Indonesian Journal of Electrical Engineering and Computer Science

Vol. 99, No. 1, Month 2022, pp. 1∼1x

ISSN: 2502-4752, DOI: 10.11591/ijeecs.v99.i1.pp1-1x ❒ 1

CNN Architecture applied for Filter Bank Detection

Hamza Ouamna1 , Zhour Madini1 , Younes Zouine1
1 Department of Electrical and Telecommunication, ISET Laboratory, National School of Applied Sciences (ENSA),
Ibn Tofail University (UIT), Kenitra, Morocco

Article Info ABSTRACT

Article history: The connected cars industry is rapidly expanding, with more and more vehicles
becoming connected to the internet and each other. This growth has led to the
Received month dd, yyyy
development of cognitive wireless communications, which are designed to im-
Revised month dd, yyyy prove the communication and connectivity of connected cars. One of the key
Accepted month dd, yyyy drivers behind this development is the need to detect the presence of primary
users, such as other vehicles, in the communication spectrum. To achieve this,
Keywords: various techniques are used, including Energy Level Detection, Cyclostationary
Feature Detection, and Filter Bank Detection. However, with the need for a dy-
First keyword namic algorithm that can adapt to changing conditions, the use of Deep learning
Cognitive radio has become increasingly important.
Deep learning This paper focuses on the use of CNN (Convolutional Neural Network) archi-
CNN tecture combined with Filter Bank method for detecting primary user presence
Energy Level Detection in connected cars. Different modulation techniques such as BPSK, QAM, FSK,
BPSK and GMSK are used to evaluate the performance of the proposed algorithm. The
QPSK use of CNN in this context allows for the detection of primary user presence to
be more accurate and efficient, as it can learn and adapt to different signal char-
acteristics. The Filter Bank method, on the other hand, allows for the separation
of different frequency bands, which improves the performance of the detection
process. Overall, the combination of these two methods provides a powerful so-
lution for detecting primary user presence in connected cars, which is essential
for improving the performance and safety of these vehicles.

This is an open access article under the CC BY-SA license.

Corresponding Author:
Hamza Ouamna
Department of Electrical and Telecommunication, ISET Laboratory, National School of Applied Sciences
Ibn Tofail University
campus universitaire, B.P 241, 14000, Kénitra, Morocco
Email: Hamza.ouamna1997@gmail.com

1. INTRODUCTION
The automotive industry is rapidly evolving with the incorporation of advanced technologies such as
autonomous vehicles, cameras, lidar, and radar. However, a new concept that has been gaining attention since
2010 is that of connected vehicles. Connected vehicles aim to enhance the safety of road users and pedestri-
ans, improve traffic management and provide infotainment services. These connections can be grouped into
categories such as vehicle-to-pedestrian (V2P), vehicle-to-infrastructure (V2I), and vehicle-to-vehicle (V2V).
To ensure proper communication and connectivity, connected vehicles operate within a bandwidth
of 75 MHz allocated by the Federal Communications Commission (FCC) in the 5.9 GHz spectrum band for
wireless access in vehicular environments (WAVE) applications. However, with the increasing demand for
diverse vehicular applications related to safety and infotainment, there is a potential for a shortage in spectral

Journal homepage: http://ijeecs.iaescore.com

2 ❒ ISSN: 2502-4752

resources for connected vehicles. To address this issue, cognitive radio is emerging as a potential solution
through the use of dynamic spectrum access (DSA) instead of fixed spectrum allocation (FSA). DSA allows
for the sharing and efficient use of the available spectrum resources among different users, this way it can be
used to address the spectrum scarcity issue.
This work focuses on the use of deep learning, specifically the Convolutional Neural Network (CNN)
architecture, combined with the Filter Bank method, using different types of modulation such as Binary Phase-
Shift Keying (BPSK), Quadrature Amplitude Modulation (QAM), Frequency-Shift Keying (FSK), and Gaus-
sian Minimum Shift Keying (GMSK). A dynamic threshold is also implemented to ensure a fixed probability
of detection. By using deep learning and the filter bank method, we can improve the accuracy of the detection
and classification of the signals, this way we can optimize the use of the spectrum resources. Additionally, the
dynamic threshold allows us to set a desired level of performance for the system and ensures that we are able
to detect signals of interest while minimizing false alarms.
The proposed solution has the potential to improve and address the spectrum scarcity issue through
the use of DSA. With this approach, we can optimize the use of the spectrum resources and make the most of
the available bandwidth, this way we can improve the communication and connectivity of connected vehicles.
The results of this work can be used in various applications such as wireless communication systems, cognitive
radio, and spectrum sensing. In the next section, we will discuss various spectrum sensing techniques, including
our proposed algorithm that employs filter bank detection and deep learning.

2. METHOD
The study is based on simulations using Matlab R2020b. It allows for exploration of different scenarios
and parameters. Results obtained from simulations may not be directly applicable to real-world situations and
should be validated with experimental data.

2.1. Spectrum Sensing Algorithms

in vehicular communication spectrum sensing is widely used to know the occupancy of a ressource
block or channel, and due to the nodes mobility this task is relatively complex to do, in general the vehicle
monitors the characteristics of the received signals which may include Radio Frequency energy levels. To this
aim, many signal detection techniques are used in sensing [2].

2.1.1. Energy Level Detection

Energy level detection is the simplest technique in term of implementation and ressources usage. the
aim of this technique is to transfrom the received signal from time domain to frequency domain by FFT (Fast
Fourier Transform), then the vehicle estimate the PSD (Power spectrql density) and by defining a threshold the
vehicle can get the occuancy of each channel to decide which one to use [2].

N
X
E= |x(n)|2 (1)
n=1

E <= λ, E >= λ (2)

2.1.2. Cyclostationary Feature Detection

Cyclostationary Feature Detection (CFD) can be used to detect primary users (PU) using periodicity
in autocorrelation of the modulated signals. These algorithms attempt to differentiate signal from noise based
on the uncorrelated nature of noise. This technique is used at very low SNR detection, robust to noise, performs
better than energy detector but has more computational complexity.

2.1.3. Matched Filter Bank Detection

A matched filter is a linear filter which provides maximum Signal-to-Noise Ratio with for a given
transmitted signal [2] [4].
The fig. 1 shows the detection using matched filter in which the received signal is band passed,
correlated with the time shifted threshold value (λ).If the output is greater than λ, H1 follows and vice-versa.
The matched filter operation is expressed as:

Indonesian J Elec Eng & Comp Sci, Vol. 99, No. 1, Month 2022: 1–1x
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 3

Figure 1. Block Diagram of Filter Bank Detection.

N
X
E= h(n − k)x(n) (3)
k=−∞

Where ’x’ is the unknown signal convolved with the impulse response of the matched filter ’h’. This
technique is far better than Energy Detection technique [27][28] [29] [30].

2.2. Deep learning applied in Cognitive radio

Deep learning is being widely used for classification problems. Deep Neural Network has high accu-
racy rate in contrast of other machine learning algorithms. One particular architecture for Deep Neural Network
is MultiLayer Perception where all the layers are densely connected to each other. We are going to give a brief
overview of the technique in this section [3] [16] .

2.2.1. Overview
Neural network is a mapping function that maps input vector to an output vector. A neural network is
consisted of different layers. Most basic layers are input layer hidden layer and output layer.

Figure 2. Multi-Layer Perception Neural Network.

2.3. Filter Bank method using deep learning

The neural network approach is different from conventional algorithms in that one cannot analytically
write the detection method and trace input to output [7][8]. for the studied cases we have proposed a neural
network architecture with the following [24] [13] [14][15] [17]:

2.3.1. Dense layers

In any neural network, a dense layer is a layer that is deeply connected with its preceding layer which
means the neurons of the layer are connected to every neuron of its preceding layer. This layer is the most com-
monly used layer in artificial neural network networks. The dense layer’s neuron in a model receives output
from every neuron of its preceding layer, where neurons of the dense layer perform matrix-vector multiplica-
tion. Matrix vector multiplication is a procedure where the row vector of the output from the preceding layers
is equal to the column vector of the dense layer. The general rule of matrix-vector multiplication is that the row
vector must have as many columns like the column vector [23].

CNN Architecture applied for Filter Bank Detection (Hamza Ouamna)

4 ❒ ISSN: 2502-4752

2.3.2. Activation functions

These are the sigmoid activation function, the Rectified linear unit (ReLU) function and the leaky
ReLU function. Leaky ReLU works similar to the ReLU function, but this function addresses the dying ReLU
problem.Leaky ReLU function speeds up the training process, as it is more balanced than the ReLU function.
We preferred the leaky ReLU function for the above said reasons [23][24].
2.3.3. Optimizer
Adam optimizer is a popular optimization algorithm, it is a type of stochastic optimization method
which uses the gradient information of the parameters to adapt the learning rate for each parameter. It is an
efficient optimizer in term of low usage of memory and computation resources. In our implementation, we
have used the Adam optimizer as it is a well-suited optimization method for deep learning tasks. The Adam
optimizer uses the first and second moment of the gradient to estimate the parameters. This adaptive moment
estimation technique helps the Adam optimizer to converge faster and achieve better performance as compared
to other optimization methods.[25].
2.3.4. Regularizer
The most common problem with any neural network training is the overfitting or underfitting. Over-
fitting is when the model learns the training data too well and bends the curve to accommodate every instances
along with the noise, while underfitting is the problem, when the model learns too less. In both situation the
model predict ineffciently for the unforeseen data. To overcome this problem we need to introduce regularizers
[26].

Figure 3. Neural Network Architecture.

2.4. Dataset Description

The signals that have been captured from the Ka-band used two types of modulation BPSK, QAM and
QPSK. BPSK is the kind of modulation where the signal uses binary phase shift. This modulation technique
uses 90° phase difference to differentiate high and low voltage signal. QPSK uses quadrature phase shift to
modulate the signal. Table below shows the properties of the dataset.
2.5. Implementation
In our implementation, we have employed a neural network with a unique architecture to tackle a
specific problem[9] [10]. The neural network was trained using a dataset of IQ samples that were collected
and preprocessed previously [18][19]. The dataset is composed of a set of 1024*2 IQ samples, providing a
large and diverse set of data for the neural network to learn from. The architecture of the neural network was
carefully designed to extract relevant features from the IQ samples and make accurate predictions. During
the training process, the neural network was continuously updated and fine-tuned to improve its performance.
The final model is capable of making accurate predictions on new IQ samples and can be used in various
applications[20][21][22].

Indonesian J Elec Eng & Comp Sci, Vol. 99, No. 1, Month 2022: 1–1x
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 5

Figure 4. Implementation of filter bank method using simulink for BPSK.

CNN Architecture applied for Filter Bank Detection (Hamza Ouamna)

6 ❒ ISSN: 2502-4752

Table 1. Neural network Architecture

Index Layers Output dimension
1 imageInputLayer 512x1
2 convolution2dLayer 512x1
3 batchNormalizationLayer 512x1
4 reluLayer 256x1
5 maxPooling2dLayer (64) 128x1
6 convolution2dLayer (64) 128x1
7 batchNormalizationLayer (128) 64x1
8 reluLayer 64x1
9 maxPooling2dLayer 64x1
10 convolution2dLayer 64x1
11 batchNormalizationLayer 1x256
12 reluLayer 1x32
13 maxPooling2dLayer 1x32
14 convolution2dLayer 1x2
15 batchNormalizationLayer 1x2
16 reluLayer 1x2
17 maxPooling2dLayer 1x2
18 convolution2dLayer 1x2
19 batchNormalizationLayer 1x2
20 reluLayer 1x2
21 maxPooling2dLayer 1x2
22 convolution2dLayer 1x2
23 batchNormalizationLayer 1x2
24 reluLayer 1x2
25 maxPooling2dLayer 1x2
26 convolution2dLayer 1x2
27 batchNormalizationLayer 1x2
28 reluLayer 1x2
29 maxPooling2dLayer 1x2
30 convolution2dLayer 1x2
31 batchNormalizationLayer 1x2
32 reluLayer 1x2
33 averagePooling2dLayer 1x2
34 fullyConnectedLayer 1x2
35 softmaxLayer 1x2
36 classificationLayer 1x2

Indonesian J Elec Eng & Comp Sci, Vol. 99, No. 1, Month 2022: 1–1x
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 7

2.6. Training process

Training was performed on the intel IRIS(R) Xe Graphics. All parameters were randomly initialized
with a Gaussian distribution. The mini-batch size was set to 128 during training, and SGD with momentum
was used as the training method with a momentum factor of 0.9. The initial learning rate was 0.02. Every 3
epochs, the learning rate was reduced to 1/10 of the previous value. The network was trained with 10 epochs
and verification was performed every 500 iterations.

Figure 5. Training.

2.7. Results and Discussion

2.7.1. Metrics for Performance Evaluation
To evaluate the classification performance of our model, we have proposed to use the probability
of detection as a metric. The probability of detection is a commonly used metric in signal processing and
machine learning to evaluate the performance of a classifier. It measures the ability of the model to correctly
identify the presence of a signal in a noisy environment. We have chosen this metric because it can provide
a comprehensive evaluation of the model’s performance for different values of Signal-to-Noise Ratio (SNR).
By varying the SNR, we can simulate different levels of noise in the signal and observe how well the model
performs under these conditions. By using the probability of detection, we can gain insight into the robustness
of the model and identify areas for improvement. Additionally, the probability of detection is a useful metric
for comparing the performance of our model to other models or traditional approaches used in the field.
2.7.2. Experimental Results
The noise floor variance in our system follows the normal distribution, as illustrated in the figures
above. To take advantage of this property, we will extract the threshold for total spectrum and bandwidth for
a desired probability of detection. The threshold is a critical parameter in determining the performance of our
system, as it determines the sensitivity of the detector. By extracting the threshold from the normal distribution,
we can determine the point at which a signal will be detected with a high probability. This allows us to set a
desired level of performance for our system and ensures that we are able to detect signals of interest while
minimizing false alarms. Furthermore, by extracting the threshold for the total spectrum and bandwidth, we
can optimize the performance of our system for different signal types and noise environments. This approach
allows us to achieve the best possible performance for our system and is a common method used in signal
processing and telecommunications systems.

CNN Architecture applied for Filter Bank Detection (Hamza Ouamna)

8 ❒ ISSN: 2502-4752

Figure 6. 2% False Alarm Probability Threshold of Total Noise Power

Figure 7. 2% False Alarm Probability Threshold of Bandwith Noise Power

Indonesian J Elec Eng & Comp Sci, Vol. 99, No. 1, Month 2022: 1–1x
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 9

Figure 8. BPSK

Figure 9. 16-QAM

Figure 10. 32-QAM

CNN Architecture applied for Filter Bank Detection (Hamza Ouamna)

10 ❒ ISSN: 2502-4752

Figure 11. 2-FSK

Figure 12. 4-FSK

Figure 13. GMSK

The neural network defined in Table 2 has been trained with the Adam optimizer. The Adam optimizer
is a popular algorithm for training neural networks because it combines the advantages of both gradient descent

Indonesian J Elec Eng & Comp Sci, Vol. 99, No. 1, Month 2022: 1–1x
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 11

and momentum-based optimization. It is computationally efficient and has been shown to converge quickly to
a good solution. We have used categorical cross-entropy as the loss function. The categorical cross-entropy
is an appropriate loss function for this problem because it measures the dissimilarity between the predicted
and true categorical probability distributions. To ensure that the model generalizes well, we divided the IQ
dataset into 70% for training instances and 30% for validation instances. This allows us to evaluate the model’s
performance on unseen data and make adjustments to the model if necessary. This division of the dataset also
allows us to monitor the overfitting, which is common in neural network models.

To measure the performance of our model, we have plotted the probability of detection versus Signal-
to-Noise Ratio (SNR) predicted by the neural network. This allows us to evaluate the ability of the model
to correctly identify the presence of a signal in a noisy environment. We have also plotted the probability of
detection versus SNR predicted using the Energy level method. This is a traditional method used in signal
processing to evaluate the performance of a classifier. By comparing the performance of the neural network
and the Energy level method, we can gain insight into the effectiveness of our model and identify areas for
improvement. The figures show the plots for the two methods, which allow us to compare them visually and
understand the performance of our model, this way we can have a better understanding of our model and eval-
uate if it is suitable for our application.

The filter bank detection using the neural network provides an improvement in the probability of de-
tection for various modulation schemes such as BPSK, 16-QAM, 2-FSK, 4-FSK and GMSK. This demonstrates
the effectiveness of our model in detecting signals in a noisy environment. However, in the case of 32-QAM
modulation, the probability of detection is slightly different from the energy level detection. This difference
may be due to the complexity of the 32-QAM modulation scheme, which is more challenging to detect than the
other modulation schemes. Despite this difference, the neural network still performs well in detecting 32-QAM
signals. Overall, the results suggest that the filter bank detection using the neural network is an effective method
for detecting signals in a noisy environment and can provide an improvement in the probability of detection for
various modulation schemes.

3. CONCLUSION
In conclusion, this study has presented an implementation of the filter bank detection method using
deep learning to identify signal modulation. The proposed convolutional neural network (CNN) architecture
has been shown to be an effective and efficient method for signal modulation detection. The simulation results
demonstrate that the proposed CNN architecture outperforms conventional energy detection of signals by 5
dB to 10 dB depending on the modulation type. This is a significant improvement, and it demonstrates the
potential of using deep learning methods in signal processing and telecommunications. The proposed CNN
architecture can be applied to other similar problems and can be used in various applications such as wireless
communication systems, cognitive radio, and spectrum sensing. Overall, this study provides a new approach
for signal modulation detection and opens up new possibilities for research in this field. Future work will focus
on the implementation of the architecture in hardware to measure the performance of the algorithm in real
conditions.

ACKNOWLEDGEMENTS
No funding has been provided for this study.

REFERENCES
[1] D. Chew and A. B. Cooper, ”Spectrum Sensing in Interference and Noise Using Deep Learning,” 2020 54th Annual Conference on
Information Sciences and Systems (CISS), 2020, pp. 1-6, doi: 10.1109/CISS48834.2020.1570617443.
[2] M. Sansoy and A. S. Buttar, ”Spectrum sensing algorithms in Cognitive Radio: A survey,” 2015 IEEE International Conference on
Electrical, Computer and Communication Technologies (ICECCT), 2015, pp. 1-5, doi: 10.1109/ICECCT.2015.7226181.
[3] B. Farhang-Boroujeny, ”Filter Bank Spectrum Sensing for Cognitive Radios,” in IEEE Transactions on Signal Processing, vol. 56,
no. 5, pp. 1801-1811, May 2008, doi: 10.1109/TSP.2007.911490.
[4] arXiv:1909.06020 [eecs] “Spectrum Sensing Based on Deep Learning Classification for Cognitive Radios”

CNN Architecture applied for Filter Bank Detection (Hamza Ouamna)

12 ❒ ISSN: 2502-4752

[5] R. Mahesh, A. P. Vinod, C. Moy and J. Palicot, ”A Low Complexity Reconfigurable Filter Bank Architecture for Spectrum Sensing
in Cognitive Radios,” 2008 3rd International Conference on Cognitive Radio Oriented Wireless Networks and Communications
(CrownCom 2008), Singapore, 2008, pp. 1-6, doi: 10.1109/CROWNCOM.2008.4562506.
[6] S. Haykin, D. J. Thomson and J. H. Reed, ”Spectrum Sensing for Cognitive Radio,” in Proceedings of the IEEE, vol. 97, no. 5, pp.
849-877, May 2009, doi: 10.1109/JPROC.2009.2015711.
[7] T. Hunziker, U. U. Rehman and D. Dahlhaus, ”Spectrum sensing in cognitive radios: Design of DFT filter banks achieving
maximal time-frequency resolution,” 2011 8th International Conference on Information, Communications & Signal Processing,
Singapore, 2011, pp. 1-5, doi: 10.1109/ICICS.2011.6173530.
[8] Trac Duy Tran and T. Q. Nguyen, ”On M-channel linear phase FIR filter banks and application in image compression,” in IEEE
Transactions on Signal Processing, vol. 45, no. 9, pp. 2175-2187, Sept. 1997, doi: 10.1109/78.622942.
[9] T. T. Nguyen, M. Hoffmann, T. Kreul and T. Kaiser, ”Experiments on multi-channel spectrum sensing with filter bank realization
for cognitive LTE-A systems,” 2015 International Conference on Advanced Technologies for Communications (ATC), Ho Chi
Minh City, 2015, pp. 83-587, doi: 10.1109/ATC.2015.7388396.M. Young, The Technical Writer’s Handbook. Mill Valley, CA:
University Science, 1989.
[10] R. Tandra and A. Sahai, ”SNR Walls for Signal Detection,” in IEEE Journal of Selected Topics in Signal Processing, vol. 2, no. 1,
pp. 4-17, Feb. 2008, doi: 10.1109/JSTSP.2007.914879.
[11] R. Tandra and A. Sahai, ”SNR Walls for Feature Detectors,” 2007 2nd IEEE International Symposium on New Frontiers in
Dynamic Spectrum Access Networks, Dublin, 2007, pp. 559-570, doi: 10.1109/DYSPAN.2007.79.
[12] Y. Arjoune and N. Kaabouch, ”On Spectrum Sensing, a Machine Learning Method for Cognitive Radio Systems,” 2019 IEEE
International Conference on Electro Information Technology (EIT), Brookings, SD, USA, 2019, pp. 333-338, doi:
10.1109/EIT.2019.8834099.
[13] MIAH, MD SIPON; Hossain, Mohammad Amzad; Ahmed, Kazi Mowdud; Rahman, Md. Mahbubur; Calhan, Ali; Cicioglu,
Murtaza (2021): Machine Learning-Based Malicious User Detection in Energy Harvested Cognitive Radio-Internet of Things.
TechRxiv. Preprint. https://doi.org/10.36227/techrxiv.13681672.v1
[14] Yang, Kai, Zhitao Huang, Xiang Wang, and Xueqiong Li. 2019. A Blind Spectrum Sensing Method Based on Deep Learning.
Sensors 19. MDPI AG: 2270. https://doi.org/10.3390/s19102270.
[15] Chauhan, V.K., Dahiya, K. & Sharma, A. Problem formulations and solvers in linear SVM: a review. Artif Intell Rev 52, 803–855
(2019). https://doi.org/10.1007/s10462-018-9614-6.
[16] Salahdine F. Spectrum sensing techniques for cognitive radio networks. arXiv. 20171710.02668
[17] Malhotra M., Aulakh I.K., Vig R. A review on energy based spectrum sensing in Cognitive Radio Networks; Proceedings of the
International Conference on Futuristic Trends on Computational Analysis and Knowledge Management (ABLAZE); Noida, India.
25–27 February 2015; pp. 561–565.
[18] Lees W.M., Wunderlich A., Jeavons P.J., Hale P.D., Souryal M.R. Deep learning classification of 3.5-GHz band spectrograms with
applications to spectrum sensing. IEEE Trans. Cogn. Commun. Netw. 2019;5:224–236. doi: 10.1109/TCCN.2019.2899871.
[19] Sri K. Biomedical Signal Analysis for Connected Healthcare. Academic Press; Cambridge, MA, USA: 2021. Advanced analysis of
biomedical signals; pp. 157–222.
[20] Elaanba A., Ridouani M., Hassouni L. Automatic detection Using Deep Convolutional Neural Networks for 11 Abnormal
Positioning of Tubes and Catheters in Chest X-ray Images; Proceedings of the 2021 IEEE World AI IoT Congress (AIIoT); Virtual
Conference. 10–13 May 2021; pp. 7–12.
[21] Rabbah J., Ridouani M., Hassouni L. A New Churn Prediction Model Based on Deep Insight Features Transformation for
Convolution Neural Network Architecture and Stacknet. Int. J. Web Based Learn. Teach. Technol. (IJWLTT) 2022;17:1–18. doi:
10.4018/ijwltt.300342.
[22] Tan M., Le Q. Efficientnet: Rethinking model scaling for convolutional neural networks. Int. Conf. Mach. Learn.
2019;97:6105–6114.
[23] Duchi J., Hazan E., Singer Y. Adaptive subgradient methods for online learning and stochastic optimization. J. Mach. Learn. Res.
2011;12:2121–2159.
[24] Ruder S. An overview of gradient descent optimization algorithms. arXiv. 20161609.04747
[25] Pak M.S., Kim S.H. A review of deep learning in image recognition; Proceedings of the International Conference on Computer
Applications and Information Processing Technology; Kuta Bali, Indonesia. 8–10 August 2017; pp. 1–3.
[26] Usama M., Qadir J., Raza A., Arif H., Yau K.L., Elkhatib Y., Hussain A., Al-Fuqaha A. Unsupervised Machine Learning for
Networking: Techniques, Applications and Research Challenges. IEEE Access. 2019;7:65579–65615. doi:
10.1109/ACCESS.2019.2916648.
[27] Benazzouza S., Ridouani M., Salahdine F., Hayar A. Chaotic Compressive Spectrum Sensing Based on Chebyshev Map for
Cognitive Radio Networks. Symmetry. 2021;13:429. doi: 10.3390/sym13030429.
[28] Salahdine F., El Ghazi H., Kaabouch N., Fihri W.F. Matched filter detection with dynamic threshold for cognitive radio networks.
In Proceedings of the International Conference on Wireless Networks and Mobile Communications (WINCOM); Marrakech,
Morocco. 20–23 October 2015; pp. 1–6.
[29] Salahdine F., Ghribi N.K.E., Kaabouch N. A Cooperative Spectrum Sensing Scheme Based on Compressive Sensing for Cognitive
Radio Networks. Int. J. Digit. Inf. Wirel. Commun. 2019;9:124–136. doi: 10.17781/P002619.
[30] Ridouani M., Hayar A., Haqiq A. Perform sensing and transmission in parallel in cognitive radio systems: Spectrum and energy
efficiency. Digit. Signal Process. 2017;62:65–80. doi: 10.1016/j.dsp.2016.10.016.

Indonesian J Elec Eng & Comp Sci, Vol. 99, No. 1, Month 2022: 1–1x
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 13

BIOGRAPHIES OF AUTHORS
Hamza Ouamna
is an electrical engineer and PhD candidate in the field of spectrum sensing in vehicular environments.
He received their Bachelor’s degree in Electrical Engineering from EST Meknès - Moulay Ismail
University and Master’s degrees in Electrical Engineering and Control of Industrial Systems from
ENSET Mohammedia - Hassan II University Casablanca. currently he is conducting research at Ibn
Tofail University under the guidance of Mr. Younes Zouine and Mrs. Zhour Madini. His research
focuses on developing efficient and robust spectrum sensing techniques for use in vehicular commu-
nications. He published several papers in leading journals and conferences in the field of wireless
communications. He is also actively involved in mentoring undergraduate and graduate students in
their research group. Finaly He aims to contribute to the advancement of the field of vehicular com-
munications through their research and make a positive impact on society.He can be contacted at
email: Hamza.ouamna1997@gmail.com.

Zhour Madini received the B.Sc. degree in computer science from the University of
Hartford, USA, then she received her Master of Engineering degree from the Optical and Microwave
Communications Research Institute, Limoges France. She received her Ph.D degree from University
of Limoges in High Frequency Telecommunications and Optic in 2008. She is now an assistant-
professor in Electronics and Telecommunications at National School of Applied Sciences of kenitra
(ENSAK). Her current research interests include electronic devices, digital signal processing for
communications, fiber-optic communication systems and cognitive radio technology. She can be
contacted at email: zmadini@gmail.com.

Younes Zouine He received his Master of Engineering degree from the National School of
Engineers of Limoges (ENSIL) in 2002. He received his Ph.D in High Frequency Telecommunica-
tions and Optics from the University of Limoges in 2005. He joined the National School of Applied
Sciences of kenitra (ENSAK), Ibn Tofail University, as Professor of Electronics and Telecommuni-
cations. His current research interests deal with digital signal processing for communications and
especially with theory and implementation of signal processing for Optical systems. He can be con-
tacted at email: yszouine@gmail.com.

CNN Architecture applied for Filter Bank Detection (Hamza Ouamna)