A communication signal identification method based on software radio and capsule neural network

Abstract

The application discloses a communication signal identification method based on a software radio and a capsule neural network, and comprises the following steps: generating communication signal data by using a software radio platform; performing channel simulation on the communication signal data by using the software radio platform, and obtaining a simulation channel data set and a real channel data set through sending and receiving; pre-processing and feature extraction are performed on the channel data, a signal category is obtained according to a judgment standard of the extracted features, and the signal category of the corresponding channel data is labeled; a training set and a test set are constructed according to the labeled simulation channel data set and the real channel data set; the capsule neural network is trained by using the training set and the test set, and a trained capsule neural network is obtained; and a target communication signal is input into the trained capsule neural network, and an identification result of the target communication signal is obtained. The open-source software radio platform provides a data set supply, and the multi-dimensional and rich features extracted based on the capsule neural network improve the identification precision.

Description

TECHNICAL FIELD

[0001] The application relates to a communication signal recognition method based on a software radio and a capsule neural network and belongs to the technical field of communication signal recognition. BACKGROUND

[0002] In the modern wireless communication background, communication resources are increasingly scarce, channel environments are increasingly complex, and signal modulation modes in a limited frequency range are increasingly diversified. Traditional communication system modulation modes are generally simple, and a receiver only needs to receive in a known modulation mode without needing to perform modulation mode recognition. Now, a radio cannot be processed by a traditional receiving method, so how to acquire and recognize a wireless communication signal modulation mode in a complex wireless communication environment is a current research focus.

[0003] Common communication signal recognition methods include a Bayesian classifier, a decision tree classification algorithm and a convolutional neural network algorithm. The Bayesian classifier has no problem in theory, but in practice, the exact distribution probability of data cannot be known, which cannot be completed. Moreover, the calculation cost of the Bayesian optimal hypothesis is very large. The decision tree classification algorithm has a simple principle and is easy to generate, but when the number of data categories is too large, more errors may be generated, and the decision tree cannot be processed when attribute values of test data are empty. This is a top-down greedy algorithm, which may be a local optimal solution and is not necessarily suitable for overall optimal consideration. The convolutional neural network algorithm is a kind of algorithm that has been studied more in recent years, but it needs a large amount of data preparation, has a long training time and has insufficient neural structure knowledge and experience. When the data amount is large, the convolutional neural network has problems such as low recognition accuracy and insufficient generalization ability of the neural network. SUMMARY

[0004] Purpose: In order to overcome the deficiencies in the prior art, the application provides a communication signal recognition method based on a software radio and a capsule neural network, which can not only generate a large and needed data set through a software system platform and general hardware peripherals of the software radio, but also can implement multi-branch work through a capsule neural network with a topological structure, so as to improve the neural network structure performance.

[0005] The capsule neural network provided in the application can achieve better accuracy than other neural networks, and the application also proposes to use real channel data sets and simulation channel data sets to optimize the capsule network, so as to improve the generalization ability of the neural network.

[0006] Technical scheme: To solve the above technical problems, the technical scheme adopted by the application is:

[0007] In a first aspect, the present application provides a communication signal identification method, comprising:

[0008] S1. generating communication signal data by using a software radio platform;

[0009] S2. performing channel simulation on the communication signal data by using the software radio platform to obtain a simulation channel data set and a real channel data set;

[0010] S3. pre-processing and feature extraction are performed on the channel data in the simulation channel data set and the real channel data set, a signal category is obtained according to a judgment standard of the extracted features, and the signal category of the corresponding channel data is labeled;

[0011] S4. constructing a training set and a test set according to the labeled simulation channel data set and real channel data set;

[0012] S5. training a capsule neural network by using the training set and the test set to obtain a trained capsule neural network;

[0013] S6. inputting a target communication signal into the trained capsule neural network to obtain an identification result of the target communication signal.

[0014] In some embodiments, S1. generating communication signal data by using a software radio platform, comprises:

[0015] S11. setting a signal category and related parameters of a communication signal in the software radio platform, wherein the signal category includes two categories of digital modulation signals and analog modulation signals, and the related parameters include a sampling rate, a symbol rate, and corresponding white noise;

[0016] S12. generating information signals for digital modulation and information signals for analog modulation by using a signal source module of the software radio platform;

[0017] S13. modulating the information signals generated in S12 by using a corresponding modulation module to obtain modulation signals;

[0018] S14. according to the set signal category and related parameters of the communication signal, using an Add module to superimpose the modulation signals and noise signals to obtain communication signal data.

[0019] In some embodiments, S2. performing channel simulation on the communication signal data by using the software radio platform to obtain a simulation channel data set and a real channel data set, comprises:

[0020] S21. dividing the communication signal data generated in S1 into a first part of communication signal data and a second part of communication signal data according to a preset ratio;

[0021] S22, generating a simulation channel data set by a channel module simulated by a software radio platform for the first part of the communication signal data; wherein the channel module simulated by the software radio platform comprises a sampling rate offset module, a center frequency offset module, and a selective fading module;

[0022] S23, transmitting and receiving the second part of the communication signal by a UHD module of the software radio platform and a universal software radio peripheral (USRP) to obtain a real channel data set.

[0023] Further, the preset ratio is 8:2.

[0024] In some embodiments, the channel data in the simulation channel data set and the real channel data set are preprocessed and feature extracted, including:

[0025] S31: performing in-band filtering processing on the channel data;

[0026] S32: performing center frequency and code rate estimation on the channel data after the in-band filtering processing;

[0027] S33: separating the communication signal according to the estimated center frequency and code rate to obtain m groups of modulation signals of different types and orders;

[0028] S34: performing feature extraction on the m groups of modulation signals obtained in S33 respectively.

[0029] In some embodiments, the extracted features include instantaneous phase, instantaneous frequency, high-order cumulant, absolute phase standard deviation, direct phase standard deviation, and absolute frequency standard deviation.

[0030] In some embodiments, the signal categories include digital modulation signals and analog modulation signals.

[0031] In some embodiments, S4. constructing a training set and a test set according to the labeled simulation channel data set and the real channel data set, comprising:

[0032] 85% of the simulation channel data set is used as the training set, and the remaining 15% of the simulation channel data set and all the real channel data set are used as the test set.

[0033] In some embodiments, the processing process of the capsule neural network comprises:

[0034] The capsule neural network comprises, in sequence, a convolution layer, m branches, a fully connected layer, and a classification layer.

[0035] The communication signal is processed into two-dimensional data input into the convolution layer to obtain a feature map, the convolution layer has a size of 64x64, uses a convolution kernel with a size of 6x6, a depth of 10, and a step of 1, and simultaneously uses a Relu activation function.

[0036] Each branch includes an initial capsule layer and a digital capsule layer, in the initial capsule layer, a convolution kernel with a size of 3*3, a depth of 20 and a step of 1 is used to flatten the feature map obtained by the convolution layer to obtain 16 primary capsules of 20-dimensional vectors; the number of output capsules in the digital capsule layer is 1, and the dimension is 10; the capsule vector of the initial capsule layer is obtained by a dynamic routing algorithm and a nonlinear activation function;

[0037] The full connection layer integrates the outputs of the m branches into an m-dimensional vector n, and each element in n represents the possibility of the corresponding modulation type appearing in the input data;

[0038] The m-dimensional vector n is input into the classification layer to obtain the classification result.

[0039] In some embodiments, the capsule neural network is trained using the training set and the test set to obtain a trained capsule neural network, including:

[0040] The one-dimensional communication signal in the training set is converted into a two-dimensional signal and input into the capsule neural network for training;

[0041] The test set is input into the trained capsule neural network, and the results are compared to improve the network, specifically including:

[0042] The simulation channel data set and the real channel data set in the test set are respectively input into the trained capsule neural network for testing, and the difference between the precision values and the corresponding loss function values of the simulation channel test set and the real channel test set after passing through the network is compared to optimize the feature extraction algorithm of the convolution layer and the capsule layer feature classification algorithm;

[0043] Or the simulation channel data set and the real channel data set are trained and tested together to improve the neural network and improve the generalization ability of the neural network.

[0044] In a second aspect, the present application provides a communication signal recognition device, including a processor and a storage medium;

[0045] The storage medium is used to store instructions;

[0046] The processor is used to operate according to the instructions to execute the method according to the first aspect.

[0047] In a third aspect, the present application provides a device, including,

[0048] A memory;

[0049] A processor;

[0050] And

[0051] Computer program

[0052] The computer program is stored in the memory and is configured to be executed by the processor to implement the method of the first aspect.

[0053] In a fourth aspect, the application provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method of the first aspect.

[0054] Beneficial effects: The communication signal identification method based on software radio and capsule neural network provided by the application has the following advantages: the communication signal identification method based on software radio and capsule neural network can not only generate a large and needed data set through a software system platform and a general hardware peripheral of software radio, but also can implement multi-branch work through a capsule neural network with a topological structure, so as to improve the performance of the neural network structure.

[0055] The capsule neural network provided by the application can achieve better accuracy than other neural networks, and the use of real channel data set and simulation channel data set for optimizing the capsule network can improve the generalization ability of the neural network. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 It is a method flowchart according to an embodiment of the application.

[0057] Figure 2 It is a GNU Radio simulation channel model diagram according to an embodiment of the application.

[0058] Figure 3 It is a capsule neural network flowchart according to an embodiment of the application.

[0059] Figure 4 It is a schematic diagram of a dynamic routing algorithm according to an embodiment of the application. DETAILED DESCRIPTION

[0060] The application will be further described below in conjunction with the drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0061] In the description of the application, if several meanings are included, the meaning of multiple is two or more, greater than, less than, more than, etc. are understood as not including the number, above, below, etc. are understood as including the number. If it is described as first, second, it is only used to distinguish the technical features for the purpose, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the sequence of indicated technical features.

[0062] In the description of the application, the description of the terms "one embodiment", "some embodiments", "exemplary embodiment", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the application. In the description, the exemplary description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0063] Embodiment 1

[0064] In a first aspect, the embodiment provides a communication signal identification method, comprising:

[0065] S1. Generating communication signal data by using a software defined radio platform;

[0066] S2. The communication signal data is subjected to channel simulation, transmission and reception by the software defined radio platform to obtain a simulation channel data set and a real channel data set;

[0067] S3. Preprocessing and feature extraction are performed on the channel data in the simulation channel data set and the real channel data set, a signal category is obtained according to the judgment standard of the extracted features, and the signal category of the corresponding channel data is labeled;

[0068] S4. A training set and a test set are constructed according to the labeled simulation channel data set and real channel data set;

[0069] S5. The training set and the test set are used to train a capsule neural network to obtain a trained capsule neural network;

[0070] S6. The target communication signal is input into the trained capsule neural network to obtain the identification result of the target communication signal.

[0071] In some embodiments, a communication signal identification method based on software defined radio (SDR) and capsule neural network comprises the following steps: Figure 1 as shown in the figure, specifically comprising the following steps:

[0072] S1. Generating communication signal data by using a software defined radio platform;

[0073] S11. Setting the signal category and related parameters of the communication signal in the software defined radio platform, wherein the signal category includes two categories of digital modulation signal and analog modulation signal, and the related parameters include sampling rate, symbol rate and corresponding white noise;

[0074] S12, generating information signals for digital modulation and information signals for analog modulation by using signal source modules of the software radio platform;

[0075] S13, modulating the information signals generated by S12 by using corresponding modulation modules to obtain modulation signals;

[0076] S14, superimposing the modulation signals and noise signals by using Add modules to obtain communication signal data according to the signal category and related parameters of the set communication signal.

[0077] The setting of the related parameters is realized by the corresponding variable of the GNU Radio platform and the QT GUI Range module; the information signal for digital modulation is generated by using the signal source module Random Source and the analog information signal for analog modulation is generated by using the Signal Source module, which represents the information transmitted by the modulation signal; according to the set modulation category, the information signal to be transmitted is modulated by using corresponding modulation modules such as PSK Mod, GFSK Mod and GMSK Mod to obtain signal data; according to the requirements of the required signal, the required single signal and noise signal are superimposed by using the Add module to obtain the superimposed signal.

[0078] S2, the communication signal data described in S1 is subjected to channel simulation and transmission and reception by the software radio platform to obtain a simulation channel data set and a real channel data set;

[0079] S21, the communication signal data generated by S1 is divided into a first part of communication signal data and a second part of communication signal data according to a preset proportion;

[0080] S22, the first part of communication signal data is generated by the channel module simulated by the software radio platform; wherein the channel module simulated by the software radio platform includes a sampling rate offset module, a center frequency offset module and a selective fading module;

[0081] S23, the second part of communication signal data is transmitted and received by the UHD module of the software radio platform and the universal software radio peripheral USRP to obtain a real channel data set.

[0082] In some embodiments, the communication signal data generated according to S1 is divided into two parts in a ratio of 8:2; 80% of the communication signal data is transmitted through channel simulation modules such as a sampling rate offset module (SRO module), a center frequency offset module (CFO module), and a selective fading module simulated by a GNU Radio platform to generate a simulation channel data set after the simulation channel; the remaining 20% of the communication signal data is transmitted through a UHD module of the GNU Radio platform and a universal software radio peripheral (USRP B210) to obtain a real channel data set; and the channel data sets obtained after completing the channel interference are processed in the next step.

[0083] S3: pre-processing and feature extraction are performed on the channel data in the simulation channel data set and the real channel data set, a signal category is obtained according to a judgment standard of the extracted features, and the signal category of the corresponding channel data is labeled;

[0084] S31: in-band filtering is performed on the channel data;

[0085] S32: center frequency and code rate estimation is performed on the channel data after in-band filtering;

[0086] S33: the communication signal is separated into m groups of modulation signals of different types and different orders according to the estimated center frequency and code rate;

[0087] S34: feature extraction is performed on the m groups of modulation signals obtained in S33.

[0088] The extracted features include instantaneous phase, instantaneous frequency, high-order cumulant, absolute phase standard deviation, direct phase standard deviation, and absolute frequency standard deviation; and the signal categories include digital modulation signals and analog modulation signals.

[0089] The relevant modulation signal categories are determined according to the judgment standard of the extracted features, and the corresponding data is labeled.

[0090] S4: a training set and a test set are constructed according to the labeled simulation channel data set and the real channel data set, including:

[0091] 85% of the simulation channel data set is used as the training set, and the remaining 15% of the simulation channel data set and all the real channel data set are used as the test set.

[0092] In some embodiments, the training set and the test set are used to train a capsule neural network to obtain a trained capsule neural network; including:

[0093] S5: initialize the capsule neural network parameters, convert the one-dimensional communication signal in the training set into a two-dimensional signal input into the capsule neural network and train it;

[0094] S6: input the simulation channel data set and the real channel data set in the test set into the trained capsule neural network for testing, optimize the feature extraction algorithm of the convolution layer and the feature classification algorithm of the capsule layer by comparing the difference between the accuracy values and the corresponding loss function values of the simulation channel test set and the real channel test set after passing through the network;

[0095] Or train and test the simulation channel data set and the real channel data set together, improve the neural network, and improve the generalization ability of the neural network.

[0096] S7: input the target communication signal into the trained capsule neural network to obtain the recognition result of the target communication signal.

[0097] In some embodiments, as shown in Figure 2 In step S2, the simulation channel through which the communication signal data passes is composed of a GNURadio dynamic channel module.

[0098] Among them, SRO is a sampling rate offset model; CFO is a center frequency offset model; AWGN is a noise model, mainly to realize the additive white Gaussian noise. 80% of the signal set is randomly divided into several parts, and sequentially passes through the sampling rate offset model, the center frequency offset model, the selective fading model and the noise model, and the related parameters in the model such as the sampling rate and the maximum jump value can be controlled through the QT GUI Range module, so as to change to the value we need.

[0099] As shown in Figure 3 The capsule neural network is a capsule network with a topological structure. The network first is a convolution layer, in which the modulated data signal is processed into a two-dimensional data as input, with a size of 64x64, a convolution kernel size of 6x6, a depth of 10, and a step of 1, and a Relu activation function is used at the same time. Secondly, m branches are generated, each of which contains an initial capsule layer and a digital capsule layer. In the initial capsule layer, a convolution kernel with a size of 3x3, a depth of 20, and a step of 1 is used to flatten the obtained feature map to obtain 16 primary capsules of 20-dimensional vectors. The output capsule number in the digital capsule layer is 1, and the dimension is 10, which is obtained from the capsule vector of the initial capsule layer through a dynamic routing algorithm and a nonlinear activation function. Thirdly, a full connection layer is used to integrate the outputs of the m branches into an m-dimensional vector n, and each element in n represents the possibility of the corresponding modulation type appearing in the input data. Finally, the m-dimensional vector is input into the classification layer to obtain the classification result.

[0100] AsFigure 4 As shown, the dynamic routing algorithm process specifically includes the following steps:

[0101] S51, each primary capsule is multiplied by an independent weight matrix to predict each digital capsule, and the calculation formula is as follows:

[0102]

[0103] Where u i is the i-th primary capsule, w ij is the weight matrix, represents the predicted vector when the input is u i ;

[0104] S52, multiply each predicted vector and the corresponding coupling coefficient h ij , and then sum to obtain the input of the digital capsule layer, and the calculation formula is as follows:

[0105]

[0106] Where the coupling coefficient h ij satisfies is obtained by a normalized exponential function (softmax function), as shown in the following formula:

[0107]

[0108] Where b ij is the logarithmic representation of the prior probability, and the initial value is 0.

[0109] S53, the input q j of the digital capsule layer is obtained through a nonlinear activation function Squashing to obtain the output of the digital capsule layer, and the calculation formula is as follows:

[0110]

[0111] Where f s is a nonlinear function, which ensures that the vector length is compressed to between 0 and 1 under the condition that the direction of the vector does not change;

[0112] S54, the consistency of the output vector and the predicted vector is measured, and the coupling coefficient is iteratively updated with the aid of the measurement consistency, and the calculation formula is shown in formula (5);

[0113]

[0114] Then return to S52 for the next iteration, and after k iterations, the final output of the digital capsule network is obtained.

[0115] Example 2

[0116] In a second aspect, based on the embodiment 1, the present embodiment provides a communication signal identification apparatus, comprising a processor and a storage medium;

[0117] The storage medium is configured to store instructions.

[0118] The processor is configured to operate according to the instructions to perform the method according to the embodiment 1.

[0119] Embodiment 3

[0120] In a third aspect, based on the embodiment 1, the present embodiment provides an apparatus, comprising,

[0121] a memory;

[0122] a processor;

[0123] and

[0124] a computer program;

[0125] The computer program is stored in the memory and configured to be executed by the processor to implement the method according to the embodiment 1.

[0126] Embodiment 4

[0127] In a fourth aspect, based on the embodiment 1, the present embodiment provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method according to the embodiment 1.

[0128] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer-usable program code.

[0129] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the computer or other programmable data processing apparatus produce a device that implements the functions described in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1means for performing the function specified by the block or blocks.

[0130] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 flow or flows and / or blocks Figure 1 means for performing the function specified by the block or blocks.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 flow or flows and / or blocks Figure 1 steps of means for performing the function specified by the block or blocks.

[0132] The above description is only preferred embodiments of the present application, it should be pointed out that for those skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for identifying communication signals, characterized in that, The method includes: S1. Generate communication signal data using a software-defined radio platform, including: S11, setting the signal category and related parameters of the communication signal in the software-defined radio platform, wherein the signal category includes two main categories: digital modulation signal and analog modulation signal, and the related parameters include sampling rate, symbol rate, and corresponding white noise; S12, generating information signals for digital modulation and information signals for analog modulation using the signal source module of the software-defined radio platform; S13, modulating the information signals generated in S12 using the corresponding modulation module to obtain a modulated signal; S14, according to the set signal category and related parameters of the communication signal, using the Add module to superimpose the modulated signal and the noise signal to obtain the communication signal data; S2. The communication signal data is simulated and transmitted / received through a software radio platform to obtain a simulated channel dataset and a real channel dataset, including: S21, dividing the communication signal data generated in S1 into a first part of communication signal data and a second part of communication signal data according to a preset ratio; S22, generating a simulated channel dataset by passing the first part of communication signal data through a channel module simulated by the software radio platform; wherein the channel module simulated by the software radio platform includes a sampling rate offset module, a center frequency offset module, and a selective fading module; S23, transmitting and receiving the second part of communication signal data through the UHD module and the general software radio peripheral USRP of the software radio platform to obtain a real channel dataset; S3. Preprocess and extract features from the channel data in the simulated channel dataset and the real channel dataset, obtain the signal category based on the judgment criteria of the extracted features, and label the signal category of the corresponding channel data; S4. Construct training and testing sets based on the labeled simulated channel dataset and real channel dataset; S5. Train the capsule neural network using the training set and test set to obtain a trained capsule neural network; S6. Input the target communication signal into the trained capsule neural network to obtain the recognition result of the target communication signal; the processing of the capsule neural network includes: the capsule neural network sequentially includes: a convolutional layer, m branches, a fully connected layer, and a classification layer; the communication signal is processed into two-dimensional data and input into the convolutional layer to obtain a feature map, the convolutional layer size is 64×64, the convolutional kernel size is 6×6, the depth is 10, the stride is 1, and the ReLU activation function is used; each branch includes an initial capsule layer and a digital capsule layer, in the initial capsule layer, A 3×3 convolutional kernel with a depth of 20 and a stride of 1 is used to flatten the feature maps obtained from the convolutional layers, resulting in 16 primary capsules of 20-dimensional vectors. The output capsules in the digital capsule layer have one capsule and a dimension of 10. The capsule vectors of the initial capsule layer are obtained through a dynamic routing algorithm and a non-linear activation function. The fully connected layer integrates the outputs of m branches into an m-dimensional vector n, where the value of each element in n represents the probability that its corresponding modulation type appears in the input data. The m-dimensional vector n is input into the classification layer to obtain the classification result.

2. The communication signal identification method according to claim 1, characterized in that, The preset ratio is 8:

2.

3. The communication signal identification method according to claim 1, characterized in that, Preprocessing and feature extraction are performed on the channel data in the simulated channel dataset and the real channel dataset, including: S31: Perform in-band filtering on the channel data; S32: Estimate the center frequency and code rate of the channel data after in-band filtering; S33: Based on the estimated center frequency and code rate, the communication signal is separated to obtain m groups of modulation signals of different types and orders; S34: Perform feature extraction on the m modulated signals obtained in S33.

4. The communication signal identification method according to claim 1, characterized in that, The extracted features include instantaneous phase, instantaneous frequency, higher-order cumulants, absolute phase standard deviation, direct phase standard deviation, and absolute frequency standard deviation.

5. The communication signal identification method according to claim 1, characterized in that, The signal categories include digitally modulated signals and analog modulated signals.

6. The communication signal identification method according to claim 1, characterized in that, S4. Construct training and testing sets based on the labeled simulated channel dataset and real channel dataset, including: 85% of the simulated channel dataset was used as the training set, and the remaining 15% of the simulated channel dataset and all real channel datasets were used as the test set.

7. The communication signal identification method according to claim 1, characterized in that, The capsule neural network is trained using the training set and the test set to obtain a trained capsule neural network, including: The one-dimensional communication signals in the training set are converted into two-dimensional signals and input into the capsule neural network for training. The test set is input into the trained capsule neural network, and the network is improved by comparing the results and errors. Specifically, this includes: The simulated channel dataset and the real channel dataset in the test set are respectively input into the trained capsule neural network for testing. By comparing the difference between the accuracy value and the corresponding loss function value after the simulated channel test set and the real channel test set pass through the network, the feature extraction algorithm of the convolutional layer and the feature classification algorithm of the capsule layer are optimized. Alternatively, the simulated channel dataset can be trained and tested together with the real channel dataset to improve the neural network and enhance its generalization ability.

8. A storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the method described in any one of claims 1 to 7.

Images