Zero sample signal modulation identification method and device based on depth attribute extraction

Abstract

The invention provides a zero sample signal modulation identification method based on depth attribute extraction, and relates to the technical field of deep learning, and the method comprises the steps: extracting the real attribute representation of each modulation category from a signal sample generated through simulation through a feature extraction module of a pre-trained classification model; extracting time sequence characteristics of the input data by using a long short-term memory (LSTM) network, and mapping the time sequence characteristics to an attribute space through a linear transformation layer to obtain prediction attribute representation of the input data; and projecting the prediction attribute representation of the input data and the real attribute representation of each modulation category into the same embedding space, calculating the similarity between the prediction attribute and the real attribute, and determining the category of the input data based on the similarity. By adopting the scheme, zero sample signal modulation identification is realized.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a method and device for zero-sample signal modulation recognition based on deep attribute extraction. Background Art

[0002] With the rapid development of wireless communication technology, the number of wireless devices has grown exponentially, and the communication environment has become increasingly complex and harsh. During the transmission process, the signal is reshaped by various adversarial factors such as noise, multipath interference, and channel fading. Its original characteristics are severely weakened or distorted, which puts higher requirements on the accurate recognition of the signal. At the same time, the modulation methods of the signal are becoming more and more complex and diverse, which further increases the difficulty of modulation recognition. Traditional modulation recognition methods rely on domain experts to manually extract features based on experience. This process is not only time-consuming, but also limited by the prior knowledge of the received signal. In the current increasingly complex communication system, it may lead to a significant decrease in modulation recognition performance. Signal modulation recognition is facing unprecedented severe challenges.

[0003] In recent years, inspired by the great success of deep learning in image recognition, natural language processing and other fields, scholars have introduced deep learning into the field of automatic modulation recognition. Deep learning-based methods can automatically learn the potential characteristics of signals, breaking through the limitations of traditional methods that rely on manual feature extraction and achieving higher recognition accuracy. However, the performance of deep learning depends largely on large-scale labeled data sets, and its recognition ability is usually limited to the categories already included in the training data. In practical applications, the signal data available for certain signal categories is extremely scarce, resulting in incomplete coverage of signal categories in the training set, which limits the modulation recognition ability of deep learning models.

[0004] Zero-Shot Learning (ZSL) is an effective method to address the above challenges. It transfers knowledge from known classes to unknown classes by using semantic descriptions or attribute information of classes, enabling the model to recognize classes that have never been seen in the training phase. It solves the problem that the model cannot effectively recognize unknown classes when data is scarce, and has been widely used in the image field. However, in the field of signal modulation recognition, signal modulation types are extremely complex and diverse, and there is a lack of clear semantic distinction between different modulation types, making it difficult to capture their complexity with simple and clear natural language descriptions. The unexplainable nature of modulated signals poses a huge challenge to zero-shot scenarios, and it is imperative to develop an effective zero-shot classification method for signal modulation recognition. Summary of the invention

[0005] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0006] To this end, the first purpose of this application is to propose a zero-sample signal modulation recognition method based on deep attribute extraction. In view of the uninterpretability of modulated signals, semantic information characterizing different modulation modes is extracted based on a pre-trained attribute extraction method; and unknown classes are identified through attribute prediction and matching, which does not rely on labeled samples of unknown classes, providing an effective solution for modulated signal recognition in zero-sample scenarios.

[0007] The second objective of the present application is to propose a zero-sample signal modulation identification device based on deep attribute extraction.

[0008] The third objective of the present application is to provide a computer device.

[0009] A fourth object of the present application is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a zero-sample signal modulation recognition method based on deep attribute extraction, including:

[0011] The feature extraction module of the pre-trained classification model extracts the true attribute representation of each modulation category from the simulated signal samples;

[0012] The long short-term memory network (LSTM) is used to extract the time series features of the input data, and the time series features are mapped to the attribute space through the linear transformation layer to obtain the predicted attribute representation of the input data;

[0013] The predicted attribute representation of the input data and the true attribute representation of each modulation category are projected into the same embedding space, the similarity between the predicted attribute and the true attribute is calculated, and the category of the input data is determined based on the similarity.

[0014] Optionally, in one embodiment of the present application, the pre-trained classification model includes:

[0015] Select a labeled data set of known classes as input, and use supervised learning methods to pre-train the classification model. During pre-training, the input data is classified through LSTM and linear transformation layers, and the classification model is optimized by minimizing the cross entropy loss function;

[0016] The feature extraction module is obtained by removing the fully connected layers from the classification model;

[0017] The signal sample generation process includes:

[0018] According to the generation mechanism of the modulation signal, signal samples are simulated and generated for all modulation categories, and a signal sample set is constructed based on the signal samples, wherein the signal sample set includes signal samples of known classes and signal samples of unknown classes;

[0019] The feature extraction module extracts the true attribute representation of each modulation category from the signal sample, including:

[0020] The signal samples in the signal sample set are input into the feature extraction module to extract the true attribute representation of each signal sample.

[0021] Optionally, in one embodiment of the present application, the signal sample set is D S,U , the signal sample set includes signal samples of known classes D S And the signal sample D of the unknown class U , D S The real property of is expressed as:

[0022] Att S =E(D S θ * )

[0023] D U The real property of is expressed as:

[0024] Att U =E(D U θ * )

[0025] Among them, Att S is the attribute matrix of the known class, Att U is the attribute matrix of the unknown class, E is the attribute extraction function, θ * is the optimal parameter learned by pre-training.

[0026] Optionally, in one embodiment of the present application, a long short-term memory network LSTM is used to extract the time series features of the input data, and the time series features are mapped to the attribute space through a linear transformation layer to obtain a predicted attribute representation of the input data, including:

[0027] Use LSTM to extract the time series features of input data;

[0028] The last time step of the time series feature is selected and mapped to the attribute space through a linear transformation layer to obtain the predicted attribute representation of the input data.

[0029] Optionally, in one embodiment of the present application, the time series characteristics of the input data are expressed as:

[0030] h i =f(x i )

[0031] Among them, x i is the input data, h i is the time series feature, f(·) represents the LSTM feature extraction function;

[0032] The predicted attributes of the input data are expressed as:

[0033] a pi =fc(h i [:,-1,:])

[0034] Where fc represents the linear transformation function, h i [:,-1,:] is the last time step of the time series feature.

[0035] Optionally, in one embodiment of the present application, calculating the similarity between the predicted attribute and the actual attribute includes:

[0036] Calculate the similarity score between the predicted attributes of the input data and the true attributes of each modulation category by performing a matrix multiplication operation;

[0037] Determine the category of input data based on similarity, including:

[0038] Use the softmax activation function for multi-classification, convert the similarity into probability distribution, and obtain the category prediction result of the input data.

[0039] Optionally, in one embodiment of the present application, the similarity score is expressed as:

[0040] z i,n =a pi ·Att S,U

[0041] Among them, z i,n For the input data x i The similarity score between the predicted attribute and the true attribute of the nth category, Att S,U is the attribute matrix of all categories including known and unknown classes;

[0042] The probability distribution is:

[0043]

[0044] Among them, P i,n For the input data x i The probability of belonging to class n, C is the number of classes.

[0045] To achieve the above-mentioned purpose, a second embodiment of the present invention proposes a zero-sample signal modulation recognition device based on deep attribute extraction, comprising:

[0046] A real attribute extraction module, used to extract the real attribute representation of each modulation category from the signal samples generated by simulation through the feature extraction module in the pre-trained classification model;

[0047] The attribute prediction module is used to extract the time series features of the input data using the long short-term memory network LSTM, and map the time series features to the attribute space through the linear transformation layer to obtain the predicted attribute representation of the input data;

[0048] The classification module is used to project the predicted attribute representation of the input data and the real attribute representation of each modulation category into the same embedding space, calculate the similarity between the predicted attribute and the real attribute, and determine the category of the input data based on the similarity.

[0049] To achieve the above-mentioned purpose, the third aspect of the present invention proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned zero-sample signal modulation identification method based on deep attribute extraction is implemented.

[0050] In order to achieve the above-mentioned objectives, the fourth aspect of the present invention proposes a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by a processor, the above-mentioned zero-sample signal modulation identification method based on depth attribute extraction can be executed.

[0051] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0053] Figure 1 A flowchart of a zero-sample signal modulation recognition method based on deep attribute extraction provided in the first embodiment of the present application;

[0054] Figure 2 A schematic diagram of the structure of a zero-sample signal modulation identification device based on deep attribute extraction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0056] When humans recognize images, they map the target category into the attribute space, and then classify the image according to the most discriminative attribute. Inspired by the human decision-making process, this application proposes a zero-sample signal modulation recognition method based on deep attribute extraction. Since the signal semantics are uninterpretable, a pre-training-based attribute extraction method is proposed. The feature extraction ability learned by the model in the pre-training stage is used to extract representative attribute representations for each modulation category from the simulated signal samples; the time series feature extraction ability of the long short-term memory network LSTM and the mapping ability of the linear transformation layer are used to predict the signal attributes; unknown classes are identified through attribute prediction and matching, without relying on labeled samples of unknown classes, which significantly improves the model's generalization ability for new categories and provides an effective solution for modulated signal recognition in zero-sample scenarios.

[0057] The following describes a method and device for zero-sample signal modulation recognition based on deep attribute extraction according to an embodiment of the present application with reference to the accompanying drawings.

[0058] Figure 1 A flowchart of a zero-sample signal modulation recognition method based on deep attribute extraction provided in Example 1 of the present application.

[0059] like Figure 1 As shown, the zero-sample signal modulation recognition method based on deep attribute extraction includes the following steps:

[0060] Step 101, extracting the true attribute representation of each modulation category from the simulated generated signal samples through the feature extraction module of the pre-trained classification model;

[0061] Step 102, using a long short-term memory network LSTM to extract the time series features of the input data, and mapping the time series features to the attribute space through a linear transformation layer to obtain a predicted attribute representation of the input data;

[0062] Step 103 : Project the predicted attribute representation of the input data and the real attribute representation of each modulation category into the same embedding space, calculate the similarity between the predicted attribute and the real attribute, and determine the category of the input data based on the similarity.

[0063] Optionally, in one embodiment of the present application, the pre-trained classification model includes:

[0064] Select a large-scale labeled dataset S with known classes s As input, a classification model is pre-trained using supervised learning methods, and the input data Classification is performed through LSTM and linear transformation layers, and the model is optimized by minimizing the cross entropy loss function;

[0065] After pre-training, the model has learned powerful feature extraction capabilities. By removing its last fully connected layer, the classifier is transformed into a powerful attribute extractor E, which is used as a feature extraction module.

[0066] The signal sample generation process includes:

[0067] According to the generation mechanism of modulation signals, signal samples are simulated and generated for all modulation types using the MATLAB platform, and a small-scale signal sample set D is constructed. S,U , which includes signal samples of known classes D S and signal samples of unknown class D U ;

[0068] The feature extraction module extracts the true attribute representation of each modulation category from the signal sample, including:

[0069] These simulated signal samples are input into the attribute extractor E, and the optimal parameters θ learned in the pre-training phase are loaded. * , extracting representative attribute representations.

[0070] Optionally, in one embodiment of the present application, D S The real property of is expressed as:

[0071] Att S =E(D S θ * )

[0072] D U The real property of is expressed as:

[0073] Att U =E(D U θ * )

[0074] Among them, E is the attribute extraction function, Att S is the attribute matrix of the known class, containing N S The attribute vector of known classes, Att U is the attribute matrix of the unknown class, containing N U The attribute vector of the unknown class, Each category has a D e -dimensional attribute vector.

[0075] Optionally, in one embodiment of the present application, a long short-term memory network LSTM is used to extract the time series features of the input data, and the time series features are mapped to the attribute space through a linear transformation layer to obtain a predicted attribute representation of the input data, including:

[0076] Use LSTM to extract rich temporal features of input data;

[0077] The last time step of the time series feature is selected and further mapped to the attribute space through a linear transformation layer to predict the signal attributes and obtain the predicted attributes of the input data.

[0078] Optionally, in one embodiment of the present application, the time series characteristics of the input data are expressed as:

[0079] h i =f(x i )

[0080] Among them, x i is the input data, h i is the time series feature, f(·) represents the LSTM feature extraction function;

[0081] The predicted attributes of the input data are:

[0082] a pi =fc(h i [:,-1,:])

[0083] Where fc represents the linear transformation function, h i [:,-1,:] is the last time step of the time series feature.

[0084] Optionally, in one embodiment of the present application, calculating the similarity between the predicted attribute and the actual attribute includes:

[0085] Calculate the similarity score between the predicted attributes of the input data and the true attributes of each modulation category by performing a matrix multiplication operation;

[0086] Determine the category of input data based on similarity, including:

[0087] Use the softmax activation function for multi-classification, convert the output similarity into a probability distribution, and obtain the category prediction result of the sample.

[0088] Optionally, in one embodiment of the present application, the similarity score is expressed as:

[0089] z i,n =a pi ·Att S,U

[0090] Among them, z i,n For the input data x i The similarity score between the predicted attribute and the true attribute of the nth category, Att S,U is the attribute matrix of all categories including known and unknown classes;

[0091] The probability distribution is:

[0092]

[0093] Among them, P i,n For the input data x i The probability of belonging to class n, C is the number of classes.

[0094] The zero-sample signal modulation recognition method based on deep attribute extraction in the embodiment of the present application, based on a thorough study of the principles of deep learning and signal modulation classification, uses a pre-trained attribute extraction method and the attribute extraction capability learned in the pre-training stage to extract representative real attribute representations for each modulation category from the simulated signal samples, thereby realizing knowledge transfer from known classes to unknown classes and solving the problem of unexplainable signal semantics; by utilizing the time series feature extraction capability of the long short-term memory network LSTM and the mapping capability of the linear transformation layer, the signal attributes are predicted; by comparing the similarity between the predicted attributes and the real attributes for classification, the model is able to not only identify known signal categories, but also effectively classify unknown categories through attribute prediction and comparison, thereby significantly improving the model's generalization capability for new categories, which is of great significance for quickly adapting to new categories without the need to re-collect and label large amounts of data.

[0095] In order to implement the above embodiments, the present application also proposes a zero-sample signal modulation recognition device based on deep attribute extraction.

[0096] Figure 2 A schematic diagram of the structure of a zero-sample signal modulation identification device based on deep attribute extraction provided in an embodiment of the present application.

[0097] like Figure 2 As shown, the zero-sample signal modulation identification device based on deep attribute extraction includes:

[0098] A real attribute extraction module, used to extract the real attribute representation of each modulation category from the signal samples generated by simulation through the feature extraction module in the pre-trained classification model;

[0099] The attribute prediction module is used to extract the time series features of the input data using the long short-term memory network LSTM, and map the time series features to the attribute space through the linear transformation layer to obtain the predicted attribute representation of the input data;

[0100] The classification module is used to project the predicted attribute representation of the input data and the real attribute representation of each modulation category into the same embedding space, calculate the similarity between the predicted attribute and the real attribute, and determine the category of the input data based on the similarity.

[0101] It should be noted that the aforementioned explanation of the embodiment of the zero-sample signal modulation identification method based on depth attribute extraction is also applicable to the zero-sample signal modulation identification device based on depth attribute extraction of this embodiment, and will not be repeated here.

[0102] In order to implement the above embodiments, the present invention further proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method described in the above embodiments is implemented.

[0103] In order to implement the above embodiments, the present invention further proposes a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method of the above embodiments is implemented.

[0104] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0105] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0106] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0107] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0108] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0109] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0110] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0111] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A zero-sample signal modulation recognition method based on deep attribute extraction, characterized in that: include: The feature extraction module of the pre-trained classification model extracts the true attribute representation of each modulation category from the simulated signal samples; The time series features of the input data are extracted using a long short-term memory network (LSTM), and the time series features are mapped to the attribute space through a linear transformation layer to obtain a predicted attribute representation of the input data; The predicted attribute representation of the input data and the true attribute representation of each modulation category are projected into the same embedding space, the similarity between the predicted attribute and the true attribute is calculated, and the category of the input data is determined based on the similarity.

2. The method according to claim 1, characterized in that Pre-training the classification model, including: Selecting a labeled data set of known classes as input, using a supervised learning method to pre-train the classification model, during pre-training, classifying the input data through LSTM and linear transformation layers, and optimizing the classification model by minimizing the cross entropy loss function; The feature extraction module is obtained by removing the fully connected layer from the classification model; The signal sample generation process includes: According to the generation mechanism of the modulation signal, signal samples are simulated and generated for all modulation categories, and a signal sample set is constructed based on the signal samples, wherein the signal sample set includes signal samples of known classes and signal samples of unknown classes; The feature extraction module extracts the true attribute representation of each modulation category from the signal sample, including: The signal samples in the signal sample set are input into the feature extraction module to extract the true attribute representation of each signal sample.

3. The method according to claim 2, characterized in that The signal sample set is D S,U , the signal sample set includes signal samples D of known classes S And the signal sample D of the unknown class U , D S The real property of is expressed as: To S =E(D S ;θ * ) D U The real property of is expressed as: To U =E(D U ;θ * ) Among them, Att S is the attribute matrix of the known class, Att U is the attribute matrix of the unknown class, E is the attribute extraction function, θ * is the optimal parameter learned by pre-training.

4. The method according to claim 1, characterized in that The method of extracting the time series features of the input data by using the long short-term memory network LSTM and mapping the time series features to the attribute space through a linear transformation layer to obtain the predicted attribute representation of the input data includes: Use LSTM to extract the time series features of input data; The last time step of the time series feature is selected and mapped to the attribute space through a linear transformation layer to obtain the predicted attribute representation of the input data.

5. The method according to claim 4, characterized in that The time series characteristics of the input data are expressed as: h i =f(x i ) Among them, x i is the input data, h i is the time series feature, f(·) represents the LSTM feature extraction function; The predicted attributes of the input data are expressed as: and pi =fc(h i [:,-1,:]) Where fc represents the linear transformation function, h i [:,-1,:] is the last time step of the time series feature.

6. The method according to claim 1, characterized in that Calculate the similarity between the predicted attributes and the true attributes, including: Calculate the similarity score between the predicted attributes of the input data and the true attributes of each modulation category by performing a matrix multiplication operation; Determine the category of input data based on similarity, including: Use the softmax activation function for multi-classification, convert the similarity into probability distribution, and obtain the category prediction result of the input data.

7. The method according to claim 6, characterized in that The similarity score is expressed as: z i,n =a pi ·To S,U Among them, z i,n For the input data x i The similarity score between the predicted attribute and the true attribute of the nth category, Att S,U is the attribute matrix of all categories including known and unknown classes; The probability distribution is: Among them, P i,n For the input data x i The probability of belonging to class n, C is the number of classes.

8. A zero-sample signal modulation recognition device based on deep attribute extraction, characterized in that: include: A real attribute extraction module, used to extract the real attribute representation of each modulation category from the signal samples generated by simulation through the feature extraction module in the pre-trained classification model; An attribute prediction module is used to extract the time series features of the input data using a long short-term memory network (LSTM), and map the time series features to the attribute space through a linear transformation layer to obtain a predicted attribute representation of the input data; The classification module is used to project the predicted attribute representation of the input data and the real attribute representation of each modulation category into the same embedding space, calculate the similarity between the predicted attribute and the real attribute, and determine the category of the input data based on the similarity.

9. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images