Optical and Quantum Electronics (2024) 56:48

https://doi.org/10.1007/s11082-023-05667-y

Performance estimation of SISO and MIMO Ro‑FSO link

under atmospheric turbulence

Sanmukh Kaur1 · Aanchal Sharma1

Received: 17 August 2023 / Accepted: 28 October 2023 / Published online: 23 November 2023
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023

Abstract
The present study focuses on analysing the bit error rate (BER) of single-input single-out-
put and multiple-input multiple-output Radio-over free space optics system under atmos-
pheric turbulence conditions employing binary phase shift keying (BPSK) modulation.
The system performance has been investigated for weak to strong scintillation conditions
considering uncoded and Reed Solomon coded BPSK signals. The performance of the
proposed system has also been analysed under strong turbulence effects considering vary-
ing signal to noise ratio, responsivity, ­Eb/No and different scattering conditions. Further,
different machine learning techniques namely, artificial neural network (ANN), decision
tree, and gradient boosting regression have been employed for estimation of received signal
quality in terms of BER. Highest (0.98) and lowest (0.0155) values of R-squared ­(R2) and
root mean square error respectively have been obtained in case of ANN model resulting in
the best fit model.

Keywords Ro-FSO · Turbulence · Scintillation · Machine learning · ANN · DT · GBR

1 Introduction

Due to the continually increasing number of mobile users and the rise of bandwidth-
intensive applications like live streaming, video calling, and fast internet, mobile ser-
vice providers are facing challenges in efficiently distributing the limited radio fre-
quency (RF) spectrum among end users. Free space optics (FSO) links offer numerous
advantages over conventional communication systems, including low cost, quick and
simple deployment, license-free spectrum availability, immunity to electromagnetic
interference (Elsayed 2023; Elsayed et al. 2022) lightweight equipment, high channel
capacity and secure data transmission (Sharma et al. 2021a). Radio-over free space
optics (Ro-FSO) links present a viable solution to meet the demand for high channel
bandwidth in future wireless networks (Kaur et al. 2023a). Ro-FSO aims to provide a
universal platform for affordable heterogeneous wireless services similar to radio over

* Sanmukh Kaur
[email protected]
1
Amity School of Engineering and Technology, Amity University Uttar Pradesh, Noida 201203,
India

13
Vol.:(0123456789)
48 Page 2 of 22 S. Kaur, A. Sharma

fiber (RoF) systems. Ro-FSO transceivers can directly transmit an optical signal from
free space into the fiber and can transmit multiple RoF signals from an optical fiber
core into the air. With its wide-ranging capabilities and applications, Ro-FSO is one
of the essential technologies for implementing future networks. It is particularly suited
for delivering high-speed communications for 5G networks, as it offers greater capacity
to provide the best user experience (Chaudhary and Amphawan 2019). High-definition
television (HDTV) (Mallick et al. 2022), cellular backhaul (Bag et al. 2018), last-mile
network access (Kashif et al. 2020), and many other applications are offered by the tech-
nology. In challenging terrains like hilly areas, it provides excellent connectivity along
with high transmission rates (Kaur et al. 2023b).
Figure 1 illustrates the model of a terrestrial Ro-FSO system, a preferred technology
for current and future wireless systems to meet the demand for higher data rates.
Losses that affect the laser beams traveling through air include geometric loss, ambi-
ent noise, and different atmospheric conditions such as turbulence, precipitation, and
scattering (Sharma et al. 2021b; Nistazakis et al. 2014; Kaur 2019; Alheadary et al.
2018; Refai et al. 2005; Ahmed et al. 2019). Geometric loss occurs when the align-
ment between transmitter and receiver is incorrect, while atmospheric losses result from
changes in temperature, pressure, and weather conditions. The performance of FSO sys-
tems is based on both internal and external parameters (Kaur 2019), such as the optical
power, bandwidth, wavelength, and loss at the transmitter and receiver. External param-
eters include air attenuation, dazzling effects, aiming loss, window loss, and meteoro-
logical and geographical structures of the area (Alheadary et al. 2018). Rain and fog are
particularly detrimental to Ro-FSO system performance and can degrade its capacity
(Refai et al. 2005). Ambient noise lowers the signal-to-noise ratio (SNR), while turbu-
lence-induced fading occurs due to the erratic changes in pressure and temperature.
Turbulent conditions in Ro-FSO systems can range from weak to strong, as a result
of refractive index fluctuations due to atmospheric effects, which are primarily caused
by temperature gradients. Under severe turbulence, the scintillation or intensity fluctua-
tions increases, leading to probability of higher bit error rate (BER) in the communica-
tion link. Scattering conditions can exacerbate this effect, as scattering of light by aero-
sols or other particles in the atmosphere may introduce additional intensity fluctuations.
There are numerous statistical models, such as the Lognormal (LN), Gamma-Gamma
(G-G), Negative Exponential, K-distribution and M-distribution that can be used to
define the atmospheric turbulence effects. M-distribution offers flexibility to capture
varying turbulence conditions, exhibits good empirical fit to actual channel data, and
provides mathematical tractability for effective channel characterization. This makes it
a valuable choice for understanding and mitigating turbulence effects in binary phase
shift keying (BPSK) based Ro-FSO communication, enabling more accurate link perfor-
mance predictions and system optimization.

Fig. 1  Schematic of basic Ro-FSO system

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 3 of 22 48

Scattering refers to the phenomenon where light waves are dispersed or redirected due
to interactions with atmospheric turbulence. It causes fluctuations in the received signal’s
intensity, leading to signal fading or degradation of system performance.
Multiple-input-multiple-output (MIMO) systems offer improved signal quality, cov-
erage, and throughput as compared to single–input–single–output (SISO) systems thus
making them a preferred choice in a wide range of wireless communication applications
(Ahmed et al. 2019).
To mitigate the impact of atmospheric turbulence and pointing errors, different meth-
ods are employed, such as diversity, relay, multiple input multiple output, aperture aver-
aging, modulation techniques, and error-correcting codes (ECCs) (Kaushal et al. 2017).
In Ro-FSO communication, different modulation schemes have been used previously to
mitigate the atmospheric turbulence effect, including On–Off Keying (OOK), pulse posi-
tion modulation (PPM), pulse width modulation (PWM) and digital pulse interval modula-
tion (DPIM). OOK modulation is commonly used because of its simplicity and low cost,
but it requires an appropriate adaptive threshold for optimal performance (Zedini and
Alouini 2015). PPM has low spectral efficiency, while PWM requires lower peak power
for transmission than PPM and exhibits better spectral efficiency (Pham et al. 2014). How-
ever, PWM needs higher average power than PPM. DPIM is an asynchronous modula-
tion scheme with variable symbol length, leading to a high possibility of error propaga-
tion during signal demodulation at the receiver. Subcarrier intensity modulation (SIM)
has also been used, but it has poor power efficiency (Fan and Green 2007; Elsayed and
Yousif 2020a). Orthogonal frequency division multiplexing (OFDM) may be employed to
overcome these limitations. Additionally, BPSK combined with other modulation schemes
offers better performance (Ghassemlooy et al. 1998).
In an OFDM-based Ro-FSO communication system, mutually orthogonal subcarri-
ers separated at a specific frequency in the frequency domain are utilized to bring signals
parallelly together. Orthogonality of sub-carriers is achieved by employing fast fourier
transform at the transmitter terminal resulting in elimination of the effects of co-channel
interference and inter-symbol interference. In Hassan et al. (2019), the authors investigate
OFDM technique with Ro-FSO links and show that the use of technique can eliminate the
aforementioned effects.
Reed-Solomon (RS) coding is one of the most widely used error-correcting technique
employed in data communication systems. It operates by adding redundant symbols to the
original data, enabling the detection and correction of errors introduced during transmis-
sion. RS codes are characterized by their ability to correct a fixed number of errors, mak-
ing them robust against burst errors, which are common in noisy communication channels.
They find extensive applications in various domains, including wireless communica-
tion, data storage, and optical communication systems. By adding redundancy, RS codes
enhance data reliability, making them an essential tool for ensuring accurate and depend-
able data transfer in challenging environments, thereby improving the overall system per-
formance and quality of service.
Different methods have been explored by researchers to increase the data-carrying
capacity of the FSO link, such as multiplexing in intensity, phase, code, polarization, and
wavelength (Fan and Green 2007; Elsayed and Yousif 2020a; Ghassemlooy et al. 1998;
Hassan et al. 2019; Tang et al. 2010). The investigation carried out in Ghassemlooy et al.
(1998), analyses the efficiency of an FSO link encoded with Bose-Chaudhuri-Hocqueng-
hem (BCH) code under the influence of a Gamma-Gamma turbulence model. In Tang et al.
(2010), the authors investigated the BER performance of a convolutional coded OFDM-
based FSO system over a Gamma-Gamma turbulence channel.

13
48 Page 4 of 22 S. Kaur, A. Sharma

The use of machine learning (ML) for estimation of Ro-FSO link performance offers
more precise forecasts, improved environmental adaptability, link parameter optimisation,
and improved problem identification and mitigation. These benefits may help Ro-FSO
communication system work more effectively, efficiently, and dependably (Sharma and
Kaur 2013).
In Yousif et al. (2019), the authors propose an analytical system design for the inte-
gration of Orbital Angular Momentum (OAM) multiplexing and MIMO or SMM (Spatial
Modulation and Multiplexing) signal processing within a FSO channel under the impact
of atmospheric turbulence. The simulation findings demonstrate that the capacity of the
OAM-based MIMO system surpasses that of the conventional MIMO system beyond a cer-
tain predefined propagation distance threshold.
In Yousif and Elsayed (2019), the authors develop an adaptive MIMO-FSO links utiliz-
ing OAM-MIMO/SMM multiplexing and evaluate their performance in the presence of
atmospheric turbulence conditions. The computational outcomes indicate that by imple-
menting OAM-based MIMO/SMM with SMD multiplexing, the system achieves signifi-
cantly improved BER performance, especially when employing adaptive MIMO/SMM
equalization techniques.
In Algedir and Elganimi (2020), the authors evaluated the quality factor (Q-factor) of
FSO systems using four ML algorithms, namely multi-linear regression, support vector
regression, decision tree (DT) regression, and random forest (RF) regression. The DT and
RF models showed higher and lower values of coefficient of determination ­(R2) and mean
square error (MSE) respectively compared to the other models.
In Elsayed and Yousif (2020b), the authors use an on–off keying non-return-to-zero
(OOK–NRZ) and digital pulse position modulation (DPPM) techniques to study and
improve the MIMO-FSO link’s performance. The numerical findings indicate that, at a
range of 2.5 km, the sensitivity of DPPM improves by approximately ­10–11 dB, resulting
in a BER of less than ­10–12. A better performance is thus achieved by employing DPPM as
compared to equivalent OOK–NRZ modulation scheme.
In Lionis et al. (2021), the authors considered real-time non-linear atmospheric condi-
tions and employed different ML techniques, including artificial neural network (ANN), to
estimate the performance of the FSO link. The analysis suggests the improvement of model
performance using ML techniques compared to standard regression techniques. In Chaud-
hary et al. (2021), the authors designed a Ro-FSO system by incorporating hybrid mode
division multiplexing (MDM) and polarization division multiplexing (PDM) schemes to
achieve an acceptable BER of 1­ 0–3. Four independent channels, each carrying a signal at
10 Gbps data rate have been used. Input channels have been upconverted to 40 GHz radio
signal for propagation over 3.4 km FSO link operating under clear weather conditions.
In Tripathi et al. (2022), different weather conditions with varying attenuation losses
have been considered to enhance the performance of Ro-FSO links. In Sinha and Kumar
(2022), the authors investigated a RF carrying signal for dual hop FSO/RF links of
500 m/50 m, employing Gamma-Gamma and Lognormal models to achieve a BER of ­10–6
in turbulent atmospheric conditions.
In Grover and Sheetal (2019), authors investigated the performance of Ro-FSO link
with MDM using spiral phased Hermite Gaussian modes under different weather condi-
tions. The results indicate that HG00 mode is more reliable under clear air or heavy rain
and achieved an improved propagation range of 7 to 24.4 km with a Q-factor ranging from
24.4 to 2.98 respectively. The use of quadrature amplitude modulation (QAM-16) modula-
tion in composite (HG00 + HG01) mode extended the transmission range by a factor of
(2.9 to 1.7) km with a tolerable BER of ­10–5.

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 5 of 22 48

In this work, Ro-FSO link has been examined considering a BPSK based spatial modu-
lation under different atmospheric weak and strong turbulence conditions. The BER per-
formance of uncoded Ro-FSO communication links is compared with RS coded Ro-FSO
links. The proposed system’s performance has been analysed for different scattering condi-
tions and varying values of responsivity, SNR and energy per bit. ANN, DT and gradient
boosting regression (GBR) ML techniques have been employed further for BER estimation
considering ­R2 and root mean square error (RMSE) as performance metrics.
The remainder of the work has been structured as follows: In Sect. 2, a system model
is presented to predict the performance of proposed Ro-FSO system considering mathe-
matical modelling. In Sect. 3, we different ML techniques that can be used to enhance the
performance of the Ro-FSO system have been discussed. The results and discussions are
presented in Sect. 4. Finally, the research work is summarized and conclusions are drawn
in the Sect. 5.

2 System model

Figure 2 depicts a MIMO system model, where the input bit stream is first modulated
using BPSK modulation. The modulated digital signals are then passed through a MIMO
encoder utilizing the orthogonal space time block code (OSTBC) technique. The channel
assumed in this model is Rayleigh fading. The section responsible for reception carries
out the opposite task, which involves detecting all the transmitted signals and forwarding
them to the OSTBC Decoder. The decoded signal is then passed through an equalizer uti-
lizing the maximum likelihood and maximum ratio combining (MRC) equalization tech-
niques. Following this, the demodulation process is carried out using a BPSK demodula-
tor, and an approximated signal is recovered. Meanwhile, the implementation of RS-coding
involves encoding the data at the sender, transmitting it, and using decoding algorithms at
the receiver to recover the original information.

2.1 Single input single output (SISO) system

One transmitting aperture and one receiving aperture along with channel make up a SISO
FSO system. The intended signals are modulated using the BPSK subcarrier intensity mod-
ulation (BPSK-SIM) signalling method and delivered through telescope into an additive
white Gaussian noise (AWGN) enhanced discrete time ergodic channel. Natural turbulence
in the air path causes the transmitted signal to be scattered. Different environmental factors
are responsible for these signal corruptions. We consider a channel of high atmospheric

Fig. 2  Schematic of MIMO system

13
48 Page 6 of 22 S. Kaur, A. Sharma

turbulence in this work. The receiver’s interpretation of the signal (y) is given by Zhu and
Kahn (2002).
y= h𝛾PFSO x + n (1)
where, ‘h’ represents the channel state, ‘γ’ represents detector’s responsiveness, ‘x’ defines
the transmitted signal, ‘n’ represents noises from different sources, and ‘PFSO’ represents
the average optical transmitted power. The optical intensity changes brought on by fad-
ing, turbulence, and air loss are modelled by the channel state ‘h’ as (Farid and Hranilovic
2007).
h = hI hs hp (2)
where ‘hs’, ‘hp’ and ‘hl’ are caused by scintillation effects, geometric spread and beam
extinction and path loss respectively.
Additionally, the received signal’s SNR, (Samimi and Azmi 2010), is stated as
2
SNR(h) = (𝛾h)2 ∕2𝜎 n (3)

where, ‘𝜎n2’ is the channel noise variance.

For a BPSK-SIM-signalled communication link, the conditional BER probability is deter-
mined by the intensity fluctuation (Popoola et al. 2008) and may be determined by
( )
h𝛾
Pec,SISO (h) = 0.5 × erfc (4)
2𝜎

where, ‘γ’ and ‘σ2’ are photo detector responsivity and variance of the channel noise
respectively.
To determine the average BER ‘Pe, SISO’ for a gamma-gamma channel affected by pointing
errors, the Eq. (4) needs to be averaged over the probability density function (PDF) of ‘h’. The
estimation can yield the desired result as:
∞

∫ (5)
Pe,SISO = Pec,SISO (h)fh (h)dh
0

2.2 Multiple‑input‑multiple output (MIMO) system

The system being evaluated consists of M and N number of transmitting and receiving aper-
tures respectively. Signals are modulated using BPSK-SIM and transmitted through a tel-
escope into a channel which is ergodic, discrete time and with AWGN in highly turbulent air.
The received signal ­(yn) is expressed as (Popoola and Ghassemlooy 2009).
∑ � �
yn = x𝛾 M m=1 hmn + vn , n = 1, 2, … … N (6)

where ‘x’ represents the transmitted signal, ‘hmn’ defines the irradiance from the mth trans-
mitter and nth receiver. ‘vn’ represents AWGN having mean value as zero and variance
given by v = N0/2. There are optical intensity fluctuations due to atmospheric loss, turbu-
lence and fading. The irradiance ‘hmn’ models the three types of fluctuations (Tsiftsis et al.
2009) and is expressed as:

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 7 of 22 48

hmn = hImn hsmn hpmn (7)

The expression involves ‘hlmn’, which accounts for the attenuation caused by beam
extinction and propagation path loss from the mth transmitter to the nth receiver. Scin-
tillation effects are represented by ‘hsmn’. ‘hpmn’ captures the effects of geometric spread
as well as pointing errors.
Spatial diversity approach can be used at the transmitter and receiver side of a mul-
tiple-input single-output (MISO) and single-input multiple-output (SIMO) systems
respectively. It may be employed at both the sides of a MIMO system and is the princi-
pal solution for a channel under strong atmospheric turbulence conditions. The average
BER in case of a MIMO-FSO link may be computed from Sharma and Kaur (2021a).
�
⎛ �N � �2 ⎞
∞ �� �M
�
∫ 0
⎜ h𝛾
Pe,MIMO = fh (h) × 0.5 × erfc hmn ⎟dh (8)
⎜ 2MN𝜎 n=1 m=1 ⎟
⎝ ⎠

The ­Pe, MIMO pertains to the joint probability density function (PDF) of vector ‘h’,
denoted as ­(h11, ­h12,..., ­hMN) with a length of MN. The joint PDF is represented by
‘fh(h)’.
Scintillation refers to the irregular changes in the intensity of light resulting from
turbulence in the atmosphere. The turbulence causes fluctuations in power and optical
beam deflections at the receiver ­(Rx). Solar radiation can lead to unpredictable swings
in air density, and rise in temperature near the surface of the planet exacerbating the tur-
bulence effects. The turbulence-induced variations in the refractive index of the atmos-
phere can be quantified by Cn2 parameter (Al Naboulsi et al. 2004). The Rytov param-
eter σ2l , expresses the impact of turbulence on the transmitted beam as a function of Cn2,
and is described by the following equation:

𝜎l2 = 1.23Cn2 K 7∕6 L11∕6 (9)

where ‘L’ stands for the path length in meters. ‘K’ is 2𝜋

𝜆
.
The three turbulence regimes, namely strong (𝜎l2> 1), moderate (𝜎l2 ~ 1) and weak
turbulence (𝜎l2 < 1) are classified based on the value of 𝜎l2 . Turbulence modelling con-
siders the PDF of the G-G and LN distributions (Nazari et al. 2016). In a weak atmos-
pheric condition, the light intensity ‘I’ follows the LN distribution, and its PDF can be
expressed as:
� � � �
� � 𝜎2
1 I I 2
P(I) = √ exp − ln + ∕2𝜎 I (10)
I𝜎 2𝜋 <I> 2
1

where, 𝜎I2 is the variance of the intensity and is given in terms of natural logarithm. It is
determined by:

⎡ ⎤
⎢ ⎥
⎢ 0.49𝜎l2 0.51𝜎l2 ⎥
(11)
2
𝜎I = exp⎢ � �7 + � �5 ⎥−1
⎢ 12 6 17 6 ⎥
⎢ + ⎥
3
1 1.11𝜎 1 + 0.69𝜎 5

⎣ l l
⎦

13
48 Page 8 of 22 S. Kaur, A. Sharma

The G-G distribution is an alternative model that can characterize all turbulence
regimes. The PDF of the received light intensity according to this distribution is as follows:
( √ )
2(𝛼𝛽) 2 ( I ) 𝛼 + 𝛽 ( )
𝛼+𝛽
I
P(I) = − 1K𝛼−𝛽 2 𝛼𝛽 (12)
Γ(𝛼)Γ(𝛽) < I > 2 <I>

−1
⎧ ⎡ ⎤ ⎫
⎪ ⎢ ⎥ ⎪
⎪ ⎢ 0.49𝜎l2 ⎥ ⎪
𝛼 = ⎨exp⎢ � �7 ⎥ − 1⎬ (13)
⎪ ⎢ 12 6 ⎥ ⎪
⎪ ⎢ + ⎥ ⎪
5
1 1.11𝜎
⎩ ⎣ l
⎦ ⎭

−1
⎧ ⎡ ⎤ ⎫
⎪ ⎢ ⎥ ⎪
⎪ ⎢ 0.51𝜎l2 ⎥ ⎪
𝛽 = ⎨exp⎢ � �5 ⎥ − 1⎬ (14)
⎪ ⎢ 12 6 ⎥ ⎪
⎪ ⎢ 1 + 0.69𝜎l ⎥ ⎪
5

⎩ ⎣ ⎦ ⎭

The G-G distribution is defined by the parameters ‘α’ and ‘β’, which capture the
impact of large-scale and small-scale scintillation, respectively. Gamma function is rep-
resented by ‘Г’. Here, ‘Kn’ refers to the modified Bessel function of the second kind of
order ‘n’, At the receiving end, the average carrier-to-noise-plus-distortion ratio (CNDR
n(h)) for each subcarrier ‘n’ is calculated as the sum of the optical noise and intermodu-
lation distortion (IMD).

(mn 𝜌ltot pt h)2

CNDRn (h) = N (15)
2
2( T0 + 𝜎IMD )
s

The obtained CNDR n per sub-carrier may be given as:

( )2
mn 𝜌ltot pt h
[CNDRn ]AV ≈ ([ ] [ 2 ] ) (16)
N
2 T0 + 𝜎IMD AV
s AV

where, the scintillation values of each individual quantity are averaged and denoted as A ­ V,
and n represents the nth sub-carrier. At the receiver end, an optical signal is converted into
an electrical signal using a positive-intrinsic-negative (PIN) type photodiode. The retrieved
message bits are demodulated using quadrature modulation (QM) and demodulator sec-
tions. The signal is further passed through a low pass filter for filtering out any unwanted
frequencies before being sent to the regenerator for data pulse rearrangement. In order to
analyse the BER, Q-factor and eye diagram of the received signal, the sent and received
bits are aligned side by side in a BER analyzer.
Q-factor is one of the widely used performance parameter in optical communication
systems to evaluate the quality of signal, as the BER is typically too low to be meas-
ured. The ratio of number of bits erroneously received by the detector to the number of
bits transmitted is known as BER. It is used to determine the probability of errors. The

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 9 of 22 48

Q-factor is represented as the minimum SNR required for a given signal to achieve a
specific BER (Sharma and Kaur 2021b).
V1 − V0
Q= (17)
𝜎1 + 𝜎0

Assuming equal amounts of Gaussian noise across all signal levels, ‘V1’ and ‘V0’
represent the mean values of signal levels 1 and 0 respectively. ‘σ1’ and ‘σ0’ represent
the noise values for each signal level, corresponding to 1 and 0 transmission probabili-
ties respectively. The relationship between the BER and Q-factor can be expressed as
(Lubana and Kaur 2023):
� �
1 Q
BER = erfc √ (18)
2 2

3 ML techniques for performance estimation

Machine learning is a subfield of artificial intelligence (AI) that aims for imitat-
ing human learning through the use of data and algorithms. This enables machines to
enhance their performance over time, improve their accuracy in making predictions or
in classifying the datasets and uncover data-driven insights (Kaur 2023).
Figure 3 depicts the schematic block diagram of BPSK Ro-FSO system employing
ML techniques. Number of transmitting and receiving antennas, atmospheric condi-
tions and modulation techniques have been given as input parameters or features to the
proposed Ro-FSO system as well as ML model. BER results obtained through simula-
tion has also been applied to the ML model as modelling targets. Training the model
with ANN, DT and GBR ML algorithms thus predicts the performance of the system in
terms of BER for unseen dataset.

3.1 Artificial neural network

An ANN is a type of computational model that simulates the functioning of nerve cells
in the human brain. These models are equipped with learning algorithms that enable
them to adapt and learn from new data, making them a powerful tool to model non-
linear statistical data. One of the most notable advantages of ANNs is their ability for
effectively learning from observation of data sets (Kaur et al. 2023a). Figure 4 illus-
trates the process of developing an ANN algorithm, while Fig. 5 displays the schematic
diagram of an ANN with input, output, and hidden layers.
It offers several advantages, including improved signal detection in challenging con-
ditions, adaptive modulation for optimized data rates, and enhanced error correction
mechanisms.
However, ANN implementation entails complexity and computational demands,
requiring significant data for effective training, and posing a risk of overfitting if not

13
48 Page 10 of 22 S. Kaur, A. Sharma

Fig. 3  Block Schematic of BPSK Ro-FSO system employing ML

adequately regularized. Careful consideration is needed to harness the benefits of ANN

while mitigating the challenges when applied to Ro-FSO systems.

3.2 Decision tree

The supervised learning technique called DT can resolve classification or regression

problems. This technique involves a classifier having tree type of structure. Each leaf
node in the structure represents the outcome of classification and the internal nodes
represent the features of the dataset. The responsibility of decision nodes is to make
choices that can lead to multiple branches, while the leaf nodes serve as the final result
of those choices and don’t possess any additional branches. The dataset’s features are
used to perform the test or make decisions. The general structure of a DT is illustrated
in the form of a flowchart in Fig. 6.
DT offers the advantages of simplicity, interpretability, and suitability for binary clas-
sification tasks. It can facilitate rapid decision-making processes, which is advantageous

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 11 of 22 48

Fig. 4  Flowchart of ANN model

Fig. 5  Schematic block of ANN

13
48 Page 12 of 22 S. Kaur, A. Sharma

Fig. 6  Flowchart of DT

for signal classification and system control. The technique however, may struggle with
complex, non-linear relationships in the data and thus may be susceptible to overfitting
if not pruned appropriately.

3.3 Gradient boosting regression

GBR is a machine learning strategy that can enhance prediction accuracy and can be
applied to different ML techniques (Kaur 2023). The primary difference between this strat-
egy and DT and RF methods is that it trains the model sequentially rather than in a con-
tinuous manner. To elaborate, the training data set is initially used to fit the first DT. The
residuals obtained from the first tree are then used to fit the second DT and the new tree
is added to the fitted function for updating the residuals. The model is further improved
in areas where it performs poorly by fitting the residuals to relatively small trees with a
limited number of terminal nodes. The addition of a shrinkage parameter, represented as
’λ,’ allows for the fitting of more diverse types of trees in regions with higher residuals.
Figure 7 illustrates the schematic diagram of GBR technique.
GBR offers advantages such as high predictive accuracy, robustness to outliers, and the
ability to model complex non-linear relationships, making it valuable for channel quality
prediction and link optimization. However, these models can be computationally inten-
sive and require careful tuning of hyperparameters. Overfitting is also a concern when not
appropriately regularized. Thus, the decision to use GBR in Ro-FSO systems guided by a
trade-off between its predictive power and computational demands.

4 Results and discussion

The present section presents the results obtained by realization of BPSK based Ro-FSO
system considering SISO and MIMO spatial modulation using MATLAB 2022. The effects
of different atmospheric turbulence under strong to weak scintillation conditions have been
analysed for an uncoded and RS coded system. The proposed system’s performance has
also been investigated for different scattering conditions and varying values of respon-
sivity, SNR and energy per bit. Performance of the system has been further estimated by

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 13 of 22 48

Fig. 7  Schematic diagram of GBR technique

Table 1  Simulation parameters Component Value

of the system
Number of transmitting antenna 2
Modulation BPSK
Noise model AWGN
Channel model Rayleigh
Atmospheric Conditions Turbulence and Scintillation
Spatial modulation SISO and MIMO
Number of receiving antenna 2

considering different ML techniques. The parameters used in simulation of the proposed

BPSK based Ro-FSO system have been listed in Table 1.
The performance comparison of uncoded and RS coded 2 × 2 MIMO Ro-FSO systems
is shown in Fig. 8. The results have been analysed in terms BER as a function of CNDR
under weak to strong turbulent and scintillation conditions. The BER of ­10–5 and ­10–23
require 5 dB and 20 dB CNDR respectively in RS coded weak turbulence conditions.
Based on the results shown in Fig. 8, it can be inferred that the RS coded Ro-FSO system
outperforms the uncoded system under weak to strong turbulence conditions.
Figure 9 presents comparison curves of an uncoded and a RS coded Ro-FSO system
under different scattering conditions considering strong turbulence (α = 1, β = 1). The
results show that, for moderate (ρ = 0.5) and maximum (ρ = 0) scattering, the RS coded Ro-
FSO system outperforms the uncoded system. However, best results have been obtained in

13
48 Page 14 of 22 S. Kaur, A. Sharma

Fig. 8  BER performance com-

parison of coded and uncoded
BPSK Ro-FSO system under
different turbulence conditions

Fig. 9  BER performance com-

parison of coded and uncoded
BPSK Ro-FSO system under
different scattering conditions

case of RS coded system for the moderate scattering condition (ρ = 0.5) with an achieved
BER of ­10–8 and ­10–23 at 5 dB and 20 dB CNDR values respectively.
RS-coded Ro-FSO systems thus outperforms the uncoded systems due to their error-
correction capabilities. RS codes are effective in mitigating the impact of burst errors
which may be caused by strong turbulence and scattering, thereby improving BER per-
formance and enhancing the system’s reliability in challenging conditions. The specific
scattering conditions for which RS-coded Ro-FSO systems outperform uncoded ones
depend on varying factors, including link distance, system design, and the severity of
turbulence.

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 15 of 22 48

Fig. 10  BER as a function of

responsivity for SISO and MIMO
Ro-FSO systems

Figure 10 depicts the BER as a function of responsivity for SISO and MIMO FSO
system under strong turbulence conditions. It is observed that an improved BER per-
formance for a given responsivity value of the photodiode has been achieved for higher
number of apertures at transmitting and receiving side for strong (α = 1, β = 1) atmos-
pheric turbulence. The Minimum value of BER of 1­ 0−25 has been achieved for MIMO
(4 × 4) system with a responsivity of 0.8 A/W.
Figure 11 illustrates the BER performance as a function of SNR under strong (α = 1,
β = 1) atmospheric turbulence. The average BER performance improves considerably with
increase in the number of transmitting and receiving apertures.
MIMO systems consistently outperform SISO systems in due to their inherent advan-
tages. MIMO leverages multiple antennas at both the transmitter and receiver, which
enhances system performance through spatial diversity and multiplexing. This leads to
improved link reliability and data rates. Factors contributing to MIMO’s superior perfor-
mance include increased link robustness against fading and interference, greater capacity
for data transmission, and the ability to exploit spatial dimensions for parallel data streams.

Fig. 11  BER as a function of

SNR for SISO and MIMO Ro-
FSO systems

13
48 Page 16 of 22 S. Kaur, A. Sharma

Fig. 12  BER analysis of 2 × 2

MIMO system

Also, better values of BERs of ­10−12 and ­10−14 are obtained for MIMO (4 × 4) system at
SNR values of 20 dB and 50 dB respectively.
Figure 12 depicts the BER analysis of a 2 × 2 MIMO system. The results indicate
that the BER values range from 1­ 0–10 to 1­ 0–18, for a variation in ­Eb/No values from 0 to
40 dB when the system is simulated with a known channel. In contrast, the theoretical
values of BER range from 1­ 0–1 to ­10–9 when the system is simulated with BPSK mod-
ulation using orthogonal space–time block codes (OSTBC) and maximum likelihood
equalizer under Rayleigh channel.
To determine the ideal trade-off between ­Eb/No and BER, the simulated ML case
employs simulations. Meanwhile, the simulated ML with a known channel makes use of
the already known information of the channel to find the best trade-off.
The proposed Ro-FSO system’s performance has been analysed using different ML
techniques, including ANN, DT, and GBR. Regression analysis has been performed
using an ANN neural fitting program to evaluate the system’s performance. The data-
set has been divided into three parts, with 15% set for testing as well as validation and
70% utilized for training. A feed-forward network with two layers and multiple hidden
neurons has been used, and a scaled conjugate gradient method has been employed to
minimize memory usage. The algorithm has been utilized to calculate the MSE and
regression values, where lower MSE and higher R values indicate a close relationship
or smaller differences between the observed data set and predicted values. Input power,
attenuation, propagation range, turbulent and scintillation conditions have been consid-
ered as input features. BER of the received signal acts as a modelling target.
Figures 13 and 14 display the error histogram and regression fit plots generated by
the ANN for the BER target set. The RMSE values for training, validation, and testing
are 0.0155, 0.0325, and 0.0675, respectively. Moreover, the R values for the training,
validation, and testing are 0.98541, 0.98609, and 0.98232, respectively.
The DT technique involves splitting the input data into training and testing sets, with
70% of the total 520 values reserved for training and the remaining 30% for testing.
The coarse DT method has been utilized to determine the RMSE and ­R2 values for the
BER target data set after each iteration. The first iteration has been found to yield the

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 17 of 22 48

Fig. 13  Error histogram plot of ANN with 20 bins for BER estimation

best point hyperparameters and the smallest MSE, and an increase in the number of
neighbours or iterations results in a higher MSE value. Figure 15 depicts the RMSE and
R-squared values for the training and testing BER data sets as a function of the number
of neighbours.
The GBR model was trained with observations ranging from 1 to 10 neighbors and a
learning rate of 0.05. The model’s performance is satisfactory, with R ­ 2 value of 0.97 and
an RMSE of 0.38 for training data set. Figure 16 illustrates the RMSE and R-squared
plots for the training and testing datasets as a function of leaf size.
Table 2 presents the outcomes of three ML techniques utilized to train and test the BER
data sets of the simulated values from the proposed Ro-FSO model. The RMSE value rep-
resents the proximity of actual and predicted values, and a lower value indicates a bet-
ter model performance. Among all the models, the ANN approach delivers the highest
R-squared value of 0.98541 and the least error of 0.0155, demonstrating excellent compat-
ibility with the observed data set. Meanwhile, Machine Learning enhances BPSK Ro-FSO
systems performance by enabling accurate bit error identification through advanced signal
estimation and monitoring, ensuring reliable data transmission.

5 Conclusion

In the present work, a BPSK based terrestrial Ro-FSO system is investigated in terms
of BER considering strong to weak scintillation conditions. The BER performance of
uncoded Ro-FSO communication links is compared with RS coded Ro-FSO links. ANN,
DT, and GBR models have been trained for performance estimation in terms of RMSE &
R-squared values. RS coded Ro-FSO system outperforms the uncoded system under weak
to strong turbulence conditions with an achievement of BER of ­10–23 at CNDR value of

13
48 Page 18 of 22 S. Kaur, A. Sharma

Fig. 14  Regression fit plots of ANN with 10 hidden neurons

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 19 of 22 48

Fig. 15  RMSE and R-squared values employing coarse DT

Fig. 16  RMSE and R-squared values employing GBR

Table 2  Comparative analysis of different ML techniques

S. No. ML Algorithm R-squared ­(R2) RMSE
Training dataset Testing dataset Training dataset Testing dataset

1 ANN 0.9854 0.9823 0.0155 0.0675

2 DT 0.7641 0.7623 1.5025 2.0542
3 GBR 0.9755 0.9642 0.3838 0.4665

13
48 Page 20 of 22 S. Kaur, A. Sharma

20 dB. Higher values of prediction accuracies and ­R2 (0.98) has been observed in case of
ANN model. The system may be further investigated for cutting-edge modulation tech-
niques and real time performance monitoring of impairments in optical networks.

Author contributions All authors contributed to the study conception and design. Manuscript prepara-
tion and review were performed by SK and AS. Data collection and analysis were performed by SKS. All
authors read and approved the final manuscript.

Funding The authors declare that no funds, Grants, or other support were received during the preparation of
this manuscript.

Availability of data and materials The datasets generated during and/or analysed during the current study
are available from the corresponding author on reasonable request.

Declarations
Conflict of interest The authors have no relevant financial or non-financial interests to disclose.

Ethical approval Not applicable.

References
Ahmed, A., et al.: Performance analysis of WDM-MIMO free space optical system under atmospheric
turbulence. In: 2019 6th International Conference on Signal Processing and Integrated Networks
(SPIN), IEEE (2019)
Al Naboulsi, M., Sizun, H., de Fornel, F.D.R.: Fog attenuation prediction for optical and infrared waves.
Opt. Eng. 43(2), 319–329 (2004)
Algedir, A.A., and Elganimi, T.Y.: Machine learning models for predicting the quality factor of FSO
systems with multiple transceivers. In 2020 2nd Global Power, Energy and Communication Confer-
ence (GPECOM), pp. 308–311. IEEE (2020)
Alheadary, W.G., Park, K.H., Alfaraj, N., Guo, Y., Stegenburgs, E., Ng, T.K., et al.: Free-space optical
channel characterization and experimental validation in a coastal environment. Opt. Express 26(6),
6614–6628 (2018)
Bag, B., Das, A., Ansari, I.S., Prokeš, A., Bose, C., Chandra, A.: Performance analysis of hybrid FSO
systems using FSO/RF-FSO link adaptation. IEEE Photon. J. 10(3), 1–17 (2018)
Chaudhary, S., Amphawan, A.: High speed MDM-Ro-FSO communication system by incorporating
AMI scheme. Int. J. Electr. Lett. 7(3), 304–310 (2019)
Chaudhary, S., Wuttisittikulkij, L., Nebhen, J., Tang, X., Saadi, M., Al Otaibi, S., Choudhary, S.: Hybrid
MDM-PDM based Ro-FSO system for broadband services by incorporating donut modes under
diverse weather conditions. Front. Phys. 9, 506 (2021)
Elsayed, E.E., Yousif, B.B.: Performance enhancement of the average spectral efficiency using an aper-
ture averaging and spatial-coherence diversity based on the modified-PPM modulation for MISO
FSO links. Optics Commun. 463, 125463 (2020a)
Elsayed, E.E., Yousif, B.B.: Performance enhancement of hybrid diversity for M-ary modified pulse-
position modulation and spatial modulation of MIMO-FSO systems under the atmospheric turbu-
lence effects with geometric spreading. Opt. Quant. Electron. 52(12), 508 (2020b)
Elsayed, E.E., et al.: Investigations on wavelength-division multiplexed fibre/FSO PON system employ-
ing DPPM scheme. Opt. Quant. Electron. 54(6), 358 (2022)
Elsayed, E. E. (2023). "Performance evaluation and enhancement of DWDM systems under the impact
of four waves mixing crosstalk in optical communications”.
Fan, Y., Green, R.J.: Comparison of pulse position modulation and pulse width modulation for applica-
tion in optical communications. Opt. Eng. 46(6), 065001 (2007)
Farid, A.A., Hranilovic, S.: Outage capacity optimization for free-space optical links with pointing
errors. J. Lightwave Technol. 25(7), 1702–1710 (2007)

13
Performance estimation of SISO and MIMO Ro‑FSO link under… Page 21 of 22 48

Ghassemlooy, Z., Hayes, A., Seed, N., Kaluarachchi, E.: Digital pulse interval modulation for optical
communications. IEEE Commun. Mag.Commun. Mag. 36(12), 95–99 (1998)
Grover, A., Sheetal, A.: Improved performance investigation of 10 Gb/s–10 GHz 4-QAM based OFDM-
Ro-FSO transmission link. J. Optic. Commun. (2019). https://​doi.​org/​10.​1515/​joc-​2019-​0223
Hassan, M.Z., Hossain, M.J., Cheng, J., Leung, V.C.: Subcarrier intensity modulated optical wireless
communications: a survey from communication theory perspective. ZTE Commun. 14(2), 2–12
(2019)
Kashif, H., Khan, M.N., Altalbe, A.: Hybrid optical-radio transmission system link quality: link budget
analysis. IEEE Access 8, 65983–65992 (2020)
Kaur, S.: Analysis of inter-satellite free-space optical link performance considering different system param-
eters. Opto-Electron. Rev.-Electron. Rev. 27(1), 10–13 (2019)
Kaur, S., et al.: Machine Learning based performance estimation of terrestrial Ro-FSO Link. Phys. Script.
(2023). https://​doi.​org/​10.​1088/​1402-​4896/​acde15
Kaur, S., Kaur, J., Sharma, A.: Predicting the performance of radio over free space optics system using
machine learning techniques. Optik 281, 170798 (2023a)
Kaur, J., Kaur, S., Sharma, A., Amphawan, A.: Performance analysis and rain attenuation modelling of Ro-
FSO link for hilly region of India. Frequenz 77(1–2), 115–124 (2023b)
Kaushal, H., Jain, V., Kar, S.: Free space optical communication, Springer, 2017.
Lionis, A., Peppas, K., Nistazakis, H.E., Tsigopoulos, A., Cohn, K., Zagouras, A.: Using machine learning
algorithms for accurate received optical power prediction of an FSO link over a maritime environment.
Photonics 8, 212 (2021)
Lubana, A., Kaur, S.: FWM crosstalk reduction and performance investigation of SC-DWDM system
employing ML techniques. Opt. Fiber Technol.Fiber Technol. 78, 103304 (2023)
Mallick, K., Atta, R., Sarkar, N., Dutta, B., Kuiri, B., Mandal, P., Patra, A.S.: Performance evaluation of free
space optics communication system in the scenario of triple play service using probabilistic shaping
scheme. Optics Commun. 522, 128699 (2022)
Nazari, Z., Gholami, A., Vali, Z., Sedghi, M., and Ghassemlooy, Z.: Experimental investigation of scintil-
lation effect on FSO channel. In: 2016 24th Iranian Conference on Electrical Engineering (ICEE), pp.
1629–1633. IEEE (2016)
Nistazakis, H.E., Stassinakis, A.N., Sandalidis, H.G., Tombras, G.S.: QAM and PSK OFDM Ro-FSO Over
$ M $-turbulence induced fading channels. IEEE Photonics J. 7(1), 1–11 (2014)
Pham, H.T., Dang, N.T., Pham, A.T.: Effects of atmospheric turbulence and misalignment fading on per-
formance of serial-relaying m-ary pulse-position modulation free-space optical systems with partially
coherent gaussian beam. IET Commun.Commun. 8(10), 1762–1768 (2014)
Popoola, W.O., Ghassemlooy, Z.: BPSK subcarrier intensity modulated free-space optical communications
in atmospheric turbulence. J. Lightwave Technol. 27(8), 967–973 (2009)
Popoola, W.O., Ghassemlooy, Z., Allen, J.I.H., Leitgeb, E., Gao, S.: Free-space optical communication
employing subcarrier modulation and spatial diversity in atmospheric turbulence channel. IET Opto-
electron.Optoelectron. 2(1), 16–23 (2008)
Refai, H.H., Sluss Jr, J.J., & Refai, H.H.: The transmission of multiple RF signals in free-space optics using
wavelength division multiplexing. In: Atmospheric Propagation II, vol. 5793, pp. 136–143. SPIE
(2005)
Samimi, H., Azmi, P.: Subcarrier intensity modulated free-space optical communications in K-distributed
turbulence channels. J. Opt. Commun. Networking 2(8), 625–632 (2010)
Sharma, V., Kaur, G.: High speed, long reach ofdm-fso transmission link incorporating ossb and otsb
schemes. Optik 124(23), 6111–6114 (2013)
Sharma, A., Kaur, S.: Performance evaluation and fog attenuation modelling of FSO link for hilly regions of
India. Opt. Quant. Electron. 53, 1–17 (2021a)
Sharma, A., Kaur, S.: Performance analysis of 1280 Gbps DWDM–FSO system employing advanced modu-
lation schemes. Optik 248, 168135 (2021b)
Sharma, A., Kaur, S., Chaudhary, S.: Performance analysis of 320 Gbps DWDM—FSO system under the
effect of different atmospheric conditions. Opt. Quant. Electron. 53(5), 239 (2021a)
Sharma, K., Kaur, S., Chaudhary, S.: Performance investigation of hybrid MDM-WDM-FSO link under
influence of rainfall and fog conditions. Optoelectr. Adv. Mater. Rapid Commun. 15, 471–479 (2021b)
Sinha, S., and Kumar, C.: Performance investigation of 2× 20 Gb/s MDM-Ro-FSO link incorporating spi-
ral-phased Hermite–Gaussian modes under strong weather conditions. J. Optics, 1–14 (2022)
Tang, X., et al.: Performance of bpsk subcarrier intensity modulation free-space optical communications
using a log-normal atmospheric turbulence model. In: 2010 Symposium on Photonics and Optoelec-
tronics, pp. 1–4. IEEE (2010)

13
48 Page 22 of 22 S. Kaur, A. Sharma

Tripathi, A., Gupta, S., Mandloi, A., Soni, G.G.: Optically assisted mm-wave-based multi-Gbps Ro-FSO
transmission link under channel fading models. J. Mod. Opt. 69(8), 419–426 (2022)
Tsiftsis, T.A., Sandalidis, H.G., Karagiannidis, G.K., Uysal, M.: Optical wireless links with spatial diversity
over strong atmospheric turbulence channels. IEEE Trans. Wireless Commun.Commun. 8, 951–957
(2009)
Yousif, B.B., Elsayed, E.E.: Performance enhancement of an orbital-angular-momentum-multiplexed free-
space optical link under atmospheric turbulence effects using spatial-mode multiplexing and hybrid
diversity based on adaptive MIMO equalization. IEEE Access 7, 84401–84412 (2019)
Yousif, B.B., et al.: Atmospheric turbulence mitigation using spatial mode multiplexing and modified pulse
position modulation in hybrid RF/FSO orbital-angular-momentum multiplexed based on MIMO wire-
less communications system. Optics Commun. 436, 197–208 (2019)
Zedini, E., Alouini, M.-S.: Multihop relaying over im/dd fso systems with pointing errors. J. Lightwave
Technol. 33(23), 5007–5015 (2015)
Zhu, X.M., Kahn, J.M.: Free-space optical communication through atmospheric turbulence channels. IEEE
Trans. Commun.Commun. 50(8), 1293–1300 (2002)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

13