CN111507205A - Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof - Google Patents
Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof Download PDFInfo
- Publication number
- CN111507205A CN111507205A CN202010231208.XA CN202010231208A CN111507205A CN 111507205 A CN111507205 A CN 111507205A CN 202010231208 A CN202010231208 A CN 202010231208A CN 111507205 A CN111507205 A CN 111507205A
- Authority
- CN
- China
- Prior art keywords
- module
- data
- signal
- deep learning
- output end
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Signal Processing (AREA)
- Filters That Use Time-Delay Elements (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses a modulation recognition device based on signal cycle spectrum and deep learning, which comprises a data preprocessing module, a data caching module, a signal data module, a deep learning training algorithm module, a model data module and a recognition algorithm module; adjusting the gain of a receiving link according to a transmitting link and a wireless channel to ensure that the signal power of the receiving end is in the dynamic range of an analog-to-digital converter and initializing the parameters of a deep neural network; converting the one-dimensional data format of the signal into a two-dimensional circular spectrum matrix data format, and storing the two-dimensional circular spectrum matrix data format in a data cache module in a fixed-length frame form; then storing the data into a signal data system, calling a deep learning training algorithm to obtain network optimal parameters, and storing the network optimal parameters in model data; and loading fixed-length frame data and model data, and calling an identification algorithm to obtain a modulation mode of the signal. The invention solves the problem of difficult signal feature extraction under low signal-to-noise ratio and improves the modulation identification effect under complex channel environment.
Description
Technical Field
The invention relates to the technical field of signal perception of a communication system, in particular to a modulation recognition device based on a signal cycle spectrum and deep learning and a use method thereof.
Background
In a platform with limited space, a large number of radar communication electronic warfare devices with different functions are deployed and are crowded in space. Besides receiving useful communication signals of friends, the communication receiver positioned on the platform also receives electromagnetic signals leaked by other electronic equipment on the platform and electromagnetic interference signals from an enemy platform, so that the electromagnetic environment is complex. The modulation mode is used as the most important characteristic of a communication radio station and the basis for further extracting other parameters of signals, and a signal modulation identification method under the condition of low signal to noise ratio is researched, so that the method has important significance for ensuring the smoothness of full link information of a platform in a complex environment. According to different implementation modes, the current modulation identification method mainly comprises the following steps: a likelihood ratio recognition method based on hypothesis testing and a statistical pattern recognition method based on feature extraction.
The likelihood ratio identification method based on hypothesis test is to establish hypothesis by using probability density function of random signal, determine cost function, find minimum cost through likelihood ratio, and establish decision criterion for various signal types. The difficulty lies in that certain prior probability information of a confirmed signal is needed, the calculation amount is large, the operation process is difficult to optimize, and the practicability in engineering practice is not high. The statistical pattern recognition method based on feature extraction can be divided into two steps of signal feature extraction and classifier construction, and common feature parameters comprise instantaneous amplitude phase features, high-order statistic features, wavelet transformation features, cyclostationary features and the like. However, they also have many defects, such as high signal-to-noise ratio requirement of transient features, precise carrier symbol synchronization required of high-order statistic features, poor scale robustness of wavelet features, and large computation load of cyclostationary features. At present, with increasingly complex signal systems and patterns, the traditional modulation identification method has increasingly outstanding problems, is limited in application under the environment of non-ideal channels, is limited in coverage of the considered signal modulation types, is difficult in engineering application, and is difficult to realize robust multi-type real-time identification.
In recent years, deep learning technology is widely applied and developed in the fields of character, image, voice recognition and the like, and the rapid development of signal modulation technology is promoted. Meanwhile, the software radio technology uses a powerful hardware platform and software programming to realize complex signal modulation and flexible setting of signal parameters, thereby bringing great challenges to the modulation identification technology.
Chinese patent discloses a software radio modulation signal identification platform and an identification method (application No. 201010516145.9), a communication signal feature extraction and modulation identification method (application No. 201210234727.7), a convolutional neural network-based modulation mode identification method (application No. 201810190856.8), a constellation orthogonal scanning feature-based modulation identification method (application No. 201811389534), an HSCA-based communication signal feature extraction modulation identification method (application No. 201811300069.0), and the like, or adopts a traditional statistical mode identification method, or only analyzes the application of a deep learning technology in identification, does not combine deep learning, software radio ideas and real channels, and cannot overcome the real-time modulation identification problem of large-bandwidth high-frequency communication signals under low signal-to-noise ratio or complex channels.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a modulation recognition device based on signal cycle spectrum and deep learning and a using method thereof, a strategy of collecting and preprocessing baseband signals by using a software radio thought and extracting the characteristics of a signal cycle spectrogram by using a deep learning technology is utilized, and the proposed structure has the advantages of noise resistance of the characteristics of the signal cycle spectrogram and the advantages of the recognition performance of the deep learning technology.
The invention provides a modulation identification device based on signal cycle spectrum and deep learning, which is characterized in that: the system comprises a data preprocessing module, a data caching module, a signal data module, a deep learning training algorithm module, a model data module and an identification algorithm module.
The input end of the data preprocessing module receives the external radio station modulation signal processed by the ADC module, the output end of the data preprocessing module is electrically connected with the input end of the data caching module, and the data preprocessing module is used for converting a received discrete sampling data format into a two-dimensional matrix data format of a gray cyclic spectrogram;
the output end of the data cache module is respectively and electrically connected with the input end of the signal data module and the input end of the recognition algorithm module; the data cache module is used for pre-reading data and identifying in real time;
the output end of the signal data module is electrically connected with the input end of the deep learning training algorithm; the signal data module is used for storing all signal data;
the output end of the deep learning training algorithm module is electrically connected with the input end of the model data module; the deep learning training algorithm module is used for training parameters of a deep convolutional neural network;
the output end of the model data module is electrically connected with the input end of the recognition algorithm module; the model data module is used for storing the trained network model parameters;
the first input end of the recognition algorithm module is connected with the output end of the data caching module, and the second input end of the recognition algorithm module is connected with the output end of the model data module; and the recognition algorithm module is used for calling the trained network parameters to perform modulation recognition and output.
In the above technical solution, the data preprocessing module includes a cyclic autocorrelation module and an FFT module, an input end of the cyclic autocorrelation module receives an external radio station modulation signal, an output end sends a result of a signal subjected to cyclic autocorrelation operation to an input end of the FFT module, and the FFT module outputs cyclic spectrum two-dimensional matrix data that is an input signal to an input end of the data cache module.
In the technical scheme, the deep learning training algorithm module comprises a deep convolutional neural network module, a back propagation training algorithm module and an optimal parameter module, wherein the output end of the initial parameter module is connected with the first input end of the deep convolutional neural network, the second input end of the deep convolutional neural network module is connected with the output end of the signal data module, the output end of the deep convolutional neural network module is connected with the back propagation training algorithm module, and the algorithm is used for calculating the optimal parameters of the neural network; the input end of the optimal parameter module is connected with the output end of the back propagation training algorithm module, and the output end of the optimal parameter module is connected with the model data module.
In the technical scheme, the identification algorithm module comprises a real-time data frame module, an optimal parameter unit, a deep convolutional neural network module and a prediction tag module, wherein the input end of the real-time data frame module is connected with the output end of the data cache module, and the input end of the optimal parameter unit is connected with the output end of the model data module; the first input end of the deep convolutional neural network module is connected with the output end of the real-time data frame module, the second input end of the deep convolutional neural network module is connected with the output end of the optimal parameter unit, the output end of the deep convolutional neural network module is connected with the input end of the prediction tag module, and the output end of the prediction tag module is the output end of the recognition algorithm module.
In the technical scheme, the radio station modulation signal is transmitted to the transmitting antenna through the transmitting link, the transmitting antenna transmits the signal to the receiving antenna through the wireless multipath channel, and the signal received by the receiving antenna is transmitted to the data preprocessing module through the receiving link and the ADC module.
The invention provides a use method of a modulation identification device based on signal cycle spectrum and deep learning, which is characterized by comprising the following steps:
step S1, initializing a model, and adjusting the gain of a receiving link according to the isolation between the transmitting antenna and the receiving antenna so that the signal power of a receiving end is in the dynamic range of the ADC module; setting the data frame length of a data caching module;
step S2, initializing the deep convolution neural network parameters of the deep learning training algorithm module and the specific architecture of the deep convolution neural network;
step S3, data preprocessing and data caching are carried out through the circular autocorrelation module and the FFT module, the one-dimensional data format of the signal is converted into a two-dimensional circular spectrum matrix data format, and meanwhile, the two-dimensional circular spectrum matrix data format is stored in the data caching module in a fixed-length frame format;
step S4, storing the data in the data cache module into a signal data module in the big data storage system, then obtaining the network optimal parameters through a deep learning training algorithm module, and storing the network optimal parameters in a model data module in the big data storage system;
step S5, the recognition algorithm module loads the data in the data cache module, loads the data in the model data module at the same time, and calls the recognition algorithm to obtain the modulation mode of the real-time frame data;
and step S6, the recognition algorithm module transmits the real-time data stream to the upper computer monitoring software, and displays the recognition result in a visual interface mode.
In the above technical solution, the step S3 specifically includes the following steps:
substep S31: for the output signal x (n) of the ADC, a cyclic autocorrelation is calculated, resulting in
Wherein τ is the time interval, α is the cycle frequency, n is the discrete sampling time, x (n) is the output signal, T is the cycle period, and x is the conjugate operation;
substep S32: the result of the cyclic autocorrelation is processed by an FFT module to obtain a cyclic spectrum, and the result is
Where τ is the time interval, f is the frequency, α is the cycle frequency,
is a cyclic autocorrelation function, T is a cycle period,
for the spectrum of the output signal x (n),
taking conjugate operation for corresponding cyclic spectrum density function;
substep S33: converting the signal x (n) into a two-dimensional cyclic spectrum matrix data format, wherein the behavior frequency axis of the matrix, the columns of the matrix are cyclic frequency axes, and the value of each element is the corresponding cyclic spectrum
Absolute value of (a).
In the above technical solution, the step S4 specifically includes the following steps:
substep S41: the data of the data preprocessing module is stored in a data cache module in a fixed-length frame format and then stored in a signal data module in a big data storage system;
substep S42: calling signal data, initializing a deep convolution neural network module through initial parameters of an initial parameter module, training a neural network by adopting a back propagation training algorithm module, and obtaining optimal parameters of the network when a convergence threshold value is reached through repeated iteration;
substep S43: and storing the obtained optimal parameters in a model data module.
In the above technical solution, the step S5 specifically includes the following steps:
substep S51: loading data in the data cache module to obtain a real-time data frame;
substep S52: loading data in the model data module to obtain optimal parameters;
substep S53: and (3) importing the real-time data frame and the optimal parameters into a deep convolutional neural network module, and obtaining a prediction label, namely a modulation mode of the real-time signal, through network forward propagation.
In the above technical solution, the step S6 specifically includes the following steps:
substep S61: the identification algorithm module packages the real-time data stream into a fixed format and transmits the real-time data stream to the monitoring module by a function calling method;
substep S62: the monitoring module gives out a real-time data stream identification result in a graphical mode.
The method can complete recognition under the complex channel environment, adopts the deep learning technology to extract the subtle features of the cyclic spectrogram of the signal, and has better performance under the complex channel environment compared with the traditional hypothesis test and statistical mode method. The invention effectively improves the identification performance under the condition of low signal-to-noise ratio, the data preprocessing module converts the discrete signals into a circular spectrum two-dimensional matrix form, and the identification performance under the condition of low signal-to-noise ratio can be improved because the circular spectrum characteristics of the communication signals are not sensitive to noise. The invention effectively improves the identification speed, adopts a data caching mode to store the signal data in the data cache in the form of data frames, and can greatly improve the speed of modulation identification. The invention effectively improves the identification bandwidth, and the receiving link directly converts the radio frequency signal into the baseband signal for processing by adopting the idea of software radio, thereby reducing the data storage capacity and improving the identification bandwidth.
Drawings
Fig. 1 is a schematic block diagram of a modulation identification apparatus based on signal cyclic spectrum and deep learning according to the present invention.
FIG. 2 is a schematic block diagram of data preprocessing.
FIG. 3 is a functional block diagram of a deep learning training algorithm.
FIG. 4 is a functional block diagram of an identification algorithm.
FIG. 5 shows the radio station signal identification effect of the present invention under the real channel
Detailed Description
The invention will be further described in detail with reference to the following drawings and specific examples, which are not intended to limit the invention, but are for clear understanding.
As shown in fig. 1, the present invention provides a modulation recognition apparatus based on signal cycle spectroscopy and deep learning, which comprises a data preprocessing module 00, a data buffer module 01, a signal data module 02, a deep learning training algorithm module 03, a model data module 04, and a recognition algorithm module 05.
The input end of the data preprocessing module 00 is connected with the output end of the ADC, and the output end of the data preprocessing module 00 is connected with the input end of the data caching module 01; the data preprocessing is used for converting a discrete sampling data format of the signal into a two-dimensional cyclic spectrum matrix data format.
The input end of the data cache module 01 is connected with the output end of the data preprocessing module 00, and the output end of the data cache module is connected with the input end of the signal data module 02 and the input end of the recognition algorithm module 05; the data cache module is used for pre-reading data and identifying in real time.
The input end of the signal data module 02 is connected with the output end of the data cache module 01, and the output end of the signal data module is connected with the input end of the deep learning training algorithm module 03; the signal data module is used for storing all signal data.
The input end of the deep learning training algorithm module 03 is connected with the output end of the signal data module 02, and the output end of the deep learning training algorithm module 03 is connected with the input end of the model data module 04; the deep learning training algorithm is used for training parameters of the deep convolutional neural network.
The input end of the model data module 04 is connected with the output end of the deep learning training algorithm module 03, and the output end of the model data module is connected with the input end of the recognition algorithm module 05; and the model data module is used for storing the trained network model parameters.
The first input end of the identification algorithm module 05 is connected with the output end of the data cache module 01, and the second input end of the identification algorithm module 05 is connected with the output end of the model data module 04; and the recognition algorithm calls the trained network parameters to perform modulation recognition.
Further, as shown in fig. 2, the data preprocessing module 00 is composed of a cyclic autocorrelation module 000 and an FFT module 001, an input end of the cyclic autocorrelation module 000 is connected to an output end of the ADC module, and an output end is a result of the cyclic autocorrelation operation performed on the signal. The input end of the FFT module 001 is connected with the output end of the cyclic autocorrelation module 000, the output end of the FFT module is the cyclic spectrum two-dimensional matrix data of the signal, and the output end of the FFT module is connected with the input end of the data cache module 01.
Further, as shown in fig. 3, the deep learning training algorithm module 03 is composed of an initial parameter module 030, a deep convolutional neural network module 031, a back propagation training algorithm module 032, and an optimal parameter module 033, an output end of the initial parameter module 030 is connected to the deep convolutional neural network 031, a second input end of the deep convolutional neural network module 031 is connected to an output end of the signal data module 02, an output end of the deep convolutional neural network module 031 is connected to the back propagation training algorithm module 032, wherein the algorithm is used to calculate optimal parameters of the neural network, an input end of the optimal parameter module 033 is connected to an output end of the back propagation training algorithm module 032, and an output end of the optimal parameter module 033 is connected to the model data 04.
Further, as shown in fig. 4, the identification algorithm module 05 is composed of a real-time data frame module 050, an optimal parameter unit 051, a deep convolutional neural network module 052 and a predictive tag module 053, an input end of the real-time data frame module 050 is connected with an output end of the data buffer module 01, an input end of the optimal parameter unit 051 is connected with an output end of the model data module 04, a first input end of the deep convolutional neural network module 052 is connected with an output end of the real-time data frame module 050, a second input end of the deep convolutional neural network module 052 is connected with an output end of the optimal parameter unit 051, and an output end of the deep convolutional neural network module 052 is connected with an input end of the.
The invention provides a modulation identification method based on signal cycle spectrum and deep learning, which mainly comprises the following specific steps:
step S1: initializing a modulation recognition system model, comprising the sub-steps of:
substep S11: according to the isolation degree between the transmitting antenna and the receiving antenna, the gain of a receiving link is adjusted, so that the signal power of a receiving end is in the dynamic range of the ADC module;
substep S12: setting the data frame length of the data caching module 01;
step S2: initializing a deep learning training algorithm, comprising the following sub-steps:
step S21: setting initial parameters W, b of a neural network in a deep learning training algorithm module 03;
substep S22: and setting a specific architecture of a deep convolutional neural network module 031 in the deep learning training algorithm module 03.
Step S3: data preprocessing, comprising the following substeps:
substep S31: for the output signal x (n) of the ADC module, the result is calculated by the circular autocorrelation module 000
Where τ is the time interval and α is the cycle frequency.
Substep S32: the cyclic autocorrelation result is processed by the FFT module 001 to obtain a cyclic spectrum, and the result is
In the formula
As a function of the cyclic spectral density of the signal x (n).
Substep S33: converting the signal x (n) into a two-dimensional cyclic spectrum matrix data format, wherein the behavior frequency axis of the matrix, the columns of the matrix are cyclic frequency axes, and the value of each element is the corresponding cyclic spectrum
Absolute value of (a).
Step S4: the method comprises the following steps of storing the parameters into a signal data module 02, calling a deep learning training algorithm module 03 to obtain network optimal parameters, and storing the network optimal parameters into a model data module 04, wherein the method comprises the following substeps:
substep S41: the data of the data preprocessing module 00 is stored in a data cache module 01 in a fixed-length frame format and then stored in a signal data module 02 in a big data storage system;
substep S42: calling signal data, initializing a deep convolutional neural network module 031 by using an initial parameter module 030, training a neural network by using a back propagation training algorithm module 032, and obtaining optimal parameters of the network when a convergence threshold is reached through repeated iteration;
substep S43: storing the obtained optimal parameters in a model data module 04;
step S5: calling an identification algorithm to obtain a signal modulation mode, wherein the method comprises the following substeps:
substep S51: loading data in the data cache module 01 to obtain a real-time data frame;
substep S52: loading model data through a model data module to obtain optimal parameters;
substep S53: the real-time data frame and the optimal parameters are led into a deep convolutional neural network module 052, and a prediction tag, namely a modulation mode of the real-time signal, is obtained through network forward propagation.
Step S6: the upper computer monitoring software display comprises the following substeps:
substep S61: packaging the real-time data stream into a fixed format, and transmitting the real-time data stream to a monitoring module by a function calling method;
substep S62: and graphically giving a real-time data stream identification result.
Fig. 5 shows the performance comparison between the conventional statistical pattern modulation recognition method based on feature extraction and the modulation recognition method proposed by the present invention. The result of fig. 5 is obtained by a practical radio station communication system, the radio station works in an ultra-short wave frequency band, and comprises 7 modulation modes such as AM, FM, BPSK, QPSK, 8PSK, MSK, TCM, and different signal-to-noise ratios are set. It can be seen that the statistical pattern recognition method based on feature extraction has a very low recognition rate under the condition of low signal-to-noise ratio because the calculated value of the traditional feature parameter has a large deviation from the theoretical value due to strong noise and multipath channels. When the modulation identification method is adopted, the cyclic spectrum is obtained by utilizing the preprocessing, the influence of noise is reduced, the extraction capability of the abstract characteristics of the signals is improved by utilizing the deep learning technology and the data storage system, and the identification performance is better under the condition of low signal-to-noise ratio.
Details not described in this specification are within the skill of the art that are well known to those skilled in the art.
Claims (10)
1. A modulation identification device based on signal cycle spectrum and deep learning is characterized in that: the system comprises a data preprocessing module, a data caching module, a signal data module, a deep learning training algorithm module, a model data module and an identification algorithm module.
The input end of the data preprocessing module receives the external radio station modulation signal processed by the ADC module, the output end of the data preprocessing module is electrically connected with the input end of the data caching module, and the data preprocessing module is used for converting a received discrete sampling data format into a two-dimensional matrix data format of a gray cyclic spectrogram;
the output end of the data cache module is respectively and electrically connected with the input end of the signal data module and the input end of the recognition algorithm module; the data cache module is used for pre-reading data and identifying in real time;
the output end of the signal data module is electrically connected with the input end of the deep learning training algorithm; the signal data module is used for storing all signal data;
the output end of the deep learning training algorithm module is electrically connected with the input end of the model data module; the deep learning training algorithm module is used for training parameters of a deep convolutional neural network;
the output end of the model data module is electrically connected with the input end of the recognition algorithm module; the model data module is used for storing the trained network model parameters;
the first input end of the recognition algorithm module is connected with the output end of the data caching module, and the second input end of the recognition algorithm module is connected with the output end of the model data module; and the recognition algorithm module is used for calling the trained network parameters to perform modulation recognition and output.
2. The modulation identification device based on the signal cyclic spectrum and the deep learning of claim 1, wherein the data preprocessing module comprises a cyclic autocorrelation module and an FFT module, an input end of the cyclic autocorrelation module receives an external radio modulation signal, an output end of the cyclic autocorrelation module sends a result of the cyclic autocorrelation operation of the signal to an input end of the FFT module, and the FFT module outputs cyclic spectrum two-dimensional matrix data as an input signal to an input end of the data caching module.
3. The modulation recognition device based on the signal cyclic spectrum and the deep learning of claim 2, wherein the deep learning training algorithm module comprises a deep convolutional neural network module, a back propagation training algorithm module and an optimal parameter module, an output end of the initial parameter module is connected with a first input end of the deep convolutional neural network, a second input end of the deep convolutional neural network module is connected with an output end of the signal data module, an output end of the deep convolutional neural network module is connected with the back propagation training algorithm module, and the algorithm is used for calculating optimal parameters of the neural network; the input end of the optimal parameter module is connected with the output end of the back propagation training algorithm module, and the output end of the optimal parameter module is connected with the model data module.
4. The modulation identification device based on the signal cycle spectrum and the deep learning of claim 3, wherein the identification algorithm module comprises a real-time data frame module, an optimal parameter unit, a deep convolutional neural network module and a prediction tag module, wherein the input end of the real-time data frame module is connected with the output end of the data cache module, and the input end of the optimal parameter unit is connected with the output end of the model data module; the first input end of the deep convolutional neural network module is connected with the output end of the real-time data frame module, the second input end of the deep convolutional neural network module is connected with the output end of the optimal parameter unit, the output end of the deep convolutional neural network module is connected with the input end of the prediction tag module, and the output end of the prediction tag module is the output end of the recognition algorithm module.
5. The modulation identification device based on the signal cyclic spectrum and the deep learning of claim 4, wherein the radio station modulation signal is transmitted to a transmitting antenna through a transmitting link, the transmitting antenna transmits the signal to a receiving antenna through a wireless multipath channel, and the signal received by the receiving antenna is transmitted to the data preprocessing module through a receiving link and an ADC module.
6. The use method of the modulation recognition device based on the signal cycle spectrum and the deep learning is characterized by comprising the following steps:
step S1, initializing a model, and adjusting the gain of a receiving link according to the isolation between the transmitting antenna and the receiving antenna so that the signal power of a receiving end is in the dynamic range of the ADC module; setting the data frame length of a data caching module;
step S2, initializing the deep convolution neural network parameters of the deep learning training algorithm module and the specific architecture of the deep convolution neural network;
step S3, data preprocessing and data caching are carried out through the circular autocorrelation module and the FFT module, the one-dimensional data format of the signal is converted into a two-dimensional circular spectrum matrix data format, and meanwhile, the two-dimensional circular spectrum matrix data format is stored in the data caching module in a fixed-length frame format;
step S4, storing the data in the data cache module into a signal data module in the big data storage system, then obtaining the network optimal parameters through a deep learning training algorithm module, and storing the network optimal parameters in a model data module in the big data storage system;
step S5, the recognition algorithm module loads the data in the data cache module, loads the data in the model data module at the same time, and calls the recognition algorithm to obtain the modulation mode of the real-time frame data;
and step S6, the recognition algorithm module transmits the real-time data stream to the upper computer monitoring software, and displays the recognition result in a visual interface mode.
7. The method for using a modulation recognition device based on signal cycle spectroscopy and deep learning of claim 6, wherein the step S3 specifically comprises the following steps:
substep S31: for the output signal x (n) of the ADC, a cyclic autocorrelation is calculated, resulting in
Wherein τ is the time interval, α is the cycle frequency, n is the discrete sampling time, x (n) is the output signal, T is the cycle period, and x is the conjugate operation;
substep S32: the result of the cyclic autocorrelation is processed by an FFT module to obtain a cyclic spectrum, and the result is
Where τ is the time interval, f is the frequency, α is the cycle frequency,
is a cyclic autocorrelation function, T is a cycle period,
for the spectrum of the output signal x (n),
taking conjugate operation for corresponding cyclic spectrum density function;
substep S33: converting the signal x (n) into a two-dimensional cyclic spectrum matrix data format, wherein the behavior frequency axis of the matrix, the columns of the matrix are cyclic frequency axes, and the value of each element is the corresponding cyclic spectrum
Absolute value of (a).
8. The method for using a modulation recognition device based on signal cycle spectroscopy and deep learning of claim 6, wherein the step S4 specifically comprises the following steps:
substep S41: the data of the data preprocessing module is stored in a data cache module in a fixed-length frame format and then stored in a signal data module in a big data storage system;
substep S42: calling signal data, initializing a deep convolution neural network module through initial parameters of an initial parameter module, training a neural network by adopting a back propagation training algorithm module, and obtaining optimal parameters of the network when a convergence threshold value is reached through repeated iteration;
substep S43: and storing the obtained optimal parameters in a model data module.
9. The method for using a modulation recognition device based on signal cycle spectroscopy and deep learning of claim 6, wherein the step S5 specifically comprises the following steps:
substep S51: loading data in the data cache module to obtain a real-time data frame;
substep S52: loading data in the model data module to obtain optimal parameters;
substep S53: and (3) importing the real-time data frame and the optimal parameters into a deep convolutional neural network module, and obtaining a prediction label, namely a modulation mode of the real-time signal, through network forward propagation.
10. The method for using a modulation recognition device based on signal cycle spectroscopy and deep learning of claim 6, wherein the step S6 specifically comprises the following steps:
substep S61: the identification algorithm module packages the real-time data stream into a fixed format and transmits the real-time data stream to the monitoring module by a function calling method;
substep S62: the monitoring module gives out a real-time data stream identification result in a graphical mode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010231208.XA CN111507205B (en) | 2020-03-27 | 2020-03-27 | Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010231208.XA CN111507205B (en) | 2020-03-27 | 2020-03-27 | Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111507205A true CN111507205A (en) | 2020-08-07 |
| CN111507205B CN111507205B (en) | 2022-09-27 |
Family
ID=71871774
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010231208.XA Active CN111507205B (en) | 2020-03-27 | 2020-03-27 | Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111507205B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114650237A (en) * | 2022-02-16 | 2022-06-21 | 中国人民解放军海军工程大学 | Isolation signal detection control device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
-
2020
- 2020-03-27 CN CN202010231208.XA patent/CN111507205B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
Non-Patent Citations (2)
| Title |
|---|
| 李晨等: "基于循环谱和ELM-LRF的调制识别算法", 《舰船电子对抗》 * |
| 陈雪等: "一种优化的卷积神经网络调制识别算法", 《电讯技术》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114650237A (en) * | 2022-02-16 | 2022-06-21 | 中国人民解放军海军工程大学 | Isolation signal detection control device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111507205B (en) | 2022-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107124381B (en) | Automatic identification method for digital communication signal modulation mode | |
| CN109995449B (en) | Millimeter wave signal detection method based on deep learning | |
| CN113630130B (en) | End-to-end digital communication demodulation method | |
| CN105158735B (en) | Null tone Two-Dimensional Spectral Estimation method based on compression sampling array | |
| CN111507205B (en) | Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof | |
| CN110380762A (en) | A kind of extensive cut-in method that calculating is merged with communication | |
| CN101026435A (en) | Low-complexity maximum likelihood detecting method and device for communication system | |
| Jiang et al. | Modulation recognition method of satellite communication based on CLDNN model | |
| Chen et al. | Automatic modulation classification using multi-scale convolutional neural network | |
| CN111507299A (en) | Method for identifying STBC (space time Block coding) signal on frequency domain by using convolutional neural network | |
| Ma et al. | CNN-based automatic modulation recognition of wireless signal | |
| CN111027614A (en) | Noise-enhanced radio frequency fingerprint identification method and device | |
| Zhang et al. | Deep radio fingerprint ResNet for reliable lightweight device identification | |
| US20230155626A1 (en) | Non-coherent long-range (lora) communication system based on multiple-input multiple-output (mimo) technology | |
| CN113259289B (en) | Single-channel aliasing signal modulation mode identification method based on residual error neural network | |
| CN105375967A (en) | Enhanced maximal ratio combining (MRC) detection method and receiver based on statistical method | |
| CN115865145A (en) | Large-scale MIMO channel state information feedback method based on Transformer | |
| Yadav et al. | Application of Machine Learning Framework for Next‐Generation Wireless Networks: Challenges and Case Studies | |
| Wang et al. | Blind Recognition Method for Non-Cooperative Communication Signals Based on Aerial Computing | |
| Zhao et al. | An End-to-End Demodulation System Based on Convolutional Neural Networks | |
| CN113242201A (en) | Wireless signal enhanced demodulation method and system based on generation classification network | |
| Jiyuan et al. | Multi-modulation recognition using convolution gated recurrent unit networks | |
| CN111917674A (en) | Modulation identification method based on deep learning | |
| Wu et al. | Modulation recognition of terahertz signals by deep learning | |
| CN112910519B (en) | Large-scale MIMO base station receiving end signal recovery method based on residual error homogeneous sequencing technology |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |