CN112101460A - Spectrum sensing method and device based on deep learning classification - Google Patents
Spectrum sensing method and device based on deep learning classification Download PDFInfo
- Publication number
- CN112101460A CN112101460A CN202010974518.0A CN202010974518A CN112101460A CN 112101460 A CN112101460 A CN 112101460A CN 202010974518 A CN202010974518 A CN 202010974518A CN 112101460 A CN112101460 A CN 112101460A
- Authority
- CN
- China
- Prior art keywords
- signal
- detected
- data
- convolutional neural
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 133
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000013135 deep learning Methods 0.000 title claims abstract description 26
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 95
- 239000013598 vector Substances 0.000 claims abstract description 63
- 238000005070 sampling Methods 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 33
- 239000011159 matrix material Substances 0.000 claims description 30
- 238000012795 verification Methods 0.000 claims description 22
- 230000004913 activation Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 7
- 239000000126 substance Substances 0.000 claims description 4
- 230000008569 process Effects 0.000 abstract description 12
- 238000012545 processing Methods 0.000 abstract description 4
- 238000013461 design Methods 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 5
- 230000001149 cognitive effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention relates to a spectrum sensing method and device based on deep learning classification, belongs to the technical field of signal processing, and solves the problems that the existing spectrum sensing method needs to involve a large number of manual processes and relies on too much professional knowledge to cause lower accuracy. The method comprises the following steps: acquiring sample data of a signal; the sample data comprises a signal sampling sequence, a signal power spectrum and a wavelet vector; constructing a convolutional neural network, and carrying out network training on the convolutional neural network based on sample data to obtain an optimal network structure corresponding to the convolutional neural network; and inputting the signal data to be detected into an optimal network structure corresponding to the convolutional neural network to obtain the confidence coefficient corresponding to the signal data to be detected, and obtaining a frequency spectrum sensing result based on the confidence coefficient. The frequency spectrum sensing of the signal to be detected is realized, and the accuracy of the frequency spectrum sensing result is improved.
Description
Technical Field
The invention relates to the technical field of signal processing, in particular to a spectrum sensing method and device based on deep learning classification.
Background
With the large-scale deployment of 5G networks, the rapid emergence of the internet of things, and the rapid increase in demand for spectrum usage by various emerging technologies, wireless spectrum resources will become increasingly strained. The cognitive radio, as an opportunistic spectrum use technology, will greatly improve the spectrum use efficiency, and has important significance for alleviating the current situation of spectrum resource shortage. Spectrum sensing is a key premise for realizing dynamic spectrum access of cognitive radio. The existing spectrum sensing methods basically design corresponding decision statistics to make decisions by studying different characteristics between signals and noises, and therefore, the design of the decision statistics is very important for spectrum sensing performance. However, the design of the decision statistic needs a lot of manual processes, and depends on too much professional knowledge, so that the existing spectrum sensing method has low accuracy.
Disclosure of Invention
In view of the foregoing analysis, embodiments of the present invention are directed to providing a spectrum sensing method and apparatus based on deep learning classification, so as to solve the problem that the existing spectrum sensing method needs to involve a large number of manual processes and relies on too much professional knowledge to cause low accuracy.
On one hand, the embodiment of the invention provides a spectrum sensing method based on deep learning classification, which comprises the following steps:
acquiring sample data of a signal; the sample data comprises a signal sampling sequence, a signal power spectrum and a wavelet vector;
constructing a convolutional neural network, and carrying out network training on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network;
and inputting the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain the confidence coefficient corresponding to the signal data to be detected, and obtaining a frequency spectrum sensing result based on the confidence coefficient.
Further, the signal sampling sequence in the sample data is a normalized signal sampling sequence; the method for acquiring the sample data comprises the following steps:
sampling and normalizing the initial signal to obtain a normalized signal sampling sequence; the initial signal comprises a pure noise signal and a main user signal;
acquiring a signal power spectrum, a signal low-frequency component and a signal high-frequency component based on the normalized signal sampling sequence, and splicing the signal low-frequency component and the signal high-frequency component to obtain a wavelet vector;
obtaining a vector matrix based on the normalized signal sampling sequence, the signal power spectrum and the wavelet vector;
and adding a label to the vector matrix to obtain sample data, wherein the label comprises a signal and a no-signal, the initial signal with the signal comprises a main user signal, and the initial signal without the signal is a pure noise signal.
Further, the convolutional neural network comprises an input layer, an intermediate layer and an output layer;
the size of the input layer is the same as the size of the vector matrix;
the middle layer comprises a convolution layer and an activation function Rule;
the output layer comprises a softmax classification layer.
Further, network training is carried out on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network, and the method comprises the following steps:
selecting T sample data from the N sample data as training data, and taking the rest N-T sample data as verification data, wherein T is more than 1 and less than N;
inputting the training data into a convolutional neural network for network training to obtain a trained convolutional neural network;
and inputting the verification data into the trained convolutional neural network to obtain the confidence coefficient corresponding to the verification data, and taking the convolutional neural network corresponding to the confidence coefficient with the highest accuracy as the optimal network structure corresponding to the convolutional neural network.
Further, the step of acquiring the signal data to be detected comprises the following steps:
acquiring a signal sampling sequence to be detected and normalizing the signal sampling sequence to be detected to obtain a normalized signal sampling sequence to be detected;
obtaining a power spectrum of the signal to be detected, a low-frequency component of the signal to be detected and a high-frequency component of the signal to be detected based on the normalized sampling sequence of the signal to be detected, and splicing the low-frequency component of the signal to be detected and the high-frequency component of the signal to be detected to obtain a wavelet vector of the signal to be detected;
and obtaining a vector matrix of the signal to be detected based on the normalized sampling sequence of the signal to be detected, the power spectrum of the signal to be detected and the wavelet vector of the signal to be detected, wherein the vector matrix of the signal to be detected is the data of the signal to be detected.
Further, obtaining a spectrum sensing result based on the confidence level comprises the following steps:
inputting the signal data to be detected into an optimal network structure corresponding to the convolutional neural network to obtain a confidence coefficient t corresponding to the signal data to be detectednoiseAnd judging the confidence coefficient tnoiseWhether or not 1-t is satisfiednoiseGamma is more than gamma, and gamma is a judgment threshold;
and if the signal data to be detected is not the signal, the spectrum sensing result corresponding to the signal data to be detected is no signal.
Further, the decision threshold is obtained based on the following steps:
obtaining N-T corresponding confidence degrees based on the N-T verification data;
sequentially arranging the confidence coefficients from large to small, and selecting the ith confidence coefficient as a decision threshold; wherein the content of the first and second substances,
i＝ceil(pfp)
in the formula, ceil is rounded up, pfTo verify the pure noise data contained in the data, p is the verification data.
On the other hand, an embodiment of the present invention provides a spectrum sensing device based on deep learning classification, including: the sample acquisition module acquires sample data of the signal; the sample data comprises a signal sampling sequence, a signal power spectrum and a wavelet vector;
the network acquisition module is used for constructing a convolutional neural network and carrying out network training on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network;
and the frequency spectrum sensing module is used for inputting the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain the confidence coefficient corresponding to the signal data to be detected and obtain a frequency spectrum sensing result based on the confidence coefficient.
Further, the sample acquiring module executes the following process:
sampling and normalizing the initial signal to obtain a normalized signal sampling sequence; the initial signal comprises a pure noise signal and a main user signal;
acquiring a signal power spectrum, a signal low-frequency component and a signal high-frequency component based on the normalized signal sampling sequence, and splicing the signal low-frequency component and the signal high-frequency component to obtain a wavelet vector;
obtaining a vector matrix based on the normalized signal sampling sequence, the signal power spectrum and the wavelet vector;
and adding a label to the vector matrix to obtain sample data, wherein the label comprises a signal and a no-signal, the initial signal with the signal comprises a main user signal, and the initial signal without the signal is a pure noise signal.
Further, the convolutional neural network comprises an input layer, an intermediate layer and an output layer;
the size of the input layer is the same as the size of the vector matrix;
the middle layer comprises a convolution layer and an activation function Rule;
the output layer comprises a softmax classification layer.
Compared with the prior art, the invention can realize at least one of the following beneficial effects:
1. a spectrum sensing method based on deep learning classification includes collecting sample data, utilizing collected sample data to conduct network training on a constructed convolutional neural network to obtain an optimal network structure corresponding to the convolutional neural network, inputting signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain a spectrum sensing result, enabling the spectrum sensing result to be mainly used for reflecting existence conditions of main user signals of the signal to be detected, being simple and easy to implement, solving the problems that design of judgment statistics of an existing spectrum sensing method needs to involve a large number of manual processes and relies on too much professional knowledge to cause low accuracy, achieving automation of spectrum sensing, reducing dependence on professional knowledge and improving accuracy of the spectrum sensing result.
2. The initial signal containing the main user signal and the pure noise signal is processed to obtain sample data, and support and basis are provided for training the constructed convolutional neural network in the later period.
3. By constructing and training the convolutional neural network, an optimal network structure corresponding to the convolutional neural network is obtained, technical support is provided for later spectrum sensing of signal data to be detected, the problem of low accuracy obtained by the existing spectrum sensing method is solved, and the accuracy of spectrum sensing is improved.
4. The confidence coefficient of the signal data to be detected is obtained through the optimal network structure corresponding to the convolutional neural network, whether the confidence coefficient meets a judgment threshold is judged, and then a spectrum sensing result is obtained.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flowchart of a spectrum sensing method based on deep learning classification according to an embodiment of the present invention;
FIG. 2 is a diagram of a convolutional neural network architecture according to one embodiment of the present invention;
FIG. 3 is a diagram of a spectrum sensing device based on deep learning classification according to another embodiment of the present invention;
reference numerals:
100-sample obtaining module, 200-network obtaining module and 300-spectrum sensing module.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
The existing spectrum sensing methods basically design corresponding decision statistics to make decisions by studying different characteristics between signals and noises, and therefore, the design of the decision statistics is very important for spectrum sensing performance. However, the design of the decision statistic needs a lot of manual processes, and depends on too much professional knowledge, so that the existing spectrum sensing method has low accuracy. Therefore, the application provides a spectrum sensing method and a device based on deep learning classification, which acquire sample data, perform network training on a constructed convolutional neural network by using the acquired sample data to obtain an optimal network structure corresponding to the convolutional neural network, and finally input signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain a spectrum sensing result, wherein the spectrum sensing result is mainly used for reflecting the existence condition of a main user signal of the signal to be detected, is simple and easy to implement, solves the problems that the existing spectrum sensing method is low in accuracy because the design of judgment of the existing spectrum sensing method needs a large number of manual processes and depends on excessive professional knowledge, realizes the automation of spectrum sensing, reduces the dependence on the professional knowledge and improves the accuracy of the spectrum sensing result, has high practical value.
One embodiment of the present invention discloses a spectrum sensing method based on deep learning classification, as shown in fig. 1, including the following steps S1-S3.
Step S1, acquiring sample data of the signal; the sample data includes a sequence of signal samples, a signal power spectrum, and a wavelet vector. Specifically, the sample data of the acquired signal is mainly used for training and verifying the convolutional neural network constructed by the following steps to obtain the optimal network model structure of the convolutional neural network. The initial signal in the application is a standard signal obtained by modulating in any one modulation mode of PSK, ASK, FSK or QAM, so that a signal sampling sequence, a power spectrum and a wavelet vector obtained by the initial signal are in a standard matrix form, the form is standard, the calculation error can be reduced in the process of constructing sample data by adopting the signal sampling sequence, the power spectrum and the wavelet vector, meanwhile, the power spectrum reflects the change condition of the signal along with the frequency, the wavelet vector obtained by the wavelet change is the basis for time domain analysis and processing of the signal, so that the sample data is constructed by using the signal sampling sequence, the power spectrum and the wavelet vector, the existence condition of a main user signal can be reflected from the frequency domain and the time domain at the same time, and the accuracy of a spectrum sensing result is improved.
Preferably, the signal sample sequence in the sample data is a normalized signal sample sequence; the method for acquiring the sample data comprises the following steps:
sampling and normalizing the initial signal to obtain a normalized signal sampling sequence; the initial signal comprises a pure noise signal and a main user signal; the radio network at least comprises a main user and a cognitive user, wherein an initial signal received by the cognitive user is y(s) (t) + n (t), wherein s (t) is a main user signal, n (t) is a noise signal, namely the initial signal comprises the main user signal and the noise signal, and when the main user signal does not exist in the initial signal, the received initial signal is a pure noise signal.
Acquiring a signal power spectrum, a signal low-frequency component and a signal high-frequency component based on the normalized signal sampling sequence, and splicing the signal low-frequency component and the signal high-frequency component to obtain a wavelet vector;
obtaining a vector matrix based on the normalized signal sampling sequence, the signal power spectrum and the wavelet vector;
and adding a label to the vector matrix to obtain sample data, wherein the label comprises a signal and a no-signal, the initial signal with the signal comprises a main user signal, and the initial signal without the signal is a pure noise signal.
Specifically, sampling to obtain an initial signal r (n), normalizing the initial signal to obtain a normalized signal sampling sequence
Wherein the content of the first and second substances,
is calculated by the formula
Wherein M is the number of sampling points. After the normalized signal sampling sequence is obtained, the calculation formula of the signal power spectrum R (k) can be obtained
A signal power spectrum is obtained. Meanwhile, the low-frequency signal component L (m) and the high-frequency signal component H (m) obtained after wavelet transformation can be calculated according to the normalized signal sampling sequence, wherein the calculation formula of the low-frequency signal component L (m) is
The high-frequency component H (m) of the signal is calculated by the formula
Wherein, M1, 2., M2, g (k) and h (k) are respectively wavelet-changed corresponding low-pass filter and high-pass filter coefficients, which is determined according to actual conditions. After obtaining the low-frequency signal component L (M) and the high-frequency signal component H (M) by calculation, splicing L (M) and H (M) to obtain a wavelet vector W (M) with the length of M, and then carrying out the wavelet transformation on the wavelet vector W (M) and the wavelet vector H (M) to obtain the wavelet vector W (M) with the length of M
R (k) and W (M) are respectively used as column vectors and combined to obtain a vector matrix x with M rows and 3 columns, if r (n) contains a main user signal, a label of 'signal is added to the vector matrix',otherwise, the label is 'no signal', wherein a signal means that the initial signal comprises a main user signal, and a no signal means that the initial signal is a pure noise signal. The vector matrix x and the corresponding label form one sample data, and N sample data can be obtained by repeating the process for multiple times. The initial signal containing the main user signal and the pure noise signal is processed to obtain sample data, and support and basis are provided for training the constructed convolutional neural network in the later period.
And S2, constructing a convolutional neural network, and performing network training on the convolutional neural network based on sample data to obtain an optimal network structure corresponding to the convolutional neural network. Deep learning has been widely used in the fields of image processing, speech recognition, natural language processing, radio signal recognition, and the like as an end-to-end machine learning method capable of automatically learning features from data. According to the spectrum sensing method based on the convolutional neural network construction, the spectrum sensing accuracy rate in the actual environment is improved.
Preferably, the convolutional neural network comprises an input layer, an intermediate layer and an output layer; the size of the input layer is the same as the size of the vector matrix; the middle layer comprises a convolution layer and an activation function Rule; the output layer comprises a softmax classification layer.
Specifically, as shown in the structure diagram of the convolutional neural network of fig. 2, the convolutional neural network includes an input layer, an intermediate layer, and an output layer, in which an output of an upper layer serves as an input of a lower layer. For example, in the present application, the size of the input layer may be 512 rows and 2 columns, then M in the aforementioned step S1 is equal to 512, the input layer includes a convolutional layer and an activation function Rule, taking "15 × 2, conv, 128" as an example, conv represents the convolutional layer, a number (15 × 2) before conv represents the size of the convolution kernel, and a number (128) after conv represents the number of convolution kernels. The ReLU active layer represents the rectifying linear unit activation. The convolutional neural network further comprises a fully connected layer fc and a Dropout layer, wherein a number (256) after the fully connected layer fc represents the number of neurons; the number (0.6) in parentheses of the Dropout layer indicates the Dropout probability. The output layer of the convolutional neural network is a softmax classification layer. Meanwhile, a normalization layer is also included between the convolution layer and the nonlinear activation layer ReLU of the convolutional neural network. After the convolutional neural network is constructed, the convolutional neural network can be trained and verified through sample data.
Preferably, the network training is performed on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network, and the method comprises the following steps:
selecting T sample data from the N sample data as training data, and taking the rest N-T sample data as verification data, wherein T is more than 1 and less than N;
inputting training data into a convolutional neural network for network training to obtain a trained convolutional neural network;
inputting the verification data into the trained convolutional neural network to obtain the optimal confidence corresponding to the verification data, and obtaining the optimal network structure corresponding to the convolutional neural network based on the optimal confidence.
For example, after obtaining sample data based on step S1, the sample data may be divided into training data and verification data according to a ratio of 8: 1. Inputting training data into a convolutional neural network for network training to obtain a trained convolutional neural network, sequentially inputting verification data into the trained convolutional neural network, obtaining corresponding confidence coefficient and network weight parameter by each verification data, comparing all confidence coefficients obtained according to the verification data with confidence thresholds respectively to obtain optimal confidence coefficient, and finally taking the network weight parameter corresponding to the optimal confidence coefficient as a network weight parameter of an optimal network structure corresponding to the convolutional neural network to obtain the optimal network structure corresponding to the convolutional neural network.
By constructing and training the convolutional neural network, an optimal network structure corresponding to the convolutional neural network is obtained, technical support is provided for later spectrum sensing of signal data to be detected, the problem of low accuracy obtained by the existing spectrum sensing method is solved, and the accuracy of spectrum sensing is improved.
And S3, inputting the signal data to be detected into an optimal network structure corresponding to the convolutional neural network to obtain a confidence coefficient corresponding to the signal data to be detected, and obtaining a frequency spectrum sensing result based on the confidence coefficient. After the optimal network structure corresponding to the convolutional neural network is obtained through training, the acquired signal data to be detected can be input into the optimal network structure corresponding to the convolutional neural network, and finally a spectrum sensing result is obtained.
Preferably, the acquiring the signal data to be detected comprises the following steps:
acquiring a signal sampling sequence to be detected and normalizing the signal sampling sequence to be detected to obtain a normalized signal sampling sequence to be detected;
obtaining a power spectrum of a signal to be detected, a low-frequency component of the signal to be detected and a high-frequency component of the signal to be detected based on the normalized sampling sequence of the signal to be detected, and splicing the low-frequency component of the signal to be detected and the high-frequency component of the signal to be detected to obtain a wavelet vector of the signal to be detected;
and obtaining a vector matrix of the signal to be detected based on the normalized sampling sequence of the signal to be detected, the power spectrum of the signal to be detected and the wavelet vector of the signal to be detected, wherein the vector matrix of the signal to be detected is the data of the signal to be detected.
Specifically, the method of acquiring signal data to be detected is the same as the method of acquiring sample data. Firstly, a signal sampling sequence to be detected is collected and normalized to obtain a normalized signal sampling sequence to be detected, and a signal vector matrix to be detected is finally obtained according to the normalized signal sampling sequence to be detected, wherein the signal vector matrix to be detected is signal data to be detected.
Preferably, obtaining the spectrum sensing result based on the confidence level includes the following steps:
inputting the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain the confidence coefficient t corresponding to the signal data to be detectednoiseAnd judging the confidence coefficient tnoiseWhether or not 1-t is satisfiednoiseGamma is more than gamma, and gamma is a judgment threshold;
and if the signal data to be detected is not the signal, the spectrum sensing result corresponding to the signal data to be detected is no signal.
Specifically, the signal data to be detected is obtained and then input into the optimal network structure corresponding to the convolutional neural network to obtain confidence, and whether the confidence is present or not is judgedSatisfies 1-tnoiseAnd gamma is greater than gamma, if the condition is met, the spectrum sensing result is a signal, namely the acquired signal to be detected comprises a main user signal, and if the condition is not met, the spectrum sensing result is a no signal, namely the acquired signal data to be detected is a pure noise signal.
Preferably, the decision threshold is obtained based on the following steps:
obtaining N-T corresponding confidence degrees based on the N-T verification data;
sequentially arranging the confidence coefficients from large to small, and selecting the ith confidence coefficient as a decision threshold; wherein the content of the first and second substances,
i＝ceil(pfp)
in the formula, ceil is rounded up, pfTo verify the pure noise data contained in the data, p is the verification data. The verification data is the verification data obtained by dividing based on the sample data in step S2, and a label has been added to the sample data in the process of obtaining the sample data, so that how many items of data in the verification data are pure noise data is known, that is, p in the above formula is pfIn known amounts. And obtaining i based on the calculation formula, and taking the corresponding confidence coefficient as a judgment threshold. The confidence coefficient of the signal data to be detected is obtained through the optimal network structure corresponding to the convolutional neural network, whether the confidence coefficient meets a judgment threshold is judged, and then a spectrum sensing result is obtained.
Compared with the prior art, the spectrum sensing method based on deep learning classification provided by the embodiment, by collecting sample data and utilizing the collected sample data to carry out network training on the constructed convolutional neural network, to obtain the optimal network structure corresponding to the convolutional neural network, and finally inputting the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain a spectrum sensing result, the spectrum sensing result is mainly used for reflecting the existence condition of the main user signal of the signal to be detected, is simple and easy to implement, solves the problem that the existing spectrum sensing method is low in accuracy due to the fact that the design of judgment statistics of the existing spectrum sensing method needs a large number of manual processes and depends on too much professional knowledge, achieves automation of spectrum sensing, reduces the dependence on the professional knowledge, and improves the accuracy of the spectrum sensing result.
In another embodiment of the present invention, a spectrum sensing apparatus based on deep learning classification is disclosed, as shown in fig. 3, including a sample obtaining module 100 for obtaining sample data of a signal; the sample data comprises a signal sampling sequence, a signal power spectrum and a wavelet vector; the network obtaining module 200 is used for constructing a convolutional neural network, and performing network training on the convolutional neural network based on sample data to obtain an optimal network structure corresponding to the convolutional neural network; the spectrum sensing module 300 inputs the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain the confidence corresponding to the signal data to be detected, and obtains a spectrum sensing result based on the confidence.
Specifically, the steps of obtaining the spectrum sensing result by the spectrum sensing device and the spectrum sensing method are the same, and are not described herein again. The spectrum sensing result is obtained through the spectrum sensing device based on deep learning classification, the spectrum sensing method is simple and easy to implement, the problem that the existing spectrum sensing method is low in accuracy due to the fact that the design of judgment statistics of the existing spectrum sensing method needs a large number of manual processes and depends on too much professional knowledge is solved, the spectrum sensing automation is achieved, dependence on the professional knowledge is reduced, and the accuracy of the spectrum sensing result is improved.
Preferably, the sample acquisition module performs the following procedure:
sampling and normalizing the initial signal to obtain a normalized signal sampling sequence; the initial signal comprises a pure noise signal and a main user signal;
acquiring a signal power spectrum, a signal low-frequency component and a signal high-frequency component based on the normalized signal sampling sequence, and splicing the signal low-frequency component and the signal high-frequency component to obtain a wavelet vector;
obtaining a vector matrix based on the normalized signal sampling sequence, the signal power spectrum and the wavelet vector;
and adding a label to the vector matrix to obtain sample data, wherein the label comprises a signal and a no-signal, the initial signal with the signal comprises a main user signal, and the initial signal without the signal is a pure noise signal.
The initial signal containing the main user signal and the pure noise signal is processed through the sample acquisition module to obtain sample data, and support and basis are provided for training the constructed convolutional neural network in the later period.
Preferably, the convolutional neural network comprises an input layer, an intermediate layer and an output layer;
the size of the input layer is the same as the size of the vector matrix;
the middle layer comprises a convolution layer and an activation function Rule;
the output layer comprises a softmax classification layer.
By constructing and training the convolutional neural network, an optimal network structure corresponding to the convolutional neural network is obtained, technical support is provided for later spectrum sensing of signal data to be detected, the problem of low accuracy obtained by the existing spectrum sensing method is solved, and the accuracy of spectrum sensing is improved.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (10)
1. A spectrum sensing method based on deep learning classification is characterized by comprising the following steps:
acquiring sample data of a signal; the sample data comprises a signal sampling sequence, a signal power spectrum and a wavelet vector;
constructing a convolutional neural network, and carrying out network training on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network;
and inputting the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain the confidence coefficient corresponding to the signal data to be detected, and obtaining a frequency spectrum sensing result based on the confidence coefficient.
2. The deep learning classification-based spectrum sensing method according to claim 1, wherein the signal sample sequence in the sample data is a normalized signal sample sequence; the method for acquiring the sample data comprises the following steps:
sampling and normalizing the initial signal to obtain a normalized signal sampling sequence; the initial signal comprises a pure noise signal and a main user signal;
acquiring a signal power spectrum, a signal low-frequency component and a signal high-frequency component based on the normalized signal sampling sequence, and splicing the signal low-frequency component and the signal high-frequency component to obtain a wavelet vector;
obtaining a vector matrix based on the normalized signal sampling sequence, the signal power spectrum and the wavelet vector;
and adding a label to the vector matrix to obtain sample data, wherein the label comprises a signal and a no-signal, the initial signal with the signal comprises a main user signal, and the initial signal without the signal is a pure noise signal.
3. The deep learning classification-based spectrum sensing method according to claim 1, wherein the convolutional neural network comprises an input layer, an intermediate layer and an output layer;
the size of the input layer is the same as the size of the vector matrix;
the middle layer comprises a convolution layer and an activation function Rule;
the output layer comprises a softmax classification layer.
4. The spectrum sensing method based on deep learning classification according to claim 2 or 3, wherein the network training is performed on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network, and the method comprises the following steps:
selecting T sample data from the N sample data as training data, and taking the rest N-T sample data as verification data, wherein T is more than 1 and less than N;
inputting the training data into a convolutional neural network for network training to obtain a trained convolutional neural network;
and inputting the verification data into the trained convolutional neural network to obtain the optimal confidence corresponding to the verification data, and obtaining the optimal network structure corresponding to the convolutional neural network based on the optimal confidence.
5. The spectrum sensing method based on deep learning classification as claimed in claim 4, wherein the obtaining of the signal data to be detected comprises the following steps:
acquiring a signal sampling sequence to be detected and normalizing the signal sampling sequence to be detected to obtain a normalized signal sampling sequence to be detected;
obtaining a power spectrum of the signal to be detected, a low-frequency component of the signal to be detected and a high-frequency component of the signal to be detected based on the normalized sampling sequence of the signal to be detected, and splicing the low-frequency component of the signal to be detected and the high-frequency component of the signal to be detected to obtain a wavelet vector of the signal to be detected;
and obtaining a vector matrix of the signal to be detected based on the normalized sampling sequence of the signal to be detected, the power spectrum of the signal to be detected and the wavelet vector of the signal to be detected, wherein the vector matrix of the signal to be detected is the data of the signal to be detected.
6. The spectrum sensing method based on deep learning classification as claimed in claim 5, wherein the obtaining of the spectrum sensing result based on the confidence coefficient comprises the following steps:
inputting the signal data to be detected into an optimal network structure corresponding to the convolutional neural network to obtain a confidence coefficient t corresponding to the signal data to be detectednoiseAnd judging the confidence coefficient tnoiseWhether or not 1-t is satisfiednoiseGamma is more than gamma, and gamma is a judgment threshold;
and if the signal data to be detected is not the signal, the spectrum sensing result corresponding to the signal data to be detected is no signal.
7. The spectrum sensing method based on deep learning classification as claimed in claim 6, wherein the decision threshold is obtained based on the following steps:
obtaining N-T corresponding confidence degrees based on the N-T verification data;
sequentially arranging the confidence coefficients from large to small, and selecting the ith confidence coefficient as a decision threshold; wherein the content of the first and second substances,
i＝ceil(pf/p)
in the formula, ceil is rounded up, pfTo verify the pure noise data contained in the data, p is the verification data.
8. A spectrum sensing apparatus based on deep learning classification, comprising:
the sample acquisition module acquires sample data of the signal; the sample data comprises a signal sampling sequence, a signal power spectrum and a wavelet vector;
the network acquisition module is used for constructing a convolutional neural network and carrying out network training on the convolutional neural network based on the sample data to obtain an optimal network structure corresponding to the convolutional neural network;
and the frequency spectrum sensing module is used for inputting the signal data to be detected into the optimal network structure corresponding to the convolutional neural network to obtain the confidence coefficient corresponding to the signal data to be detected and obtain a frequency spectrum sensing result based on the confidence coefficient.
9. The deep learning classification-based spectrum sensing device according to claim 6, wherein the sample obtaining module performs the following procedures:
sampling and normalizing the initial signal to obtain a normalized signal sampling sequence; the initial signal comprises a pure noise signal and a main user signal;
acquiring a signal power spectrum, a signal low-frequency component and a signal high-frequency component based on the normalized signal sampling sequence, and splicing the signal low-frequency component and the signal high-frequency component to obtain a wavelet vector;
obtaining a vector matrix based on the normalized signal sampling sequence, the signal power spectrum and the wavelet vector;
and adding a label to the vector matrix to obtain sample data, wherein the label comprises a signal and a no-signal, the initial signal with the signal comprises a main user signal, and the initial signal without the signal is a pure noise signal.
10. The deep learning classification-based spectrum sensing device according to claim 9, wherein the convolutional neural network comprises an input layer, an intermediate layer and an output layer;
the size of the input layer is the same as the size of the vector matrix;
the middle layer comprises a convolution layer and an activation function Rule;
the output layer comprises a softmax classification layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010974518.0A CN112101460A (en) | 2020-09-16 | 2020-09-16 | Spectrum sensing method and device based on deep learning classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010974518.0A CN112101460A (en) | 2020-09-16 | 2020-09-16 | Spectrum sensing method and device based on deep learning classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112101460A true CN112101460A (en) | 2020-12-18 |
Family
ID=73760193
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010974518.0A Pending CN112101460A (en) | 2020-09-16 | 2020-09-16 | Spectrum sensing method and device based on deep learning classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112101460A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115642968A (en) * | 2022-12-22 | 2023-01-24 | 中南大学 | Communication interference signal and spectrum cavity joint cognition method, device and medium based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107547460A (en) * | 2017-08-21 | 2018-01-05 | 西安电子科技大学 | Radio communication Modulation Signals Recognition method based on deep learning |
| CN108234370A (en) * | 2017-12-22 | 2018-06-29 | 西安电子科技大学 | Modulation mode of communication signal recognition methods based on convolutional neural networks |
| US20180324595A1 (en) * | 2017-05-05 | 2018-11-08 | Ball Aerospace & Technologies Corp. | Spectral sensing and allocation using deep machine learning |
| CN109450573A (en) * | 2018-12-17 | 2019-03-08 | 电子科技大学 | A kind of frequency spectrum sensing method based on deep neural network |
-
2020
- 2020-09-16 CN CN202010974518.0A patent/CN112101460A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180324595A1 (en) * | 2017-05-05 | 2018-11-08 | Ball Aerospace & Technologies Corp. | Spectral sensing and allocation using deep machine learning |
| CN107547460A (en) * | 2017-08-21 | 2018-01-05 | 西安电子科技大学 | Radio communication Modulation Signals Recognition method based on deep learning |
| CN108234370A (en) * | 2017-12-22 | 2018-06-29 | 西安电子科技大学 | Modulation mode of communication signal recognition methods based on convolutional neural networks |
| CN109450573A (en) * | 2018-12-17 | 2019-03-08 | 电子科技大学 | A kind of frequency spectrum sensing method based on deep neural network |
Non-Patent Citations (7)
| Title |
|---|
| P.Y. DIBAL ET AL.: "Application of wavelet transform in spectrum sensing for cognitive radio: A survey", 《PHYSICAL COMMUNICATION》, vol. 28, 30 June 2018 (2018-06-30), pages 45 - 57 * |
| SHILIAN ZHENG ET AL.: "Spectrum Sensing Based on Deep Learning Classification for Cognitive Radios", 《CHINA COMMUNICATIONS》, vol. 17, no. 2, 28 February 2020 (2020-02-28), pages 1 - 4, XP011775874, DOI: 10.23919/JCC.2020.02.012 * |
| 许达: "基于频谱数据的时频域预测算法研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》, 1 April 2019 (2019-04-01), pages 3 * |
| 许达: "基于频谱数据的时频域预测算法研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》, no. 12, 15 December 2019 (2019-12-15), pages 3 * |
| 郑仕链等: "用于认知无线电协作频谱感知的混合蛙跳算法群体初始化技术", 《物理学报》, vol. 62, no. 7, 8 April 2013 (2013-04-08), pages 1 - 6 * |
| 韩冬: "基于深度学习的频谱感知方法研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》, 1 June 2018 (2018-06-01), pages 4 * |
| 韩冬: "基于深度学习的频谱感知方法研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》, no. 01, 15 January 2019 (2019-01-15), pages 4 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115642968A (en) * | 2022-12-22 | 2023-01-24 | 中南大学 | Communication interference signal and spectrum cavity joint cognition method, device and medium based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021160189A1 (en) | Csi method for recognizing human fall in wi-fi interference environment | |
| Zeng et al. | Spectrum analysis and convolutional neural network for automatic modulation recognition | |
| Zhou et al. | Adaptive genetic algorithm-aided neural network with channel state information tensor decomposition for indoor localization | |
| CN107124381B (en) | Automatic identification method for digital communication signal modulation mode | |
| CN109890043B (en) | Wireless signal noise reduction method based on generative countermeasure network | |
| Wu et al. | A deep learning approach for modulation recognition via exploiting temporal correlations | |
| Lin et al. | Learning of time-frequency attention mechanism for automatic modulation recognition | |
| CN111585922A (en) | Modulation mode identification method based on convolutional neural network | |
| CN111800359B (en) | Method, device, equipment and medium for identifying communication signal modulation mode | |
| CN114157539B (en) | Data-aware dual-drive modulation intelligent identification method | |
| CN112069883A (en) | Deep learning signal classification method fusing one-dimensional and two-dimensional convolutional neural network | |
| Sang et al. | Application of novel architectures for modulation recognition | |
| CN109512390A (en) | Sleep stage method and wearable device based on EEG time domain various dimensions feature and M-WSVM | |
| CN112910812A (en) | Modulation mode identification method for deep learning based on space-time feature extraction | |
| CN113315593A (en) | Frequency spectrum sensing algorithm based on FLOM covariance matrix and LSTM neural network | |
| CN112101460A (en) | Spectrum sensing method and device based on deep learning classification | |
| CN115186712A (en) | Modulated signal identification method and system | |
| CN114980122A (en) | Small sample radio frequency fingerprint intelligent identification system and method | |
| CN114255487A (en) | Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction | |
| CN109309538A (en) | A kind of frequency spectrum sensing method, device, equipment, system and storage medium | |
| CN116738278A (en) | Automatic modulation classification method based on wavelet transformation and multi-mode feature fusion | |
| CN112580598B (en) | Radio signal classification method based on multichannel Diffpool | |
| CN113259289B (en) | Single-channel aliasing signal modulation mode identification method based on residual error neural network | |
| CN113076925B (en) | M-QAM signal modulation mode identification method based on CNN and ELM | |
| CN111917674A (en) | Modulation identification method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |