PSK (Phase Shift Keying) signal intelligent modulation identification method based on spectral correlation density

Abstract

The invention provides a PSK signal intelligent modulation identification method based on spectral correlation density, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the simulation of various PSK signals and non-PSK signals; performing sliding window cutting on the signal; carrying out spectrum correlation density calculation on the cut signal; performing Hilbert transform on the signal; respectively normalizing the spectral correlation density characteristics and the IQ path data, constructing a data set, and dividing the data set into a training set and a verification set; inputting the data into a self-attention mechanism neural network to generate a deep learning model; a cyclic frequency is determined through a cyclic autocorrelation blind search algorithm; calculating spectral correlation density characteristics and IQ path data, and inputting the data into the model after normalization; and outputting probability distribution of the plurality of prediction samples, and taking a result corresponding to the highest probability as a final modulation type identification result. According to the method, the PSK signal can be identified, the unknown signal can be identified, and the method has very high identification accuracy and anti-noise and anti-interference capability.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a PSK signal intelligent modulation recognition method based on spectral correlation density. Background Art

[0002] Traditional PSK signal modulation recognition methods rely primarily on time-frequency analysis and feature extraction techniques. These methods have shown some effectiveness in processing PSK signals in simple scenarios. However, with the advancement of communication technology and the increasing complexity of electromagnetic environments, existing modulation recognition methods are no longer able to meet the needs of modern communication systems.

[0003] In recent years, the rapid development of deep learning technology has provided a novel solution for PSK signal modulation recognition. Research has shown that deep learning-based PSK signal recognition methods can maintain high accuracy even in low signal-to-noise ratio environments, significantly improving over traditional methods. While deep learning offers new promise for PSK signal modulation recognition, integrating deep learning with existing technologies in practical applications still faces numerous challenges.

[0004] First, the non-stationary nature of PSK signals means that the feature matrix generated by traditional time-frequency analysis may suffer from information loss, making it difficult to meet the high-discrimination feature requirements of deep learning. Second, in real-world applications, the dynamic changes in signal parameters (such as carrier frequency and symbol rate) require models with strong generalization capabilities. Furthermore, specialized scenarios such as military operations place stringent demands on real-time and lightweight algorithms, making existing complex deep learning models difficult to deploy on edge devices. Summary of the Invention

[0005] In order to solve the technical problems in the related art, the present invention provides a PSK signal intelligent modulation identification method based on spectral correlation density.

[0006] In order to achieve the above object, the technical solution adopted by the present invention includes: According to a first aspect of the present invention, a method for intelligent modulation identification of a PSK signal based on spectral correlation density is provided, comprising the following steps: Step S1: Simulate a variety of PSK signals and non-PSK signals to generate signal samples containing random parameters; perform sliding window clipping on the signals and discard samples whose length is less than a predetermined window length; Step S2: Calculate the spectral correlation density of the clipped signal to obtain spectral correlation density features; and perform Hilbert transform on the signal to extract I-channel data and Q-channel data. Step S3: Normalize the spectrum correlation density features, I-channel data, and Q-channel data respectively, establish a mapping relationship with the signal modulation type, construct a data set, and divide it into a training set and a validation set; Step S4: input the preprocessed data into the self-attention mechanism neural network, optimize the network parameters through back propagation until the loss function converges, and generate a trained deep learning model; Step S5: determine the cycle frequency of the predicted signal through a cyclic autocorrelation blind search algorithm; calculate the spectral correlation density feature, I data and Q data after cutting the predicted signal, and input the normalized data into the trained model; Step S6: output the probability distribution of multiple prediction samples, and take the result corresponding to the highest probability as the final modulation type recognition result.

[0007] Optionally, in the step S1, The PSK signal includes at least one of the following types: BPSK, QPSK, 8PSK, 16PSK, QPSK; The non-PSK signal includes at least one of the following types: AM, FM, FSK, QAM; The signal simulation parameters include randomly set carrier frequency, symbol rate, signal-to-noise ratio and phase offset.

[0008] Optionally, in the step S2, the spectral correlation density calculation satisfies the following formula: In the formula, is the spectral correlation function, is the center frequency, is the cycle frequency, is the data length, is the discrete sampling value of the input signal, The function is to move the signal to , is the conjugate complex of The function is to move the signal to .

[0009] Optionally, in the step S2, the Hilbert transform satisfies: In the formula, is the in-phase component, equal to the original signal, representing the real part of the signal, is the value of the original signal at time t, is the current observation time point, is the quadrature component, which is the Hilbert transform of the original signal, is the value of the original signal at time , the current observation time point.

[0010] Optionally, in the step S3, the feature of the non-PSK signal is represented by a zero vector, and the data set is proportionally divided into a training set and a verification set, and the proportion ranges from 6:4 to 8:2.

[0011] Optionally, in the step S4, the self-attention mechanism neural network comprises three input branches, a feature extraction module, an attention module and an output layer. The three input branches are respectively used for processing the spectral correlation density feature, the I-channel data and the Q-channel data. The feature extraction module is used for sequentially performing convolution, feature superposition, standardization and pooling operations on each input. The attention module adopts a multi-head attention mechanism to weight the features, and performs full connection fusion after Dropout and regularization processing. The output layer is used for convolution compression and full connection classification of the fused features, and outputs a modulation type probability distribution.

[0012] Optionally, the input size of the I-channel and the Q-channel data is 1× , and the input size of the spectral correlation density feature is 1× , wherein, and are predetermined lengths; and the final output dimension of the network is consistent with the type of signal modulation.

[0013] Optionally, in the step S5, the cyclic autocorrelation blind search algorithm specifically comprises: Step S5-1: setting a cyclic frequency candidate value to scan a candidate cyclic frequency range with a fixed step size; Step S5-2: calculating a cyclic autocorrelation amplitude corresponding to each candidate frequency; Step S5-3: selecting a cyclic frequency corresponding to the maximum amplitude as the main cyclic frequency; wherein the cyclic autocorrelation formula is: In the formula, is a cyclic autocorrelation function, which describes the correlation of a signal at a cyclic frequency and a time delay , is the length of a sampling sequence, is the value of the signal at , is the current observation position, is a delay parameter, is the value of the signal at , is a rotation factor, which multiplies the signal by a periodic complex weight in different time windows to capture the Frequency-dependent components, is the sampling rate, is the cycle frequency.

[0014] According to the second aspect of the present invention, a computer device is also provided, comprising 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 steps of the PSK signal intelligent modulation identification method based on spectral correlation density described in any one of the technical solutions in the first aspect of the present invention can be implemented.

[0015] According to the third aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it can implement the steps of the PSK signal intelligent modulation identification method based on spectral correlation density described in any technical solution in the first aspect of the present invention.

[0016] Beneficial effects: 1. Through the above technical solution, first, the method of the present invention extracts high-resolution features of PSK signals, including their frequency domain and phase information, through spectral correlation density calculation and Hilbert transform. This method not only considers traditional time-frequency analysis (such as short-time Fourier transform) but also deeply explores the details in the IQ data, providing a more comprehensive signal description. In particular, the application of the self-attention mechanism allows the model to automatically adjust weights based on different parts of the input signal, enhancing the learning of key features. For example, in a low signal-to-noise ratio environment, the network can automatically increase its attention to strongly cyclic frequency components; while under high signal-to-noise ratio conditions, it may place greater emphasis on phase changes in the IQ data. Therefore, the method of the present invention can significantly improve the accuracy and robustness of PSK modulation recognition, enabling it to maintain high performance even in complex electromagnetic environments.

[0017] Second, unlike traditional methods that rely on fixed feature engineering, the method of the present invention adopts a data-driven framework based on deep learning to capture the essential characteristics of PSK signals by learning from a large amount of simulation data. This method is not limited to a specific type of known signal, but can adapt to and identify new or unknown modulation methods. In addition, when using the trained model to analyze the predicted samples, the stability of the final recognition is improved by combining the prediction results of multiple samples and adopting strategies such as voting algorithms. This makes the present invention not only suitable for ideal signal recognition in laboratory environments, but also can show good adaptability and reliability when facing variable electromagnetic interference and new signal modulation methods in practical applications.

[0018] Third, the self-attention mechanism neural network designed in the present invention exhibits excellent anti-noise and anti-interference properties. This mechanism enables the network to intelligently focus on the most relevant features in the signal, while ignoring those that may be noise or irrelevant information. This selective feature attention not only improves the efficiency of feature extraction, but also enhances the model's ability to handle complex nonlinear relationships. During the training process, the network parameters are continuously updated and optimized through the optimizer's backpropagation algorithm to ensure that the model is always in the best state. Therefore, even in the presence of severe noise and interference, the model proposed in the present invention can still provide accurate and reliable modulation recognition results, greatly improving the stability and practicality of the entire system. In addition, the cyclic frequency of the signal is determined by the cyclic autocorrelation blind search algorithm, which further enhances the accurate capture of signal features and lays a solid foundation for subsequent classification tasks.

[0019] 2. Other beneficial effects or advantages of the present invention will be described in detail in the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments. Obviously, the 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 paying any creative labor.

[0021] in: Figure 1 1 is a flowchart of a method for intelligent modulation identification of a PSK signal based on spectral correlation density provided by an exemplary embodiment of the present invention; Figure 2 1 is a schematic diagram of the overall process of a PSK signal intelligent modulation identification method based on spectral correlation density provided by an exemplary embodiment of the present invention; Figure 3 is a schematic diagram of a process for constructing a data set in model training provided by an exemplary embodiment of the present invention; Figure 4 1 is a schematic diagram of a process for determining a PSK signal for an unknown signal provided by an exemplary embodiment of the present invention; Figure 5 This is a schematic diagram of the PSK prediction network structure based on the deep learning attention mechanism provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0023] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0024] In addition, the terms "including" and "having" and any variations thereof mentioned in the description of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices. It should also be noted that in the embodiments of the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way.

[0025] In order to facilitate relevant technical personnel to have a clearer and more accurate understanding of the technical solution of the present invention, the existing related technologies are further explained below.

[0026] Traditional PSK signal modulation recognition methods mainly rely on time-frequency analysis and feature extraction techniques. These methods have shown some effectiveness when processing PSK signals in simple scenarios. However, they face significant challenges in complex electromagnetic environments.

[0027] First, commonly used time-frequency analysis tools (such as short-time Fourier transform and wavelet transform) are limited by fixed window resolution, making it difficult to simultaneously consider the time and frequency domain characteristics of high-frequency transition signals. This means that in highly dynamic signal environments, traditional methods may lose key information, resulting in reduced recognition performance.

[0028] Secondly, manually designed feature extraction methods rely heavily on expert experience, which not only increases the time cost of algorithm development but also limits their generalization capabilities. For example, certain designed features may only be applicable to a certain type of PSK signal and perform poorly when faced with new or unknown signal variants.

[0029] Furthermore, under low signal-to-noise ratio conditions, the recognition accuracy of traditional algorithms significantly decreases, failing to meet the high standards required in practical applications. Therefore, despite the widespread use of these methods over the past few decades, with the advancement of communication technology and the increasing complexity of the electromagnetic environment, existing modulation recognition methods are no longer able to meet the needs of modern communication systems.

[0030] The rapid development of deep learning technology in recent years has provided a new solution for PSK signal modulation recognition. Through end-to-end training, deep neural networks can automatically learn the signal's essential laws directly from the raw signal or pre-processed features, eliminating the need for manual feature engineering.

[0031] Specifically, convolutional neural networks can automatically extract local features of signals and are suitable for capturing the spectral characteristics of PSK signals; long short-term memory networks are good at processing sequence data and can effectively capture the temporal correlation of signals; and the attention mechanism can dynamically focus on the key parts of the signal during the processing process, thereby improving recognition accuracy.

[0032] Research shows that the PSK signal recognition method based on deep learning can still maintain a high accuracy in a low signal-to-noise ratio environment, which is significantly improved compared to traditional methods.

[0033] Furthermore, the application of data augmentation techniques and transfer learning further reduces the model's need for large amounts of labeled data, improving its practicality. These advantages make deep learning a powerful tool for solving the problem of PSK recognition in complex electromagnetic environments, showing great potential in application scenarios such as communications reconnaissance and spectrum monitoring.

[0034] Although deep learning has brought new hope for PSK signal modulation recognition, there are still many challenges in combining deep learning with existing technologies in practical applications.

[0035] First, the non-stationary nature of PSK signals means that the feature matrix generated by traditional time-frequency analysis may suffer from information loss, making it difficult to meet the high-discrimination feature requirements of deep learning. To overcome this problem, researchers need to explore more effective feature extraction methods, such as spectral correlation density calculation, to provide more representative input data.

[0036] Secondly, in real-world applications, the dynamic changes in signal parameters (such as carrier frequency and symbol rate) require the model to have strong generalization capabilities. This means that the model must not only be able to recognize PSK signals under standard conditions, but also adapt to various complex real-world situations.

[0037] In addition, special scenarios such as military operations place stringent demands on the real-time and lightweight nature of algorithms, and existing complex deep learning models are often difficult to deploy on edge devices.

[0038] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0039] like Figure 1 As shown, this embodiment provides a PSK signal intelligent modulation identification method based on spectral correlation density according to the first aspect of the present invention, including the following steps: Step S1: Simulate a variety of PSK signals and non-PSK signals to generate signal samples containing random parameters; perform sliding window clipping on the signals and discard samples whose length is less than a predetermined window length; Step S2: Calculate the spectral correlation density of the clipped signal to obtain spectral correlation density features; and perform Hilbert transform on the signal to extract I-channel data and Q-channel data. Step S3: Normalize the spectrum correlation density features, I-channel data, and Q-channel data respectively, establish a mapping relationship with the signal modulation type, construct a data set, and divide it into a training set and a validation set; Step S4: Input the preprocessed data into the self-attention mechanism neural network, optimize the network parameters through back propagation until the loss function converges, and generate a trained deep learning model; Step S5: Determine the cyclic frequency of the prediction signal through a cyclic autocorrelation blind search algorithm; calculate the spectral correlation density characteristics, I-channel data and Q-channel data after clipping the prediction signal, and input them into the trained model after normalization; Step S6: Output the probability distribution of multiple prediction samples, and take the result corresponding to the highest probability as the final modulation type recognition result.

[0040] Through the above technical solution, first, the method of the present invention extracts high-resolution features of PSK signals, including their frequency domain and phase information, through spectral correlation density calculation and Hilbert transform. This method not only considers traditional time-frequency analysis (such as short-time Fourier transform) but also deeply explores the details in the IQ data, providing a more comprehensive signal description. In particular, the application of the self-attention mechanism allows the model to automatically adjust weights based on different parts of the input signal, enhancing the learning of key features. For example, in low signal-to-noise ratio environments, the network can automatically increase attention to strongly cyclic frequency components; while in high signal-to-noise ratio conditions, it may place greater emphasis on phase changes in the IQ data. Therefore, the method of the present invention can significantly improve the accuracy and robustness of PSK modulation recognition, enabling it to maintain high performance even in complex electromagnetic environments.

[0041] Secondly, unlike traditional methods that rely on fixed feature engineering, the method of the present application adopts a deep learning-based data-driven framework to capture the essential features of PSK signals through learning from a large amount of simulation data. This method is not limited to specific types of known signals, but can adapt to and identify new or unknown modulation methods. In addition, when analyzing the prediction samples using the trained model, the prediction results of multiple samples are combined, and strategies such as voting algorithms are used to improve the stability of the final identification. This makes the present application not only suitable for ideal signal identification in laboratory environments, but also able to exhibit good adaptability and reliability in the face of changing electromagnetic interference and new signal modulation methods in actual applications.

[0042] Thirdly, the self-attention mechanism neural network designed in the present application exhibits excellent anti-noise and anti-interference characteristics. This mechanism enables the network to intelligently focus on the most relevant features in the signal while ignoring those that may be noise or irrelevant. This selective feature attention not only improves the efficiency of feature extraction, but also enhances the model's ability to handle complex nonlinear relationships. During training, the network parameters are continuously updated and optimized through the backpropagation algorithm of the optimizer, ensuring that the model is always in the best state. Therefore, even in the presence of severe noise and interference, the model proposed in the present application can still provide accurate and reliable modulation identification results, greatly improving the stability and practicality of the entire system. In addition, the cyclic frequency of the signal is determined through the cyclic autocorrelation blind search algorithm, further enhancing the accurate capture of signal features and laying a solid foundation for subsequent classification tasks.

[0043] The technical solutions of the present application will be described below in conjunction with an exemplary embodiment.

[0044] In an exemplary embodiment, the present application provides a PSK signal intelligent modulation recognition method based on spectral correlation density, the overall process of which can be referred to Figure 2 , including the following specific steps: Step one, use signal simulation software to simulate PSK signals, including BPSK, QPSK, 8PSK, 16PSK, / 4-QPSK, as well as AM, FM, FSK, and QAM signals. Generate multiple signals by setting carrier frequency, symbol rate, signal-to-noise ratio, phase offset, and other parameters in a random parameter space to ensure data diversity; Step two, slide window clipping is performed on the signal, with a sliding step of 4096 and a sliding window length of 8192. Samples with a length less than 8192 are discarded. Step 3: Calculate the spectrum correlation density of the simulation signal, set the frequency expansion length to 2048, obtain the real and imaginary part results of the spectrum correlation density with a size of 2048, and calculate the absolute value of the real and imaginary part results to obtain the spectrum correlation density calculation with a size of 2048. The calculation formula is: in, is the spectral correlation function, is the center frequency, is the cycle frequency, is the data length, is the discrete sampling value of the input signal, The function is to make the signal move, for The complex conjugate of The function is to make the signal move.

[0045] Step 4: Perform Hilbert transform on the signal to obtain I-channel data and Q-channel data with a size of 8192. The calculation formula is: , Where, is the in-phase component, equal to the original signal, representing the real part of the signal, is the value of the original signal at time t, is the current observation time point, is the orthogonal component, is the Hilbert transform of the original signal, is the original signal at time The value of is the current observation time point.

[0046] Step 5: Normalize the maximum and minimum values ​​of the spectrum correlation density calculation results and the signal IQ path data respectively; Step 6: Establish a correspondence between the spectral correlation density calculation results and the IQ data, and use the one-hot method to construct a dataset based on the corresponding signal type. The label of non-PSK signals is all zero, and the dataset is divided into a training set and a validation set in a ratio of 7:3; Step 7: Combine the processed data into a training batch and feed it into the designed self-attention mechanism neural network for learning. Backpropagation is performed through the optimizer to update the parameters in the network. When the loss function converges, the deep learning model file is obtained and the model training stops. Step 8: Use the cyclic autocorrelation blind search algorithm to obtain the cyclic frequency of the signal. The cyclic autocorrelation formula is: where, is the cyclic autocorrelation function, describes the signal at the cyclic frequency with time delay , is the length of the sampled sequence, is the value of the signal at , is the current observation position, is the delay parameter, is the value of the signal at , is the rotation factor, this factor will multiply the signal with a periodic complex weight on different time windows, used to capture the components in the signal related to frequency, is the sampling rate, is the cyclic frequency, usually the frequency of a certain periodic component of the signal.

[0047] Step nine, after uniform clipping processing of the predicted signal, the spectral correlation features and IQ data are obtained through Hilbert transform, and the spectral correlation density is calculated using the calculated cyclic frequency, and the maximum and minimum value normalization operations are performed on the calculation results respectively; Step ten, using the designed self-attention mechanism neural network and the trained model weight file to predict all signal prediction samples, the prediction result with the highest probability value is the prediction result of modulation recognition. In this embodiment, it needs to be explained that, please refer to Figure 2 , the method of the application first uses signal simulation software to simulate PSK related signals, and then clips the simulation signals; the clipped data is subjected to spectral correlation density calculation and Hilbert transform to obtain the spectral correlation features and IQ data of the signals; after normalization and uniform clipping operations are respectively performed, a mapping relationship is established to construct a data set, and the data set is divided into a training set and a verification set according to a certain proportion; the processed data is combined as a training batch and sent into the designed self-attention mechanism neural network for learning, the optimizer is used for back propagation, and the parameters in the network are updated, and when the loss function converges, the deep learning model file can be obtained, and the model training stops; the cyclic frequency is obtained by using the cyclic autocorrelation blind search algorithm for the predicted signal, the spectral correlation features and IQ data of the signal are obtained by spectral correlation density calculation and Hilbert transform after the prediction data are clipped; after normalization and relationship mapping, a plurality of prediction samples are obtained; finally, the designed self-attention mechanism neural network and the trained model weight file are used to predict a plurality of signal prediction samples, a plurality of prediction results are obtained, and the prediction result with the highest probability value is the prediction result of modulation recognition.

[0048] Wherein, the method for calculating the cycle frequency by the cyclic autocorrelation blind search algorithm is as follows: 1. Set the cycle frequency candidate value, and scan from 0 to half of the sampling frequency by 1 kHz step; 2. Calculate the cyclic autocorrelation result of the signal for each candidate cycle frequency; 3. Extract the maximum amplitude of the cyclic autocorrelation, and store the current cycle frequency and the corresponding maximum amplitude; 4. Find the cycle frequency with the maximum amplitude among all candidate cycle frequencies, which is the main cycle frequency of the signal.

[0049] Wherein, the data set construction process of the application is as shown in Figure 3 The spectrum correlation density calculation and Hilbert transform are performed on the simulation signal after uniform cutting operation, the spectrum correlation features and IQ road data are obtained, normalization and label relationship mapping are performed, and the data set is constructed.

[0050] The PSK signal discrimination process of the unknown signal of the application is as shown in Figure 4 The cycle frequency is obtained by the cyclic autocorrelation blind search algorithm, the spectrum correlation features and IQ road data of the signal are obtained by spectrum correlation density calculation and Hilbert transform after cutting the prediction data, normalization and relationship mapping are performed to obtain multiple prediction samples, and finally the self-attention mechanism neural network and the trained model weight file are used to predict the multiple signal prediction samples to obtain multiple prediction results, and the prediction result with the highest probability value is the prediction result of modulation recognition.

[0051] The PSK prediction network based on deep learning attention mechanism designed by the application is shown in Figure 5 The network structure is a three-input single-output network structure, and the feature extraction module and the attention module are constructed and used in the network structure, wherein the feature extraction module first uses the convolution kernel to extract the features of the input feature matrix, the extraction result and the input data are stacked and standardized, and finally the pooling processing is performed; the attention module uses the multi-head attention module to extract the features of the input feature matrix, the feature extraction result and the input feature are stacked, respectively using dropout and regularization for generalization processing, and the result is processed by twice full connection and dropout, and then stacked and processed by regularization method for secondary generalization.

[0052] The input of the network is respectively I path data, Q path data, spectrum correlation density three kinds of data, the input format of the IQ data is 1*8192, and the input format of the spectrum correlation density is 1*2048; the network first uses the feature extraction modules with 512, 512, 256 and 128 convolution kernels to perform feature extraction processing on the IQ data in turn, and obtains two feature matrices with the size of 128*512; the spectrum correlation density feature is extracted by using the feature extraction modules with 512 and 128 convolution kernels in turn, and a feature matrix with the size of 128*512 is obtained; the feature matrices extracted from the three kinds of data are processed by the attention mechanism module and then superimposed, and a feature matrix with the size of 768*512 is obtained; the fusion feature matrix is processed by using the feature extraction modules with 256, 128 and 64 convolution kernels in turn, then flattened, and then converged by using 512-dimensional full connection, and finally converged by using 5-dimensional full connection.

[0053] According to the second aspect of the present application, a computer device is also provided, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor can implement the steps of the spectrum correlation density-based PSK signal intelligent modulation recognition method in any of the technical solutions of the first aspect of the present application when executing the computer program.

[0054] It can be understood that, in this embodiment, the memory can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a read-only memory, a flash memory, a hard disk or a solid state disk; in addition, the memory can also include a combination of the above-mentioned kinds of memories. The present application does not make specific limitation on this.

[0055] Similarly, the processor can be used to implement or execute the various exemplary logical steps described in combination with the disclosure of the present application. The processor can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It can be used to implement or execute the various exemplary logical steps described in combination with the disclosure of the present application. The processor can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0056] According to the third aspect of the present application, a computer readable storage medium is also provided, which stores a computer program, and the computer program is characterized in that the computer program can implement the steps of the spectrum correlation density-based PSK signal intelligent modulation recognition method in any of the technical solutions of the first aspect of the present application when executed by a processor.

[0057] In this embodiment, the computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a register, a hard disk, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an application-specific integrated circuit (ASIC). In embodiments of the present invention, computer-readable storage media may be any tangible media that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0058] The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A PSK signal intelligent modulation recognition method based on spectral correlation density, characterized in that: The steps include: Step S1: Simulate a variety of PSK signals and non-PSK signals to generate signal samples containing random parameters; perform sliding window clipping on the signals and discard samples whose length is less than a predetermined window length; Step S2: Calculate the spectral correlation density of the clipped signal to obtain spectral correlation density features; and perform Hilbert transform on the signal to extract I-channel data and Q-channel data. Step S3: Normalize the spectrum correlation density features, I-channel data, and Q-channel data respectively, establish a mapping relationship with the signal modulation type, construct a data set, and divide it into a training set and a validation set; Step S4: Input the preprocessed data into the self-attention mechanism neural network, optimize the network parameters through back propagation until the loss function converges, and generate a trained deep learning model; Step S5: Determine the cyclic frequency of the prediction signal through a cyclic autocorrelation blind search algorithm; calculate the spectral correlation density characteristics, I-channel data and Q-channel data after clipping the prediction signal, and input them into the trained model after normalization; Step S6: Output the probability distribution of multiple prediction samples, and take the result corresponding to the highest probability as the final modulation type recognition result.

2. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 1 is characterized in that: In the step S1, The PSK signal includes at least one of the following types: BPSK, QPSK, 8PSK, 16PSK, -QPSK; The non-PSK signal includes at least one of the following types: AM, FM, FSK, QAM; Signal simulation parameters include randomly set carrier frequency, symbol rate, signal-to-noise ratio, and phase offset.

3. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 1 is characterized in that: In step S2, the spectrum correlation density calculation satisfies the following formula: Where, is the spectral correlation function, is the center frequency, is the cycle frequency, is the data length, is the discrete sampling value of the input signal, The function is to make the signal move, for The complex conjugate of The function is to make the signal move.

4. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 1 is characterized in that: In step S2, the Hilbert transform satisfies: Where, is the in-phase component, equal to the original signal, representing the real part of the signal, is the value of the original signal at time t, is the current observation time point, is the orthogonal component, is the Hilbert transform of the original signal, is the original signal at time The value of is the current observation time point.

5. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 1 is characterized in that: In step S3, the features of the non-PSK signal are represented by an all-zero vector, and the data set is divided into a training set and a validation set in a ratio ranging from 6:4 to 8:

2.

6. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 1 is characterized in that: In step S4, the self-attention mechanism neural network includes three input branches, a feature extraction module, an attention module and an output layer; The three input branches are used to process spectrum-related density features, I-channel data, and Q-channel data, respectively; The feature extraction module is used to perform convolution, feature superposition, normalization and pooling operations on each input in sequence; The attention module uses a multi-head attention mechanism to weight features and performs full-connection fusion after Dropout and regularization processing; The output layer is used to perform convolution compression and full connection classification on the fused features, and output the modulation type probability distribution.

7. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 6, characterized in that: The I-channel and Q-channel data input sizes are both 1× , the spectral correlation density feature input size is 1× ,in, and is a predetermined length; and the final output dimension of the network is consistent with the signal modulation type.

8. The PSK signal intelligent modulation identification method based on spectral correlation density according to claim 1 is characterized in that: In step S5, the cyclic autocorrelation blind search algorithm specifically includes: Step S5-1: Setting candidate cyclic frequency values ​​and scanning the candidate cyclic frequency range with a fixed step size; Step S5-2: Calculate the cyclic autocorrelation amplitude corresponding to each candidate frequency; Step S5-3: Select the cyclic frequency corresponding to the maximum amplitude as the main cyclic frequency; Among them, the cyclic autocorrelation formula is: Where, is the cyclic autocorrelation function, which describes the signal In the cycle frequency Next, the time delay is The correlation between is the length of the sampling sequence, For the signal The value of is the current observation position, is the delay parameter, For the signal The value of is a rotation factor, which multiplies the signal by a periodic complex weight in different time windows to capture the Frequency-dependent components, is the sampling rate, is the cycle frequency.

9. A computer device comprising 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 steps of the PSK signal intelligent modulation identification method based on spectral correlation density according to any one of claims 1 to 8 can be implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the PSK signal intelligent modulation identification method based on spectral correlation density according to any one of claims 1 to 8 can be implemented.

Images