CN114024808A - Modulation signal identification method and system based on deep learning - Google Patents
Modulation signal identification method and system based on deep learning Download PDFInfo
- Publication number
- CN114024808A CN114024808A CN202111107840.4A CN202111107840A CN114024808A CN 114024808 A CN114024808 A CN 114024808A CN 202111107840 A CN202111107840 A CN 202111107840A CN 114024808 A CN114024808 A CN 114024808A
- Authority
- CN
- China
- Prior art keywords
- noise
- modulation signal
- neural network
- modulation
- deep neural
- 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
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 238000001228 spectrum Methods 0.000 claims abstract description 80
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 78
- 238000013528 artificial neural network Methods 0.000 claims abstract description 75
- 238000001914 filtration Methods 0.000 claims abstract description 49
- 238000010586 diagram Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 32
- 230000009467 reduction Effects 0.000 claims abstract description 29
- 238000012545 processing Methods 0.000 claims abstract description 19
- 239000011159 matrix material Substances 0.000 claims description 46
- 230000008569 process Effects 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 15
- 230000007274 generation of a signal involved in cell-cell signaling Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 9
- 230000003595 spectral effect Effects 0.000 claims description 9
- 238000005316 response function Methods 0.000 claims description 6
- 230000000737 periodic effect Effects 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 3
- 238000005311 autocorrelation function Methods 0.000 claims 1
- 238000000605 extraction Methods 0.000 description 7
- 230000007547 defect Effects 0.000 description 3
- 238000007476 Maximum Likelihood Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 101150049479 CCNC gene Proteins 0.000 description 1
- 241000160765 Erebia ligea Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0012—Modulated-carrier systems arrangements for identifying the type of modulation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a modulation signal identification method based on deep learning, which comprises the following steps: generating different kinds of modulation signals containing noise; carrying out wiener filtering noise reduction on the modulation signal containing the noise; performing cyclic spectrum estimation on the modulated signal subjected to noise reduction, and extracting a cyclic spectrum two-dimensional sectional view; constructing a deep neural network, inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network, and training the deep neural network; and identifying the modulation mode of the unknown signal by using the trained deep neural network. The invention also provides a modulation signal identification system based on deep learning, which is characterized in that the noise reduction processing is carried out on the modulation signal through wiener filtering, so that the influence of noise on the identification precision can be effectively reduced; meanwhile, the two-dimensional section diagram of the cyclic spectrum is used as an input characteristic, so that on one hand, the influence of noise on an identification result can be effectively reduced due to the fact that the two-dimensional section diagram of the cyclic spectrum is insensitive to the noise, on the other hand, the complexity of an algorithm can be greatly reduced, and the identification efficiency is improved.
Description
Technical Field
The invention relates to the technical field of communication technology application, in particular to a modulation signal identification method and system based on deep learning.
Background
The purpose of modulation identification techniques is to identify modulation schemes from signals received or intercepted in a short time in civilian or military applications, and thus they can give the radio transceiver cognitive capabilities in application scenarios such as surveillance, electronic warfare, etc. Modulation recognition algorithms can be divided into two main types, one being maximum likelihood based algorithms and the other being feature extraction based algorithms. The maximum likelihood method can achieve better recognition accuracy than the feature extraction method, but requires prior knowledge of Channel State Information (CSI) or statistical properties of the received signal. In contrast, the feature extraction system does not require priority information, and therefore is better in practicality.
In the existing application, modulation identification is often required to be realized by performing blind extraction on artificial features such as carrier frequency, instantaneous amplitude, phase and the like of a received signal or an intercepted signal. Since these parameters, which directly characterize the signal, are sensitive to noise and interference, researchers have attempted to extract features of the signal in an indirect manner. For example, spectral features are extracted using wavelet transform, and it is proposed to evaluate statistical features using spectral circulation functions, high order moments, and cumulants.
In recent years, related technologies have been used to define the cyclostationary characteristics of signals and identify the modulation mode of digital signals by using the cyclostationary spectrum characteristics of signals. Several calculation methods for estimating the Cyclic spectrum characteristics are proposed in the documents [1] R.S.Roberts, W.A.Brown and H.H.Loomy, Jr, computational efficiency Algorithms for Cyclic Spectral Analysis, IEEE Signal Processing Magazine, vol.8, No.2, pp.38-49, April 1991.doi:10.1109/79.81008. Ramkumar, Automatic modulation classification for coherent radio using cyclic feature detection, IEEE Circuits and Systems Magazine, vol.9, No.2, pp.27-45, Second quater 2009, doi:10.1109/MCAS 2008.931739. modulation type identification of signals is realized by using cyclic spectrum feature analysis and a classifier based on a decision tree, but the method needs to manually extract several features of a cyclic spectrum, and the operation complexity is high. In order to overcome the problem of high computational complexity, the documents [3] z.wu, e.like and v.chakravarty, "Reliable Modulation Classification at Low SNR Using Spectral Correlation,"20074th IEEE condition Communications and Networking reference, 2007, pp.1134-1138, doi:10.1109/ccnc.2007.228 "analyze that the extraction is performed by Using an alpha section and an f section in a cyclic spectrum feature, and the section feature is classified by Using a deep learning neural network.
Disclosure of Invention
In order to solve at least one technical defect, the invention provides a modulation signal identification method and system based on deep learning, which can reduce the operation complexity and improve the identification precision of the modulation mode of the modulation signal.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a modulation signal identification method based on deep learning comprises the following steps:
s1: generating different kinds of modulation signals containing noise;
s2: carrying out wiener filtering on the modulation signal containing the noise, and carrying out noise reduction processing;
s3: performing cyclic spectrum estimation on the modulated signal subjected to noise reduction, and extracting a cyclic spectrum two-dimensional sectional view;
s4: constructing a deep neural network, inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network, and training the deep neural network;
s5: and identifying the modulation mode of the unknown signal by using the trained deep neural network.
In the scheme, the noise reduction processing is carried out on the modulation signal through wiener filtering, so that the influence of noise on the identification precision can be effectively reduced; meanwhile, the two-dimensional section diagram of the cyclic spectrum is used as an input characteristic, so that on one hand, the influence of noise on an identification result can be effectively reduced due to the fact that the two-dimensional section diagram of the cyclic spectrum is insensitive to the noise, on the other hand, the complexity of an algorithm can be greatly reduced, and the identification efficiency is improved.
In order to improve the identification precision and robustness, the scheme provides that a data set which reflects the statistical characteristics of signals and is insensitive to interference is used for forming a data set to train the neural network; in addition, a method of wiener filtering is also proposed to suppress noise so as to improve the accuracy of modulation identification. And then extracting the two-dimensional SCF contour to reduce the size of the data set, and effectively overcoming the defects of high calculation complexity and large noise interference of the existing characteristic parameters in the extraction process.
Wherein, the step S1 specifically includes: selecting code element rate and carrier signal to generate different kinds of modulation signal, and transmitting the modulation signal through white Gaussian noise channel to obtain the modulation signal containing noise.
In step S2, the process of performing wiener filtering on the modulation signal specifically includes:
calculating a cross-correlation matrix R of the modulation signal s (n) and the noise-containing modulation signal x (n)xs(j) The concrete formula is as follows:
in the formula, Rxs(j) A cross-correlation matrix representing the modulation signal s (N) and the noise-containing modulation signal x (N), N representing the filter length, and h (m) representing the frequency response function of the filter.
Then, calculate the autocorrelation matrix R of the modulation signal x (n) containing noisexxThe concrete formula is as follows:
in the formula, Rxx(j) An autocorrelation matrix representing a noisy signal.
The filter matrix H of the wiener filtering is specifically represented as:
in the formula, rxsRepresenting a cross-correlation matrix Rxs(j) The column vector of the first column of (1).
And filtering the modulation signal containing the noise through a filtering matrix to realize the noise reduction processing of the modulation signal.
In step S3, performing cyclic spectrum estimation on the noise-reduced modulation signal, and specifically expressing a calculation formula for extracting a cyclic spectrum two-dimensional cross-sectional diagram as follows:
wherein: f is the frequency spectrum and alpha is the period frequency.
In order to be a function of the cyclic auto-correlation,<·>tis a time average statistic;
and then extracting a two-dimensional section diagram of the cyclic spectrum, which specifically comprises the following steps:
wherein f is the frequency spectrum, alpha is the period frequency, fsIs the maximum periodic frequency, fmaxThe maximum spectral frequency.
Wherein, in the step S4, the training process of the deep neural network includes an offline training phase and an online deployment phase; wherein:
in the off-line training stage, a plurality of cyclic spectrum two-dimensional cross-sectional views obtained in the steps S1-S3 are used as a data set, and the data set is divided into a training data set and a verification data set; performing off-line training on the deep neural network by using a training data set to obtain bn weight and variable weight of the deep neural network;
in an online deployment stage, the acquired bn weight and the variable weight are used as parameters of the deep neural network to complete the deployment of the deep neural network, so that the deep neural network can identify the modulation mode of any input modulation signal.
The scheme also provides a modulation signal recognition system based on deep learning, which comprises a noise-containing modulation signal generation module, a wiener filtering module, a cyclic spectrum estimation module, a neural network construction module, a neural network training module and a recognition module; wherein:
the noise-containing modulation signal generation module is used for generating different types of noise-containing modulation signals;
the wiener filtering module is used for carrying out wiener filtering on the modulation signal containing the noise and carrying out noise reduction processing;
the cyclic spectrum estimation module is used for carrying out cyclic spectrum estimation on the modulated signal subjected to noise reduction and extracting a cyclic spectrum two-dimensional sectional diagram;
the neural network construction module is used for constructing a deep neural network;
the neural network training module is used for inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network and training the deep neural network;
the identification module is used for reserving the trained deep neural network and identifying the modulation mode of the unknown signal.
Wherein, the noise-containing modulation signal generation module comprises a signal generation unit and a white gaussian noise channel unit, wherein: the signal generating unit generates different types of modulation signals according to the selected code element rate and the carrier signal, and the modulation signals are transmitted through a Gaussian white noise channel provided by the Gaussian white noise channel unit, so that the modulation signals containing noise are obtained.
In the wiener filtering module, the process of performing wiener filtering on the modulation signal specifically includes:
calculating a cross-correlation matrix R of the modulation signal s (n) and the noise-containing modulation signal x (n)xs(j) The concrete formula is as follows:
in the formula, Rxs(j) A cross-correlation matrix representing the modulation signal s (N) and the noise-containing modulation signal x (N), N representing the filter lengthDegree, h (m) represents the frequency response function of the filter.
Then, calculate the autocorrelation matrix R of the modulation signal x (n) containing noisexxThe concrete formula is as follows:
in the formula, Rxx(j) An autocorrelation matrix representing a noisy signal.
The filter matrix H of the wiener filtering is specifically represented as:
in the formula, rxsRepresenting a cross-correlation matrix Rxs(j) The column vector of the first column of (1).
And filtering the modulation signal containing the noise through a filtering matrix to realize the noise reduction processing of the modulation signal.
In the cyclic spectrum estimation module, cyclic spectrum estimation is performed on the modulated signal after noise reduction, and a calculation formula for extracting a cyclic spectrum two-dimensional sectional view is specifically expressed as:
wherein: f is the frequency spectrum and alpha is the period frequency.
In order to be a function of the cyclic auto-correlation,<·>tis a time average statistic;
and then extracting a two-dimensional section diagram of the cyclic spectrum, which specifically comprises the following steps:
wherein f is the frequency spectrum, alpha is the period frequency, fsIs the maximum periodic frequency, fmaxThe maximum spectral frequency.
In the neural network construction module, a softmax function is used as an output layer of the deep neural network, and an output matrix obtained after characteristic input is calculated by the neural network is converted into probability distribution in a range of [0,1 ].
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
drawings
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a schematic diagram of a wiener filtering algorithm framework according to an embodiment of the invention;
fig. 3 is a two-dimensional cross-sectional view of cyclic spectrum functions of five modulation signals of 2ASK,2FSK, BPSK, QPSK, and 4FSK according to an embodiment of the present invention;
FIG. 4 is a flow chart of a deep neural network according to an embodiment of the present invention;
fig. 5 is a schematic diagram illustrating the results of the modulation identification precision of five modulation signals 2ASK,2FSK, BPSK, QPSK, and 4FSK according to an embodiment of the present invention;
fig. 6 is a diagram illustrating the total modulation recognition accuracy result of five modulation signals according to an embodiment of the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
As shown in fig. 1, a modulation signal identification method based on deep learning includes the following steps:
s1: generating different kinds of modulation signals containing noise;
s2: carrying out wiener filtering on the modulation signal containing the noise, and carrying out noise reduction processing;
s3: performing cyclic spectrum estimation on the modulated signal subjected to noise reduction, and extracting a cyclic spectrum two-dimensional sectional view;
s4: constructing a deep neural network, inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network, and training the deep neural network;
s5: and identifying the modulation mode of the unknown signal by using the trained deep neural network.
In the specific implementation process, the scheme performs noise reduction processing on the modulation signal through wiener filtering, so that the influence of noise on the identification precision can be effectively reduced; meanwhile, the two-dimensional section diagram of the cyclic spectrum is used as an input characteristic, so that on one hand, the influence of noise on an identification result can be effectively reduced due to the fact that the two-dimensional section diagram of the cyclic spectrum is insensitive to the noise, on the other hand, the complexity of an algorithm can be greatly reduced, and the identification efficiency is improved.
In order to improve the identification precision and robustness, the scheme provides that a data set which reflects the statistical characteristics of signals and is insensitive to interference is used for forming a data set to train the neural network; in addition, a method of wiener filtering is also proposed to suppress noise so as to improve the accuracy of modulation identification. And then extracting the two-dimensional SCF contour to reduce the size of the data set, and effectively overcoming the defects of high calculation complexity and large noise interference of the existing characteristic parameters in the extraction process.
More specifically, the step S1 specifically includes: selecting a symbol rate and a carrier signal to generate different types of modulation signals, in this embodiment, five different types of modulation signals containing noise are generated, including 2ASK,2FSK, BPSK, QPSK, and 4FSK, and the generation parameters are: sampling frequency fs16kHz, carrier frequency fc10kHz, symbol rate fdThe modulated signal is then transmitted over a white gaussian noise channel to obtain a modulated signal containing noise.
More specifically, in step S2, as shown in fig. 2, the process of performing wiener filtering on the modulation signal specifically includes:
calculating a cross-correlation matrix R of the modulation signal s (n) and the noise-containing modulation signal x (n)xs(j) The concrete formula is as follows:
in the formula, Rxs(j) A cross-correlation matrix representing the modulation signal s (N) and the noise-containing modulation signal x (N), where s (N) is x (N) + v (N), v (N) is the noise added to the modulation signal, N represents the filter length, and h (m) represents the frequency response function of the filter.
Then, calculate the autocorrelation matrix R of the modulation signal x (n) containing noisexxThe concrete formula is as follows:
in the formula, Rxx(j) An autocorrelation matrix representing a noisy signal.
The filter matrix H of the wiener filtering is specifically represented as:
in the formula, rxsRepresenting a cross-correlation matrix Rxs(j) The column vector of the first column of (1).
Filtering the modulation signal containing noise through a filtering matrix to realize the noise reduction processing of the modulation signal, and obtaining a signal y (n), wherein e (n) is the error of the modulation signal x (n) containing noise and the signal y (n).
In step S3, performing cyclic spectrum estimation on the noise-reduced modulation signal, and specifically expressing a calculation formula for extracting a cyclic spectrum two-dimensional cross-sectional diagram as follows:
wherein: f is the frequency spectrum and alpha is the period frequency.
In order to be a function of the cyclic auto-correlation,<·>tis a time average statistic;
and then extracting a two-dimensional section diagram of the cyclic spectrum, which specifically comprises the following steps:
wherein f is the frequency spectrum, alpha is the period frequency, fsIs the maximum periodic frequency, fmaxThe maximum spectral frequency.
In the specific implementation process, the two-dimensional cross-sectional diagram of the extracted cyclic spectrum is shown in fig. 3 and is input into the deep neural network for training as a characteristic parameter.
More specifically, in the step S4, the training process of the deep neural network includes an offline training phase and an online deployment phase; wherein:
in the off-line training stage, a plurality of cyclic spectrum two-dimensional cross-sectional views obtained in the steps S1-S3 are used as a data set, and the data set is divided into a training data set and a verification data set; performing off-line training on the deep neural network by using a training data set to obtain bn weight and variable weight of the deep neural network;
in an online deployment stage, the acquired bn weight and the variable weight are used as parameters of the deep neural network to complete the deployment of the deep neural network, so that the deep neural network can identify the modulation mode of any input modulation signal.
In the specific implementation process, the specific execution flow of the deep neural network is shown in fig. 4:
the first layer is used as an input layer, and parameters of the cyclic spectrum two-dimensional sectional diagram are received and used as the input of the deep neural network; the second layer is a batch standardized bn (batch normalization) layer. The batch standardization has the function of normalizing the data to the effective range of the activation function as much as possible, so that each network input has the same distribution, the influence of the previous layer on the next layer is reduced, and the CN layer is used to ensure that the input parameter meets the condition that the mean value is 0 and the variance is 1; the third layer is a hidden layer, and nonlinear processing is performed by using a ReLU activation function, so that the network has certain sparsity, and the network and the operation speed are increased. The fourth layer is a BN layer, the fifth layer and the sixth layer are hidden layers, and the seventh layer uses a sofxmax function to convert output values of output matrixes of a plurality of neurons into probability distribution with the range of [0,1] and 1. And finally, generating an independent thermal matrix y [ n ] according to the obtained probability distribution, and outputting a predicted label, wherein the label is the modulation type.
Firstly, wiener filtering is applied to processing of a signal containing noise, so that noise interference is reduced; secondly, the extracted feature is SCF (cyclic spectral density function) which is insensitive to noise; in order to reduce the data set, alpha section characteristics of the cyclic spectrum of the data set are extracted to train a deep neural network, and the complexity of the algorithm is greatly reduced.
In a specific implementation process, as shown in fig. 5, the scheme can improve the modulation identification accuracy of cyclic spectrum features extracted from 2ASK,2FSK, BPSK, QPSK, and 4FSK signals after wiener filtering noise reduction to more than 90% under the condition that the SNR is as low as-25, while the highest identification accuracy of the original five signals based on the modulation identification accuracy of the cyclic spectrum can only reach about 60% under the condition that the SNR is as low as-25, and the modulation identification accuracy of the invention is greatly improved under the condition that the SNR is as high as-25 to-15.
In the specific implementation process, as shown in fig. 6, the scheme can improve the total modulation identification accuracy of five signals, namely 2ASK,2FSK, BPSK, QPSK, and 4FSK, to more than 90% when the SNR is as low as-25, and when the SNR is greater than-15, the identification accuracy reaches 100%, while when the SNR is as low as-25, the total identification accuracy is less than 30% and when the SNR is greater than-12, the identification accuracy can reach more than 90%, which indicates that the modulation identification accuracy is greatly improved.
Example 2
More specifically, on the basis of embodiment 1, the present disclosure further provides a modulation signal recognition system based on deep learning, which is used to implement a modulation signal recognition method based on deep learning, and specifically includes a noise-containing modulation signal generation module, a wiener filtering module, a cyclic spectrum estimation module, a neural network construction module, a neural network training module, and a recognition module; wherein:
the noise-containing modulation signal generation module is used for generating different types of noise-containing modulation signals;
the wiener filtering module is used for carrying out wiener filtering on the modulation signal containing the noise and carrying out noise reduction processing;
the cyclic spectrum estimation module is used for carrying out cyclic spectrum estimation on the modulated signal subjected to noise reduction and extracting a cyclic spectrum two-dimensional sectional diagram;
the neural network construction module is used for constructing a deep neural network;
the neural network training module is used for inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network and training the deep neural network;
the identification module is used for reserving the trained deep neural network and identifying the modulation mode of the unknown signal.
More specifically, the noise-containing modulation signal generation module includes a signal generation unit and a white gaussian noise channel unit, wherein: the signal generating unit generates different types of modulation signals according to the selected code element rate and the carrier signal, and the modulation signals are transmitted through a Gaussian white noise channel provided by the Gaussian white noise channel unit, so that the modulation signals containing noise are obtained.
More specifically, in the wiener filtering module, the process of performing wiener filtering on the modulation signal specifically includes:
calculating a cross-correlation matrix R of the modulation signal s (n) and the noise-containing modulation signal x (n)xs(j) The concrete formula is as follows:
in the formula, Rxs(j) A cross-correlation matrix representing the modulation signal s (N) and the noise-containing modulation signal x (N), where s (N) is x (N) + v (N), v (N) is the noise added to the modulation signal, N represents the filter length, and h (m) represents the frequency response function of the filter.
Then, calculate the autocorrelation matrix R of the modulation signal x (n) containing noisexxThe concrete formula is as follows:
in the formula, Rxx(j) An autocorrelation matrix representing a noisy signal.
The filter matrix H of the wiener filtering is specifically represented as:
in the formula, rxsRepresenting a cross-correlation matrix Rxs(j) The column vector of the first column of (1).
Filtering the modulation signal containing noise through a filtering matrix to realize the noise reduction processing of the modulation signal, and obtaining a signal y (n), wherein e (n) is the error of the modulation signal x (n) containing noise and the signal y (n).
More specifically, in the cyclic spectrum estimation module, cyclic spectrum estimation is performed on the modulated signal after noise reduction, and a calculation formula for extracting a cyclic spectrum two-dimensional cross-sectional diagram is specifically expressed as:
wherein: f is the frequency spectrum and alpha is the period frequency.
In order to be a function of the cyclic auto-correlation,<·>tis a time average statistic;
and then extracting a two-dimensional section diagram of the cyclic spectrum, which specifically comprises the following steps:
wherein f is the frequency spectrum, alpha is the period frequency, fsIs the maximum periodic frequency, fmaxThe maximum spectral frequency.
More specifically, in the neural network construction module, a softmax function is adopted as an output layer of the deep neural network, and an output matrix obtained after characteristic input is calculated by the neural network is converted into probability distribution in a range of [0,1 ].
In the specific implementation process, the system firstly carries out noise reduction processing on the modulation signal through wiener filtering, so that the influence of noise on the identification precision can be effectively reduced; meanwhile, the two-dimensional section diagram of the cyclic spectrum is used as an input characteristic, so that on one hand, the influence of noise on an identification result can be effectively reduced due to the fact that the two-dimensional section diagram of the cyclic spectrum is insensitive to the noise, on the other hand, the complexity of an algorithm can be greatly reduced, and the identification efficiency is improved.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A modulation signal identification method based on deep learning is characterized by comprising the following steps:
s1: generating different kinds of modulation signals containing noise;
s2: carrying out wiener filtering on the modulation signal containing the noise, and carrying out noise reduction processing;
s3: performing cyclic spectrum estimation on the modulated signal subjected to noise reduction, and extracting a cyclic spectrum two-dimensional sectional view;
s4: constructing a deep neural network, inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network, and training the deep neural network;
s5: and identifying the modulation mode of the unknown signal by using the trained deep neural network.
2. The method for recognizing the modulation signal based on the deep learning of claim 1, wherein the step S1 specifically includes: selecting code element rate and carrier signal to generate different kinds of modulation signal, and transmitting the modulation signal through white Gaussian noise channel to obtain the modulation signal containing noise.
3. The method as claimed in claim 1, wherein in step S2, the process of performing wiener filtering on the modulation signal specifically includes:
calculating a cross-correlation matrix R of the modulation signal s (n) and the noise-containing modulation signal x (n)xs(j) The concrete formula is as follows:
in the formula, Rxs(j) A cross-correlation matrix representing the modulation signal s (n) and the noise-containing modulation signal x (n); n represents the filter length; h (m) represents the frequency response function of the filter; the value range of j is [0, N-1 ]]M is a summation variable;
then, calculate the autocorrelation matrix R of the modulation signal x (n) containing noisexxThe concrete formula is as follows:
in the formula, Rxx(j) An autocorrelation matrix representing the noise-containing signal;
the filter matrix H of the wiener filtering is specifically represented as:
in the formula, rxsRepresenting a cross-correlation matrix Rxs(j) The column vector of the first column of (1).
And filtering the modulation signal containing the noise through a filtering matrix to realize the noise reduction processing of the modulation signal.
4. The method according to claim 3, wherein in step S3, the modulation signal after noise reduction is subjected to cyclic spectrum estimation, and the calculation formula for extracting the cyclic spectrum two-dimensional cross-sectional view is specifically represented as:
wherein:
where f is the frequency spectrum and α is the period frequency.
5. The method for recognizing the modulation signal based on the deep learning of claim 4, wherein in the step S4, the training process of the deep neural network comprises an off-line training phase and an on-line deployment phase; wherein:
in the off-line training stage, a plurality of cyclic spectrum two-dimensional cross-sectional views obtained in the steps S1-S3 are used as a data set, and the data set is divided into a training data set and a verification data set; performing off-line training on the deep neural network by using a training data set to obtain bn weight and variable weight of the deep neural network;
in an online deployment stage, the acquired bn weight and the variable weight are used as parameters of the deep neural network to complete the deployment of the deep neural network, so that the deep neural network can identify the modulation mode of any input modulation signal.
6. A modulation signal identification system based on deep learning is characterized by comprising a noise-containing modulation signal generation module, a wiener filtering module, a cyclic spectrum estimation module, a neural network construction module, a neural network training module and an identification module; wherein:
the noise-containing modulation signal generation module is used for generating different types of noise-containing modulation signals;
the wiener filtering module is used for carrying out wiener filtering on the modulation signal containing the noise and carrying out noise reduction processing;
the cyclic spectrum estimation module is used for carrying out cyclic spectrum estimation on the modulated signal subjected to noise reduction and extracting a cyclic spectrum two-dimensional sectional diagram;
the neural network construction module is used for constructing a deep neural network;
the neural network training module is used for inputting the two-dimensional section diagram of the cyclic spectrum as an input feature into the deep neural network and training the deep neural network;
the identification module is used for reserving the trained deep neural network and identifying the modulation mode of the unknown signal.
7. The system of claim 6, wherein the noise-containing modulation signal generation module comprises a signal generation unit and a white gaussian noise channel unit, and wherein: the signal generating unit generates different types of modulation signals according to the selected code element rate and the carrier signal, and the modulation signals are transmitted through a Gaussian white noise channel provided by the Gaussian white noise channel unit, so that the modulation signals containing noise are obtained.
8. The system according to claim 6, wherein in the wiener filtering module, the process of wiener filtering the modulation signal is specifically:
calculating a cross-correlation matrix R of the modulation signal s (n) and the noise-containing modulation signal x (n)xs(j) The concrete formula is as follows:
in the formula, Rxs(j) A cross-correlation matrix representing the modulation signal s (N) and the noise-containing modulation signal x (N), N representing the filter length, and h (m) representing the frequency response function of the filter.
Then, calculate the autocorrelation matrix R of the modulation signal x (n) containing noisexxThe concrete formula is as follows:
in the formula, Rxx(j) An autocorrelation matrix representing the noise-containing signal;
the filter matrix H of the wiener filtering is specifically represented as:
in the formula, rxsRepresenting a cross-correlation matrix Rxs(j) The column vector of the first column of (1).
And filtering the modulation signal containing the noise through a filtering matrix to realize the noise reduction processing of the modulation signal.
9. The system according to claim 6, wherein in the cyclic spectrum estimation module, the cyclic spectrum estimation is performed on the modulated signal after noise reduction, and a calculation formula for extracting a cyclic spectrum two-dimensional cross-sectional view is specifically represented as:
wherein: f is the frequency spectrum frequency, and alpha is the cycle frequency;
in order to be a function of the cyclic auto-correlation,<·>tis a time average statistic;
and then extracting a two-dimensional section diagram of the cyclic spectrum, which specifically comprises the following steps:
in the formula (f)sIs the maximum periodic frequency, fmaxThe maximum spectral frequency.
10. The method according to claim 9, wherein in the neural network construction module, a softmax function is used as an output layer of the deep neural network, and an output matrix obtained by calculating the characteristic input through the neural network is converted into a probability distribution in a range of [0,1 ].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111107840.4A CN114024808A (en) | 2021-09-22 | 2021-09-22 | Modulation signal identification method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111107840.4A CN114024808A (en) | 2021-09-22 | 2021-09-22 | Modulation signal identification method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114024808A true CN114024808A (en) | 2022-02-08 |
Family
ID=80054472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111107840.4A Pending CN114024808A (en) | 2021-09-22 | 2021-09-22 | Modulation signal identification method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114024808A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114818839A (en) * | 2022-07-01 | 2022-07-29 | 之江实验室 | Deep learning-based optical fiber sensing underwater acoustic signal identification method and device |
| CN117376067A (en) * | 2023-07-10 | 2024-01-09 | 电子科技大学 | Automatic modulation recognition method for composite modulation signal of aerospace measurement and control link |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109802905A (en) * | 2018-12-27 | 2019-05-24 | 西安电子科技大学 | Digital signal Automatic Modulation Recognition method based on CNN convolutional neural networks |
| CN110321953A (en) * | 2019-07-03 | 2019-10-11 | 中山大学 | Deep learning intelligent modulation recognition methods based on circulation Power estimation |
| CN111259798A (en) * | 2020-01-16 | 2020-06-09 | 西安电子科技大学 | Modulation signal identification method based on deep learning |
| CN113271273A (en) * | 2021-05-11 | 2021-08-17 | 中山大学 | Modulation identification method based on wiener filtering preprocessing |
-
2021
- 2021-09-22 CN CN202111107840.4A patent/CN114024808A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109802905A (en) * | 2018-12-27 | 2019-05-24 | 西安电子科技大学 | Digital signal Automatic Modulation Recognition method based on CNN convolutional neural networks |
| CN110321953A (en) * | 2019-07-03 | 2019-10-11 | 中山大学 | Deep learning intelligent modulation recognition methods based on circulation Power estimation |
| CN111259798A (en) * | 2020-01-16 | 2020-06-09 | 西安电子科技大学 | Modulation signal identification method based on deep learning |
| CN113271273A (en) * | 2021-05-11 | 2021-08-17 | 中山大学 | Modulation identification method based on wiener filtering preprocessing |
Non-Patent Citations (3)
| Title |
|---|
| LIN ZHANG等: ""Intelligent Denoising-Aided Deep Learning Modulation Recognition With Cyclic Spectrum Features for Higher Accuracy"", 《IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS》 * |
| 李红光等: ""基于时频能量谱纹理特征的跳频调制方式识别"", 《通信学报》 * |
| 管鹏辉等: ""一种基于循环谱分析的调制信号识别算法"", 《数据通信》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114818839A (en) * | 2022-07-01 | 2022-07-29 | 之江实验室 | Deep learning-based optical fiber sensing underwater acoustic signal identification method and device |
| CN114818839B (en) * | 2022-07-01 | 2022-09-16 | 之江实验室 | Deep learning-based optical fiber sensing underwater acoustic signal identification method and device |
| CN117376067A (en) * | 2023-07-10 | 2024-01-09 | 电子科技大学 | Automatic modulation recognition method for composite modulation signal of aerospace measurement and control link |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108509911B (en) | Interference signal identification method based on convolutional neural network | |
| CN111582320B (en) | Dynamic individual identification method based on semi-supervised learning | |
| CN114024808A (en) | Modulation signal identification method and system based on deep learning | |
| CN111310700B (en) | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature recognition | |
| CN112422213B (en) | Efficient spectrum sensing method based on support vector machine | |
| CN111050315B (en) | Wireless transmitter identification method based on multi-core two-way network | |
| Jeong et al. | Spectrogram-based automatic modulation recognition using convolutional neural network | |
| CN112347871A (en) | Interference signal modulation identification method for communication carrier monitoring system | |
| CN112395941B (en) | Radio signal radiation source individual identification method, computer device and storage medium | |
| CN114239749A (en) | Modulation identification method based on residual shrinkage and bidirectional long-short term memory network | |
| CN115964670B (en) | Spectrum anomaly detection method | |
| Shen et al. | Node identification in wireless network based on convolutional neural network | |
| CN111680737A (en) | Radar radiation source individual identification method under differential signal-to-noise ratio condition | |
| CN115982613A (en) | Signal modulation identification system and method based on improved convolutional neural network | |
| CN116866129A (en) | Wireless communication signal detection method | |
| CN112613443A (en) | Robustness communication radiation source intelligent identification method based on deep learning | |
| Huang et al. | Radar waveform recognition based on multiple autocorrelation images | |
| Ghasemzadeh et al. | GGCNN: An efficiency-maximizing gated graph convolutional neural network architecture for automatic modulation identification | |
| CN112821968B (en) | Efficient spectrum sensing method based on compressed sensing and support vector machine | |
| Wang et al. | A new method of automatic modulation recognition based on dimension reduction | |
| CN113076925A (en) | M-QAM signal modulation mode identification method based on CNN and ELM | |
| Gai et al. | Spectrum sensing method based on residual cellular network | |
| Liu et al. | Incremental learning based radio frequency fingerprint identification using intelligent representation | |
| CN113347175B (en) | Method and system for fingerprint feature extraction and equipment identity identification of optical communication equipment | |
| CN114580476A (en) | Unmanned aerial vehicle signal identification model construction method and corresponding identification method and system |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220208 |