A method and apparatus for classifying radio signals

Abstract

This invention discloses a method and apparatus for radio signal classification. The method includes: acquiring a raw dataset; training a signal classification model using the raw dataset to obtain an optimized signal classification model; acquiring radio signal data; and processing the radio signal data using the optimized signal classification model to obtain radio signal modulation information. This invention provides a radio signal classification method and apparatus that effectively identifies complex modulation signals using a signal classification model, improving the modulation recognition accuracy of high-order modulation signals. This is of great significance for ensuring the reliability and stability of communication systems.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method and apparatus for classifying radio signals. Background Technology

[0002] In modern communication systems, high-order modulation techniques have been widely used due to their ability to significantly improve spectral efficiency. However, traditional modulation recognition methods face challenges of high computational complexity and limited recognition efficiency. Existing deep learning models often struggle to achieve ideal discriminative power when dealing with the classification of high-order modulated signals. Therefore, achieving the classification and recognition of high-order modulated signals is an urgent problem to be solved. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a radio signal classification method and apparatus, which uses a signal classification model to effectively identify complex modulation signals and improves the modulation recognition accuracy of high-order modulation signals.

[0004] To address the aforementioned technical problems, a first aspect of the present invention discloses a radio signal classification method, the method comprising:

[0005] S1, Obtain the original dataset;

[0006] S2, Using the original dataset, the signal classification model is trained to obtain an optimized signal classification model;

[0007] S3, acquire radio signal data, process the radio signal data using the signal classification optimization model, and obtain radio signal modulation information.

[0008] As an optional implementation, in the first aspect of the present invention, the signal classification model includes: a data input layer, a data convolution processing layer, a batch normalization layer, a first function processing layer, a random deactivation layer, a first long short-term memory layer, a second function processing layer, a second long short-term memory layer, a third function processing layer, a third long short-term memory layer, a fourth function processing layer, a first fully connected layer, a fifth function processing layer, a second fully connected layer, and a data output layer;

[0009] The data input layer, the data convolution processing layer, the batch normalization layer, the first function processing layer, the random deactivation layer, the first long short-term memory layer, the second function processing layer, the second long short-term memory layer, the third function processing layer, the third long short-term memory layer, the fourth function processing layer, the first fully connected layer, the fifth function processing layer, the second fully connected layer, and the data output layer are sequentially connected.

[0010] As an optional implementation, in the first aspect of the present invention, the step of training the signal classification model using the original dataset to obtain an optimized signal classification model includes:

[0011] S21, preprocess the original dataset to obtain a preprocessed original dataset;

[0012] S22, using the preprocessed original dataset, the signal classification model is trained to obtain an optimized signal classification model.

[0013] As an optional implementation, in the first aspect of the present invention, the step of training the signal classification model using the preprocessed original dataset to obtain an optimized signal classification model includes:

[0014] S221, the preprocessed original dataset is divided according to a set ratio to obtain a training dataset and a test dataset;

[0015] Preset training stopping threshold;

[0016] S222, The training dataset is processed using the signal classification model and the multi-class cross-entropy loss function to obtain the first difference value;

[0017] S223, using the adaptive estimation function and weight decay coefficient, the parameters of the signal classification model are updated to obtain the stopping parameters;

[0018] S224, determine whether the stopping parameter is greater than the preset training stopping threshold, and obtain the stopping parameter determination result;

[0019] If the result of the stop parameter determination is yes, execute S225;

[0020] If the result of the stop parameter determination is negative, execute S222;

[0021] S225, determine whether the first difference value is less than a preset difference value threshold, and obtain the difference value discrimination result;

[0022] If the difference value determination result is negative, execute S221;

[0023] If the difference value discrimination result is yes, the training process of the signal classification model is completed, and the signal classification optimization model is obtained.

[0024] As an optional implementation, in the first aspect of the present invention, the expression for the multi-class cross-entropy loss function is:

[0025]

[0026] Among them, Q i This represents the difference between the probability distribution of a sample belonging to class i and the true probability distribution; i represents the index of the modulation type; k represents the index of the sample data; N represents the number of modulation categories; and K represents the number of sample data in class i. This represents the true label value of the k-th sample data of the i-th modulation category; This represents the predicted probability value that the k-th sample data belongs to the i-th modulation category;

[0027] The expression for the adaptive estimation function is:

[0028]

[0029] Among them, w i+1 w represents the gradient update value at time i+1; i Represents the gradient update value at time i; μ represents the learning rate; Let represent the second moment estimate at time i; β represents the second moment estimate after bias correction at time i; i τ represents the attenuation coefficient at time i; τ represents the first reference constant.

[0030] As an optional implementation, in the first aspect of the present invention, the step of acquiring radio signal data and processing the radio signal data using the signal classification optimization model to obtain radio signal modulation information includes:

[0031] S31, receives and acquires radio signal data in real time;

[0032] S32, preprocess the radio signal data to obtain preprocessed signal data;

[0033] S33, Based on the signal classification optimization model, the preprocessed signal data is processed to obtain radio signal modulation information.

[0034] As an optional implementation, in the first aspect of the present invention, the step of processing the preprocessed signal data based on the signal classification optimization model to obtain radio signal modulation information includes:

[0035] S3301, The data convolution processing layer is used to perform convolution processing on the preprocessed signal data to obtain the first processed data;

[0036] S3302, using the batch normalization layer, the first processed data is normalized to obtain the second processed data;

[0037] S3303, using the first function processing layer, the second processed data is calculated and processed to obtain the third processed data;

[0038] S3304, using the random deactivation layer, the third processed data is generalized to obtain the fourth processed data;

[0039] S3305, using the first long short-term memory layer, the fourth processing data is processed to obtain the fifth processing data;

[0040] S3306, using the second function processing layer, the fifth processed data is calculated and processed to obtain the sixth processed data;

[0041] S3307, using the second long short-term memory layer, the sixth processed data is processed to obtain the seventh processed data;

[0042] S3308, using the third function processing layer, the seventh processed data is calculated and processed to obtain the eighth processed data;

[0043] S3309, using the third long short-term memory layer, the eighth processed data is processed to obtain the ninth processed data;

[0044] S3310, using the fourth function processing layer, the ninth processed data is calculated and processed to obtain the tenth processed data;

[0045] S3311 uses the first fully connected layer to process the tenth processing data to obtain the eleventh processing data;

[0046] S3312, using the fifth function processing layer, the eleventh processed data is calculated and processed to obtain the twelfth processed data;

[0047] S3313, using the second fully connected layer, the twelfth processing data is processed to obtain radio signal modulation information.

[0048] A second aspect of this invention discloses a radio signal classification device, the device comprising:

[0049] Data acquisition module, training processing module, and recognition processing module;

[0050] The data acquisition module is used to acquire the original dataset;

[0051] The training processing module is used to train the signal classification model using the original dataset to obtain an optimized signal classification model.

[0052] The identification and processing module is used to acquire radio signal data, process the radio signal data using the signal classification optimization model, and obtain radio signal modulation information.

[0053] A third aspect of the present invention discloses a radio signal classification device, the device comprising:

[0054] Memory containing executable program code;

[0055] A processor coupled to the memory;

[0056] The processor calls the executable program code stored in the memory to execute some or all of the steps in the radio signal classification method disclosed in the first aspect of the present invention.

[0057] The fourth aspect of the present invention discloses a computer-readable storage medium storing computer instructions, which, when invoked, execute some or all of the steps in the radio signal classification method disclosed in the first aspect of the present invention.

[0058] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0059] In this embodiment of the invention, a signal classification model is used to achieve effective identification of low-order complex modulation signals, thereby improving the modulation identification accuracy of high-order modulation signals. This is of great significance for ensuring the reliability and stability of communication systems. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of a scenario for the radio signal classification system provided in an embodiment of the present invention;

[0062] Figure 2 This is a flowchart illustrating a radio signal classification method disclosed in an embodiment of the present invention;

[0063] Figure 3 This refers to the signal modulation recognition accuracy under four preprocessing methods of the Convolutional Neural Network (CNN) model disclosed in this invention. Figure 3 (a) is a line graph showing the change in recognition accuracy of the CNN model with signal-to-noise ratio under the original IQ signal preprocessing method. Figure 3 (b) is a line graph showing the change in recognition accuracy of the CNN model with signal-to-noise ratio under different signal amplitude and phase preprocessing methods. Figure 3(c) is a line graph showing the change in recognition accuracy of the CNN model with signal-to-noise ratio under the signal amplitude, phase, and frequency preprocessing methods. Figure 3 (d) is a line graph showing the change in recognition accuracy of the CNN model with signal-to-noise ratio under the two-dimensional Gram angle field preprocessing method;

[0064] Figure 4 This is a confusion matrix diagram for signal modulation recognition under four preprocessing methods of the convolutional neural network (CNN) model disclosed in this invention. Figure 4 (a) is the recognition confusion matrix diagram of the CNN model under the original IQ signal preprocessing method. Figure 4 (b) is the recognition confusion matrix diagram of the CNN model under the signal amplitude and phase preprocessing methods. Figure 4 (c) is the recognition confusion matrix diagram of the CNN model under the signal amplitude, phase and frequency preprocessing methods. Figure 4 (d) is the recognition confusion matrix diagram of the CNN model under the two-dimensional Gram angle field preprocessing method;

[0065] Figure 5 This invention discloses the signal modulation recognition accuracy under four preprocessing methods for the model combining convolutional neural networks and long short-term memory networks (CNN-LSTM). Figure 5 (a) is a line graph showing the change in recognition accuracy of the CNN-LSTM model with signal-to-noise ratio under the original IQ signal preprocessing method. Figure 5 (b) is a line graph showing the change in recognition accuracy of the CNN-LSTM model with signal-to-noise ratio under different signal amplitude and phase preprocessing methods. Figure 5 (c) is a line graph showing the change in recognition accuracy of the CNN-LSTM model with signal-to-noise ratio under different signal amplitude, phase, and frequency preprocessing methods. Figure 5 (d) is a line graph showing the change in recognition accuracy of the CNN-LSTM model with signal-to-noise ratio under the two-dimensional Gram angular field preprocessing method;

[0066] Figure 6 This is a confusion matrix diagram for signal modulation recognition under four preprocessing methods of the CNN-LSTM model, which combines convolutional neural networks and long short-term memory networks, as disclosed in this embodiment of the invention. Figure 6 (a) is the recognition confusion matrix diagram of the CNN-LSTM model under the original IQ signal preprocessing method. Figure 6 (b) is the recognition confusion matrix diagram of the CNN-LSTM model under the signal amplitude and phase preprocessing methods. Figure 6 (c) is the recognition confusion matrix diagram of the CNN-LSTM model under the signal amplitude, phase and frequency preprocessing methods. Figure 6(d) is the recognition confusion matrix diagram of the CNN-LSTM model under the two-dimensional Gram angle field preprocessing method;

[0067] Figure 7 This refers to the signal modulation recognition accuracy under two preprocessing methods of the signal classification model (Convolutional Long Short Term Memory Networks with Fully Connected Layer, CLF model) disclosed in this invention. Figure 7 (a) is a line graph showing the recognition accuracy of the signal classification model as a function of the signal-to-noise ratio under different signal amplitude and phase preprocessing methods. Figure 7 (b) is a line graph showing the change in recognition accuracy of the signal classification model with signal-to-noise ratio under signal amplitude, phase and frequency preprocessing methods;

[0068] Figure 8 This is a confusion matrix diagram of signal modulation recognition under two preprocessing methods of the signal classification model (Convolutional Long Short-Term Memory Networks with Fully Connected Layer, CLF model) disclosed in this embodiment of the invention. Figure 8 (a) is the confusion matrix diagram of the signal classification model under the signal amplitude and phase preprocessing methods. Figure 8 (b) is the recognition confusion matrix diagram of the signal classification model under the signal amplitude, phase and frequency preprocessing methods;

[0069] Figure 9 This is a schematic diagram of the structure of a radio signal classification device disclosed in an embodiment of the present invention;

[0070] Figure 10 This is a schematic diagram of another radio signal classification device disclosed in an embodiment of the present invention. Detailed Implementation

[0071] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0073] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0074] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0075] It should be noted that since the method in this application embodiment is executed in a computer device, the processing objects of each computer device exist in the form of data or information, such as time, which is essentially time information. It is understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data that exist so that the computer device can process them. Specific details will not be elaborated here.

[0076] This application provides a radio signal classification method, system, apparatus, computer device, and computer-readable storage medium, which will be described in detail below.

[0077] Please see Figure 1 , Figure 1 This is a schematic diagram of a signal analysis system provided in an embodiment of this application. The system may include a computer device 100, which integrates a radio signal classification device, such as... Figure 1Computer equipment in the country.

[0078] In this embodiment, the computer device 100 can be a standalone server, a server network, or a server cluster. For example, the computer device 100 described in this embodiment includes, but is not limited to, a computer, a network host, a single network server, a set of multiple network servers, or a cloud server composed of multiple servers. The cloud server is composed of a large number of computers or network servers based on cloud computing.

[0079] It is understood that the computer device 100 used in the embodiments of this application can be a device that includes both receiving and transmitting hardware, that is, a device having receiving and transmitting hardware capable of performing bidirectional communication on a bidirectional communication link. Such a device may include: cellular or other communication devices having a single-line display, a multi-line display, or a cellular or other communication device without a multi-line display. Specifically, the computer device 100 may be a desktop terminal or a mobile terminal, and may also be one of a mobile phone, tablet computer, laptop computer, etc.

[0080] Those skilled in the art will understand that Figure 1 The application environment shown is merely one application scenario of the solution in this application and does not constitute a limitation on the application scenario of the solution in this application. Other application environments may include those that are more specific to this application. Figure 1 The number of computer devices shown is more or less, for example Figure 1 Only one computer device is shown in the diagram. It is understood that the system may also include one or more other services, which are not limited here.

[0081] In addition, such as Figure 1 As shown, the signal analysis system may also include a data storage device 200 for storing recognition result data and sample data, such as simulation result data.

[0082] It should be noted that, Figure 1 The schematic diagram of the signal analysis system shown is merely an example. The signal analysis system and scenario described in this application are intended to more clearly illustrate the technical solutions of this application and do not constitute a limitation on the technical solutions provided in this application. As those skilled in the art will know, with the evolution of signal analysis and control management systems and the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0083] This invention discloses a radio signal classification method and apparatus, which utilizes a signal classification model to effectively identify complex modulated signals, thereby improving the modulation recognition accuracy of high-order modulated signals. Detailed descriptions follow.

[0084] Example 1

[0085] Please see Figure 2 , Figure 2 This is a flowchart illustrating a radio signal classification method disclosed in an embodiment of the present invention. Figure 2 The described radio signal classification method is applied in a signal analysis and management system, such as a local server or cloud server for the system; this embodiment of the invention is not limited thereto. Figure 2 As shown, the radio signal classification method may include the following operations:

[0086] S1, Obtain the original dataset;

[0087] It should be noted that the acquisition of the raw dataset includes, but is not limited to, raw IQ data directly collected by the receiver and publicly available datasets;

[0088] S2, Using the original dataset, the signal classification model is trained to obtain an optimized signal classification model;

[0089] S3, acquire radio signal data, process the radio signal data using the signal classification optimization model, and obtain radio signal modulation information.

[0090] As can be seen, implementing the radio signal classification method described in the embodiments of the present invention enables effective identification of complex modulation signals by utilizing a signal classification model, thereby improving the modulation identification accuracy of high-order modulation signals.

[0091] In an optional embodiment, in step S2 above, the signal classification model includes: a data input layer, a data convolution processing layer, a batch normalization layer, a first function processing layer, a random deactivation layer, a first long short-term memory layer, a second function processing layer, a second long short-term memory layer, a third function processing layer, a third long short-term memory layer, a fourth function processing layer, a first fully connected layer, a fifth function processing layer, a second fully connected layer, and a data output layer;

[0092] The data input layer, the data convolution processing layer, the batch normalization layer, the first function processing layer, the random deactivation layer, the first long short-term memory layer, the second function processing layer, the second long short-term memory layer, the third function processing layer, the third long short-term memory layer, the fourth function processing layer, the first fully connected layer, the fifth function processing layer, the second fully connected layer, and the data output layer are sequentially connected.

[0093] It should be noted that the data input layer is used to receive the dataset to be processed; the dataset to be processed includes the original dataset or radio signal data, which is a 3×128 array containing the amplitude information, phase information and frequency information of the data to be processed;

[0094] It should be noted that the data convolution processing layer is used to perform convolution processing on the dataset to be processed to obtain the first processed data;

[0095] It should be noted that in this embodiment, the data convolution processing layer is a one-dimensional convolution layer with 2 input channels (3 under the data preprocessing method of amplitude, phase and frequency) and 50 output channels; the convolution kernel size is set to 7, the stride is 1, and edge padding of size 1 is used.

[0096] It should be noted that the batch normalization layer is used to normalize the first processed data to obtain the second processed data;

[0097] The batch normalization layer is used to improve the generalization ability of the model;

[0098] It should be noted that the expression for the normalization process is:

[0099]

[0100] Among them, A i The normalized value of the i-th sample data represents the normalized value; α represents the first learning parameter; β represents the second learning parameter; m represents the total number of sample data; x i Represents the i-th sample data; ε represents the second reference constant;

[0101] It should be noted that in this embodiment, ε is a very small constant used to prevent division by zero; the value of ε is 10. -5 ;

[0102] It should be noted that the first function processing layer is used to perform calculations on the second processed data to obtain the third processed data;

[0103] It should be noted that the calculation expression is as follows:

[0104]

[0105] Where x represents the second processed data;

[0106] It should be noted that the random deactivation layer is used to generalize the third processed data to obtain the fourth processed data;

[0107] The random deactivation layer is used to reduce overfitting during the training of convolutional neural networks. The ratio is set to 0.6, which means that 60% of the neuron outputs will be randomly set to zero during training. This can prevent the network from being overly sensitive to the training data and improve the model's generalization ability.

[0108] It should be noted that the generalization expression is:

[0109] y=x ⊙Bernoulli(p)

[0110] Where ⊙ denotes element-wise multiplication, Bernoulli() represents the Bernoulli distribution; p represents the probability of discarding an element; and x represents the third-processed data.

[0111] In this embodiment, p = 0.6;

[0112] It should be noted that the first long short-term memory layer is used to process the fourth processed data to obtain the fifth processed data;

[0113] It should be noted that the size of the hidden layer of the first long short-term memory layer is 64, the corresponding input is set to 50, and the size of the hidden state and the cell state is 64.

[0114] It should be noted that the processing expression for the first long short-term memory layer is:

[0115] f t =σ(w f ·[h t-1 x t ]+b f )

[0116] i t =σ(w i ·[h t-1 x t ]+b i )

[0117]

[0118] o t =σ(w o ·[h t-1 x t ]+b o )

[0119] h t =o t ·tanh(c t )

[0120] Among them, f t Indicates forgetting the gate calculation; i tIndicates input gate calculation; Indicates the candidate cell state; c t Indicates cell state; o t Indicates output gate calculation; h t Represents hidden-state computation; σ represents the sigmoid function, used to generate values ​​between 0 and 1; * represents element-wise multiplication; w f w represents the first weight matrix; i w represents the second weight matrix. c Represents the third weight matrix; w o b represents the fourth weight matrix; f b represents the first bias vector; i b represents the second bias vector; c b represents the third bias vector; o Represents the fourth bias vector; [h t-1 x t The symbol ] represents the connection between the hidden state and the input; tanh() represents the hyperbolic tangent function, used to map values ​​to the range [-1, 1].

[0121] It should be noted that the second function processing layer is used to perform calculations on the fifth processed data to obtain the sixth processed data;

[0122] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0123] It should be noted that the second long short-term memory layer is used to process the sixth processed data to obtain the seventh processed data;

[0124] It should be noted that the size of the hidden layer of the second long short-term memory layer is 100, and the corresponding input is set to 64.

[0125] It should be noted that the second long short-term memory layer is in the same manner as the first long short-term memory layer;

[0126] It should be noted that the third function processing layer is used to perform calculations on the seventh processed data to obtain the eighth processed data;

[0127] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0128] It should be noted that the third long short-term memory layer is used to process the eighth processed data to obtain the ninth processed data;

[0129] It should be noted that the size of the hidden layer of the third long short-term memory layer is 128, and the corresponding input is set to 100;

[0130] It should be noted that the third long short-term memory layer is in the same manner as the first long short-term memory layer;

[0131] It should be noted that the fourth function processing layer is used to perform calculations on the ninth processed data to obtain the tenth processed data;

[0132] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0133] It should be noted that the first fully connected layer is used to process the tenth processed data to obtain the eleventh processed data;

[0134] It should be noted that the output feature of the first fully connected layer is set to 128;

[0135] It should be noted that in this embodiment, the fully connected layer uses a linear transformation, which reduces the number of input features from 128 to 64.

[0136] y = Wx + a

[0137] Where x represents the tenth processed data, with a dimension of (N, 128), where N represents the batch size; W represents the first weight matrix, with a dimension of (64, 128); a represents the first bias vector, with a dimension of (64, 1); and y represents the eleventh processed data, with a dimension of (N, 64).

[0138] It should be noted that the fifth function processing layer is used to perform calculations on the eleventh processed data to obtain the twelfth processed data;

[0139] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0140] It should be noted that the second fully connected layer is used to process the twelfth processing data to obtain radio signal modulation information;

[0141] It should be noted that the output features of the second fully connected layer are the results of classification processing;

[0142] It should be noted that in this embodiment, the second fully connected layer also adopts a linear transformation, which reduces the number of input features from 128 to 64;

[0143] q = μp + b

[0144] Where p represents the twelfth processed data, with a dimension of (N, 128), where N represents the batch size; μ represents the second weight matrix, with a dimension of (64, 128); b represents the second bias vector, with a dimension of (64, 1); and y represents the radio signal modulation information, with a dimension of (N, 64).

[0145] It is evident that by implementing the radio signal classification method described in the embodiments of the present invention, constructing and utilizing a signal classification model to effectively identify complex modulation signals, the modulation recognition accuracy of high-order modulation signals is improved.

[0146] In another optional embodiment, step S2 above, which involves training the signal classification model using the original dataset to obtain an optimized signal classification model, includes:

[0147] S21, preprocess the original dataset to obtain a preprocessed original dataset;

[0148] S22, using the preprocessed original dataset, the signal classification model is trained to obtain an optimized signal classification model.

[0149] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, the signal classification model is trained to obtain an optimized signal classification model, thereby achieving effective identification of complex modulation signals and improving the modulation recognition accuracy of high-order modulation signals.

[0150] In another optional embodiment, step S21 above, which involves preprocessing the original dataset to obtain a preprocessed original dataset, includes:

[0151] Optionally, the original dataset is subjected to a first processing to obtain a first preprocessed dataset;

[0152] The first processing expression is:

[0153]

[0154] Where N represents the length of the original dataset; r q ri(n) represents the Q component of the nth raw data; ri(n) represents the I component of the nth raw data; Φ n Represents the phase information of the nth raw data; A n This represents the amplitude information of the nth raw data.

[0155] Optionally, the original dataset is subjected to a second processing to obtain a first preprocessed dataset;

[0156] The second processing expression is:

[0157]

[0158] Where N represents the length of the original dataset; r q (n) represents the Q component of the nth original data; r i (n) represents the I component of the nth original data; φ n Represents the phase information of the nth raw data; A n The amplitude information of the nth original data is represented; x[n] represents the time-domain discrete signal of the nth original data, and X[k] is the frequency-domain discrete signal of the nth original data; M represents the total number of samples in the original dataset; k represents the kth frequency-domain component of the original dataset; j represents the imaginary unit;

[0159] The first preprocessed dataset is normalized to obtain the second preprocessed dataset;

[0160] The normalization expression is:

[0161]

[0162] Where norm(x) represents the second preprocessed dataset; x i This represents the i-th first preprocessed data; n represents the total number of first preprocessed datasets.

[0163] The second preprocessed dataset is processed to obtain the preprocessed original dataset;

[0164] The processing expression is:

[0165]

[0166] Where X′ represents the preprocessed original dataset; x i This represents the i-th first preprocessed data (i = 1, ..., n).

[0167] It should be noted that, in this embodiment, Figure 3 and Figure 4 The diagrams show the recognition accuracy and confusion matrix of the Convolutional Neural Network (CNN) model for four different data preprocessing methods. Figure 5 and Figure 6 The graphs show the recognition accuracy and confusion matrix of a model combining convolutional neural networks and long short-term memory networks (CNN-LSTM) for four different data preprocessing methods. Figure 3 , Figure 4 , Figure 5 and Figure 6It can be seen that the data preprocessing method using amplitude, phase, and frequency has the best recognition accuracy, followed by the data preprocessing method using amplitude and phase. Therefore, the signal classification model adopts the data preprocessing method using amplitude, phase, and frequency, or the data preprocessing method using amplitude and phase.

[0168] It should be noted that in this embodiment, data preprocessing is performed using both amplitude, phase, and frequency preprocessing methods, and amplitude and phase preprocessing methods. The signal classification model (CLF model) is trained using data from both preprocessing methods, with a learning rate set to 1e-5. Figure 7 and Figure 8 As can be seen, compared to commonly used deep learning models, the signal classification model (CLF model) exhibits a significantly clear diagonal and high recognition accuracy. It effectively addresses the recognition errors in 8PSK and QPSK seen in commonly used deep learning models. Furthermore, the CLF model shows a significantly lower overall number of misidentifications in the confusion matrix. For higher-order modulation, the CLF model demonstrates a substantial advantage. For 16QAM and 64QAM, the confusion matrix reveals a clearly distinguishable diagonal compared to the largely indistinguishable rectangle displayed by commonly used deep learning models. Regarding the two different data preprocessing methods, using amplitude and phase is more convenient, requires less data, and yields better results. Therefore, the signal classification model can achieve excellent modulation recognition performance by utilizing amplitude and phase, or amplitude, phase, and frequency data preprocessing methods.

[0169] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, the original dataset is preprocessed to obtain a preprocessed original dataset, which meets the data requirements of the signal classification model, thereby achieving effective identification of complex modulation signals and improving the modulation recognition accuracy of high-order modulation signals.

[0170] In another optional embodiment, step S22 above, which involves training the signal classification model using the preprocessed original dataset to obtain an optimized signal classification model, includes:

[0171] S221, the preprocessed original dataset is divided according to a set ratio to obtain a training dataset and a test dataset;

[0172] It should be noted that in this embodiment, the set ratio is 8:2, that is, the division of the training set and the test set of the data is 8:2;

[0173] It should be noted that in this embodiment, at each signal-to-noise ratio, 8800 signals are used for training and 2200 signals are used for testing;

[0174] Preset training stopping threshold;

[0175] It should be noted that, in this embodiment of the invention, the training stopping threshold is set to 3;

[0176] S222, The training dataset is processed using the signal classification model and the multi-class cross-entropy loss function to obtain the first difference value;

[0177] S223, using the adaptive estimation function and weight decay coefficient, the parameters of the signal classification model are updated to obtain the stopping parameters;

[0178] S224, determine whether the stopping parameter is greater than the preset training stopping threshold, and obtain the stopping parameter determination result;

[0179] If the result of the stop parameter determination is yes, execute S225;

[0180] If the result of the stop parameter determination is negative, execute S222;

[0181] S227, determine whether the first difference value is less than a preset difference value threshold, and obtain the difference value discrimination result;

[0182] If the difference value determination result is negative, execute S221;

[0183] If the difference value discrimination result is yes, the training process of the signal classification model is completed, and the signal classification optimization model is obtained.

[0184] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, the signal classification model is trained by preprocessing the original dataset, thereby obtaining an optimized signal classification model, which lays the foundation for effective identification of complex modulation signals and improves the modulation recognition accuracy of high-order modulation signals.

[0185] In another optional embodiment, in step S223 above, the expression for the multi-class cross-entropy loss function is:

[0186]

[0187] Among them, Q i This represents the difference between the probability distribution of a sample belonging to class i and the true probability distribution; i represents the index of the modulation type; k represents the index of the sample data; N represents the number of modulation categories; and K represents the number of sample data in class i. This represents the true label value of the k-th sample data of the i-th modulation category; This represents the predicted probability value that the k-th sample data belongs to the i-th modulation category;

[0188] The expression for the adaptive estimation function is:

[0189]

[0190] Among them, w i+1 w represents the gradient update value at time i+1; i Represents the gradient update value at time i; μ represents the learning rate; Let represent the second moment estimate at time i; β represents the second moment estimate after bias correction at time i; i τ represents the attenuation coefficient at time i; τ represents the first reference constant.

[0191] It should be noted that the first reference constant is set to 10. -5 This is used to prevent division by zero.

[0192] It should be noted that the β i The value can be 0.99 or 0.9.

[0193] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, the signal classification model is trained by preprocessing the original dataset, thereby obtaining an optimized signal classification model, which lays the foundation for effective identification of complex modulation signals and improves the modulation recognition accuracy of high-order modulation signals.

[0194] In an optional embodiment, step S3 above, which involves acquiring radio signal data and processing the radio signal data using the signal classification optimization model to obtain radio signal modulation information, includes:

[0195] S31, receives and acquires radio signal data in real time;

[0196] It should be noted that the radio signal data refers to the raw IQ data directly acquired by the receiver;

[0197] S32, preprocess the radio signal data to obtain preprocessed signal data;

[0198] S33, Based on the signal classification optimization model, the preprocessed signal data is processed to obtain radio signal modulation information.

[0199] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, and utilizing the signal classification optimization model, the classification and identification processing of radio signal data is realized, laying the foundation for the effective identification of complex modulation signals and improving the modulation identification accuracy of high-order modulation signals.

[0200] In another optional embodiment, step S32 above, the preprocessing of the radio signal data to obtain preprocessed signal data, includes:

[0201] Optionally, the radio signal data is subjected to a first processing to obtain a first preprocessed signal set;

[0202] The first processing expression is:

[0203]

[0204] Wherein, N represents the radio signal data; r q (n) represents the Q component of the nth radio signal data; r i (n) represents the I component of the nth radio signal data; Φ n A represents the phase information of the nth radio signal data; n This represents the amplitude information of the nth radio signal data;

[0205] Optionally, the radio signal data is subjected to a second processing to obtain a first preprocessed signal set;

[0206] The second processing expression is:

[0207]

[0208] Where N represents the length of the radio signal data; rq(n) represents the Q component of the nth radio signal data; r i (n) represents the I component of the nth radio signal data; φ n A represents the phase information of the nth radio signal data; n The amplitude information of the nth radio signal data is represented; x[n] represents the time-domain discrete signal of the nth radio signal data, X[k] is the frequency-domain discrete signal of the nth radio signal data; M represents the total number of samples of the radio signal data; k represents the kth frequency-domain component of the radio signal data; j represents the imaginary unit;

[0209] The first preprocessed signal set is normalized to obtain the second preprocessed signal set;

[0210] The normalization expression is:

[0211]

[0212] Where norm(x) represents the second preprocessed signal set; x i This represents the i-th first preprocessed signal data; n represents the total number of first preprocessed signals.

[0213] The second preprocessed signal set is processed to obtain a preprocessed signal dataset;

[0214] The processing expression is:

[0215]

[0216] Where X′ represents the preprocessed signal dataset; x i This represents the i-th second preprocessed signal (i = 1, ..., n).

[0217] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, radio signal data is preprocessed to obtain preprocessed signal data, which meets the data requirements of the signal classification optimization model, thereby achieving effective identification of complex modulation signals and improving the modulation identification accuracy of high-order modulation signals.

[0218] In another optional embodiment, step S33 above, which involves processing the preprocessed signal data based on the signal classification optimization model to obtain radio signal modulation information, includes:

[0219] S3301, The data convolution processing layer is used to perform convolution processing on the preprocessed signal data to obtain the first processed data;

[0220] Optionally, in the data preprocessing method of amplitude, phase and frequency, the input channel of the convolution process is 3, the output channel is 50, the convolution kernel size is set to 7, the stride is 1, and edge padding with a size of 1 is used;

[0221] Optionally, in the amplitude and phase data preprocessing method, the input channel of the convolution process is 2, the output channel is 50, the convolution kernel size is set to 7, the stride is 1, and edge padding with a size of 1 is used;

[0222] S3302, using the batch normalization layer, the first processed data is normalized to obtain the second processed data;

[0223] It should be noted that the expression for the normalization process is:

[0224]

[0225] Among them, A iThe normalized value of the i-th sample data represents the normalized value; α represents the first learning parameter; β represents the second learning parameter; m represents the total number of sample data; x i Represents the i-th sample data; ε represents the second reference constant;

[0226] It should be noted that in this embodiment, ε is set to 10. -5 ;

[0227] S3303, using the first function processing layer, the second processed data is calculated and processed to obtain the third processed data;

[0228] It should be noted that the calculation expression is as follows:

[0229]

[0230] Where x represents the second processed data;

[0231] S3304, using the random deactivation layer, the third processed data is generalized to obtain the fourth processed data;

[0232] It should be noted that the generalization expression is:

[0233] y=x ⊙ Bernoulli(p)

[0234] Where ⊙ denotes element-wise multiplication, Bernoulli() represents the Bernoulli distribution; p represents the probability of discarding an element; and x represents the third-processed data.

[0235] S3305, using the first long short-term memory layer, the fourth processing data is processed to obtain the fifth processing data;

[0236] It should be noted that the size of the hidden layer of the first long short-term memory layer is 64, the corresponding input is set to 50, and the size of the hidden state and the cell state is 64.

[0237] It should be noted that the processing expression for the first long short-term memory layer is:

[0238] f t =σ(w f ·[h t-1 x t ]+b f )

[0239] i t =σ(w i ·[h t-1 x t ]+b i )

[0240]

[0241] o t =σ(w o ·[h t-1 x t ]+b o )

[0242] h t =o t ·tanh(c t )

[0243] Among them, f t Indicates forgetting the gate calculation; i t Indicates input gate calculation; Indicates the candidate cell state; c t Indicates cell state; o t Indicates output gate calculation; h t Represents hidden-state computation; σ represents the sigmoid function, used to generate values ​​between 0 and 1; * represents element-wise multiplication; w f w represents the first weight matrix; i w represents the second weight matrix. c Represents the third weight matrix; w o b represents the fourth weight matrix; f b represents the first bias vector; i b represents the second bias vector; c b represents the third bias vector; o Represents the fourth bias vector; [h t-1 x t The symbol ] represents the connection between the hidden state and the input; tanh() represents the hyperbolic tangent function, used to map values ​​to the range [-1, 1].

[0244] S3306, using the second function processing layer, the fifth processed data is calculated and processed to obtain the sixth processed data;

[0245] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0246] S3307, using the second long short-term memory layer, the sixth processed data is processed to obtain the seventh processed data;

[0247] It should be noted that the size of the hidden layer of the second long short-term memory layer is 100, and the corresponding input is set to 64;

[0248] It should be noted that the second long short-term memory layer is in the same manner as the first long short-term memory layer;

[0249] S3308, using the third function processing layer, the seventh processed data is calculated and processed to obtain the eighth processed data;

[0250] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0251] S3309, using the third long short-term memory layer, the eighth processed data is processed to obtain the ninth processed data;

[0252] It should be noted that the size of the hidden layer of the third long short-term memory layer is 128, and the corresponding input is set to 100;

[0253] It should be noted that the third long short-term memory layer is in the same manner as the first long short-term memory layer;

[0254] S3310, using the fourth function processing layer, the ninth processed data is calculated and processed to obtain the tenth processed data;

[0255] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0256] S3311 uses the first fully connected layer to process the tenth processing data to obtain the eleventh processing data;

[0257] It should be noted that the output feature of the first fully connected layer is set to 128;

[0258] It should be noted that in this embodiment, the fully connected layer uses a linear transformation, which reduces the number of input features from 128 to 64.

[0259] y = Wx + a

[0260] Where x represents the tenth processed data, with a dimension of (N, 128), where N represents the batch size; W represents the first weight matrix, with a dimension of (64, 128); a represents the first bias vector, with a dimension of (64, 1); and y represents the eleventh processed data, with a dimension of (N, 64).

[0261] S3312, using the fifth function processing layer, the eleventh processed data is calculated and processed to obtain the twelfth processed data;

[0262] It should be noted that the calculation and processing method described is consistent with the calculation and processing method of the first function processing layer;

[0263] S3313, using the second fully connected layer, the twelfth processing data is processed to obtain radio signal modulation information;

[0264] It should be noted that the output features of the second fully connected layer are the results of classification processing;

[0265] It should be noted that in this embodiment, the second fully connected layer also adopts a linear transformation, which reduces the number of input features from 128 to 64;

[0266] q = μp + b

[0267] Where p represents the twelfth processed data, with dimensions (N, 128), where N represents the batch size; μ represents the second weight matrix, with dimensions (64, 128); b represents the second bias vector, with dimensions (64, 1); and y represents the radio signal modulation information, with dimensions (N, 64).

[0268] As can be seen, by implementing the radio signal classification method described in the embodiments of the present invention, the preprocessed signal data is processed using a signal classification optimization model, thereby achieving effective identification of complex modulation signals and improving the modulation recognition accuracy of high-order modulation signals.

[0269] Example 2

[0270] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of a radio signal classification device disclosed in an embodiment of the present invention. Figure 9 The described apparatus can be applied in signal analysis and management systems, such as local servers or cloud servers for signal analysis systems, and the embodiments of the present invention are not limited thereto. Figure 9 As shown, the device may include:

[0271] Data acquisition module 101, training processing module 102, and recognition processing module 103;

[0272] The data acquisition module 101 is used to acquire the original dataset;

[0273] The training processing module 102 is used to train the signal classification model using the original dataset to obtain an optimized signal classification model.

[0274] The identification and processing module 103 is used to acquire radio signal data, process the radio signal data using the signal classification optimization model, and obtain radio signal modulation information.

[0275] Example 3

[0276] Please see Figure 10 , Figure 10 This is a schematic diagram of the structure of a radio signal classification device disclosed in an embodiment of the present invention. Figure 10The described apparatus can be applied in signal analysis and management systems, such as local servers or cloud servers for signal analysis systems, and the embodiments of the present invention are not limited thereto. Figure 10 As shown, the device may include:

[0277] Memory 201 storing executable program code;

[0278] Processor 202 coupled to memory 201;

[0279] The processor 202 calls the executable program code stored in the memory 201 to execute the steps in the radio signal classification method described in Embodiment 1.

[0280] Example 4

[0281] This invention discloses a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform the steps in the radio signal classification method described in Embodiment 1.

[0282] Example 5

[0283] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform the steps in the radio signal classification method described in Embodiment 1.

[0284] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0285] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0286] Finally, it should be noted that the radio signal classification method, system, and apparatus disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for classifying radio signals, characterized in that, The method includes: S1, Obtain the original dataset; S2, Using the original dataset, the signal classification model is trained to obtain an optimized signal classification model; The signal classification model includes: a data input layer, a data convolution processing layer, a batch normalization layer, a first function processing layer, a random deactivation layer, a first long short-term memory layer, a second function processing layer, a second long short-term memory layer, a third function processing layer, a third long short-term memory layer, a fourth function processing layer, a first fully connected layer, a fifth function processing layer, a second fully connected layer, and a data output layer. The data input layer, the data convolution processing layer, the batch normalization layer, the first function processing layer, the random deactivation layer, the first long short-term memory layer, the second function processing layer, the second long short-term memory layer, the third function processing layer, the third long short-term memory layer, the fourth function processing layer, the first fully connected layer, the fifth function processing layer, the second fully connected layer, and the data output layer are sequentially connected. S3, acquire radio signal data, process the radio signal data using the signal classification optimization model, and obtain radio signal modulation information; The step of training the signal classification model using the original dataset to obtain an optimized signal classification model includes: S21, preprocess the original dataset to obtain a preprocessed original dataset; S22, Using the preprocessed original dataset, the signal classification model is trained to obtain an optimized signal classification model; The step of preprocessing the original dataset to obtain a preprocessed original dataset includes: The original dataset is processed to obtain a first preprocessed dataset; The processing expression is: Where N represents the length of the original dataset; This represents the Q component of the nth original data; This represents the I component of the nth original data; This represents the phase information of the nth raw data. This represents the amplitude information of the nth raw data. This represents the time-domain discrete signal of the nth original data. It is the frequency domain discrete signal of the nth original data; M represents the total number of samples in the original dataset; k represents the kth frequency domain component of the original dataset; j represents the imaginary unit; The first preprocessed dataset is normalized to obtain the second preprocessed dataset; The normalization expression is: in, This represents the second preprocessed dataset; This represents the i-th first preprocessed data; n represents the total number of first preprocessed datasets. The second preprocessed dataset is processed to obtain the preprocessed original dataset; The processing expression is: in, This refers to the preprocessed original dataset; Indicates the first The first preprocessed data; The step of training the signal classification model using the preprocessed original dataset to obtain an optimized signal classification model includes: S221, the preprocessed original dataset is divided according to a set ratio to obtain a training dataset and a test dataset; Preset training stopping threshold; S222, The training dataset is processed using the signal classification model and the multi-class cross-entropy loss function to obtain the first difference value; S223, using the adaptive estimation function and weight decay coefficient, the parameters of the signal classification model are updated to obtain the stopping parameters; S224, determine whether the stopping parameter is greater than the preset training stopping threshold, and obtain the stopping parameter determination result; If the result of the stop parameter determination is yes, execute S225; If the result of the stop parameter determination is negative, execute S222; S225, determine whether the first difference value is less than a preset difference value threshold, and obtain the difference value discrimination result; If the difference value determination result is negative, execute S221; If the difference value discrimination result is yes, the training process of the signal classification model is completed, and the signal classification optimization model is obtained.

2. The radio signal classification method according to claim 1, characterized in that, The expression for the multi-class cross-entropy loss function is as follows: in, Indicates that the sample belongs to the first The difference between the probability of a class and the true probability distribution; An index representing the modulation type; Indicates the sample data index; N represents the number of modulation categories; K represents the index of the first sample data. Number of class sample data; Indicates the first Modulation category number The true label value of each sample data; Indicates the first The nth sample data is the nth The predicted probability value of the modulation category; The expression for the adaptive estimation function is: in, Indicates the first Gradient update values ​​at different times; Indicates the first Gradient update values ​​at different times; Indicates the learning rate; Indicates the first The estimated value of the second moment at time t; Indicates the first The estimated value of the second moment after time deviation correction; Indicates the first The decay coefficient at time; This represents the first reference constant.

3. The radio signal classification method according to claim 1, characterized in that, The process of acquiring radio signal data and processing the radio signal data using the signal classification optimization model to obtain radio signal modulation information includes: S31, receives and acquires radio signal data in real time; S32, preprocess the radio signal data to obtain preprocessed signal data; S33, Based on the signal classification optimization model, the preprocessed signal data is processed to obtain radio signal modulation information.

4. The radio signal classification method according to claim 3, characterized in that, The step of processing the preprocessed signal data based on the signal classification optimization model to obtain radio signal modulation information includes: S3301, The data convolution processing layer is used to perform convolution processing on the preprocessed signal data to obtain the first processed data; S3302, using the batch normalization layer, the first processed data is normalized to obtain the second processed data; S3303, using the first function processing layer, the second processed data is calculated and processed to obtain the third processed data; S3304, using the random deactivation layer, the third processed data is generalized to obtain the fourth processed data; S3305, using the first long short-term memory layer, the fourth processing data is processed to obtain the fifth processing data; S3306, using the second function processing layer, the fifth processed data is calculated and processed to obtain the sixth processed data; S3307, using the second long short-term memory layer, the sixth processed data is processed to obtain the seventh processed data; S3308, using the third function processing layer, the seventh processed data is calculated and processed to obtain the eighth processed data; S3309, using the third long short-term memory layer, the eighth processed data is processed to obtain the ninth processed data; S3310, using the fourth function processing layer, the ninth processed data is calculated and processed to obtain the tenth processed data; S3311 uses the first fully connected layer to process the tenth processing data to obtain the eleventh processing data; S3312, using the fifth function processing layer, the eleventh processed data is calculated and processed to obtain the twelfth processed data; S3313, using the second fully connected layer, the twelfth processing data is processed to obtain radio signal modulation information.

5. A radio signal classification device, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the radio signal classification method as described in any one of claims 1-4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when invoked, are used to perform the radio signal classification method as described in any one of claims 1-4.

Images