Distance Doppler processing method and device of target and storage medium

Abstract

The invention provides a target distance Doppler processing method and device and a storage medium, and the method comprises the steps: carrying out the pulse sampling of a radar echo signal of a target, and determining radar observation data in a one-dimensional vector form; reconstructing the radar observation data according to the number of pulses and the number of sampling times when pulse sampling is carried out on the radar echo signal, and determining a two-dimensional observation matrix corresponding to the radar observation data; the sampling frequency is greater than the pulse number; dividing the two-dimensional observation matrix into a plurality of batches of reconstruction signals according to computing power parameters of the GPU; each batch of reconstruction signal comprises multiple columns of data in the two-dimensional observation matrix; and performing distance Doppler signal recovery processing on the reconstructed signals of each batch in parallel by using a GPU, determining a target recovery signal corresponding to the reconstructed signals of each batch, and determining a distance Doppler processing result corresponding to the target according to the target recovery signals of each batch. By adopting the technical scheme of the invention, the efficiency of distance Doppler processing can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing, and in particular to a method, device and storage medium for processing range Doppler of a target. Background Art

[0002] In radar signal processing, when facing high-throughput radar systems, it is generally necessary to process large-scale observation data and recover target features from them. High-throughput radar systems can generate huge data streams. Especially in the detection of high-speed dynamic targets, frequent and large amounts of measurement data / observation data will be generated, and target Doppler distance processing must be performed on a large amount of observation data to recover target features from them.

[0003] In the related art, during the range-Doppler processing of a target, a central processing unit (CPU) generally performs range processing and Doppler processing on a large amount of observation data in sequence, and finally restores the range-Doppler information of the target.

[0004] However, the above technology has a problem of slow processing speed. Summary of the Invention

[0005] The present invention provides a method, device and storage medium for range Doppler processing of a target, which are used to overcome the drawback of slow processing speed in the prior art when performing range Doppler processing on observation data of a high-throughput radar system. The method achieves this by sampling and reconstructing radar echo signals into a two-dimensional observation matrix consisting of fast time and slow time, and dividing the two-dimensional observation matrix into multiple batches of reconstructed signals based on the computing power of a GPU. The GPU is then used to simultaneously perform range Doppler signal recovery processing on multiple columns of data in each batch of reconstructed signals in parallel, thereby accelerating the speed of range Doppler processing and improving the efficiency of range Doppler processing.

[0006] The present invention provides a range Doppler processing method for a target, comprising the following steps.

[0007] Acquire the radar echo signal of the target and perform pulse sampling on the radar echo signal to determine the radar observation data in the form of a one-dimensional vector; Reconstructing radar observation data based on the number of pulses and the number of sampling times when pulse sampling the radar echo signal, and determining a two-dimensional measurement matrix corresponding to the radar observation data; the number of rows of the two-dimensional measurement matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional measurement matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; Divide the two-dimensional measurement matrix into a plurality of batches of first reconstructed signals according to the computing power parameter of the GPU; each batch of the first reconstructed signals includes a plurality of columns of data in the two-dimensional measurement matrix; GPUs are used to perform range Doppler signal recovery processing on the first reconstructed signals of each batch in parallel, and the first target recovery signals corresponding to the first reconstructed signals of each batch are determined. The range Doppler processing results corresponding to the targets are determined based on the first target recovery signals of each batch.

[0008] According to a target range Doppler processing method provided by the present invention, the GPU is used in parallel to perform range Doppler signal recovery processing on the first reconstructed signals of each batch, and the first target recovery signal corresponding to the first reconstructed signals of each batch is determined, including: According to the sampling times of the radar echo signal, a range-dimensional measurement matrix is ​​constructed; the number of rows of the above-mentioned range-dimensional measurement matrix is ​​related to the sampling times, and the columns of the range-dimensional measurement matrix represent the single pulse echo signal at the time corresponding to each sampling time; Constructing a first compressed sensing model corresponding to the first reconstructed signal of each batch according to the distance dimension measurement matrix and the first reconstructed signal of each batch; Determine initialization parameters corresponding to the first compressed sensing model of each batch; According to the initialization parameters of the first compressed sensing model of each batch, the gradient descent optimization algorithm and GPU parallelism are used to solve the first reconstructed signal in the first compressed sensing model of each batch to determine the first target recovery signal corresponding to the first reconstructed signal of each batch.

[0009] According to a range Doppler processing method for a target provided by the present invention, the initialization parameters include an initial recovery signal and an initial auxiliary variable. Based on the initialization parameters of the first compressed sensing model of each batch, a gradient descent optimization algorithm and GPU parallelism are used to solve the first reconstructed signal in the first compressed sensing model of each batch, and determine the first target recovery signal corresponding to the first reconstructed signal of each batch, including: For each batch of first compressed sensing models, initializing the corresponding first compressed sensing models using an initial recovery signal and an initial auxiliary variable; the initial auxiliary variable is determined according to the initial recovery signal; Performing an iterative operation, the iterative operation comprising: solving the first reconstructed signal in the first compressed sensing model using a gradient descent optimization algorithm according to the first reconstructed signal, the distance-dimensional measurement matrix, and the first auxiliary variable of the current iteration round to obtain an intermediate variable in the current iteration round; determining a first restored signal in the current iteration round according to the intermediate variable in the current iteration round; and updating the first auxiliary variable according to the first restored signal in the current iteration round to determine a second auxiliary variable; Determine whether the second auxiliary variable of the current iteration round meets the first accuracy requirement. If not, use the second auxiliary variable as the new first auxiliary variable, return to perform the above iterative operation until the second auxiliary variable of the current iteration round meets the first accuracy requirement, and determine the second auxiliary variable that meets the first accuracy requirement as the first target recovery signal corresponding to the first reconstructed signal of the corresponding batch.

[0010] According to a range Doppler processing method for a target provided by the present invention, the method of determining the first recovered signal of the current iteration round according to the intermediate variable in the current iteration round includes: Determine the step size parameter corresponding to the gradient descent optimization algorithm and the regularization parameter corresponding to the range Doppler signal recovery process; Determine the signal comparison threshold based on the step size parameter and the regularization parameter; the regularization parameter is related to the sparsity of the restored signal; The first restored signal of the current iteration round is determined according to the intermediate variables and the signal comparison threshold in the current iteration round.

[0011] According to a target range Doppler processing method provided by the present invention, the above-mentioned updating of the first auxiliary variable and determining the second auxiliary variable according to the first recovered signal of the current iteration round include: Obtaining the first recovery signal of the previous iteration round of the current iteration round; Calculate the momentum adjustment parameter for the next iteration round based on the momentum adjustment parameter for the current iteration round; The first auxiliary variable is updated and the second auxiliary variable is determined according to the first recovery signal of the current iteration round, the first recovery signal of the previous iteration round, the momentum adjustment parameter of the current iteration round, and the momentum adjustment parameter of the next iteration round.

[0012] According to a range Doppler processing method for a target provided by the present invention, the GPU is used in parallel to perform range Doppler signal recovery processing on the first reconstructed signals of each batch, determine the first target recovery signal corresponding to the first reconstructed signal of each batch, and determine the range Doppler processing result corresponding to the target based on the first target recovery signal of each batch, including: Using GPUs to perform range signal recovery processing on the first reconstructed signals of each batch in parallel, and determining the first target recovery signal corresponding to the first reconstructed signals of each batch; The first target recovery signals of each batch are spliced ​​together to determine the range recovery signal; the number of rows and columns of the range recovery signal is the same as the number of rows and columns of the two-dimensional measurement matrix; Perform Doppler signal recovery processing on the range recovery signal to determine the range Doppler processing result corresponding to the target.

[0013] The application provides a target range-Doppler processing method, which comprises the following steps: The range recovery signal is transposed to obtain a transposed range recovery signal, and the transposed range recovery signal is divided into a plurality of batches of second reconstruction signals according to the GPU computing power parameter; each batch of second reconstruction signals comprises a plurality of columns of data in the transposed range recovery signal; The GPU is used to perform Doppler signal recovery processing on each batch of second reconstruction signals in parallel to determine a second target recovery signal corresponding to each batch of second reconstruction signals, and the target range-Doppler processing result is determined according to the second target recovery signals of the batches.

[0014] The application provides a target range-Doppler processing method, which comprises the following steps: A Doppler-velocity observation matrix is constructed by using a dftmtx function in MATLAB according to the number of pulses of the radar echo signal; the number of rows and columns of the Doppler-velocity observation matrix is the same as the number of pulses; A second compressed sensing model corresponding to each batch of second reconstruction signals is constructed according to the range dimension observation matrix and the second reconstruction signals of each batch; An initialization parameter corresponding to each batch of second compressed sensing models is determined; The second reconstruction signals in each batch of second compressed sensing models are solved by using a gradient descent optimization algorithm and the GPU in parallel according to the initialization parameter of each batch of second compressed sensing models, so that a second target recovery signal corresponding to each batch of second reconstruction signals is determined.

[0015] The application further provides a target range-Doppler processing device, which comprises the following modules: An echo signal sampling module is configured to acquire radar echo signals of a target and perform pulse sampling on the radar echo signals to determine radar observation data in the form of a one-dimensional vector; An echo signal reconstruction module is configured to reconstruct the radar observation data according to the number of pulses and the sampling times when the pulse sampling is performed on the radar echo signals to determine a two-dimensional observation matrix corresponding to the radar observation data; the number of rows of the two-dimensional observation matrix is related to the sampling times, the number of columns of the two-dimensional observation matrix is related to the number of pulses, and the sampling times are greater than the number of pulses; A batch division module is configured to divide the two-dimensional observation matrix into a plurality of batches of first reconstruction signals according to the GPU computing power parameter; each batch of first reconstruction signals comprises a plurality of columns of data in the two-dimensional observation matrix; The distance Doppler processing module is configured to perform distance Doppler signal recovery processing on each batch of the first reconstructed signals in parallel by using the GPU, determine a first target recovery signal corresponding to each batch of the first reconstructed signals, and determine a distance Doppler processing result corresponding to the target according to the first target recovery signals of the batches.

[0016] The application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the distance Doppler processing method of any of the above targets when executing the computer program.

[0017] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program implements the distance Doppler processing method of any of the above targets when executed by a processor.

[0018] The application further provides a computer program product, which comprises a computer program, and the computer program implements the distance Doppler processing method of any of the above targets when executed by a processor.

[0019] The distance Doppler processing method, device, and storage medium provided by the application determine radar observation data in the form of a one-dimensional vector by obtaining a radar echo signal of a target and performing pulse sampling on the radar echo signal, reconstruct the radar observation data according to the number of samplings and the number of pulses when the radar echo signal is pulse sampled, determine a two-dimensional observation matrix, divide the two-dimensional observation matrix into reconstructed signals of multiple batches according to the computing power parameters of a GPU, perform distance Doppler signal recovery processing on each batch of the reconstructed signals in parallel by using the GPU, determine a target recovery signal corresponding to each batch of the reconstructed signals, and determine a distance Doppler processing result of the target according to the target recovery signals of the batches. The number of rows of the two-dimensional observation matrix is related to the number of samplings, and the number of columns of the two-dimensional observation matrix is related to the number of pulses. The number of samplings is greater than the number of pulses, and each batch of the reconstructed signals comprises multiple column data of the two-dimensional observation matrix. In the method, the radar echo signal can be sampled and reconstructed into a two-dimensional observation matrix composed of fast time and slow time, and the two-dimensional observation matrix can be divided into reconstructed signals of multiple batches according to the computing power of the GPU, so that the multiple column data in the reconstructed signals of the batches can be simultaneously subjected to distance Doppler signal recovery processing in parallel by using the GPU, the speed of distance Doppler processing of the target is accelerated, the efficiency of distance Doppler processing of the target is improved, and the distance Doppler processing is suitable for high-throughput radar systems. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the 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 one of the flow charts of the range Doppler processing method for a target provided by the present invention.

[0022] Figure 2 This is the second flow chart of the range Doppler processing method for a target provided by the present invention.

[0023] Figure 3 It is a schematic diagram of the range Doppler processing result of the target provided by the present invention.

[0024] Figure 4 It is a detailed flow chart of the range Doppler processing method of a target provided by the present invention.

[0025] Figure 5 It is a structural schematic diagram of the range Doppler processing device of the target provided by the present invention.

[0026] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0028] In radar signal processing, especially when dealing with high-throughput radar systems, processing large-scale observation data and recovering target features from it is an extremely challenging task. High-throughput radar systems can generate huge data streams. Especially in the detection of high-speed dynamic targets, the frequent and large amount of measurement data makes data processing speed and computational efficiency a bottleneck. In particular, in the process of range Doppler processing of the target, traditional methods face problems such as large computational workload, slow processing speed, and high storage pressure. Furthermore, radar signals are often subject to noise interference, and the target's state exhibits sparseness in the time and frequency domains. For most time periods, the target signal's energy is negligible, and the target's characteristics only become apparent at specific times and frequency ranges. Consequently, radar signals exhibit high redundancy and sparsity. In light of these characteristics, compressed sensing (CS), a general signal processing technique, has been gradually introduced into radar data processing. CS exploits the sparsity of signals to recover them with fewer sampled data. This theoretically significantly reduces storage and transmission requirements, and particularly in high-throughput data processing, CS methods demonstrate superior efficiency compared to traditional methods, enabling them to reduce data acquisition while maintaining high signal recovery accuracy. However, the CS computational process itself involves extensive optimization calculations, significantly increasing its computational complexity in practical applications, especially when processing large amounts of data, thus hindering its effectiveness in high-throughput environments. In addition, GPU (Graphic Processing Unit) acceleration methods are applied to the field of high-throughput data processing. Through parallel computing, GPU acceleration can significantly increase processing speed and shorten the time of the entire processing flow, showing advantages over traditional computing methods. However, although existing GPU acceleration methods have improved computing speed, there is still room for optimization. In particular, when processing high-throughput radar data, its acceleration effect has not reached the optimal level.

[0029] It can be seen that the current technology has the problem of slow data processing speed for high-throughput radar systems. Based on this, the embodiments of the present invention provide a target range Doppler processing method, device and storage medium to solve this technical problem.

[0030] It should be noted that the execution subject of the embodiments of the present invention can be a range Doppler processing device of the target, or an electronic device, or a radar system, or other devices or equipment. The following embodiments will be described using an electronic device as an example of the execution subject. The electronic device can be an independent electronic device or an electronic device in a radar system, and no specific limitation is made here.

[0031] Figure 1This is one of the flow charts of the range Doppler processing method of the target provided by the present invention, such as Figure 1 As shown, the method includes the following steps: Step 102 : Acquire the radar echo signal of the target and perform pulse sampling on the radar echo signal to determine radar observation data in the form of a one-dimensional vector.

[0032] In this embodiment, the radar echo signal may be generated by a signal transmitting module in the radar system transmitting a signal to a target, and then a signal receiving module receiving the signal reflected from the target. This reflected signal is the radar echo signal. The radar system may be a high-throughput radar system or a non-high-throughput radar system, and the processed radar echo data may be high-throughput data or non-high-throughput data.

[0033] In addition, the transmitted signal in this embodiment can be a pulse signal, or alternatively, a multi-pulse chirp signal, with the number of pulses being, for example, 2048. In this embodiment, the initial target distance can also be pre-set to 10,000 meters and the speed to -100 meters / second. This setting allows the acquisition of a radar echo signal from the target. Sampling can then be performed within a pulse time (i.e., in the time domain) at a preset sampling frequency (e.g., 8192 times). After complete echo sampling, radar observation data in the form of a one-dimensional vector can be obtained. For example, with a pulse number of 2048 and a sampling frequency of 8192, radar observation data of a scale of 16,777,216 × 1 can be obtained after sampling.

[0034] Furthermore, in order to better simulate the actual radar echo signal, Gaussian white noise with a noise variance of 0.1 can be added to the radar observation data to make the radar observation data closer to the actual situation.

[0035] Step 104: Reconstruct the radar observation data based on the number of pulses and the number of sampling times when pulse sampling the radar echo signal, and determine the two-dimensional observation matrix corresponding to the radar observation data; the number of rows of the above-mentioned two-dimensional observation matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional observation matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses.

[0036] Among them, the radar observation data in the form of a one-dimensional vector can be reconstructed into data in the form of a two-dimensional matrix with fast time and slow time as dimensions through the sampling times of the radar echo signal during pulse sampling and the pulse data, thereby obtaining a two-dimensional observation matrix.

[0037] The number of rows in the two-dimensional measurement matrix is ​​related to the number of sampling times, for example, the number of rows is equal to the number of sampling times, and the number of columns in the two-dimensional measurement matrix is ​​related to the number of pulses, for example, the number of columns is equal to the number of pulse data. The number of sampling times is greater than the number of pulses, that is, the number of rows in the two-dimensional measurement matrix is ​​greater than the number of columns.

[0038] Furthermore, the fast time dimension of the two-dimensional observation matrix corresponds to the matrix columns, and the slow time dimension corresponds to the matrix rows. This reconstruction of the one-dimensional vector radar observation data into a two-dimensional observation matrix with fast and slow time dimensions facilitates subsequent batch processing with GPU frameworks, improving data processing efficiency.

[0039] For example, assuming that the radar observation data is a one-dimensional vector of size 16777216×1, it can be reconstructed into a two-dimensional observation matrix of size 8192×2048.

[0040] Step 106 : dividing the two-dimensional measurement matrix into a plurality of batches of first reconstructed signals according to the computing power parameter of the GPU; each batch of the first reconstructed signals includes a plurality of columns of data in the two-dimensional measurement matrix.

[0041] In this embodiment, the range-Doppler processing of the target is performed in the MATLAB (Matrix Lab) software environment, i.e., the operating platform is MATLAB. The actual processing process is demonstrated through simulation examples within this operating platform. Therefore, it is necessary to first ensure that the MATLAB environment has the Parallel Computing Toolbox add-on, which is a fundamental tool for GPU-based data processing. Next, all data (i.e., the two-dimensional observation matrix) is converted from CPU (Central Processing Unit) format to GPU format using the gpuArray (GPU array) function built into MATLAB. It should be noted that the radar echo signal, radar observation data, and two-dimensional observation matrix in steps 102 and 104 are all processed by the CPU and are therefore CPU-formatted data. Subsequent parallel processing using the GPU requires data conversion to GPU format.

[0042] Before using a GPU to process high-throughput data such as a two-dimensional observation matrix, you can first obtain the GPU's computing power parameters and use them to determine the amount of data the GPU can process simultaneously. The GPU's computing power parameters can include parameters such as the GPU's parallel computing capability and the amount of video memory on the device where the GPU is located.

[0043] Furthermore, before converting the data from CPU format to GPU format, the data processing precision can be selected based on the computing capabilities of the graphics card of the device hosting the GPU for data of varying precision. The data in the two-dimensional observation matrix can then be converted to the same precision based on the selected data processing precision before the GPU format conversion is performed. For example, the GPU in an NVIDIA GeForce RTX 4050 notebook has a single-precision computing capability of approximately 8.986 TFLOPS (teraflops), while its double-precision computing capability is only approximately 140.4 GFLOPS (gigaflops). This indicates that single-precision computing efficiency is significantly higher than double-precision computing efficiency. Therefore, in this embodiment, the data in the two-dimensional observation matrix can be converted to single-precision data, significantly improving data processing efficiency.

[0044] After the data in the two-dimensional observation matrix is ​​converted to data processing accuracy and GPU format, the amount of data that the GPU can process in parallel at the same time can be determined based on the GPU's computing power parameters (such as the GPU's parallel computing capability, the size of the video memory of the device where the GPU is located, etc.). The amount of data here refers to the number of columns of column data that can be processed in parallel in the two-dimensional observation matrix. After determining the number of columns that can be processed in parallel each time, optionally, the column data of the two-dimensional observation matrix can be directly batched according to the number of columns, or the column data of the two-dimensional observation matrix after format conversion can also be batched according to the number of columns. Here, after the two-dimensional observation matrix after format conversion is batched, multiple batches of matrices can be obtained, and the matrices of each batch are recorded as first reconstructed signals. Each first reconstructed signal is a two-dimensional matrix, and the number of rows therein is the same as the number of rows of the two-dimensional observation matrix, and each first reconstructed signal includes multiple columns of data in the two-dimensional observation matrix.

[0045] It can be understood that here the column data of the two-dimensional observation matrix is ​​divided into multiple batches of first reconstructed signals according to a certain number of columns, which can ensure that the data volume of each batch of first reconstructed signals during processing can fully utilize the resources of the GPU without exceeding the video memory limit, and the divided data batches can be processed one by one subsequently.

[0046] For example, assuming that during the target distance processing, the GPU processes a batch of first reconstructed signals each time, and the first reconstructed signals include 128 groups / columns of data, that is, the GPU can process 128 groups / columns of data simultaneously / in parallel each time, then after batch division, the scale of the first reconstructed signals obtained in each batch can be 8196×128, and a total of 2048÷128=16 batches are obtained. The scale of data processed each time is, for example, 8192×128. In this case, the utilization rate of the GPU can be maintained between 85% and 90%, which is relatively reasonable.

[0047] Step 108 , using the GPU to perform range Doppler signal recovery processing on the first reconstructed signals of each batch in parallel, determining the first target recovery signal corresponding to the first reconstructed signal of each batch, and determining the range Doppler processing result corresponding to the target based on the first target recovery signal of each batch.

[0048] Among them, after dividing the two-dimensional observation matrix into multiple batches of first reconstructed signals, the GPU can process one batch of the first reconstructed signals at a time and process multiple columns of data in the first reconstructed signals in parallel, which can significantly improve data processing efficiency.

[0049] Specifically, the GPU can perform range-Doppler signal recovery processing on multiple columns of data in a first reconstructed signal in parallel, obtaining a recovered signal corresponding to each first reconstructed signal, recorded as a first target recovered signal. The range-Doppler processing result of the target can then be obtained by combining the first target recovered signals of each batch, or by combining the first target recovered signals of each batch and then further processing them to obtain the range-Doppler processing result of the target. The range-Doppler processing result includes the velocity of the target at a specific distance.

[0050] The range-Doppler signal recovery process described above may include both range signal recovery and Doppler signal recovery. For example, range signal recovery and Doppler signal recovery may be performed simultaneously to obtain a range-Doppler processing result for the target. Alternatively, range signal recovery may be performed first, followed by Doppler signal recovery on the result of the range signal recovery process, ultimately obtaining a range-Doppler processing result for the target. Here, range signal recovery refers to recovering the distance information in the signal to obtain the target's range information, while Doppler signal recovery refers to recovering the speed information or frequency information in the signal to obtain the target's speed information or frequency information given the recovered distance information.

[0051] In this embodiment, the radar echo signal of the target is acquired and pulse sampled to determine the radar observation data in the form of a one-dimensional vector, the radar observation data is reconstructed according to the number of samplings and the number of pulses when the radar echo signal is pulse sampled, a two-dimensional observation matrix is determined, the two-dimensional observation matrix is divided into multiple batches of reconstructed signals according to the computing power parameter of the GPU, the distance Doppler signal recovery processing is performed on each batch of reconstructed signals in parallel by using the GPU, the target recovery signal corresponding to each batch of reconstructed signals is determined, and the distance Doppler processing result of the target is determined according to the target recovery signals of each batch; wherein the number of rows of the two-dimensional observation matrix is related to the number of samplings and the number of columns is related to the number of pulses, the number of samplings is greater than the number of pulses, and each batch of reconstructed signals includes multiple column data of the two-dimensional observation matrix. In this method, the radar echo signal can be sampled and reconstructed into a two-dimensional observation matrix composed of fast time and slow time, and the two-dimensional observation matrix can be divided into multiple batches of reconstructed signals according to the computing power of the GPU, so that the distance Doppler signal recovery processing of the multiple column data in each batch of reconstructed signals can be performed simultaneously by the GPU in parallel, the speed of distance Doppler processing of the target can be accelerated, the efficiency of distance Doppler processing of the target can be improved, and the distance Doppler processing of the high-throughput radar system can be applied.

[0052] The following embodiments describe an implementation of determining the recovery signal through each batch of reconstructed signals.

[0053] In one embodiment, the distance Doppler signal recovery processing on each batch of first reconstructed signals in parallel by using the GPU in step 108 described above to determine the first target recovery signal corresponding to each batch of first reconstructed signals can include: Step A1, constructing a distance dimension observation matrix according to the number of samplings of the radar echo signal; the number of rows of the distance dimension observation matrix is related to the number of samplings, and the column of the distance dimension observation matrix represents the monopulse echo signal at each sampling time.

[0054] In this embodiment, the process of recovering the distance information of the target is described first. As mentioned above, the transmission signal corresponding to the radar echo signal can be a multi-pulse Chirp signal, the multi-pulse Chirp signal can be sampled according to a preset number of samplings, such as 8192 samplings in one pulse time, then the Chirp sampling data of the first pulse is taken as an initial vector, and the distance dimension observation matrix is generated according to the initial vector as a reference. Each column in the distance dimension observation matrix corresponds to the monopulse echo signal that should be obtained at different time delays of the transmission signal.

[0055] The echo signals in the distance dimension observation matrix can reflect the time delay in the signal propagation process, each row of the distance dimension observation matrix represents an echo signal at a time point, and is used to describe the detection unit of the radar system in the distance dimension. In the simulation, a preset number (such as 8192) of distance detection units are divided, and the distance dimension observation matrix can be constructed according to the number of sampling times and the number of divided distance detection units. For example, a distance dimension observation matrix with a size of 8192*8192 can be constructed.

[0056] It can be understood that the distance dimension observation matrix here is a square matrix, that is, the number of rows and the number of columns of the distance dimension observation matrix are equal.

[0057] Further, in order to simplify the difficulty of solving the recovered signal by using the gradient descent optimization algorithm subsequently, the distance dimension observation matrix and the two-dimensional observation matrix described above can be normalized. When performing normalization, the product of the distance dimension observation matrix and the transpose of the distance dimension observation matrix can be calculated first, the eigenvalues of the product are solved, the largest eigenvalue in the solved eigenvalues is selected, and then each element in the distance dimension observation matrix is divided by the square root of the largest eigenvalue to realize the normalization of the distance dimension observation matrix. At the same time, each element in the two-dimensional observation matrix can also be divided by the square root of the largest eigenvalue to realize the normalization of the two-dimensional observation matrix, or each element in the radar observation data can also be divided by the square root of the largest eigenvalue to realize the normalization of the radar observation data. After the above normalization, the largest eigenvalue is set to 1, the eigenvalue range of the distance dimension observation matrix is set to [0, 1], and the eigenvalue range of the two-dimensional observation matrix is set to [0, 1]. Through normalization, the eigenvalue range of the distance dimension observation matrix and the two-dimensional observation matrix can be standardized, which can improve the stability of subsequent calculation and improve the optimization efficiency, and can also simplify the step size parameter setting in the calculation process of the gradient descent optimization algorithm subsequently.

[0058] In addition, as an option, in step 106 of the above embodiment, when the two-dimensional observation matrix is divided into a plurality of batches of first reconstructed signals, the normalized two-dimensional observation matrix can also be divided into a plurality of batches of first reconstructed signals.

[0059] Step A2, constructing a first compressed sensing model corresponding to each batch of first reconstructed signals according to the distance dimension observation matrix and each batch of first reconstructed signals.

[0060] Step A3, determining an initialization parameter corresponding to each batch of first compressed sensing models.

[0061] Step A4: Based on the initialization parameters of the first compressed sensing model of each batch, a gradient descent optimization algorithm and GPU parallelism are used to solve the first reconstructed signal in the first compressed sensing model of each batch to determine the first target recovered signal corresponding to the first reconstructed signal of each batch.

[0062] To adapt to the characteristics of high-throughput radar data and fully utilize the parallel processing capabilities of the GPU, the entire matrix data corresponding to each first reconstructed signal can be input into the GPU processing framework as a whole. At the same time, combined with the distance dimension observation matrix constructed above, the data processing process is converted into a mathematical model. In this way, a compressed sensing model in matrix form can be obtained, as follows: Y = AX + N ; in, Y represents the reconstructed matrix echo signal, A Represents the observation matrix, such as the distance dimension observation matrix in this step, X Indicates the signal carrying target information (range information or Doppler information) that needs to be recovered by reconstructing the signal. N is a noise signal.

[0063] For the distance signal recovery process, the above-mentioned compressed sensing model can be recorded as the first compressed sensing model, in which Y is the first reconstructed signal, A is the distance dimension observation matrix, X The signal carrying the target distance information is recovered by the first reconstructed signal. For the Doppler signal recovery process, the above compressed sensing model can be recorded as the second compressed sensing model, where Y is the second reconstructed signal, A is the Doppler dimension observation matrix, X It is the signal carrying the Doppler information of the target that needs to be restored through the second reconstructed signal (the part about Doppler signal restoration will be explained later).

[0064] For example, taking the test data corresponding to the above 8192×2048 two-dimensional measurement matrix as an example, in the test data, during the signal recovery process of the distance dimension of each batch, , , , so the signal recovered in each batch should be: ,in, Indicates plural.

[0065] After constructing the first compressed sensing model corresponding to the first reconstructed signal of each batch in the distance-dimensional signal recovery process, the first compressed sensing model for each batch can be solved to recover the recovery signal of the corresponding batch. This solution can be performed using a gradient descent optimization algorithm, which generally involves an iterative process. Therefore, initialization parameters corresponding to the first compressed sensing model can be set. The initialization parameters can include an initial recovery signal, initial auxiliary variables (described later), and step size parameters. The initial recovery signal can be calculated based on the distance-dimensional observation matrix and the first reconstructed signal of the corresponding batch.

[0066] For the first compressed sensing model of each batch, after the initialization parameters of the first compressed sensing model are set, a gradient descent optimization algorithm and GPU parallel calculation can be used based on the initialization parameters to restore the required signal.

[0067] Taking a first compressed sensing model as an example, in the first compressed sensing model, the matrix X Each column corresponds to a sparse signal in the fast time dimension, and each column is independent of each other, so the matrix X Each column of is treated as a separate sparse signal vector for processing, and the process of processing matrix data can be regarded as a batch recovery using the gradient descent optimization algorithm, with each column updated independently and in parallel.

[0068] Update the matrix signal through the gradient descent optimization algorithm X , the update rule of the gradient descent method can be expressed as: ; in, k Indicates the current iteration round; X k Indicates the k The signal recovered from the round; T k Indicates the intermediate variables calculated in the current iteration round; Indicates the k The gradient of the signal recovered in each round can be expressed as follows: , among which x Represents a variable, which can be X k ; Represents the step size parameter, which must be greater than or equal to A T AThe maximum eigenvalue of the distance dimension measurement matrix has been normalized as mentioned above to standardize the eigenvalue range of the distance dimension measurement matrix, ensuring the calculation stability and improving the optimization efficiency. Therefore, after normalization, the maximum eigenvalue of the distance dimension measurement matrix is ​​set to 1, that is, It can be directly set to 1, which can simplify the algorithm design and speed up the convergence.

[0069] During the solution process, the recovered signal can be ultimately obtained by continuously iterating the gradient descent optimization calculation on the above-mentioned first compressed sensing model. This method is a standard gradient descent process. When faced with high-throughput radar data, its computational efficiency needs to be improved. Based on this, this embodiment further proposes a technical solution to accelerate the solution, that is, further accelerate the convergence process, by introducing an auxiliary variable. This is explained below.

[0070] Optionally, the initialization parameters may include an initial recovery signal and an initial auxiliary variable. In step A4, based on the initialization parameters of the first compressed sensing model of each batch, a gradient descent optimization algorithm and GPU parallelism are used to solve the first reconstructed signal in the first compressed sensing model of each batch, and determining the first target recovery signal corresponding to the first reconstructed signal of each batch may include: Step A41 : For each batch of first compressed sensing models, the corresponding first compressed sensing model is initialized using the initial restored signal and initial auxiliary variables; the initial auxiliary variables are determined according to the initial restored signal.

[0071] Step A42, performing an iterative operation, the above-mentioned iterative operation includes: solving the first reconstructed signal in the first compressed sensing model according to the first reconstructed signal, the distance dimension observation matrix and the first auxiliary variable of the current iteration round, obtaining the intermediate variables in the current iteration round, determining the first recovered signal of the current iteration round according to the intermediate variables in the current iteration round, and updating the first auxiliary variable according to the first recovered signal of the current iteration round to determine the second auxiliary variable.

[0072] Step A43, determine whether the second auxiliary variable of the current iteration round meets the first accuracy requirement. If not, use the second auxiliary variable as the new first auxiliary variable, return to perform the above iterative operation until the second auxiliary variable of the current iteration round meets the first accuracy requirement, and determine the second auxiliary variable that meets the first accuracy requirement as the first target recovery signal corresponding to the first reconstructed signal of the corresponding batch.

[0073] Among them, this embodiment introduces the Nesterov acceleration strategy, which mainly updates the current solution by weighted average of the previous iterative solutions. Based on this, this embodiment introduces an auxiliary variableZ , the auxiliary variable represents the weighted average of the previous solution and the current solution, and is used to adjust the update direction of each iteration. The initialization parameter of the auxiliary variable is the initial auxiliary variable, denoted as Z 1 , the initial recovery signal (the initialization parameter of the signal to be recovered) is recorded as X 0 , X 0 = A T Y In this embodiment, the auxiliary variables are initialized, which can be specifically set Z 1 = X 0 Based on this, the recovery signal in the above gradient descent optimization algorithm is represented by auxiliary variables, and its expression can be updated to the following expression: ; in, Z k Indicates the k Auxiliary variables of round, according to k The signal recovered in this round and the signal recovered in the previous round are calculated. Indicates the k Gradients of auxiliary variables for this episode.

[0074] After the auxiliary variables and the signal to be recovered are initialized, the first compressed sensing model of the first reconstructed signal is initialized. Then, an iterative operation in the gradient descent optimization algorithm can be performed to solve the signal to be recovered in the first compressed sensing model. The iterative operation includes the following steps: Step 1: Calculate the initial recovery signal of the first iteration according to the first reconstructed signal and the distance dimension measurement matrix, that is, X 0 = A T Y , and then determine the auxiliary variable of the first iteration round according to the initial recovery signal of the first round, which is recorded as the first auxiliary variable, that is, Z 1 = X 0 The gradient descent optimization algorithm is used to solve the first compressed sensing model to obtain the intermediate variables in the current iteration round, that is, the expression of the gradient descent optimization algorithm represented by the auxiliary variables is used to calculate the intermediate variables of the first iteration round. T 1 .

[0075] Starting from the second iteration, the gradient in the current iteration is calculated using the above gradient formula according to the first reconstructed signal, the distance dimension measurement matrix and the first auxiliary variable in the current iteration, and the intermediate variable in the current iteration is obtained by subtracting the gradient in the current iteration from the first auxiliary variable in the current iteration. T k .

[0076] Step 2: Update the recovery signal X k , X k It can be recorded as the first recovery signal of the current iteration round, that is, according to the intermediate variables in the current iteration round T k Determining / updating the first recovered signal for the current iteration round can be done using a soft threshold comparison operation. This process may include: determining a step size parameter corresponding to the gradient descent optimization algorithm and a regularization parameter corresponding to the range Doppler signal recovery process; determining a signal comparison threshold based on the step size parameter and the regularization parameter; the regularization parameter is related to the sparsity of the recovered signal; and determining the first recovered signal for the current iteration round based on the intermediate variables in the current iteration round and the signal comparison threshold. This step primarily imposes sparsity constraints on the recovered signal to enable updating of the recovered signal.

[0077] The update resumes the signal X k The specific process can be expressed by the following formula: ; in, Represents the regularization parameter. The regularization parameters for the range signal recovery process and the Doppler signal recovery process can be different. The regularization parameter for the range signal recovery process can be smaller than the regularization parameter for the Doppler signal recovery process. For example, the regularization parameter for the range signal recovery process can be set to 0.0025, and the regularization parameter for the Doppler signal recovery process can be set to 1.6. is the step size parameter, which can be set to 1. It can represent the signal comparison threshold.

[0078] above It can be expressed as: , among which sign ( T k ) is a symbolic function, and its specific meaning is: when T k When it is greater than 0, the sign function value is equal to 1. T k When it is equal to 0, the sign function value is equal to 0. Tk When it is less than 0, the sign function value is -1. The max function takes the larger of the two values. If it is greater than 0, the max function takes Otherwise, the max function takes 0.

[0079] The above X k The updating process can be understood as follows: if the intermediate variable of the current iteration round is greater than the signal comparison threshold, the intermediate variable of the current iteration round is shrunk; conversely, if the intermediate variable of the current iteration round is less than the signal comparison threshold, the intermediate variable of the current iteration round is compressed to 0. Here, a sparse constraint is imposed on the estimated value of the updated restored signal. Specifically, a soft threshold comparison operation is performed to map the restored signal into the sparse solution space to ensure that the sparse constraint is met. This reduces the signal components smaller than the signal comparison threshold, retains the signal components larger than the signal comparison threshold, and removes the noise component, thereby enhancing the sparsity of the signal.

[0080] Step 3: Update the first auxiliary variable of the current iteration round according to the first recovery signal of the current iteration round, and determine the second auxiliary variable. Optionally, it can include: obtaining the first recovery signal of the previous iteration round of the current iteration round; calculating the momentum adjustment parameter of the next iteration round according to the momentum adjustment parameter of the current iteration round; updating the first auxiliary variable and determining the second auxiliary variable according to the first recovery signal of the current iteration round, the first recovery signal of the previous iteration round, the momentum adjustment parameter of the current iteration round, and the momentum adjustment parameter of the next iteration round.

[0081] Among them, the momentum adjustment parameter can be recorded as t , the initialization parameter of the momentum adjustment parameter is t 1=1, the momentum adjustment parameter of the current iteration round can be recorded as t k , the momentum adjustment parameter for the next iteration can be recorded as t k+1 , which is calculated as follows: ; The above formula can be used to update the momentum adjustment parameter and obtain the momentum adjustment parameter for the next iteration round.

[0082] Afterwards, the auxiliary variable can be updated. The updated auxiliary variable is recorded as the second auxiliary variable, which can be specifically expressed by the following formula: ; in, Z k Indicates the k Round is the first auxiliary variable of the current iteration round,Z k+1 The second auxiliary variable representing the update of the current iteration round is also the k+ Round 1 is the first auxiliary variable of the next iteration round, X k-1 Indicates the k- Round 1 is the first recovery signal of the previous iteration round.

[0083] After obtaining the second auxiliary variable updated in the current iteration, it is possible to determine whether the data accuracy of the second auxiliary variable updated in the current iteration meets the first accuracy requirement and obtain a determination result. The first accuracy requirement is the data accuracy requirement corresponding to the distance dimension signal recovery processing, which can be set according to actual conditions, for example, the data accuracy can reach 10 -6 If the above judgment result is that the data accuracy of the second auxiliary variable updated in the current iteration round meets the first accuracy requirement, the second auxiliary variable updated in the current round is used as the signal restored corresponding to the first reconstructed signal of the batch, and is recorded as the first target restored signal. If the above judgment result is that the data accuracy of the second auxiliary variable updated in the current iteration round does not meet the first accuracy requirement, then k = k +1, and use the second auxiliary variable updated in the current iteration round as the first auxiliary variable in the next iteration round, and continue to iterate and execute steps 1 to 3 in the above iterative operation until the data accuracy of the second auxiliary variable updated in the current iteration round meets the first accuracy requirement. The iteration is terminated, and the second auxiliary variable updated in the current round at the end of the iteration is used as the first target recovery signal corresponding to the first reconstructed signal of the batch.

[0084] In the above distance dimension signal recovery process, since each column in the matrix data of each batch of the first reconstructed signal represents an independent observation signal in fast time, the columns are independent of each other, and the first target recovery signal obtained by processing each batch of the first reconstructed signal is also in the form of a matrix. X , in the test data, .

[0085] The above iterative operation can be used to solve the first target recovery signal corresponding to each batch of the first reconstructed signals. In this way, the first target recovery signal corresponding to each batch of the first reconstructed signals can be obtained. Then, the range Doppler processing result corresponding to the target can be determined based on the first target recovery signal of each batch.

[0086] In the embodiment, the distance dimension observation matrix is constructed by the sampling number of the radar echo signal, and the compressed sensing model is constructed by the distance dimension observation matrix and the reconstructed signal, so that the reconstructed signal in the compressed sensing model is solved by using the GPU and the gradient descent optimization algorithm, and the target recovery signal is obtained, which can effectively improve the efficiency of signal recovery. In addition, auxiliary variables representing the average of multiple recovery signals and intermediate variables are introduced in the signal recovery process to perform gradient descent optimization calculation, which can further improve the efficiency of signal recovery. Further, when updating the recovery signal in each iteration round, the signal comparison threshold can be determined by the step parameter and the regularization parameter, and the recovery signal in each iteration round is updated by the comparison result of the intermediate variable and the signal comparison threshold, which can eliminate the noise component in the recovery signal, enhance the sparsity of the signal, and further improve the accuracy of subsequent signal recovery. At the same time, when updating the auxiliary variable in each iteration round, the recovery signal in the previous iteration round can be updated by combining the momentum adjustment parameter in the current and next iteration round, which can improve the accuracy of the auxiliary variable update and improve the efficiency of signal recovery.

[0087] The above embodiment describes the processing process of distance dimension signal recovery by distance dimension observation matrix, and then the Doppler dimension signal recovery processing can be performed. The following embodiment describes the processing process of Doppler dimension signal recovery.

[0088] Figure 2 is a flowchart of the distance Doppler processing method of the target provided by the application, as Figure 2 As shown in the above step 108, the distance Doppler signal recovery processing of the first reconstructed signal of each batch is performed by using the GPU in parallel, the first target recovery signal corresponding to the first reconstructed signal of each batch is determined, and the distance Doppler processing result corresponding to the target is determined according to the first target recovery signal of each batch, which can include the following steps: Step 202, the distance signal recovery processing of the first reconstructed signal of each batch is performed by using the GPU in parallel, and the first target recovery signal corresponding to the first reconstructed signal of each batch is determined.

[0089] In the above embodiment, the distance dimension signal recovery processing of the first reconstructed signal of each batch is performed by using the GPU in parallel, and the distance information recovered by the first reconstructed signal of each batch is obtained, which is denoted as the first target recovery signal. Then the Doppler dimension speed or frequency information can be recovered in the first target signal of each batch.

[0090] Step 204, the first target recovery signal of each batch is spliced to determine the distance recovery signal; the row and column numbers of the distance recovery signal are the same as those of the two-dimensional observation matrix.

[0091] In this step, after obtaining each batch of first target recovered signals, the scale of each first target recovered signal is the same as the scale of its corresponding first reconstructed signal. Since each batch of first reconstructed signals is divided into columns of the two-dimensional measurement matrix, the first target recovered signals of each batch can be spliced ​​together in the order of the columns in which their corresponding first reconstructed signals reside to obtain the final spliced ​​distance recovered signal. The number of rows in this distance recovered signal is equal to the number of rows in the two-dimensional measurement matrix, and the number of columns is also equal to the number of columns in the two-dimensional measurement matrix, that is, the distance recovered signal has the same scale as the two-dimensional measurement matrix.

[0092] In addition, after obtaining the first target recovery signals of all batches, these first target recovery signals can be spliced ​​by column to obtain the distance recovery signal on the GPU side (the scale is, for example, 8192×2048). Then, the gather function in MATLAB is used to convert the distance recovery signal on the GPU side from the GPU format to the CPU format to obtain the distance recovery signal in the CPU format, which is convenient for subsequent Doppler signal recovery processing.

[0093] Step 206: Perform Doppler signal recovery processing on the range recovery signal to determine the range Doppler processing result corresponding to the target.

[0094] In this step, after obtaining the distance recovery signal, Doppler signal recovery processing can be continued based on the distance recovery signal to recover the Doppler information of the target, and finally obtain the target range and Doppler processing results.

[0095] Optionally, performing Doppler signal recovery processing on the range recovery signal in step 206 to determine a range Doppler processing result corresponding to the target may include: Step B1, transpose the distance recovery signal to obtain a transposed distance recovery signal, and divide the transposed distance recovery signal into multiple batches of second reconstructed signals according to the computing power parameters of the GPU; each batch of second reconstructed signals includes multiple columns of data in the transposed distance recovery signal.

[0096] During the target Doppler signal recovery process, the target's Doppler information resides in the slow time dimension. Unlike the previous range-dimensional signal recovery process, the data units processed here are row data, not column data. Therefore, the input matrix data (i.e., the range-recovered signal) must be transposed to obtain the transposed range-recovered signal. For example, the resulting transposed range-recovered signal can be a 2048×8192 matrix.

[0097] After obtaining the transposed distance recovery signal, similar to the above-mentioned distance recovery signal processing process, the transposed distance recovery signal is high-throughput data and can also be divided into batches. Specifically, the transposed distance recovery signal can be divided into multiple batches of second reconstructed signals according to the computing power parameters of the GPU.

[0098] In addition, before the above division, the values ​​of each element in the transposed distance recovery signal can be converted to single-precision data. The transposed distance recovery signal after data precision conversion can then be converted from CPU format to GPU format before batch division.

[0099] It can be understood that when the transposed distance recovery signal is divided into batches here, the multiple columns of data of the transposed distance recovery signal are divided according to the determined number of columns that the GPU can process in parallel each time, and multiple batches of second reconstructed signals are obtained. The number of rows of each second reconstructed signal is the same as the number of rows of the transposed distance recovery signal, and the number of columns is the number of columns that the GPU can process in parallel each time as determined above.

[0100] For example, the scale of the above-mentioned distance recovery signal is 8192×2048, and the scale of the transposed distance recovery signal after transposition is 2048×8192. Since the number of rows of the transposed distance recovery signal is small, the number of columns that the GPU can process in parallel each time will be more than the number of columns determined in the distance dimension signal recovery processing. For example, here the number of columns that the GPU can process in parallel each time is 512. After batch division, the scale of the second reconstructed signal obtained in each batch can be 2048×512, and a total of 8192÷512=16 batches are obtained. The data scale processed each time is, for example, 2048×512. In this case, the utilization rate of the GPU can be maintained between 85% and 90%, which is relatively reasonable.

[0101] Step B2: Use GPU to perform Doppler signal recovery processing on each batch of second reconstructed signals in parallel, determine the second target recovery signal corresponding to each batch of second reconstructed signals, and determine the range Doppler processing result corresponding to the target based on the second target recovery signal of each batch.

[0102] In this step, after obtaining multiple batches of second reconstructed signals, the GPU can perform Doppler signal recovery processing on multiple columns of data in a second reconstructed signal in parallel each time, and obtain the recovery signal corresponding to each second reconstructed signal, which is recorded as the second target recovery signal. Then, the range Doppler processing result of the target can be obtained by splicing the second target recovery signals of each batch together by column.

[0103] Optionally, in step B2, using a GPU in parallel to perform Doppler signal recovery processing on each batch of second reconstructed signals to determine the second target recovered signal corresponding to each batch of second reconstructed signals may include: Step B21: construct a Doppler measurement matrix based on the number of pulses of the radar echo signal using the dftmtx function in MATLAB; the number of rows and columns of the Doppler measurement matrix is ​​the same as the number of pulses.

[0104] Step B22: constructing a second compressed sensing model corresponding to the second reconstructed signals of each batch according to the distance dimension measurement matrix and the second reconstructed signals of each batch.

[0105] Step B23: Determine the initialization parameters corresponding to the second compressed sensing model of each batch.

[0106] Step B24: Based on the initialization parameters of the second compressed sensing model of each batch, a gradient descent optimization algorithm and GPU parallelism are used to solve the second reconstructed signal in the second compressed sensing model of each batch to determine the second target recovered signal corresponding to the second reconstructed signal of each batch.

[0107] Similar to the range-dimensional signal recovery process described above, the Doppler-dimensional signal recovery process also requires constructing an observation matrix. This observation matrix can be referred to as the Doppler-dimensional observation matrix. This Doppler-dimensional observation matrix is ​​constructed based on the discrete Fourier transform (DFT). The size and structure of the matrix must match the number of pulses. Specifically, the DFT matrix can be constructed using the dftmtx function in MATLAB. The dftmtx function is a MATLAB function that generates a discrete Fourier transform (DFT) matrix. It is understood that the range-dimensional observation matrix here is a square matrix, with the number of rows and columns equal to the number of pulses sampled in the radar echo signal.

[0108] After constructing the Doppler measurement matrix, its inverse matrix can be calculated to facilitate signal recovery or transformation in subsequent processing. Specifically, the inverse matrix calculation process includes performing a conjugate transpose operation on the Doppler measurement matrix, and then dividing the transposed matrix by the number of pulses to obtain its inverse matrix.

[0109] Furthermore, to simplify the subsequent difficulty of solving the recovered signal using a gradient descent optimization algorithm, both the Doppler measurement matrix and the transposed range-recovered signal can be normalized. Specifically, the normalization process can be performed by first calculating the product of the Doppler measurement matrix (or its inverse matrix) and its transposed matrix, then solving for the eigenvalues ​​of the matrix corresponding to the multiplication result. The largest eigenvalue among the solved eigenvalues ​​is then selected, and the Doppler measurement matrix is ​​normalized by dividing each element in the Doppler measurement matrix by the square root of the largest eigenvalue. Alternatively, each element in the transposed range-recovered signal can be normalized by dividing it by the square root of the largest eigenvalue. Alternatively, each element in the radar observation data can be normalized by dividing it by the square root of the largest eigenvalue. After the above normalization processing, the largest eigenvalue is set to 1, so that the step size parameter in the gradient descent optimization algorithm in the Doppler signal recovery process can be directly set to 1, the eigenvalue range of the Doppler measurement matrix is ​​set between [0, 1], and the eigenvalue range of the transposed distance recovery signal is set between [0, 1]. In this way, through the normalization processing, the range of the eigenvalues ​​of the Doppler measurement matrix and the transposed distance recovery signal matrices can be standardized, which can improve the stability of subsequent calculations and improve the optimization efficiency. At the same time, it can simplify the step size parameter setting of the subsequent gradient descent optimization algorithm in the calculation process.

[0110] The second target recovery signal corresponding to the second reconstructed signal for each batch can then be solved according to the processes in steps A2 through A4 and A41 through A43. The second target recovery signal may include the target's Doppler information, i.e., velocity or frequency. Although the data organization methods for range- and Doppler-dimensional signal recovery differ, the iterative solution process is consistent. Both utilize a gradient descent optimization algorithm, soft thresholding, and auxiliary variable updates to achieve a step-by-step optimization. Ultimately, the matrix data after range-Doppler processing is output, i.e., the target recovery signal corresponding to each batch of reconstructed signals is obtained, recorded as the second target recovery signal.

[0111] After obtaining all batches of second target recovery signals, these second target recovery signals can be spliced ​​by column to obtain a matrix signal (for example, a scale of 2048×8192). The matrix signal can then be transposed to obtain the final GPU-side range-Doppler processing result (for example, a scale of 8192×2048). The gather function in MATLAB is then used to convert the GPU-side range-Doppler processing result from GPU format to CPU format to obtain the CPU-format range-Doppler processing result for subsequent applications.

[0112] Finally, the range-doppler processing result of the high-throughput radar observation data after the range-doppler signal recovery processing can be output, for example, the Range-Doppler Map (range-doppler map) can be used to intuitively represent the distance information and Doppler information of the target, as shown in Figure 3 The range-doppler processing result of the target is shown in the schematic diagram, wherein the horizontal axis is Range distance, the unit is m, and the vertical axis is speed Speed, the unit is m / s. As can be seen from the local enlarged view, there is a target point (i.e. white point in the figure) at a distance of 10000 m and a speed of -100 m / s, which is the obtained distance and speed information of the target. This is the same as the distance and speed information of the target set when the radar echo signal is first obtained, which indicates the feasibility and processing accuracy of the embodiment scheme of the present application.

[0113] In addition, in the high-throughput radar echo data processing process, after processing by the GPU framework, the data processed by the CPU framework in series is compared, and the results show that the GPU processing significantly accelerates the calculation efficiency. At the same time, through experiments, under the condition that the processing environment CPU is AMD Ryzen 9 7940H and the GPU is NVIDIA GeForce RTX 4050 notebook, the time acceleration ratio of the double-precision data format using the GPU framework to the double-precision data format using the CPU framework reaches 12.1 times, and the time acceleration ratio of the single-precision data format using the GPU framework to the single-precision data format using the CPU framework reaches 84.6 times, and the calculation error is controlled within 1x10 -6 orders of magnitude, which can meet the engineering requirements and significantly improve the calculation efficiency. It can be seen that the GPU framework and single-precision data format processing have significant advantages in calculation efficiency, and on the basis of ensuring accuracy, can significantly improve the calculation efficiency, and become the preferred scheme for efficient calculation.

[0114] In summary, the test results show that the high-throughput radar observation data target range-doppler processing method proposed in the embodiment of the present application can significantly improve the data processing efficiency and ensure the recovery accuracy of the signal. By combining the gradient descent optimization algorithm and the sparse signal reconstruction technology, and using the GPU processing framework and the adaptive precision selection mechanism, the calculation bottleneck problem of large-scale data processing in the high-throughput radar system can be effectively solved, the real-time and accuracy requirements in the high-throughput radar system can be met, and the method has wide engineering application value and promotion potential.

[0115] In this embodiment, a GPU is first used to perform range recovery on each batch of first reconstructed signals in parallel to obtain each first target recovery signal. These first target recovery signals are then spliced ​​together to obtain a range recovery signal, which is then subjected to Doppler signal recovery processing to obtain the target's range Doppler processing. This improves the efficiency of range Doppler processing and ensures the accuracy of the processing results. Furthermore, when performing Doppler signal recovery processing, the input signal can be divided into multiple batches, and the GPU can be used to perform parallel calculations on multiple columns of data in each batch, further improving computational efficiency. Furthermore, the Doppler dimension measurement matrix is ​​reconstructed based on the number of pulses, and a compressed sensing model is constructed for solution to obtain the second target recovery signal for each batch, thereby improving the accuracy of the ultimately recovered signal.

[0116] A detailed embodiment is given below for illustration. Based on the above embodiment, the specific idea of ​​the technical solution of the present invention is: reconstructing the echo vector data received by the radar into a matrix form with fast time and slow time as dimensions to facilitate batch processing, constructing and normalizing the distance dimension observation matrix, using the GPU processing framework, automatically selecting the appropriate data accuracy according to the processing environment, batch processing the matrix data, calculating the gradient of the signal and performing gradient descent update, performing soft threshold processing, updating the auxiliary variables and calculating the new signal estimate, after each iteration, judging whether the signal meets the accuracy requirements, if not, continuing to process until the requirements are met, and converting the data back to CPU format after completing the distance dimension data processing; Doppler dimension data processing is similar to the distance dimension data processing, and finally outputting the CPU format data after distance Doppler processing. See Figure 4 The detailed flow chart of the range Doppler processing method of the target shown in FIG. 1 is a flowchart of the range Doppler processing method of the target, and the method may specifically include the following steps: (1) Obtain radar echo data and sample; (1a) Obtain radar echoes and sample the data in the time domain to obtain radar observation data in the form of a one-dimensional vector; (2) Reconstruct the radar observation data in the form of one-dimensional vectors into a two-dimensional observation matrix with fast time and slow time as dimensions; (2a) The received radar observation data in vector form is reconstructed into a two-dimensional observation matrix with fast time and slow time as the dimensions, corresponding to the target's range information and Doppler information respectively, to facilitate subsequent batch processing of data with the GPU framework and improve data processing efficiency.

[0117] (3) Construct and normalize the distance dimension observation matrix; (3a) Construct a range-dimensional measurement matrix based on the radar model, calculate the product of the range-dimensional measurement matrix by itself, obtain the eigenvalue matrix, and extract the maximum eigenvalue; (3b) The distance measurement matrix and echo data are divided by the square root of the largest eigenvalue to complete the normalization operation, standardizing the eigenvalue range of the matrix, ensuring computational stability and improving optimization efficiency. After normalization, the largest eigenvalue is set to 1, so that the step size parameter in the gradient descent algorithm can be directly set to 1, thereby simplifying the algorithm design and accelerating convergence. (4) Using the GPU processing framework, data processing batches are divided according to computing power parameters and data processing accuracy is adaptively selected: (4a) Using the GPU processing framework to convert data from CPU format to GPU format; (4b) Adaptively select data precision based on the computing efficiency of the corresponding graphics card for different precisions. Taking the NVIDIA GeForce RTX 4050 laptop GPU as an example, its single-precision computing capability is about 8.986 TFLOPS, while its double-precision computing capability is only about 140.4 GFLOPS. The computing efficiency of single precision is much higher than that of double precision, so the data precision is converted to single precision; (4c) Based on the GPU's parallel computing capability and video memory size, the distance dimension observation data in matrix form is further divided into multiple batches suitable for GPU processing, ensuring that the amount of data processed each time can fully utilize the GPU resources without exceeding the video memory limit. The divided data batches are then processed one by one. (5) Calculate the gradient of the current signal estimate and perform gradient descent update; (5a) Calculate the gradient of the objective function based on the current auxiliary variable and obtain the optimization direction of the signal estimation; (5b) Using the set step size parameter, the current auxiliary variable is adjusted based on the gradient value to generate a new signal estimate; (5c) Mapping the new signal estimate into the sparse solution space to ensure compliance with the sparsity constraint; (6) Perform soft threshold processing on the updated signal; (6a) Applying sparse constraints to the updated signal estimate, and reducing the components of the signal that are smaller than the threshold through a soft threshold operation; (6b) retain the signal components that are larger than the set threshold and remove the noise components to enhance the sparsity of the signal; (7) Update auxiliary variables and calculate new signal estimates; (7a) Calculate the momentum adjustment parameter based on the previous signal estimate and auxiliary variables; (7b) Generate new auxiliary variables using momentum adjustment parameters; (8) Determine whether the signal meets the first accuracy requirement. If not, increase the number of iterations / iteration rounds by one and continue iterating from step (5) to step (7); (9) After completing the distance dimension signal recovery processing, convert the data format back to the CPU format; (10) Construct and normalize the Doppler measurement matrix; (10a) Construct the Doppler dimension measurement matrix based on the radar model, calculate the self-multiplication result of the range dimension measurement matrix, obtain the eigenvalue matrix, and extract the maximum eigenvalue; (10b) The Doppler measurement matrix and the transposed range recovery signal are divided by the square root of the largest eigenvalue to complete the normalization operation, standardizing the eigenvalue range of the matrix, ensuring computational stability and improving optimization efficiency. After normalization, the largest eigenvalue is set to 1, so that the step size parameter in the gradient descent algorithm can be directly set to 1, thereby simplifying the algorithm design and accelerating convergence. (11) Execute steps (4) to (7); (11a) In the process of Doppler dimension signal recovery, since the target's Doppler information is in the slow time dimension, unlike the previous range dimension processing, the data unit processed here is row data rather than column data. The column data of the matrix is ​​the signal that is processed independently and in parallel, so the input matrix data needs to be transposed; (12) Determine whether the signal meets the second accuracy requirement. If not, continue iterating steps (5) to (7). The second accuracy requirement can be the same as the first accuracy requirement, indicating the data accuracy requirement during Doppler signal recovery processing, and its size can be set according to actual conditions. (13) After completing the Doppler signal recovery process, convert the final output data format back to the CPU format; (13a) After all Doppler-dimensional batches of data are processed, the obtained matrix signal data is transposed again and then converted from GPU format to CPU format; (14) Output the final radar detection data after range Doppler processing.

[0118] As can be seen from the above description, in the embodiment of the present invention, by reconstructing radar observation data into a two-dimensional observation matrix with fast time and slow time as dimensions, utilizing the parallel computing capability of the GPU, combined with the compressed gradient descent algorithm, sparse signals are restored through batch matrixization, large-scale data is processed efficiently and in parallel, and the efficiency of high-throughput data processing is improved. It is suitable for high-throughput radar data range Doppler processing.

[0119] The target range Doppler processing device provided by the present invention is described below. The target range Doppler processing device described below and the target range Doppler processing method described above can be referred to each other.

[0120] Figure 5This is a schematic diagram of the structure of the range Doppler processing device of the target provided by the present invention, see Figure 5 As shown, the device may include: The echo signal sampling module 510 is used to obtain the radar echo signal of the target and perform pulse sampling on the radar echo signal to determine the radar observation data in the form of a one-dimensional vector; an echo signal reconstruction module 520 for reconstructing radar observation data based on the number of pulses and the number of sampling times when pulse sampling the radar echo signal, and determining a two-dimensional measurement matrix corresponding to the radar observation data; wherein the number of rows of the two-dimensional measurement matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional measurement matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; A batch division module 530 is configured to divide the two-dimensional measurement matrix into a plurality of batches of first reconstructed signals according to a computing power parameter of the GPU; each batch of first reconstructed signals includes a plurality of columns of data in the two-dimensional measurement matrix; The range Doppler processing module 540 is used to use the GPU in parallel to perform range Doppler signal recovery processing on the first reconstructed signals of each batch, determine the first target recovery signal corresponding to the first reconstructed signal of each batch, and determine the range Doppler processing result corresponding to the target based on the first target recovery signal of each batch.

[0121] It should be noted here that the above-mentioned device provided by the embodiment of the present invention can implement all the method steps implemented by the above-mentioned method embodiment and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.

[0122] Figure 6 An example of a physical structure diagram of an electronic device is shown below. Figure 6As shown, the electronic device may include: a processor 610 , a communications interface 620 , a memory 630 and a communication bus 640 , wherein the processor 610 , the communications interface 620 and the memory 630 communicate with each other via the communication bus 640 . The processor 610 can call the logic instructions in the memory 630 to execute the range Doppler processing method of the target, which includes: obtaining the radar echo signal of the target and pulse sampling the radar echo signal to determine the radar observation data in the form of a one-dimensional vector; reconstructing the radar observation data according to the number of pulses and the number of sampling times when the radar echo signal is pulse sampled, and determining the two-dimensional observation matrix corresponding to the radar observation data; the number of rows of the above-mentioned two-dimensional observation matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional observation matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; dividing the two-dimensional observation matrix into multiple batches of first reconstructed signals according to the computing power parameters of the GPU; the first reconstructed signal of each batch includes multiple columns of data in the two-dimensional observation matrix; using the GPU in parallel to perform range Doppler signal recovery processing on the first reconstructed signal of each batch, determining the first target recovery signal corresponding to the first reconstructed signal of each batch, and determining the range Doppler processing result corresponding to the target based on the first target recovery signal of each batch.

[0123] Furthermore, the logic instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0124] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the range Doppler processing method of the target provided by the above-mentioned methods, the method including: obtaining a radar echo signal of the target and performing pulse sampling on the radar echo signal to determine radar observation data in the form of a one-dimensional vector; reconstructing the radar observation data based on the number of pulses and the number of sampling times when the radar echo signal is pulse sampled, and determining a two-dimensional observation matrix corresponding to the radar observation data; the number of rows of the above-mentioned two-dimensional observation matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional observation matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; dividing the two-dimensional observation matrix into multiple batches of first reconstructed signals according to the computing power parameter of the GPU; the first reconstructed signal of each batch includes multiple columns of data in the two-dimensional observation matrix; using the GPU in parallel to perform range Doppler signal recovery processing on the first reconstructed signal of each batch, determining a first target recovery signal corresponding to the first reconstructed signal of each batch, and determining the range Doppler processing result corresponding to the target based on the first target recovery signal of each batch.

[0125] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the range Doppler processing method for the target provided by the above-mentioned methods, the method comprising: acquiring a radar echo signal of the target and performing pulse sampling on the radar echo signal to determine radar observation data in the form of a one-dimensional vector; reconstructing the radar observation data based on the number of pulses and the number of sampling times when the radar echo signal is pulse sampled, and determining a two-dimensional observation matrix corresponding to the radar observation data; the number of rows of the above-mentioned two-dimensional observation matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional observation matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; dividing the two-dimensional observation matrix into multiple batches of first reconstructed signals according to the computing power parameters of the GPU; the first reconstructed signals of each batch include multiple columns of data in the two-dimensional observation matrix; using the GPU to perform range Doppler signal recovery processing on the first reconstructed signals of each batch in parallel, determining the first target recovery signal corresponding to the first reconstructed signal of each batch, and determining the range Doppler processing result corresponding to the target based on the first target recovery signals of each batch.

[0126] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0127] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A range Doppler processing method for a target, characterized in that: include: Acquiring a radar echo signal of the target and performing pulse sampling on the radar echo signal to determine radar observation data in a one-dimensional vector form; Reconstructing the radar observation data based on the number of pulses and the number of sampling times when pulse sampling the radar echo signal, and determining a two-dimensional measurement matrix corresponding to the radar observation data; the number of rows of the two-dimensional measurement matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional measurement matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; Dividing the two-dimensional measurement matrix into a plurality of batches of first reconstructed signals according to a computing power parameter of a graphics processing unit (GPU); each batch of the first reconstructed signals includes a plurality of columns of data in the two-dimensional measurement matrix; The GPU is used to perform range Doppler signal recovery processing on the first reconstructed signals of each batch in parallel, determine a first target recovery signal corresponding to the first reconstructed signals of each batch, and determine a range Doppler processing result corresponding to the target based on the first target recovery signals of each batch.

2. The range Doppler processing method of a target according to claim 1, characterized in that: The using the GPU to perform range Doppler signal recovery processing on each batch of the first reconstructed signals in parallel to determine a first target recovered signal corresponding to each batch of the first reconstructed signals includes: Constructing a distance-dimensional measurement matrix according to the sampling times of the radar echo signal; the number of rows of the distance-dimensional measurement matrix is ​​related to the sampling times, and the columns of the distance-dimensional measurement matrix represent the single pulse echo signal at the time corresponding to each sampling times; Constructing a first compressed sensing model corresponding to the first reconstructed signal of each batch according to the distance dimension measurement matrix and the first reconstructed signal of each batch; Determining initialization parameters corresponding to the first compressed sensing model for each batch; According to the initialization parameters of the first compressed sensing model of each batch, the first reconstructed signal in the first compressed sensing model of each batch is solved in parallel using a gradient descent optimization algorithm and the GPU to determine the first target recovered signal corresponding to the first reconstructed signal of each batch.

3. The range Doppler processing method of a target according to claim 2, characterized in that: The initialization parameters include an initial recovery signal and an initial auxiliary variable. The first reconstructed signal in each batch of the first compressed sensing model is solved in parallel using a gradient descent optimization algorithm and the GPU according to the initialization parameters of the first compressed sensing model of each batch, and the first target recovery signal corresponding to the first reconstructed signal of each batch is determined, including: For each batch of the first compressed sensing models, initializing the corresponding first compressed sensing models using the initial restored signal and the initial auxiliary variables; the initial auxiliary variables are determined according to the initial restored signal; performing an iterative operation, the iterative operation comprising: solving the first reconstructed signal in the first compressed sensing model using a gradient descent optimization algorithm according to the first reconstructed signal, the distance-dimensional measurement matrix, and a first auxiliary variable of a current iteration round to obtain an intermediate variable in the current iteration round; determining a first restored signal in the current iteration round according to the intermediate variable in the current iteration round; and updating the first auxiliary variable according to the first restored signal in the current iteration round to determine a second auxiliary variable; Determine whether the second auxiliary variable of the current iteration round meets the first accuracy requirement. If not, use the second auxiliary variable as the new first auxiliary variable, return to execute the iterative operation until the second auxiliary variable of the current iteration round meets the first accuracy requirement, and determine the second auxiliary variable that meets the first accuracy requirement as the first target recovery signal corresponding to the first reconstructed signal of the corresponding batch.

4. The range Doppler processing method of a target according to claim 3, characterized in that: The determining the first recovery signal of the current iteration round according to the intermediate variable in the current iteration round includes: Determining a step size parameter corresponding to the gradient descent optimization algorithm and a regularization parameter corresponding to the range Doppler signal recovery process; Determining a signal comparison threshold according to the step size parameter and the regularization parameter; the regularization parameter is related to the sparsity of the restored signal; A first recovery signal of the current iteration round is determined according to the intermediate variable and the signal comparison threshold in the current iteration round.

5. The range Doppler processing method of a target according to claim 3, characterized in that: The updating of the first auxiliary variable according to the first recovery signal of the current iteration round to determine the second auxiliary variable includes: Obtaining the first recovery signal of the previous iteration round of the current iteration round; Calculate the momentum adjustment parameter for the next iteration round based on the momentum adjustment parameter for the current iteration round; The first auxiliary variable is updated and the second auxiliary variable is determined according to the first recovery signal of the current iteration round, the first recovery signal of the previous iteration round, the momentum adjustment parameter of the current iteration round and the momentum adjustment parameter of the next iteration round.

6. The range Doppler processing method of a target according to any one of claims 1 to 5, characterized in that: The step of using the GPU to perform range Doppler signal recovery processing on each batch of the first reconstructed signals in parallel, determining a first target recovery signal corresponding to each batch of the first reconstructed signals, and determining a range Doppler processing result corresponding to the target based on the first target recovery signal of each batch, includes: Using the GPU to perform range signal recovery processing on each batch of the first reconstructed signals in parallel, and determining a first target recovery signal corresponding to the first reconstructed signals of each batch; Splicing the first target recovery signals of each batch to determine a distance recovery signal; the number of rows and columns of the distance recovery signal is the same as the number of rows and columns of the two-dimensional measurement matrix; Performing Doppler signal recovery processing on the range recovery signal to determine a range Doppler processing result corresponding to the target.

7. The range Doppler processing method of a target according to claim 6, characterized in that: The performing Doppler signal recovery processing on the range recovery signal to determine the range Doppler processing result corresponding to the target includes: transposing the distance recovery signal to obtain a transposed distance recovery signal, and dividing the transposed distance recovery signal into a plurality of batches of second reconstructed signals according to a computing power parameter of the GPU; each batch of the second reconstructed signals includes a plurality of columns of data in the transposed distance recovery signal; The GPU is used to perform Doppler signal recovery processing on each batch of the second reconstructed signals in parallel, determine the second target recovery signal corresponding to each batch of the second reconstructed signals, and determine the range Doppler processing result corresponding to the target based on the second target recovery signal of each batch.

8. The range Doppler processing method of a target according to claim 7, characterized in that: The using the GPU to perform Doppler signal recovery processing on each batch of the second reconstructed signals in parallel to determine a second target recovered signal corresponding to each batch of the second reconstructed signals includes: According to the number of pulses of the radar echo signal, a Doppler dimension measurement matrix is ​​constructed using the dftmtx function in MATLAB; the number of rows and columns of the Doppler dimension measurement matrix is ​​the same as the number of pulses; Constructing a second compressed sensing model corresponding to the second reconstructed signal of each batch according to the distance dimension measurement matrix and the second reconstructed signal of each batch; Determining initialization parameters corresponding to the second compressed sensing model for each batch; According to the initialization parameters of the second compressed sensing model of each batch, the second reconstructed signal in the second compressed sensing model of each batch is solved in parallel using a gradient descent optimization algorithm and the GPU to determine the second target recovered signal corresponding to the second reconstructed signal of each batch.

9. A range Doppler processing device for a target, characterized in that: include: An echo signal sampling module is used to obtain a radar echo signal of a target and perform pulse sampling on the radar echo signal to determine radar observation data in a one-dimensional vector form; an echo signal reconstruction module, configured to reconstruct the radar observation data based on the number of pulses and the number of sampling times when pulse sampling is performed on the radar echo signal, and determine a two-dimensional measurement matrix corresponding to the radar observation data; the number of rows of the two-dimensional measurement matrix is ​​related to the number of sampling times, the number of columns of the two-dimensional measurement matrix is ​​related to the number of pulses, and the number of sampling times is greater than the number of pulses; a batch division module, configured to divide the two-dimensional measurement matrix into a plurality of batches of first reconstructed signals according to a computing power parameter of a graphics processing unit (GPU); each batch of the first reconstructed signals includes a plurality of columns of data in the two-dimensional measurement matrix; a range Doppler processing module, configured to use the GPU to perform range Doppler signal recovery processing on each batch of the first reconstructed signals in parallel, determine a first target recovery signal corresponding to each batch of the first reconstructed signals, and determine a range Doppler processing result corresponding to the target based on the first target recovery signal of each batch.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the range Doppler processing method according to any one of claims 1 to 8 is implemented.

Images