Low sidelobe pulse compression method, device and equipment for distributed coherent radar

Abstract

The invention discloses a low sidelobe pulse compression method, device and equipment for a distributed coherent radar, and relates to the field of radar pulse compression, and the method comprises the steps: carrying out the channel separation of a mixed target echo signal received by the distributed coherent radar, and obtaining a plurality of single-channel echo signals; for any single-channel echo signal, determining probability distribution of a signal peak value in the single-channel echo signal in each distance unit by adopting a pulse compression neural network; according to probability distribution of a signal peak value in the single-channel echo signal in each distance unit, determining a peak value position so as to obtain a low sidelobe pulse compression signal corresponding to the single-channel echo signal; and performing coherent synthesis processing on the low-sidelobe pulse compression signals corresponding to the single-channel echo signals to obtain a target detection result. According to the invention, through channel separation, neural network-assisted peak positioning and phase synthesis, low-sidelobe and high-precision target detection of distributed coherent radar pulse compression is realized.

Description

Technical Field

[0001] This application relates to the field of radar pulse compression, and in particular to a method, apparatus and equipment for low sidelobe pulse compression of distributed coherent radar. Background Technology

[0002] Distributed coherent radar, as a new type of radar system, spatially disperses multiple small-aperture radar units. Through multi-node collaboration, signal-level coherent processing, and intelligent networking technology, it effectively constructs a large-aperture radar system. It has advantages such as high synthetic gain, strong mobility, and strong anti-interference capability, and has extremely high application value in fields such as military reconnaissance, civil aviation control, and ship navigation.

[0003] Pulse compression technology is a key supporting technology for achieving high range resolution in distributed coherent radar. Its core principle is to transform the radar echo signal through filtering, compressing wide pulse signals into narrow pulses, thereby improving the time-domain resolution of radar echo detection. Traditional pulse compression technology mainly uses matched filters, but matched filters suffer from high sidelobe interference in practical applications. High sidelobe signals may be misidentified as real targets, leading to a decrease in target detection accuracy. For distributed coherent radar, its coherent synthesis step relies on high-precision alignment of echoes transmitted from different stations in the time domain. If the pulse compression output has high sidelobes, it will impair the coherent synthesis efficiency. Therefore, high sidelobes in pulse compression have become a key bottleneck restricting the performance improvement of distributed coherent radar.

[0004] To address the high sidelobe interference problem caused by pulse compression using matched filters, various sidelobe suppression methods have been proposed, mainly including the following technical approaches.

[0005] (1) Window function method: By applying window functions, such as Hamming window, Hanning window, Blackman window, etc., to the impulse response of the matched filter, the sidelobe amplitude can be reduced. However, this method will cause the main lobe to widen, increasing the ambiguity of the time delay estimation. Although it suppresses the output sidelobe of pulse compression to a certain extent, it will reduce the range resolution of the radar and cannot help improve the coherent synthesis accuracy of distributed coherent radar.

[0006] (2) Least square filter method: Based on the least square criterion, a mismatch filter is designed to seek a balance between sidelobe suppression and main lobe width. However, this method is complex to design and sensitive to noise. Its performance deteriorates in low signal-to-noise ratio environments and cannot adapt to the complex electromagnetic environment of distributed coherent radar operation.

[0007] (3) Neural Network Method: The mismatch filter is designed using the nonlinear fitting capability of artificial neural networks. Currently, this method has achieved certain sidelobe suppression effects in radar systems. Although the signal-to-noise ratio performance of the mismatch filter is not optimal, it can significantly reduce the output sidelobes of pulse compression and play a role in suppressing interference. However, existing neural network methods are only designed for single-channel signals and can only realize the pulse compression function of single-station radar. They lack the ability to separate multi-channel echo signals of distributed coherent radar and cannot be directly adapted to the multi-node architecture of distributed coherent radar.

[0008] The main shortcomings of existing technologies can be summarized as follows: First, there is a trade-off between sidelobe suppression and main lobe width in the radar pulse compression process based on matched filtering. Traditional sidelobe suppression techniques such as window function method and least squares filter cannot achieve low sidelobe and narrow main lobe at the same time, resulting in a trade-off between range resolution and sidelobe suppression performance, which makes it difficult to meet the requirements of distributed coherent radar for high resolution and low interference.

[0009] Secondly, distributed coherent radar needs to process orthogonal frequency division echo signals from multiple sub-unit radars simultaneously at the receiver. Traditional matched filters undertake the dual tasks of pulse compression and channel separation. However, existing neural network-based pulse compression methods are only designed for single-channel signals and do not consider the separation requirements of multi-channel signals. They are not suitable for the multi-channel architecture of distributed coherent radar, which makes them unsuitable for direct application in distributed coherent radar systems.

[0010] Finally, while complex neural networks such as deep convolutional neural networks can improve sidelobe suppression performance, their large number of parameters, high computational complexity, and long inference latency make them unsuitable for the millisecond-level real-time processing requirements of practical radar systems. Furthermore, training datasets are often generated under ideal conditions, lacking simulation of real-world complex electromagnetic environments, resulting in poor robustness and insufficient generalization ability of the networks in real-world noise and interference scenarios.

[0011] Due to the above drawbacks, a new pulse compression method is needed to meet the requirements of distributed coherent radar. Summary of the Invention

[0012] The purpose of this application is to provide a method, apparatus, and device for low sidelobe pulse compression in distributed coherent radar. By channel separation, neural network-assisted peak localization, and coherent synthesis, low sidelobe and high-precision target detection can be achieved through distributed coherent radar pulse compression.

[0013] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a low sidelobe pulse compression method for distributed coherent radar, comprising: Channel separation is performed on the mixed target echo signals received by the distributed coherent radar to obtain multiple single-channel echo signals; For any single-channel echo signal, a pulse compression neural network is used to determine the probability distribution of the signal peak value in each distance cell of the single-channel echo signal; the pulse compression neural network is obtained by training a feedforward neural network in advance using a training sample set, the training sample set including linear frequency modulated signals with different peak positions and different signal-to-noise ratio ranges and corresponding distance cell position labels; Based on the probability distribution of the signal peak in each distance cell in the single-channel echo signal, the peak position is determined to obtain the low sidelobe pulse compression signal corresponding to the single-channel echo signal. The low sidelobe pulse compression signals corresponding to each single-channel echo signal are subjected to coherent synthesis processing to obtain the target detection results.

[0014] Secondly, this application provides a low sidelobe pulse compression device for a distributed coherent radar, comprising: The channel separation module is used to separate the mixed target echo signals received by the distributed phased coherent radar into multiple single-channel echo signals. The pulse compression module is used to determine the probability distribution of the signal peak in each distance cell of any single-channel echo signal using a pulse compression neural network. The pulse compression neural network is obtained by training a feedforward neural network with a training sample set, which includes linear frequency modulated signals with different peak positions and different signal-to-noise ratio ranges, as well as corresponding distance cell position labels. The compressed signal generation module is used to determine the peak position based on the probability distribution of the signal peak in each distance unit in the single-channel echo signal, so as to obtain the low sidelobe pulse compressed signal corresponding to the single-channel echo signal. The coherent synthesis module is used to perform coherent synthesis processing on the low sidelobe pulse compression signals corresponding to each single-channel echo signal to obtain the target detection result.

[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned low sidelobe pulse compression method for distributed phased coherent radar.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects: The distributed coherent radar low sidelobe pulse compression method provided in this application addresses the technical pain points of traditional pulse compression technology in distributed coherent radar applications where the sidelobe is high and the peak positioning accuracy is insufficient in scenarios with low signal-to-noise ratio. It achieves significant technical gains by relying on the collaborative processing logic of channel separation, pulse compression neural network and coherent synthesis. Specifically, by separating the channels of the mixed target echo signal, signal interference between multiple channels can be eliminated, laying the foundation for accurate single-channel processing. A pulse compression neural network trained with linear frequency modulated signals covering different peak positions and signal-to-noise ratio ranges, along with corresponding tags, can accurately output the probability distribution of the single-channel echo signal peak in each range cell. This breaks through the dependence of traditional algorithms on signal-to-noise ratio, enabling accurate determination of peak positions under low signal-to-noise ratio conditions, thereby effectively reducing the sidelobes of the pulse-compressed signal and improving target resolution. Subsequent coherent synthesis processing integrates the advantages of each single-channel low-sidelobe pulse compression signal, enhancing the coherent accumulation gain of the distributed radar, significantly improving the sensitivity and reliability of target detection, and meeting the application requirements of distributed coherent radar for high-precision, high-stability target detection in complex electromagnetic environments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a low sidelobe pulse compression method for a distributed phased coherent radar, provided as an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the received signal processing link of a distributed phased coherent radar with independent channel separation and pulse compression in one embodiment of this application.

[0020] Figure 3 This is a schematic diagram of the structure of a feedforward neural network in one embodiment of this application.

[0021] Figure 4 This is a schematic diagram illustrating the process of generating a training sample set through cyclic shifting in one embodiment of this application.

[0022] Figure 5 This is a comparison diagram in the time domain of the pulse compression effect achieved by different methods in one embodiment of this application.

[0023] Figure 6This is a comparison chart of the distance-Doppler plot results output in distributed coherent working mode for pulse compression effects achieved by different methods in one embodiment of this application.

[0024] Figure 7 This is a functional module diagram of a low sidelobe pulse compression device for a distributed phased coherent radar provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] The purpose of this application is to overcome the aforementioned deficiencies in the prior art and provide a low-sidelobe pulse compression method for distributed coherent radar. Specifically, this application achieves the following objectives: (1) Solve the problem that distributed coherent radar cannot directly apply neural networks for pulse compression, and realize the coordinated processing of channel separation and low sidelobe pulse compression.

[0027] (2) Significantly reduces the output sidelobe of distributed coherent radar pulse compression.

[0028] (3) Design a lightweight neural network architecture to reduce computational complexity and meet the real-time processing requirements of distributed coherent radar.

[0029] (4) By using the above methods, the accuracy of coherent parameter estimation and the efficiency of coherent synthesis can be improved, thereby enhancing the target detection performance of distributed coherent radar.

[0030] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, a low sidelobe pulse compression method for distributed coherent radar is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the low sidelobe pulse compression method for distributed coherent radar includes the following steps 101 to 104.

[0032] Step 101: Perform channel separation on the mixed target echo signal received by the distributed coherent radar to obtain multiple single-channel echo signals (BPF1~BPF2). N ).

[0033] In a specific application example, the hybrid target echo signal includes the radio frequency echo signal received by each radar element in the distributed coherent radar. Step 101 includes steps 11 to 12.

[0034] Step 11: The hybrid target echo signal is amplified and down-converted to baseband to obtain a hybrid baseband echo signal. Specifically, the distributed coherent radar starts operating in cooperative detection mode, with each radar unit simultaneously transmitting orthogonal frequency division linear frequency modulated (OFDM) signals. The carrier frequency and bandwidth of each signal are allocated according to a preset rule to ensure orthogonality between signals. The receiving antennas (RX-N) of each radar unit receive the hybrid target echo signals transmitted by all radar units and reflected by the target. After amplification, the signals are input to the downconverter to remove the influence of the carrier frequency, converting the hybrid target echo signal into a hybrid baseband echo signal.

[0035] Step 12: The mixed baseband echo signal is subjected to finite impulse response bandpass filtering using a bandpass filter bank to obtain multiple single-channel echo signals, thereby suppressing cross-interference.

[0036] The center frequency of the bandpass filter bank corresponds one-to-one with the center frequency of the orthogonal frequency division linear frequency modulated (OFDM) signal transmitted by each radar element of the distributed coherent radar. The bandpass filter bank performs frequency-differentiated channel separation on the received mixed target echo signals while maintaining the linear phase response and consistent group delay of the signals. The bandpass filters in the bandpass filter bank are Finite Impulse Response (FIR) digital filters.

[0037] The design of bandpass filters in a bandpass filter bank must meet the following conditions to ensure the effectiveness of subsequent coherent processing: the phase response of the bandpass filter is linearly related to the frequency to ensure that the phase of the signal is not distorted after filtering; the group delay deviation of the bandpass filter corresponding to each channel is ≤10ns to avoid time synchronization errors in multi-channel signals; the stopband attenuation of the bandpass filter is ≥60dB to ensure the isolation between channels and effectively suppress cross-interference.

[0038] In a specific application example, the bandwidth of each bandpass filter in the bandpass filter bank is 1.1 to 1.3 times the bandwidth of the corresponding orthogonal frequency division linear frequency modulated (OFDM) signal. The order of the bandpass filters ranges from 16th to 64th, with passband ripple ≤0.1dB and stopband attenuation ≥40dB. The baseline distance between each radar unit is 10m to 1000m, the carrier frequency range of the transmitted OFDM signal is 100MHz to 10GHz, the bandwidth is 1MHz to 10MHz, and the pulse width is 0.5µs to 5µs.

[0039] After bandpass filtering, the echo signals from the different channels are input into the pulse compression neural network. This structural design can realize the separation of channel separation and pulse compression functions.

[0040] Step 102: For any single-channel echo signal, a pulse compression neural network is used to determine the probability distribution of the signal peak value in each distance cell of the single-channel echo signal.

[0041] First, the single-channel echo signal is standardized by scaling the signal amplitude to the [-1,1] range to match the input requirements of the pulse compression neural network. Then, the signal is input into the pulse compression neural network for compression processing.

[0042] In a specific application example, such as Figure 3 As shown, the pulse compression neural network includes an input layer, a hidden layer, and an output layer.

[0043] The input layer is used to receive the real and imaginary components of the single-channel echo signal to obtain a two-dimensional input tensor. The dimension of the input layer is N×1, where N is the number of signal sampling points.

[0044] The hidden layer is used to extract the nonlinear features of the two-dimensional input tensor. The dimension of the hidden layer is 2×K, where K is the number of neurons in each hidden layer. To achieve real-time processing, the number of neurons in the hidden layer used in this application is 80 to 120. The activation function of the hidden layer is the Sigmoid function, whose expression is: This function is characterized by continuous differentiability and strong nonlinear mapping ability, and can effectively extract the nonlinear features of signals. Here, x is the Sigmoid function, and x is the input parameter of the Sigmoid function.

[0045] The output layer is used to convert the nonlinear features into a normalized probability distribution, obtaining the probability distribution of the signal peak in each range cell of the single-channel echo signal. The output layer has an N×1 dimension, consistent with the number of range cells, and its activation function is the Softmax function, expressed as follows: , For the first i Nonlinear characteristics corresponding to each signal sampling point For the first n The nonlinear features correspond to each signal sampling point. The Softmax function can convert the network's original output into a normalized probability distribution, making it easier to directly determine the location of the signal peak. The output layer's role is to map the nonlinear features to the probability distribution of the signal peak in each distance unit, thus achieving peak localization.

[0046] In this application, the pulse compression neural network transforms the pulse compression problem into a multi-classification problem by directly outputting the probability distribution of peak positions, thus avoiding the error accumulation in peak detection in traditional methods. Furthermore, through data-driven training, it can adapt to linear frequency modulated signals with different bandwidths, pulse widths, and noise environments, exhibiting strong generalization ability. Additionally, the network structure is relatively simple, with a small number of parameters, and supports real-time inference.

[0047] The pulse compression neural network is obtained by pre-training a feedforward neural network using a training sample set, which includes linear frequency modulated signals with different peak positions and signal-to-noise ratio ranges, along with corresponding distance unit position labels. The inference process of the pulse compression neural network is accelerated by hardware using a field-programmable gate array (FPGA) or a graphics processing unit (GPU), with a processing latency of ≤1ms, and supports online updates of network parameters.

[0048] The training process of the pulse compression neural network includes the following steps 21 to 24.

[0049] Step 21: Generate linear frequency modulated (LFM) signals with different peak positions and signal-to-noise ratio ranges. Taking the orthogonal frequency division linear frequency modulated (OFDM) signal widely used in distributed coherent radar as an example, the signal model is as follows: ,in, It is an orthogonal frequency division linear frequency modulation signal. The pulse width is rectangular pulse, t For time, The pulse width. The center frequency of the carrier. j It is an imaginary number. For frequency modulation coefficients, , B For bandwidth.

[0050] Step 22: The peak value of the linear frequency modulated (LFM) signal is located to all distance cells through a cyclic shift operation. The distance cell position labels of the LFM signal are determined, and signal samples covering all distance cells are generated. This ensures that the feedforward neural network can learn the peak features at different locations, such as... Figure 4 As shown.

[0051] Step 23: Add Gaussian white noise to each signal sample and perform data augmentation to obtain the training sample set. By adding Gaussian white noise with a signal-to-noise ratio ranging from -20dB to +10dB to each signal sample, interference factors such as transmission noise and receiver internal noise in the actual electromagnetic environment are simulated. The size of the dataset is expanded through data augmentation operations such as random flipping, amplitude scaling, and phase shifting to avoid overfitting of the feedforward neural network and enhance its environmental adaptability.

[0052] The generated training sample set contains 10,000 to 50,000 samples, including 10% zero-signal samples, which are used to train the network's ability to recognize targetless scenes.

[0053] Step 24: Based on the training sample set, the quantized conjugate gradient algorithm is used as the optimizer, and the cross-entropy loss function is used as the training objective function to iteratively train the feedforward neural network to obtain the pulse compression neural network.

[0054] The expression for the cross-entropy loss function is: ,in, The loss value. M The total number of training samples, For the first m The true label values ​​of each training sample are set, with the label set to 1 for the distance cell corresponding to the peak position and 0 for the other positions. The model predicts the first m The probability of the peak position of each training sample.

[0055] Specifically, the training sample set is divided into a training set, a validation set, and a test set in a 7:1.5:1.5 ratio. The training set is used for updating network parameters, the validation set is used to monitor the training process and prevent overfitting, and the test set is used to evaluate the final performance of the network. The feedforward neural network is trained using the cross-entropy loss function and the quantized conjugate gradient algorithm. The number of training iterations ranges from 20 to 500 epochs. Training stops when the validation set loss shows no decrease for 5 consecutive epochs, and the current optimal network parameters are saved.

[0056] After training the model according to the different actual operating frequencies of the distributed coherent radar, the trained pulse compression neural network is deployed in the distributed coherent radar receiving link.

[0057] Step 103: Determine the peak position based on the probability distribution of the signal peak in each distance cell of the single-channel echo signal, so as to obtain the low sidelobe pulse compression signal corresponding to the single-channel echo signal.

[0058] Step 104: Perform coherent synthesis processing on the low sidelobe pulse compression signals corresponding to each single-channel echo signal to obtain the target detection result.

[0059] In a specific application example, step 104 includes steps 41 to 44.

[0060] Step 41: For any single-channel echo signal, determine the coherent parameters of the single-channel echo signal based on the peak position of the low-sidelobe pulse compression signal corresponding to the single-channel echo signal. Specifically, a combined method of "peak detection + cross-correlation" is used to estimate the coherent parameters of each single-channel echo signal. The coherent parameters include time delay. and phase .

[0061] Step 42: Based on the coherent parameters of the single-channel echo signal, perform time alignment and phase compensation on the low sidelobe pulse compression signal corresponding to the single-channel echo signal to obtain the compensation signal corresponding to the channel echo signal.

[0062] Step 43: Superimpose and synthesize the compensation signals corresponding to each single-channel echo signal to achieve a synthesis gain improvement and obtain the synthesized signal.

[0063] Step 44: Based on the synthesized signal, an adaptive threshold adjustment strategy is used to dynamically adjust the detection threshold based on the environmental noise power to determine the target detection result.

[0064] Specifically, the synthesized signal is input into the detector, and an adaptive threshold adjustment strategy is adopted to dynamically adjust the detection threshold based on the ambient noise power, and the target detection result is output to realize the complete workflow of distributed coherent radar.

[0065] This application addresses the technical shortcomings of existing distributed coherent radars, such as severe sidelobe interference from traditional matched filters and the inability of existing neural network pulse compression methods to achieve multi-channel signal separation. By introducing a bandpass filtering step to improve the received signal processing link, designing a lightweight feedforward neural network, constructing an adaptive training dataset, and optimizing the pulse compression process, this application achieves the coordinated processing of multi-channel signal separation and low-sidelobe pulse compression. It is suitable for target detection scenarios requiring high resolution and low interference in distributed coherent radars, and can improve the target detection accuracy and environmental adaptability of distributed coherent radars, showing broad application prospects.

[0066] Compared with existing technologies, this application improves radar sidelobe suppression performance, increases the coherent synthesis efficiency of distributed coherent radar, has high real-time processing performance and strong generalization ability, and is robust in complex environments.

[0067] The following example illustrates the processing flow of a low sidelobe pulse compression method for distributed coherent radar. Table 1 shows the system parameter configuration for distributed coherent radar pulse compression.

[0068] Table 1. Specific parameter configurations for the distributed coherent radar pulse compression embodiment.

[0069] Step 1: Construct a receive signal processing link that is independent of channel separation and pulse compression.

[0070] like Figure 2 As shown, a distributed coherent reception link is constructed, from signal reception → down-conversion → bandpass filtering → neural network pulse compression → coherent parameter estimation → coherent synthesis → target detection. A bandpass filter is added to separate the functions of channel separation and pulse compression.

[0071] During signal reception, the receiving antennas of the three radar stations simultaneously receive the mixed echo signal reflected by the target. After being amplified by a low-noise amplifier, the signal is mixed with the signal of the local oscillator (LO) by the downconverter, and the radio frequency signal is downconverted to baseband to remove the influence of the carrier frequency, thus obtaining the mixed baseband echo signal.

[0072] Next, the hybrid baseband echo signal is input into a bandpass filter bank. The center frequency of each bandpass filter corresponds to the carrier frequency of the transmitted signal from the three stations. By filtering and suppressing interference from non-target channel signals, the system separates three single-channel echo signals. The three separated single-channel echo signals are then input into the subsequent standardization processing module to complete the connection with the neural network pulse compression module, achieving the design goal of channel separation and pulse compression separation.

[0073] Step 2: Design a lightweight feedforward neural network. For example... Figure 3 As shown, the feedforward neural network constructed in this step consists of an input layer, a hidden layer, and an output layer. The specific design and implementation process is as follows.

[0074] Input layer design: The input layer has a dimension of N=128 and is divided into real input branches and imaginary input branches, which respectively receive the real and imaginary components of the single-channel echo signal separated in step 1 to form a two-dimensional input vector.

[0075] Hidden layer design: The hidden layer has a dimension of K=100×2, and the activation function is the Sigmoid function. 100 Sigmoid neurons are constructed using PyTorch to extract nonlinear features from the input signal, providing feature support for peak location detection.

[0076] Output layer design: The output layer has a dimension of N=128 and uses the Softmax function as the activation function. The features output by the hidden layer are mapped to the normalized probability distribution of the signal peak in N distance units, which is used to directly locate the peak position in the subsequent process.

[0077] Step 3: Generate an adaptive training dataset.

[0078] First, a reference signal is generated; then, based on the orthogonal frequency division linear frequency modulation signal model... , where pulse width Take 2μs, the carrier center frequency of sites 1 to 3 We take 1GHz, 1.01GHz, and 1.02GHz respectively. The bandwidth in the frequency modulation coefficient. Choose 5MHz.

[0079] Next, the sample expansion process is as follows: Figure 4 As shown, the generated reference signal is cyclically shifted to sequentially locate the signal peak at each of the 128 distance cells, generating 128 signal samples at different peak positions, ensuring that the network can learn the peak features across the entire distance domain.

[0080] Next, noise is added to each signal sample by adding Gaussian white noise with a signal-to-noise ratio ranging from -20dB to 10dB to simulate interference such as transmission noise and receiver internal thermal noise in the actual electromagnetic environment, thereby improving the robustness of the network. Data augmentation is then performed by expanding the diversity of samples through operations such as random flipping, amplitude scaling, and phase shifting to avoid overfitting during training.

[0081] Finally, the training sample set is constructed by integrating the samples generated in the previous process to generate a total of 10,000 training sample sets, including 500 signal samples with all amplitudes being zero, which are used to train the network's ability to recognize targetless scenes.

[0082] Step 4: Train the pulse compression neural network. The 10,000 training samples built in Step 3 are proportionally divided into a training set of 7,500 samples, a validation set of 1,500 samples, and a test set of 1,500 samples. The training set is used for parameter updates, the validation set is used to monitor overfitting, and the test set is used to evaluate the final performance. The training process uses the quantized conjugate gradient algorithm as the optimizer.

[0083] Step 5: Test the received echo signal and perform pulse compression and target detection.

[0084] Three radar stations simultaneously transmit orthogonal frequency division linear frequency modulated (OFDM) signals. The carrier frequency and bandwidth of each signal are allocated according to the preset parameters in Table 1 to ensure signal orthogonality and avoid interference at the transmitting end. The receiving antenna receives the mixed echo signal reflected from the target, amplifies it, down-converts it to baseband, and then inputs it into the bandpass filter bank in the radar receiving processing link designed in step 1 to separate three single-channel echo signals and suppress cross-interference. The amplitude of the three single-channel echo signals is scaled to the [-1,1] interval to adapt to the input requirements of the pulse compression neural network. The standardized signal is input into the pulse compression neural network, which outputs the peak position probability distribution to locate the peak and obtain a low sidelobe pulse compression signal.

[0085] Figure 5The time-domain compression effect of station 1 in this embodiment is shown in the comparison. The pulse compression output sidelobe amplitude of this application (black curve) is significantly lower than that of the traditional matched filtering method (blue curve) and Hamming window method (red curve), and the main lobe is narrower and the peak is sharper.

[0086] The peak detection method is further employed to estimate the time delay and phase of the three signals. Based on the estimated time delay and phase, the three compressed signals are time-aligned and phase-compensated before being superimposed and synthesized. The synthesized signal is input into a detector, and an adaptive threshold adjustment strategy is used to dynamically adjust the detection threshold based on the ambient noise power. Figure 6 The distance-Doppler image comparison shown is as follows: Figure 6 The three sub-figures on the left, representing traditional matched filter structures, exhibit significant sidelobe interference in their distance-Doppler plots. Figure 6 The middle part is characterized by a large amount of yellow trailing against a blue background; while Figure 6 The three sub-figures on the right represent the application where the sidelobe interference of the distance-Doppler image is significantly suppressed, the target location is clearer, and the interference suppression effect is significant.

[0087] In summary, this embodiment verifies the effectiveness of this application in distributed coherent radar scenarios, realizing the coordinated processing of channel separation and low sidelobe pulse compression, which significantly improves target detection accuracy and environmental adaptability.

[0088] Based on the same inventive concept, this application also provides an apparatus for implementing the method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, specific limitations in one or more apparatus embodiments provided below can be found in the limitations of the method described above, and will not be repeated here.

[0089] In one exemplary embodiment, such as Figure 7 As shown, a low sidelobe pulse compression device for distributed phased coherent radar is provided, which includes the following functional modules.

[0090] The channel separation module 701 is used to separate the mixed target echo signals received by the distributed phased coherent radar into multiple single-channel echo signals.

[0091] The pulse compression module 702 is used to determine the probability distribution of the signal peak value in each range cell of any single-channel echo signal using a pulse compression neural network. The pulse compression neural network is obtained by pre-training a feedforward neural network with a training sample set, which includes linear frequency modulated signals with different peak positions and different signal-to-noise ratio ranges, along with corresponding range cell position labels.

[0092] The compressed signal generation module 703 is used to determine the peak position based on the probability distribution of the signal peak in each distance unit in the single-channel echo signal, so as to obtain the low sidelobe pulse compressed signal corresponding to the single-channel echo signal.

[0093] The coherent synthesis module 704 is used to perform coherent synthesis processing on the low sidelobe pulse compression signals corresponding to each single-channel echo signal to obtain the target detection result.

[0094] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0095] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0096] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0098] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0099] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0100] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0101] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A low-sidelobe pulse compression method for distributed coherent radar, characterized in that, The low sidelobe pulse compression method of the distributed coherent radar includes: Channel separation is performed on the mixed target echo signals received by the distributed coherent radar to obtain multiple single-channel echo signals; For any single-channel echo signal, a pulse compression neural network is used to determine the probability distribution of the signal peak value in each distance cell of the single-channel echo signal; the pulse compression neural network is obtained by training a feedforward neural network in advance using a training sample set, the training sample set including linear frequency modulated signals with different peak positions and different signal-to-noise ratio ranges and corresponding distance cell position labels; Based on the probability distribution of the signal peak in each distance cell in the single-channel echo signal, the peak position is determined to obtain the low sidelobe pulse compression signal corresponding to the single-channel echo signal. The low sidelobe pulse compression signals corresponding to each single-channel echo signal are subjected to coherent synthesis processing to obtain the target detection results.

2. The low sidelobe pulse compression method for distributed phased coherent radar according to claim 1, characterized in that, The hybrid target echo signal includes the radio frequency echo signal received by each radar unit in the distributed phased coherent radar; Channel separation is performed on the mixed target echo signals received by the distributed coherent radar to obtain multiple single-channel echo signals, including: The hybrid target echo signal is amplified and then downconverted to baseband to obtain a hybrid baseband echo signal; The hybrid baseband echo signal is subjected to finite impulse response bandpass filtering using a bandpass filter bank to obtain multiple single-channel echo signals.

3. The low sidelobe pulse compression method for distributed phased coherent radar according to claim 2, characterized in that, The center frequency of the bandpass filter bank corresponds one-to-one with the center frequency of the orthogonal frequency division linear frequency modulated signal transmitted by each radar unit of the distributed coherent radar.

4. The low sidelobe pulse compression method for distributed coherent radar according to claim 2, characterized in that, The phase response of each bandpass filter in the bandpass filter bank is linearly related to the frequency, and the group delay deviation of the bandpass filter is ≤10ns, and the stopband attenuation is ≥60dB.

5. The low sidelobe pulse compression method for distributed phased coherent radar according to claim 1, characterized in that, The pulse compression neural network includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the real and imaginary components of a single-channel echo signal to obtain a two-dimensional input tensor. The hidden layer is used to extract the nonlinear features of the two-dimensional input tensor; The output layer is used to convert the nonlinear features into a normalized probability distribution to obtain the probability distribution of the signal peak in each distance cell in the single-channel echo signal.

6. The low sidelobe pulse compression method for distributed phased coherent radar according to claim 1, characterized in that, The training process of the pulse compression neural network includes: Generate linear frequency modulated signals with different peak positions and different signal-to-noise ratio ranges; The peak value of the linear frequency modulated signal is located to all distance cells through a cyclic displacement operation, the distance cell position label of the linear frequency modulated signal is determined, and a signal sample covering all distance cells is generated. Gaussian white noise is added to each signal sample and data augmentation is performed to obtain the training sample set; Based on the training sample set, the quantized conjugate gradient algorithm is used as the optimizer, and the cross-entropy loss function is used as the training objective function to iteratively train the feedforward neural network to obtain the pulse compression neural network.

7. The low sidelobe pulse compression method for distributed phased coherent radar according to claim 1, characterized in that, The low-sidelobe pulse compression signals corresponding to each single-channel echo signal are coherently synthesized to obtain the target detection results, including: For any single-channel echo signal, the coherent parameters of the single-channel echo signal are determined based on the peak position of the low sidelobe pulse compression signal corresponding to the single-channel echo signal. Based on the coherent parameters of the single-channel echo signal, the low sidelobe pulse compression signal corresponding to the single-channel echo signal is time-aligned and phase-compensated to obtain the compensation signal corresponding to the single-channel echo signal. The compensation signals corresponding to each single-channel echo signal are superimposed and synthesized to obtain the composite signal; Based on the synthesized signal, an adaptive threshold adjustment strategy is used to dynamically adjust the detection threshold based on the environmental noise power to determine the target detection result.

8. The low sidelobe pulse compression method for distributed phased coherent radar according to claim 7, characterized in that, The coherent parameters include time delay and phase.

9. A low sidelobe pulse compression device for a distributed coherent radar, characterized in that, The low-sidelobe pulse compression device of the distributed coherent radar performs the low-sidelobe pulse compression method of the distributed coherent radar according to any one of claims 1-8, and the low-sidelobe pulse compression device of the distributed coherent radar comprises: The channel separation module is used to separate the mixed target echo signals received by the distributed phased coherent radar into multiple single-channel echo signals. The pulse compression module is used to determine the probability distribution of the signal peak in each distance cell of any single-channel echo signal using a pulse compression neural network. The pulse compression neural network is obtained by training a feedforward neural network with a training sample set, which includes linear frequency modulated signals with different peak positions and different signal-to-noise ratio ranges, as well as corresponding distance cell position labels. The compressed signal generation module is used to determine the peak position based on the probability distribution of the signal peak in each distance unit in the single-channel echo signal, so as to obtain the low sidelobe pulse compressed signal corresponding to the single-channel echo signal. The coherent synthesis module is used to perform coherent synthesis processing on the low sidelobe pulse compression signals corresponding to each single-channel echo signal to obtain the target detection result.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the low sidelobe pulse compression method for distributed phased-coherent radar according to any one of claims 1-8.

Images