A method, apparatus, device and medium for identifying a radar signal modulation type
By using Hilbert transform and CNN-LSTM cascaded models to extract features and calculate probabilities from radar signals, the problem of low efficiency and accuracy in existing radar signal modulation type identification is solved, and efficient identification in complex environments is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 成都华日通讯技术股份有限公司
- Filing Date
- 2026-02-28
- Publication Date
- 2026-06-19
AI Technical Summary
Existing radar signal modulation type identification methods suffer from low identification efficiency and accuracy, especially with a sharp decline in performance under low signal-to-noise ratio environments. Furthermore, fully connected neural networks lack the ability to model the local structure and temporal correlation of signals.
The Hilbert transform is used to separate the real and imaginary features of the signal. The instantaneous phase and frequency are obtained by differentiation. The CNN-LSTM cascaded model is combined for feature extraction and probability calculation. The weight matrix is used to map the high-dimensional feature vector to the category space to achieve fast recognition.
It improves the efficiency and accuracy of radar signal modulation type identification, solves the problems of identification accuracy and robustness of existing methods in complex electromagnetic environments, and achieves accurate identification of signal features.
Smart Images
Figure CN122239013A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of signal recognition technology, and in particular to a method, apparatus, device and medium for identifying radar signal modulation types. Background Technology
[0002] Intra-pulse modulation identification of radar signals is a core component of electronic reconnaissance and is crucial for target identification and threat analysis. Different types of radars, such as search radars, fire control radars, and guidance radars, as well as different combat missions, such as reconnaissance, tracking, and attack, will employ drastically different intra-pulse modulation methods. Therefore, by identifying the modulation type, it is possible to decode the radar's functional intent and target attributes for subsequent targeted processing.
[0003] Existing signal modulation type identification methods rely on time-frequency analysis (such as WVD and SPWVD) and manual feature extraction (such as instantaneous frequency and phase difference statistics), which have problems such as computational complexity, feature design relying on prior knowledge, and performance degradation in low signal-to-noise ratio environments. While neural network-based identification methods, such as using fully connected neural networks (BP networks) to classify the instantaneous frequency curves of signals, simplify feature engineering, the fully connected networks used have insufficient ability to model the local structure and time-series correlation of the input data, which limits their identification accuracy and robustness in complex electromagnetic environments.
[0004] Therefore, existing signal modulation type identification methods suffer from low identification efficiency and accuracy. Summary of the Invention
[0005] This application provides a method, apparatus, device, and medium for identifying radar signal modulation types, in order to solve the problems of low identification efficiency and accuracy in existing signal modulation type identification methods.
[0006] In a first aspect, this application provides a method for identifying radar signal modulation types, the method comprising: Perform a Hilbert transform on the target signal to obtain the instantaneous phase of the target, and then differentiate the instantaneous phase of the target to obtain the instantaneous frequency; The instantaneous frequency is discretized and normalized to obtain the corresponding target frequency sequence. Based on the target CNN-LSTM cascaded model, the target frequency sequence is feature extracted to obtain the corresponding feature vector. Based on the weight matrix and eigenvector, the corresponding target probability vector is determined, and based on the vector data of each dimension in the target probability vector, the probability values corresponding to multiple preset modulation types are determined respectively. Based on multiple probability values, determine the target probability value and its corresponding target modulation type, and output the target modulation type and target probability value.
[0007] In some embodiments of this application, performing a Hilbert transform on the target signal to obtain the instantaneous phase of the target includes: Signal acquisition is performed based on the target sampling rate and bandwidth, and the acquired signal is extracted according to the preset amplitude threshold to obtain a time-domain complex signal. Based on high-pass filtering and adaptive noise suppression algorithms, the DC component and noise in the time-domain complex signal are removed to obtain the target signal. Then, Hilbert transform is performed on the target signal to obtain the instantaneous phase of the target.
[0008] In some embodiments of this application, the instantaneous phase of the target is obtained by performing a Hilbert transform based on the target signal, including: Perform a Hilbert transform on the target signal to obtain the signal to be analyzed and convert it into the corresponding polar coordinate form; For the signal to be analyzed in polar coordinates, the corresponding instantaneous phase of the target is determined based on the arctangent model.
[0009] In some embodiments of this application, feature extraction is performed on the target frequency sequence based on the target CNN-LSTM cascaded model to obtain the corresponding feature vector, including: Multiple radar signal samples are acquired and their corresponding modulation type labels are determined according to their respective modulation types to obtain the model training set; Based on the target frequency sequence and modulation type label corresponding to each radar signal sample, the initial CNN-LSTM cascaded model is trained to obtain the target CNN-LSTM cascaded model. Based on the target CNN-LSTM cascaded model, feature extraction is performed on the target frequency sequence to obtain the feature vector.
[0010] In some embodiments of this application, feature vectors are extracted from the target frequency sequence based on the target CNN-LSTM cascaded model, including: Based on the CNN network layers in the target CNN-LSTM cascade model, local structured features of the target frequency sequence are extracted to obtain the corresponding feature sequence. Then, based on the LSTM network layers, the temporal dependencies of the feature sequence are determined to obtain the corresponding feature vector.
[0011] In some embodiments of this application, the corresponding target probability vector is determined based on the weight matrix and the eigenvector, including: Based on the fully connected layers in the target CNN-LSTM cascade model, the weight matrix is determined, and the target probability vector is determined by the product of the weight matrix and the feature vector.
[0012] In some embodiments of this application, determining a target probability value and its corresponding target modulation type based on multiple probability values includes: By comparing the magnitudes of each probability value, the target probability value with the largest value is obtained, and the target modulation type is determined based on the modulation type corresponding to the target probability value.
[0013] Secondly, this application provides a radar signal modulation type identification device, the device comprising: The differentiation module is used to perform Hilbert transform on the target signal to obtain the instantaneous phase of the target, and to differentiate the instantaneous phase of the target to obtain the instantaneous frequency; The extraction module is used to discretize and normalize the instantaneous frequency to obtain the corresponding target frequency sequence, and to extract features from the target frequency sequence based on the target CNN-LSTM cascaded model to obtain the corresponding feature vector. The determination module is used to determine the corresponding target probability vector based on the weight matrix and feature vector, and to determine the probability values corresponding to multiple preset modulation types based on the vector data of each dimension in the target probability vector. The output module is used to determine the target probability value and its corresponding target modulation type based on multiple probability values, and output the target modulation type and target probability value.
[0014] Thirdly, this application provides a computer device, including: a processor, and a memory communicatively connected to the processor; The memory stores instructions that the computer executes; The processor executes computer execution instructions stored in memory to implement the method of this application.
[0015] Fourthly, this application provides a computer-readable storage medium storing program code, which, when executed by a processor, is used to implement the method of this application.
[0016] Compared with existing technologies, the method in this application separates the real and imaginary parts of the target signal by performing a Hilbert transform, which facilitates the subsequent calculation of the instantaneous phase. By differentiating the instantaneous phase, the dynamic change of the signal frequency over time is captured, thus determining the instantaneous frequency features that characterize the essence of intra-pulse modulation and solving the problem of accurately depicting the temporal frequency characteristics of signals. Furthermore, discretizing the instantaneous frequency transforms continuous frequency changes into regular sequence data, eliminating data scale interference caused by differences in different signal frequency ranges and avoiding recognition bias due to uneven data distribution. Feature extraction of the target frequency sequence is performed based on a target CNN-LSTM cascaded model. The CNN module can determine the local structured features in the sequence, while the LSTM module can model the long-term temporal dependencies of frequency changes. The cascaded effect of these two modules achieves accurate recognition of signal features, solving the problem of insufficient utilization of local and temporal features in existing fully connected networks. By combining the weight matrix and feature vector, high-dimensional features are mapped to target probability vectors, thereby quickly determining the probability values corresponding to each preset modulation type, improving the efficiency and accuracy of signal modulation type recognition. Attached Figure Description
[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0018] Figure 1 A flowchart illustrating a radar signal modulation type identification method provided in an embodiment of this application; Figure 2 This is a schematic diagram of a radar signal modulation type identification method provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a radar signal modulation type identification device provided in an embodiment of this application; Figure 4 This is a structural block diagram of an apparatus for performing a radar signal modulation type identification method according to an embodiment of this application. Detailed Implementation
[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0020] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0021] Figure 1 This is a flowchart illustrating a radar signal modulation type identification method provided in an embodiment of this application. Figure 1 As shown, the method for identifying this type of radar signal modulation may include the following steps: S110. Perform a Hilbert transform on the target signal to obtain the instantaneous phase of the target, and differentiate the instantaneous phase of the target to obtain the instantaneous frequency.
[0022] The target signal refers to the effective intra-pulse IQ signal obtained after preprocessing the acquired raw signal, removing the DC component and noise. The raw IQ data contains background noise and DC component, while the target signal is obtained through preprocessing operations such as filtering and threshold extraction, which can eliminate invalid interference.
[0023] The Hilbert transform is used to convert real signals into analytic signals, that is, to convert real signals into complex analytic signals; by causing a -90° phase shift in each frequency component of the signal, the instantaneous amplitude, frequency, and phase information of the signal can be extracted.
[0024] The instantaneous phase of the target is the phase angle of the analytical signal at each moment, which can also be understood as the phase state of the signal at each moment. Its change over time is directly equivalent to the essential law of intra-pulse modulation. The differences between different intra-pulse modulation types will ultimately be reflected in the change curve of the instantaneous phase of the target.
[0025] Based on this, the target signal is obtained by preprocessing the original signal and then Hilbert transform is performed to obtain the corresponding instantaneous phase of the target. Since the instantaneous phase of the target can characterize the signal characteristics of different intra-pulse modulation types, the instantaneous frequency is obtained by differentiating the instantaneous phase of the target, so that the modulation type of the target signal can be determined based on the instantaneous frequency.
[0026] S120. Discretize and normalize the instantaneous frequency to obtain the corresponding target frequency sequence. Based on the target CNN-LSTM cascaded model, extract features from the target frequency sequence to obtain the corresponding feature vector.
[0027] The target frequency sequence is a one-dimensional numerical sequence that represents the instantaneous frequency temporal change of a pulse signal after discretization, sampling, and normalization. The input requirement for the CNN-LSTM cascaded model is a structured, dimensionless, and temporally ordered numerical sequence, while the target frequency sequence is a one-dimensional fixed-length sequence, which meets the CNN's requirement for input dimension. After normalization, all elements are in the [0,1] interval, avoiding gradient explosion or vanishing during training and improving the model's convergence speed.
[0028] The target CNN-LSTM cascaded model is a pre-trained deep learning model designed for instantaneous frequency sequences of radar pulses. Through the cascaded structure of CNN layers and LSTM layers, it achieves end-to-end processing of automatic feature extraction, feature fusion, and probability output. In practical applications, a single CNN is good at local feature extraction but cannot capture long-term temporal dependencies, while a single LSTM is good at temporal modeling but weak in extracting fine local features. By defining the cascaded model, the CNN first extracts local features such as frequency modulation slope and transition points, and the LSTM then models temporal dependencies based on these local features. This adapts to the characteristic that local details of intrapulse modulation determine the type and temporal patterns verify the type, thus improving the accuracy of modulation type identification.
[0029] Based on this, the instantaneous frequency is discretized and normalized to determine the corresponding target frequency sequence, thus obtaining a one-dimensional data sequence; and based on the target CNN-LSTM cascaded model, the target frequency sequence is feature extracted to obtain the corresponding feature vector.
[0030] S130. Based on the weight matrix and eigenvector, determine the corresponding target probability vector, and based on the vector data of each dimension in the target probability vector, determine the probability values corresponding to multiple preset modulation types respectively.
[0031] The weight matrix is a pre-trained parameter matrix in the fully connected layer, used to map high-dimensional feature vectors to the category score space, thereby transforming abstract high-dimensional feature vectors into quantized scores related to specific modulation types. Each weight parameter w_ij represents the importance of the i-th feature in distinguishing the j-th modulation type. If w_ij is a large positive number, it means that the more obvious the i-th feature is, the higher the probability of the j-th modulation type. If w_ij is a small or negative number, it means that the i-th feature has a weak distinguishing effect on the j-th modulation type, or even has an inhibitory effect. The feature vector is an abstract feature of H dimensions (such as 64 dimensions), which cannot directly correspond to M-dimensional modulation types. Since the two dimensions are different, the weight matrix maps the H-dimensional feature vector to the M-dimensional original score vector through matrix multiplication, completing the transformation from the feature space to the category space.
[0032] The target probability vector is a vector obtained by multiplying the weight matrix and the eigenvector. It is a quantitative expression of the probability of various modulation types and intuitively reflects the model's confidence in each modulation type. For example, the vector P = [0.98, 0.015, 0.005], where each element of the vector directly corresponds to the confidence probability of a modulation type. For example, the probability value corresponding to modulation type A is 0.98, modulation type B is 0.015, and modulation type C is 0.00. The higher the score of the category, the greater the corresponding probability value. The sum of all probabilities is 1, which conforms to the physical logic that a signal belongs to only one modulation type.
[0033] The preset modulation type is a pre-determined set of radar pulse modulation types that need to be identified, including linear frequency modulation (LFM), binary frequency shift keying (2FSK), binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), frequency coding (FSK), etc. Each type corresponds to one dimension of the target probability vector, and the probability value corresponding to this modulation type is determined based on the specific value of that dimension.
[0034] The probability value is an element of a certain dimension in the target probability vector, representing the model's confidence quantification result that the input signal belongs to the j-th preset modulation type. The closer the value is to 1, the higher the confidence level. For example, p_j=0.98 indicates that there is a high probability of 98% that the signal belongs to the j-th modulation type. Conversely, the closer the value is to 0, the lower the confidence level. For example, p_j=0.005 indicates that the signal is almost impossible to belong to the j-th modulation type.
[0035] Based on this, the corresponding target probability vector is obtained by multiplying the weight matrix and feature vector of the fully connected layer in the model, and the probability values corresponding to multiple preset modulation types are determined according to the vector data of each dimension in the target probability vector.
[0036] S140. Based on multiple probability values, determine the target probability value and its corresponding target modulation type, and output the target modulation type and target probability value.
[0037] The target probability value is the maximum value of the probability values of all dimensions in the target probability vector. For example, if the vector P=[0.98,0.015,0.005], then the corresponding target probability value is 0.98. The target probability value is the optimal confidence level of the model in that the input signal belongs to a certain type of modulation.
[0038] The target modulation type is a preset modulation type that uniquely corresponds to the target probability value; for example, if vector P=[0.98,0.015,0.005], the target probability value is 0.98, then the modulation type A corresponding to this target probability value is the target modulation type.
[0039] Based on this, by determining the probability value with the largest value among multiple probability values, the target probability value is obtained, and the target modulation type corresponding to the target probability value is further determined, thereby outputting the target modulation type and the target probability value, and realizing the identification of the signal modulation type.
[0040] Based on the feasible implementation of S110 described above, this application further provides a method for performing a Hilbert transform on the target signal to obtain the instantaneous phase of the target, including: Signal acquisition is performed based on the target sampling rate and bandwidth, and the acquired signal is extracted according to the preset amplitude threshold to obtain a time-domain complex signal. Based on high-pass filtering and adaptive noise suppression algorithms, the DC component and noise in the time-domain complex signal are removed to obtain the target signal. Then, Hilbert transform is performed on the target signal to obtain the instantaneous phase of the target.
[0041] The target sampling rate is a pre-set time frequency standard used to collect signals. In practical applications, the target sampling rate must satisfy the Nyquist sampling theorem, that is, the target sampling rate fs must be greater than twice the highest frequency f_max of the signal. Otherwise, aliasing distortion of the signal will occur, that is, high-frequency signals will be misjudged as low-frequency signals and the original information will be lost.
[0042] The preset amplitude threshold is a pre-set threshold value for determining the effective signal segment. In practical applications, when radar signals are acquired, background noise and effective pulse signals are mixed in simultaneously. The core difference between the two lies in their amplitude. Effective pulse signals carry modulation information and usually have a larger amplitude, while background noise is meaningless interference and usually has a smaller amplitude. Therefore, the preset amplitude threshold filters based on amplitude differences. When the acquired signal amplitude is not less than the preset amplitude threshold, it is determined to be an effective signal segment; when the signal amplitude is less than the preset amplitude threshold, it is determined to be noise or invalid interference segment.
[0043] A time-domain complex signal is a time-domain digital signal containing both real and imaginary parts, obtained after being acquired at a target sampling rate and extracted using a preset amplitude threshold.
[0044] Based on this, signal acquisition is performed using the target sampling rate and bandwidth, and the acquired signal is effectively extracted according to the preset amplitude threshold to obtain a time-domain complex signal. Then, the signal is processed using high-pass filtering and adaptive noise suppression algorithms to obtain the target signal.
[0045] Based on the feasible implementation of S110 described above, this application further provides a method for obtaining the instantaneous phase of the target by performing a Hilbert transform on the target signal, including: Perform a Hilbert transform on the target signal to obtain the signal to be analyzed and convert it into the corresponding polar coordinate form; For the signal to be analyzed in polar coordinates, the corresponding instantaneous phase of the target is determined based on the arctangent model.
[0046] The signal to be analyzed is a complex signal constructed by performing a Hilbert transform on the target signal. Since the real and imaginary parts of the target signal are coupled, it is impossible to directly separate the phase state at each moment. The amplitude and phase information of the target signal can be decoupled by the Hilbert transform, thereby obtaining the signal to be analyzed.
[0047] The polar coordinate form is a mathematical expression that converts a complex signal from a real-imaginary form to an amplitude-phase form, so as to calculate the instantaneous frequency. The instantaneous frequency is the derivative of the instantaneous phase with respect to time. The signal to be analyzed separated by the polar coordinate form is a continuous phase-time sequence, and the instantaneous frequency can be obtained directly by differentiation without the need for additional processing of amplitude interference.
[0048] The arctangent model is a mathematical model that uses the arctangent function to solve for the instantaneous phase of the signal to be analyzed, based on the geometric meaning in polar coordinates.
[0049] Based on this, the target signal is subjected to Hilbert transform to obtain the signal to be analyzed, and the signal to be analyzed is further converted into the corresponding polar coordinate form. Then, for the signal to be analyzed in polar coordinate form, the corresponding instantaneous phase of the target is determined based on the arctangent model.
[0050] Based on the feasible implementation of S120 described above, this application further provides a feature extraction method for the target frequency sequence based on a target CNN-LSTM cascaded model to obtain the corresponding feature vector, including: Multiple radar signal samples are acquired and their corresponding modulation type labels are determined according to their respective modulation types to obtain the model training set; Based on the target frequency sequence and modulation type label corresponding to each radar signal sample, the initial CNN-LSTM cascaded model is trained to obtain the target CNN-LSTM cascaded model. Based on the target CNN-LSTM cascaded model, feature extraction is performed on the target frequency sequence to obtain the feature vector.
[0051] Among them, radar signal samples are a collection of sample radar pulse signal data containing specific intra-pulse modulation types and having a clear signal-to-noise ratio. They contain signal characteristics under different modulation types and different noise environments, which are then used by the model to learn the mapping relationship between modulation types and frequency sequence characteristics.
[0052] Modulation type label is an identifier of the true modulation type for each radar signal sample. It serves as a supervisory signal for model training, enabling the model to learn which frequency sequence features correspond to which modulation type by comparing the prediction results with the label.
[0053] The initial CNN-LSTM cascade model is a deep learning network framework designed according to the logic of local feature extraction, temporal dependency modeling, and classification decision-making, without sample training. Its structure is consistent with the final target CNN-LSTM cascade model, except that the parameters are not optimized.
[0054] Model training involves inputting a radar signal sample set into an initial CNN-LSTM cascaded model and optimizing the network parameters through an iterative process of forward propagation, loss calculation, backpropagation, and parameter updates, enabling the model to accurately learn the mapping relationship between the target frequency sequence and modulation type.
[0055] Based on this, in practical applications, a training set is formed by collecting radar pulse samples containing multiple modulation types and multiple signal-to-noise ratios; a standardized instantaneous frequency sequence and its modulation type label are generated for each sample; the gradient descent algorithm is used to optimize the model parameters with the sequence as input and the label as supervision until the model loss function converges.
[0056] Based on the feasible implementation of S120 described above, this application further provides a feature vector extraction method for the target frequency sequence based on a target CNN-LSTM cascaded model, including: Based on the CNN network layers in the target CNN-LSTM cascade model, local structured features of the target frequency sequence are extracted to obtain the corresponding feature sequence. Then, based on the LSTM network layers, the temporal dependencies of the feature sequence are determined to obtain the corresponding feature vector.
[0057] Among them, local structured features are a set of features that reflect the local positional correlation of the target frequency sequence after performing one-dimensional convolution operation on the target frequency sequence through the CNN network layer. The target frequency sequence is the trajectory of the instantaneous frequency change over time. Its local structured features are essentially the combination rules of several consecutive sampling points on the trajectory, corresponding to the local characteristics of intrapulse modulation. Through CNN, local features with constant continuous differences are extracted, which can be used to distinguish different modulation types.
[0058] Temporal dependence refers to the correlation between features at different time positions in a target frequency sequence or its local structured feature sequence. Through the gating mechanism of the LSTM network, including the input gate, forget gate, and output gate, it is modeled as a global temporal context vector and reflected in the feature vector output by the LSTM.
[0059] Based on this, by determining the cascaded model of CNN-LSTM, CNN first extracts local features such as frequency modulation slope and transition points, and LSTM then models temporal dependencies based on these local features. This adapts to the characteristic that local details of intrapulse modulation determine the type and temporal patterns verify the type, thereby improving the accuracy of modulation type recognition.
[0060] Based on the feasible implementation of S130 described above, this application further provides a method for determining the corresponding target probability vector based on the weight matrix and eigenvectors, including: Based on the fully connected layers in the target CNN-LSTM cascade model, the weight matrix is determined, and the target probability vector is determined by the product of the weight matrix and the feature vector.
[0061] The fully connected layer is an important component of the classification decision module in the target CNN-LSTM cascade model. It is a parameterized layer composed of input neurons, output neurons, and fully connected weights between neurons. Since the feature vector output by LSTM is an abstract high-dimensional encoding, each dimension does not directly correspond to a specific modulation type, but only contains local features and temporal dependent fusion information. Based on the weight matrix learned through training by the fully connected layer, these abstract features can be transformed into quantized scores that are directly related to each modulation type.
[0062] Based on this, the target probability vector is determined by the fully connected layer in the target CNN-LSTM cascade model, which learns the weight matrix during training, and the product of the weight matrix and the feature vector.
[0063] Based on the feasible implementation of S140 described above, this application further provides a method for determining a target probability value and its corresponding target modulation type based on multiple probability values, including: By comparing the magnitudes of each probability value, the target probability value with the largest value is obtained, and the target modulation type is determined based on the modulation type corresponding to the target probability value.
[0064] Based on this, the target probability value is obtained by determining the probability value with the largest value among multiple probability values, so as to further determine the target modulation type corresponding to the target probability value.
[0065] Please refer to Figure 2 , Figure 2 A schematic diagram illustrating the framework of a radar signal modulation type identification method provided in this application embodiment; as shown Figure 2 As shown, the original IQ signal is obtained by inputting and merging the IQ signals, thereby acquiring the effective pulse and performing instantaneous frequency calculation and data normalization. The network structure includes an input layer, a one-dimensional convolutional layer, a pooling layer, a flattening layer, an LSTM layer, a fully connected layer, a Softmax layer, and an output layer. Thus, through a defined classification network, namely the target CNN-LSTM cascaded model network, the processed effective signal is identified to obtain the identification result of the corresponding intra-pulse modulation type.
[0066] Based on the above steps, it can be seen that this application separates the real and imaginary parts of the target signal by performing a Hilbert transform, so as to solve the instantaneous phase in the subsequent calculation. By differentiating the instantaneous phase, the dynamic change law of the signal frequency over time is captured, thus realizing the determination of the instantaneous frequency features that characterize the essence of intra-pulse modulation, solving the problem of the difficulty in accurately depicting the temporal frequency characteristics of signals in existing methods. Discretizing the instantaneous frequency transforms continuous frequency changes into regular sequence data, eliminating data scale interference caused by differences in the frequency range of different signals and avoiding recognition bias caused by uneven data distribution. Based on the target CNN-LSTM cascaded model, feature extraction is performed on the target frequency sequence. The CNN module can determine the local structured features in the sequence, and the LSTM module can model the long-term temporal dependence of frequency changes. The cascaded effect of the two achieves accurate recognition of signal features, solving the problem of insufficient utilization of local and temporal features in existing fully connected networks. By combining the weight matrix and feature vector, high-dimensional features are mapped to target probability vectors, thereby quickly determining the probability values corresponding to each preset modulation type, improving the recognition efficiency and accuracy of signal modulation type.
[0067] Figure 3 This is a schematic diagram of a radar signal modulation type identification device provided in an embodiment of this application. Figure 3 As shown, the radar signal modulation type identification device includes: a differentiation module, an extraction module, a determination module, and an output module; wherein: The differentiation module is used to perform Hilbert transform on the target signal to obtain the instantaneous phase of the target, and to differentiate the instantaneous phase of the target to obtain the instantaneous frequency; The extraction module is used to discretize and normalize the instantaneous frequency to obtain the corresponding target frequency sequence, and to extract features from the target frequency sequence based on the target CNN-LSTM cascaded model to obtain the corresponding feature vector. The determination module is used to determine the corresponding target probability vector based on the weight matrix and feature vector, and to determine the probability values corresponding to multiple preset modulation types based on the vector data of each dimension in the target probability vector. The output module is used to determine the target probability value and its corresponding target modulation type based on multiple probability values, and output the target modulation type and target probability value.
[0068] In this embodiment of the application, the differentiation module can also be specifically used for: Signal acquisition is performed based on the target sampling rate and bandwidth, and the acquired signal is extracted according to the preset amplitude threshold to obtain a time-domain complex signal. Based on high-pass filtering and adaptive noise suppression algorithms, the DC component and noise in the time-domain complex signal are removed to obtain the target signal. Then, Hilbert transform is performed on the target signal to obtain the instantaneous phase of the target.
[0069] In this embodiment of the application, the differentiation module can also be specifically used for: Perform a Hilbert transform on the target signal to obtain the signal to be analyzed and convert it into the corresponding polar coordinate form; For the signal to be analyzed in polar coordinates, the corresponding instantaneous phase of the target is determined based on the arctangent model.
[0070] In this embodiment of the application, the extraction module can also be specifically used for: Multiple radar signal samples are acquired and their corresponding modulation type labels are determined according to their respective modulation types to obtain the model training set; Based on the target frequency sequence and modulation type label corresponding to each radar signal sample, the initial CNN-LSTM cascaded model is trained to obtain the target CNN-LSTM cascaded model. Based on the target CNN-LSTM cascaded model, feature extraction is performed on the target frequency sequence to obtain the feature vector.
[0071] In this embodiment of the application, the extraction module can also be specifically used for: Based on the CNN network layers in the target CNN-LSTM cascade model, local structured features of the target frequency sequence are extracted to obtain the corresponding feature sequence. Then, based on the LSTM network layers, the temporal dependencies of the feature sequence are determined to obtain the corresponding feature vector.
[0072] In this embodiment of the application, the determining module can also be specifically used for: Based on the fully connected layers in the target CNN-LSTM cascade model, the weight matrix is determined, and the target probability vector is determined by the product of the weight matrix and the feature vector.
[0073] In this embodiment of the application, the output module can also be specifically used for: By comparing the magnitudes of each probability value, the target probability value with the largest value is obtained, and the target modulation type is determined based on the modulation type corresponding to the target probability value.
[0074] Figure 4 This is a schematic diagram of the structure of an apparatus for performing a radar signal modulation type identification method according to an embodiment of this application. Figure 4 As shown, the device includes: The device may include one or more processors with processing cores, one or more computer-readable storage media such as memory, communication components, etc. The processor, memory, and communication components are connected via a bus.
[0075] In the specific implementation process, at least one processor executes computer execution instructions stored in memory, causing at least one processor to execute the radar signal modulation type identification method described above.
[0076] The specific implementation process of the processor can be found in the above method embodiments, and its implementation principle and technical effect are similar, so it will not be repeated here.
[0077] Furthermore, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.
[0078] The memory may include Random Access Memory (RAM) and may also include Non-volatile Memory (NVM), such as at least one disk storage device.
[0079] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.
[0080] In some embodiments, a computer program product is also provided, comprising a computer program or instructions that, when executed by a processor, implement the steps in any of the radar signal modulation type identification methods described above.
[0081] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0082] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0083] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of program codes, which can be loaded by a processor to execute the steps in any of the radar signal modulation type identification methods provided in embodiments of this application.
[0084] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0085] According to one aspect of this application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium.
[0086] Since the instructions stored in the storage medium can execute the steps in any of the radar signal modulation type identification methods provided in the embodiments of this application, the beneficial effects that any of the radar signal modulation type identification methods provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.
[0087] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope of this application is indicated by the appended claims.
[0088] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.
Claims
1. A method for identifying radar signal modulation types, characterized in that, The method includes: Perform a Hilbert transform on the target signal to obtain the instantaneous phase of the target, and then differentiate the instantaneous phase of the target to obtain the instantaneous frequency; The instantaneous frequency is discretized and normalized to obtain the corresponding target frequency sequence. Based on the target CNN-LSTM cascaded model, the target frequency sequence is feature extracted to obtain the corresponding feature vector. Based on the weight matrix and the feature vector, the corresponding target probability vector is determined, and based on the vector data of each dimension in the target probability vector, the probability values corresponding to multiple preset modulation types are determined respectively. Based on the multiple probability values, a target probability value and its corresponding target modulation type are determined, and the target modulation type and the target probability value are output.
2. The method according to claim 1, characterized in that, The step of performing a Hilbert transform on the target signal to obtain the instantaneous phase of the target includes: Signal acquisition is performed based on the target sampling rate and bandwidth, and the acquired signal is extracted according to the preset amplitude threshold to obtain a time-domain complex signal. Based on high-pass filtering and adaptive noise suppression algorithms, the DC component and noise in the time-domain complex signal are removed to obtain the target signal. Then, Hilbert transform is performed on the target signal to obtain the instantaneous phase of the target.
3. The method according to claim 2, characterized in that, The step of performing a Hilbert transform on the target signal to obtain the instantaneous phase of the target includes: Perform a Hilbert transform on the target signal to obtain the signal to be analyzed and convert it into the corresponding polar coordinate form; For the signal to be analyzed in polar coordinates, the corresponding instantaneous phase of the target is determined based on the arctangent model.
4. The method according to claim 1, characterized in that, The target-based CNN-LSTM cascaded model extracts features from the target frequency sequence to obtain the corresponding feature vector, including: Multiple radar signal samples are acquired and their corresponding modulation type labels are determined according to their respective modulation types to obtain the model training set; Based on the target frequency sequence and modulation type label corresponding to each radar signal sample, the initial CNN-LSTM cascaded model is trained to obtain the target CNN-LSTM cascaded model. Based on the target CNN-LSTM cascaded model, feature extraction is performed on the target frequency sequence to obtain the feature vector.
5. The method according to claim 1, characterized in that, The feature vector is obtained by extracting features from the target frequency sequence based on the target CNN-LSTM cascaded model, including: Based on the CNN network layer in the target CNN-LSTM cascaded model, local structured features of the target frequency sequence are extracted to obtain the corresponding feature sequence. Based on the LSTM network layer, the temporal dependency of the feature sequence is determined to obtain the corresponding feature vector.
6. The method according to claim 1, characterized in that, The step of determining the corresponding target probability vector based on the weight matrix and the feature vector includes: Based on the fully connected layers in the target CNN-LSTM cascaded model, the weight matrix is determined, and the target probability vector is determined according to the product of the weight matrix and the feature vector.
7. The method according to claim 1, characterized in that, Determining the target probability value and its corresponding target modulation type based on multiple probability values includes: By comparing the numerical values of each probability value, the target probability value with the largest value is obtained, and the target modulation type is determined according to the modulation type corresponding to the target probability value.
8. A radar signal modulation type identification device, characterized in that, The device includes: The differentiation module is used to perform Hilbert transform on the target signal to obtain the instantaneous phase of the target, and to differentiate the instantaneous phase of the target to obtain the instantaneous frequency; The extraction module is used to discretize and normalize the instantaneous frequency to obtain the corresponding target frequency sequence, and to extract features from the target frequency sequence based on the target CNN-LSTM cascaded model to obtain the corresponding feature vector. The determination module is used to determine the corresponding target probability vector based on the weight matrix and the feature vector, and to determine the probability values corresponding to multiple preset modulation types based on the vector data of each dimension in the target probability vector. The output module is used to determine a target probability value and its corresponding target modulation type based on multiple probability values, and output the target modulation type and the target probability value.
9. A computer device, characterized in that, include: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in memory and configured to be executed by one or more processors, the one or more programs being configured to perform the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code that can be called by a processor to perform the method as described in any one of claims 1 to 7.