CN111404852B - Modulation mode identification method based on amplitude and spectral amplitude characteristics - Google Patents
Modulation mode identification method based on amplitude and spectral amplitude characteristics Download PDFInfo
- Publication number
- CN111404852B CN111404852B CN202010138848.6A CN202010138848A CN111404852B CN 111404852 B CN111404852 B CN 111404852B CN 202010138848 A CN202010138848 A CN 202010138848A CN 111404852 B CN111404852 B CN 111404852B
- Authority
- CN
- China
- Prior art keywords
- signal
- sample
- row vector
- amplitude
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
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)
- Image Analysis (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention relates to a modulation mode identification method based on amplitude and spectral amplitude characteristics, which comprises the following specific steps: 1. obtaining a sample: a receiving end receives and respectively processes N sample signals to obtain N row vectors H and N labels; the method for acquiring the row vector H comprises the following steps: the receiving end samples the signal to obtain a discrete sequence, and extracts the amplitude of the discrete sequence to obtain a row vector A; fourier transform is carried out on the discrete sequence to obtain a frequency spectrum sequence, and the amplitude of the frequency spectrum sequence is extracted to obtain a row vector FFT(ii) a Transversely merging a row vector A and a row vector FFTObtaining a row vector H; the label obtaining method comprises the following steps: coding the modulation mode of the signal by applying a one-hot coding rule to obtain a label; forming a sample by the N row vectors H and the N labels; 2. constructing a one-dimensional convolutional neural network; 3. training a one-dimensional convolutional neural network by using a sample; 4. the modulation mode of the signal is identified by applying the trained one-dimensional convolution neural network; the method has the advantages of high identification accuracy, good anti-noise performance and strong robustness.
Description
The technical field is as follows:
the invention relates to a modulation mode identification method, in particular to a modulation mode identification method based on amplitude and spectral amplitude characteristics.
(II) background art:
in a communication system, a transmitting end converts a baseband signal into a form suitable for transmission in a channel through modulation, and a receiving end needs to demodulate the signal. In cooperative communication, the transmitter and the receiver pre-define the modulation mode, modulation parameters, carrier frequency, etc. used for communication, and in some cases, pre-transmit a pilot sequence to help the receiver to complete carrier synchronization, symbol synchronization, channel estimation, etc. In a non-cooperative communication environment, the known communication parameters of the receiving end are few, so the receiving end needs to complete operations such as carrier frequency estimation, symbol rate estimation, signal-to-noise ratio estimation, baseband waveform estimation, modulation mode identification and the like, and then complete synchronous demodulation according to the estimated parameters to acquire useful information.
The modulation mode identification is a technology between signal detection and signal demodulation, and the main task is to complete intelligent receiving and processing of a modulation signal. The modulation mode identification technology is related to many fields of the national civilians, military and national security, and is a research hotspot at home and abroad in recent years. In addition, the recognition of the modulation mode of the authorized user signal can configure reliable reconstruction parameters for the cognitive radio technology, and interference of the cognitive user on normal communication is avoided. With the commercialization of 5G technology, cognitive radio technology as one of 5G key technologies poses new challenges to modulation scheme identification technology.
The modulation mode identification method comprises the following steps: a recognition method based on a likelihood decision theory and a pattern recognition method based on feature extraction. The recognition method based on the likelihood decision theory has the highest recognition performance, but needs a lot of priori knowledge of definite signals, and has large calculation amount. The traditional pattern recognition method based on feature extraction needs manual feature extraction and classification rule design, most methods select a decision tree as a classifier, but the decision tree classifier can only recognize one feature at a time, and the difference information among classes in a single feature set is less, so that the traditional pattern recognition method based on feature extraction has limited recognizable modulation modes; in addition, under the condition of low signal-to-noise ratio, the sample is seriously polluted by noise, and the accuracy of the decision tree classifier for identifying the modulation mode is low.
The deep neural network has strong self-learning capability, and given input and output, the neural network can automatically adjust the weight and the threshold according to a certain algorithm to find the rule of the sample in a self-adaptive manner. The deep neural network has strong fault tolerance, and even if the test sample and the training sample have some differences, such as noise pollution or distortion due to some reason, the deep neural network can still process incomplete data according to the similarity between the samples. In view of this, deep learning is widely used in the fields of image recognition, natural language processing, data mining, and the like. In recent years, researchers try to apply deep learning to identify modulation modes, and simulation results show that a modulation mode identification method based on deep learning is feasible.
The modulation mode recognition system based on deep learning is composed of five parts, namely signal acquisition, preprocessing, feature extraction, neural network training and modulation mode recognition, wherein the feature extraction and neural network training are key parts. The signal characteristics are divided into two categories, namely image characteristics and sequence characteristics, and the modulation mode becomes a research hotspot based on sequence characteristic identification, for example, a patent application with the application publication number of CN107547460A and the name of 'wireless communication modulation signal identification method based on deep learning' discloses a wireless communication modulation signal identification method based on deep learning, and firstly, samples captured signals; normalizing the sampling sequence obtained by sampling, and manufacturing a two-dimensional histogram of the signal according to the normalized sampling sequence; then constructing a deep convolutional neural network; training a deep convolutional neural network by using a training sample; and finally, recognizing the wireless communication modulation signal by using the trained deep convolutional neural network. The method mainly solves the problem that the identification performance of the traditional pattern identification method based on feature extraction excessively depends on the feature extraction by manpower, and has the defects that the two-dimensional histogram is concentrated and has little representation category difference information, the convolutional neural network can only learn partial features of a sample, and the identification performance is poor, especially under the condition of low signal-to-noise ratio.
(III) the invention content:
the technical problem to be solved by the invention is as follows: the modulation mode identification method based on the amplitude and spectral amplitude characteristics is high in identification accuracy, good in anti-noise performance and strong in robustness.
The technical scheme of the invention is as follows:
a modulation mode identification method based on amplitude and spectral amplitude characteristics comprises the following steps of firstly preprocessing a sample signal to obtain a discrete sequence, then respectively extracting the amplitude and the spectral amplitude of the signal to obtain a sample, then training a one-dimensional convolutional neural network, and finally identifying a modulation mode by using the trained one-dimensional convolutional neural network, wherein the method specifically comprises the following steps:
step 1, obtaining a sample:
a receiving end receives N sample signals, respectively processes the N sample signals, and obtains N row vectors H and N labels; the method for obtaining the row vector H of each sample signal comprises the following steps: the receiving end samples the signal to obtain a discrete sequence, and extracts the amplitude of the discrete sequence to obtain a row vector A; fourier transform is carried out on the discrete sequence to obtain a frequency spectrum sequence, and the amplitude of the frequency spectrum sequence is extracted to obtain a row vector FFT(ii) a Transversely merging a row vector A and a row vector FFTObtaining a row vector H; the label of each sample signal is obtained by the following method: coding the modulation mode of the signal by applying a one-hot coding rule to obtain a label;
forming a sample by N row vectors H and N labels of the obtained sample signal, and dividing the sample into a training sample and a test sample;
step 2, constructing a one-dimensional convolution neural network;
step 3, training a one-dimensional convolution neural network:
inputting the training sample into a one-dimensional convolutional neural network, selecting an Adam optimization algorithm to accelerate the training speed, and training the one-dimensional convolutional neural network by adopting an early-stopping method to prevent the one-dimensional convolutional neural network from generating an overfitting phenomenon;
step 4, identifying the modulation mode of the signal by applying the trained one-dimensional convolutional neural network:
the receiving end receives an identified signal, acquires the row vector H of the identified signal according to the acquisition method of the row vector H of the sample signal, and then inputs the row vector H of the identified signal into the trained one-dimensional convolutional neural network to obtain the modulation mode of the identified signal.
In step 1, the sample signal includes BPSK signal, QPSK signal, 2ASK signal, 4FSK signal, 16QAM signal, 64QAM signal, MSK signal, DSB signal, and FM signal, and these sample signals may be generated in advance.
In step 1, the method for obtaining the row vector H of each sample signal specifically includes:
step 1.1, the receiving end receives a signal R, performs M-point sampling on the signal R to obtain a discrete sequence R', R ═ R1,...ri,...rM]I is more than or equal to 1 and less than or equal to M, wherein riIs a complex point, riIs represented by the mathematical expression ofi＝xi+jyiCalculating the amplitude R 'of the discrete sequence R', R [ | R |)1|,...|ri|,...|rM|]Wherein
Extracting the amplitude of M points in the amplitude | R' | to obtain a row vector A with the dimension of 1 multiplied by M;
step 1.2, determining Fourier transform points M': if M is equal to an integer power of 2, then M' ═ M; if M is not equal to the integer power of 2, M' is the shortest integer power of 2;
step 1.3, performing fourier transform on the discrete sequence R' to obtain a frequency spectrum sequence F, [ F ═ F1,...Fi...FM′]Since the spectral sequence F is symmetrical, the front of the spectral sequence F is calculated
The amplitude of the point | F' |,
extracting in amplitude | F' |
The amplitude of a point is given by dimension
Row vector F ofFT；
The fourier transform equation is as follows:
step 1.4, transversely merging the row vector A and the row vector FFTObtaining a row vector H, H ═ A, FFT]。
In step 2, the one-dimensional convolutional neural network comprises two convolutional layers and two full-connection layers; the sizes of convolution kernels in the convolution layers are all set to be 256 multiplied by 3, a max posing layer and a dropout layer are sequentially connected behind each convolution layer to prevent the one-dimensional convolution neural network from generating an overfitting phenomenon in the training process, and the dropout rate is set to be 0.5; the number of neurons of the two fully-connected layers is set to be 256 and 9; and selecting a cross entropy function as a cost function of the one-dimensional convolutional neural network, and determining an initialization method of the one-dimensional convolutional neural network weight.
In step 3, the "early-stop method" means: when the one-dimensional convolutional neural network has good recognition performance on the training sample, but the recognition performance on the test sample begins to decline, the training is stopped.
The invention has the beneficial effects that:
1, analyzing a signal from the angles of a time domain and a frequency domain, extracting amplitude characteristics and spectral amplitude characteristics of the signal, then obtaining a sample, wherein the time domain characteristics and the frequency domain characteristics are mutually complemented, so that the sample can concentrate more information representing category differences, the effect of training a one-dimensional convolutional neural network is better, the accuracy of the one-dimensional convolutional neural network in a modulation mode is finally improved, the identification accuracy of the one-dimensional convolutional neural network under a low signal-to-noise ratio is still high, the anti-noise performance is good, the robustness is strong, and the method is particularly suitable for the identification fields of electronic countermeasure, spectrum monitoring and the like in uncooperative communication.
2. The method can identify the common modulation mode in the communication system, has strong generalization capability, and has the advantages of simple structure of the one-dimensional convolutional neural network, less parameters and less memory occupied by the stored parameters.
(IV) specific embodiment:
the modulation mode identification method based on amplitude and spectral amplitude characteristics comprises the following steps of firstly preprocessing a sample signal to obtain a discrete sequence, then respectively extracting the amplitude and the spectral amplitude of the signal to obtain a sample, then training a one-dimensional convolutional neural network, and finally identifying a modulation mode by using the trained one-dimensional convolutional neural network, wherein the method specifically comprises the following steps:
step 1, obtaining a sample:
first of all, a sample signal is generatedThe signal comprises a BPSK signal, a QPSK signal, a 2ASK signal, a 4FSK signal, a 16QAM signal, a 64QAM signal, a MSK signal, a DSB signal and an FM signal, then a receiving end receives 130000 sample signals, and respectively processes the 130000 sample signals to obtain 130000 row vectors H and 130000 labels; the method for obtaining the row vector H of each sample signal comprises the following steps: the receiving end samples the signal to obtain a discrete sequence, and extracts the amplitude of the discrete sequence to obtain a row vector A; fourier transform is carried out on the discrete sequence to obtain a frequency spectrum sequence, and the amplitude of the frequency spectrum sequence is extracted to obtain a row vector FFT(ii) a Transversely merging a row vector A and a row vector FFTObtaining a row vector H; the label of each sample signal is obtained by the following method: coding the modulation mode of the signal by applying a one-hot coding rule to obtain a label; the modulation scheme of the sample signal and the table of the labels are shown in table 1:
TABLE 1
| Modulation system | Label (R) |
| BPSK | [1,0,0,0,0,0,0,0,0] |
| QPSK | [0,1,0,0,0,0,0,0,0] |
| 2ASK | [0,0,1,0,0,0,0,0,0] |
| 4FSK | [0,0,0,1,0,0,0,0,0] |
| 16QAM | [0,0,0,0,1,0,0,0,0] |
| 64QAM | [0,0,0,0,0,1,0,0,0] |
| MSK | [0,0,0,0,0,0,1,0,0] |
| DSB | [0,0,0,0,0,0,0,1,0] |
| FM | [0,0,0,0,0,0,0,0,1] |
130000 row vectors H and 130000 labels of the obtained sample signals form a sample, and the sample is divided into a training sample and a test sample;
step 2, constructing a one-dimensional convolution neural network;
step 3, training a one-dimensional convolution neural network:
inputting the training sample into a one-dimensional convolutional neural network, selecting an Adam optimization algorithm to accelerate the training speed, and training the one-dimensional convolutional neural network by adopting an early-stopping method to prevent the one-dimensional convolutional neural network from generating an overfitting phenomenon;
step 4, identifying the modulation mode of the signal by applying the trained one-dimensional convolutional neural network:
the receiving end receives an identified signal, acquires the row vector H of the identified signal according to the acquisition method of the row vector H of the sample signal, and then inputs the row vector H of the identified signal into the trained one-dimensional convolutional neural network to obtain the modulation mode of the identified signal.
In step 1, the method for obtaining the row vector H of each sample signal specifically includes:
step 1.1, receiving by receiving endOne signal R is sampled at 500 points to obtain a discrete sequence R', R ═ R1,...ri,...r500]I is more than or equal to 1 and less than or equal to 500, wherein riIs a complex point, riIs represented by the mathematical expression ofi＝xi+jyiCalculating the amplitude R 'of the discrete sequence R', R [ | R |)1|,...|ri|,...|r500|]Wherein
Extracting the amplitude of 500 points in the amplitude | R' | to obtain a row vector A with the dimensionality of 1 multiplied by 500;
step 1.2, taking the Fourier transform point number as the power of 2 with the shortest distance of 500, wherein the Fourier transform point number is 512;
step 1.3, performing fourier transform on the discrete sequence R' to obtain a frequency spectrum sequence F, [ F ═ F1,...Fi...F512]Since the spectrum sequence F is symmetrical, the amplitude | F '|, | F' | [ | F ] of the first 256 points of the spectrum sequence F is calculated1|,...|Fi|,...|F256|]I is more than or equal to 1 and less than or equal to 256, and the amplitudes of 256 points in the amplitude | F' | are extracted to obtain a row vector F with the dimension of 1 multiplied by 256FT；
The fourier transform equation is as follows:
step 1.4, transversely merging the row vector A and the row vector FFTTo obtain a row vector H with dimension 1 × 756, [ a, F ═ HFT]。
In step 2, the one-dimensional convolutional neural network comprises two convolutional layers and two full-connection layers; the sizes of convolution kernels in the convolution layers are all set to be 256 multiplied by 3, a max posing layer and a dropout layer are sequentially connected behind each convolution layer to prevent the one-dimensional convolution neural network from generating an overfitting phenomenon in the training process, and the dropout rate is set to be 0.5; the number of neurons of the two fully-connected layers is set to be 256 and 9; and selecting a cross entropy function as a cost function of the one-dimensional convolutional neural network, and determining an initialization method of the one-dimensional convolutional neural network weight.
In step 3, the "early-stop method" means: when the one-dimensional convolutional neural network has good recognition performance on the training sample, but the recognition performance on the test sample begins to decline, the training is stopped.
The technical effect of the modulation mode identification method based on the amplitude and spectral amplitude characteristics is described below with the combination of simulation experiments:
1. simulation conditions and contents:
the simulation experiment is carried out on a hardware platform of an Intel (R) core (TM) i5-8400 CPU @2.80GHZ2.81GHZ and a 64-bit windows operating system, and simulation software adopts Matlab2018b and a deep learning framework keras with a base TensorFlow as a rear end. The comparison of the identification accuracy of the modulation mode identification method based on amplitude and spectral amplitude characteristics and the identification accuracy of the modulation mode identification based on spectral characteristics is shown in table 2:
TABLE 2
2. And (3) simulation result analysis:
as shown in table 2, the recognition accuracy of the modulation scheme recognition method based on amplitude and spectral amplitude characteristics is significantly higher than that of the modulation scheme recognition based on spectral characteristics. When the signal to noise ratio is equal to-10 db, the identification accuracy of the modulation mode identification method based on the amplitude and spectral amplitude characteristics reaches 82.2%, the identification accuracy under the high signal to noise ratio is close to 100%, and the identification performance and robustness of the modulation mode identification method based on the amplitude and spectral amplitude characteristics are good.
Claims (4)
1. A modulation mode identification method based on amplitude and spectral amplitude features is characterized in that: comprises the following steps:
step 1, obtaining a sample:
a receiving end receives N sample signals, respectively processes the N sample signals, and obtains N row vectors H and N labels; row vector of each sample signalThe acquisition method of H comprises the following steps: the receiving end samples the signal to obtain a discrete sequence, and extracts the amplitude of the discrete sequence to obtain a row vector A; fourier transform is carried out on the discrete sequence to obtain a frequency spectrum sequence, and the amplitude of the frequency spectrum sequence is extracted to obtain a row vector FFT(ii) a Transversely merging a row vector A and a row vector FFTObtaining a row vector H; the label of each sample signal is obtained by the following method: coding the modulation mode of the signal by applying a one-hot coding rule to obtain a label;
forming a sample by N row vectors H and N labels of the obtained sample signal, and dividing the sample into a training sample and a test sample;
step 2, constructing a one-dimensional convolution neural network;
step 3, training a one-dimensional convolution neural network:
inputting the training sample into a one-dimensional convolution neural network, selecting an Adam optimization algorithm to accelerate the training speed, and training the one-dimensional convolution neural network by adopting an early-stopping method; "early arrest" means: when the one-dimensional convolutional neural network has good identification performance on a training sample, and the identification performance on a test sample begins to decline, stopping training;
step 4, identifying the modulation mode of the signal by applying the trained one-dimensional convolutional neural network:
the receiving end receives an identified signal, acquires the row vector H of the identified signal according to the acquisition method of the row vector H of the sample signal, and then inputs the row vector H of the identified signal into the trained one-dimensional convolutional neural network to obtain the modulation mode of the identified signal.
2. The method of claim 1, wherein the method further comprises: in step 1, the sample signal includes a BPSK signal, a QPSK signal, a 2ASK signal, a 4FSK signal, a 16QAM signal, a 64QAM signal, an MSK signal, a DSB signal, and an FM signal.
3. The method of claim 1, wherein the method further comprises: in step 1, the method for obtaining the row vector H of each sample signal specifically includes:
step 1.1, the receiving end receives a signal R, performs M-point sampling on the signal R to obtain a discrete sequence R', R ═ R1,...ri,...rM]I is more than or equal to 1 and less than or equal to M, wherein riIs a complex point, riIs represented by the mathematical expression ofi＝xi+jyiCalculating the amplitude R 'of the discrete sequence R', R [ | R |)1|,...|ri|,...|rM|]Wherein
Extracting the amplitude of M points in the amplitude | R' | to obtain a row vector A with the dimension of 1 multiplied by M;
step 1.2, determining Fourier transform points M': if M is equal to an integer power of 2, then M' ═ M; if M is not equal to the integer power of 2, M' is the shortest integer power of 2;
step 1.3, performing fourier transform on the discrete sequence R' to obtain a frequency spectrum sequence F, [ F ═ F1,...Fi...FM′]Calculating the front of the spectral sequence F
The amplitude of the point | F' |,
extracting in amplitude | F' |
The amplitude of a point is given by dimension
Row vector F ofFT；
Step 1.4, transversely merging the row vector A and the row vector FFTObtaining a row vector H, H ═ A, FFT]。
4. The method of claim 1, wherein the method further comprises: in the step 2, the one-dimensional convolutional neural network comprises two convolutional layers and two full-connection layers; the sizes of convolution kernels in the convolution layers are all set to be 256 multiplied by 3, a max posing layer and a dropout layer are sequentially connected behind each convolution layer, and the dropout rate is set to be 0.5; the number of neurons of the two fully-connected layers is set to be 256 and 9; and selecting a cross entropy function as a cost function of the one-dimensional convolution neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010138848.6A CN111404852B (en) | 2020-03-03 | 2020-03-03 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010138848.6A CN111404852B (en) | 2020-03-03 | 2020-03-03 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111404852A CN111404852A (en) | 2020-07-10 |
| CN111404852B true CN111404852B (en) | 2021-04-13 |
Family
ID=71428657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010138848.6A Active CN111404852B (en) | 2020-03-03 | 2020-03-03 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111404852B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113312996B (en) * | 2021-05-19 | 2023-04-18 | 哈尔滨工程大学 | Detection and identification method for aliasing short-wave communication signals |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104618073A (en) * | 2015-03-03 | 2015-05-13 | 北京邮电大学 | Signal modulation mode identification method |
| CN106059969A (en) * | 2016-06-16 | 2016-10-26 | 中国科学院信息工程研究所 | Modulation identification method and device based on envelope square spectrum analysis |
| US9729362B1 (en) * | 2013-03-20 | 2017-08-08 | Georgia Tech Research Corporation | Systems and methods for autonomous signal modulation format identification |
| CN107979554A (en) * | 2017-11-17 | 2018-05-01 | 西安电子科技大学 | Radio signal Modulation Identification method based on multiple dimensioned convolutional neural networks |
| CN108764013A (en) * | 2018-03-28 | 2018-11-06 | 中国科学院软件研究所 | A kind of automatic Communication Signals Recognition based on end-to-end 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 |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107317778B (en) * | 2017-07-28 | 2019-11-26 | 西安电子科技大学 | BPSK modulating signal phase transition detection method based on 1D-CNN |
| CN108234370B (en) * | 2017-12-22 | 2020-12-15 | 西安电子科技大学 | Communication signal modulation mode identification method based on convolutional neural network |
-
2020
- 2020-03-03 CN CN202010138848.6A patent/CN111404852B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9729362B1 (en) * | 2013-03-20 | 2017-08-08 | Georgia Tech Research Corporation | Systems and methods for autonomous signal modulation format identification |
| CN104618073A (en) * | 2015-03-03 | 2015-05-13 | 北京邮电大学 | Signal modulation mode identification method |
| CN106059969A (en) * | 2016-06-16 | 2016-10-26 | 中国科学院信息工程研究所 | Modulation identification method and device based on envelope square spectrum analysis |
| CN107979554A (en) * | 2017-11-17 | 2018-05-01 | 西安电子科技大学 | Radio signal Modulation Identification method based on multiple dimensioned convolutional neural networks |
| CN108764013A (en) * | 2018-03-28 | 2018-11-06 | 中国科学院软件研究所 | A kind of automatic Communication Signals Recognition based on end-to-end 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 |
Non-Patent Citations (1)
| Title |
|---|
| 基于机器学习的调制方式识别;范涛涛;《中国优秀硕士学位论文全文数据库》;20190630;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111404852A (en) | 2020-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108234370B (en) | Communication signal modulation mode identification method based on convolutional neural network | |
| CN110163282B (en) | Modulation mode identification method based on deep learning | |
| Zhang et al. | Automatic modulation classification using CNN-LSTM based dual-stream structure | |
| WO2018176889A1 (en) | Method for automatically identifying modulation mode for digital communication signal | |
| CN111832417B (en) | Signal modulation pattern recognition method based on CNN-LSTM model and transfer learning | |
| CN113014524B (en) | Digital signal modulation identification method based on deep learning | |
| CN109887047B (en) | Signal-image translation method based on generation type countermeasure network | |
| CN114422311B (en) | Signal modulation recognition method and system combining deep neural network and expert priori features | |
| Jeong et al. | Spectrogram-based automatic modulation recognition using convolutional neural network | |
| CN109120563A (en) | A kind of Modulation Identification method based on Artificial neural network ensemble | |
| Sang et al. | Application of novel architectures for modulation recognition | |
| CN111404852B (en) | Modulation mode identification method based on amplitude and spectral amplitude characteristics | |
| CN112613443A (en) | Robustness communication radiation source intelligent identification method based on deep learning | |
| Bouchenak et al. | A semi-supervised modulation identification in MIMO systems: A deep learning strategy | |
| Wang et al. | A spatiotemporal multi-stream learning framework based on attention mechanism for automatic modulation recognition | |
| Yin et al. | Co-channel multi-signal modulation classification based on convolution neural network | |
| CN114255487A (en) | Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction | |
| He et al. | Deep learning-based automatic modulation recognition algorithm in non-cooperative communication systems | |
| CN115913850B (en) | Open set modulation identification method based on residual error network | |
| Zhang et al. | A spectrum sensing algorithm for OFDM signal based on deep learning and covariance matrix graph | |
| CN115935252A (en) | Communication signal modulation pattern recognition method and device based on self-supervision contrast learning | |
| CN118115853A (en) | Broadband signal detection method based on lightweight full convolution neural network | |
| CN116016071A (en) | Modulation signal identification method based on double-flow fusion CNN-BiLSTM network | |
| CN115276857A (en) | Total-blind spectrum detection method based on combination of Cholesky decomposition and convolutional neural network | |
| CN111917674A (en) | Modulation identification method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |