A signal in-pulse modulation pattern recognition method and system

Abstract

The application relates to the technical field of data processing, and discloses a signal intra-pulse modulation pattern identification method and system, which comprises the following steps: receiving a radar pulse signal; performing short-time Fourier transform on the radar pulse signal; in a time-frequency matrix, calculating the energy distribution gradient of each time point along the frequency axis; identifying a gradient mutation point according to the energy distribution gradient of each time point; determining the filter bandwidth boundary based on the gradient mutation point by adopting a negative feedback adjustment mechanism of bandwidth and signal-to-noise ratio; performing bandwidth filtering on the radar pulse signal based on the filter bandwidth boundary; calculating the instantaneous frequency, the first-order differential frequency and the second-order differential frequency of the filtered signal through Hilbert transform, and forming a feature vector; predicting the clustering number range through a long short-term memory neural network according to the feature vector; performing clustering analysis within the clustering number range through a K-Means clustering algorithm according to the feature vector; and outputting a modulation pattern label according to the clustering analysis result.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for recognizing intra-pulse modulation patterns. Background Technology

[0002] In the field of intra-pulse modulation feature extraction for electronic signals, a complete processing flow typically includes key steps such as signal reception, preprocessing, feature extraction, and analysis. After the signal receiving device captures the raw signal from a complex electromagnetic environment, it needs to undergo preprocessing operations such as filtering and noise reduction to remove interference and improve signal quality.

[0003] In existing technologies, preprocessing for low-intercept targets such as electronic signals mainly employs fixed-bandwidth filtering. This method filters noise from the received signal by pre-setting a fixed filtering bandwidth, relying on manually or empirically set bandwidth parameters. Its core logic is to retain signal energy within the preset bandwidth range while filtering out noise interference outside the bandwidth.

[0004] However, in complex battlefield electromagnetic environments, the signal-to-noise ratio (SNR) is extremely low, and target signals are easily submerged in noise. Fixed-bandwidth filtering methods cannot dynamically adjust bandwidth parameters according to changes in the actual electromagnetic environment, and therefore cannot adapt to the dynamic distribution of signal time-frequency energy. When noise frequencies fall within the preset bandwidth, interference cannot be effectively filtered; conversely, when the signal bandwidth exceeds the preset range due to environmental changes, key signal characteristics (such as instantaneous frequency) are lost, resulting in a high signal recognition failure rate under low SNR conditions, making it difficult to meet the detection requirements of low-interception targets. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides a method for identifying intra-pulse modulation patterns, comprising:

[0006] S1: Receive radar pulse signals;

[0007] S2: Perform a short-time Fourier transform on the radar pulse signal to generate a time-frequency matrix;

[0008] S3: In the time-frequency matrix, calculate the energy distribution gradient at each time point along the frequency axis;

[0009] S4: Identify gradient abrupt change points based on the energy distribution gradient at each time point;

[0010] S5: A negative feedback adjustment mechanism for bandwidth and signal-to-noise ratio is adopted to determine the filter bandwidth boundary based on the gradient abrupt change point;

[0011] S6: Based on the filtering bandwidth boundary, perform bandwidth filtering on the radar pulse signal;

[0012] S7: Calculate the instantaneous frequency, first-order differential frequency, and second-order differential frequency of the filtered signal using Hilbert transform, and form a feature vector.

[0013] S8: Based on the feature vector, predict the range of the number of clusters using a long short-term memory neural network;

[0014] S9: Based on the feature vector, perform cluster analysis within the range of the number of clusters using the K-Means clustering algorithm;

[0015] S10: Output modulation style labels based on cluster analysis results.

[0016] A second aspect of the present invention provides a signal pulse modulation pattern recognition system, comprising: a processor and a memory;

[0017] The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the signal pulse modulation pattern recognition method as described in the first aspect.

[0018] The beneficial effects of this invention are reflected in the fact that by adopting a negative feedback adjustment mechanism of bandwidth and signal-to-noise ratio, the filtering bandwidth can be dynamically adjusted according to the time-frequency characteristics of the signal. When the signal-to-noise ratio is high, the bandwidth can be contracted to avoid unnecessary noise interference, while when the signal-to-noise ratio is low, the bandwidth can be automatically expanded to ensure that the key features of the signal are not lost. This effectively improves the accuracy of signal recognition under low signal-to-noise ratio, enhances the robustness of signal processing, reduces signal loss due to bandwidth mismatch, improves the reliability in low intercept target detection, and ultimately achieves more accurate signal recognition and modulation pattern classification. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a signal pulse modulation pattern recognition method provided by the present invention.

[0020] Figure 2 This is a flowchart illustrating a method for dynamically adjusting the window length of a sliding time window provided by the present invention.

[0021] Figure 3 This is a flowchart illustrating a method for determining the filter bandwidth boundary provided by the present invention.

[0022] Figure 4 This is a flowchart illustrating a clustering analysis method provided by the present invention.

[0023] Figure 5 This is a schematic diagram of the structure of a signal pulse modulation pattern recognition system provided by the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1:

[0026] Reference Figures 1-4 The present invention provides a method for identifying intra-pulse modulation patterns, comprising:

[0027] S1: Receive radar pulse signals.

[0028] It should be noted that this step is a fundamental input stage for signal processing.

[0029] In one possible implementation, after S1 and before S2, the process further includes: performing analog-to-digital conversion on the radar pulse signal to convert it into a digital signal.

[0030] It should be noted that converting radar pulse signals into digital signals via analog-to-digital conversion (ADC) enables them to adapt to modern digital signal processing technologies, facilitating subsequent time-frequency analysis, feature extraction, and intelligent recognition. Furthermore, digital signals offer strong anti-interference capabilities, high processing accuracy, flexible parameter adjustment, and support for storage and reproduction, thereby improving the accuracy of modulation pattern recognition and system compatibility.

[0031] S2: Perform a short-time Fourier transform on the radar pulse signal to generate a time-frequency matrix.

[0032] Short-Time Fourier Transform (STFT) is an analytical method that decomposes a signal into time-frequency domains. It divides the signal into short-time segments by applying a sliding window process, and then applies a Fourier transform to each segment to obtain the frequency distribution information at each time point. STFT effectively reveals the changes in a signal in both time and frequency, and is widely used in the analysis of non-stationary signals, such as speech, audio, and radar signals. Its main feature is its ability to provide local time-frequency information of the signal, making it suitable for real-time monitoring and analysis of time-varying signals.

[0033] In one possible implementation, S2 specifically includes sub-steps S201 and S202:

[0034] S201: Dynamically adjust the window length of the sliding time window according to the signal carrier frequency.

[0035] The default window length of the sliding time window is 512 sampling points.

[0036] Optionally, S201 specifically includes S2011 to S2015:

[0037] S2011: Before processing each frame of signal, the spectrum of the previous frame of signal is analyzed by fast Fourier transform to extract the main peak frequency.

[0038] The Fast Fourier Transform (FFT) is an efficient method for calculating the Discrete Fourier Transform (DFT). By reducing redundant computations, it significantly improves the computational speed of the Fourier Transform. The core principle of the FFT is to use the divide-and-conquer method to break down a large-scale Fourier Transform problem into multiple smaller problems, thereby improving computational efficiency. It is widely used in signal processing fields, such as audio analysis, image processing, and communication systems, to convert time-domain signals into frequency-domain signals and reveal the spectral information of the signal.

[0039] The peak frequency refers to the frequency point in the signal spectrum where the energy is at its maximum; it is the most prominent peak position on the spectrum. In practical applications, the peak frequency usually represents the center frequency or carrier frequency of the signal and is the main frequency component of the signal.

[0040] S2012: Using the main peak frequency as the current carrier frequency value, compare it with the initial carrier frequency to calculate the carrier frequency deviation:

[0041] ;

[0042] in, Indicates carrier frequency deviation. Indicates the main peak frequency. Indicates the initial carrier frequency.

[0043] S2013: When the carrier frequency deviation satisfies At the same time, the window length of the sliding time window remains unchanged.

[0044] It should be noted that when the carrier frequency deviation is small, the signal frequency does not change much. Maintaining the current window length can reduce the complexity of adjustment and improve processing efficiency. In this case, frequent adjustments are unnecessary, reducing the computational burden.

[0045] S2014: When the carrier frequency deviation satisfies Adjust the window length of the sliding time window as needed:

[0046] ;

[0047] in, This indicates the adjusted window length. L Indicates the current window length.

[0048] It should be noted that when the carrier frequency changes relatively gently (2-5MHz), the window length can adapt to the signal frequency changes. By adjusting the window length, it can be ensured that the short-time Fourier transform (STFT) can always capture the key frequency components of the signal, thus ensuring the accuracy of time-frequency analysis.

[0049] S2015: When the carrier frequency deviation satisfies When this happens, the window length is reset, and the initial value setting is re-executed to ensure that the window length of the time window is adapted to the new carrier frequency.

[0050] It should be noted that resetting the window length ensures flexibility and adaptability in signal analysis when the carrier frequency changes significantly. Resetting the window length is necessary when the carrier frequency changes considerably, as it prevents a fixed window from being unable to adapt to wideband signals, ensuring that time-frequency analysis of the signal covers the new frequency range and avoiding signal information loss.

[0051] S202: Calculate the frequency components of each window using short-time Fourier transform to construct a three-dimensional time-frequency matrix containing time, frequency, and energy.

[0052] In the prior art, the STFT window is fixed. In this application, the window length can be adaptively adjusted according to the signal carrier frequency, thus avoiding the problem of insufficient adaptability of fixed windows to wideband and narrowband signals.

[0053] S3: In the time-frequency matrix, calculate the energy distribution gradient at each time point along the frequency axis.

[0054] In one possible implementation, S3 specifically involves: calculating the energy distribution gradient based on the energy differences between neighboring frequency points.

[0055] ;

[0056] in, express t The frequency of time is f The energy distribution gradient at that location, express t The frequency of time is f The energy value at +1 express t The frequency of time is f The energy value at -1.

[0057] It should be noted that calculating the gradient using the energy difference between adjacent frequency points can quickly capture the local variation characteristics of energy in the time-frequency matrix, highlighting abrupt changes in the signal along the frequency axis (such as the slope or phase jump of linear frequency modulation), while suppressing interference from smooth noise. This calculation method is simple and efficient, suitable for real-time processing, and provides accurate local variation information for subsequent gradient abrupt point detection and bandwidth boundary extraction.

[0058] S4: Identify gradient abrupt change points based on the energy distribution gradient at each time point.

[0059] Specifically, data points whose absolute gradient value exceeds a preset threshold are considered gradient abrupt change points. The preset threshold is 10% of the average energy of the current frequency band.

[0060] S5: A negative feedback adjustment mechanism for bandwidth and signal-to-noise ratio is adopted, and the filter bandwidth boundary is determined based on the gradient mutation point.

[0061] Signal-to-noise ratio (SNR) is an important indicator of signal quality, representing the ratio of signal strength to noise strength. A higher SNR means the signal is more distinct relative to noise, indicating better signal quality; a lower SNR indicates greater noise interference, making the signal harder to identify. In fields such as communications, radar, and audio, SNR is a key metric for evaluating system performance, often used to measure signal clarity and reliability.

[0062] In one possible implementation, S5 specifically includes sub-steps S501 and S502:

[0063] S501: When the signal-to-noise ratio (SNR) satisfies At that time, the bandwidth is extended to 20% beyond the gradient abrupt change point in the frequency domain.

[0064] It should be noted that in low signal-to-noise ratio (SNR) environments, signals may be weak and easily overwhelmed by noise. Extending the bandwidth allows for better capture of key signal components, preventing signal loss due to insufficient bandwidth. Simultaneously, extending the bandwidth improves the robustness of filtering, enabling more signal features to be preserved and identified, thereby enhancing signal recognition capabilities in low SNR conditions.

[0065] S502: When the signal-to-noise ratio (SNR) satisfies At that time, the bandwidth is reduced to 10% of the frequency domain range within the gradient abrupt change point.

[0066] It should be noted that in environments with a high signal-to-noise ratio, the impact of noise is relatively small, and interference with unrelated frequency components can be reduced by narrowing the bandwidth. This helps to extract key features of the signal more accurately, avoiding the inclusion of unnecessary noise and irrelevant frequency information, thereby improving the accuracy and efficiency of signal processing, while also reducing the computational burden.

[0067] Existing technologies lack energy gradient analysis and rely directly on fixed bandwidth. This application locates the effective bandwidth boundary of the signal by identifying gradient abrupt change points, providing a basis for adaptive filtering.

[0068] In this embodiment of the invention, a negative feedback adjustment mechanism for bandwidth and signal-to-noise ratio is adopted. The filtering bandwidth can be dynamically adjusted according to the time-frequency characteristics of the signal. When the signal-to-noise ratio is high, the bandwidth can be contracted to avoid unnecessary noise interference, while when the signal-to-noise ratio is low, the bandwidth can be automatically expanded to ensure that the key features of the signal are not lost. This effectively improves the accuracy of signal recognition under low signal-to-noise ratio, enhances the robustness of signal processing, reduces signal loss due to bandwidth mismatch, improves the reliability in low intercept target detection, and ultimately achieves more accurate signal recognition and modulation pattern classification.

[0069] S6: Based on the filtering bandwidth boundary, perform bandwidth filtering on the radar pulse signal.

[0070] Specifically, dynamic bandwidth filtering is achieved through FIR filters, preserving key features such as instantaneous frequency.

[0071] Existing technologies employ fixed-bandwidth filtering, which is prone to recognition failure in low signal-to-noise ratio (SNR) environments due to noise falling into or signal overflowing the bandwidth. This invention achieves dynamic matching between the filtering bandwidth and the signal's time-frequency characteristics through energy gradient extraction and negative feedback bandwidth adjustment. Real-world testing data shows that for low intercept signals with an SNR ≤ -5dB, the SNR is improved by ≥15dB, and the signal saturation distortion tolerance is increased by 300%, solving the problem that traditional fixed-bandwidth filtering cannot adapt to complex electromagnetic environments.

[0072] S7: Calculate the instantaneous frequency, first-order differential frequency, and second-order differential frequency of the filtered signal using Hilbert transform, and form an eigenvector.

[0073] The Hilbert transform is a mathematical transformation used to convert a real-valued signal into its corresponding complex-valued signal, generating an "analytical signal". Its core function is to construct an imaginary part that is 90 degrees out of phase with the original signal, thereby obtaining the instantaneous amplitude and frequency of the signal.

[0074] S8: Based on the feature vectors, predict the range of the number of clusters using a long short-term memory neural network.

[0075] Long Short-Term Memory (LSTM) neural networks are a special type of recurrent neural network (RNN) designed to address the vanishing or exploding gradient problems that traditional RNNs encounter when processing long sequences of data. By introducing memory units and control gate mechanisms, LSTM can selectively save and forget information, thereby capturing long-term dependencies.

[0076] Existing technologies require manual pre-setting of the number of clusters. This application utilizes LSTM to dynamically predict the range of cluster numbers, avoiding the bias caused by manual setting. LSTM has the ability to process long-sequence data and capture long-term dependencies, and can predict the number of clusters based on the signal characteristics of the previous time step, thereby effectively reducing the bias and inaccuracy caused by manually setting the number of clusters in traditional methods. This method improves the adaptability and accuracy of cluster analysis, especially in complex and ever-changing signal environments, where it can flexibly adjust the clustering strategy according to the characteristics of the actual signal, further enhancing the performance and robustness of signal recognition.

[0077] S9: Based on the feature vectors, perform cluster analysis within the range of the number of clusters using the K-Means clustering algorithm.

[0078] K-Means clustering is a commonly used unsupervised learning method for dividing a dataset into K distinct clusters, where the center of each cluster is the mean of the samples within that cluster. The algorithm iteratively selects K initial cluster centers, then assigns samples to the nearest cluster based on their distance from the cluster center, updates the center of each cluster, and repeats this process until the cluster centers no longer change.

[0079] In one possible implementation, S9 specifically includes:

[0080] S901: Within the range of the number of clusters, initialize different numbers of cluster centers.

[0081] S902: Clustering is performed according to the cluster centers using the K-Means clustering algorithm.

[0082] S903: Silhouette coefficients for comparing cluster analysis results with different numbers of clusters.

[0083] The silhouette coefficient is a measure of cluster quality, combining intra-cluster compactness and inter-cluster separation.

[0084] S904: Outputs the cluster analysis result with the largest silhouette coefficient.

[0085] It should be noted that by initializing different numbers of cluster centers within a range of cluster sizes, performing cluster analysis using the K-Means clustering algorithm, and then comparing the silhouette coefficients of the clustering results under different numbers of clusters, the most suitable number of clusters can be effectively selected. This method avoids the bias of manually setting the number of clusters and, by utilizing the silhouette coefficient as a metric for cluster quality, ensures that the selected number of clusters reflects the true structure of the data to the greatest extent. By outputting the clustering result with the largest silhouette coefficient, the accuracy and stability of cluster analysis can be improved, thereby enhancing the precision and reliability of signal modulation pattern recognition.

[0086] Existing technologies rely on single-dimensional features or manually preset cluster numbers, leading to high error rates at low signal-to-noise ratios. This invention addresses this by employing three-dimensional feature extraction (instantaneous frequency + differential frequency) and LSTM clustering prediction, combined with parallel clustering optimization, providing dual assurance in terms of feature dimensionality and clustering rationality. At a signal-to-noise ratio of -5dB, the modulation recognition accuracy is increased from 62% in existing technologies to 89%, significantly improving the reliability of signal sorting and threat assessment.

[0087] S10: Output modulation style labels based on cluster analysis results.

[0088] Optionally, the modulation style labels include: linear frequency modulation and phase coding.

[0089] In one possible implementation, the signal pulse modulation pattern recognition method further includes: using a CPU and GPU parallel processing architecture to perform short-time Fourier transform, bandwidth filtering, and cluster analysis in parallel.

[0090] Existing technologies, based on a CPU serial architecture, often suffer from single-pulse processing latency exceeding hundreds of milliseconds, failing to meet the demands of high-density signal scenarios. This invention achieves efficient utilization of computing resources through GPU parallel optimization (CUDA Kernel splitting, pulsed pipeline, and shared memory). With a 1080Ti GPU deployment, single-pulse processing latency is reduced to 35ms, supporting real-time analysis of >500 pulses per second. Successfully applied to border reconnaissance systems, it resolves the real-time bottleneck of traditional architectures.

[0091] Furthermore, a pulsed pipeline can be designed to initiate STFT calculation for the next frame during clustering of the previous frame.

[0092] Furthermore, preprocessed data (such as time-frequency matrices) is stored in the GPU's global memory and transferred via shared memory in thread blocks, reducing CPU-GPU data copy latency.

[0093] It should be noted that this invention supports both CPU and GPU deployment. It improves signal detectability through adaptive filtering, enhances recognition accuracy through multi-dimensional clustering, and improves real-time response capabilities through parallel architecture, forming a full-link optimization from signal detection to feature extraction. This provides key technical support for electronic warfare and intelligence gathering in complex battlefield environments.

[0094] Example 2:

[0095] Reference Figure 5 An intra-pulse modulation pattern recognition system 20 provided in this embodiment of the invention includes: a processor 201 and a memory 202.

[0096] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the signal pulse modulation pattern recognition method as in Embodiment 1 and achieve the same technical effect. To avoid repetition, the present invention will not repeat the above.

[0097] In the description of the embodiments of the present invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "center," "top," "bottom," "top," "bottom," "inner," "outer," "inner side," and "outer side," etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. "Inner side" refers to the interior or enclosed area or space. "Outer perimeter" refers to the area surrounding a specific component or specific area.

[0098] In the description of embodiments of the present invention, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of that feature. In the description of the present invention, unless otherwise stated, "a plurality of" means two or more.

[0099] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," "joining," and "assembly" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0100] In the description of embodiments of the present invention, specific features, structures, materials or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0101] In the description of the embodiments of the present invention, it should be understood that "-" and "~" represent a range of two numerical values, and this range includes the endpoints. For example, "AB" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0102] In the description of embodiments of the present invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of identifying a signal modulation pattern within a pulse, characterized by, Comprising: S1: receiving a radar pulse signal; S2: performing short-time Fourier transform on the radar pulse signal to generate a time-frequency matrix; S3: calculating the energy distribution gradient of each time point along the frequency axis in the time-frequency matrix; S4: identifying the gradient mutation point according to the energy distribution gradient of each time point; S5: determining the filter bandwidth boundary based on the gradient mutation point using a negative feedback adjustment mechanism of bandwidth and signal-to-noise ratio; S6: performing bandwidth filtering on the radar pulse signal based on the filter bandwidth boundary; S7: calculating the instantaneous frequency, first-order differential frequency and second-order differential frequency of the filtered signal through Hilbert transform, and forming a feature vector; S8: predicting the clustering number range through a long short-term memory neural network according to the feature vector; S9: performing clustering analysis within the clustering number range through a K-Means clustering algorithm according to the feature vector; S10: outputting a modulation pattern label according to the clustering analysis result; Wherein, the S2 specifically comprises: S201: dynamically adjusting the window length of the sliding time window according to the signal carrier frequency; S202: calculating the frequency component of each window through short-time Fourier transform to construct a three-dimensional time-frequency matrix containing time, frequency and energy; Wherein, the S201 specifically comprises: S2011: before processing each frame of signal, performing frequency spectrum analysis on the previous frame of signal through fast Fourier transform to extract the main peak frequency; S2012: taking the main peak frequency as the current carrier frequency value, comparing it with the initial carrier frequency, and calculating the carrier frequency deviation; S2013: when the carrier frequency offset satisfies the window length of the sliding time window is kept unchanged; S2014: when the carrier frequency offset satisfies adjust the window length of the sliding time window; S2015: When the carrier frequency offset satisfies the window length resetting is triggered, the initial value setting is re-executed, and the window length of the time window is ensured to be adapted to the new carrier frequency.

2. The method of claim 1, wherein After the S1 and before the S2, it further comprises: Analog-to-digital conversion of the radar pulse signal into a digital signal.

3. The method of claim 1, wherein The S3 specifically is: Calculating the energy distribution gradient through the energy difference of adjacent frequency points.

4. The method of claim 1, wherein The S5 specifically comprises: S501: when the signal-to-noise ratio SNR satisfies , the bandwidth is extended to a frequency domain range 20% outside the gradient mutation point. S502: When the signal-to-noise ratio SNR satisfies shrink the bandwidth to a frequency domain range that is 10% within the gradient mutation point.

5. The method of claim 1, wherein The S9 specifically comprises: S901: initializing different numbers of clustering centers within the clustering number range; S902: performing clustering according to the clustering centers through the K-Means clustering algorithm; S903: comparing the silhouette coefficients of the clustering analysis results under different clustering numbers; S904: outputting the clustering analysis result with the largest silhouette coefficient.

6. The method of claim 1, wherein The modulation pattern label includes: linear frequency modulation and phase coding.

7. The method of claim 1, wherein Further comprising: Using CPU and GPU parallel processing architecture to parallelly process short-time Fourier transform, bandwidth filtering and clustering analysis.

8. A signal in-pulse modulation pattern recognition system, characterized by, Comprising: A processor and a memory; The memory stores programs or instructions that can be run on the processor, and the programs or instructions are executed by the processor to realize the steps of the signal intra-pulse modulation pattern identification method according to any one of claims 1 to 7.

Images