Radar signal CFAR algorithm implementation method based on GPU acceleration

Abstract

The invention relates to the field of radar signal CFAR processing, in particular to a radar signal CFAR algorithm implementation method based on GPU acceleration. According to the technical scheme, the method comprises the following steps: calculating module values of a main channel signal and an auxiliary channel signal after being subjected to an MTI / MTD algorithm; after signal noise parameter estimation of different sampling points is completed, signal noise parameter estimation data of different sampling points are compared with a signal threshold, and a distance dimension peak value is searched; the maximum value of each sampling point signal in the same distance dimension and different Doppler dimensions is calculated, when two-dimensional CFAR processing is carried out, zero Doppler output is detected through a clutter map, output of each of other Doppler filters is detected through a unit average CFAR detector, and the maximum value of each sampling point signal in the same distance dimension and different Doppler dimensions is calculated. The output of the clutter map detection and the output of the CFAR detector are combined in a large selection mode to serve as a final detection result. The method is suitable for radar signal CFAR processing.

Description

Technical Field

[0001] This invention relates to the field of radar signal CFAR processing, and specifically to a method for implementing a radar signal CFAR algorithm based on GPU acceleration. Background Technology

[0002] With the continuous improvement of the detection capabilities and intelligence level of modern radar systems, higher demands are placed on the accuracy, real-time performance, and adaptability of radar target detection. Phased array radar, as an important representative of digital array radar, possesses outstanding advantages such as multi-beam, multi-target, and high resolution. However, its echo signal processing often needs to cope with complex and ever-changing clutter environments and electromagnetic interference. To ensure the target detection probability under different backgrounds and strictly control the false alarm probability, CFAR (Constant False Alarm Rate) technology has become one of the core algorithms in modern radar signal processing. CFAR adaptively adjusts the detection threshold based on the statistical characteristics of the reference cells surrounding the detection unit, effectively suppressing false alarm rate fluctuations caused by clutter, noise, and active interference, thereby improving the robustness and reliability of target detection.

[0003] In the processing of multi-channel, multi-pulse echo signals in phased array radar, the CFAR algorithm needs to perform sliding window statistics, threshold calculation, and decision operations on large-scale two-dimensional or even three-dimensional data. With the increasing number of array elements and data channels, the amount of data per frame and the real-time processing pressure rise sharply. Traditional processing architectures based on DSPs (Digital Signal Processors) or CPUs (Central Processing Units) are no longer sufficient to meet the application requirements of high channel count, high throughput, and high real-time performance. Especially in multi-beam, broadband, and high-resolution systems, the parallel data processing capability of the CFAR algorithm becomes a key bottleneck restricting target detection performance. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a GPU-accelerated radar signal CFAR algorithm implementation method, which improves the efficiency of radar signal CFAR processing.

[0005] The present invention achieves the above objectives by adopting the following technical solution: the present invention provides a method for implementing a radar signal CFAR algorithm based on GPU acceleration, comprising:

[0006] S1. Calculate the magnitude values ​​of the main channel signal and the auxiliary channel signal after passing through the MTI (Moving Target Indication) / MTD (Moving Target Detection) algorithm;

[0007] Multiple thread workgroups are set up using the GPU, with each workgroup containing multiple threads. Each thread is responsible for calculating the modulus operation for one sampling point to obtain the signal and noise parameter estimates for each sampling point.

[0008] S2. After estimating the signal and noise parameters at different sampling points, compare the estimated signal and noise parameters at different sampling points with the signal threshold and find the peak value in the distance dimension.

[0009] Multiple thread workgroups are set up using the GPU, with each workgroup containing multiple threads. Each thread completes the comparison between the estimated signal and noise parameters of a sampling point and the signal threshold.

[0010] S3. Calculate the maximum value of the signal at each sampling point in different Doppler dimensions at the same distance dimension;

[0011] During two-dimensional CFAR processing, the zero-Doppler output is detected by clutter map, and the output of each other Doppler filter is detected by a unit-averaged CFAR detector. The CFAR detector, together with the sidelobe anechoic channel data and the anechoic threshold, realizes the sidelobe anechoic operation. The clutter map output and the CFAR detector output are combined by selecting the largest value to obtain the final detection result.

[0012] Furthermore, the detection of zero Doppler output via clutter maps specifically includes:

[0013] Signal intensity of each distance cell in the zero Doppler channel after MTI / MTD processing. And set the clutter map threshold corresponding to the current azimuth angle. With signal strength If a comparison is made, Then it is determined that a target exists in the distance cell, where , This is the false alarm control factor, based on the expected false alarm probability. and noise distribution model settings, For distance cells in clutter diagrams and azimuth Clutter power estimation at the location.

[0014] Furthermore, the output of each Doppler filter is detected by a unit-averaged CFAR detector, specifically including:

[0015] The average signal energy of all cells within the reference window is used as the background noise power estimate for the current cell under test. , This represents the background noise power estimate for the i-th cell under test. This represents the set of indices for the reference window. An adaptive detection threshold is constructed based on the sum of the reference window indices. If the signal energy of the unit under test If the distance is within a certain range, then it is determined that a target exists in that range cell; otherwise, no target exists.

[0016] Furthermore, the CFAR detector, in conjunction with sidelobe cloning channel data and cloning threshold, achieves sidelobe cloning specifically by including:

[0017] Extracting the complex signal power of the main channel With auxiliary channel complex signal power ,in Indicates the main channel signal. Indicates the main channel I signal. Indicates the main channel Q signal, subscript This represents the main channel, r represents the range element, and d represents the corresponding Doppler element. Indicates auxiliary channel signal, Indicates auxiliary channel I signal, Indicates the auxiliary channel Q signal, subscript This refers to the auxiliary channel; and the energy difference between the primary and auxiliary channels is defined. ,when hour, If a stealth threshold is set for the system, the signal is considered to enter from the sidelobe and is not a valid target.

[0018] Furthermore, firstly, for each distance unit in the two-dimensional distance-Doppler plane... The clutter map detection results in the zero-Doppler channel and the CFAR detection results in other non-zero-Doppler channels were obtained respectively. Each detection result contains two pieces of information: one is whether the range-Doppler cell is identified as a target, and the other is the corresponding signal energy intensity value, i.e., the squared magnitude. ;

[0019] Next, for each distance unit traversing all its Doppler channels The system filters out all channels that are identified as targets and compares the signal energy intensities of these channels. Then, it selects the channel with the highest signal energy from all the channels that detect targets as the optimal detection result for that range unit. If multiple channels detect targets at the same time, only the result of the channel with the highest energy is retained, and the rest are discarded. If no channel detects targets, it is determined that there are no targets in that range unit.

[0020] The beneficial effects of this invention are as follows:

[0021] This invention improves clutter suppression by adding comparison with clutter images to the standard CFAR algorithm. It also implements a CFAR algorithm for echo signals with a variable number of channels, utilizing GPU parallel processing to meet real-time requirements. With sufficient GPU memory, it can support CFAR algorithm processing for multi-pulse signals while maintaining real-time processing speed. Attached Figure Description

[0022] Figure 1 This is a flowchart of a GPU-accelerated radar signal CFAR algorithm implementation method provided by the present invention;

[0023] Figure 2 This is a flowchart provided by the present invention for obtaining signal noise parameter estimation for each sampling point through GPU parallel computing;

[0024] Figure 3 This is a flowchart provided by the present invention that compares the estimated signal noise parameters at different sampling points with the signal threshold using GPU parallel computing;

[0025] Figure 4 This is a flowchart provided by the present invention for calculating the maximum value of the signal at each sampling point in different Doppler dimensions at the same distance dimension;

[0026] Figure 5 This is a schematic diagram of the CFAR pre-pulse pressure results provided by the present invention;

[0027] Figure 6 This is a schematic diagram of the distance-dimensional CFAR output results provided by the present invention;

[0028] Figure 7 This is a schematic diagram of the CFAR pre-MTD results provided by the present invention;

[0029] Figure 8 This is the range-Doppler two-dimensional CFAR output result provided by the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0031] This invention provides a GPU-accelerated implementation method for radar signal CFAR algorithm, such as... Figure 1 As shown, it specifically includes:

[0032] S1. Calculate the magnitude values ​​of the main channel signal and the auxiliary channel signal after applying the MTI / MTD algorithm;

[0033] Before performing CFAR, modulo operations need to be performed on the data containing MTI and MTD. This modulo operation only needs to be performed on the data from the MTI and MTD channels. When implementing this on a GPU, each thread in the workgroup can calculate the modulus of one signal channel, with one thread in the workgroup responsible for calculating the modulo operation for one sample point.

[0034] like Figure 2 As shown, multiple thread workgroups are set up using the GPU. Each thread workgroup contains multiple threads, and each thread is responsible for calculating the modulus operation for one sampling point to obtain the signal and noise parameter estimates for each sampling point.

[0035] S2. After estimating the signal and noise parameters at different sampling points, compare the estimated signal and noise parameters at different sampling points with the signal threshold and find the peak value in the distance dimension.

[0036] To perform a CFAR operation on a signal with n points, noise parameter estimation needs to be performed on each of the n sampling points using a reference window, and then compared with a threshold. Since the noise parameter estimations for these n reference windows are independent, when implementing this algorithm in parallel using a GPU, the number of thread workgroups can be reasonably arranged according to the number of sampling points n. One thread workgroup completes the noise parameter estimation and threshold comparison for a portion of the sampling points, and all thread workgroups work together to complete the work for all n sampling points. Each thread in the thread workgroup completes the reference window noise parameter estimation and threshold comparison for one sampling point.

[0037] like Figure 3 As shown, multiple thread workgroups are set up using the GPU. Each thread workgroup contains multiple threads, and each thread completes the comparison between the estimated signal and noise parameters of a sampling point and the signal threshold.

[0038] S3. Calculate the maximum value of the signal at each sampling point in different Doppler dimensions at the same distance dimension;

[0039] like Figure 4 As shown, during two-dimensional CFAR processing, the zero-Doppler output is first detected using a clutter map. Specifically, the detection method involves extracting the signal strength of each distance cell in the zero-Doppler channel after MTI / MTD processing. And set the clutter map threshold corresponding to the current azimuth angle. With signal strength If a comparison is made, Then it is determined that a target exists in the distance cell, where , This is the false alarm control factor, based on the expected false alarm probability. And the noise distribution model is set, while For distance cells in clutter diagrams and azimuth Clutter power estimation at the location.

[0040] Each of the other Doppler filter outputs is detected by a unit-averaged CFAR detector. Specifically, the detection method involves averaging the signal energy of all units within the reference window to obtain an estimate of the background noise power of the current unit under test. , This represents the set of indices for the reference window. The sum of the reference window indices. Construct an adaptive detection threshold. If the signal energy of the unit under test If the distance is within a certain range, then it is determined that a target exists in that range cell; otherwise, no target exists.

[0041] The CFAR detector, in conjunction with sidelobe cloning channel data and cloning threshold, enables sidelobe cloning operations. Specifically, it extracts the complex signal power of the main channel. With auxiliary channel complex signal power ,in Indicates the main channel signal. Indicates the main channel I signal. Indicates the main channel Q signal, subscript This represents the main channel, r represents the range element, and d represents the corresponding Doppler element. Indicates auxiliary channel signal, Indicates auxiliary channel I signal, Indicates the auxiliary channel Q signal, subscript This refers to the auxiliary channel; and the energy difference between the primary and auxiliary channels is defined. ,when hour, If a stealth threshold is set for the system, the signal is considered to enter from the sidelobe and is not a valid target.

[0042] The clutter map detection output and the CFAR detector output are combined using a maximum-valued merging process to obtain the final detection result. Specifically, for each range cell in the two-dimensional range-Doppler plane... The clutter map detection results in the zero-Doppler channel and the CFAR detection results in other non-zero-Doppler channels were obtained. Each detection result contains two pieces of information: first, whether the range-Doppler cell is identified as a target (binary decision flag); and second, the corresponding signal energy intensity value (i.e., the squared magnitude). Next, for each distance unit... traversing all its Doppler channels The process involves filtering out all channels identified as targets and comparing the signal energy intensities of these channels. Then, from all channels that detected the target, the channel with the highest signal energy is selected as the optimal detection result for that range unit. If multiple channels detect the target simultaneously, only the result of the channel with the highest energy is retained, and the others are discarded; if no channel detects the target, the range unit is determined to have no target.

[0043] The advantages of this invention are verified below through a practical data processing example. The actual data signal is a linear frequency modulated signal with a pulse width of 30μs and a bandwidth of 10MHz. Figure 5 This is the result of signal pulse compression on actual data. Figure 6 The output of the distance-dimensional CFAR processing on the signal pulse compression result. Figure 7 This is the MTD result before CFAR. Figure 8 This is the output of a two-dimensional CFAR analysis of the MTD results.

[0044] The following table 1 compares the time performance of the GPU parallel processing method of the present invention with that of the existing CPU processing method.

[0045] Table 1. Comparison of CPU and GPU processing times for distance-Doppler 2D CFAR.

[0046]

[0047] The table shows that when the signal pattern is a linear frequency modulated signal with a pulse width of 30μs and a bandwidth of 10MHz, the time required for the GPU to complete the distance-Doppler two-dimensional CFAR is significantly lower than that of the CPU, resulting in a significant performance improvement.

[0048] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for implementing a GPU-accelerated radar signal CFAR algorithm, characterized in that, include: S1. Calculate the magnitude values ​​of the main channel signal and the auxiliary channel signal after applying the MTI / MTD algorithm; Multiple thread workgroups are set up using the GPU, with each workgroup containing multiple threads. Each thread is responsible for calculating the modulus operation for one sampling point to obtain the signal and noise parameter estimates for each sampling point. S2. After estimating the signal and noise parameters at different sampling points, compare the estimated signal and noise parameters at different sampling points with the signal threshold and find the peak value in the distance dimension. Multiple thread workgroups are set up using the GPU, with each workgroup containing multiple threads. Each thread completes the comparison between the estimated signal and noise parameters of a sampling point and the signal threshold. S3. Calculate the maximum value of the signal at each sampling point in different Doppler dimensions at the same distance dimension; During two-dimensional CFAR processing, the zero-Doppler output is detected by clutter map, and the output of each other Doppler filter is detected by a unit-averaged CFAR detector. The CFAR detector, together with the sidelobe anechoic channel data and the anechoic threshold, realizes the sidelobe anechoic operation. The clutter map output and the CFAR detector output are combined by selecting the largest value to obtain the final detection result.

2. The method for implementing the GPU-accelerated radar signal CFAR algorithm according to claim 1, characterized in that, Zero Doppler output is detected via clutter maps, specifically including: Signal intensity of each distance cell in the zero Doppler channel after MTI / MTD processing. And set the clutter map threshold corresponding to the current azimuth angle. With signal strength If a comparison is made, If the distance cell contains a target, then it is determined that a target exists. , This is the false alarm control factor, which is based on the expected false alarm probability. and noise distribution model settings, For distance cells in clutter diagrams and azimuth Clutter power estimation at the location.

3. The method for implementing the GPU-accelerated radar signal CFAR algorithm according to claim 1, characterized in that, Each Doppler filter output is detected by a unit-averaged CFAR detector, specifically including: The average signal energy of all cells within the reference window is used as the background noise power estimate for the current cell under test. , This represents the background noise power estimate for the i-th cell under test. This represents the set of indices for the reference window. An adaptive detection threshold is constructed based on the sum of the reference window indices. If the signal energy of the unit under test If the distance is within a certain range, then it is determined that a target exists in that range cell; otherwise, no target exists.

4. The method for implementing the GPU-accelerated radar signal CFAR algorithm according to claim 1, characterized in that, The CFAR detector, in conjunction with sidelobe cloning channel data and cloning threshold, enables sidelobe cloning operations, specifically including: Extracting the complex signal power of the main channel ,in Indicates the main channel signal. Indicates the main channel I signal. Indicates the Q-channel signal of the main channel, subscript This represents the main channel, r represents the range cell, and d represents the Doppler cell corresponding to the range cell; Extracting the complex signal power of the auxiliary channel ,in, Indicates auxiliary channel signal, Indicates auxiliary channel I signal, Indicates the auxiliary channel Q signal, subscript This refers to the auxiliary channel; And define the energy difference between the main and auxiliary channels. ,when hour, If a stealth threshold is set for the system, the signal is considered to enter from the sidelobe and is not a valid target.

5. The method for implementing the GPU-accelerated radar signal CFAR algorithm according to claim 1, characterized in that, The clutter map detection output and the CFAR detector output are combined using the largest value to obtain the final detection result. Specifically, this includes: First, for each distance unit in the two-dimensional distance-Doppler plane The clutter map detection results in the zero-Doppler channel and the CFAR detection results in other non-zero-Doppler channels were obtained respectively. Each detection result contains two pieces of information: one is whether the range-Doppler cell is identified as a target, and the other is the corresponding signal energy intensity value, i.e., the squared magnitude. ; Next, for each distance unit traversing all its Doppler channels The system filters out all channels that are identified as targets and compares the signal energy intensities of these channels. Then, it selects the channel with the highest signal energy from all the channels that detect targets as the optimal detection result for that range unit. If multiple channels detect targets at the same time, only the result of the channel with the highest energy is retained, and the rest are discarded. If no channel detects targets, it is determined that there are no targets in that range unit.

Images