 Acknowledgments
employing ORA, OPRA and TIFR algorithms, respectively.
We have encountered a difficulty in simulation procedure when This paper is supported by Ministry of Education, Science and
trying to confirm zero-capacity of CIFR transmission with Technological Development of the Republic of Serbia, under
heterodyne detection. Let N denote the number of independent projects TR-32051 and III-44006. Last author is funded by the
simulation runs. By increasing N one should expect to obtain the framework of Competitiveness Enhancement Program of the
results that are progressively in closer agreement with analytical National Research Tomsk Polytechnic University, Russia.
results. However, there are cases where the required number of
simulation runs is higher than practically feasible. One such 8 References
example is the CIFR adaptive transmission for the case of
[1] Hranilovic, S.: ‘Wireless optical communication systems’ (Springer, New
heterodyne detection considered in this paper (see Fig. 7). By York, 2005)
examining closely the cumulative distribution function of the [2] Arnon, S., Barry, J.R., Karagiannidis, G.K., et al.: ‘Advanced optical wireless
random variable involved, we notice that for average SNR of 50 dB communication systems’ (Cambridge University Press, Cambridge UK, 2012)
[3] Andrews, L.C., Phillips, R.L.: ‘Laser beam propagation through random
and number of simulation runs of N = 106, we can rarely expect media’ (SPIE, Bellingham WA, USA, 2005)
any realisations of random variable that are lower than −10 dB. [4] Willebrand, H., Ghuman, B.S.: ‘Free space optics: enabling optical
Cumulative probability that the SNR falls below this threshold connectivity in today's networks’ (SAMS, Indianopolis USA, 2002)
level is about 10−6, i.e. approximately 1/N . On the other hand, the [5] Lapidoth, A., Moser, S.M., Wigger, M.A.: ‘On the capacity of free-space
optical intensity channels’, IEEE Trans. Inf. Theory, 2009, 55, (10), pp. 4449–
obtained simulation dataset is equivalent of CIFR adaptation 4461
excluding realisations below the −10 dB threshold, since we [6] Al-Habash, M.A., Andrews, L.C., Phillips, R.L.: ‘Mathematical model for the
almost never get any. Even if we increase the number of simulation irradiance probability density function of a laser beam propagating through
turbulent media’, Opt. Eng., 2001, 40, (8), pp. 1554–1562
runs to 1012, we would only be able to account for realisations [7] Garrido-Balsells, J.M., Jurado-Navas, A., Paris, J.F., et al.: ‘Novel
above −70 dB in this example, which is still not adequate to formulation of the m model through the generalized – k distribution for
demonstrate close agreement with analytical results. Accurate atmospheric optical channels’, Opt. Express, 2015, 23, (5), pp. 6345–6358
simulation of such processes requires advanced simulation [8] Garrido-Balsells, J.M., Lopez-Martinez, F.J., Castillo-Vázquez, M., et al.:
‘Performance analysis of FSO communications under los blockage’, Opt.
techniques which are not considered here further. Therefore, the Express, 2017, 25, (21), pp. 12550–12562
perceived SNR threshold in our simulations decreases as 1/N for [9] Simon, M.K., Alouini, M.S.: ‘Digital communication over fading channels’
the heterodyne case. When considering IM/DD systems, the (John Wiley & Sons Inc., Hoboken, NJ, 2005)
[10] Jurado–Navas, A., Maria, J., Francisco, J., et al.: ‘A unifying statistical model
problem is easily overcome with increasing the number of for atmospheric optical scintillation’, in Awrejcewicz, Jan (Eds.): ‘Numerical
simulation runs, as the perceived threshold decreases with 1/N 2. simulations of physical and engineering processes’ (InTech, Croatia, 2011),
pp. 181–206
[11] Alheadary, W.G., Park, H.K., Alouini, M.S.: ‘Performance analysis of
6 Conclusion subcarrier intensity modulation using rectangular QAM over Malaga
turbulence channels with integer and non-integer β’, Wirel. Commun. Mob.
We have investigated capacity for FSO communication over Comput., 2016, 16, pp. 2730–2742
Málaga atmospheric turbulence channel with pointing error and [12] López-González, F.J., Jurado-Navas, A., Garrido-Balsells, J.M., et al.:
path loss for IM/DD and heterodyne detection. ‘Characterization of sub-channel based Málaga atmospheric optical links with
Analytical expression in closed form for ORA, OPRA and real β parameter’, Opt. Appl., 2017, XLVII, (4), pp. 545–556
[13] Ansari, I.S., Yilmaz, F., Alouini, M.S.: ‘Performance analysis of free-space
TIFR adaptive transmission algorithms are derived. We show that it optical links over málaga (m) turbulence channels with pointing errors’, IEEE
is not possible to achieve any definitive channel capacity when Trans. Wirel. Commun., 2016, 15, (1), pp. 91–102
using CIFR adaptive transmission algorithm over the proposed [14] Amirabadi, M.A., Tabataba Vakili, V.: ‘A new optimization problem in FSO
channel model. For the proposed model of atmospheric turbulence communication system’, IEEE Commun. Lett., 2018, 22, (7), pp. 1442–1445
[15] Amirabadi, M.A., Tabataba Vakili, V.: ‘A novel hybrid FSO/RF
(for both types of detection and for different atmospheric communication system with receive diversity’, Signal Process., 2018, pp. 1–
turbulence strengths), channel capacity of CIFR adaptive 5. Available at http://arxiv.org/abs/1802.07348
transmission algorithm tends to zero. Obtained results are [16] Balaji, K.A., Prabu, K.: ‘Performance evaluation of FSO system using
numerically evaluated and graphically presented for different wavelength and time diversity over Malaga turbulence channel with pointing
errors’, Opt. Commun., 2018, 410, pp. 643–651
strength of atmospheric turbulence (weak, moderate and strong) [17] Chen, L., Wang, W.: ‘Multi-diversity combining and selection for relay-
and for both types of detection (IM/DD and heterodyne). assisted mixed RF/FSO system’, Opt. Commun., 2017, 405, pp. 1–7
Results of asymptotic high-SNR approximate expressions for
ORA, OPRA and TIFR adaptive transmission algorithms are

IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585 1583

© The Institution of Engineering and Technology 2019
[18] Nguyen, N.T.T., Vu, M.Q., Pham, H.T.T., et al.: ‘Performance enhancement of 1, ξ2 + 1
HAP-based relaying M-PPM FSO system using spatial diversity and and we write: I12 = ln g0 G2,3, 41 ηg0 , for g = g0. Second term
heterodyne detection receiver’, J. Opt. Commun., 2018, pp. 1–10 ξ2, α, k, 0
[19] Ansari, I.S., Alouini, M.S., Cheng, J.: ‘Ergodic capacity analysis of free-space in partial integration equals
optical links with nonzero boresight pointing errors’, IEEE Trans. Wirel.
Commun., 2015, 14, (8), pp. 4248–4264

∫ ∫
[20] Nistazakis, H.E., Karagianni, E.A., Tsigopoulos, A.D., et al.: ‘Average b ∞ 1, ξ2 + 1
capacity of optical wireless communication systems over atmospheric vdu = g−1G2,3, 41 ηg dg (35)
turbulence channels’, J. Lightwave Technol., 2009, 27, (8), pp. 974–979 a g0 ξ2, α, k, 0
[21] Peppas, K.P., Stassinakis, A.N., Topalis, G.K., et al.: ‘Average capacity of
optical wireless communication systems over I-K atmospheric turbulence We solve this integral using equation (2.24.2/3) from [32]
channels’, J. Opt. Commun. Netw., 2012, 4, (12), pp. 1026–1032
[22] Nistazakis, H.E., Tombras, G.S., Tsigopoulos, A.D., et al.: ‘Capacity
estimation of optical wireless communication systems over moderate to 1, ξ2 + 1, 1
strong turbulence channels’, Journal of Communications and Networks, 2009, I13 = G3,4, 51 ηg0 (36)
11, (4), pp. 384–389 0, ξ2, α, k, 0
[23] Peppas, K.P., Stassinakis, A.N., Nistazakis, H.E., et al.: ‘Capacity analysis of
dual amplify-and-forward relayed free-space optical communication systems β
over turbulence channels with pointing errors’, J. Opt. Commun. Netw., 2013, Integral I1 therefore equals I1 = ∑ bk I11 − I12 − I13 . Integral
5, (9), pp. 1032–1042 k=1
[24] Hassan, M.Z., Hossain, M.J., Cheng, J.: ‘Ergodic capacity comparison of ∞
optical wireless communications using adaptive transmissions’, Opt. Express,
I2 = ln g0 ∫g0 f g g dg is solved using (07.34.21.0085.01) from [33],
2013, 21, (17), pp. 20346–20362 yielding
[25] Anees, S., Bhatnagar, M.R.: ‘Information theoretic analysis of a dual-hop
fixed gain af based mixed RF-FSO system’. 2015 IEEE 26th Annual Int. β
ln g0 η2 Ξopra
Symp. on Personal, Indoor, and Mobile Radio Communications (PIMRC),
Hong Kong, 2015, pp. 927–931
I2 DD =
16π ∑ bk2α +k
G3,7, 70 g
16 0 Θopra
(37)
[26] Gappmair, W.: ‘Further results on the capacity of free-space optical channels k=1
in turbulent atmosphere’, IET Commun., 2011, 5, (9), pp. 1262–1267
β
[27] Kikuchi, K.: ‘High spectral density optical communication technologies’ ξ2 + 1, 1
(Springer, Berlin, Heidelberg, 2010) I2(Het) = ln g0 ∑ bkG 4, 0
2, 4 ηg0 (38)
[28] Andrews, L.C., Phillips, R.L., Young, C.Y.: ‘Laser beam scintillation with k=1 0, ξ2, α, k
applications’ (SPIE, WA, 2001)
[29] Petković, M.I., Djordjević, G.T., Milić, D.N.: ‘Ber performance of IM/DD
FSO system with ook using apd receiver’, Radioengineering, 2014, 23, (1), Final closed-form expressions for OPRA adaptive transmission
pp. 480–487 algorithms are given in Section (4.2).
[30] Petković, M., Djordjević, G.T: ‘Average BER of dual-branch FSO system
employing sim-bpsk influenced by Malaga atmospheric turbulence with
pointing errors’. Proc. of 4th Int. Conf. on Electrical,Electronics and 9.2 Appendix 2: CIFR adaptive transmission algorithm
Computing Engineering,(IcETRAN 2017). (Society for Electronics,
Telecommunications, Computers, Automatic Control and Nuclear To determine the capacity of CIFR transmission, we need to
Engineering), Kladovo, 2017, pp. TEI1.4.1–TEI1.4.6 evaluate the following expression:
[31] Stefanović, M.Č., Anastasov, J.A., Panić, S.R., et al.: ‘Channel capacity
analysis under various adaptation policies and diversity techniques over
∫
+∞
1
fading channels’, in Ali, Eksim (Eds.): ‘Wireless communications and f (x) dx, (39)
networks - recent advances’ (InTech, Croatia, 2012), pp. 281–302 0
x g
[32] Prudnikov, A.P., Brychkov, J.A.: ‘Integrasl and series’ (Fizmatlit, Moscow,
2003)
[33] W.R. Inc.: ‘Wolfram functions site’. (Wolfram Research, Inc., 2019. Available where f g(x) corresponds to (10). Results of Monte-Carlo
at: http://functions.wolfram.com simulations indicate that the value diverges, but we are set to
formally prove so. Since there are multiple summands if β > 1, we
9 Appendix need to prove that at least one of them
 
∫
+∞ 2
1 1 3 1/ r ξ + 1
9.1 Appendix 1: OPRA adaptive transmission algorithm 2 G3 0 η x dx (40)
0 x 2
ξ , α, k
∞
Integral I1 = ∫g0 ln g f g(g)dg for OPRA is somewhat complicated
diverges for the specified k. The constants that are non-essential for
to solve. It consists of 1/g, logarithm function, and Meijer's G- integration are neglected. Simple change of variable allows the
function, from combining equations (10) and (20). We solve this scaling factor η to be left out, and we further concentrate on the
integral using the partial integration method case of k = 1. The improper integral has a singular point x = 0 that
b b
∫a udv = (uv) ba − ∫a vdu where we take u = ln(g) and presents a problem for integration. Therefore, we want to prove
1
2
1 ξ + 1 that the integral diverges in arbitrary small ϵ-neighbourhood of the
dv = G1,3, 30 ηg r 2 dg. singularity
g ξ , α, k
For IM/DD and heterodyne detection, antiderivative v is
∫
ϵ 2
1 3, 0 1/ r ξ + 1
obtained using equation (07.34.21.0003.01) from [33] J= 2 G1, 3 x dx → + ∞ (41)
0 x 2
ξ , α, 1
1, ξ2 + 1
v = G2,3, 41 ηg (33) We first notice that the following limit exists:
ξ2, α, k, 0
2
1 3, 0 1/ r ξ + 1 Γ(α − 1)
First term in partial integration rule is lim+ G x 2
1/ r 1, 3
= 2 (42)
x→0 x ξ , α, 1 ξ −1
1, ξ2 + 1 ∞
(uv) ba = ln g G2,3, 41 ηg g0 Furthermore, if the function under the limit operator monotonically
ξ2, α, k, 0 decreases with x, we can simply use the comparison test to write
For g → ∞, we apply the L' Hopital's rule, yielding ξ2 + 1
G1,3, 30 ϵ1/ r
∫ ∫
ϵ ϵ
1, ξ2 + 1 Γ(α − 1) dx ξ2, α, 1 dx (43)
I11 = lim ln g G2,3, 41 ηg = 0, ≥J≥
g→∞
(34) ξ2 − 1 x 2 − 1/ r
ϵ 1/ r
x 2 − 1/ r
ξ2, α, k, 0 0 0

1584 IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585

© The Institution of Engineering and Technology 2019
Since the p-integrals in the above inequality diverge for r ≥ 1, this The form of approximate expression for channel capacity is then
would be enough to prove that J also diverges. In the case of
monotonically increasing function, we get similar inequality with ⟨C⟩(a)
the sense of comparison operators reversed, and the same outcome = constant × n0 + k0 ⋅ SNR[dB] , (46)
B
that J diverges. Thus, we prove that CIFR adaptive transmission in
the strict sense is not feasible for the analysed system model. where we denote k0 = limz → 0+ k. Parameter n0 is obtained as

9.3 Appendix 3: Approximate capacity for high SNR a1, …, ap z

n0 = lim+ Gmp,,qn z + 10 k0log10 (47)
Analytical expressions for the channel capacities of ORA, OPRA z→0 b1, …, bq C0
and TIFR adaptive transmission algorithms for IM/DD and
heterodyne detection are presented in (17), (18), (23), (24), (27) Approximations of channel capacity are obtained by applying the
and (30). Moreover, these expressions can be closely approximated procedure to (17), (18), (23), (24), (27) and (30). For example, the
by simpler ones for high SNR values. We start the procedure by approximate high-SNR channel capacity of direct-detection ORA
finding the derivative of Meijer G function with respect to SNR transmission becomes

d SNR a1, …, ap β β
⟨C⟩ora
k= Gm, n C ⋅ 10− 10 , (44) ≃ ∑ bini + ∑ biki SNR dB , (48)
d SNR p, q 0 b1, …, bq B [ ]
DD i=1 i=1

where where
1 αβκ γ + Ω′ r ln 10 1
C0 = . ki = Γ(α)Γ(i)
c γβ + Ω′ 10 2ξ2
Γ(α)Γ(i) βγ + Ω′
By applying simple change of variable z = C0 ⋅ 10− SNR /10, and ni = ln (49)
equation (07.34.20.0002.01) from [33], we express the derivative ξ2 αβκ(γ + Ω′) 2π
as 1 1
+ − 2 + Ψ(i) + Ψ(α)
2 ξ
ln 10 m, n + 1 0, a1, …, ap
k= − G z (45) In the above expression for coefficients ni, symbol Ψ( ⋅ ) stands for
10 p + 1, q + 1 b1, …bm, 1, bm + 1, …, bq
the digamma function defined as (06.14.27.0002.01) in [33].

IET Commun., 2019, Vol. 13 Iss. 11, pp. 1578-1585 1585

© The Institution of Engineering and Technology 2019