CN108234370B - Communication signal modulation mode identification method based on convolutional neural network - Google Patents
Communication signal modulation mode identification method based on convolutional neural network Download PDFInfo
- Publication number
- CN108234370B CN108234370B CN201711403640.7A CN201711403640A CN108234370B CN 108234370 B CN108234370 B CN 108234370B CN 201711403640 A CN201711403640 A CN 201711403640A CN 108234370 B CN108234370 B CN 108234370B
- Authority
- CN
- China
- Prior art keywords
- neural network
- convolutional neural
- signal
- layer
- modulation mode
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 140
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000004891 communication Methods 0.000 title claims abstract description 47
- 239000011159 matrix material Substances 0.000 claims abstract description 27
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 238000006243 chemical reaction Methods 0.000 claims abstract description 13
- 239000000654 additive Substances 0.000 claims abstract description 11
- 230000000996 additive effect Effects 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims description 50
- 238000012360 testing method Methods 0.000 claims description 22
- 238000007781 pre-processing Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 12
- 238000010586 diagram Methods 0.000 claims description 9
- 230000002779 inactivation Effects 0.000 claims description 8
- 230000005540 biological transmission Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 30
- 210000002569 neuron Anatomy 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 9
- 238000004422 calculation algorithm Methods 0.000 description 9
- 239000000284 extract Substances 0.000 description 9
- 238000004088 simulation Methods 0.000 description 8
- 230000004913 activation Effects 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 7
- 238000005070 sampling Methods 0.000 description 7
- 238000001228 spectrum Methods 0.000 description 7
- 230000010363 phase shift Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 230000018109 developmental process Effects 0.000 description 5
- 241000209094 Oryza Species 0.000 description 4
- 235000007164 Oryza sativa Nutrition 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000006073 displacement reaction Methods 0.000 description 4
- 235000009566 rice Nutrition 0.000 description 4
- 238000005562 fading Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 230000001149 cognitive effect Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 230000009849 deactivation Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012567 pattern recognition method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 230000001131 transforming effect Effects 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses a modulation mode identification system and method based on a convolutional neural network, and solves the problems that the prior art has complex feature extraction steps and low identification rate under low signal-to-noise ratio. The simple feature in the identification system is constructed by taking the same-direction component and the orthogonal component of a baseband signal as the simple features of the signal and sending the simple features to a convolutional neural network module for identification; the identification method comprises the following implementation steps: modulating a sending signal and performing pulse forming; the sending signal is sent through an additive white Gaussian noise channel after up-conversion; the receiving end carries on the preconditioning first, get the syntropy component r (t) of the analytic signal; the simple structure is characterized by analyzing the homodromous component r (t) and the orthogonal component of the signal
Constructing a two-dimensional matrix; learning and classifying through the characteristics of a convolutional neural network; and sending the modulation mode to a demodulation end to obtain a demodulated signal. The invention has low complexity of feature design, high classification accuracy and high recognition performance, avoids explicit feature extraction and can be applied to communication systems with higher requirements on recognition performance.
Description
Technical Field
The invention belongs to the technical field of communication, mainly relates to the technical field of communication signal identification and deep learning, and particularly relates to a convolutional neural network-based communication signal modulation mode identification method which is used for the fields of cognitive radio, interference identification, frequency spectrum management, electronic countermeasure and the like.
Background
The identification of the communication signal modulation mode is an intermediate step of signal detection and demodulation, has important research value in military and civil fields, and is widely applied to the fields of cognitive radio, interference identification, frequency spectrum management, electronic countermeasure and the like.
At present, in documents about modulation identification published at home and abroad, modulation identification methods can be divided into two categories, namely judgment mode identification and statistical theory identification. The method for judging pattern recognition is based on probability theory and Bayes theory in hypothesis test, and obtains a specific modulation mode of the signal by comparing the likelihood ratio of the likelihood function of the received signal with a threshold value, but the mode can obtain the optimal recognition performance, but the likelihood function is complex to deduce, the calculated amount is large when the unknown variable is large, and some priori knowledge of the signal is needed to be known; the statistical pattern recognition method firstly extracts the characteristic parameters of the signals and then judges the modulation mode of the signals according to the characteristic parameters, and the method has the advantages of simple theory, simple pretreatment and easy realization, and has the defects that a recognition frame has no complete theoretical basis, usually needs additional training samples and is difficult to realize in engineering.
A method for automatically identifying a neural network classifier based on a hierarchical structure is provided in the journal article published by Zhouyanjuan, Lizecheng and Wanguojin in 2008, in feature extraction and application thereof in digital modulation mode identification, wherein in Signal processing, 24(2), 201 and 203 are provided. The method provides four characteristic parameters, and then adopts a neural network classifier with a hierarchical structure to carry out automatic identification, and has the defect of large acquisition and calculation amount of the characteristic parameters. A journal article, "a digital signal modulation mode identification method", published in 2011 xu, leiban and libo has proposed a modulation mode identification scheme based on signal power spectrum and spectrum correlation as features, communication technology, 11(44):22-24,102. The method extracts various characteristics of a signal power spectrum as main characteristic parameters, identifies signals according to the number, peak value and distribution of spectral peaks in signal circulating density, has limited types of signals which can be identified, and cannot be expanded to some common signals. A classification algorithm based on a constellation diagram is provided in a journal article of 'digital modulation mode identification based on a constellation diagram' of Wangjian Xin and Song Hui Sen Tao in 2004, the journal of the Communications, 25(6), 166-. The algorithm is to firstly perform subtraction clustering on signals, and then extract a clustering center to match with an ideal constellation diagram model. The method has complex algorithm and is difficult to realize real-time modulation mode identification.
Modulation scheme identification algorithms, although successful, do not cover the deficiencies in their development: the existing modulation mode recognition algorithm has complex feature extraction steps, almost all features required by system recognition are manually designed, the recognition rate is low under low signal-to-noise ratio, and the like.
Disclosure of Invention
The invention aims to provide a system and a method for identifying a modulation mode based on a convolutional neural network without setting complex characteristics in advance aiming at the defects in the background.
The invention relates to a modulation mode identification system based on a convolutional neural network, which is characterized in that a system sending end, a channel and a receiving end are sequentially connected according to the transmission direction of signals, the signals are demodulated after sequentially passing through the units, a preprocessing operation, a simple feature construction module and a convolutional neural network module are sequentially connected in the receiving end, and the modulation mode of the modulation signals identified by the convolutional neural network is used for signal demodulation.
The invention also relates to a modulation mode identification method based on a convolutional neural network, which is characterized by comprising the following steps:
(1) the sending end modulates a sending signal and carries out pulse forming: at a transmitting end, mapping and modulating a transmitting signal according to different constellation diagrams, performing pulse forming by adopting rectangular pulses, and selecting a code element length M;
(2) the signal passes through the channel: the pulse signal is sent out after up-conversion and passes through an additive white Gaussian noise channel;
(3) and (3) preprocessing operation: the receiving end firstly carries out carrier frequency estimation on the signal, and then carries out down-conversion operation on the received signal to obtain a baseband signal r (t), wherein the signal is the homodromous component of the analytic signal;
(4) simple structure characteristics: the simple-structure characteristic module performs Hilbert transform on the homodromous component r (t) of the analysis signal to obtain the orthogonal component of the analysis signal
Combining the homodromous component r (t) and the quadrature component of the complex signal
Constructed as a two-dimensional matrix R;
(5) the convolutional neural network classifies the signals: the convolutional neural network module is used for learning, training and classifying the two-dimensional matrix R;
(5.1) convolutional neural network learning training stage:
randomly initializing the weight and the bias of each layer in the convolutional neural network; inputting a training sample consisting of a two-dimensional matrix and a label thereof, and calculating an output value of each layer by using a forward conduction formula; calculating the residual error of the last layer according to the label value and the output value of the last layer; and updating the weight of the network according to a gradient descent method to finish the training of the convolutional neural network.
(5.2) a convolutional neural network classification stage:
keeping the parameters in the convolutional neural network unchanged, and calculating the output values of all layers of the test sample before passing through the output layer of the convolutional neural network layer by layer; calculating a class vector value output by an output layer of the convolutional neural network; determining a modulation mode of the test sample according to the position of the maximum value in the convolutional neural network output category vector values;
(6) and sending the modulation mode of the signal identified by the convolutional neural network to a signal demodulation module, and demodulating the signal to obtain a demodulated signal.
The invention has simple characteristic structure, and the network can capture the characteristics of the original data by means of automatic learning and improve the identification performance of the system.
Compared with the prior art, the invention has the following advantages:
firstly, the capability that the convolutional neural network can identify the displacement, the scaling and other forms of distortion invariance in deep learning is applied to the field of modulation mode identification, so that the explicit characteristic extraction is avoided, and the requirement on the characteristic effectiveness in the traditional modulation mode identification system is weakened;
secondly, the required input is a two-dimensional matrix formed by an in-phase component and a quadrature component of an analytic signal, and the required input is easily obtained in a communication system, so that the complexity of the structural characteristics is obviously reduced;
thirdly, compared with the traditional modulation mode identification method, the method overcomes the defect that the modulation signal information can not be fully utilized in the prior art, can obtain higher identification accuracy under the same condition, and improves the system performance.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a block diagram of a convolutional neural network used in the present invention;
fig. 3 is a comparison graph of the modulation mode identification performance simulation results proposed by the present invention and a.liu.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Example 1
Modulation scheme recognition is an important application of pattern recognition in the field of communications, and although many efforts have been made using conventional methods of feature value acquisition in combination with trainable classifiers, these advances have not masked the deficiencies in development: the existing modulation mode identification algorithm has complex characteristic parameter extraction steps, almost all the characteristics required by system identification are manually designed, the identification rate is low under low signal-to-noise ratio, and the like. The invention develops and studies the problems, and provides a communication signal modulation mode identification system based on a convolutional neural network, a system sending end, a channel and a receiving end are sequentially connected according to the transmission direction of signals, the signals are demodulated after sequentially passing through the units, a preprocessing operation, a simple feature construction module and a convolutional neural network module are sequentially connected in the receiving end, the modulation mode of the modulation signals identified by the convolutional neural network is used for signal demodulation, the simple feature construction module takes the homodromous component and the orthogonal component of the preprocessed analytic signals as the simple features of the signals, and the simple features of the signals are sent to the convolutional neural network module to extract and classify the abstract features to obtain the modulation mode of the communication signals, wherein the simple features of the signals are extracted and classified.
According to the method, an input end carries out constellation mapping on binary bits to obtain a modulation signal, the modulation mode of the signal selects a common modulation mode in satellite communication, such as binary phase shift keying, quaternary phase shift keying, octal phase shift keying, hexadecadrature amplitude modulation and the like, then the modulation signal reaches a receiving end of a system after passing through a transmission channel, the receiving end firstly carries out preprocessing operation on the signal to obtain an in-phase component of an analytic signal, the in-phase component is sent to a simple feature construction module to obtain simple features, and the simple features are sent to a convolutional neural network module to carry out abstract feature extraction and specific modulation mode identification.
In the field of satellite signal communication and telemetry, modulation mode identification is of great significance, is an important step between signal detection and signal demodulation, and is also a necessary precondition for a demodulator to correctly demodulate signals and acquire satellite link information. The invention is provided under the background of rapid development of deep learning, pixel points of an image are inspired for mode recognition input from handwriting digital recognition, a modulation mode recognition system of the invention innovatively takes a time domain signal as input, then acquires the characteristics of data by means of the extraction capability of a convolutional neural network on abstract characteristics of the data, and classifies modulation signals according to the acquired characteristics.
Example 2
The general construction of the convolutional neural network-based modulation mode identification system is the same as that of embodiment 1, and the convolutional neural network of the present invention is a multilayer deep convolutional neural network, and includes an input layer, four convolutional layers, four downsampling layers, two full-link layers, and an output layer. Referring to fig. 2, in this example, the input layer is a 2 × 512 matrix, and after passing through the first layer of convolutional layer, that is, after 32 convolutional cores of 1 × 3 perform convolution operations on input data, the input data is changed into 32 characteristic maps of 2 × 512; in order to prevent overfitting of data in the training process, the 32 2 × 512 feature maps need to be subjected to 1 × 2 downsampling operation, and the 32 2 × 256 feature maps are obtained after the downsampling operation. In order to fully extract the features in the input data, the number of the convolutional layers and the down-sampling layers is set to 4 layers in the example, the output obtained by passing the input through the four convolutional layers and the four down-sampling layers is elongated into a 1024-dimensional vector, the 1024-dimensional vector is fully connected to a hidden layer with the length of 128, and the 128-dimensional vector is fully connected to the output layer through the softmax function. The convolution aims at extracting different characteristics of input data by different convolution kernels, and the operation complexity is reduced through weight sharing. The down-sampling process is equivalent to filtering operation, so that the dimensionality of the data is reduced, secondary feature extraction is carried out on the data, and the generalization capability of the convolutional neural network and the robustness of a modulation mode identification system are improved. The number of convolution kernels and the length of a full connection layer in the convolutional neural network can be changed, but the larger the number of convolution kernels is, the higher the feature extraction capability of data is, and correspondingly, the larger the number of neurons needing to be trained in the network is. The recognition performance of the system can be improved by deepening the depth of the convolutional neural 92 network.
Example 3
The present invention is also a modulation scheme identifying method based on a convolutional neural network, which is used in a modulation scheme identifying system based on a convolutional neural network, and the overall configuration of the modulation scheme identifying system based on a convolutional neural network is the same as that of embodiments 1-2,
referring to fig. 1, the identification method includes the following steps:
(1) the sending end modulates a sending signal and carries out pulse forming: at a transmitting end of a modulation mode identification system based on a convolutional neural network, mapping and modulating a transmitted communication signal according to different constellation diagrams, and performing pulse forming operation by adopting rectangular pulses to obtain rectangular pulse forming signals. When the signal is modulated by different modes, the length of code element is selected, in this example, the length of code element M is 32. The length of input data in the system is mainly determined by the length M of symbol data in the truncated signal,The shorter the input data is, the fewer the number of neurons in the convolutional neural network needs to be trained, which is determined by the symbol rate of the transmitting end and the sampling rate of the receiving end. E.g. taking M32 and symbol rate RbCarrier rate of fc＝2RbSampling rate fs＝32RbThen the input data to the convolutional neural network is 2 × 1024.
(2) The signal passes through the channel: the formed pulse signal is carried on a high-frequency carrier, namely, the pulse signal is sent out after up-conversion and passes through an additive white gaussian noise Channel, namely, a modulation signal is carried on the high-frequency carrier and is transmitted to a system receiving end through the additive white gaussian noise Channel (AWGN Channel). In the real environment, a Rayleigh Fading Channel (Rayleigh Fading Channel) or a rice Channel (Rician Channel) can be selected, and the performance of the identification method in the real Channel environment can be obtained by observing the performance of the Rayleigh Fading Channel. When a channel is transmitted in a real environment, the energy of a signal is attenuated when the signal is shielded by surrounding objects, a plurality of paths often exist between a transmitting end and a receiving end, and the signal time delays of signals in different paths are different, so that the signals received by the receiving end are superposed by the paths.
(3) And (3) preprocessing operation: the receiving end of the modulation mode identification system based on the convolutional neural network firstly utilizes the costas loop to carry out carrier frequency estimation on a communication signal, and then utilizes the estimated carrier frequency to carry out down-conversion operation on a receiving signal transmitted through a channel to obtain a down-converted baseband signal r (T), wherein the signal is the homodromous component of an analytic signal, and the baseband signal in a time interval T is represented as:
wherein,
is a complex baseband signal, N (t) is complex additive white Gaussian noise, and the bilateral power spectral density of the noise is N0/2，
Is the carrier phase tracking error and N is the number of symbols sampled. s [ k ]]Representing the symbol of the kth recovered constellation, the amplitude of this symbol being skM is the number of phases determined by the particular modulation scheme, g (T) is a square pulse, TsIs the symbol interval.
The invention firstly carries out preprocessing operation, and down-converts the received high-frequency signal into a baseband signal, namely, a homodromous component r (t) of an analytic signal.
(4) Simple structure characteristics: in a modulation mode identification system based on a convolutional neural network, a simple feature module is constructed to carry out Hilbert transform on the homodromous component r (t) of the analysis signal obtained in the step (3) to obtain an orthogonal component of the analysis signal
Then the homodromous component r (t) and the orthogonal component of the signal are combined
Constructed as a two-dimensional matrix R, which is simply a characteristic of the constructed input communication signal.
The invention sends the simple characteristics R of the constructed communication signal into the convolutional neural network for learning, training and classifying the neural network.
(5) The convolutional neural network classifies the signals: and (4) the convolutional neural network module performs learning training and classification on the two-dimensional matrix R obtained in the step (4).
The convolutional neural network has the characteristics that:
the invention applies the convolutional neural network to the signal modulation mode identification and classification, and has the following two reasons: the convolutional neural network is mainly used for identifying two-dimensional graphs with displacement, scaling and other form distortion invariance, so that explicit feature extraction is avoided, and learning is performed from training data implicitly; and secondly, the weights of the neurons on the same characteristic diagram in the structure of the convolutional neural network are the same, and the complexity of the network is reduced by means of weight sharing. The convolutional neural network for identifying and classifying the communication signal modulation modes comprises two stages: a learning training phase and a classification phase.
(5.1) convolutional neural network learning training stage: randomly initializing the weight and the bias of each layer in the convolutional neural network; inputting a training sample consisting of a two-dimensional matrix and a label thereof, and calculating an output value of each layer of the convolutional neural network by using a forward conduction formula; calculating the residual error of the last layer according to the label value and the output value of the output layer; and updating the weight of the neural network according to a gradient descent method to finish the training of the convolutional neural network.
In the training process, in order to prevent overfitting of the neural network model caused by too little training data, the convolutional layer random inactivation probability needs to be set in the forward conduction process of the convolutional network, and the method can effectively improve the generalization capability of the neural network and improve the identification performance of the network.
The training samples are obtained by randomly generated modulation signals, each modulation signal corresponds to a unique label, the labels are manually added when the signals are generated and are in a one-dimensional vector form, and only one element of the label vector is one, and the rest elements are zero. And the number of training samples for each modulation signal is the same. When the convolutional neural network is trained, all training samples are randomly disturbed so as to improve the generalization capability of the convolutional neural network.
(5.2) a convolutional neural network classification stage: keeping the parameters in the convolutional neural network unchanged, and calculating the output values of the test samples passing through all hidden layers of the convolutional neural network layer by layer; calculating a class vector value output by an output layer of the convolutional neural network; and determining the modulation mode of the test sample, namely the modulation mode of the obtained sending signal according to the position of the maximum value in the output class vector values of the convolutional neural network, wherein the position of the maximum value corresponds to the modulation mode identified by the invention.
The test samples and labels were obtained in the same manner as the training samples. The number of training samples and the number of test samples can be the same or different, and when the training samples and the test samples are different, the number of test samples is taken: the number of training samples is 2: 1.
(6) And sending the modulation mode of the communication signal identified by the convolutional neural network to a signal demodulation module, demodulating the received signal to obtain a demodulated communication signal, and completing the identification of the whole modulation mode.
The invention expends little energy on the manual design characteristics, obtains a two-dimensional matrix which fully represents the modulation mode characteristics of the baseband communication signals after Hilbert conversion, and the input is easy to obtain in a communication system, while the traditional manual design characteristic method can only extract partial modulation information from high-dimensional data input and can not fully utilize the information carried by the communication signals; the invention utilizes the advantage that the multilayer deep convolution neural network can extract the features of the input data as much as possible to implicitly extract the features in the two-dimensional data, thereby simplifying the steps of setting complex features, obviously reducing the complexity of constructing the features, and being a development direction of modulation mode identification.
Example 4
The system and the method for identifying the modulation mode based on the convolutional neural network have the characteristics of simple structure as those described in the embodiments 1 to 3 and the step 4, and specifically include the following steps:
(4.1) performing Hilbert transform on the homodromous component r (t) obtained in the step (3) to obtain an orthogonal component of the signal
The formula is as follows:
where r (τ) is the homodromous component of the analytic signal,
is the quadrature component of the resolved signal being found, τ is the introduced variable, t represents the signal duration;
let the Hilbert transform of the signal y (t)
Then engineering, Hill of the signalThe bert transform is usually represented in the form of a convolution as:
rather than transforming a signal from the time domain to another domain, the hilbert transform transforms a signal from the time domain to the time domain. The frequency domain corresponding to the hilbert transform corresponds to:
H(f)＝-jsgn(f) (4)
wherein j is2＝-1，
Fourier transform of the hilbert transform of signal y (t):
wherein, y (f) is a frequency domain expression obtained by performing fourier transform on the signal y (t).
(4.2) combining the co-directional component r (t) and the quadrature component of the communication signal
Constructed as a two-dimensional matrix
From the perspective of generating modulated signal, two-dimensional matrix formed from homodromous component and orthogonal component of signal is formed
As the basic characteristics of the signal, the basic characteristics fully represent the modulation mode information carried by the modulation signal.
The invention applies Hilbert transform to the feature construction of communication signals, avoids the process of mapping high-dimensional input data to low-dimensional features, only uses a two-dimensional matrix formed by the same-dimensional component and the orthogonal component of the communication signals as the representation of the original communication signals, and the required input is easy to obtain in a communication system, so that the complexity of the feature construction is obviously reduced.
Example 5
The system and method for identifying modulation modes based on the convolutional neural network are the same as embodiments 1-4, and the convolutional neural network used for identifying and classifying the modulation modes of the communication signals in step 5 comprises two stages: a training phase and a classification phase.
(5.1) the training phase of the convolutional neural network comprises the following steps:
(5.1.a) randomly initializing weights and biases of each layer in the convolutional neural network: randomly generating weight W between input unit and hidden unitlAnd bias of hidden unit blL, L is the depth of the neural network, and L is the total depth of the network;
(5.1.b) inputting training data x consisting of two-dimensional matrix and labels y thereof, wherein the training sample set is { (x)(1),y(1)),…(x(n),y(n)) The samples belong to U categories, x (i) is the ith sample, y (i) is the corresponding category label of the ith sample x (i), and the output value h of each layer of the convolutional neural network is calculated by utilizing a forward conduction formula(k):
Wherein a (a)k)Is the input value h of the k layer of the convolutional neural network corresponding to the input sample x(k)To correspond to the input value a(k)An output value obtained through a k-th layer network, wherein k is a network depth; in other words, a(k)Is the input value of the k-th layer, h(k)Is the output value of the k-th layer.
First, the training sample is assigned to the input h of the network(0)Then according to the network input value a(k)And a network output value h(k)The formula (A) is operated layer by layer, and the output value h of the L-th layer is obtained by L times of calculation iteration termination(L)The value is the output of the input data through the convolutional neural network. In the calculation process, in orderThe method and the device prevent overfitting of a neural network model caused by too little training data, namely prevent the phenomenon that a convolutional neural network model has a lower loss function on training data and a higher prediction accuracy rate but has a higher loss function on test data and a lower accuracy rate, and set the probability of convolutional layer random inactivation (dropout) in the forward transmission process of the convolutional network, wherein the value range of the probability is 40-50%. The specific operation is as in this example the output value h of the k-th network(k)The value is zero with a certain probability. The random inactivation rate refers to that the output of a part of neurons is set to be zero in the convolutional neural network with a certain probability. In this example, the probability of the random deactivation of the neuron is taken to be 40%, that is, the output value h of the k-th network in this example(k)The value is zero with a probability of 40%.
(5.1.c) selecting a cross entropy cost function from the cost function of the network:
calculating the residual error of the last layer according to the label value and the output value of the last layer
Wherein,
representing the output of the ith neuron of the L layer of the output layer, yiIs the ith element in the label and U is the total number of classes of modulation.
(5.1.d) updating weights and biases in the convolutional neural network according to a gradient descent rule:
wherein, α is a constant and represents a learning rate, and the smaller the learning rate is, the finer the network learning becomes, but the function convergence speed is too slow; if the learning rate is too large, the function may not converge. The reference values of the learning rate are 0.001, 0.003, 0.01, 0.03, 0.1, 0.3, and the like. The two equations above stop updating if the iteration reaches the maximum number of iterations or if the network cost function reaches a certain threshold. In this example, the condition for ending the selected iteration is that the maximum number of iterations is reached, and the maximum number of iterations is 200. Solving partial derivatives in the above two equations
And
calculated using a back propagation algorithm. When a back propagation algorithm is applied, firstly forward propagation is carried out, then the residual error of each layer is obtained by utilizing the residual error value calculated in the step (5.1.c), so that the residual error of each layer is obtained, and finally the residual error of each layer is obtained
And
the value of (c).
(5.2) the classification stage of the convolutional neural network comprises the following steps:
and (5.2.a) keeping all parameters in the convolutional neural network unchanged, and calculating the output values of the test samples passing through all hidden layers of the convolutional neural network layer by layer.
(5.2.b) calculating a class vector value output by the output layer of the convolutional neural network, wherein the one-dimensional class vector value is the probability of classifying the signal into each sample in the training samples, and the dimension of the one-dimensional vector is the total class number U of the modulation mode in the invention. Selecting softmax function as the last layer of the convolutional neural network, and inputting the function as netjThe output value through the function is
netjRepresenting the jth element in the input vector net.
And (5.2.c) determining the modulation mode of the test sample according to the position of the maximum value in the output class vector values of the convolutional neural network, namely, the position of the maximum value in the output class vector corresponds to the modulation mode of the identified communication signal.
With the development and maturity of deep learning, the invention innovatively provides a modulation mode identification system and method based on the convolutional neural network, the classification capability of the convolutional neural network to the two-dimensional graph in the deep learning is applied to the field of modulation mode identification, the displayed feature extraction is avoided, and the complexity of the network is greatly reduced by a weight sharing mode.
Example 6
The system and the method for identifying the modulation mode based on the convolutional neural network are the same as the embodiments 1 to 5, and referring to fig. 1, the implementation steps of the invention are as follows:
(1) the sending end modulates a sending signal and carries out pulse forming: the signals selected for use in this example are quaternary quadrature amplitude modulation 4QAM signals, hexadecimal quadrature amplitude modulation 16QAM signals, thirty-two quadrature amplitude modulation 32QAM signals, and sixty-four quadrature amplitude modulation 64QAM signals. The signal selected in this time is four signals satisfying high-speed signal transmission, in this example, M ═ 256 symbols are randomly generated in each of four modulation modes, and the symbol rate R iss1Kbps, carrier rate fc2KHz, sample rate FsThe number of samples of the truncated signal is 8192 in this case at 32 KHz.
(2) The signal passes through the channel, in this example, the channel environment selects the rice fading channel, and the performance of the identification method in the rice channel propagation environment is observed.
(3) A preprocessing module: at a receiving end, carrier frequency estimation is carried out on the signal by using a costas loop, and then down-conversion operation is carried out on the received signal by using the estimated carrier frequency to obtain a down-converted baseband signal r (t), wherein the signal is also the homodromous component of the analytic signal. Base in time interval T after down conversionThe band signal r (t) is expressed in equation 1. In this example g (T) is a square root raised cosine pulse, TsIs the symbol interval. Referring to fig. 1, in the system of the present invention, the preprocessor of fig. 1 is used to perform the preprocessing. The preprocessing mainly completes the functions of carrier frequency estimation, signal down-conversion, filtering operation and the like.
(4) Simple structure characteristics: although the four signals in the experiment are difficult to distinguish at low signal-to-noise ratios, the truncated signal selected this time is long enough so that the truncated signal is sufficient to characterize the modulated signal. The method comprises the following specific steps: subjecting the in-phase component r (t) to Hilbert transform to obtain the quadrature component of the signal
Combining the co-directional component r (t) and the quadrature component of the signal
Constructed as a two-dimensional matrix
This two-dimensional matrix is a simple feature of construction.
(5) The convolution neural network fuses local receptive fields, weight sharing and down sampling together to ensure displacement, scale and rotation invariance to a certain degree. In a convolutional neural network, the input to each cell in each layer is derived from several adjacent neurons in the previous layer. This approach of connecting together neurons and the input local receptive fields dates back to 1960. The local receptive field may extract some primary features such as edges, corners, etc. of the image. The convolutional neural network is used for the identification process of the modulation mode of the communication signal and comprises a learning training stage and a classification stage, wherein the learning training stage comprises the following steps:
(5.1.a) randomly generating a weight W between the input unit and the hidden unitlAnd bias of hidden unit blL, L is the depth of the network, L is the total number of depths of the network;
(5.1.b) inputting training data x consisting of two-dimensional matrix and labels y thereof, wherein the training sample set is { (x(1),y(1)),…(x(n),y(n)) They belong to U categories, x(i)Is a sample, y(i)Is a sample x(i)A corresponding category label. The specific steps are the same as above.
In the calculation process, in order to prevent the phenomenon that the loss function of the model on the training data is higher than the smaller prediction accuracy, but the loss function on the test data is higher than the larger prediction accuracy, the probability of convolutional layer random inactivation (dropout) is set, and in the example, the probability of the random inactivation of the neuron is 45%.
And (5.1.c) selecting a cross entropy cost function from the cost function of the network, and calculating the residual error of the last layer, which is the same as the formula (8).
(5.1.d) updating the weights and biases in the convolutional neural network according to the gradient descent rule, as in equation (9). In this example, the learning rate is 0.001.
(5.2) the classification stage of the convolutional neural network comprises the following steps:
and (5.2.a) keeping all parameters in the convolutional neural network unchanged, calculating the output values of the test samples passing through all hidden layers of the convolutional neural network layer by layer, wherein when the output values of all the hidden layers are calculated, the activation function can be a sigmoid function, a tanh activation function and a Relu activation function, and in the example, the activation function is the Relu activation function.
And (5.2.b) calculating a class vector value output by the output layer of the convolutional neural network, wherein the activation of the output layer of the convolutional neural network selects a softmax function.
(5.2.c) determining the modulation mode of the test sample according to the position of the maximum value in the output class vector values of the convolutional neural network, namely determining the position of the maximum value in the output class vector to be the identified modulation mode.
A more detailed example is given below to further illustrate the present invention.
Example 7
The system and the method for identifying the modulation mode based on the convolutional neural network are the same as the embodiments 1 to 6, and referring to fig. 1, the implementation steps of the invention are as follows:
(1) the sending end modulates a sending signal and carries out pulse forming: the signal selected for use in this example isBinary Phase Shift Keying (BPSK) signals, Quaternary Phase Shift Keying (QPSK) signals, octal phase shift keying (8 PSK) signals and hexadecimal amplitude modulation (16 QAM) signals. When the number of the types of the signals increases, the length of the signals can be increased to improve the recognition rate of the modulation system. The length of input data in the modulation mode identification system is mainly determined by the length M of code element data in a truncated signal and a sampling rate of a receiving end, and the shorter the input data is, the fewer the number of neurons needing to be trained in the convolutional neural network is. In this example, M ═ 16 symbols are randomly generated for each of the four modulation schemes, and the symbol rate R iss1Mbps, carrier rate fc2MHz, sample rate Fs＝16MHz。
(2) The signal passes through the channel: the pulse signal is carried on a high-frequency carrier wave and then sent out, and the additive white Gaussian noise is added to the modulation signal through an additive white Gaussian noise channel to obtain a mixed signal subjected to the interference of the additive white Gaussian noise. The channel environment can select a Rayleigh fading channel or a Rice channel, the performance of the identification method in a real channel propagation environment can be obtained by observing the identification result of the system under the multipath Rayleigh fading channel, and the identification rate of the modulation mode identification system under the Gaussian white noise channel is the highest under the condition that other residual conditions are kept the same through research and experiments.
(3) A preprocessing module: in this example, at the receiving end, the carrier frequency estimation is performed on the signal by using the costas loop, and then the down-conversion operation is performed on the received signal by using the estimated carrier frequency, so as to obtain a down-converted baseband signal r (t), which is also the homodromous component of the analytic signal. The down conversion operation brings carrier phase tracking error to the signal
This error is inevitable in real systems.
Referring to fig. 1, in the system of the present invention, the preprocessor of fig. 1 is used to perform the preprocessing. The preprocessing mainly completes the functions of carrier frequency estimation, signal down-conversion, filtering operation and the like.
(4) Simple structure characteristics: in general useThe traditional characteristics comprise instantaneous amplitude, instantaneous phase, instantaneous angular frequency, high-order cumulant parameters, frequency spectrum, power spectrum and the like of the signal, the traditional characteristics are obtained by a large amount of calculation, and the length of the truncated signal is long enough. The invention can process high-dimensional data by utilizing the deep learning capability, directly takes the communication signal as the original input without starting from the traditional characteristic angle, fully utilizes the characteristics of the signal, and comprises the following specific steps: subjecting the in-phase component r (t) to Hilbert transform to obtain the quadrature component of the signal
Combining the co-directional component r (t) and the quadrature component of the signal
Constructed as a two-dimensional matrix
This two-dimensional matrix is a simple feature of construction.
(5) In the conventional method, after the characteristics of the signals are confirmed, a conventional classifier is required to be added to classify the signals, such as a decision tree classifier, an artificial neural network classifier, a hierarchical network and the like. However, the traditional classifier has the disadvantages of slow iteration process and low recognition efficiency. The invention adopts a convolutional neural network with the capability of extracting abstract features from input data to a high degree as a classifier. The convolutional neural network is mainly used for various tasks of image and video analysis, such as image classification, face recognition, object recognition, image segmentation and the like, and the accuracy rate of the convolutional neural network is generally far higher than that of other neural network models. In a convolutional neural network, feature maps are features extracted from an input by convolution, and each feature map is a class of features extracted from the input. For the representation capability of the convolutional network, a number of different feature maps may be used in each layer to better represent the features of the image. In this example, referring to fig. 2, the convolutional neural network uses 32, 16 feature maps to characterize the input features layer by layer. The weights on each feature map are the same, and the weight sharing mechanism enables the number of parameters needing to be trained of the convolutional neural network to be greatly reduced compared with the number of parameters needing to be trained of a fully-connected network.
The convolutional neural network is used for the identification process of the modulation mode of the communication signal and comprises a learning training stage and a classification stage, wherein the learning training stage comprises the following steps:
(5.1.a) randomly generating a weight W between the input unit and the hidden unitlAnd bias of hidden unit blL, L is the depth of the network, L is the total number of depths of the network.
(5.1.b) inputting training data x consisting of two-dimensional matrix and labels y thereof, wherein the training sample set is { (x)(1),y(1)),…(x(n),y(n)) They belong to m classes, x(i)Is a sample, y(i)Is a sample x(i)The corresponding class label, the output value of each layer is calculated using equation (6). In the calculation process, in order to prevent the data from generating overfitting, the probability of random inactivation (dropout) of the convolutional layer is set, and the overfitting can be obviously reduced by setting the random inactivation probability. In this example, the probability of taking the neurons to inactivate randomly was 50%.
In this example, 10000 training samples are the same as 10000 testing samples. The number of training samples and the test samples are generated from randomly generated communication signals, and the sample data and the labels are randomly shuffled before being fed into the convolutional neural network.
(5.1.c) the cost function of the network selects a cross entropy cost function, specifically shown in a formula (7), and a residual error formula is shown in a formula (8).
(5.1.d) updating the weights and biases in the convolutional neural network according to the gradient descent rule, see equation (9).
In this example, the learning rate is 0.0035.
(5.2) the convolutional neural network for classification stage steps are as follows:
and (5.2.a) keeping all parameters in the convolutional neural network unchanged, and calculating the output values of the test samples passing through all hidden layers of the convolutional neural network layer by layer.
(5.2.b) calculating a class vector value output by the output layer of the convolutional neural network, wherein the one-dimensional class vector value is the probability of classifying the signal into each sample in the training samples, and the dimension of the one-dimensional vector is the total class number U of the modulation mode in the invention. And selecting a softmax function as the final layer of the activation function of the convolutional neural network.
(5.2.c) determining the modulation mode of the test sample according to the position of the maximum value in the output class vector values of the convolutional neural network, namely determining the position of the maximum value in the output class vector to be the identified modulation mode.
(6) And sending the modulation mode of the signal identified by the convolutional neural network to a communication signal demodulation module, and demodulating the received signal to obtain a demodulated communication signal.
The method applies the properties that the convolutional neural network can identify the distortion invariance of the two-dimensional graph in deep learning to the identification field of the modulation mode of the communication signal, solves the problems of simplifying and representing the characteristics of the communication signal and fully utilizing the modulation information carried by the communication signal, avoids explicit characteristic extraction, and weakens the requirement on the characteristic effectiveness in the traditional modulation mode identification system. The ability of convolutional neural networks to identify two-dimensional patterns of displacement, scaling and other forms of distortion invariance.
Compared with the traditional modulation mode identification method, the method overcomes the defect that the modulation signal information cannot be fully utilized in the prior art, can obtain higher identification accuracy under the same condition, and improves the system performance.
The effects of the present invention can be explained again by the following simulations:
example 8
The convolutional neural network-based modulation mode identification system and method are the same as those of embodiments 1 to 7,
1. simulation conditions
Matlab R2016a and Pycharm simulation software are used, system simulation parameters are consistent with those described in the examples, a transmission channel is an additive white Gaussian noise channel, 10000 groups of training data and 10000 groups of test data are randomly generated. The signal-to-noise ratio is selected within a range of 0-10 db.
2. Emulated content
The method provided by the invention and the method provided by the A.LIU are respectively subjected to modulation mode identification performance simulation, and the average value of the identification performance is calculated.
3. Simulation result
The classification performance simulation curve, as shown in fig. 3, in which the curve with "box" represents the recognition performance of the system of the present invention, and the curve with "meter shape" represents the performance curve of the algorithm proposed by a.liu. In fig. 3, the horizontal axis represents the signal power to noise power ratio in decibels, and the vertical axis represents the classification accuracy.
As can be seen from the simulation results of fig. 3, under the condition of the same communication signal power to noise power ratio, the classification accuracy of the present invention is better than the performance of the algorithm proposed by a.liu. As can be seen from fig. 3, the recognition accuracy of the method at 0db is 94.5%, which is 12% higher than the performance of the method of a.liu; the invention can reach 99.99% accuracy at 1 dB, and the performance is stable with the improvement of SNR, while A.LIU can only reach 91.2% accuracy, it can obtain higher identification accuracy at low SNR, and improve the system performance, which proves that the invention can fully extract the characteristics carried by the communication signal and make classification.
In summary, the modulation mode identification system and method based on the convolutional neural network disclosed by the invention mainly solve the problems that the prior art has complex feature extraction steps, almost all the features required by the system are manually designed, and the identification rate is low under a low signal-to-noise ratio. The modulation mode identification system is sequentially connected with a system sending end, a channel and a receiving end according to the transmission direction of signals, the signals are sequentially subjected to demodulation operation after passing through the units, and the receiving end is sequentially connected with preprocessing operation, a simple feature construction module and a convolutional neural network module. The identification method comprises the following steps: at a transmitting end, mapping and modulating a transmitting signal according to different constellation diagrams; the sending signal is sent out through an additive white Gaussian noise channel after up-conversionRemoving; the receiving end firstly carries out preprocessing operation and then obtains a signal r (t) transmitted through a channel, wherein the signal is the homodromous component of the signal; constructing a simple feature module to convert the homodromous component r (t) and the quadrature component of the baseband signal
Constructed as a two-dimensional matrix
Carrying out feature classification on the signals by the convolutional neural network; and sending the modulation mode of the signal identified by the convolutional neural network to a signal demodulation module, and demodulating the received signal to obtain a demodulated communication signal. The invention has the advantages of low complexity of feature design, avoidance of explicit feature extraction and high classification accuracy, and can be used in a communication system with higher requirement on identification performance.
Claims (5)
1.A modulation mode identification system based on a convolutional neural network is sequentially connected with a system sending end, a channel and a receiving end according to the transmission direction of signals, the signals sequentially pass through the units and then are demodulated, the receiving end is sequentially connected with a preprocessing module, a simple characteristic construction module and a convolutional neural network module, and the modulation mode of the modulation signals identified by the convolutional neural network is used for signal demodulation
Two-dimensional matrix of constructions
And as the simple features of the signals, the simple features of the signals are sent to a convolutional neural network module for abstract feature extraction and classification, and the modulation mode of the communication signals is obtained.
2. The convolutional neural network-based modulation scheme recognition system of claim 1, wherein the convolutional neural network is a multi-layer deep convolutional neural network comprising an input layer, four convolutional layers and four downsampling layers, two fully-connected layers and an output layer.
3. A modulation mode identification method based on a convolutional neural network is characterized by comprising the following steps:
(1) the sending end modulates a sending signal and carries out pulse forming: at a transmitting end, mapping and modulating a transmitting signal according to different constellation diagrams, performing pulse forming by adopting rectangular pulses, and selecting a code element length M;
(2) the signal passes through the channel: the pulse signal is sent out after up-conversion and passes through an additive white Gaussian noise channel;
(3) and (3) preprocessing operation: the receiving end firstly carries out carrier frequency estimation on the signal, then carries out down-conversion operation on the received signal to obtain a baseband signal r (t) transmitted through a channel, and the signal is the in-phase component of an analytic signal;
(4) simple structure characteristics: the simple-structure characteristic module performs Hilbert transform on the in-phase component r (t) of the analytic signal to obtain the orthogonal component of the analytic signal
The in-phase component r (t) and the quadrature component of the analytic signal
Constructed as a two-dimensional matrix R;
(5) the convolutional neural network classifies the signals: the convolutional neural network module is used for learning, training and classifying the two-dimensional matrix R;
(5.1) convolutional neural network learning training stage:
randomly initializing the weight and the bias of each layer in the convolutional neural network; inputting a training sample consisting of a two-dimensional matrix and a label thereof, and calculating an output value of each layer by using a forward conduction formula; calculating the residual error of the last layer according to the label value and the output value of the last layer; updating the weight of the network according to a gradient descent method to finish the training of the convolutional neural network;
(5.2) a convolutional neural network classification stage:
keeping the parameters in the convolutional neural network unchanged, and calculating the output values of the test samples passing through all layers of the hidden layer of the convolutional neural network layer by layer; calculating a class vector value output by an output layer of the convolutional neural network; determining a modulation mode of the test sample according to the position of the maximum value in the convolutional neural network output category vector values;
(6) and sending the modulation mode of the signal identified by the convolutional neural network to a signal demodulation module, and demodulating the received signal to obtain a demodulated communication signal.
4. The convolutional neural network-based modulation scheme identification method as claimed in claim 3, wherein the simple structure feature in step 4 specifically comprises:
(4.1) performing Hilbert transform on the in-phase component r (t) obtained in the step (3) to obtain an orthogonal component
Wherein the parameter t represents the signal duration;
(4.2) combining the in-phase component r (t) and the quadrature component
Constructed as a two-dimensional matrix
5. The convolutional neural network-based modulation scheme identification method as claimed in claim 3, wherein the training phase in step 5 specifically comprises the following steps:
(5.1.a) randomly initializing weights W of each layer in the convolutional neural networklAnd bias blL, L is the depth of the network, and L is the total depth of the network;
(5.1.b) inputting training data x consisting of a two-dimensional matrix and a label y thereof, calculating an output value of each layer by using a forward conduction formula, and setting the probability of random inactivation (dropout) of the convolutional layer to be 40-50% in the calculation process;
(5.1.c) calculating the residual error of the last layer according to the label value and the output value of the output layer;
and (5.1.d) updating the weight of the network according to a gradient descent method to finish the training of the convolutional neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711403640.7A CN108234370B (en) | 2017-12-22 | 2017-12-22 | Communication signal modulation mode identification method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711403640.7A CN108234370B (en) | 2017-12-22 | 2017-12-22 | Communication signal modulation mode identification method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108234370A CN108234370A (en) | 2018-06-29 |
| CN108234370B true CN108234370B (en) | 2020-12-15 |
Family
ID=62647728
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711403640.7A Active CN108234370B (en) | 2017-12-22 | 2017-12-22 | Communication signal modulation mode identification method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108234370B (en) |
Families Citing this family (54)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109120563B (en) * | 2018-08-06 | 2020-12-29 | 电子科技大学 | Modulation recognition method based on neural network integration |
| CN109246048B (en) * | 2018-10-30 | 2021-02-02 | 广州海格通信集团股份有限公司 | Physical layer secure communication method and system based on deep learning |
| WO2020087293A1 (en) * | 2018-10-30 | 2020-05-07 | 华为技术有限公司 | Communication receiver and method for processing signal |
| CN109523023A (en) * | 2018-11-16 | 2019-03-26 | 泰山学院 | A kind of deep learning network and system for subsurface communication Modulation Identification |
| CN109274626B (en) * | 2018-11-21 | 2020-11-13 | 电子科技大学 | Modulation identification method based on constellation diagram orthogonal scanning characteristics |
| CN109361635B (en) * | 2018-11-23 | 2021-06-04 | 泰山学院 | Underwater communication modulation mode identification method and system based on depth residual error network |
| CN109657604B (en) * | 2018-12-17 | 2021-02-12 | 中国人民解放军战略支援部队信息工程大学 | Satellite amplitude-phase signal identification and demodulation method and device based on cyclic neural network |
| CN109886075A (en) * | 2018-12-27 | 2019-06-14 | 成都数之联科技有限公司 | A kind of signal modulation pattern recognition methods based on planisphere |
| CN109787927A (en) * | 2019-01-03 | 2019-05-21 | 荆门博谦信息科技有限公司 | Modulation Identification method and apparatus based on deep learning |
| CN109672639B (en) * | 2019-01-22 | 2021-07-16 | 哈尔滨工业大学 | Signal demodulation method based on machine learning |
| CN109714086B (en) * | 2019-01-23 | 2021-09-14 | 上海大学 | Optimized MIMO detection method based on deep learning |
| CN109787929A (en) * | 2019-02-20 | 2019-05-21 | 深圳市宝链人工智能科技有限公司 | Signal modulate method, electronic device and computer readable storage medium |
| CN110086737B (en) * | 2019-03-13 | 2021-07-02 | 西安电子科技大学 | Communication signal modulation mode identification method based on graph neural network |
| CN109951409B (en) * | 2019-04-04 | 2020-06-16 | 四川九洲电器集团有限责任公司 | Method and system for determining modulation signal category |
| CN110071885A (en) * | 2019-04-17 | 2019-07-30 | 成都华日通讯技术有限公司 | A kind of deep learning method of discrimination of PSK digital signal subclass Modulation Identification |
| CN110048980A (en) * | 2019-04-19 | 2019-07-23 | 中国电子科技集团公司第三十六研究所 | A kind of blind demodulation method of digital communication and device |
| CN110210330B (en) * | 2019-05-13 | 2021-05-28 | 清华大学 | Electromagnetic signal identification method and device for constructing graph convolution network based on implicit knowledge |
| CN110324080A (en) * | 2019-06-28 | 2019-10-11 | 北京邮电大学 | A kind of method, apparatus of optical information networks, electronic equipment and medium |
| CN110191073B (en) * | 2019-06-28 | 2021-08-31 | 华侨大学 | Modulation mode identification method based on deep learning and suitable for changing scene |
| CN110427835B (en) * | 2019-07-11 | 2022-03-11 | 清华大学 | Electromagnetic signal identification method and device for graph convolution network and transfer learning |
| CN110515096B (en) * | 2019-08-20 | 2023-03-31 | 东南大学 | Convolutional neural network-based satellite navigation interference signal identification device and method |
| CN110536257B (en) * | 2019-08-21 | 2022-02-08 | 成都电科慧安科技有限公司 | Indoor positioning method based on depth adaptive network |
| CN110826425B (en) * | 2019-10-21 | 2022-11-18 | 大连理工大学 | VHF/UHF frequency band radio signal modulation mode identification method based on deep neural network |
| CN111079510A (en) * | 2019-10-25 | 2020-04-28 | 北京百卓网络技术有限公司 | Signal modulation type identification method and device |
| CN110956197A (en) * | 2019-10-28 | 2020-04-03 | 新绎健康科技有限公司 | Method and system for establishing pulse wave noise signal identification model based on convolutional neural network |
| CN111049770B (en) * | 2019-12-06 | 2021-11-02 | 西安电子科技大学 | Modulation signal identification method based on high-order cumulant |
| CN110855591B (en) * | 2019-12-09 | 2021-10-29 | 山东大学 | QAM and PSK signal intra-class modulation classification method based on convolutional neural network structure |
| CN111404852B (en) * | 2020-03-03 | 2021-04-13 | 西安电子科技大学 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
| CN111507205B (en) * | 2020-03-27 | 2022-09-27 | 中国人民解放军海军工程大学 | Modulation recognition device based on signal cyclic spectrum and deep learning and use method thereof |
| CN111563827B (en) * | 2020-04-10 | 2023-08-04 | 杭州电子科技大学 | Load decomposition method based on physical characteristics of electrical appliance and electricity utilization behaviors of residents |
| CN111786915B (en) * | 2020-05-09 | 2023-03-14 | 广州番禺职业技术学院 | Wireless communication method and system based on deep learning under nonlinear distortion condition |
| CN111614398B (en) * | 2020-05-12 | 2021-06-11 | 北京邮电大学 | Method and device for identifying modulation format and signal-to-noise ratio based on XOR neural network |
| CN111711586B (en) * | 2020-06-15 | 2021-07-13 | 吉林大学 | Time-frequency collaborative filtering method and system applied to communication modulation mode identification |
| CN111935040B (en) * | 2020-07-07 | 2022-08-09 | 同济大学 | Wireless signal modulation mode identification system based on USRP and neural network |
| CN112101460B (en) * | 2020-09-16 | 2024-07-12 | 中国电子科技集团公司第三十六研究所 | Spectrum sensing method and device based on deep learning classification |
| CN112347871B (en) * | 2020-10-23 | 2023-04-07 | 中国电子科技集团公司第七研究所 | Interference signal modulation identification method for communication carrier monitoring system |
| CN112364729A (en) * | 2020-10-29 | 2021-02-12 | 成都明杰科技有限公司 | Modulation identification method based on characteristic parameters and BP neural network |
| CN112241724B (en) * | 2020-10-30 | 2023-12-15 | 南京信息工程大学滨江学院 | Automatic identification method and system based on double-path convolution long-term neural network |
| CN112615804B (en) * | 2020-12-12 | 2022-08-02 | 中国人民解放军战略支援部队信息工程大学 | Short burst underwater acoustic communication signal modulation identification method based on deep learning |
| CN112887239B (en) * | 2021-02-15 | 2022-04-26 | 青岛科技大学 | Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network |
| CN112910812B (en) * | 2021-02-25 | 2021-10-22 | 电子科技大学 | Modulation mode identification method for deep learning based on space-time feature extraction |
| CN113014524B (en) * | 2021-03-03 | 2021-12-14 | 电子科技大学 | Digital signal modulation identification method based on deep learning |
| CN113128394A (en) * | 2021-04-14 | 2021-07-16 | 山东大学 | Ultra-wideband channel classification method and system |
| CN113141325B (en) * | 2021-04-14 | 2022-07-19 | 北京邮电大学 | Training method, identification method and device for optical OFDM signal subcarrier modulation format identification model |
| CN113259289B (en) * | 2021-05-13 | 2022-06-17 | 曾泓然 | Single-channel aliasing signal modulation mode identification method based on residual error neural network |
| CN113395225B (en) * | 2021-06-10 | 2023-01-31 | 中国人民解放军战略支援部队信息工程大学 | Universal intelligent processing method and device for directly receiving communication signal waveform to bit |
| CN113630130B (en) * | 2021-07-31 | 2022-12-23 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | End-to-end digital communication demodulation method |
| CN114362881B (en) * | 2021-12-16 | 2023-06-06 | 中国人民解放军战略支援部队信息工程大学 | Space-time block code blind identification method and system based on space-time correlation matrix |
| CN114301747B (en) * | 2021-12-28 | 2023-12-29 | 杭州电子科技大学 | Modulation signal identification method based on SVM-SRNN |
| CN114615118B (en) * | 2022-03-14 | 2023-09-22 | 中国人民解放军国防科技大学 | Modulation identification method based on multi-terminal convolution neural network |
| CN114745248B (en) * | 2022-04-06 | 2023-11-14 | 山东大学 | DM-GSM signal detection method based on convolutional neural network |
| CN114553650B (en) * | 2022-04-27 | 2022-07-29 | 南京信息工程大学 | Multi-level neural network-based anti-mode coupling signal complex format analysis method |
| CN115314348B (en) * | 2022-08-03 | 2023-10-24 | 电信科学技术第五研究所有限公司 | QAM signal modulation identification method based on convolutional neural network |
| CN117807529B (en) * | 2024-02-29 | 2024-05-07 | 南京工业大学 | Modulation mode identification method and system for output signals of signal generator |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5271036A (en) * | 1990-11-16 | 1993-12-14 | Thomson-Csf | Method and device for the recognition of modulations |
| CN104866810A (en) * | 2015-04-10 | 2015-08-26 | 北京工业大学 | Face recognition method of deep convolutional neural network |
| CN107220606A (en) * | 2017-05-22 | 2017-09-29 | 西安电子科技大学 | The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks |
| CN107342962A (en) * | 2017-07-03 | 2017-11-10 | 北京邮电大学 | Deep learning intelligence Analysis On Constellation Map method based on convolutional neural networks |
-
2017
- 2017-12-22 CN CN201711403640.7A patent/CN108234370B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5271036A (en) * | 1990-11-16 | 1993-12-14 | Thomson-Csf | Method and device for the recognition of modulations |
| CN104866810A (en) * | 2015-04-10 | 2015-08-26 | 北京工业大学 | Face recognition method of deep convolutional neural network |
| CN107220606A (en) * | 2017-05-22 | 2017-09-29 | 西安电子科技大学 | The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks |
| CN107342962A (en) * | 2017-07-03 | 2017-11-10 | 北京邮电大学 | Deep learning intelligence Analysis On Constellation Map method based on convolutional neural networks |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108234370A (en) | 2018-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108234370B (en) | Communication signal modulation mode identification method based on convolutional neural network | |
| CN110855591B (en) | QAM and PSK signal intra-class modulation classification method based on convolutional neural network structure | |
| Peng et al. | A survey of modulation classification using deep learning: Signal representation and data preprocessing | |
| CN110163282B (en) | Modulation mode identification method based on deep learning | |
| CN109657604B (en) | Satellite amplitude-phase signal identification and demodulation method and device based on cyclic neural network | |
| CN112702294B (en) | Modulation recognition method for multi-level feature extraction based on deep learning | |
| CN108718288B (en) | Method for recognizing digital signal modulation mode based on convolutional neural network | |
| CN109120563B (en) | Modulation recognition method based on neural network integration | |
| CN108052956A (en) | Wireless light communication subcarrier modulation constellation recognition methods under a kind of atmospheric turbulance | |
| CN114422311B (en) | Signal modulation recognition method and system combining deep neural network and expert priori features | |
| CN112347871A (en) | Interference signal modulation identification method for communication carrier monitoring system | |
| Lin et al. | A real-time modulation recognition system based on software-defined radio and multi-skip residual neural network | |
| An et al. | Robust automatic modulation classification in low signal to noise ratio | |
| Zhang et al. | Training images generation for CNN based automatic modulation classification | |
| Almohamad et al. | Dual-determination of modulation types and signal-to-noise ratios using 2D-ASIQH features for next generation of wireless communication systems | |
| Chen et al. | Automatic modulation classification using multi-scale convolutional neural network | |
| CN116628566A (en) | Communication signal modulation classification method based on aggregated residual transformation network | |
| CN114615118A (en) | Modulation identification method based on multi-terminal convolution neural network | |
| Zhang et al. | Research on modulation recognition algorithm based on channel and spatial self-attention mechanism | |
| Cai et al. | The performance evaluation of big data-driven modulation classification in complex environment | |
| CN115913850B (en) | Open set modulation identification method based on residual error network | |
| CN115086123B (en) | Modulation identification method and system based on fusion of time-frequency diagram and constellation diagram | |
| Hazza et al. | Robustness of digitally modulated signal features against variation in HF noise model | |
| Ma et al. | Deep learning based cognitive radio modulation parameter estimation | |
| Zhou et al. | Automatic modulation classification with genetic backpropagation neural network |
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 |