electronics

Article
A Deep-Learning-Based Method for Spectrum Sensing with
Multiple Feature Combination
Yixuan Zhang 1 and Zhongqiang Luo 1,2, *

1 School of Automation and Information Engineering, Sichuan University of Science and Engineering,
Yibin 644000, China; [email protected]
2 Artificial Intelligence Key Laboratory of Sichuan Province, Sichuan University of Science and Engineering,
Yibin 644000, China
* Correspondence: [email protected]

Abstract: Cognitive radio networks enable the detection and opportunistic access to an idle spectrum
through spectrum-sensing technologies, thus providing services to secondary users. However, at a
low signal-to-noise ratio (SNR), existing spectrum-sensing methods, such as energy statistics and
cyclostationary detection, tend to fail or become overly complex, limiting their sensing accuracy
in complex application scenarios. In recent years, the integration of deep learning with wireless
communications has shown significant potential. Utilizing neural networks to learn the statistical
characteristics of signals can effectively adapt to the changing communication environment. To
enhance spectrum-sensing performance under low-SNR conditions, this paper proposes a deep-
learning-based spectrum-sensing method that combines multiple signal features, including energy
statistics, power spectrum, cyclostationarity, and I/Q components. The proposed method used
these combined features to form a specific matrix, which was then efficiently learned and detected
through the designed ‘SenseNet’ network. Experimental results showed that at an SNR of −20 dB,
the SenseNet model achieved a 58.8% spectrum-sensing accuracy, which is a 3.3% improvement over
the existing convolutional neural network model.

Keywords: multiple feature combination; deep learning; I/Q components; spectrum sensing;
low SNR

Citation: Zhang, Y.; Luo, Z. A

Deep-Learning-Based Method for
Spectrum Sensing with Multiple 1. Introduction
Feature Combination. Electronics 2024, Over the last ten years, the relentless advancement of communication services and
13, 2705. https://doi.org/10.3390/
the rapid increase in mobile user numbers have catalyzed the introduction of innovative
electronics13142705
wireless mobile applications and services. On one hand, the demand for spectrum resources
Academic Editor: Ricardo Martins is increasing due to the diverse services that require massive connectivity and high-quality
communication [1]. On the other hand, communication systems are constrained by multiple
Received: 24 May 2024
practical factors, such as antenna size, electromagnetic wave properties, device performance,
Revised: 6 July 2024
and transmission and reception power, making the genuinely available spectrum resources
Accepted: 9 July 2024
increasingly scarce and strained [2]. Therefore, enhancing spectrum utilization under the
Published: 10 July 2024
limited and scarce conditions of spectrum resources is a current research focus [3].
In order to address this issue, experts have suggested using cognitive radio and its
spectrum-sensing technology [4]. With a significant focus on secondary users, this method
Copyright: © 2024 by the authors. aims to enhance spectrum usage by opportunistically utilizing unlicensed primary users’
Licensee MDPI, Basel, Switzerland. unused spectrum by detecting their lack of communication activity. Secondary users must
This article is an open access article detect and refrain from using the same frequency bands as primary users while the latter are
distributed under the terms and active, especially in regions where primary user signals are weak. Improving the system’s
conditions of the Creative Commons ability to detect signals in low-SNR conditions can enhance secondary users’ utilization of
Attribution (CC BY) license (https:// peripheral spectrum resources. This will enable them to effectively strategize and regulate
creativecommons.org/licenses/by/ spectrum usage, reducing overall system interference.
4.0/).

Electronics 2024, 13, 2705. https://doi.org/10.3390/electronics13142705 https://www.mdpi.com/journal/electronics

Electronics 2024, 13, 2705 2 of 13

Conventional methods for spectrum sensing primarily encompass energy statistics [5],
cyclostationary detection [6], and matched filter detection [7]. While these techniques are
effective under specific conditions, their general applicability is constrained. For instance,
cyclostationary detection is effective at lower SNR and is considered the most efficient
among the traditional spectrum-sensing techniques. However, it requires prior information
about the primary user’s signal, resulting in longer detection times and increased complex-
ity. Overall, traditional methods are constrained by their application scenarios and face
challenges, such as noise and multipath fading, making extending them to other fields and
scenarios difficult.
In recent years, applying deep learning in various fields has become a focal point and
hot topic among researchers. Deep learning, with its excellent adaptive advantages and
nonlinear modeling capabilities, has been effectively applied in the field of communications,
offering new research ideas and approaches [8]. Deep learning can effectively address the
aforementioned issues, and this paper focuses on deep-learning-based approaches.
Researchers have employed well-known network models, including convolutional
neural network (CNN) [9], long short-term memory network (LSTM) [10], and artificial neu-
ral networks (ANN) [11], to extract different characteristics of signals for spectrum-sensing
tasks. These models have demonstrated superior performance compared to conventional
single-node spectrum-sensing methods. Many deep learning approaches lack awareness
of the underlying structure of the source signal. Most deep learning methods use blind
sensing to recognize the underlying structure of the primary signal. For instance, in [12],
Tekbıyık K and colleagues proposed a deep learning method that calculates the cyclosta-
tionary characteristics of signals through the spectrum correlation function to perform
spectrum sensing on cellular communication data, achieving better performance than the
support vector machine (SVM) model [13]. In the fields of modulation recognition and
Industrial Scientific Medical (ISM), researchers typically use the fast Fourier transform, I/Q
features, and CNN classifiers to identify signals and interference [14].
Current research in spectrum sensing in deep learning focuses on using better net-
works or optimizing existing neural networks in combination with other modules to
further solve problems in this field. In [15], the researchers adopted CNN as the benchmark
model. On this basis, they integrated new transformer networks and bidirectional long
short-term memory to form a novel network, which enhanced the detection probability
under −20 dB. Moreover, the F1 score was superior to that of the original network. Scholar
Salma Benazzouza [16] employed lightweight CNN as the detection network, concur-
rently extracted the time–frequency information of the signal, and transformed it into a
spectrogram as the neural network input. Based on the original approach, this method
demands fewer computing resources and enhances the detection efficiency. In [17], the
researchers proposed an optimized CNN network to learn the underlying structure of
modulated signals extracted by energy detectors, addressing the impact of the SNR wall on
the detection process. The network is based on the convolutional long short-term memory
deep neural network (CLDNN) model and includes a feature fusion layer in the neural
network. Experimental results showed that the optimized model performed better than
typical CNN and deep neural network (DNN) [18], showing stable performance across
different modulation orders. This experiment considered only a single feature, similar to
cyclostationary characteristics in deep learning applications.
In recent years, some scholars have tried to use neural networks to learn various
features of signals to further improve the detection probability under low SNR. Scholars
Dong Han et al. [19] extracted and further preprocessed the energy features and cyclo-
stationary features of the primary user signal and noise and subsequently input them
into CNN. Experimental results demonstrated that the detection probability of the deep
learning method based on two features at −20 dB was higher than that of the deep learning
method based on cyclostationary features. However, in spectrum-sensing tasks that need to
detect transient signals or signals that change with time, CNN’s time correlation extraction
and learning may not be effective. As a fundamental feature of signals, energy is the
Electronics 2024, 13, 2705 3 of 13

primary research focus for scholars. In [20], scholars proposed a hybrid spectrum-sensing
technology based on artificial neural networks. They used the likelihood ratio test statistics
and the energy feature of signals as the training data for the neural networks. The optimal
detection performance of this approach was improved by approximately 60% compared to
traditional energy detection methods. Meanwhile, scholar Yue Geng, in [21], considered
combining multiple signal features as inputs for CNN, achieving good results in the range
from −10 dB to 5 dB, but the general two-dimensional CNN could not effectively extract the
spatial and temporal correlation features of signals. System performance was poor when
the SNR was below −14 dB, suggesting the introduction of gated recurrent units (GRU)
and redesigning the model to further optimize performance. In summary, the above-related
research shows the great potential of deep learning combined with multiple signal features
in spectrum-sensing tasks. By exploring different combinations of signal features and
designing appropriate neural networks, we can solve relevant problems in specific fields
and further improve detection accuracy and robustness.
Based on the above insights and findings in the literature, we propose a deep-learning-
based multi-feature spectrum-sensing method, which focuses on solving the problem of
unstable detection capability of a single feature in low-SNR environments and is char-
acterized by further extracting and learning the underlying information of the signals.
Compared to existing multi-feature spectrum-sensing methods, we further combine the
phase information of the signal, namely the I/Q features, so that the feature matrix can
more comprehensively represent the signal’s information. We then introduce the LSTM
layer to enhance the processing ability of time series information, further improving the
detection probability. The primary contributions of this paper are as follows.
(1) We decompose the co-directional and orthogonal vectors of the signal, process them to
obtain the I/Q features, and combine them with the existing feature matrix to express
the underlying information of the received signal more completely.
(2) We introduce the LSTM structure and optimize the design of a new neural network
concerning CLDNN to improve the model’s ability to extract relevant features.
(3) By learning multiple features of the signal through the designed SenseNet, the detec-
tion ability of the model at low SNR is improved.
Therefore, this paper mainly studies the deep learning method and detection of
multiple feature combinations of extracted signals through a neural network under low
SNR. The structure of this paper is as follows: Section 2 introduces the general model of
spectrum sensing and the model of signal and noise required for simulation. In Section 3, the
method of extracting four features of the signal and the combination process is introduced
in detail, and the structure of SenseNet proposed in this paper is further explained. In
Section 4, the simulation results of the multi-feature combination method based on deep
learning are presented and analyzed. In Section 5, we analyze and discuss the limitations
of this study and possible future research efforts. Finally, the critical conclusion is reached
in Section 6.

2. System Model
2.1. Sensing Model
In the realm of spectrum sensing within cognitive radio systems, the sensing approach
is commonly represented as a binary classification model. This involves observing and
determining whether the primary users’ channels are idle or busy. Signal detection by the
secondary users is thereby framed as a binary hypothesis testing problem, as illustrated in
Equation (1) below:
H0 : y(n) = w(n)
(1)
H1 : y(n) = h · s(n) + w(n)
Within a defined frequency spectrum, H0 and H1 correspond to the hypotheses that
the primary cognitive signal is, respectively, absent and present. The equation explicates the
variables as follows: y(n) symbolizes the nth received sample, w(n) symbolizes the additive
Electronics 2024, 13, 2705 4 of 13

noise, which is characterized by a circularly symmetric complex Gaussian distribution with

a zero mean, s(n) symbolizes the signal originating from the primary user, and h symbolizes
the constant channel gain throughout the sensing period.
The evaluation of a spectrum-sensing algorithm is principally gauged by several
performance indicators: detection probability, false-alarm probability, and missed detec-
tion probability, among others. Detection probability is defined as the probability that
the secondary user accurately identifies the presence of the primary user, denoted by
Pd = Pr { H1 | H1 }, when the primary user is indeed present.
The false-alarm probability is defined as the likelihood, denoted by Pf = Pr { H1 | H0 },
that a secondary user erroneously perceives the presence of the primary user when the
primary user is actually absent.
The effectiveness of an improved model is measured by its ability to maintain a
consistent false-alarm probability while demonstrating an enhanced detection probability,
particularly in environments characterized by low SNR.

2.2. Signal and Noise Model

This work examines the placement of sensing nodes for secondary users and the trans-
mitters of primary users’ signals linked to these nodes within a specific frequency range.
The proposed approach involves performing M sensing and sampling. The experiment
uses two modulation signals, namely QPSK and 8PSK, to serve as the fundamental signals
for the primary user signals. In order to replicate a reasonably accurate situation, the
carrier frequencies of these two modulation signals are set to be 10 kHz, and they traverse
a multipath Rayleigh fading channel. To assess the effectiveness of this model in situations
with weak signals and high noise, the tests are predominantly carried out within an SNR
range of −20 dB to 5 dB.
The noise signal consistently adheres to a generalized Gaussian distribution to more
accurately match the data and facilitate later enhancements to the model. The probability
density distribution provides a more precise understanding of the observable condition:

1 |x|β
f X (x) = exp (− ), x ∈ ℜ (2)
2αΓ( β1 ) α

The scale parameter, α, and the shape parameter, β, govern the probability density
function of the noise signal. By setting the shape parameter β to 2, we obtain the additive
white Gaussian noise (AWGN).

3. Using the SenseNet Network to Learn the Combined Features of the Signal
Figure 1 illustrates the flowchart of the deep learning spectrum-sensing algorithm
proposed in this paper. Initially, we extracted energy, power spectrum, and cyclostation-
ary features from the obtained observational data. Then, we extracted the in-phase and
quadrature components of the signal and performed energy calculations and normalization.
The four retrieved features were organized in sequence. Subsequently, the four retrieved
features were organized in the respective sequence to construct the feature matrix, and
the dimensions of the matrix were adjusted to match the neural network it will traverse.
Next, we input the feature matrix into the specialized neural network, called SenseNet. In
SenseNet, we trained this feature matrix to perform binary classification tasks and achieve
the corresponding detection performance. Ultimately, it determines whether the primary
user’s status is occupied.
 trieved features were organized in the respective sequence to construct the feature matrix,
and the dimensions of the matrix were adjusted to match the neural network it will trav-
erse. Next, we input the feature matrix into the specialized neural network, called Sense-
Net. In SenseNet, we trained this feature matrix to perform binary classification tasks and
Electronics 2024, 13, 2705 achieve the corresponding detection performance. Ultimately, it determines whether 5 ofthe
13
primary user’s status is occupied.

Figure 1. Flowchart of
Figure 1. of the
the deep
deep learning
learning spectrum-sensing
spectrum-sensing algorithm
algorithm proposed
proposed in
inthis
thispaper.
paper.

3.1. Multiple Feature Combination

3.1. Multiple Feature Combination
3.1.1. Energy Statistics
3.1.1. Energy Statistics
Energy statistics are a prevalent approach in traditional spectrum sensing due to
their Energy statistics are
straightforward a prevalent
detection approachand
mechanism in traditional spectrum sensing
ease of implementation. duemethod
This to their
straightforward detection mechanism and ease of implementation. This method
fundamentally operates by calculating the energy of the received signal, either in the funda-
time
mentally operates by calculating the energy of the received signal, either in
or frequency domain, and then comparing it to a predetermined threshold to ascertain the the time or
frequencyofdomain,
presence and user.
the primary then The
comparing it toisa executed
calculation predetermined threshold to ascertain the
as follows:
presence of the primary user. The calculation is executed as follows:
1 X −1 X 1 2
ES (y)X n∑
ESˆ(y)^ = 1 |y(n)| , n = 1, 2, . . . , N − 1 (3)
=0  | y( n ) |2 , n  1, 2,  , N  1 (3)
X n 0

The
The formula
formulaprocesses
processesthe sampled
the sampledreceived signal,
received y(n),
signal, X samples
overover
y(n), X samples and computes
and com-
the average energy statistics for each segment of the received signal sequence.
putes the average energy statistics for each segment of the received signal sequence. The derived
The
T
energy statistical feature is subsequently denoted by E = [ ESˆ 1 ,ESˆ 2 , . .^. ESˆ^M ] . ^ T
derived energy statistical feature is subsequently denoted by E  [ ES1 , ES 2 ,  ES M ] .
3.1.2. Power Spectrum
3.1.2.The
Power Spectrum
signal power is a crucial characteristic of a signal, and the power spectral density
(PSD)The signal insights
provides power isintoa crucial characteristic
the distribution of a signal,
of this power. and theconsider
Let us power spectral density
a finite-length
(PSD) provides insights into the distribution of this power. Let us consider a finite-length
received signal sequence, denoted as y(k). The estimation of the power spectral density can
received
be signal
expressed sequence, denoted as y(k). The estimation of the power spectral density
as follows:
can be expressed as follows:
1 1
Sy (k ) = |y(k )|2 = | FFT [y(k)]|2 , k = 0, 1, . . . N − 1 (4)
N N
Let FFT (y(k )) represent the Fourier transform of the received signal sequence, y(k),
with the period N of the Fourier transform chosen as the period for calculating the power
spectrum. The computed power spectrum features of the received signal sequence are
stored in the feature vector P = [ P1 , P2 , . . . PM ] T .

3.1.3. Cyclostationary Features

Different modulation and noise signals have distinct statistical characteristics, such as
means, variances, and covariances. Observing a signal’s cyclostationary features (including
mean, variance, and autocovariance) can help distinguish signals from different sources.
The procedure for calculating cyclostationary features is outlined below.
Electronics 2024, 13, 2705 6 of 13

The periodic autocorrelation function of a signal is usually used to measure the

autocorrelation of a signal at different time lags and takes into account the periodicity that
may be present in the signal, which is calculated as follows:

Z
T
τ τ
Rxα (τ ) = lim 2 x (t + ) x (t − )e− j2παt dt (5)
T →∞ T 2 2
−
2
Afterward, the acquired autocorrelation function is subjected to a discrete Fourier
transform to obtain the spectrum autocorrelation function:
Z +∞
Sxα ( f ) = Rαx (τ )e− j2π f τ dτ (6)
−∞

Z
1
Z
Sxα ( f ) 2 X (t, f + ατ ) X ∗ (t, f − ατ )dτ
= lim lim (7)
Z T
T →∞ Z →∞ TZ 2 T
2
−
2
Finally, the obtained computational results are saved in the feature vector:
CS = [CS1 , CS2 , . . . CS M ] T .

3.1.4. I/Q Component

In communication systems, two carriers can represent a modulated signal: one is the
in-phase vector, and the other is the quadrature vector. Decomposing the composite signal
into these two independent components allows for better identification of the actual part of
the signal, resistance to interference, and improved analysis of the signal’s amplitude and
phase information. The formulas for the two independent vectors are as follows:

I = Image(y(n))
(8)
Q = Real (y(n))

MATLAB R2022a obtained the I/Q characteristics by segregating the real and imag-
inary components of the signal, computing the mean energy of each component, and
subsequently standardizing the highest value. The extracted I/Q component features were
saved in the feature vectors, I = [ I1 , I2 , . . . I M ] T and Q = [ Q1 , Q2 , . . . Q M ] T , respectively.
Finally, the feature matrices of the four types of signals were combined. First, for
the energy statistics features, the average energy of M segments was calculated and then
transposed to adjust the size of the matrix. Next, the power spectrum features were
obtained by magnifying the results calculated using the Fourier transform and the pwelch
method by a factor of 100 and were stored in an M × 1 vector. Finally, the four types of
features were arranged in order to form a matrix. Its size matched the input layer of the
designed SenseNet, and the final feature matrix was represented as: [ ESˆ(y), P, CS, I, Q],
with a matrix size of M × 6. The process of feature extraction of the signal is shown in
Figure 2 below.

3.2. SenseNet Network Infrastructure

Two-dimensional convolutional neural networks are the most commonly used foun-
dational networks in the fields of image classification and image data processing. They
possess two spatial dimensions, height and width, which make them suitable for processing
image data. Regarding spectrum sensing, local features of the spatial correlation charac-
teristics of received signals can be extracted. However, for dealing with time series data
that have significant spatiotemporal correlations, recurrent neural networks (RNN), such
as LSTM and GRU, may be a more suitable and efficient choice.
 Electronics 2024, 13, x FOR PEER REVIEW 7 of 15

^
Electronics 2024, 13, 2705 [ ES ( y ), P, CS , I , Q] , with a matrix size of M × 6. The process of feature extraction of the 7 of 13
signal is shown in Figure 2 below.

Figure 2. Schematic diagram of the extraction process of each feature of the signal sequence.

3.2. SenseNet Network Infrastructure

Two-dimensional convolutional neural networks are the most commonly used foun-
dational networks in the fields of image classification and image data processing. They
possess two spatial dimensions, height and width, which make them suitable for pro-
cessing image data. Regarding spectrum sensing, local features of the spatial correlation
characteristics of received signals can be extracted. However, for dealing with time series
Figure 2. Schematic diagram of the extraction process of each feature of the signal sequence.
data
Figure 2.that have significant
Schematic diagram spatiotemporal
of the extractioncorrelations,
process of each recurrent
featureneural
of thenetworks (RNN),
signal sequence.
such as LSTM and GRU, may be a more suitable and efficient choice.
3.2. SenseNet Network Infrastructure
Figure
Figure 3 illustrates
3 illustratesthe theSenseNet
SenseNet network structurewe
network structure weproposed
proposed forfor spectrum
spectrum sens-sensing,
which isTwo-dimensional
ing, which primarily
is primarily convolutional
composed
composed of of neural networks
convolutional
convolutional andare
and the layers.
LSTM
LSTM most commonly
layers. TheThe used
signals
signals foun-
receivedreceived
dational
by by secondary
secondary networks
users in
userswere
weretheprocessed
fields of image
processed through
throughclassification
a feature
a feature and image
extractor
extractor data
to obtain processing. They
a multi-feature
to obtain a multi-feature
possess
combination two spatial
matrix.We dimensions,
Wetreated
treatedthe height
thesignal’s and width,
signal’sfeature which
feature matrix
matrix as make
as a a 11 ×them
× 64 suitable for pro-
combination matrix. 64××6 6image,
image,which
which was
cessing
was image
thenintoinputthe data.
into Regarding
the SenseNet spectrum
network. sensing, local features of the spatial correlation
then input SenseNet network. ThisThis network
network firstfirst
used used twotwo 2D 2Dconvolutional
convolutional layers
characteristics
layers to process ofthe
received
input signals
data to can be extracted.
obtain sufficiently However,
accuratefor dealing
local withthereby
features, time series
ex-
to process
data that the
haveinput data tospatiotemporal
significant obtain sufficiently accurate
correlations, local features,
recurrent neural thereby
networks extracting
(RNN),
tracting features from multiple time points to reduce irrelevant data. The data entered
features
such as from
LSTM multiple
and GRU, time pointsbe ato reduce irrelevant data. The data entered convolutional
convolutional layer 1 frommay a single more suitable
channel, and efficient
generating choice.
four feature maps, and convolu-
layer 1 from
tionalFigure
a single
layer 23further
channel,
illustrates generating
the SenseNet
extracted featuresnetwork
four feature
from thestructure
four feature
maps,
we proposed and
maps of the
convolutional
for spectrumlayer,
previous sens-layer 2
further extracted
ing, which
outpu features
is primarily
ing eight from
featurecomposed the four
maps. Theoffunctionsfeature
convolutional maps
performed of
and LSTMthe previous
by thelayers. layer, outputting
layers eight
The signals received
two convolutional
feature
by maps.
secondary The
can be expressed as: functions
users were performed
processed by
through the a two convolutional
feature extractor to layers
obtain can
a be expressed
multi-feature as:
combination matrix. We treated the signal’s feature matrix as a 1 × 64 × 6 image, which
was then input into the SenseNet X (Xt)(t= f (fW(W
)network. (t()tThis
) XX
⊗ (t(−t 11)) +
network bb((tt))
first )) two 2D convolutional
used (9) (9)
layers to process the input data to obtain sufficiently accurate local features, thereby ex-
tracting features from multiple time points to reduce irrelevant data. The data entered
convolutional layer 1 from a single channel, generating four feature maps, and convolu-
tional layer 2 further extracted features from the four feature maps of the previous layer,
outpu ing eight feature maps. The functions performed by the two convolutional layers
can be expressed as:

X (t )  f (W (t )  X (t  1)  b(t )) (9)

Figure Proposed
3. 3.
Figure ProposedSenseNet
SenseNet network architecture.
network architecture.

The convolutional layer’s output was transformed into a feature vector with 64 di-
mensions by replicating the output of the fully connected layer. This layer is designed to
combine and re-map the learned local features for higher-level abstract representations.
Next, the data passed through two LSTM layers, whose excellent learning ability for tem-
poral correlation makes them particularly effective for processing sequential data. The
operation of the LSTM layers can be represented as:
Figure 3. Proposed SenseNet network architecture.
(1) (1) (1) (1)
ht , Ct = LSTM1( xt , ht−1 , Ct−1 )
(2) (2) (2) (2) (10)
ht , Ct = LSTM2( xt , ht−1 , Ct−1 )

(1) (1)
Here, xt represents the input to the first-layer LSTM at time step t, while ht−1 and Ct−1
represent the hidden state and cell state of the first-layer LSTM from the previous time
(2) (2)
step, respectively. Similarly, ht−1 and Ct−1 represent the hidden state and cell state of the
second-layer LSTM from the previous time step. The ultimate result was categorized by
the successive fully connected layers.
Electronics 2024, 13, 2705 8 of 13

Following the convolutional layers, the technique of maximum pooling (MaxPool)

was employed without padding to decrease the spatial dimensions of the features and
improve their resilience. The second pooling layer further decreased the spatial dimensions.
The Rectified Linear Unit (ReLU) was the activation function used after each layer. Table 1
provides a comprehensive overview of the hyperparameters calculated using a rigorous
Electronics 2024, 13, x FOR PEER REVIEW 8 of 15
cross-validation process.

Table 1. Hyperparameters of the proposed SenseNet network.

The convolutional layer’s output was transformed into a feature vector with 64 di-
mensions by replicating the output of the fully connected layer. This layer is designed
Hyperparameters Value to
combine and re-map the learned local features for higher-level abstract representations.
Number of convolution kernels per convolutional layer 4, 8
Next, the data passed through two LSTM layers, whose excellent learning ability for tem-
Convolution
poral correlation makes kernel size effective for processing sequential
them particularly 2× ×2
2, 2 The
data.
operation of the LSTM layers can be represented as:
Cell unit for each LSTM layer 32, 16
(1) (1) (1) (1)
Output dimensions for h , C each
t fully
t  LSTM 1( x , h , C )
connectedt layer
t 1 t 1 64, 32, 2
(2) (2) (2) (2)
(10)
h ,C
Optimizer
t t  LSTM 2( xt , h , C )
t 1 t 1 Stochastic Gradient Descent
Learning rate 0.01
(1)
Here, xt represents the input to the first-layer LSTM at time step , while ht 1 and
Batch size 32
Ct(1)
1 represent the hidden state and cell state of the first-layer LSTM from the previous

time step, respectively. Similarly, ht(2) (2)

1 and Ct 1 represent the hidden state and cell state
Compared to CNN, the CLDNN model offers better flexibility and scalability. It
of the second-layer LSTM from the previous time step. The ultimate result was categorized
combines the advantages of
by the successive fully connected layers.CNN and LSTM to capture signals’ spatial and temporal
characteristics
Following the more effectively,
convolutional demonstrating
layers, the technique excellent
of maximum potential
pooling in processing complex
(MaxPool)
signal tasks [22].
was employed By padding
without adjusting the number
to decrease of LSTM
the spatial and DNN
dimensions of thelayers
featuresand
andthe activation
improve their
functions resilience.each
following The second
layer’s pooling layer further
structure, the modeldecreased
can the
oftenspatial
be dimen-
further adapted to
sions.
the Theof
type Rectified Linearinto
data input Unitthe
(ReLU) was theachieving
network, activation function used after each We
better performance. layer.
used a dataset
Table 1 provides a comprehensive overview of the hyperparameters calculated using a
comprising four feature combination matrices of QPSK signals and conducted model-tuning
rigorous cross-validation process.
experiments
Comparedand extensive
to CNN, the CLDNNcross-validation
model offers bebyeraltering
flexibilitythe
andpositioning of the LSTM and
scalability. It com-
DNN
bines the advantages of CNN and LSTM to capture signals’ spatial and temporal charac- parameters
layers and varying the activation functions, thus obtaining the optimal
and model.
teristics Figure 4demonstrating
more effectively, displays theexcellent
parts of the models
potential with complex
in processing a relatively
signalhigh average
detection
tasks [Error!accuracy
Reference(i.e., thenot
source average
found.].of Bythe best performance
adjusting of theand
the number of LSTM models
DNN at each SNR
layers and
during thethe activation
training functions
process) following
during theeach layer’s structure,
experiments. the modelsome
By changing can often
of the network’s
be further and
structure adapted to the type
activation of data the
functions, input into the
neural network,
network achieving
showed be er performance
different perfor- trends
mance. We used a dataset comprising four feature combination matrices of QPSK signals
for signal feature data under low-SNR conditions.
and conducted model-tuning experiments and extensive cross-validation by altering the
The model
positioning with two
of the LSTM convolutional
and DNN layers and layers,
varying one fully connected
the activation functions,layer, two long short-
thus ob-
term memory layers, and two final completely connected layers
taining the optimal parameters and model. Figure 4 displays the parts of the models with had the greatest perfor-
mance, according
a relatively to the
high average data in
detection Table (i.e.,
accuracy 2. Thethe model
average attained
of the bestaperformance
mean accuracy of of 91.03%
the models
within theat−each
20 dBSNR toduring the training
5 dB range, process) during
establishing it as the
theexperiments.
most optimal By changing
model among those
some of the network’s
examined. Therefore, structure and activation
we defined model functions, the neural
4 as SenseNet, network
a deep showed
learning dif- suitable for
model
ferent performance trends for signal feature data under low-SNR conditions.
learning a combination of four features.

Figure 4.4.Detection
Figure Detectionperformance of some
performance of neural
some network models with
neural network different
models withinternal structures.
different internal structures.
Electronics 2024, 13, 2705 9 of 13

Table 2. Average performance of selected neural network models.

Average Detection
Model Architecture
Accuracy
Model 1 2COV + Sigmoid + 3FC + Sigmoid 88.49%
Model 2 2 COV + ReLU + 2FC + ReLU 2LSTM + 1FC + ReLU 88.89%
Model 3 2 COV + ReLU + 1FC + ReLU + 2LSTM + 2FC + SoftMax 88.87%
Model 4 2 COV + ReLU + 1FC + ReLU + 2LSTM + 1FC + ReLU 91.03%

4. Experimental Analysis and Result Discussion

4.1. Experimental Environment
The software required for the experiment was the R2022a version of MATLAB, Python
3.7, Pytorch 2.0.1 deep learning framework. The hardware platform for the experimental
environment was the Windows 11 64-bit operating system. The processor was AMD Ryzen
7 5800H (3.20 GHz) and the graphics card was NVIDIA GeForce RTX 3050 Ti Laptop GPU.

4.2. Data Generation and Model Training

The experimental datasets utilized in this study were generated by MATLAB. These
datasets principally consisted of noisy modulated signal samples of QPSK and 8PSK, as
well as signal samples containing only noise. The SNR ranged from −20 dB to 5 dB.
The sampling rate of each signal sample was 2000 Hz, and it underwent transmission
across a multipath Rayleigh fading channel, with a path delay vector of [0 s, 0.001 s]. The
negative samples were comprised of the previously specified additive white Gaussian
noise. This study partitioned the dataset into training, validation, and testing sets using the
conventional split ratio of 6:2:2. The parameters of the dataset are shown in Table 3 below.

Table 3. Dataset parameters.

Type of modulation QPSK, 8PSK

Signal sampling frequency 2000 Hz
Length of generated random-bit sequence 400
SNR range −20 dB~5 dB in 1 dB increments
Number of training samples 15,600

4.3. Performance with QPSK and 8PSK Signals

In this experiment, two signal sequences, QPSK and 8PSK, were generated by MAT-
LAB as a dataset for the model SenseNet, which was tested by completing the training and
saving the parameters, thus evaluating SenseNet and comparing the performance of the
two modulated signals, QPSK and 8PSK, on this model.
The performance of the two signals under this model is shown in Figure 5 below.
Figure 5a demonstrates that when the SNR was higher than −14 dB, the detection proba-
bility of QPSK under this model was higher than 80%. When the SNR reached −9 dB, the
detection probability could reach 100%, which fulfilled the spectrum-sensing task well. Due
to the small phase interval of the 8PSK signal, the characteristics were not as apparent as
QPSK. Its detection performance was lower than that of QPSK under low-SNR conditions,
and it could only achieve a better detection effect when the SNR was −11 dB. On the other
hand, in the range of −20 dB to −15 dB, the declining trend of detection performance
for the QPSK signal was relatively stable with the decrease in SNR, while the detection
performance for the 8PSK signal was further decreased.
 QPSK, which resulted in an overall higher false-alarm probability than QPSK signals un-
der low-SNR conditions.
In summary, the multi-feature combination method based on deep learning demon-
strated be er detection performance for primary user signals dominated by QPSK signals
Electronics 2024, 13,compared
2705 to the 8PSK signal under low-SNR conditions, and it also achieved a lower 10 of 13
overall false-alarm probability.

Figure 5. (a) Detection performance of SenseNet with QPSK and 8PSK signals. (b) False-alarm prob-
Figure 5. (a) Detection performance of SenseNet with QPSK and 8PSK signals. (b) False-alarm
ability of SenseNet with QPSK and 8PSK signals.
probability of SenseNet with QPSK and 8PSK signals.

4.4. Comparison of Feature Types

From the false-alarm probabilities of the two signals demonstrated in Figure 5b, the
In order to evaluate the
false-alarm effectiveness
probability of deep
decreased learning
with techniques
the increase utilizing
in SNR, and themultiple
false-alarm probability
feature combinations, we compared
of both was theirwhen
less than 10% performance
the SNRwith
was methods based
greater than −10ondB.
a single
When the SNR was
cyclostationarity feature
greater thanand
−6methods based on combining
dB, the false-alarm probabilityenergy andwas
of both power
almostspectra.
0%. InInaddition, since
addition, experiments were conducted
8PSK signals by adding
are susceptible onlytheir
to noise, I/Q features
featureswithout changing
were not the and stable as
as distinct
CNN model toQPSK,validate the resulted
which performance
in an of the added
overall higherfeatures. Figure
false-alarm 6 displays
probability thanthe re- signals under
QPSK
sults. low-SNR conditions.
In summary, the multi-feature combination method based on deep learning demon-
strated better detection performance for primary user signals dominated by QPSK signals
compared to the 8PSK signal under low-SNR conditions, and it also achieved a lower
overall false-alarm probability.

4.4. Comparison of Feature Types

In order to evaluate the effectiveness of deep learning techniques utilizing multiple
feature combinations, we compared their performance with methods based on a single
ics 2024, 13, x FOR PEER REVIEW cyclostationarity feature and methods based on combining energy and 12 ofpower
15 spectra. In
addition, experiments were conducted by adding only I/Q features without changing the
CNN model to validate the performance of the added features. Figure 6 displays the results.

Figure 6. (a) Detection performance of deep learning models with different feature types under
Figure
QPSK signals. (b) 6. (a) Detection
False-alarm probabilityperformance of deep
of deep learning learning
models withmodels
differentwith different
feature types feature
un- types under
QPSK signals. (b) False-alarm probability of deep learning models with different feature types under
der QPSK signals.
QPSK signals.
Using QPSK as the received signal and under the same experimental setup, Figure
6a demonstrates that the model based on a single cyclostationarity feature (“CS+CNN” in
Figure 6) usually performed poorly at low SNR and generally required an SNR greater
than 0 dB to complete the detection. In contrast, the deep learning method that combines
energy and power spectra (“ES+P+CNN” in Figure 6) performed relatively well and could
achieve a detection probability of about 70% at −16 dB. However, the overall performance
Electronics 2024, 13, 2705 11 of 13

Using QPSK as the received signal and under the same experimental setup, Figure 6a
demonstrates that the model based on a single cyclostationarity feature (“CS+CNN” in
Figure 6) usually performed poorly at low SNR and generally required an SNR greater
than 0 dB to complete the detection. In contrast, the deep learning method that combines
energy and power spectra (“ES+P+CNN” in Figure 6) performed relatively well and could
achieve a detection probability of about 70% at −16 dB. However, the overall performance
was lower than the spectrum-aware method based on three features (“ES+P+CS+CNN”
in Figure 6). Regarding the false-alarm probability, Figure 6b demonstrates that the false-
alarm rate of the model using the cyclostationarity feature approach was not ideal, as it
did not decrease with the increase in SNR, and the value always remained around 10% to
15%. In this paper, we proposed signal I/Q feature components for existing deep learning
methods that combine energy, power spectrum, and cyclostationarity features and designed
a SenseNet model suitable for learning the feature types.
The results demonstrated that the multi-feature method based on deep learning
proposed in this study (“Ours” in Figure 6) enhanced the detection performance by an
average of 2.3% compared to the existing method based on three features within the
range of −20 dB to −7 dB, with a notable improvement of 3.3% at −20 dB. With only
the introduction of I/Q features, while retaining the CNN model (“ES+P+CS+IQ+CNN”
in Figure 6), the average detection probability of the model increased by about 0.33%.
In addition, analyzing the relationship between the number of features in the feature
combination and the detection performance, it was observed that with an increase in the
number of features, the deep learning model exhibited a stronger ability to extract and learn
signal features, resulting in a generally improved detection performance. Compared with
the single cyclostationary feature method, the performance of the two feature combination
methods based on energy features and power features was significantly enhanced, and the
reduction in false-alarm probability remained relatively stable. When the cyclostationary
feature was combined with the other two features, the improvement in detection probability
began to decrease, while the false-alarm probability further decreased. On this basis, the
multi-feature method proposed in this study further improved the detection probability of
primary user signals under low-SNR conditions while keeping the false-alarm probability
relatively stable.
On the other hand, Figure 6b illustrates that with changes to the feature matrix and
the model’s redesign, the false-alarm rate of the method presented in this paper was almost
identical to that of existing methods, consistent with the model’s performance evaluation
standards under practical conditions, as previously mentioned.
In addition, we further calculated the complexity of the original CNN model and the
SenseNet model in this paper, and compared the results, as shown in Table 4. It can be seen
from Table 4 that in the training process of the two models, for the feature matrix dataset
composed of four features, the optimal detection accuracy of SenseNet was about 1% higher
than that of the CNN model. In terms of model complexity, the SenseNet proposed in
this paper had slightly lower complexity compared to the original CNN model, whether
measured by the number of parameters or Floating-Point Operations Per Second (FLOPs).
This is because the CNN model contains multiple consecutive fully connected layers, which
are the main source of the total number of parameters in the model. For SenseNet, the
total number of parameters was reduced by decreasing the number of fully connected
layers. Additionally, the computational complexity of the LSTM layer was lower than that
of multiple fully connected layers, resulting in the total FLOPs and complexity of SenseNet
being slightly lower than those of the CNN model.

Table 4. Model complexity calculation.

Model Best Accuracy Parameter Count FLOPs

CNN 89.69% 80,834 287,408
SenseNet 90.65% 65,726 256,720
Electronics 2024, 13, 2705 12 of 13

5. Limitations and Future Work

Compared with the existing three-feature combination methods, the multi-feature
combination method proposed in this paper improved the detection performance in the
low-SNR range and slightly reduced the model complexity. However, the experimental
results showed that as the number of features in the feature combination increased, the
growth rate of the spectrum-sensing and detection performance of the model gradually
decreased. This is because the feature matrix research can more completely express the
underlying information of the signal, and the impact of increasing the feature types of the
signal on system detection performance was gradually reduced.
In the future, it can be considered that the feature extraction process and combination of
the signal could also be completed by the neural network, with the neural network directly
learning the signal and fusing different features through the concatenate layer to further
improve the model’s expression ability and performance. Additionally, our deep learning
approach is supervised and requires a large number of labeled datasets to improve model
performance. In a real-world communication environment, obtaining and labeling large
amounts of communication data is relatively difficult and costly. Therefore, unsupervised
learning can be considered for spectrum sensing in the future to further reduce costs.
Unsupervised learning mainly analyzes and models unlabeled data, summarizing the
structure and rules in the data to enable prediction and classification.

6. Conclusions
This study proposed a spectrum-sensing method based on the SenseNet neural net-
work, which primarily learns and analyzes by examining the internal information of
modulated signals. By decomposing the signal into its co-phase and quadrature compo-
nents, we extracted the features of the I/Q vectors and combined them with existing feature
matrices in a specific manner to form a feature matrix suitable for deep learning. First,
the noise interference suffered by the signals received by secondary users conformed to
a generalized Gaussian distribution and underwent multipath Rayleigh fading during
transmission. Subsequently, through extensive cross-validation and model tuning, we
determined a deep learning model that fit the multi-feature matrix of the signal. We then
discussed the performance of the spectrum-sensing methods using a combination of dif-
ferent signal features under the same parameter conditions and the deep learning model.
Simulation results showed that at a low SNR, the proposed spectrum-sensing method sub-
stantially improved the detection probability compared to the single-feature deep learning
method based on cyclostationarity features and had a lower false-alarm probability than
the combination of the power spectrum and energy statistics. Furthermore, compared to
existing CNN-based multi-feature spectrum-sensing schemes, this model also improved
the detection probability while maintaining a stable false-alarm rate, and it advanced the
SNR point at which a 100% detection probability was first achieved to −9 dB.

Author Contributions: Conceptualization, Z.L.; methodology and investigation, Y.Z.; writing—original

draft preparation, Y.Z.; writing—review and editing, Z.L.; supervision, Z.L.; project administra-
tion, Z.L.; funding acquisition, Z.L. All authors have read and agreed to the published version of
the manuscript.
Funding: This work was supported in part by the National Natural Science Foundation of China
under Grant 61801319, in part by the Sichuan Science and Technology Program under Grants
2020JDJQ0061 and 2021YFG0099, in part by the Innovation Fund of Chinese Universities under
Grant 2020HYA04001, and in part by the Innovation Fund of Engineering Research Center of
the Ministry of Education of China, Digital Learning Technology Integration and Application
(No. 1221009).
Data Availability Statement: The settings for generating the relevant data are included. For further
information, please contact the first author.
Conflicts of Interest: The authors declare no conflicts of interest.
Electronics 2024, 13, 2705 13 of 13

References
1. Chen, W.; Demirkol, I.; Mostafa, M.; Perotti, A. Mobile Communications And Networks. IEEE Commun. Mag. 2021, 59, 52–53.
[CrossRef]
2. Liu, B.; Bai, X.; Zhang, G.; Shen, J.; Xie, J.; Zhao, L.; Hong, T. A Review of Research on Deep Learning for Cognitive Radio
Applications. J. East China Norm. Univ. (Nat. Sci. Ed.) 2021, 1, 36–52.
3. Ali, M.; Yasir, M.N.; Bhatti, D.M.S.; Nam, H. Optimization of Spectrum Utilization Efficiency in Cognitive Radio Networks. IEEE
Wirel. Commun. Lett. 2022, 12, 426–430. [CrossRef]
4. Mitola, J. Cognitive Radio. An Integrated Agent Architecture for Software Defined Radio. Ph.D. Thesis, KTH Royal Institute of
Technology, Stockholm, Sweden, 2002.
5. Luo, J.; Zhang, G.; Yan, C. An Energy Detection-Based Spectrum-Sensing Method for Cognitive Radio. Wirel. Commun. Mob.
Comput. 2022, 2022, 3933336. [CrossRef]
6. Yang, M.; Li, Y.; Liu, X.; Tang, W. Cyclostationary Feature Detection Based Spectrum Sensing Algorithm under Complicated
Electromagnetic Environment in Cognitive Radio Networks. China Commun. 2015, 12, 35–44. [CrossRef]
7. Brito, A.; Sebastião, P.; Velez, F.J. Hybrid Matched Filter Detection Spectrum Sensing. IEEE Access 2021, 9, 165504–165516.
[CrossRef]
8. Dai, L.; Jiao, R.; Adachi, F.; Poor, H.V.; Hanzo, L. Deep Learning for Wireless Communications: An Emerging Interdisciplinary
Paradigm. IEEE Wirel. Commun. 2020, 27, 133–139. [CrossRef]
9. Seo, D.; Nam, H. Deep Rp-Cnn for Burst Signal Detection in Cognitive Radios. IEEE Access 2020, 8, 167164–167171. [CrossRef]
10. Alkhouri, I.; Joneidi, M.; Hejazi, F.; Rahnavard, N. Large-Scale Spectrum Occupancy Learning via Tensor Decomposition and LSTM
Networks; IEEE: Washington, DC, USA, 2020; pp. 677–682.
11. Nasser, A.; Chaitou, M.; Mansour, A.; Yao, K.C.; Charara, H. A deep neural network model for hybrid spectrum sensing in
cognitive radio. Wirel. Pers. Commun. 2021, 118, 281–299. [CrossRef]
12. Tekbıyık, K.; Akbunar, Ö.; Ekti, A.R.; Görçin, A.; Kurt, G.K.; Qaraqe, K.A. Spectrum Sensing and Signal Identification with Deep
Learning Based on Spectral Correlation Function. IEEE Trans. Veh. Technol. 2021, 70, 10514–10527. [CrossRef]
13. Tekbiyik, K.; Akbunar, Ö.; Ekti, A.R.; Görçin, A.; Kurt, G.K. Multi–Dimensional Wireless Signal Identification Based on Support
Vector Machines. IEEE Access 2019, 7, 138890–138903. [CrossRef]
14. Kulin, M.; Kazaz, T.; Moerman, I.; De Poorter, E. End-to-End Learning from Spectrum Data: A Deep Learning Approach for
Wireless Signal Identification in Spectrum Monitoring Applications. IEEE Access 2018, 6, 18484–18501. [CrossRef]
15. Vijay, E.V.; Aparna, K. Spectrum Sensing Using Deep Learning for Proficient Data Transmission in Wireless Sensor Networks for
Wireless Communication. Ain Shams Eng. J. 2024, 15, 102784. [CrossRef]
16. Benazzouza, S.; Ridouani, M.; Salahdine, F.; Hayar, A. A Novel Spectrogram Based Lightweight Deep Learning for IoT Spectrum
Monitoring. Phys. Commun. 2024, 64, 102364. [CrossRef]
17. Gao, J.; Yi, X.; Zhong, C.; Chen, X.; Zhang, Z. Deep Learning for Spectrum Sensing. IEEE Wirel. Commun. Lett. 2019, 8, 1727–1730.
[CrossRef]
18. Zhang, X.; Ma, Y.; Liu, Y.; Wu, S.; Jiao, J.; Gao, Y.; Zhang, Q. Robust DNN-Based Recovery of Wideband Spectrum Signals. IEEE
Wirel. Commun. Lett. 2023, 12, 1712–1715. [CrossRef]
19. Han, D.; Sobabe, G.C.; Zhang, C.; Bai, X.; Wang, Z.; Liu, S.; Guo, B. Spectrum Sensing for Cognitive Radio Based on Convolution
Neural Network. In Proceedings of the 2017 10th International Congress on Image and Signal Processing, BioMedical Engineering
and Informatics (CISP-BMEI), Shanghai, China, 14–16 October 2017; pp. 1–6.
20. Patel, D.K.; López-Benítez, M.; Soni, B.; García-Fernández, Á.F. Artificial Neural Network Design for Improved Spectrum Sensing
in Cognitive Radio. Wirel. Netw. 2020, 26, 6155–6174. [CrossRef]
21. Geng, Y.; Huang, J.; Yang, J.; Zhang, S. Spectrum Sensing for Cognitive Radio Based on Feature Extraction and Deep Learning; IOP
Publishing: Bristol, UK, 2022; Volume 2261, p. 012016.
22. Salahdine, F.; Han, T.; Zhang, N. 5G, 6G, and Beyond: Recent Advances and Future Challenges. Ann. Telecommun. 2023,
78, 525–549. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.