This article has been accepted for publication in IEEE Photonics Journal.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

1
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

An Improved Adaptive Coding and Modulation

Scheme with Hybrid Switching Standard for
UAV‑to‑Ground Free Space Optical Communication
Qianwu Zhang1,*, Boyang Liu1, Guanwen Chen1, Shucheng Zhan1, Zhiyu Li1, Jing Zhang2, Ning Jiang2, Bingyao
Cao1, Zhengxuan Li1
1
Key Laboratory of Specialty Optics and Optical Access Networks, Shanghai Institute for Advanced Communication
and Data Science, Shanghai University, Shanghai 200072, China
2
School of Information and Communication Engineering, University of Electronic Science and Technology of China,
Chengdu, Sichuan 611731, China
*zhangqianwu@shu.edu.cn

 Abstract—In this paper, an improved adaptive coding and be overcame in the practical implementation of FSO systems
modulation scheme with hybrid switching standard is proposed [6-8]. One of the main causes for performance degradation in
for UAV-to-Ground free space optical (FSO) communication to FSO systems is atmospheric turbulence, which is caused by
mitigate the degradation of turbulence intensity variation. In the uneven distribution of temperature and pressure in the
order to improve bits efficiency in different turbulence, a channel
atmosphere [9,10]. In practice, atmospheric turbulence induces
estimator based on Gated Recurrent Unit (GRU) and a hybrid
switching standard based on frame error rate (FER) of LDPC
phase disturbances, resulting in issues such as intensity
decoding are applied. Simulation results show that the GRU fluctuations, beam wander, and beam broadening of FSO
estimator achieves approximately 34.9% improvement in signals.
training time compared to the LSTM estimator, without The fluctuations of atmospheric turbulence are described
degradation in system performance. Moreover, the hybrid by statistical models based on the intensity of turbulence,
switching standard significantly improves the signal-to-noise which are used to assess its impact on the reliability of FSO
ratio (SNR) switching threshold deviation of the ACM scheme communication [11,12]. The widely used models include the
from 1.8 dB and 4 dB, which were observed in moderate Log-Normal distribution, which is suitable for weak
turbulence conditions, to an average of 0.35 dB. This
turbulence, the Gamma-Gamma distribution for moderate to
improvement allows for the effective implementation of the ACM
scheme in different turbulence intensities at a relatively small
strong turbulence, and the K-distribution for strong turbulence
cost, thereby enhancing the communication quality of FSO [13-15]. These models introduce probability density functions
systems in atmospheric channels characterized by turbulent based on the turbulence intensity to describe the effects of
variations. turbulence on FSO signals. Among them, the most widely
used which is employed in this paper is the Gamma-Gamma
Index Terms—UAV, FSO, ACM model, which utilizes the properties of the Gamma distribution
to describe the influence of atmospheric turbulence in vertical
I. INTRODUCTION links [16]. It effectively describes the impacts of FSO

F ree space optical (FSO) communication is a wireless channels from UAV to ground.
method that utilizes laser beams to transmit data The performance of space optical communication systems
through the air instead of optical fibers [1]. In areas is deeply influenced by Channel State Information (CSI). Due
where deploying optical fibers is challenging, it is considered to the relatively slow variations of channel fading in FSO
as an alternative, cost-effective, and fast-deploying solution. systems compared to the signal rate, CSI can be accurately
Furthermore, It is regarded as a strong competitor to estimated by the estimator of receiver. Consequently, to
traditional RF technology due to its advantages such as high- enhance communication reliability and throughput, the
speed transmission data rate, security, absence of bandwidth, transmitter can adjust its transmitting parameters, which has
and low power consumption [2]. FSO can be used in both been proven effective in RF communication systems [17,18].
space-to-ground and ground-to-ground communications, In existing research of FSO communication system,
particularly in last-mile connectivity [3, 4]. With the recent various types of adaptive transmission systems have been
popularity of unmanned-aerial-vehicle (UAV), FSO proposed and their feasibility and advantages have been
communication for high-speed UAV-to-ground links has validated. Zhou from Auburn University investigated adaptive
gained significant attention due to the increasing demand for power allocation schemes to support high data rats in WDM
transmission rates and bandwidth [5]. Despite its many transmission of RoFSO system [19]. Safi provided an adaptive
advantages, there are still challenges and obstacles that need to coding and power control and analyzed the impacts of channel
estimation errors in practical FSO systems [20].
Chatzidiamantis raised an adaptive S-PSK modulation

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

2
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

schemes in ground FSO systems and offered efficient ACM scheme with hybrid switching standard and GRU
utilization of channel capacity [21]. Alshaer from Tanta channel estimator.
University provided an adaptive modulation scheme
A. FSO System model
employing SIMO techniques in a satellite-to-ground FSO
system in presence of scintillation and beam wander to The UAV-to-Ground FSO channel model is given by:
improve spectrum efficiency [22]. Djordjevic proposed an y  RIx  n, (1)
algorithm to adjust coding and modulation schemes based on where x is the transmitted optical intensity signal, R is the
CSI while keeping the transmission power constant [23]. photodiode responsivity and I is the optical signal after
In the above literature, the authors proposed different atmosphere turbulence. Besides, y is the received electrical
methods of adaptive transmission schemes based on CSI in signal and n is the additive white Gaussian noise (AWGN)
FSO communication, aiming to improve the reliability and with zero-mean and variance.
throughput of communication systems. However, they all We use Gamma-Gamma distribution to model the random
considered adaptive transmission schemes under the atmospheric variable I whose probability density function is
assumption of unchanged turbulence. Because of the presence given by：
of turbulence, different turbulence intensities at the same SNR
 
2  2
 
level can also affect communication quality. This implies that  
1
while these schemes work well when turbulence remains f GG I   I 2
K   2  I ; I  0 (2)
constant, the situation that adaptive transmission schemes vary   
for different turbulence strength indicates that turbulence where Γ (.) is the well-known gamma function, K(.) is the
variations can render previously formulated adaptive Modified Bessel function of the second kind, and 1/β and 1/α
transmission schemes ineffective or even degrade transmission are the variances of the small and large scale eddies,
performance. For instance, applying an adaptive coding and respectively, the α and β are shown as:
modulation (ACM) scheme designed for weak turbulence in 1
conditions of moderate or strong turbulence would inevitably   0.49 R2  
  exp   1 (3)
result in the selected coding and modulation scheme being
unable to meet the required BER threshold, thereby leading to 
  1  1.11 12
R
/5
7/6 
 
a reduction in communication quality or even communication 1
disruption under the same signal-to-noise ratio (SNR).   0.51 R2  
  exp   1 (4)
Similarly, employing an ACM scheme tailored for strong
turbulence in conditions of moderate or weak turbulence, in 
  1  0.69 12
R
/5
 5/ 6 
 
the pursuit of achieving an excessively low BER, may lead to
the selection of a simpler coding and modulation scheme.  R2  1.23Cn2 k 7 / 6 z11/ 6 , (5)
However, this would decrease the channel utilization rate,
subsequently impacting the bits rate of communication. where σ2R denotes the Rytov variance, k = 2π/λ is the optical
Therefore, it is necessary to format an improved ACM scheme wave number, and z is the propagation distance of FSO links,
to adapt to FSO channel with variable turbulence intensity. and C2n is the parameter of index refraction structure.
In this paper, an improved ACM scheme with hybrid Generally, σ2R is used to differentiate the turbulence intensity.
switching standard is proposed for UAV-to-Ground FSO The σ2R below 0.5 is classified as weak turbulence, between
channels to address the problems associated with portability 0.5 and 1.6 as moderate turbulence, and values exceeding 1.6
across different turbulence strength, which is validated by are considered strong turbulence [24].
simulation. It is observed that the GRU estimator achieves
B. Principle of ACM with hybrid switching standard
approximately 34.9% improvement in training time compared
to the LSTM estimator, without any degradation in system The traditional ACM scheme aims to maximize the bits
performance. Additionally, a hybrid switching standard based transmitted per symbol interval by utilizing the highest
on frame error rate of LDPC decoding is introduced. This possible coding and modulation order, as long as the bit error
significantly improves the SNR switching threshold deviation rate (BER) remains below an acceptable threshold P [22]. This
of the ACM scheme from 1.8 dB and 4 dB to an average of is achieved by estimating the channel state and instantaneous
0.35 dB. This improvement allows for the effective SNR through the transmission of pilot symbols at the
implementation of the ACM scheme in different turbulence beginning of communication symbols. Through a feedback
intensities at a relatively small cost, thereby enhancing the link (a low data rate radio signal which is assumed to be error-
communication quality of FSO systems in atmospheric free), the channel state information is sent from the receiver
channels characterized by turbulent variations. (Rx) to the transmitter (Tx).
The adaptive bit efficiency is defined as the number of bits
II. PRINCIPLE OF PROPOSED SCHEME transmitted per symbol interval after removing redundancy, is
given as:
RPadp  maxM i , Rc j , Pe h   P 
In this section, we present the system model used in the
UAV-to-Ground FSO system and demonstrate the principle of (6)

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

1
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

Fig. 1. Structure of the proposed UAV-to-Ground FSO communication system

where Mi and Rcj denotes the modulation and coding order of Recurrent Unit (GRU) model exhibits advantages in handling
transmission system separately, Pe is the BER of received sequential signals. In comparison to the conventionally Long
signal, and P represents the target BER of communication. Short-Term Memory (LSTM) neural network model, GRU
When considering the presence of turbulence in the channel, can also be applied for channel estimator. When used to
the traditional ACM scheme formulated based on a fixed estimate the CSI in the FSO system proposed in this paper,
turbulence intensity becomes ineffective since different GRU demonstrates shorter training time and requires fewer
turbulence intensity can affect the BER at the same SNR. training resources without compromising performance. The
Considering the practical limitation that Rx can not estimate principle of the GRU model is illustrated in the Fig. 2.
the BER without pilot sequences to adjust the ACM scheme in
real-time communication. Therefore, we propose a hybrid
switching standard based on the FER of LDPC decoding.
The hybrid switching standard is established by determining
the lower and upper limits of FER, denoted as f0 and f1,
respectively, based on a predefined ACM scheme. At the Rx
side, if the FER exceeds f0, the encoding and modulation order
is decreased, while if the FER is below f1, the encoding and Fig. 2. Principle of GRU.
modulation order is increased. This approach aims to
maximize the bit efficiency within the range of f1 < f < f0. The GRU can handle time sequences that have correlation with
hybrid switching standard we proposed at the Rx side is as each backward and forward better by reset gate, update gate
follows: and candidate hidden state [25]. The functions of each part in
RPadp  maxM i , Rc j , Pe h   P, f1  f  f 0 
GRU are described as:
(7) rt   Wxr xt  Whr ht 1  br  (8)
M i  2i i  1,2 , Rc j  1 / 2,2 / 3,4 / 5 zt   Wxz xt  Whz ht 1  bz  (9)
where the threshold P in terms of BER represents the forward ~
error correction (FEC) threshold, while the thresholds f1 and f0 ht  tanhWxh xt  Whh rt  ht 1   bh  (10)
for FER are determined based on the ACM scheme with ~
certain redundancy in the presence of moderate turbulence. In ht  1  zt   ht  zt  ht 1 (11)
the intensity modulation/direct detection (IM/DD) system, the
modulation formats considered are OOK and PAM4, while where rt, zt, and are the outputs of reset gate, update gate and
candidate hidden state, respectively. Besides, σ is the sigmoid
LDPC encoding is used with code rates of {1/2, 2/3, 4/5}.
In a fixed turbulence scenario, the ACM scheme formulated function, ht, which is updated by zt and indicates the output
value of the block. xt is the time sequence, and ○ denotes the
for the given turbulence conditions does not require FER-
Hadamard product. W and b are the parameters determined by
based corrections. However, under changing turbulence
training of the GRU network.
conditions, the ACM scheme, which is no longer suitable for
the current channel, can be improved using the hybrid
III. SIMULATION SYSTEM SETUP
switching standard. This allows the system to adapt to the
attenuation caused by turbulence variations and enhance the In this paper, the proposed scheme is simulated and validated
overall communication quality of the FSO system. by Optisystem and MATLAB. The structure of the proposed
UAV-to-Ground FSO communication system is illustrated in the
C. Principle of GRU model Fig. 1. Firstly, at the transmitter, PRBS15 data is encoded using
As a type of Recurrent Neural Network (RNN), The Gated LDPC encoder and modulated by OOK/PAM4 modulator. The

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

2
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

signal is modulated by Mach-Zehndar modulator (MZM) at 1550 poor portability of the ACM scheme across different turbulence
nm laser provided by continuous wave (CW) laser. The optical conditions. These two methods are necessary in atmospheric
signal is transmitted through the UAV-to-Ground FSO channel, channels where turbulence variations are likely to occur. The
which primarily encompasses a 20 km atmospheric channel. results and analysis of these improvements are presented as
Consequently, the channel is affected by both atmospheric- follows.
induced signal attenuation and the gamma-gamma distribution
representing atmospheric turbulence. These parameters are
A. GRU channel estimator
adjusted by modifying the channel attenuation rate and the
atmospheric refractive index parameter, thereby influencing the The performance of the GRU estimator model is presented in
channel state. At the receiver, an avalanche photodiode (APD) Fig. 3(a), where the turbulence intensities varies at values of 0.5,
converts the optical signal into an electrical signal. Subsequently, 1, 1.6, and 3.5, evaluated by the mean squared error (MSE) metric,
the SNR of the channel is estimated by the GRU channel which indicates the accuracy of the model’s predictions (smaller
estimator through Python. After demodulation and decoding of values indicate better performance). It can be observed that the
the electrical signal, which is realized by belief propagation MSE gradually decreases as the SNR increases, and the overall
algorithm [26], the LDPC decoding FER is analyzed by the FER MSE of the GRU estimator remains consistently below 0.01.
analyzer. The hyper switching standard, composed of CSI and Additionally, with increasing turbulence intensity, the MSE
FER, is used to select the most suitable coding and modulation exhibits a slight upward trend at the same SNR level.
scheme for the current channel, based on the predefined ACM Consequently, the GRU estimator model demonstrates
scheme. This ACM scheme is then transmitted back to the excellent performance in the variable UAV-to-Ground FSO
transmitter via a feedback channel (a low data rate radio signal channel.
which is assumed to be error-free), which the encoder and Moreover, Fig. 3(b) presents the MSE curves of the GRU
modulator at the transmitter adjust the coding and modulation and LSTM estimators under weak turbulence conditions. As
formats according to. Finally, the BER analyzer evaluates the the SNR increases, it can be observed that the MSE of the
BER of the communication system. The simulation parameters GRU estimator is comparable to that of the LSTM estimator.
are listed in Table I. As shown in Fig. 3(c), the MSE of both the GRU and LSTM
estimators exhibit a decreasing trend as the hidden size
TABLE I. PARAMETERS USED IN SIMULATION increases, particularly at SNR of 20 dB. However, this
declining trend becomes less noticeable when the hidden size
Parameters Value
exceeds 120. Although the MSE of the GRU estimator is
Laser frequency 1550 nm
consistently slightly higher than that of the LSTM estimator
Link length 20 km when the hidden size is less than 120, they exhibit parallel
Bit rate 10 Gbps MSE behavior when the hidden size surpasses 120. In
APD gain 20 dB summary, the GRU estimator model demonstrates similar
APD responsivity 1 A/W
goodness of fit and prediction accuracy to the LSTM estimator
model in the FSO channel.
Attenuation coefficient 0.1 dB/km
Zenith angle 60 °
Rytov variance 0.5, 1, 1.6, 3.5
Modulation type OOK/PAM4
CW laser power 10 dBm
Transmit power 20 dBm
CW laser linewidth 10MHz
MZM bandwidth 40 GHz
APD bandwidth 48 GHz
APD dark current 10 nA
(a)
GRU hidden size 120

IV. SIMULATION RESULTS AND DISCUSSIONS

The proposed methods in the ACM scheme for UAV-to-
Ground FSO channels are primarily focused on the hybrid
switching standard. In the part of channel estimation, a more
resource-efficient and faster channel estimator based on GRU is
proposed. Additionally, the hybrid switching standard based on
FER of LDPC decoding is introduced to address the challenge of

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

3
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

UAV-to-Ground FSO channel estimation.

(b)
(a)

(c)

(b)
Fig. 3. MSE performance of (a) increasing SNR of GRU
estimator at different turbulence intensity (b) GRU and LSTM
estimator at weak turbulence (c) different hidden size of GRU
and LSTM. Fig. 4.(a) Training flops (b) Training time of GRU and LSTM
estimator with different hidden size.
The Fig. 4(a) provides a comparison of the flops calculation
B. Hybrid switching standard
for channel estimation between GRU and LSTM models at
different hidden sizes. It is evident that the GRU model
requires fewer computational resources during training in
comparison to LSTM, and this difference becomes more
pronounced as the hidden size increases. These results indicate
that GRU is more computationally efficient than LSTM,
especially during multiple training iterations, making it a more
resource-saving option.
Additionally, as shown in The Fig. 4(b), there is a
significant disparity in training time between GRU and LSTM
for various hidden sizes. As the hidden size increases, the
training time of GRU exhibits a steady rise, whereas the
training time of LSTM demonstrates a sharp upward trend.
Moreover, the training time of GRU remains consistently
below 1.7 hours, while the training time of LSTM exceeds 1.7
hours at a hidden size of 140. Furthermore, at a hidden size of
Fig. 5. BER and Bits Efficiency versus SNR with different
120, the training time of GRU is 65.1% of the training time of
coding and modulation order under medium turbulence.
LSTM, despite both models exhibiting similar performance.
Hence, based on the observations from Fig. 3 and Fig. 4, it can
The results depicted in the Fig. 5 illustrate the performance
be concluded that the GRU estimator outperforms the LSTM
of conventional adaptive coding and modulation scheme with
estimator in terms of computational resource consumption and
a single switching standard under medium turbulence intensity.
training time, while demonstrating similar performance in

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

4
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

This ACM scheme , as the blue curve shows, enhances the bit changing turbulent channels.
efficiency of FSO transmission systems based on the
prevailing channel conditions. When the received SNR
degrease below the requirements of the current coding and
modulation scheme, the scheme progressively reduces the
coding and modulation index until the desired SNR is
achieved. Conversely, in prefect channel conditions, the ACM
scheme will increase the coding and modulation index as long
as the BER under FEC threshold. As illustrated in the figure,
the bit efficiency (R) of the ACM scheme improves from 0.5
bits/symbol (poor channel conditions) to 1.6 bits/symbol
(perfect channel conditions), while ensuring that the maximum
average BER of the channel is lower than the FEC threshold
(3.8×10-3).

Fig. 7. SNR difference of ACM scheme and ACM scheme

with FER switching compared with ideal ACM scheme under
different turbulence.

The ACM scheme from medium turbulence, directly

transplanted or adopted with FER switching standard, is
transplanted to other turbulence intensity scenarios as
illustrated in the Fig. 7. It can be observed that the SNR
difference between the directly transplanted ACM scheme and
the ideal ACM scheme under the current turbulence is positive
when the turbulence is lower than that of the original ACM
scheme. This indicates a delay in the coding and modulation
Fig. 6. Frame Error Rate versus SNR with different coding index switching, resulting in increased waste in bits efficiency
and modulation order under medium turbulence. as the SNR decreases. Conversely, when the turbulence
intensity is higher than that of the original ACM scheme, the
To obtain the upper and lower threshold values for LDPC SNR difference is negative. This suggests that the coding and
decoding error rate switching in the Hybrid Switching modulation index switching will occur earlier. The larger the
Standard, the LDPC decoding error rate at the receiving end absolute value of the interpolation as the turbulence increases,
under moderate turbulence is shown in the Fig. 6. It can be the more likely it is to cause BER to exceed the FEC threshold,
observed that with the established ACM scheme, the LDPC leading to communication interruptions.
decoding error rate remains within a certain range. To allow In contrast, the ACM scheme with FER switching standard,
for some redundancy in the switching scheme, significant when compared with the ideal scheme under different
margins are reserved at both the maximum and minimum turbulence conditions, exhibits relatively stable SNR
frame error rates. Subsequently, upper and lower thresholds of differences, averaging around 0.35 dB. Due to the inclusion of
1×10-1 and 4×10-2 were chosen [27]. a certain redundancy in selecting the LDPC decoding FER
In other words, when there is a change in turbulence switching threshold, the SNR difference will be slightly
intensity, the LDPC decoding error rate of current greater than zero in turbulence different from the original
communication may surpass the upper or lower thresholds. if scheme, indicating a slight delay in switching. In summary,
the LDPC decoding error rate exceeds the upper threshold, to the conventional ACM scheme, when directly transplanted to
maintain the current communication quality, the coding and different turbulence scenarios, leads to waste in bits efficiency
modulation index should be reduced until it meets the or communication interruptions. Moreover, as the difference
threshold. Similarly, if the decoding error rate falls below the from the original turbulence increases, these issues are
lower threshold, the coding and modulation index should be exacerbated. However, the ACM scheme with FER switching
increased until it meets the threshold. This LDPC decoding standard effectively addresses the transplanted issues of the
error rate-based switching criterion, apart from SNR, enables ACM scheme under different turbulence conditions, with an
the effective transfer of an ACM scheme obtained under average cost of around 0.35 dB. This approach aligns better
specific turbulence conditions to different turbulence with the rapidly changing atmospheric channels caused by
intensities, thus catering to the requirements of rapidly turbulence.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

5
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

communication interruption. However, with the adoption of

(a) FER threshold switching, the sensitivity of the ACM scheme
improves by an average of approximately 4 dB. The ACM
scheme is able to delay the switch and thereby meet the FEC
threshold, ensuring regular communication and achieving a
relatively ideal communication quality.
The traditional ACM scheme has advantages in bits
efficiency compared to fixed coding and modulation order.
However, due to the presence of turbulence, the ACM scheme
developed for moderate turbulence may result in waste of bits
efficiency or communication interruptions when used under
different turbulence intensities. In this paper, an ACM scheme
with hybrid switching standards is proposed, utilizing LDPC
decoding error rate determination at the receiver. This
approach enables the ACM scheme to be effectively
transplanted to different turbulence intensities at the cost of
about 0.35 dB. It enhances communication quality and speed,
(b) making it more suitable for atmospheric channels with rapidly
changing turbulence conditions.

V. CONCLUSIONS
In this paper, an improved ACM scheme with hybrid
switching standard is proposed for UAV-to-Ground FSO
channels to address the challenges associated with portability
across different turbulence conditions, which is validated by
simulation. It is observed that the GRU estimator achieves
approximately 34.9% improvement in training time compared
to the LSTM estimator, without any degradation in system
performance. Additionally, a hybrid switching standard based
on FER of LDPC decoding is introduced. This standard
significantly improves the SNR switching threshold deviation
of the ACM scheme from 1.8 dB and 4 dB, which were
Fig. 8. BER and Bits Efficiency versus SNR with different
observed in moderate turbulence conditions, to an average of
coding and modulation order under (a) weak turbulence (b)
0.35 dB. This improvement allows for the effective
strong turbulence.
implementation of the ACM scheme in different turbulence
intensities at a relatively small cost, thereby enhancing the
Fig. 8 (a) illustrates the specific scenario of weak turbulence communication quality of FSO systems in atmospheric
(0.5) from Fig. 7. It can be observed that under weak channels characterized by turbulent variations.
turbulence, the same coding and modulation order exhibit The ACM scheme proposed in this paper is implemented
better BER performance. Therefore, directly transplanting the based on IM/DD. With some modifications, this system can
ACM scheme designed for moderate turbulence to weak also be applied to coherent communication. For instance, in
turbulence leads to a delay in switching process. In other PSK modulation, more complex switches between modulation
words, it fails to timely switch to a higher coding and formats, such as QPSK, 8PSK, 16PSK, or even higher-order
modulation index that satisfies the FEC threshold, resulting in modulation schemes, can be implemented. Additionally, after
waste of bits efficiency. However, with the adoption of FER optimizing the system to facilitate an increase in DSP speed,
threshold switching, the sensitivity of the ACM scheme there is potential for higher communication bit rates. These
improves by an average of 1.8 dB, enabling an earlier switch advancements will be explored in future research.
in coding and modulation index. This improves bits efficiency
to a certain extent while meeting the FEC threshold and ACKNOWLEDGMENT
enhances the communication rate. This work was supported in part by the National Key
Fig. 8 (b) depicts the specific scenario of strong turbulence Research and Development Program of China
(3.5) from Fig. Similarly, the same coding and modulation (2021YFB2900800); Science and Technology Commission of
format exhibit poorer BER performance under strong Shanghai Municipality (22511100902, 22511100502,
turbulence. When directly transplanting the ACM scheme 20511102400); 111 Project (D20031).
designed for moderate turbulence to strong turbulence, it leads
to the issue of switching to a coding and modulation scheme
that does not meet the FEC threshold, resulting in

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2023.3329648

6
> REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) <

REFERENCES [24]Z. Ghassemlooy, W. Popoola, S. Rajbhandari, Optical Wireless

Communications System and Channel Modelling with MATLAB, 131-
[1] H. Kaushal, G. Kaddoum, Optical Communication in Space: Challenges 134.
and Mitigation Techniques, IEEE Commun. Surv. and Tutor. 19 (2017) [25]K. Greff, R. K. Srivastava, J. Koutník, B. R. Steunebrink, J. Schmidhuber,
57-96. LSTM: A Search Space Odyssey, IEEE Trans. on N.a. Netw. and Learn.
[2] H. Huang, Y. Peng, Y. Lin, W. Fan, Y. Chen, H. Lu, 5G NR fiber-FSO- Syst. 28 (2017) 2222-2232.
wireless systems with dual-polarization and single-carrier optical [26]Sonali, A. Dixit, V. K. Jain, SNR-and Rate-Optimized LDPC Codes for
modulation schemes, Opt. Commun. 530 (2023). Free-Space Optical Channels, IEEE Access, 9(2021) 13212-13223.
[3] H. Al-Mohammed, M. Al-Ali, E. Yaacoub, FSO communication system [27]Z. Zhong, Y. Huang, Z. Zhang, X. You, C. Zhang, A Flexible and High
for high-speed trains under varying visibility conditions, Veh. Commun. Parallel Permutation Network for 5G LDPC Decoders, IEEE Transactions
43 (2023). on Circuits and Systems II: Express Briefs, 67(2020) 3018-3022.
[4] H. Liang, Y. Li, M. Miao, C. Gao, X. Li, Analysis of selection combining
hybrid FSO/RF systems considering physical layer security and
interference, Opt. Commun. 497 (2021).
[5] R. Lu, J. Wang, X. Fu, S. Lin, Q. Wang, B. Zhang, Performance analysis
and optimization for UAV-based FSO communication systems, Phys.
Commun. 51 (2022).
[6] M. Mohsin, I. A. Murdas, Performance analysis of an outdoor Li-Fi
system-based AO-OFDM architecture under different FSO turbulence and
weather conditions, Opt. 273 (2023).
[7] B. Dutta, B. Kuiri, R. Atta, N. Sarkar, A. Patra, Numerical evaluation of
bidirectional high-speed data transmission over turbulence tolerable FSO
link employing WDM-OAM multiplexing and DP-QPSK modulation
techniques, Opt. Commun. 546 (2023).
[8] M. A. Esmail, Experimental performance evaluation of weak turbulence
channel models for FSO links, Opt. Commun. 486 (2021).
[9] Y. Bian, Y. Li, D. Zheng, W. Li, X. Hong, J. Qiu, J. Wu, Low-complexity
parallel real-valued weight adaptive digital combining algorithm for
coherent FSO communications employing modes diversity reception
under atmospheric turbulence channel, Opt. Commun. 474 (2020).
[10]B. K. Saw, V. Janyani, G. Singh, Analysis of coherent-OFDM modulation
for FSO communication over distribution with pointing errors, Opt.
Commun. 545 (2023).
[11]E. Erdogan, I. Altunbas, G. K. Kurt, H. Yanikomeroglu, The Secrecy
Comparison of RF and FSO Eavesdropping Attacks in Mixed RF-FSO
Relay Networks, IEEE Photonics J. 14 (2022). 1-8.
[12]C. H. Yeh, Y. R. Xie, C. M. Luo, C. W. Chow, Integration of FSO Traffic
in Ring-Topology Bidirectional Fiber Access Network With Fault
Protection, IEEE Commun. Lett. 24 (2020) 589-592.
[13]R. Samy, H. C. Yang, T. Rakia and M. S. Alouini, Ergodic Capacity
Analysis of Satellite Communication Systems With SAG-FSO/SH-
FSO/RF Transmission, IEEE Photonics J. 14 (2022) 1-9.
[14]V. Mai, H. Kim, Beam Steering and Divergence Control Using Variable
Focus Liquid Lenses for WDM FSO Communications, IEEE Photonics
Technol. Lett. 34 (2022) 1226-1229.
[15]M. J. Saber, J. Mazloum, A. M. Sazdar, A. Keshavarz, M. J. Piran, On
Secure Mixed RF-FSO Decode-and-Forward Relaying Systems With
Energy Harvesting, IEEE Sys. J. 14 (2019) 4402-4405.
[16]C. Abou-Rjeily, Packet Unloading Strategies for Buffer-Aided Multiuser
Mixed RF/FSO Relaying, IEEE Wirel. Commun. Lett. 9 (2020) 1051-
1055.
[17]R. Li, T. Chen, L. Fan, A. Dang, Performance Analysis of a Multiuser
Dual-Hop Amplify-and-Forward Relay System With FSO/RF Links, J. of
Opt. Commun. and Netw. 11 (2019) 362-370.
[18]S. Song, Y. Liu, T. Xu, L. Guo, Hybrid FSO/RF System Using Intelligent
Power Control and Link Switching, IEEE Photonics Technol. Lett. 33
(2021) 1018-1021.
[19]H. Zhou, S. Mao, P. Agrawal, Optical power allocation for adaptive
transmissions in wavelength-division multiplexing free space optical
networks, Digit. Commun. Netw. 1 (2015) 171–180.
[20]S. H , S. A , D.T , Adaptive Channel Coding and Power Control for
Practical FSO Communication Systems Under Channel Estimation Error,
Trans. on Veh. Technol. 68(2019) 7566-7577.
[21]N. D. Chatzidiamantis, A. S. Lioumpas, G. K. Karagiannidis, S. Arnon,
Adaptive Subcarrier PSK Intensity Modulation in Free Space Optical
Systems, IEEE Trans. on Commun. 59 (2011) 1368-1377.
[22]N. Alshaer, T. Ismail, M. E. Nasr, Enhancing earth-to-satellite FSO
system spectrum efficiency with adaptive M-ary PSK and SIMO in
presence of scintillation and beam wander, AEU - Int. J. of Electron. and
Commun. 125 (2020).
[23]I. B. Djordjevic, Adaptive modulation and coding for free-space optical
channels, IEEE/OSA J. Opt. Commun. Netw. 2 (2010) 221–229.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/