Radar signal intelligent detection method and system based on digital channelization and PointNet + +
By combining digital channelization and the PointNet++ network, the difficulty of radar signal detection under low signal-to-noise ratio conditions is solved, intelligent radar signal detection is realized, detection accuracy and adaptability are improved, and it can adapt to complex electromagnetic environments.
Patent Information
- Application Number
- CN202510936959.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-09-19
AI Technical Summary
Existing electronic reconnaissance systems have difficulty detecting single and multiple signals under low signal-to-noise ratio conditions. Traditional radar signal detection methods have application limitations and human subjectivity, making them difficult to effectively combine with digital channelized receiver technology.
An intelligent radar signal detection method based on digital channelization and PointNet++ is adopted. By constructing a signal reception model, designing a prototype low-pass filter, and receiving signals through a polyphase filter bank, and using the PointNet++ network to perform three-dimensional point cloud segmentation, intelligent signal detection is achieved.
It improves the accuracy and adaptability of radar signal detection under low signal-to-noise ratio conditions, can effectively handle multi-signal detection, adapt to complex electromagnetic environments, and has strong learning and adaptability.
Smart Images
Figure CN120669203A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of radar signal detection, and in particular relates to an intelligent radar signal detection method based on digital channelization and PointNet++. Background Art
[0002] With technological advancements and the evolution of warfare, the importance of electronic warfare and information warfare in information warfare continues to grow. However, the increasingly complex electromagnetic environment places higher demands on the application of electronic warfare. Electronic reconnaissance, a key component of electronic warfare, is primarily used to intercept radar emitter signals in complex electromagnetic environments. By analyzing signal properties and other information, it detects and locates radar emitters of interest, enabling appropriate countermeasures. As the information input to electronic reconnaissance systems, radar signal detection is the foundation and prerequisite for radar electronic reconnaissance receivers to measure signal parameters.
[0003] Because the power of radiation source signals is typically low, they are susceptible to significant noise interference during transmission. The signals reaching the receiver often have very low signal-to-noise ratios, making signal detection and interception more difficult. The pulse density of radiation source signals on modern battlefields continues to increase, along with the increasing probability of pulse overlap. This makes single and multiple signal detection increasingly difficult for electronic reconnaissance systems under low signal-to-noise ratio conditions.
[0004] Traditional radar signal detection methods use manually set thresholds for signal detection. While adaptive thresholds can be achieved, they have certain application limitations and are subject to human subjectivity. Compared to traditional signal processing systems, intelligent processing has stronger learning capabilities and can effectively solve signal detection problems through training. Furthermore, with a sufficient number of training samples, it is highly adaptable to the environment. Therefore, intelligent signal processing is a new and effective approach to solving problems related to radar signal detection.
[0005] In view of this, it is necessary to develop an intelligent radar signal detection technology to solve the above problems. The intelligent radar signal detection technology proposed in this paper can detect radar signals with a high probability. At the same time, under the premise of a high signal-to-noise ratio of the training sample, it can also detect low signal-to-noise ratio radar signals with a high probability, effectively solving problems related to intelligent radar signal detection. Summary of the Invention
[0006] To address the difficulty existing electronic reconnaissance systems face in detecting single and multiple signals under low signal-to-noise ratio conditions, this paper proposes an intelligent radar signal detection method based on digital channelization and PointNet++. This method also addresses the difficulty of integrating traditional intelligent technologies with digital channelized receiver technology.
[0007] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a radar signal intelligent detection method based on digital channelization and PointNet++, the method comprising the following steps: Step S1: constructing a signal receiving model; Step S2: Design a The prototype low-pass filter of the order is modulated to obtain a polyphase filter bank; Step S3: passing the signal received by the signal receiving model through a polyphase filter bank to complete reception by a digital channelized receiver; Step S4: Processing the multi-channel signal received by the digital channelized receiver to obtain a data set, including a training set and a test set; Step S5: Set the PointNet++ network parameters and train the PointNet++ network using the training set; Step S6: Call the trained PointNet++ network, input the test set, and obtain the segmentation task results.
[0008] Furthermore, the parameters of the prototype low-pass filter are: The passband frequency is MHz, the stop band frequency is MHz, the passband ripple is , the stop-band attenuation is defined as .
[0009] Furthermore, the multi-channel signal processing is specifically as follows: The multi-channel signal is converted into a three-dimensional point cloud image, and the points in the point cloud image are labeled as signal points or no signal points. After obtaining the three-dimensional point cloud image, a data set is created.
[0010] Furthermore, the above-mentioned settings of PointNet++ network parameters include 3 SA layers, 3 FP layers, and output of segmentation results through BN+Relu.
[0011] Furthermore, the SA layer includes a sampling layer, a grouping layer, and a PointNet layer; The sampling layer is used to select a subset from a given input point set by iterative farthest point sampling; The grouping layer is used to group the samples in the subset according to the set radius and number of points using the Ball Query method; The PointNet layer is used to extract features using PointNet in each group of the grouping layer.
[0012] Furthermore, the above FP layer includes interpolation and Unit PointNet layers.
[0013] Furthermore, the above-mentioned FP layer is used to adopt a hierarchical propagation strategy based on distance interpolation and cross-layer skip links to propagate features from the downsampled points of the SA layer to the original points. It is also used to concatenate the obtained features with the features in the SA layer using skip links. The concatenated features are subjected to feature unit convolution through the Unit PointNet layer.
[0014] Secondly, the radar signal intelligent detection method based on digital channelization and PointNet++ described in the present invention can be fully implemented using computer software. Therefore, correspondingly, the present invention also provides a radar signal intelligent detection system based on digital channelization and PointNet++.
[0015] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, executes any one of the above-mentioned methods for intelligent detection of radar signals based on digital channelization and PointNet++.
[0016] In a fourth aspect, the present invention also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes any one of the above-mentioned methods for intelligent detection of radar signals based on digital channelization and PointNet++.
[0017] The beneficial effects of the present invention are: In modern battlefields, signals reaching receivers often have very low signal-to-noise ratios. Furthermore, the pulse density of radiation source signals in modern battlefields continues to increase, and the probability of pulse overlap is also increasing. This makes single and multiple signal detection increasingly difficult for electronic reconnaissance systems under low signal-to-noise ratio conditions. Therefore, the present invention provides an intelligent radar signal detection method based on digital channelization and PointNet++. This method, combined with digital channelization and the point cloud neural network PointNet++, effectively solves the problem of single and multiple signal detection under low signal-to-noise ratio conditions, providing a new solution for intelligent detection.
[0018] Furthermore, existing digital channelized receivers have the advantages of high sensitivity, large instantaneous bandwidth, wide dynamic range, and high probability of interception. They can adapt to most radar signals and can process multiple signals arriving simultaneously, solving problems such as broadband monitoring and multi-signal reception. They also have strong adaptability in complex electromagnetic environments. However, due to the multi-channel characteristics of the output data of digital channelized receivers, traditional intelligent technologies are difficult to combine with digital channelized receiver technology. Therefore, the present invention uses a deep learning-based PointNet++ network to process the output data of the digital channelized receiver into a three-dimensional point cloud, transforming the detection task into a three-dimensional point cloud segmentation task, thereby realizing intelligent detection of digital channelized receivers.
[0019] The present invention is applicable to the field of intelligent detection of single signals and multiple signals under low signal-to-noise ratio conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0021] Figure 1 This is a structural diagram of the radar signal intelligent detection method based on digital channelization and Pointnet++ according to the present invention; Figure 2 The uniform channelization structure based on the polyphase filter bank of the present invention; Figure 3 The uniform channelized receiving structure after decimation and forward shifting according to the present invention; Figure 4 The structure of the high-efficiency digital channelized receiver according to the present invention; Figure 5 This is a schematic diagram of outputting spectrum of each sub-channel according to the present invention; Figure 6 The point cloud sample data described in the present invention; Figure 7 This is the low signal-to-noise ratio detection result described in the present invention. DETAILED DESCRIPTION
[0022] In the following description, the specific implementation details of the "Intelligent Detection Method of Radar Signals Based on Digital Channelization and PointNet++" provided in this specification (such as experimental equipment, operating procedures, data processing steps and example parameters) are for illustrative purposes rather than restrictive definitions, and are intended to help those skilled in the art to thoroughly understand the principles and implementation of the present invention. However, those skilled in the art should be aware that these details only represent one of the feasible embodiments, and the core concept of the present invention can be fully implemented through other technical means or workarounds that are not fully described without departing from its spirit. In addition, the omission of conventional experimental methods and device details known in the art in the specification is to avoid redundant information interfering with the understanding of the innovation points. This does not mean that these known technologies are not required for implementation, and those skilled in the art should be able to supplement and apply them on their own based on their professional knowledge.
[0023] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. The following embodiments will help those skilled in the art further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that those skilled in the art may make various changes and improvements without departing from the scope of the present invention, and these are all within the scope of protection of the present invention.
[0024] Implementation 1: This implementation proposes an intelligent radar signal detection method based on digital channelization and PointNet++. This method integrates the PointNet++ network into a digital channelized receiver, achieving high detection accuracy while also being applicable to detecting radar signals with low signal-to-noise ratios. The method comprises the following steps: Step S1: constructing a signal receiving model; Step S2: Design a The prototype low-pass filter of the order is modulated to obtain a polyphase filter bank; Step S3: passing the signal received by the signal receiving model through a polyphase filter bank to complete reception by a digital channelized receiver; Step S4: Processing the multi-channel signal received by the digital channelized receiver to obtain a data set, including a training set and a test set; Step S5: Set the PointNet++ network parameters and train the PointNet++ network using the training set; Step S6: Call the trained PointNet++ network, input the test set, and obtain the segmentation task results.
[0025] In modern battlefields, signals reaching receivers often have very low signal-to-noise ratios. Furthermore, the pulse density of radiation source signals in modern battlefields continues to increase, and the probability of pulse overlap is also increasing. This makes single and multiple signal detection increasingly difficult for electronic reconnaissance systems under low signal-to-noise ratio conditions. Therefore, this embodiment provides an intelligent radar signal detection method based on digital channelization and PointNet++. This method, combined with digital channelization and the point cloud neural network PointNet++, effectively solves the problem of single and multiple signal detection under low signal-to-noise ratio conditions, providing a new solution for intelligent detection.
[0026] Implementation Method 2: Combination Figures 1 to 7 This embodiment is to describe the radar signal intelligent detection method based on digital channelization and PointNet++ described in the first embodiment. Step S1: constructing a signal receiving model; Specifically: Construct a receiving signal model, which acts as a receiving end and receives signals including target signals and noise signal .
[0027] Input signal for:
[0028] in, It can be a single target signal or a signal superimposed by multiple target signals. is Gaussian white noise.
[0029] Step S2: Design a The prototype low-pass filter of the order is modulated to obtain a polyphase filter bank; Specifically: Design a The prototype low-pass filter of order is modulated by the prototype low-pass filter to obtain a polyphase filter bank.
[0030] Input signal After sampling, it is converted into a discrete signal ;set up ,in is the number of channels, is the decimation factor. Divided into multiphase structures as follows:
[0031] Prototype filter After modulation, A polyphase filter, taking , for the channels, then:
[0032] From the above formula we can see that The output of the filter is actually a branch of the polyphase filter After IDFT, Through this process, a filter composed of multiple polyphase filters can be constructed to extract The uniform channelization structure of sub-channels, such as Figure 2 shown.
[0033] make is the output of the polyphase filter bank. When the decimation is not forward-shifted, the IDFT output It can be expressed as:
[0034] Perform the Z transform:
[0035] The system can be obtained Path transfer function:
[0036] Output of each sub-channel It can be expressed as:
[0037] Furthermore, in order to avoid wasting computing resources, the extraction is performed forward, such as Figure 3 As shown. After decimation and forward shift, the output of the polyphase filter is:
[0038] Perform the Z transform:
[0039] Among them, due to is an integer, then , the above formula can be further simplified as:
[0040] Then we can get:
[0041] in, , is the filter order. Then, its output after IDFT can be expressed as:
[0042] The constraint that adjacent channels do not alias is , this embodiment takes , so the multiphase components can be expressed as:
[0043] when When, due to , so even-numbered output subchannels do not need to be multiplied by any coefficient factor, while odd-numbered subchannels need to be multiplied by The IDFT operation is replaced by the IFFT fast algorithm. In summary, the efficient digital channelized receiver structure adopted in this embodiment is as follows: Figure 4 shown.
[0044] This structure is applicable to both real and complex channelization. If the input signal is a real signal, then and The output is real number, and the outputs of other sub-channels are all complex numbers. and ( ) are conjugate to each other. Therefore, for real number channelization, only the first If the input signal is a complex signal, then all channel outputs are complex numbers, and all channels are independent and there is no case where the outputs are conjugate to each other.
[0045] Step S3: passing the signal received by the signal receiving model through a polyphase filter bank to complete reception by a digital channelized receiver; Specifically: This embodiment adopts a digital channelized receiver front-end ADC sampling frequency of MHz, the monitoring frequency range is MHz, and the number of channels is 32. The relevant parameters involved in the prototype filter design are as follows: (1) The passband frequency of the filter is set to MHz; (2) The stopband frequency is set to MHz; (3) The passband ripple is defined as ; (4) The stopband attenuation is defined as .
[0046] This results in a 192-order prototype low-pass filter. Functional simulation of the efficient digital channelization model was performed. There are two types of radar signals input to the digital channelization module: signal 1 is a single-frequency signal with a carrier frequency of 906 MHz and a power of 0 dBm; signal 2 is a linear frequency modulation signal with a center frequency of 820 MHz and a bandwidth of 10 MHz, and a power of 0 dBm. The input signal-to-noise ratio is set to 10 dB. The frequency domain simulation results corresponding to the first 16 sub-channels of the real channelization output are shown in the figure below. Figure 5 shown.
[0047] Step S4: Processing the multi-channel signal received by the digital channelized receiver to obtain a data set, including a training set and a test set; Specifically: The multi-channel signals received by a digital channelized receiver are converted into a 3D point cloud image. The points in the point cloud image are labeled as either signal points or no signal points. After obtaining enough 3D point cloud images, a dataset is created, including a training set and a test set.
[0048] For the PointNet++ point cloud neural network, a data set containing six types of signals was created: single-frequency signal, LFM signal, 2FSK signal, 4FSK signal, Barker code, and Frank code signal. The data includes 14,400 sample signals, each of which contains the three-dimensional coordinates of 24,000 sampling points. The label corresponding to the sampling point. Define the sampling frequency of the signal MHz, carrier range MHz, the interval between each two frequencies is 75MHz, there are 12 frequency points in total, the signal-to-noise ratio range is 0~20dB, each signal has 4 pulses, the pulse width is 2.5us, the bandwidth of the LFM signal is 30MHz, and the frequency interval of 2FSK and 4FSK signals is 10MHz. The same parameters as above are used to generate a test data set for subsequent network testing. Point cloud data samples are as follows Figure 6 shown.
[0049] Step S5: Set the PointNet++ network parameters and train the PointNet++ network using the training set; Specifically: Set the PointNet++ network parameters, including 3 SA layers and 3 FP layers, and output the segmentation results through BN+Relu. Some network parameters include batch_size (number of samples per training), epochs (number of training rounds), learning_rate (learning rate), npoint (number of sampling points), etc., which should be appropriately set according to different data sets and computing resources. Use the data set in step S4 to train the PointNet++ network and save the training model. The PointNet++ network model used is as follows: Figure 1 shown.
[0050] For the SA layer, it contains three key layers: sampling layer, grouping layer and PointNet layer.
[0051] For the sampling layer, given an input point set , select a subset from it by iterative farthest point sampling (FPS) , the specific steps are: (1) Select an initial point from the point set , define the set ; (2) Traverse all points in the point set ,calculate , find the maximum value as Add to collection ; (3) Repeat the above process until the selected points.
[0052] Compared with random sampling, FPS can better cover the entire point set with the same number of sampling points.
[0053] For the grouping layer, the Ball Query method is used to group the sample points according to the set radius and number of points. The specific steps are: Draw a sphere with the radius of the sphere, and take the point in the sphere that is closest to the center of the sphere. points, and these points and the sampling points constitute a local area.
[0054] For the PointNet layer, PointNet is used to extract features in each group of the grouping layer. PointNet directly processes point cloud data and maps the coordinates and other features of the local point cloud into a high-dimensional space through a multi-layer perceptron (MLP).
[0055] For FP layers, it includes interpolation and Unit PointNet layers.
[0056] For interpolation, since the original point set is downsampled in the SA layer, and in the segmentation task, it is hoped to generate features for all points in the point set, a hierarchical propagation strategy based on distance interpolation and cross-layer jump links is adopted to propagate features from the downsampled points to the original points. Interpolation is performed by the following formula:
[0057] in, Indicates the feature dimension of the point set input. In this implementation, .
[0058] After interpolation, the obtained features are concatenated with the features in SA using skip links. The concatenated features are convolved with Unit PointNet, and finally the features of each point are updated using BN+ReLU.
[0059] Initialize the network parameters to batch_size=8, epochs=10, learning_rate=0.0001, and npoint=1024. Define the error function and let the network perform gradient descent based on the error function. The error function uses the negative log-likelihood loss function, as shown below:
[0060] in, is the true label, is the model’s predicted probability distribution, Is the model's true category The predicted probability of .
[0061] After training is complete, save the network model.
[0062] Step S6: Call the trained PointNet++ network, input the test set, and obtain the segmentation task results.
[0063] Specifically: The test data set has the same parameters as the training set, with a total of 1440 sample data. The test results are shown in Table 1.
[0064] Table 1
[0065] In practical application, this embodiment further adds step S7: generating low signal-to-noise ratio sample data, calling the model trained in step S5 to test the segmentation task results, analyzing the results, and evaluating the low signal-to-noise ratio detection results.
[0066] Specifically: Other parameters remain unchanged, generate sample data with a signal-to-noise ratio of -10~0dB, call the model trained in step S5 to test the segmentation task results, and compare the detection results with the digital channelization + OS-CFAR method (set the false alarm probability ), the result is as follows Figure 7 shown.
[0067] Depend on Figure 7 It can be seen that when the signal-to-noise ratio is low, the detection probability of the present invention is significantly higher than that of the traditional adaptive threshold method, which shows that the present invention is different from the traditional adaptive algorithm, has a strong learning ability, and is more adaptable to low signal-to-noise ratio environments.
[0068] Embodiment 3: The radar signal intelligent detection method based on digital channelization and PointNet++ described in Embodiment 1 or 2 above can be implemented entirely using computer software. Therefore, correspondingly, this embodiment provides a radar signal intelligent detection system based on digital channelization and PointNet++, the system comprising: a storage device for constructing a signal reception model; To design a The prototype low-pass filter of the order is modulated to obtain the storage device of the polyphase filter bank; a storage device for passing the signal received by the signal receiving model through a polyphase filter bank to complete reception by the digital channelized receiver; A storage device for processing multi-channel signals received by a digital channelized receiver to obtain a data set, including a training set and a test set; A storage device used to set PointNet++ network parameters and train the PointNet++ network using the training set; A storage device used to call the trained PointNet++ network, input the test set, and obtain the segmentation task results.
[0069] Implementation 4. This implementation provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the radar signal intelligent detection method based on digital channelization and PointNet++ described in any one of the above implementations is executed.
[0070] Implementation 5. This implementation provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes a radar signal intelligent detection method based on digital channelization and PointNet++ as described in any one of the above implementations.
[0071] This embodiment provides a computer device. The hardware devices in this part are of general models and are not shown in the diagram. The system includes a processor and a memory, wherein the processor and the memory can be connected via a bus or other means. The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and modules, and corresponding program instructions / modules. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, so as to implement the radar signal intelligent detection method and steps based on digital channelization and PointNet++ in the above-mentioned method embodiment.
[0072] In the above description, it should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.
[0073] The foregoing description is merely an embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of the claims.
Claims
1. A radar signal intelligent detection method based on digital channelization and PointNet++, characterized in that: The method is: S1: Build a signal reception model; S2: Design a The prototype low-pass filter of the order is modulated to obtain a polyphase filter bank; S3: The signal received by the signal receiving model is passed through a polyphase filter bank to complete reception by a digital channelized receiver; S4: Processing the multi-channel signal received by the digital channelized receiver to obtain a data set, including a training set and a test set; S5: Set the PointNet++ network parameters and train the PointNet++ network using the training set; S6: Call the trained PointNet++ network, input the test set, and obtain the segmentation task results.
2. The radar signal intelligent detection method based on digital channelization and PointNet++ according to claim 1, characterized in that: The prototype low-pass filter parameters are: The passband frequency is MHz, the stop band frequency is MHz, the passband ripple is , the stop-band attenuation is defined as .
3. The radar signal intelligent detection method based on digital channelization and PointNet++ according to claim 1, characterized in that: Multi-channel signal processing is specifically as follows: The multi-channel signal is converted into a three-dimensional point cloud image, and the points in the point cloud image are labeled as signal points or no signal points. After obtaining the three-dimensional point cloud image, a data set is created.
4. The radar signal intelligent detection method based on digital channelization and PointNet++ according to claim 1, characterized in that: Set the PointNet++ network parameters, including 3 SA layers, 3 FP layers, and output the segmentation results through BN+Relu.
5. The radar signal intelligent detection method based on digital channelization and PointNet++ according to claim 4, characterized in that: The SA layer includes the sampling layer, the grouping layer, and the PointNet layer; The sampling layer is used to select a subset from a given input point set by iterative farthest point sampling; The grouping layer is used to group the samples in the subset according to the set radius and number of points using the Ball Query method; The PointNet layer is used to extract features using PointNet in each group of the grouping layer.
6. The radar signal intelligent detection method based on digital channelization and PointNet++ according to claim 4, characterized in that: The FP layer contains the interpolation and Unit PointNet layers.
7. The radar signal intelligent detection method based on digital channelization and PointNet++ according to claim 6, characterized in that: The FP layer is used to adopt a hierarchical propagation strategy based on distance interpolation and cross-layer skip links to propagate features from the downsampled points of the SA layer to the original points. It is also used to concatenate the obtained features with the features in the SA layer using skip links. The concatenated features are then convolved with feature units through the Unit PointNet layer.
8. A radar signal intelligent detection system based on digital channelization and PointNet++, characterized in that: The system is implemented based on a radar signal intelligent detection method based on digital channelization and PointNet++ as described in any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the radar signal intelligent detection method based on digital channelization and PointNet++ according to any one of claims 1 to 7.
10. A computer device, characterized in that: The device includes a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the radar signal intelligent detection method based on digital channelization and PointNet++ as described in any one of claims 1 to 7.