CN116684233A - Communication signal modulation identification method based on image significance detection - Google Patents
Communication signal modulation identification method based on image significance detection Download PDFInfo
- Publication number
- CN116684233A CN116684233A CN202310710751.1A CN202310710751A CN116684233A CN 116684233 A CN116684233 A CN 116684233A CN 202310710751 A CN202310710751 A CN 202310710751A CN 116684233 A CN116684233 A CN 116684233A
- Authority
- CN
- China
- Prior art keywords
- communication signal
- image
- frequency
- time
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004891 communication Methods 0.000 title claims abstract description 144
- 238000000034 method Methods 0.000 title claims abstract description 69
- 238000001514 detection method Methods 0.000 title claims abstract description 47
- 238000012549 training Methods 0.000 claims description 83
- 238000012360 testing method Methods 0.000 claims description 47
- 238000012795 verification Methods 0.000 claims description 41
- 238000010586 diagram Methods 0.000 claims description 28
- 230000000694 effects Effects 0.000 claims description 24
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 10
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 239000000654 additive Substances 0.000 claims description 3
- 230000000996 additive effect Effects 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000011176 pooling Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/32—Carrier systems characterised by combinations of two or more of the types covered by groups H04L27/02, H04L27/10, H04L27/18 or H04L27/26
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
A communication signal modulation recognition method based on image significance detection relates to a communication signal modulation recognition method. The invention aims to solve the problem that the existing communication signal modulation recognition method can not effectively remove background noise in a time-frequency image under the condition of low signal-to-noise ratio, so that the modulation recognition accuracy of a communication signal is low. The invention can effectively remove the background noise in the time-frequency image under the condition of low signal-to-noise ratio and improve the modulation recognition accuracy of the communication signal. The invention belongs to the technical field of communication signal modulation and identification.
Description
Technical Field
The invention relates to a communication signal modulation recognition method, and belongs to the technical field of communication signal modulation recognition.
Background
In the field of communication, communication signal modulation and identification is an important technology, which provides a key foundation for realizing reliable information data transmission and efficient operation of a communication system, and is widely applied to civil and military fields such as radio detection, spectrum management, battlefield reconnaissance, electronic countermeasure and the like. However, with the increasing complexity of the communication environment in recent years, there is a lot of noise and interference in the communication channel, and the received signal is heavily distorted due to the doping of a lot of noise, so that the characteristics of the received signal are affected, and the modulation recognition accuracy of the communication signal is reduced. Therefore, many researchers aim to improve and explore new methods and technologies in the field of communication signal modulation recognition so as to overcome the influence of a large amount of noise on modulation recognition and effectively improve the modulation recognition effect of communication signals under low signal-to-noise ratio.
The university of western security electronics science and technology in 2019 discloses a modulation mode identification method (Chinese patent application number: 201910430613.1) based on deep learning, wherein the method comprises the steps of preprocessing an acquired communication signal sample set and extracting characteristics, and dividing the communication signal sample set into a training set and a testing set; setting up NNs of the neural network as the neural network for realizing modulation, identification and classification; dividing the training set into a plurality of sub-training sets based on the signal-to-noise ratio, and respectively training NNs of the neural network by using the sub-training sets to obtain a plurality of trained NNs neural network models; and finally, evaluating the signal-to-noise ratio of the modulation signal to be tested, selecting a suitable trained neural network according to the interval where the signal-to-noise ratio is located, and identifying the modulation mode of the modulation signal to be tested. Although the method can obtain high modulation recognition accuracy under different signal-to-noise ratios, different communication signal modulation recognition models are required to be selected according to the interval where the signal-to-noise ratio is, which means that priori information of the signal-to-noise ratio level of the signal to be detected is required to be known or estimated in advance, and the selection and storage of different modulation recognition models can bring additional complexity and workload to the whole modulation recognition process, and the universality and the practicability are lacked.
The late Wen Sheng et al of the university of electronic science and technology in 2021 discloses a communication signal modulation recognition method based on a BP neural network (Chinese patent application number: 202110071360.0), wherein the method segments an obtained original communication signal in advance, performs noise reduction treatment through low-pass filtering one by one, extracts characteristic parameters based on instantaneous statistics to construct multidimensional vectors as a training set, trains the built BP neural network, finally reconstructs the signal and the characteristic parameter vectors as a test set, and tests the modulation recognition classification effect of a generated model. The method utilizes the communication signal of the low-pass filter to perform noise reduction pretreatment, but the design of the low-pass filter needs to perform parameter selection according to specific application scenes and signal characteristics, and in actual use, the signal is subjected to smoothing treatment, so that details of the signal are possibly lost or blurred, and the subsequent characteristic extraction, recognition and classification of the signal are affected.
Qin Bowei et al at the air force engineering university of 2022 published an article "modulation recognition algorithm based on residual error generation countermeasure network" and disclosed a novel residual error generation countermeasure network designed by the article, the network was trained by using 10 communication signal data sets with different signal to noise ratios, and finally a generation model was tested, and the result showed that the modulation recognition accuracy of the network model could reach 91% under the conditions of a small sample and a signal to noise ratio of more than 0dB, and the validity of the method was verified. In this approach, however, the training set contains a variety of different signal-to-noise ratio communication signal data, which can increase the complexity and computational cost of the training process. In addition, the generalization capability of the network may be limited by the distribution of training data, and particularly in practical application, unknown signal-to-noise ratio conditions may be encountered, so that the modulation recognition classification effect on the actual acquisition signal is greatly reduced.
Disclosure of Invention
The invention aims to solve the problem that the conventional communication signal modulation recognition method cannot effectively remove background noise in a time-frequency image under the condition of low signal-to-noise ratio, so that the modulation recognition accuracy of a communication signal is low, and further provides a communication signal modulation recognition method based on image significance detection.
The technical scheme adopted by the invention for solving the problems is as follows: the method comprises the following specific steps:
setting related parameters of communication signals, constructing a communication signal data set, and carrying out noise adding processing;
step two, performing time-frequency analysis transformation on the communication signal data set to obtain a communication signal time-frequency image data set;
step three, processing the communication signal time-frequency image data set by using an image saliency detection method to obtain an image saliency communication signal time-frequency image data set, and dividing the image saliency communication signal time-frequency image data set into a training set, a verification set and a test set;
step four, building a VGG-16 network model;
training the VGG-16 network model built in the fourth step by using the training set divided in the third step, and adjusting super parameters in the model training process and carrying out preliminary evaluation on the correlation capacity of the super parameters by using the verification set divided in the third step;
step six, inputting the test set divided in the step three into the VGG-16 network model generated after training and verification, and outputting a classification test result of a related signal modulation mode, so as to evaluate the performance of a corresponding model after the image significance detection method is applied.
Further, the step one of setting related parameters of the communication signal, constructing a communication signal data set, and performing noise adding processing includes:
setting the number of code elements, carrier frequency, code element period and sampling frequency, and constructing eight communication signals of 2ASK, 2FSK, 4ASK, BPSK, QPSK, 8PSK, 16QAM and MSK according to the communication signal parameters, wherein the eight communication signals are denoted as s (n); the specific process of the noise adding processing is to add the additive white Gaussian noise v (n) with the signal-to-noise ratio interval of [0,10dB ] and the step length of 1dB into s (n), finally obtain the noise added communication signal x (n) =s (n) +v (n), and integrate all the generated x (n), thereby constructing the original communication signal data set to be used.
Further, the second step performs time-frequency analysis transformation on the communication signal data set, and the specific method for obtaining the communication signal time-frequency image data set comprises the following steps: and processing the original communication signal data set by a smooth pseudo-Wiggner-Willi distributed time-frequency conversion method, and finally obtaining the SPWVD time-frequency image data set of the communication signal.
Further, in the third step, the communication signal time-frequency image dataset is processed by using an image saliency detection method to obtain the image saliency communication signal time-frequency image dataset, and the image saliency communication signal time-frequency image dataset is divided into a training set, a verification set and a test set, wherein the specific steps are as follows:
processing the communication signal time-frequency image dataset by adopting an image saliency detection algorithm based on global contrast, and calculating a certain pixel point I of any time-frequency image I k Is (I) k ) The method comprises the following steps:
in the formula (1), I i The gray value of any pixel point in the time-frequency image I is in the range of [0,255 ]]And (3) further reconstructing to obtain:
in the formula (2), f n The frequency of the nth pixel point in the time-frequency image I is presented in a histogram form; the contrast between the gray value of each pixel point in the communication signal time-frequency diagram and the surrounding area can be compared through an LC algorithm to determine the significance degree of each pixel point, so that the influence of noise in a background area with a lower pixel value in the time-frequency diagram is weakened, the characteristics of a useful signal area are more obvious, and finally an image significance communication signal time-frequency diagram data set is obtained;
for the division of the image saliency communication signal time-frequency diagram data set, the data under the 6dB signal-to-noise ratio is divided into a training set, a verification set and a test set according to a certain proportion, the training set, the verification set and the test set are mainly used for training a network to generate a modulation recognition model, and then the data under other signal-to-noise ratios are used as the test set, so that whether the generalization capability of a corresponding network model can be improved by using the image saliency method for subsequent experiment verification is padded.
Further, the specific steps of the preliminary evaluation in the fifth step are as follows:
training the VGG-16 network model built in the fourth step by utilizing the time-frequency diagram data of the image significance communication signals in the training set, and setting corresponding super parameters of the network and training before training to obtain a corresponding modulation recognition network model; and checking the classification recognition effect of the model on the verification set in each iteration in the training process, finding the problem of the model or the parameter in time according to the effect, stopping training in time to perform corresponding adjustment, and finally primarily evaluating the generalization capability of the generated model by comparing with the classification effect of the training set, and searching the modulation recognition network model with the best training and verification effect based on image significance detection.
Further, in the step six, the step of evaluating the performance of the corresponding model after the image saliency detection method is applied is as follows:
and (3) inputting the time-frequency diagram data of the image saliency communication signals in the test set into the modulation recognition network model generated after training and verification in the step five for testing, judging the modulation recognition classification effect of the model in the set confidence-to-noise ratio interval according to the test result, thereby verifying the effectiveness of the used image saliency method in the aspect of noise reduction, and improving the classification recognition accuracy of the corresponding model in the low signal-to-noise ratio.
The beneficial effects of the invention are as follows:
1. the invention can effectively remove the background noise in the time-frequency image under the condition of low signal-to-noise ratio, and improve the modulation recognition accuracy of the communication signal;
2. compared with the traditional method for preprocessing and reducing noise of communication signals, the method has the following advantages: first, no significant a priori information of the communication signal is required; secondly, the calculation cost and the workload are low; thirdly, the problems of detail expression of the signals and the like are not affected; fourth, the generalization capability of the communication signal can be improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of SPWVD time-frequency conversion of 8 communication signals;
FIG. 3 is a schematic diagram of the SPWVD time-frequency graph effect of 8 communication signals after LC algorithm is applied;
FIG. 4 is a schematic diagram of a VGG-16 network architecture;
FIG. 5 is a graph showing the comparison of classification accuracy without applying a model corresponding to the application of the LC algorithm for different SNRs;
fig. 6 is a graph of a test confusion matrix for a corresponding network model using LC algorithms at a 0dB signal-to-noise ratio.
Fig. 7 is a graph of a test confusion matrix for a network model without using the image saliency detection method at a 0dB signal-to-noise ratio.
Detailed Description
The first embodiment is as follows: referring to fig. 1 to 7, a communication signal modulation and identification method based on image saliency detection according to the present embodiment is implemented by the following steps:
setting related parameters of communication signals, constructing a communication signal data set, and carrying out noise adding processing;
step two, performing time-frequency analysis transformation on the communication signal data set to obtain a communication signal time-frequency image data set;
step three, processing the communication signal time-frequency image data set by using an image saliency detection method to obtain an image saliency communication signal time-frequency image data set, and dividing the image saliency communication signal time-frequency image data set into a training set, a verification set and a test set;
step four, building a VGG-16 network model;
training the VGG-16 network model built in the fourth step by using the training set divided in the third step, and adjusting super parameters in the model training process and carrying out preliminary evaluation on the correlation capacity of the super parameters by using the verification set divided in the third step;
step six, inputting the test set divided in the step three into the VGG-16 network model generated after training and verification, and outputting a classification test result of a related signal modulation mode, so as to evaluate the performance of a corresponding model after the image significance detection method is applied.
In the first step, the communication signal data set is generated through MATLAB simulation, wherein the setting of the relevant parameters of the communication signal data set is shown in table 1. The communication signal types to be identified comprise 8 types of 2ASK, 2FSK, 4ASK, BPSK, QPSK, 8PSK, 16QAM and MSK, each modulation mode has 600 signal samples under the 6dB signal-to-noise ratio, each modulation mode has 100 signal samples under the other signal-to-noise ratios, and each signal sample contains 2000 signal sampling values;
table 1 communication signal data set parameter table
| Signal parameters | Corresponding numerical value |
| Number of symbols N | 20 |
| Symbol period T | 0.05s |
| Sampling frequency f s | 2000Hz |
| Carrier frequency f c | 200Hz |
| Signal to noise ratio SNR | 0dB to 10dB, the step length is 1dB |
In the second step, the time-frequency analysis transformation is realized by SPWVD time-frequency transformation method, the basic idea is that the signal is decomposed into convolution form of window function and time-frequency kernel function, then the window function at each moment is Fourier transformed to obtain frequency domain representation, then the frequency domain representation is adjusted by polynomial weighting mode, finally the inverse transformation is performed to obtain time-frequency distribution diagram; the SPWVD time-frequency distribution is defined as:
wherein SPWVD s (t, ω) is the result of the time-frequency transformation of the communication signal; g (u), h (τ) are window functions; superscript x denotes complex conjugate; the SPWVD time-frequency distribution not only has higher time-frequency resolution, but also can effectively inhibit cross term interference in the superimposed signal, so that the obtained time-frequency graph result is more accurate; the SPWVD time-frequency transformation diagram of the 8 communication signals is shown in fig. 2, and finally the SPWVD time-frequency image data set of the communication signals is obtained.
The VGG-16 network structure in the fourth step is shown in figure 4, contains 16 layers of network depth and consists of a convolution layer, a pooling layer, a full connection layer and a Softmax classifier; the input is an image with 224 multiplied by 3, a very small 3 multiplied by 3 convolution kernel and a small pooling window are adopted, a plurality of convolution and pooling layers are stacked, and finally an 8-dimensional vector is output to represent the probability distribution of 8 modulation categories to which the input salient time-frequency image belongs.
The second embodiment is as follows: referring to fig. 1 to 7, a description is given of the steps of setting relevant parameters of a communication signal, constructing a communication signal data set, and performing noise adding processing in the step one of the communication signal modulation recognition method based on image saliency detection according to the present embodiment, wherein the steps include:
setting the number of code elements, carrier frequency, code element period and sampling frequency, and constructing eight communication signals of 2ASK, 2FSK, 4ASK, BPSK, QPSK, 8PSK, 16QAM and MSK according to the communication signal parameters, wherein the eight communication signals are denoted as s (n); the specific process of the noise adding processing is to add the additive white Gaussian noise v (n) with the signal-to-noise ratio interval of [0,10dB ] and the step length of 1dB into s (n), finally obtain the noise added communication signal x (n) =s (n) +v (n), and integrate all the generated x (n), thereby constructing the original communication signal data set to be used.
And a third specific embodiment: referring to fig. 1 to 7, a specific method for obtaining a communication signal time-frequency image dataset by performing time-frequency analysis transformation on the communication signal dataset in the second step of the communication signal modulation recognition method based on image significance detection according to the present embodiment is as follows: and processing the original communication signal data set by a smooth pseudo-Wiggner-Willi distributed time-frequency conversion method, and finally obtaining the SPWVD time-frequency image data set of the communication signal.
The specific embodiment IV is as follows: referring to fig. 1 to 7, in the third step of the communication signal modulation recognition method based on image saliency detection according to this embodiment, the communication signal time-frequency image dataset is processed by using the image saliency detection method, so as to obtain the image saliency communication signal time-frequency image dataset, and the specific steps of dividing the image saliency communication signal time-frequency image dataset into a training set, a verification set and a test set are as follows:
processing the communication signal time-frequency image dataset by adopting an image saliency detection algorithm based on global contrast, and calculating a certain pixel point I of any time-frequency image I k Is (I) k ) The method comprises the following steps:
in the formula (1), I i The gray value of any pixel point in the time-frequency image I is in the range of [0,255 ]]And (3) further reconstructing to obtain:
in the formula (2), f n The frequency of the nth pixel point in the time-frequency image I is presented in a histogram form; the contrast between the gray value of each pixel point in the communication signal time-frequency diagram and the surrounding area can be compared through an LC algorithm to determine the significance degree of each pixel point, so that the influence of noise in a background area with a lower pixel value in the time-frequency diagram is weakened, the characteristics of a useful signal area are more obvious, and finally an image significance communication signal time-frequency diagram data set is obtained;
for the division of the image saliency communication signal time-frequency diagram data set, the data under the 6dB signal-to-noise ratio is divided into a training set, a verification set and a test set according to a certain proportion, the training set, the verification set and the test set are mainly used for training a network to generate a modulation recognition model, and then the data under other signal-to-noise ratios are used as the test set, so that whether the generalization capability of a corresponding network model can be improved by using the image saliency method for subsequent experiment verification is padded.
The method and the device can effectively weaken the influence of noise in a background area with a lower pixel value in a time-frequency image, so that the characteristics of a useful signal area are more obvious; the corresponding time-frequency diagram of the 8 communication signals after the LC algorithm is applied is shown in figure 3, and finally the time-frequency diagram data set of the image significance communication signals is obtained.
In addition, for the division of the training set, the verification set and the test set, the time-frequency diagram of 600 image saliency communication signals corresponding to each communication signal under the 6dB signal-to-noise ratio is as follows: 1: the 1 proportion is divided into a training set, a verification set and a test set, and the time-frequency diagram of 100 image significance communication signals corresponding to each communication signal under other signal-to-noise ratios is used as the test set, so that the performance of the model is verified more comprehensively.
Fifth embodiment: referring to fig. 1 to 7, a specific step of preliminary evaluation in the fifth step of the communication signal modulation recognition method based on image saliency detection according to the present embodiment is as follows:
training the VGG-16 network model built in the fourth step by utilizing the time-frequency diagram data of the image significance communication signals in the training set, and setting corresponding super parameters of the network and training before training to obtain a corresponding modulation recognition network model; and checking the classification recognition effect of the model on the verification set in each iteration in the training process, finding the problem of the model or the parameter in time according to the effect, stopping training in time to perform corresponding adjustment, and finally primarily evaluating the generalization capability of the generated model by comparing with the classification effect of the training set, and searching the modulation recognition network model with the best training and verification effect based on image significance detection.
The training and verifying of the modulation recognition network model specifically comprises the following steps:
step five (one), pre-configuring the environment and super parameters required by the network training and verification, wherein the environment configuration situation and the super parameter setting situation are shown in tables 2 and 3:
table 2 network training and verification of environmental configuration conditions
| Software and hardware | Configuration version |
| Operating system | Windows |
| Processor and method for controlling the same | 12thGenIntel(R)Core(TM)i7-12700H |
| GPU | NVDIAGeForceRTX3070Ti |
| Cudnn | Cudnn8.2.1 |
| Cuda | Cuda11.3 |
| Python | Python3.8 |
| Deep learning frame | Pytorch1.10.0 |
| Development tool | PyCharmCommunityEdition2022.2.3x64 |
Table 3 super parameter settings for network training and verification
Fifthly (II), carrying out normalization processing on the training set divided in the step three, expanding the pixel sizes of all images into 224 multiplied by 224, and inputting the image pixel sizes into the VGG-16 network model built in the step four for training;
the specific process for training the VGG-16 network model comprises the following steps:
step 1, initializing VGG-16 network model parameters;
step 2, randomly dividing the training set into a plurality of batch_size;
step 3, randomly selecting one batch_size as X n ；
Step 4, X is taken as n Sending the data into a network, extracting features through a convolution layer and a pooling layer in the VGG-16 network, and finally obtaining a classification result through a plurality of full-connection layers and a softmax layer;
step 5, calculating an error between the prediction probability distribution and the real label distribution;
step 6, updating parameters in a network by adopting a random gradient descent algorithm (Stochastic Gradient Descent, SGD), wherein a loss function adopts a cross entropy loss function;
step 7, repeating the steps 2 to 7 until the iteration times epoch are maximum, so that the performance of the generated model is not changed any more;
checking the classification recognition effect of the network model on the verification set in each iteration in the training process, finding the problem of the model or training parameters in time according to the effect, stopping training in time, and correspondingly adjusting, wherein the super-parameter setting of the model obtained after adjustment is shown in a table 3; finally, the generalization capability of the generated model is primarily evaluated by comparing the generalized capability with the classification effect of the training set, and a modulation recognition network model with the best training and verification effect and based on image significance detection is searched;
further, in the repeated iteration process, the performance of the training model is observed through the verification set, the model with the highest recognition accuracy rate of the verification set in all the iterative training is finally output after adjustment, and the model is subjected to preliminary evaluation, so that the average classification recognition accuracy rates of the model in the training set and the verification set are shown in a table 4;
table 4 average classification accuracy of model on training set and validation set
| Data set | Average classification accuracy |
| Training set | 0.999 |
| Verification set | 0.996 |
Specific embodiment six: referring to fig. 1 to 7, a description is given of the present embodiment, in which the step of evaluating the performance of the corresponding model after the application of the image saliency detection method in the step six of the communication signal modulation recognition method based on the image saliency detection according to the present embodiment is:
and (3) inputting the time-frequency diagram data of the image saliency communication signals in the test set into the modulation recognition network model generated after training and verification in the step five for testing, judging the modulation recognition classification effect of the model in the set confidence-to-noise ratio interval according to the test result, thereby verifying the effectiveness of the used image saliency method in the aspect of noise reduction, and improving the classification recognition accuracy of the corresponding model in the low signal-to-noise ratio.
The test of the modulation recognition network model specifically comprises the following steps:
step six, firstly, testing VGG-16 network models corresponding to the unused image significance detection method by using different signal-to-noise ratio test sets divided in the step three, wherein the obtained classification recognition effect comparison chart is shown in figure 5; it can be seen that, compared with the method without using the image saliency detection method, the LC detection algorithm has better modulation recognition classification performance at low noise ratio, compared with the method without using the saliency detection method, the modulation recognition accuracy is greatly improved, and along with the improvement of the signal to noise ratio, the modulation recognition classification accuracy of the corresponding model of the LC image saliency detection method is continuously and rapidly improved, and the modulation recognition accuracy can reach more than 95% under the condition of more than 4 dB;
step six, testing a VGG-16 network model corresponding to the unused image significance detection method by utilizing the test set divided in the step three under the signal-to-noise ratio of 0dB, wherein the obtained test confusion matrix is shown in the attached figures 6 and 7; after the LC algorithm is used, the modulation recognition accuracy of other communication signals except the 2ASK signal is improved; in addition, the average classification recognition accuracy of the model corresponding to the test set with the 0dB signal-to-noise ratio by using the LC algorithm and the image saliency detection method is counted as shown in the table 5; compared with the model generated by an unused image saliency method, the modulation recognition accuracy can be improved by more than 30% after the LC algorithm is applied.
Table 5 average classification accuracy for 0dB snr test set using LC algorithm and no image saliency detection algorithm
| Pretreatment of | Means for identifying classification accuracy |
| LC | 85.875％ |
| No image saliency detection algorithm is applied | 50.375％ |
All the test results prove that the LC algorithm has effectiveness and feasibility in the aspects of denoising the time-frequency diagram and displaying the characteristics of the useful signals, and compared with the traditional communication signal modulation identification preprocessing denoising method, the method does not need prior information of the signals, does not need a large amount of calculation cost, does not influence the useful characteristics of the signals, only has more obvious characteristics and has stronger generalization capability.
The present invention is not limited to the preferred embodiments, but is capable of modification and variation in detail, and other embodiments, such as those described above, of making various modifications and equivalents will fall within the spirit and scope of the present invention.
Claims (6)
1. A communication signal modulation identification method based on image significance detection is characterized in that: the communication signal modulation identification method based on image significance detection is realized through the following steps:
setting related parameters of communication signals, constructing a communication signal data set, and carrying out noise adding processing;
step two, performing time-frequency analysis transformation on the communication signal data set to obtain a communication signal time-frequency image data set;
step three, processing the communication signal time-frequency image data set by using an image saliency detection method to obtain an image saliency communication signal time-frequency image data set, and dividing the image saliency communication signal time-frequency image data set into a training set, a verification set and a test set;
step four, building a VGG-16 network model;
training the VGG-16 network model built in the fourth step by using the training set divided in the third step, and adjusting super parameters in the model training process and carrying out preliminary evaluation on the correlation capacity of the super parameters by using the verification set divided in the third step;
step six, inputting the test set divided in the step three into the VGG-16 network model generated after training and verification, and outputting a classification test result of a related signal modulation mode, so as to evaluate the performance of a corresponding model after the image significance detection method is applied.
2. The communication signal modulation recognition method based on image saliency detection according to claim 1, wherein: setting related parameters of communication signals, constructing a communication signal data set, and carrying out noise adding processing, wherein the steps are as follows:
setting the number of code elements, carrier frequency, code element period and sampling frequency, and constructing eight communication signals of 2ASK, 2FSK, 4ASK, BPSK, QPSK, 8PSK, 16QAM and MSK according to the communication signal parameters, wherein the eight communication signals are denoted as s (n); the specific process of the noise adding processing is to add the additive white Gaussian noise v (n) with the signal-to-noise ratio interval of [0,10dB ] and the step length of 1dB into s (n), finally obtain the noise added communication signal x (n) =s (n) +v (n), and integrate all the generated x (n), thereby constructing the original communication signal data set to be used.
3. The communication signal modulation recognition method based on image saliency detection according to claim 1, wherein: step two, carrying out time-frequency analysis transformation on the communication signal data set, and obtaining the communication signal time-frequency image data set by the specific method comprising the following steps: and processing the original communication signal data set by a smooth pseudo-Wiggner-Willi distributed time-frequency conversion method, and finally obtaining the SPWVD time-frequency image data set of the communication signal.
4. The communication signal modulation recognition method based on image saliency detection according to claim 1, wherein: processing the communication signal time-frequency image data set by using an image saliency detection method to obtain an image saliency communication signal time-frequency image data set, and dividing the image saliency communication signal time-frequency image data set into a training set, a verification set and a test set, wherein the method comprises the following specific steps of:
processing the communication signal time-frequency image dataset by adopting an image saliency detection algorithm based on global contrast, and calculating a certain pixel point I of any time-frequency image I k Is (I) k ) The method comprises the following steps:
in the formula (1), I i The gray value of any pixel point in the time-frequency image I is in the range of [0,255 ]]And (3) further reconstructing to obtain:
in the formula (2), f n The frequency of the nth pixel point in the time-frequency image I is presented in a histogram form; the contrast between the gray value of each pixel point in the communication signal time-frequency diagram and the surrounding area can be compared through an LC algorithm to determine the significance degree of each pixel point, so that the influence of noise in a background area with a lower pixel value in the time-frequency diagram is weakened, the characteristics of a useful signal area are more obvious, and finally an image significance communication signal time-frequency diagram data set is obtained;
for the division of the image saliency communication signal time-frequency diagram data set, the data under the 6dB signal-to-noise ratio is divided into a training set, a verification set and a test set according to a certain proportion, the training set, the verification set and the test set are mainly used for training a network to generate a modulation recognition model, and then the data under other signal-to-noise ratios are used as the test set, so that whether the generalization capability of a corresponding network model can be improved by using the image saliency method for subsequent experiment verification is padded.
5. The communication signal modulation recognition method based on image saliency detection according to claim 1, wherein: the specific steps of the preliminary evaluation in the fifth step are as follows:
training the VGG-16 network model built in the fourth step by utilizing the time-frequency diagram data of the image significance communication signals in the training set, and setting corresponding super parameters of the network and training before training to obtain a corresponding modulation recognition network model; and checking the classification recognition effect of the model on the verification set in each iteration in the training process, finding the problem of the model or the parameter in time according to the effect, stopping training in time to perform corresponding adjustment, and finally primarily evaluating the generalization capability of the generated model by comparing with the classification effect of the training set, and searching the modulation recognition network model with the best training and verification effect based on image significance detection.
6. The communication signal modulation recognition method based on image saliency detection according to claim 1, wherein: in the sixth step, the step of evaluating the performance of the corresponding model after the image saliency detection method is applied is as follows:
and (3) inputting the time-frequency diagram data of the image saliency communication signals in the test set into the modulation recognition network model generated after training and verification in the step five for testing, judging the modulation recognition classification effect of the model in the set confidence-to-noise ratio interval according to the test result, thereby verifying the effectiveness of the used image saliency method in the aspect of noise reduction, and improving the classification recognition accuracy of the corresponding model in the low signal-to-noise ratio.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310710751.1A CN116684233A (en) | 2023-06-15 | 2023-06-15 | Communication signal modulation identification method based on image significance detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310710751.1A CN116684233A (en) | 2023-06-15 | 2023-06-15 | Communication signal modulation identification method based on image significance detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116684233A true CN116684233A (en) | 2023-09-01 |
Family
ID=87788814
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310710751.1A Pending CN116684233A (en) | 2023-06-15 | 2023-06-15 | Communication signal modulation identification method based on image significance detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116684233A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018176889A1 (en) * | 2017-03-27 | 2018-10-04 | 华南理工大学 | Method for automatically identifying modulation mode for digital communication signal |
| CN108805039A (en) * | 2018-04-17 | 2018-11-13 | 哈尔滨工程大学 | The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features |
| CN110163282A (en) * | 2019-05-22 | 2019-08-23 | 西安电子科技大学 | Modulation Mode Recognition method based on deep learning |
| CN114943245A (en) * | 2022-04-07 | 2022-08-26 | 中国人民解放军海军航空大学航空作战勤务学院 | Automatic modulation recognition method and device based on data enhancement and feature embedding |
-
2023
- 2023-06-15 CN CN202310710751.1A patent/CN116684233A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018176889A1 (en) * | 2017-03-27 | 2018-10-04 | 华南理工大学 | Method for automatically identifying modulation mode for digital communication signal |
| CN108805039A (en) * | 2018-04-17 | 2018-11-13 | 哈尔滨工程大学 | The Modulation Identification method of combination entropy and pre-training CNN extraction time-frequency image features |
| CN110163282A (en) * | 2019-05-22 | 2019-08-23 | 西安电子科技大学 | Modulation Mode Recognition method based on deep learning |
| CN114943245A (en) * | 2022-04-07 | 2022-08-26 | 中国人民解放军海军航空大学航空作战勤务学院 | Automatic modulation recognition method and device based on data enhancement and feature embedding |
Non-Patent Citations (1)
| Title |
|---|
| 陈晋音;成凯回;郑海斌;: "低信噪比下基于深度学习的调制模式识别方法", 计算机科学, no. 1, 15 June 2020 (2020-06-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109035149B (en) | License plate image motion blur removing method based on deep learning | |
| CN109597043B (en) | Radar signal identification method based on quantum particle swarm convolutional neural network | |
| CN112308008B (en) | Radar radiation source individual identification method based on working mode open set of transfer learning | |
| CN113094993B (en) | Modulation signal denoising method based on self-coding neural network | |
| CN111126471A (en) | Microseism event detection method and system | |
| CN113014524B (en) | Digital signal modulation identification method based on deep learning | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| CN111680737B (en) | Radar radiation source individual identification method under differential signal-to-noise ratio condition | |
| CN110929842B (en) | Accurate intelligent detection method for non-cooperative radio signal burst time region | |
| CN112115821B (en) | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy | |
| CN114896887B (en) | Frequency-using equipment radio frequency fingerprint identification method based on deep learning | |
| CN113792685B (en) | Microseism event detection method based on multi-scale convolutional neural network | |
| CN114285545B (en) | Side channel attack method and system based on convolutional neural network | |
| CN114980122A (en) | Small sample radio frequency fingerprint intelligent identification system and method | |
| CN111612130A (en) | Frequency shift keying communication signal modulation mode identification method | |
| CN114528876A (en) | Small sample signal automatic modulation identification method based on improved prototype network | |
| CN113076925B (en) | M-QAM signal modulation mode identification method based on CNN and ELM | |
| CN116684233A (en) | Communication signal modulation identification method based on image significance detection | |
| CN116055026A (en) | ARMA model-based particle filter side channel attack noise reduction preprocessing method | |
| CN115809426A (en) | Radiation source individual identification method and system | |
| CN113409213B (en) | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump | |
| CN112069987B (en) | Interference type automatic identification method based on statistical manifold optimization dimension reduction | |
| CN114626412A (en) | Multi-class target identification method and system for unattended sensor system | |
| CN113014361A (en) | BPSK signal confidence test method based on graph | |
| He et al. | Modulation recognition of communication signals 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 |