Content uploaded by Masood Asim
Author content
All content in this area was uploaded by Masood Asim on Dec 08, 2023
Content may be subject to copyright.
2023 3rd International Conference on Advancement
in Electronics & Communication Engineering(AECE)
979-8-3503-3072-4/23/$31.00 ©2023 IEEE
DOI:
Application Of Deep Learning in Free Space
Optical Communication using Malaga Channel
Masood Asim
Electronics & Communication
Engineering
Galgotias College of Engineering and
Technology, India
Greater Noida, India
Masoodasim2001@gmail.com
Lakshmanan. M
Electronics & Communication
Engineering
Galgotias College of Engineering and
Technology, India
Greater Noida, India
tmlakshmanan@gmail.com
Manoranjan Kumar Singh
Electronics & Communication
Engineering
Galgotias College of Engineering and
Technology, India
Greater Noida, India
manoranjan4152@gmail.com
Kunwar Nitesh Singh
Electronics & Communication
Engineering
Galgotias College of Engineering and
Technology, India
Greater Noida, India
Niteshsingh41562@gmail.com
Abstract—The use of Free Space Optical (FSO)
communication systems is growing as a result of their
ability to deliver high data speeds through unlicensed
spectrum with large bandwidth, higher power
efficiency, and more security. These systems are also
suitable candidates for backhaul lines for the next-
generation communication networks, as well as for
bottleneck and last-mile applications. However, the
performance of FSO systems is harmed by
atmospheric turbulence, which is caused by variations
in the temperature and pressure of the atmosphere
along the propagation path. As a result, researchers
and communication system designers can benefit while
investigating and enhancing the performance of FSO
links with Malaga distribution. At the receiver end the
received signal is fed into an ML detector with CSI,
theoretically the Maximum Likelihood (ML) detector
is the ideal detector and channel State Information
(CSI) that can be provided either in perfect or blind
forms. From the simulation results, we observe that Bit
error rate decreases with increase in average electrical
SNR for Malaga turbulence channel. The outcome of
the research done shows that Malaga Distribution fits
best for a wide range of atmospheric turbulence
conditions also the DNN detector gives better results
for Bit error rate vs Signal to noise ratio plot.
Keywords - Bit Error Rate , Malaga Distribution , Free Space
Optical Communication , DNN detector , BPSK modulation .
I. INTRODUCTION
Free-space optical communication is method of
transferring information through modulated optical signal
from one end to the other, where the medium for
transmitting this information is free space. This medium
is often called as channel which is either free space or
vacuum, and it is the characteristics of this channel that
provides us in improving the performance and reliability
of FSO communication. FSO has gained interest of
researchers through many aspects, some of which include
its advantages like high speed data transmission, low cost
installation, low power consumption, more security, and
wide bandwidth with an unlicensed spectrum in
comparison. When FSO is compared with RF systems ,
the major drawbacks in RF systems is less security of
data and licensed spectrum and because of this licensed
spectrum RF communication is more costly than FSO . It
is also seen as the best alternative for future development
in communication systems especially in the field where
high- speed data communication is required. In
comparison with fiber optics communication systems,
FSO are more flexible and for maximum exchange of
information, the transmitter and receiver need to be
aligned i.e. there should be Line of Sight (LOS)
communication. The performance of FSO system highly
depends on atmospheric conditions like wind, fog, rain,
earthquake,as well as phenomenon like scattering,
absorption and pointing errors but the major problem is
caused due to atmosphericturbulence.Scattering,
absorption, and turbulence impact the transmitted light
signal when the atmospheric channel conditions are poor.
The non- uniformity in temperature, pressure, and wind
speed over the channel change the atmosphere's
refractive index, which alters the optical signal strength. .
When the signal is received it has to be demodulated, this
process of demodulation can be done by using various
modulation-demodulation schemes like BPSK, MSK ,
OOK etc., but the ML detector is a generalized method
for demodulation and detection of the received signal .
The ML detection is done with CSI (channel state
information).
II. SYSTEM MODEL
Fig. 1 System model of deep learning-based FSO
system
In FSO communication system, the channel models are
categorized according to the turbulence level of channels.
In this study, we have used Malaga Distribution without
pointing error. The Malaga channel is valid for a wide
range of turbulence conditions also it unifies a number of
previously existing statistical models such as log-normal,
gamma-gamma, K-distribution etc. For this distribution
wehave found the expression for channel also referred as
observed Irradiance as ‘h’. Thus analysis and
improvements in the FSO system done by using Malaga
distribution would be beneficial for researchers and
communication system designers. In table 1.1 an
overview of various channels conditions and their
corresponding statistical distributions
Table 1.1 list of distributions used in various turbulence
conditions
The system model of an FSO communication system is
similar to any wireless communication. First, the Sender
sends a message in any form, then this signal is encoded
into binary data by an Encoder, this encoded form of
signal is fed into a Modulator, the modulator operates by
increasing the frequency of the input signal for further
transmission. Various modulation techniques are used, in
this study, we have worked on the BPSK modulation
scheme which transmits the binary data in form of 1 and -
1 i.e 0 as -1 and 1 as 1.
ML detector is used for detecting and demodulating the
received signal, various other detectors are available, but
since at the receiver end the detector does not know about
the modulation scheme used while transmitting the
information, therefore, ML detection is a generalized way
of detecting the received signal that can be compared to
the original signal transmitted and when the difference
between the two is known it can be reduced for better
performance of the FSO system. In the following
sections, we have shown how BER expression is
calculated for analysis and the implementation of ML
detectors with and without CSI.
2.1 Channel Model
The received signal is expressed as ‘Y’ and the
transmitted signal as ‘X’, AWGN noise is also considered
and represented as ‘W’, the channel gain is described by
‘h’.The equation for describing the System can be
expressed as: Y = h × X + W. The gain ‘h’ varies
according to the distribution selected. In this study we are
using Malaga Distribution and the channel characteristics
are obtained from [3] and [5] .
The PDF of h for Malaga distribution is derived in [4,
Eqn(24) ] as :
)
Where, α and β are the fading parameters that denote the
large-scale and small-scale fluctuations, respectively. In
(i), Kv(.) stands for the modified Bessel function of the
second kind and order ѵ (.) mentioned in [19, sec. (8.432)]
Parameter Ω represents the average power of the optical
signal for the line-of-sight (LOS) component, it is the
average power of the total scatter component, and defines
the amount of scattering power coupled to the LOS
component when g =2bo (1 −ρ. Furthermore, ΦA and
ΦBdenote deterministic angles for the LOS and coupled-
to-LOS scatter components
Ω'=Ω+ρ2bo+2 2boΩpcos(ФA+ФB)
The modified Bessel function in (i) rewritten in terms
of the Meijer-G function from [20] is as follows:
)
From [3] we get the final expression for PDF of
Malaga distributions, ‘h’ parameter as:
The Bit error rate is he number of bit errors per unit time.
The average Bit error rate can be calculated using:
=
(viii)
Where P(e) is the conditional error probability given as:
WEAK
TURBULENCE
MODERATE
TURBULENCE
STRONG
TURBULENCE
Log Normal
Distribution
Log Normal
Distribution
G-G Distribution
G-G Distribution
Negative
Exponential
K-Distribution
Malaga
Distribution
Malaga
Distribution
Malaga
Distribution
For BPSK modulation [20], where erfc is the
complementary error function. Finally, the P(e) is
expressed in [11] as:
(ix)
After applying Integration property of Meijer-G function
[21]. The average BER for FSO system is expressed as:
(x)
2.3 Implementing ML detector with CSI
Further we have applied ML detector with perfect CSI on
our simulation results and the outcome is shown in fig
2.3.1. For blind CSI conditions the results are displayed
in figure 2.3.2. If the receiver has perfect CSI, data can be
approximated from the obtained statistical data by
comparing each ‘y’ to a pre-calculated decision threshold.
In fact, due to the symmetric distribution of AWGN noise,
the threshold is a simple function of instantaneous
atmospheric turbulence and background radiation
mean.In theory, if the receiver has perfect CSI, the
information bearing data can be determined from the
received statistics by comparing each ‘y’ to a pre-
calculated decision threshold. For detection with CSI, the
maximum likelihood (ML) decision rule for the kth bit is
used[6]:
sˆ[k]
=1 r[k]
≷τCSI
sˆ[k]=0
Where τCSI = Ib + Is/ 2
a) We modeled the above equation in MATLAB by
eliminating the BPSK demodulator and using above
decision threshold.
b) We plotted the curve for Perfect CSI for SNR [0:35]
in db and we verified the curve from [6].
c) The final stage of our system is the deployment of a
deep learning-based detector. We removed the ML
detector and introduced our perceptron model at the
receiver end. The perceptron model is a DNN that
uses weights and bias vectors to train itself and
update these parameters in a cycle on completion of
each single iteration. This model uses sigmoid
activation function and the training is done with the
help of the user-specified number of epochs. Greater
the number of epochs, the better the system's
performance. To reduce complexity, we used only
three epochs here. The primary function of this
perceptron model is to generate bits with the least
amount of error, commonly known as target output.
The model essentially attempts to attain higher
performance than perfect CSI by training itself by
adjusting its parameters at the end of each iteration.
Here, we trained our model for each SNR value ranging
from 0 to 35, for a total of 36 values. The output of the
perceptron model is now saved in a variable, and the
graph is presented beside the perfect CSI curve. We
discovered that the deep learning curve outperforms the
perfect CSI curve
2.5 Deep Neural Network Model at Reciever end
There are several detection techniques for combating the
effect of atmospheric turbulence, such as the perfect CSI
ML , we need to compare the DNN with the ideal CSI
ML detector to evaluate its performance. All of these
detectors require channel state information and have high
computational complexity, making them unsuitable for
real- time applications. In this research, we develop a
deep learning-based detector to improve system
performance.
A deep learning-based detector is a generalized detector
that can receive bits at the receiver with performance
levels comparable to or better than the perfect CSI. If the
detector's training cycle lengthens, it signifies that the
bigger the number of training cycles, the better the
detector's performance. At the transmitter, we employ
BPSK modulation, and the channel is a Malaga
distribution with AWGN (additive white Gaussian noise
with 0-db variance). Now that the DNN (deep neural
network) has been implemented at the receiver, it is vital
to create the perfect CSI ML detector with known channel
characteristics since we need to compare the DNN with
the ideal CSI ML detector to evaluate its performance.
III. RESULT AND DISCUSSION
In this work we studied the performance of FSO system
in terms of Bit error rate. The Performance of FSO
system is degraded by various atmospheric turbulences,
Malaga distribution is used to describe closed form
expression for bit error rate and probability distribution
function for the channel gain ‘h’. Further ML detector is
applied on the received signal, for perfect and imperfect
CSI conditions and the outcome is verified with that in
[2]. Various parameter values are adjusted for obtaining
the desired outcomes, Table 2.1 shows an overview of all
the parameters used and their values for Malaga
turbulence channel.
Table 2.1 parameters used for Malaga turbulence channel.
S.No
Distribution Parameters
Value
1
α
2.1
2
β
4
3
Ω
0.75
4
ρ
0.85
5
N
2,4,8,16
6
ΦA
90
7
ΦB
0
8
bo
0.5
Following the ML detector, a deep neural network is
simulated in MATLAB using a perceptron model as the
deep neural network (DNN). In MATLAB, we compared
the ideal CSI ML detector against our deep learning-based
detector. Parameters that are used in deep learning model
are given below
3.1 Bit Error Rate performance for Malaga
distribution without ML detector
Fig. 3.1 Bit Error rate for Malaga fading
(Condition: α = 2.1, β = 3, Ω = 0.75, ρ= 0.85, ΦA=90, ΦB=0,
bo=0.5) for FSO system
Fig 3.1 represents the Bit Error Rate and Average
Electrical SNR over Malaga channel (ρ= 0.85) for FSO
system under Malaga Turbulence channel. This shows
that for a particular threshold as the average electrical
SNR increases the Bit error rate decreases.
3.2 Bit Error Rate performance for Malaga
distribution with ML detector for perfect CSI
Fig 3.2 Bit Error rate for Malaga fading
(condition: α = 2.1, β = 3, Ω = 0.75, ρ= 0.85, ΦA = 90, ΦB = 0 ,bo
= 0.5 ) for FSO system
Fig 3.2 Represents Bit error performance with ML
detector under perfect CSI conditions for Malaga
turbulence channel.
It can be observed that with ML detector for perfect CSI
conditions the graph obtained is closer to that of
simulation done without ML detector.
3.3 BER for BPSK modulation in Malaga channel
with deep learning
Fig 3.2
(deep learning results for BER vs SNR plot in Malaga
channel)
3.4 Comparision of Ml detector vs Deep learning
results
Fig 3.3
(comparing deep learning based results with that obtained from ML
detector)
3.5 Training performance of implemented system
model
Fig.3.4
plot mean absolute error vs. Epochs
It signifies that the mean absolute error of FSO system
is reduced by deep learning model.
IV. CONCLUSIONS
In this study, we used Malaga Channel and applied a
deep learning detector based on multilayer perceptron
models for the FSO system, for th BPSK modulation
method. As we've shown, the DL-based detector
operates efficiently and yields results that are superior
to those of the ideal CSI-based ML detector in terms of
BER. The ML detector has a drawback in that it
requires channel state information, is more
sophisticated, and has a low data rate. The perceptron
model must be trained for a large number of iterations
to increase its effectiveness. Additionally, this deep
learning-based detector can be used as a generalised
detector at the receiver and is applicable to any
modulation scheme. Thus we can conclude Malaga
Distribution can be used for wide range of turbulence
conditions and provided better results for the deep
learning detector.
REFERENCES
[1] Amirabadi, M. A., Kahaei, M. H., & Nezamalhosseini, S. A.
(2020). Deep learning- based detection technique for FSO
communication systems. Physical Communication, 43, 101229.
https://doi.org/10.1016/j.phycom.2020.101229
[2] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and
Products, Academic Press, 2014. Table of Integrals, Series, and
Products. (2015). Elsevier. https://doi.org/10.1016/c2010-0-64839-
5.
[3] Riediger, M. L. B., Schober, R., & Lampe, L. (2008). Blind
detection of on-off keying for free- space optical communications.
(2008) Canadian Conference on Electricaland Computer
Engineering. https://doi.org/10.1109/ccece.2008.456476
[4] Jurado-Navas, A., Maria, J., Francisco, J., & Puerta-Notario, A.
(2011). A Unifying Statistical Model for Atmospheric Optical
Scintillation. Numerical Simulations of Physical and Engineering
Processes https://doi.org/10.5772/25097
[5] Milosevic, N. D., Petkovic, M. I., & Djordjevic, G.T. (2017,
August). Average BER of SIM-DPSK FSO System With Multiple
Receivers over distributed Atmospheric Channel With Pointing
Errors. IEEE Photonics Journal, 9(4), 1–10.
https://doi.org/10.1109/jphot.2017.2710320
[6] Milosevic, N. D., Petkovic, M. I., & Djordjevic, G. (2017, August).
Average BER of SIM-DPSK FSO System With Multiple Receivers
over distributed Atmospheric Channel With Pointing Errors. IEEE
PhotonicsJournal,9(4),1–10.
https://doi.org/10.1109/jphot.2017.2710320
[7] Zixiong Wang, Wen-De Zhong, Songnian Fu, & Chinlon Lin.
(2009). Performance comparison of different modulation formats
over free-space optical (FSO) turbulence links with space diversity
reception technique. IEEE Photonics Journal, 1(6), 277–285.
ttps://doi.org/10.1109/jphot.2009.2039015
[8] Kumar, R., Singh, P., & Kumar, N. (2020). Alamouti Code
Generator in Optical Domain Using Mach–Zehnder
Interferometer,79–82. https://doi.org/10.1007/978-981-15-5546-
6_7
[9] Goldsmith, A. (2005). Wireless Communications. Cambridge:
Cambridge University Press. doi:10.1017/CBO9780511841224
[10] Uysal, M., Jing Li, & Meng Yu. (2006, June). Error rate
performance analysis of coded free- space optical links over
gamma-gamma atmospheric turbulence channels. IEEE
Transactions on wireless communications,5(6),1229-1233.
https://doi.org/10.1109/twc.2006.1638639
[11] Alsemmeari, R. A., Bakhsh, S. T., & Alsemmeari, H. (2016). Free
space optics vs radio frequency wireless communication. Int. J. Inf.
Technol. Comput. Sci, 8(9), 1-8.
[12] Rani, R., & Kaur, G. (2021, November). Design and analysis of
MIMO FSO system and WDM FSO system for leh (ladakh), India
under worst weather conditions. In 2021 7th international
conference on signal processing and communication (ICSC) (pp.
31-36). IEEE.
[13] Petkovic, M. I., Zdravkovic, N. M., & Dordevic, G.T. (2014,
November). Outage performance of switch-and-examine
combining receiver over FSO Gamma-Gamma atmospheric
turbulence with pointing errors. 2014 22nd Telecommunications
[14] Wang, Z., Dedo, M. I., Guo, K., Zhou, K., Shen, F., Sun, Y., ... &
Guo, Z. (2019). Efficient recognition of the propagated orbital
angular momentum modes in turbulences with the convolutional
neural network. IEEE Photonics Journal, 11(3), 1-14.
[15] Li, J., Zhang, M., Wang, D., Wu, S., & Zhan, Y. (2018). Joint
atmospheric turbulence detection and adaptive demodulation
technique using the CNN for the OAM-FSO
communication. Optics express, 26(8), 10494-10508.
[16] Amirabadi, M. A., Kahaei, M. H., & Nezamalhosseini, S. A.
(2020). Deep learning based detection technique for FSO
communication systems. Physical Communication, 43, 101229.
[17] Zhao, Q., Hao, S., Wang, Y., Wang, L., Wan, X., & Xu, C. (2018).
Mode detection of misaligned orbital angular momentum beams
based on convolutional neural network. Applied Optics, 57(35),
10152-10158.
[18] Li, Z., & Zhao, X. (2017). BP artificial neural network based wave
front correction for sensor-less free space optics
communication. Optics Communications, 385, 219-228.
[19] Al-Gailani, S. A., Salleh, M. F. M., Salem, A. A., Shaddad, R. Q.,
Sheikh, U. U., Algeelani, N. A., & Almohamad, T. A. (2020). A
survey of free space optics (FSO) communication systems, links,
and networks. IEEE Access, 9, 7353-7373.
[20] Tian, Q., Li, Z., Hu, K., Zhu, L., Pan, X., Zhang, Q., ... & Xin, X.
(2018). Turbo-coded 16-ary OAM shift keying FSO
communication system combining the CNN-based adaptive
demodulator. Optics express, 26(21), 27849-27864.
