Adaptive radar constant false alarm rate detection method based on dynamic length truncation and segmented threshold compensation

Abstract

This invention discloses an adaptive radar constant false alarm rate (CFAR) detection method based on dynamic length truncation and segmented threshold compensation, mainly addressing the problem that existing technologies cannot simultaneously achieve both detection accuracy and real-time computation in dense multi-target scenarios. The scheme includes: 1) acquiring the amplitude data of the radar echo signal and constructing reference sliding windows on both sides of each target unit; 2) calculating the mean ratio of the left and right reference units and determining whether to perform truncation processing based on it; 3) obtaining the estimated noise power value after truncation using a preset gradient strategy; 5) determining the compensation factor value based on the effective sample length; 7) obtaining the nominal factor through a dynamic sample generation module, and calculating the final detection threshold by combining the estimated value and the compensation factor, and outputting the detection result. This invention has a constant data throughput and short processing time, and can achieve high-precision anti-masking radar signal detection while maintaining deterministic low latency, meeting the requirements of autonomous driving for deterministic low latency.

Claims

1. An adaptive radar constant false alarm rate (CFAR) detection method based on dynamic length truncation and piecewise threshold compensation, characterized in that, Includes the following steps: (1) Acquire the amplitude data of the radar echo signal, construct reference sliding windows on both sides of each unit CUT to be detected, and divide them into left and right reference unit sets. The total length of the reference sliding window is NSlide; (2) Calculate the mean values ​​mu_left and mu_right of the left and right reference cells respectively, then obtain the mean ratio MR, and compare MR with the preset edge determination threshold to make the following determination: if MR is greater than the preset edge determination threshold, it is determined to be a clutter edge mode, and the mean of the larger side is taken as the background estimate. The effective sample length m is the length of the single-side reference sliding window NSlide / 2, and step (7) is executed directly; otherwise, it is determined to be a uniform or multi-target interference mode, and step (3) is executed to perform truncation processing. (3) Merge all samples in the reference sliding window into a sequence cell, sort them in ascending order to obtain an ordered sequence, and select the median as the noise baseline value of the current window. (4) A preset gradient strategy is adopted. By setting the gradient factor Kgrad and combining it with the median to calculate the decision threshold, the system performs a reverse scan and removes interfering samples to ensure the accuracy of background estimation. This includes: (4.1) Preset the gradient factor Kgrad according to the signal characteristics and scenario requirements; (4.2) The result of multiplying the median by the preset gradient factor Kgrad is used as the decision threshold Limit Value; (4.3) Perform reverse scanning from large to small values ​​for the ordered sequence. If the gradient between the current sample value and the median is greater than the preset decision threshold, the sample is determined to be an interference target and is removed. Otherwise, the sample and all remaining smaller samples are determined to be background noise and the scanning stops. The samples that are not removed are valid samples with lengths m=1, 2, ..., NSlide. (4.3) Calculate the arithmetic mean of the effective samples to obtain the truncated noise power estimate. ; (5) Determine if truncation has occurred. If the effective sample length m is less than the total length of the reference sliding window NSlide, then interference is detected and truncation has occurred. Continue to step (6) to obtain the compensation factor through the compensation strategy. If the effective sample length m equals the total reference sliding window length NSlide, then it is determined that no truncation has occurred, and the compensation factor is adjusted accordingly. Take 1 and execute step (7) directly. (6) Calculate the cutoff ratio and preset the threshold, construct a segmented threshold compensation strategy that includes two compensation modes: mild interference and severe interference, dynamically select the compensation mode according to the cutoff ratio, and obtain the compensation factor. ; (7) Generate a dynamic sample length matching lookup table (LUT) through the dynamic sample generation module, including: under the preset false alarm probability Pfa, for each possible effective sample length m, calculate its corresponding nominal factor α, and store the mapping relationship (m, α) in the storage unit to generate the lookup table LUT. The query operation of the table includes: directly reading the corresponding nominal factor from the LUT according to the effective sample length m. ; (8) Calculate the final detection threshold according to the following formula. : ； (9) Compare the amplitude of the unit to be detected with the final detection threshold T, and output the logical result of whether the target exists or not.

2. The method according to claim 1, characterized in that: The mean ratio MR mentioned in step (2) is calculated according to the following formula: 。 3. The method according to claim 1, characterized in that: The truncation ratio mentioned in step (6) is obtained by calculating the ratio of the effective sample length m to the total length NSlide of the reference sliding window.

4. The method according to claim 3, characterized in that: The step (6) of dynamically selecting the compensation mode based on the truncation ratio includes: when the truncation ratio is less than a preset threshold, selecting the mild interference compensation mode; when the truncation ratio is greater than or equal to the preset threshold, automatically switching to the severe interference compensation mode.

5. The method according to claim 4, characterized in that: The aforementioned mild interference compensation mode specifically employs a linear compensation function to calculate the compensation factor. To maintain high-precision false alarm control capabilities: ； in, This is the preset linear adjustment slope coefficient.

6. The method according to claim 5, characterized in that: The compensation effect of the linear compensation function increases monotonically with the increase of the truncation depth, and is used to accurately offset the noise power estimation bias caused by the reduction of samples.

7. The method according to claim 5, characterized in that: The preset linear adjustment slope coefficient is used to control the rate and amplitude of the change of the compensation factor with the cutoff ratio; the value is determined by simulation or actual measurement according to the performance requirements of the radar system, the noise environment and the target detection scenario.

8. The method according to claim 4, characterized in that: The severe interference compensation mode specifically uses a fixed threshold compensation method to obtain the compensation factor. Let the preset fixed compensation factor be ,but .

9. The method according to claim 8, characterized in that: The fixed compensation factor , is a real number greater than 1, whose value is determined through simulation or actual measurement. It is used to ensure that the detection threshold meets the preset false alarm rate index in the case of heavy interference, so as to prevent false alarms caused by too low a threshold or false alarms caused by too high a threshold.

10. The method according to claim 1, characterized in that: Step (9) compares the amplitude of the unit to be detected with the final detection threshold T. If the amplitude of the unit to be detected is greater than or equal to T, the logical result "target exists" is output; otherwise, the logical result "target does not exist" is output.

Images