Ensemble Learning Approach for Digital Communication Modulation’s Classification

International Journal of Artificial Intelligence and Applications (IJAIA), Vol.15, No.1, January 2024
DOI:10.5121/ijaia.2024.15103 43
ENSEMBLE LEARNING APPROACH FOR DIGITAL
COMMUNICATION MODULATION’S CLASSIFICATION
Yahya Benremdane, Said Jamal ,Oumaima Taheri, Jawad Lakziz
and Said Ouaskit
Faculty of science Ben M’Sik, University Hassan II, Casablanca, Morocco
ABSTRACT
This work uses artificial intelligence to create an automatic solution for the modulation's classification of
various radio signals. This project is a component of a lengthy communications intelligence process that
aims to find an automated method for demodulating, decoding, and deciphering communication signals. As
a result, the work we did involved selecting the database required for supervised deep learning, assessing
the performance of current methods on unprocessed communication signals, and suggesting a deep
learning network-based method that would enable the classification of modulation types with the best
possible ratio between computation time and accuracy. In order to use the current automatic classification
models as a guide, we first conducted study on them. As a result, we suggested an ensemble learning
strategy based on Transformer Neural Network and adjusted ResNet that takes into account the difficulty
of forecasting in low Signal Noise Ratio (SNR) scenarios while also being effective at extracting multi-
scale characteristics from the raw I/Q sequence data. Ultimately, we produced an architecture for
communication signals that is simple to work with and implement. With an accuracy of up to 95%, this
solution's optimum and sturdy architecture decides the type of modulation on its own.
KEYWORDS
Automatic modulation classification, Artificial Intelligence, Deep Learning & Modulation
recognition.
1. INTRODUCTION
Intelligence systems entrusted with keeping an eye on the communications spectrum have a tough
problem when it comes to communication signals, which are characterized by a variety of
modulations to achieve high data rates while limiting interference. When modulations get more
complex, so do the identification and demodulation processes. This is especially true when trying
to extract useful information in the field of Communications Intelligence (COMINT).
Within the COMINT area, the study concentrates on the complex problem of identifying and
categorizing modulations in intercepted signals, where the main goal is to extract relevant
information. This is essential for figuring out the kind of transmission and making the
demodulation procedure easier. Decoding communication signals, whether voice or data, is the
focus of Communications Intelligence (COMINT), as opposed to Electronic Intelligence
(ELINT), which mostly works with radars.
Modulation is a basic mechanism in the communication signals used to transmit information. By
modulating the data into a certain frequency, air attenuation issues are resolved and high-speed
transmission is made possible. The Intelligence process begins with the interception of an
unknown signal in the wide communications spectrum. Frequency and signal strength
measurements are involved. But figuring out the modulation to employ while sending the radio

44
signal is the first and most important step. In the past, intelligence methods required repeatedly
using different demodulators. However, this strategy has been shown to be slow and
unproductive, especially when dealing with contemporary modulations. The development of
Artificial Intelligence (AI) has drastically changed this environment. The wireless research
community has paid close attention to the automatic classification of modulation types at the
receiver, which has the notable benefit of increasing spectrum use efficiency. Early attempts used
Convolutional Neural Network (CNN) architectures and applied spectrogram images produced
by various modulations.
The Inphase and Quadrature signals (I/Q) of a signal also known as the "DNA" of any signal
have been used in recent studies in a novel way. In automatic modulation recognition, I/Q data
has proven to perform better than conventional methods. Any signal consists of two parts,
essentially: the quadrature (Sinus) and the in-phase (Cosine) components. The complex baseband
signal described by these I/Q samples has waveforms denoting its real and imaginary
components, I(t) and Q(t).
The substance of the signal is captured in the whole signal description, X(t) = I(t)+jQ(t), which is
encoded into a matrix with two rows for I and Q. This I/Q-based methodology for modulation
categorization turns out to be a potent tool, offering a thorough and efficient way to interpret the
complex modulations seen in contemporary communication signals.
2. BACKGROUND
Automatic Modulation Classification (AMC) techniques encompass a spectrum of
methodologies, broadly categorized into traditional approaches, where most of them are basically
categorized into the likelihood-based (LB) and feature-based (FB) approaches and advanced
techniques leveraging deep learning.
2.1. Traditional Approaches
2.1.1. Likelihood-Based Methods
In the early days of Automatic Modulation Classification (AMC), most approaches were
likelihood-based. These techniques entail precisely determining likelihood functions for various
modulation kinds. The basic idea is to find the most likely modulation type by comparing the
received signal to a set of specified likelihood functions. In situations where there is both known
and unknown channel information, modulation identification problems are addressed using
likelihood-based techniques, which make use of theoretical models and probability theories [1].
These methods need significant computing complexity for model parameter estimation, even if
they can reach ideal classification accuracy provided complete knowledge of the signal and
channel models is assumed [2], [3].
2.1.2. Feature-Based Approaches
Feature-based methods are a basic method for differentiating modulation patterns in the context
of AMC. This approach, which strikes a practical compromise between classification accuracy
and computing efficiency, is based on feature extraction and classifier construction. The basic
idea is to capture the unique properties of different signals without having to deduce the signal's
likelihood function in great detail. Two crucial steps in the feature-based method are pre-
processing the signal and extracting pertinent features. The signal is then classified using a
classifier that is applied in accordance with these features. The careful selection of signal
characteristics and the development of reliable classifiers are essential components of this

45
approach's success. Real-time applications and resource-constrained contexts benefit greatly from
feature-based approaches, which are especially beneficial in situations when algorithm
complexity must be reduced [4].
Although feature-based methods exhibit adaptability to various channel models, they encounter
significant limitations, including the weak discriminatory capability of manually crafted
features and the constrained learning capacity of conventional classification algorithms [5], [6].
2.2. Advanced Approaches
The remarkable data processing powers of deep learning (DL) have attracted a lot of attention
and have been used in many different fields due to the quick development of Artificial
Intelligence (AI) technology, which includes radio signal processing for communications.
Research on the application of deep learning for AMC is ongoing, with new methods and
architectures being put forth to increase classification accuracy and lower computing complexity.
In fact, using DL to solve traditional feature-based signal classification problems offers a
productive and affordable substitute for AMC. To overcome the current shortcomings of
conventional techniques, a number of deep network-based AMC techniques, including deep
neural networks (DNNs), convolutional neural networks (CNNs) [7], recurrent neural networks
(RNNs), and long short-term memory networks (LSTMs) [8], have been developed.
Nevertheless, the over-fitting problem caused by a significant number of network parameters
may still have an impact on the effectiveness of these deep learning-based AMC techniques [9].
2.3. Ensemble Learning for AMC
With notable performance across a range of domains, ensemble learning has become a potent
paradigm in machine learning. Combining predictions from various models is the idea behind
improving overall performance, which improves accuracy and robustness [10]. Because
ensemble models can handle the complex and dynamic character of communication signals, they
have gained interest in the context of Automatic Modulation Classification (AMC). Ensemble
models incorporate information from several sources, which allows them to capture complex
patterns present in different forms of modulation and different SNR situations [11], [12]. Several
benefits are provided by ensemble models in the context of AMC. They perform well with a
variety of modulation styles, adjust well to changes in signal-to-noise ratio, and yield higher
classification accuracy results. Innovative architectures and approaches are among the latest
developments in ensemble models for AMC. Prominent instances comprise deep learning
ensemble models that utilize architectures like recurrent neural networks (RNNs), convolutional
neural networks (CNNs), and deep neural networks (DNNs). These models show promise for
lowering computational complexity and increasing classification accuracy [13].
Notwithstanding their achievements, there are still difficulties in creating efficient group models
for AMC. Achieving the ideal balance between coherence and model variation is essential.
Furthermore, there is continuous research being conducted to solve overfitting-related concerns
and guarantee that ensemble models are generalizable across various signal circumstances. The
progress and difficulties described in the literature on ensemble models in AMC serve as a
motivation for the ensemble model that this study proposes. In order to take use of
complementary characteristics, ResNet and Transformer neural networks were combined.
Existing ensemble models in AMC have gaps and room for development, as shown by a rigorous
examination. By combining cutting-edge architectures and improving the ensemble learning
procedure for more accurate modulation categorization, the suggested model seeks to close these
gaps.

46
3. THE PROPOSED APPROACH
The new method we present in this section for modulation classification (Figure 1) uses an
ensemble of two powerful neural network models: Transformer Neural Network (TNN) and
Residual Network (ResNet). One model is tuned to accurately predict signals with low SNRs,
while the other model is tuned for signals with high SNRs. The primary objective is to mitigate
the challenges posed by varying Signal-to-Noise Ratios (SNRs) by tailoring each model to
function well in a given SNR scenario. The goal of the ensemble design is to take advantage of
TNN's and ResNet's complementary skills in spatial feature extraction and sequential data
handling and temporal relationships, respectively.
Figure 1. Proposed ensemble model (Resnet with TNN).
3.1. Residual Net: High SNR Model
Microsoft researchers unveiled ResNet, or Residual Network, as a kind of convolutional neural
network (CNN) architecture in 2015 [14]. The use of "residual connections," which enable the
network to learn a residual mapping rather than an explicit mapping from the input to the output,
is the main innovation of ResNet. This allows for much deeper networks to be trained than could
previously be achieved without running into the vanishing gradient problem and with acceptable
performance. ResNet may be applied to AMC as well as achieving state-of-the-art results on a
range of image classification tasks [15].
The residual stack and the residual unit are the two primary components of the ResNet
architecture. The residual stack is a series of residual units, each having several levels within it.
The residual stack is in charge of using the layer inputs to learn a residual function. To achieve
this, first feed the input through the previous layer and then add it to the output of the same layer.
It is in charge of deepening the network and enabling it to pick up intricate details from the
information. The fundamental component of the ResNet architecture is the residual unit. It is
composed of two or more convolutional layers, where the first layer's output is added to the
second layer's input.
In addition to helping to preserve data from the source, this enables the network to pick up a
residual function. In order to improve the network's stability and normalize the convolutional
output, the residual unit additionally contains a batch that normalizes the layer.

47
3.2. Transformer Neural Network: Low SNR Model
Sequence-to-sequence tasks with long-range dependencies can be solved with ease by the TNN
architecture [16]. Transformer models detect the influence and dependency of distant data items
by using a series of mathematical operations called attention or self-attention. The network can
concentrate on the most crucial components of the signal, such as the signal of interest, and
ignore the noise by using attention mechanisms to weight the various input signal components
differently. By creating a weighting coefficient to weight the input to sum up for a specific
objective, this mechanism's primary job is to identify which features in the input are significant
for the target and which ones are not.
There are multiple layers in the Transformer neural network architecture, including encoding and
decoding. Multiple layers of feed-forward and self-attention neural networks make up the
encoder. When generating predictions, the model is able to balance the significance of various
input components thanks to the self-attention process. The output of the self-attention layer is
processed using a feed-forward neural network.
Multiple layers of feed-forward and self-attentional neural networks make up the decoder as well. In
order to stop the model from "peeping" at future tokens in the input sequence while making
predictions, the decoder also employs a technique known as "masked self-attention." Additionally, the
transformer architecture has a Multi-Head Attention mechanism that enhances the model's
comprehension of the input by allowing it to attend to different sections of the input simultaneously. It
is computationally efficient and extremely parallelizable. The following architecture is employed:
• Transformer Block: a 256-node Feed-Forward neural network (FFN) that adds
nonlinearity to the model and boosts its capacity.
• The average of every value in the input tensor is called global average pooling.
• Alpha Dropout (0.3): in order to avoid overfitting, a specific percentage of the activations
are randomly removed. By leaving the mean and variance of the input at their initial
values, it preserves those values.
• Two fully connected networks with an Alpha Dropout of 0.2: Scaled Exponential Linear
Unit, or SeLU, is the activation function used.
Due to its ability to handle sequential data, such as time series, and its demonstrated efficacy in
tasks requiring comprehension of the context and dependencies among many inputs, the
Transformer neural network has been selected for low SNR signals. In fact, our technique
consists of extracting features from a low SNR signal using a transformer encoder, and then using
those features to reconstruct the signal using a transformer decoder. In order to minimize the
error of reconstitution between the input and output signals, the encoder and decoder are
concurrently trained.
3.3. Ensemble Model Implementation
The ensemble model put out in this work makes use of the synergies between Transformer Neural
Network (TNN) and Residual Network (ResNet), two different deep learning architectures. The
purpose of this integration is to take use of TNN's efficiency in managing temporal dependencies
and sequential data, as well as ResNet's skill in capturing spatial information.
ResNet performs exceptionally well at differentiating modulation signals in clean, noise-free
situations because it is optimized for high Signal-to-Noise Ratio (SNR) environments. ResNet's
spatial feature extraction output becomes an essential input for a smooth ensemble integration.
With the help of spatial features, the model is trained to generate predictions. On the other hand,

48
the Transformer Network can capture temporal dependencies because it is designed for low SNR
conditions and can process sequential data with ease.
The TNN's output, enhanced by its self-attention mechanisms, adds predictions to the ensemble
based on signal sequences. Our ensemble model is unique in that it can forecast simultaneously
using both ResNet and TNN. Every model processes the incoming signal independently and
produces a forecast. Then, using the ensemble decision-making mechanism, the final output is
chosen based on which of the two predictions is the maximum. This tactic guarantees that the
ensemble gains from the advantages of both models, offering a reliable and flexible classification
method. The models are jointly optimized during the ensemble's joint training.
This entails combining the decision-making process that chooses the maximum prediction with
the optimization of ResNet and TNN's parameters. The ensemble is a flexible solution since it
can dynamically adjust to the many obstacles presented by various SNR situations. The ensemble
model is improved architecturally to support the decision-making process and the two forecasts.
To preserve the distinct contributions that each ResNet and TNN provide to the overall
classification process, further layers and links are added to help with the information flow
between them. The ultimate output of the suggested ensemble model is chosen by taking into
account the simultaneous forecasts of TNN and ResNet in a unique way. The utilization of a
dynamic technique guarantees the ensemble's resilience and ability to leverage the advantages of
both models, hence improving the modulation classification accuracy under a range of SNR
situations.
4. EXPERIMENTS ANDRESULTS
4.1. Dataset Characteristics
In order to verify the effectiveness of our suggested approach, we have assembled an extensive
dataset comprising both artificial and actual data. This dataset includes both synthetic and
simulated channel effects, and it is thoughtfully constructed to cover a wide variety of modulation
settings.
4.1.1. Dataset Structure
The primary components of the dataset are as follows:
• Synthetic Data: To capture the complexity of real-world communication, our synthetic
dataset includes twenty-four distinct modulation kinds. Among these are, notably, high-order
modulations that are common in high-SNR low-fading channel settings.
• Real Gathered Data: We included 44,876 real-world gathered frames, each of which
represented a distinct modulation at a range of noise levels, to further improve the realism of
our dataset. Real-world noise presents difficulties that represent real-world communications..
4.1.2. Types of Modulation
Our dataset covers a spectrum of modulation types (See examples in figures 2 and 3), including:
• PSK Modulations: QPSK, 8PSK, 16PSK, 32PSK, 16APSK, 32APSK, 64APSK, 128APSK.
• QAM Modulations: 16QAM, 32QAM, 64QAM, 128QAM, 256QAM.
• Others: AM-SSB-WC (Amplitude Modulation - Single Sideband – Wideband Carrier).

49
4.1.3. Synthetic Dataset
The following describe the synthetic dataset:
• SNR Levels: Offering a wide variety of noise situations, each modulation type has 26
levels of Signal-to-Noise Ratio (SNR).
• Frame Composition: 1,024 complicated time-series samples are contained in each of the
4,096 frames that make up each modulation-SNR combination.
• Data Format: In-phase and quadrature (I/Q) floating-point components are used to
represent samples.
Figure 2. FSK2 modulation.
Figure 3. PSK 8 modulation in constellation representation.
4.1.4. Real Dataset
To achieve this goal, we have recorded many real signals via a communications receiver. Then
this data has been labeled with its modulation type and level of noise (SNR).
The real dataset introduces authentic challenges:
• Frame Count: Containing 44,876 frames, each representing different modulations in
thepresence of real-world noise.
• Classification Challenges: The noise component in the real dataset enhances
thedifficulty of modulation classification, reflecting the complexities encountered in
practical applications. Both synthetic and real datasets were thoughtfully merged into a
unified dataset of 2,600,780 samples, ensuring a holistic representation of diverse
modulation scenarios.

50
4.1.5. Technical Implementation
All neural network implementations are constructed using Keras, with Tensorflow serving as the
backend, ensuring a robust and standardized framework for model development and evaluation.
In summary, our dataset composition, structure, and inclusion of both synthetic and real-world
data positions it as a robust foundation for evaluating the performance of our proposed ensemble
model under varied and realistic conditions.
4.2. Results for high SNR (ResNet)
Following a series of tests, it was found that the high signal-to-noise ratio (SNR) dataset could be
classified with nearly perfect accuracy using the ResNet model. The model's maximum accuracy,
which it obtained at 30dB, was 95.9% (Figure 4). However, the categorization across low SNR
signals was very poor (35% for -4 dB). This is caused by the noise effect and is also connected to
specific modulation signals that are undoubtedly harder to categorize because of their unique
signal properties.
This deep learning model is resilient and generalizable for predicting high SNR signals rather
than those in low SNR situations, as evidenced by the consistency of our results across all test
cases (Figure 4). Furthermore, our suggested ResNet-based solution outperforms existing
approaches in terms of accuracy when compared to other state-of-the-art techniques for high SNR
circumstances.
This demonstrates our method's promise as a dependable solution for high SNR automatic
modulation classification challenges.
Figure 4. Resnet accuracy versus SNR levels
4.3. Results for low SNR (Transformer network)
After extensive testing under very low SNR settings, the Transformer Network (TNN)
demonstrated its remarkable performance. With an exceptional accuracy rate of 72.6%, the model
demonstrated a remarkable capacity to both precisely rebuild and identify input signals (figure 5).
This result represents a significant improvement over earlier approaches that found it difficult to
achieve high accuracy rates under low signal-to-noise ratio circumstances. The Transformer
Network is a game-changer in the field of modulation categorization under unfavorable signal-to-
noise scenarios because of its effectiveness in such demanding environments.

51
The Transformer Network's superiority is immediately obvious when compared to the ResNet
model, which had trouble recognizing signals in low SNR situations (which varied from -20 dB
to 0 dB) and only managed a maximum accuracy of 20%. This striking difference emphasizes
how conventional deep learning models, like ResNet, are inherently limited when it comes to
signal processing in low signal-to-noise ratio settings.
The architectural elements of the Transformer Network, including the inclusion of self-attention
mechanisms, are responsible for its resilience in low signal-to-noise ratio situations. These
processes enable the model to successfully filter out noise while choosing focusing on pertinent
components of the incoming signal. The Transformer Network exhibits a distinct resistance to the
difficulties presented by low SNR settings by intelligently attending to considerable portions of
the signal, which leads to a notable improvement in classification accuracy. Promising
implications for real-world applications arise from the Transformer Network's effectiveness in
low SNR environments, especially in communication systems where noise interference is a major
concern. For modulation classification jobs where signal integrity maintenance in low SNR
situations is critical, the model's capacity to maneuver across complex signal environments makes
it a useful tool.
Conclusively, the Transformer Network's exceptional low-signal-to-noise ratio performance,
when combined with its architectural advantages, represents a noteworthy advancement in the
creation of reliable and precise modulation classification models, especially when dealing with
difficult noise-filled communication channels.
Figure 5. TNN accuracy versus SNR levels
4.4. Results of ensemble model
Figure 6 illustrates the experimental findings of the deep ensemble learning model and provides a
thorough understanding of the model's performance throughout a range of Signal-to-Noise Ratios
(SNRs). The selected architecture exhibits superior performance for both low and high SNR
circumstances, consistently outperforming baseline models [18]. When different SNR
circumstances are taken into account, it becomes evident that the suggested ensemble design
excels in achieving higher overall accuracy, which is a noteworthy advantage. Figure 6 shows
how well the model performs in low SNR settings and validates its effectiveness in high SNR
settings.
These results highlight the effectiveness of the ensemble learning strategy in improving the
model's accuracy and stability under a range of SNR circumstances. The ensemble model's
potential to adapt and perform optimally under various signal difficulties is demonstrated by its
persistent ability to outperform individual baseline models. Interestingly, our findings show an
interesting trend:

52
 The ensemble model's classification performance is almost 50% better than the single
baseline model, ResNet, at decreasing signal-to-noise ratios (SNR).
 This significant improvement in performance in low signal-to-noise ratio (SNR)
circumstances emphasizes ensemble learning's innate ability to reduce noise and enhance
classification accuracy in situations where signal clarity is impaired.
 The ensemble model's observed performance has important ramifications for modulation
categorization tasks in real-world communication.
 The model’s ability to maintain high accuracy across a range of SNR conditions positions
it as a robust solution for real-world applications, where signal quality can vary widely.
In conclusion, the ensemble learning model’s superior performance across different SNR levels
signifies its adaptability and resilience in the face of varying signal challenges. These results
strengthen the case for employing ensemble learning as an effective strategy for improving the
stability and accuracy of modulation classification models, particularly in dynamic
communication environments where SNR fluctuations are prevalent.
Figure 6. Ensemble learning model accuracy versus SNR levels
4.5. Advantages in Practical Applications
To elucidate the advantages of our chosen models in practical applications, we consider the
following factors:
4.5.1. ResNet in High SNR Environments:
• Acquiring Spatial Features in depth: ResNet's ability to extract spatial features from
high-dimensional data is the foundation of its efficacy in high Signal-to-Noise Ratio
(SNR) circumstances. The network is able to recognize complex patterns and structures.
ResNet is particularly good at extracting and understanding spatial characteristics in
modulation classification tasks with high SNR and little fading. This capacity is essential
for correctly differentiating modulation signals in settings that are clear and noise-free.
• Robust Signal discrimination in Low Noise Conditions: ResNet exhibits a remarkable
capacity to recognize minor differences in modulation signals in clean surroundings with
high SNR. When signal clarity is critical, the model's ability to maneuver across complex
spatial patterns guarantees a high degree of accuracy in recognizing modulation schemes,
adding to its dependability. ResNet is a solid option for applications where the integrity

53
of the transmitted signal is crucial, including in high-quality communication channels,
because of its robustness in noise-free environments.
• Pertinence in Real-World High-SNR, Low-Fading Channels: ResNet's suitability also
extends to high-SNR, low-fading channel settings found in real life. Because of its
flexibility in handling different signal complexity, it works well in situations where the
signal strength is constantly strong. Because of its flexibility, it can be used more
effectively in communication systems where keeping a high signal-to-noise ratio (SNR)
is crucial. This means that it can function reliably under situations similar to those seen in
stable communication channels.
4.5.2. TNN in Low SNR Environments:
• Processing Sequential Data accurately: In low Signal-to-Noise Ratio (SNR)
conditions, the Transformer Neural Network (TNN) shows promise as a reliable solution
for modulation classification tasks. One of its strongest points is how well it handles
sequential data, which is very useful in situations with low signal-to-noise ratios. The
attention-based architecture of TNN makes it possible for it to precisely analyze
sequential input signals, which facilitates the efficient extraction of temporal
connections.
• Adaptability to Real-World Noisy Communication Channels: Noise- and
interference-filled real-world communication channels can benefit from TNN's capability
for modulation classification under low SNR circumstances. Because of its attention
mechanisms and capacity to handle sequential data. The model's flexibility in these kinds
of noisy communication channels emphasizes its applicability in real-world situations
where signal clarity varies.
• Selective Focus on Relevant Signal Components: The ability of TNN's unique
attention mechanisms to selectively focus on pertinent portions of the input signal gives
the model this ability. TNN's capacity to identify and rank informative signal portions is
useful in low SNR environments, where noise can obscure important signal components.
This narrow focus makes the model more resilient to noise interference and improves its
classification accuracy for modulation schemes in difficult, low signal-to-noise ratio
settings.
5. CONCLUSION
With the advancement of Artificial Intelligence, encompassing Deep Learning, neural networks,
and other technologies, automated modulation classification (AMC), a fundamental component
of communication signal processing, has become more and more important in fields like
cognitive electronic warfare and cognitive radio (CR). Its main objective is to correctly
categorize the received signals' modulated modes. This research presents an end-to-end deep
learning model for classifying modulation signals that integrates the prediction ability of many
features and improves model stability through the use of an ensemble learning network.
Techniques for ensemble learning are frequently used to handle multi-class classification issues
and improve classification accuracy overall. These techniques function by making features more
functional and by highlighting each model. In order to create a solid algorithmic framework with
great adaptability, our method involves utilizing the advantages of two deep learning
architectures, ResNet and Transformer network, and learning from one another. Our tests showed
that the suggested deep ensemble approach achieves good stability and accuracy for both high
and low SNR categorization recognition.

54
REFERENCES
[1] W. Su, J. L. Xu, and M. Zhou, “Real-time modulation classification based on maximum likelihood,”
IEEE Commun. Lett., vol. 12, no. 11,pp. 801–803, Dec. 2008.
[2] J. L. Xu, W. Su, and M. Zhou, “Likelihood-ratio approaches to automatic modulation classification,”
IEEE Trans. Syst., Man, Cybern. C, Appl.Rev., vol. 41, no. 4, pp. 455–469, Jul. 2011.
[3] O. Ozdemir, R. Li, and P. K. Varshney, “Hybrid maximum likelihood modulation classification using
multiple radios,” IEEE Communication. Lett.,vol. 17, no. 10, pp. 1889–1892, Oct. 2013
[4] S. Majhi, R. Gupta, W. Xiang, and S. Glisic, “Hierarchical hypothesis and feature-based blind
modulation classification for linearly modulated signals,” IEEE Trans. Veh. Technol., vol. 66, no. 12,
pp. 11057–11069, Dec. 2017.
[5] V. D. Orlic and M. L. Dukic, “Automatic modulation classification algorithm using higher-order
cumulants under real-world channel conditions,” IEEE Commun. Lett., vol. 13, no. 12, pp. 917–919,
Dec. 2009.
[6] W. Su, “Feature space analysis of modulation classification using very high-order statistics,”
IEEECommun. Lett., vol. 17, no. 9, pp. 1688–1691, Sep. 2013.
[7] T. O’Shea, J. Corgan, and T. Clancy, “Convolutional radio modulation recognition networks,”
inProc. International conference on engineering applications of neural networks, Aberdeen, UK,
pp.213-226, Sept. 2016.
[8] Z. Zhang, HH. Luo, C. Wang, C. Gan, and Y. Xiang, “Automatic modulation classification using
CNN-LSMT based dual-stream structure,” IEEE Transactions on Vehicular Technology, vol. 69, no.
11, pp. 69-70, Nov. 2020.
[9] S. Huang, Y. Jiang, Y. Gao, Z. Feng, and P. Zhang, “Automatic modulation classification using
contrastive fully convolutional network,” IEEE Wireless Communications Letters, vol. 8, no. 4, pp.
1044-1047, Aug. 2019.
[10] Dietterich, Thomas G. ”Ensemble methods in machine learning.” International workshop on
multipleclassifier systems. Berlin, Heidelberg: Springer Berlin Heidelberg, 2000.
[11] Roy, Chirag, et al. ”An Ensemble Deep Learning Model for Automatic Modulation Classification in
5G and Beyond IoT Networks.” Computational Intelligence and Neuroscience 2021 (2021).
[12] Venkata Subbarao, M., and P. Samundiswary. ”Automatic modulation classification using cumulants
and ensemble classifiers.” Advances in VLSI, Signal Processing, Power Electronics,
IoT,Communication and Embedded Systems: Select Proceedings of VSPICE 2020. Springer
Singapore,2021.
[13] Le, Ha-Khanh, Van-Sang Doan, and Van-Phuc Hoang. ”Ensemble of Con-volution Neural Networks
for Improving Automatic Modulation Classification Performance.” (2022)
[14] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc.
IEEEconference on computer vision and pattern recognition, Las Vegas NV, USA, Dec. 2016, pp.
770-778.
[15] H. Zhang, L. Yuan, G. Guangyu Wu, F. Zhou, and Q. Wu, “ Automatic modulation classification
using involution enabled residual networks,” IEEE Wireless Communications Letters, vol. 10, no. 11,
pp. 2417-2420, Nov. 2021.
[16] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. Gomez, L. Kaiser, and I. Sukhin.
[17] “Attention is all you need,” in Proc. Conference on Advances in neural information processing
systems, Long Beach California, USA, pp. 30-31, Dec. 2017.
[18] Y.Benremdane. S.Jamal, O.Taheri, J.Lakziz, and S.Ouaskit, “Communication Signals Modulations
Classification based on Neural Network Algorithms”, in Proc.12th International Conference on Soft
Computing, Artificial Intelligence and Applications (SCAI 2023), Sydney, Australia, pp 263-
276,Dec.2023.
AUTHORS
Mr Yahya BENREMDANE Engineer and Phd Student, Faculty of Science Ben M’sik,
University Hassan II, Casablanca, Morocco As a Telecommunications engineer, Yahya
Benremdane has been working on signal analysis and proprieties of electromagnetic waves.
He’s interested in new technologies of radio transmissions, radars and receiver’s
technologies. Since 2019 he’s preparing a thesis on classification of the modulations of

55
telecommunications based on artificial Intelligence, at the Faculty of Science Ben M’sik in Casablanca.
Ben M’sik in Casablanca.
Mr Said JAMAL Engineer and Phd Student, Faculty of Science Ben M’sik,
UniversityHassan II, Casablanca, Morocco Said JAMAL is an engineer and Phd student
at the university Hassan II. He’s preparing his thesis on the application of AI in the
context of the classification of submarine acoustic waves.
Mrs Oumaima TAHERI Engineer and Phd Student, Faculty of Science Ben
M’sik ,University Hassan II, Casablanca, Morocco Oumaima Taheri holds a Master's in
Modeling and Data Science. With a passion for AI and data-driven solutions, she excels
in applying predictive modeling techniques. Oumaima's expertise lies in navigating
complex challenges through innovative data analysis, making her a valuable asset in the
evolving landscape
Dr Jawad LAKZIZ Phd, Faculty of Science Ben M’sik,University Hassan II,
Casablanca,Morocco Dr Jawad LAKZIZ holds a Phd in Modeling acoustic waves in
submarine environment. He is currently an Associate researcher at the Faculty of Science
Ben M’sik, University Hassan II, Casablanca, Morocco, and guest Editor of the Journal
of Marine Technology and Environment. His research interests are modeling acoustic
waves, and signal processing, including underwater acoustic classification based on IA .
Pr Said OUASKIT Professor, Faculty of Science Ben M’sik,University Hassan
II,Casablanca, Morocco Professor Said OUASKIT is a professor at the Faculty of Science
Ben M’sik, University of Hassan II Casablanca. He’s an expert in experimental physics,
condensed Matter, atomic, molecular and Optical Physics.

Ensemble Learning Approach for Digital Communication Modulation’s Classification

More Related Content

Similar to Ensemble Learning Approach for Digital Communication Modulation’s Classification (20)

More from gerogepatton (20)

Recently uploaded (20)

Ensemble Learning Approach for Digital Communication Modulation’s Classification