A velocity deambiguation method for 4D millimeter-wave radar imaging

Abstract

A velocity deambiguation method suitable for 4D millimeter-wave radar imaging comprises the following steps: Step 1: obtaining radar original target sampling data, performing two-dimensional fast Fourier transform processing, and obtaining a target range-Doppler matrix; performing constant false alarm detection on the target range-Doppler matrix; Step 2: finding the best representative point as the target's velocity representative point; Step 3: applying a hypothetical phase compensation velocity deambiguation algorithm to the target's velocity representative point to obtain the representative point's corresponding velocity and phase compensation value; Step 4: compensating all target points with the phase compensation value calculated based on the representative point, and performing DOA estimation to obtain an azimuth angle value; Step 5: performing coordinate conversion based on the obtained target point velocity and azimuth angle value to obtain the target's true position in a three-dimensional coordinate system, with the representative point velocity representing the target's true velocity. The present invention can enhance the detection capability of a MIMO radar imaging system for high-speed targets.

Description

Technical Field

[0001] The present invention belongs to the field of radar imaging technology, and in particular relates to a velocity deambiguation method suitable for 4D millimeter-wave radar imaging. Background Art

[0002] In radar systems, Doppler shift is used to measure a target's velocity relative to the radar. However, velocity ambiguity occurs when the Doppler shift generated by the target's velocity exceeds the maximum frequency that the radar system can process. While MIMO radar can provide richer target information, such as additional information like target height and profile, its operating principle significantly reduces the radar system's unambiguous detection velocity range. Inaccurate target velocity estimation can affect the accuracy of target angle estimation, making velocity ambiguity resolution crucial for MIMO imaging radars.

[0003] Methods based on pulse repetition frequency (PRF) variation are usually common solutions to velocity ambiguity problems, including the Chinese Remainder Theorem (CRT) method, table lookup method, and one-dimensional set algorithm.

[0004] The Chinese Remainder Theorem (CRT) is an effective method for eliminating phase uncertainty in Doppler shift detection. It resolves ambiguity by establishing a system of congruence equations between the measured value and the target velocity. However, in the classic CRT, the greatest common divisor (GDC) of any pair of PRFs is equal to 1. While this method for resolving Doppler ambiguity is quite accurate and fast, its accuracy degrades rapidly when the remainder exhibits some deviation due to noise. The classic lookup table method precalculates and stores velocity-Doppler shift relationships for different scenarios to quickly match the observed Doppler shift to the target's true velocity. While this method is simple and easy to implement in engineering, the table storage requires a large amount of storage space. Furthermore, the lookup table divides the distance units into degrees, wasting a significant amount of time on unnecessary numerical calculations. This makes it difficult to achieve real-time processing requirements and is not adaptable to complex scenarios. Furthermore, the computational complexity of the one-dimensional set algorithm increases exponentially with the number of targets within the radar's detection range, making it difficult to achieve real-time radar detection.

[0005] The hypothetical phase compensation (HPC) velocity deambiguation algorithm can estimate the true target velocity under large velocity ambiguity periods and perform velocity interpretation and deambiguation within a single frame. This approach is suitable for MIMO radar systems, but for imaging, due to its high spatial resolution, it can decompose a single target into multiple target points. This high resolution increases the number of target points. Applying the HPC velocity deambiguation algorithm to each target point in the process of processing each target point to eliminate velocity ambiguity significantly increases the algorithm's time complexity, thereby reducing system processing speed. Furthermore, the higher energy of sidelobe points can lead to velocity expansion, which can generate noise in the point cloud data.

[0006] The patent document "A method, device and millimeter wave radar for speed deambiguation of radar" (application date: August 15, 2022, announcement number: CN115421134B) discloses a method, device and millimeter wave radar for speed deambiguation of radar. This method receives corresponding echo signals through at least four antennas, and calculates the estimated speed of the target in combination with the phase relationship of each antenna. According to the estimated speed of the target, combined with the known maximum fuzzy speed, the ambiguity is obtained; based on the ambiguity, the Doppler shift is performed and the reference speed of the target is calculated as the true speed. However, this method is not suitable for imaging radar, and multiple frames of data are required for deambiguation.

[0007] The patent document "A method for obtaining entropy to resolve velocity ambiguity in millimeter-wave MIMO traffic radar" (application date: February 27, 2020, announcement number: CN111308437B) discloses a method for obtaining entropy to resolve velocity ambiguity in millimeter-wave MIMO traffic radar. This method obtains the preliminary distance, speed, and angle information of each target, calculates the unambiguous speed of the target, establishes an optimization model, and obtains 2L groups of different compensation phases; obtains the minimum entropy value e, and calculates the corresponding unambiguous speed at this time based on its position. This method requires processing each target point, but for 4D millimeter-wave radar, because of its high spatial resolution, one target will usually be identified as multiple target points. If the above processing is performed on each target point, the real-time performance of the algorithm will be greatly reduced. Summary of the Invention

[0008] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a velocity deambiguation method suitable for 4D millimeter-wave radar imaging. By adopting a hypothetical phase compensation velocity deambiguation algorithm, the velocity deambiguation problem of moving targets with a large velocity ambiguity period is solved using single-frame data, thereby improving the detection capability of MIMO radar imaging systems for high-speed targets.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is:

[0010] A velocity deambiguation method suitable for 4D millimeter-wave radar imaging, comprising the following steps:

[0011] Step 1: The 4D millimeter-wave radar mixes the received and transmitted chirp signals to generate a difference frequency signal, which is then processed by a two-dimensional fast Fourier transform (2D-FFT) to obtain the target range-Doppler (RDM) matrix.

[0012] And through constant false alarm detection, the target point distance, target point speed and phase information are obtained;

[0013] Step 2: Find the best representative point as the velocity representative point of the target, which is located near the midpoint of the line connecting the peak point of the target point cloud image and the geometric center point of the target RDM;

[0014] Step 3: Use the hypothetical phase compensation velocity defuzzification algorithm on the velocity representative point of the target to obtain the velocity and phase compensation values ​​corresponding to the velocity representative point;

[0015] Step 4: Select the velocity ambiguity period corresponding to the peak point of the angular power spectrum of the target point under the assumptions in step 3, compensate all target points with the phase compensation value calculated based on the velocity representative point, and perform DOA estimation to obtain the azimuth angle value;

[0016] Step 5: Perform coordinate transformation based on the obtained target point speed and azimuth angle values ​​to obtain the true position of the target in the three-dimensional coordinate system. The speed representative point speed represents the true speed of the target.

[0017] The step 1 is specifically as follows:

[0018] Step (1): Sample the difference frequency signal generated by mixing the chirp signal transmitted and received by the 4D millimeter-wave radar. The sampling results are stored in the form of a matrix. The ADC module implemented based on FPGA performs sampling. The specific sampling parameter calculation is shown below.

[0019] The number of radar range FFT points is determined by the radar's maximum detection range and range resolution. The formula can be expressed as:

[0020]

[0021] Where N FFTR is the number of FFT points, R max Represents the maximum detection range of the radar, R res is the distance resolution.

[0022] ADC sampling rate F s According to the Nyquist sampling theorem, the effective sampling frequency f of the radar is IFmax The maximum detection range of the radar is as follows:

[0023]

[0024] Where f IFmax Indicates the effective sampling frequency, c = 3 × 10 8 m / s represents the speed of light, and S is the frequency modulation slope of the Chirp signal. Under the ADC complex sampling condition, the ADC sampling frequency F S Just need to meet F S ≥f IFmax , the intermediate frequency signal can be restored, so the sampling frequency must meet the following requirements:

[0025]

[0026] Step (2): performing FFT transformation on each row and each column of the matrix to obtain a range-Doppler spectrum, and estimating the range and speed of the target by analyzing the target range-Doppler spectrum (RDM);

[0027] Step (3): Use the ordered statistics (OS-CFAR) detection algorithm to perform constant false alarm detection on the target RDM; obtain the target point distance, target point speed and phase information.

[0028] The step (2) is specifically as follows:

[0029] The echo delay is calculated by the spectrum of the difference frequency signal to determine the target distance. The transmitted Chirp signal is expressed by the following formula:

[0030] S TX =Aexp(j×(2πf c t+πkt 2 )), t∈(0,T c )

[0031] Among them, f c is the carrier frequency, T c is the period, A is the amplitude, is the slope of frequency versus time, B is the signal bandwidth, and t is the time coordinate of the signal frequency changing with time within one cycle;

[0032] The difference frequency signal is generated by mixing the target reflected signal with the reference signal, and is expressed as follows:

[0033] S IF =Aexp(j×(2π·kτ·t+2π·f c τ-πkt 2 ))

[0034] in is the round-trip delay of the electromagnetic wave, c is the speed of light, R represents the distance between the target and the radar, the frequency of the difference frequency signal is linearly related to the time delay τ, and an FFT is performed on the difference frequency signal to obtain the signal frequency of the difference frequency signal, and the distance R of the target is calculated. The formula is:

[0035]

[0036] In digital domain processing, the number of sampling points is N r , the sampling rate is F s Under the sampling condition of , the distance resolution R of the distance information obtained after digital signal processing is res , expressed as:

[0037]

[0038] For the same moving target, perform FFT on two adjacent difference frequency signals separately to obtain different phase information. The one-dimensional FFT peaks of the two difference frequency signals are the same, but the phases are different. The speed information of the target can be obtained based on this phase difference.

[0039] The process of applying the two-dimensional FFT algorithm to process a frame of target difference frequency signal is expressed as:

[0040] First, the echo signals of multiple chirps in each frame are sampled and stored in the form of a matrix. The sampling result of each chirp signal is stored in the matrix row by row. Then, each row and column of the matrix is ​​transformed by FFT to obtain the range-Doppler spectrum. By analyzing the range-Doppler spectrum (RDM), the range and speed of the target are estimated.

[0041] The step (3) is specifically as follows:

[0042] 1) First, the threshold of the distance dimension of the tested unit in the RDM matrix is ​​solved. According to the principle of CACFAR algorithm, the detection threshold of the tested unit in the distance dimension is: Among them, M is the number of training units, rx i , r y i, respectively, are the training units on the left and right of the unit under test, and K is the threshold factor;

[0043] 2) Then, the threshold of the unit under test in the velocity dimension of the RDM matrix is ​​solved in the same way as the distance dimension. The threshold of the unit under test in the velocity dimension is dx i , dy are the training units above and below the tested unit respectively;

[0044] 3) After calculating the distance dimension and speed dimension thresholds, the tested unit is tested for crossing the threshold. The crossing threshold test can be expressed as (A>T r )&(A>Td ), A is the power of the unit under test in the RDM. Only when it exceeds the distance dimension and speed dimension threshold at the same time, it is identified as a target;

[0045] After CFAR, the distance, velocity and phase information of the target point are obtained;

[0046] The velocity and phase information of the target point serve as the input for subsequent velocity deambiguation. The target point distance corresponds to the distance index, and the target point velocity corresponds to the Doppler index. The target velocity is calculated by the Doppler frequency shift of the target. The phase information and the echo intensity of the target signal are the power amplitude of the target point.

[0047] The step 2 is specifically as follows:

[0048] First, find the point with the highest energy in the target's point cloud image, i.e., the peak point, and connect it with the target's RDM geometric center to obtain the midpoint. The target RDM geometric center is obtained by performing a 2D FFT on the RDM, then extracting the target area after target detection and calculating the geometric center point. Then, traverse all target points and select the point in the RDM closest to this midpoint as the target's velocity representative point.

[0049] The geometric center point of the RDM is obtained by RDM calculation, and then the optimal point is found near the midpoint between the geometric center point of the RDM and the peak point.

[0050] In step 3, the velocity defuzzification algorithm using the assumed phase compensation is applied to the velocity representative point of the target. The method is as follows:

[0051] First, the detection phase shift of the CFAR detection output of the target point is Enumerate all possible phase values ​​that may need to be compensated:

[0052]

[0053] Secondly, according to formula (1), the radar data is compensated with each hypothetical phase compensation value in turn, and the compensated data is expressed as

[0054]

[0055] Among them, S i Represents the signal data in the i-th channel, i starts from 2, Indicates the phase value that needs to be compensated. At the same time, data is extracted from the same index position in each channel RDM to form a channel data set as follows:

[0056]

[0057] The value of N depends on the number of transmitting antennas M TX To calculate, N satisfies:

[0058] |N|<=(M ТХ -1) / 2 (4)

[0059] When the number of transmitting antennas is odd;

[0060] |N|<=(M TX / 2)-1 (5)

[0061] When the number of transmitting antennas is even;

[0062] Use formulas (4) and (5) to get the velocity blur period N, where M TX is the number of transmitting antennas;

[0063] Then, according to the N value, all HPC situations are listed, and the radar data are compensated in turn with each hypothetical phase compensation value. H satisfies formula (2). Then, FFT operation is performed on each compensated data (2N+1 in total) to obtain the angular power spectrum under each hypothetical condition;

[0064] H=H -N ,...,H k ,...,H N (6)

[0065] By selecting the peak point in the angular power spectrum, the true speed of each speed representative point is determined, and the speed corresponding to the speed representative point is the true speed of the entire target.

[0066] The step 4 is specifically as follows:

[0067] When the RDM containing the target speed and distance information is processed by CFAR detection, the estimated value of the target speed and the Doppler phase difference are obtained;

[0068] Set the result sequence passing through the CFAR target point to X (m , n) (0 <m<N TX ,0 <n<N RX ), then the phase compensation satisfies the formula:

[0069] Where X(m,n) is the original signal matrix, is the phase factor, m is the transmit antenna index, n is the receive antenna index, N TX is the number of transmitting antennas, N RX is the number of receiving antennas;

[0070] Use angular FFT to estimate DOA. When using FFT to process signals, the direction information of the target signal is "mapped" to the frequency domain, and the direction of arrival of the signal is determined by calculating the energy spectrum at different angles.

[0071] The formula is:

[0072]

[0073] Where M represents the number of receiving antennas;

[0074] The result of FFT processing is an angle spectrum P(θ). The arrival angle of the target is estimated by finding the peak position of the power spectrum. In the calculated angle spectrum P(θ), the angle corresponding to the peak is the arrival angle of the target.

[0075] The step 5 is specifically as follows:

[0076] The target point distance, velocity and phase information are used as inputs of the improved hypothetical phase compensation velocity ambiguity resolution algorithm model. Based on the representative point selection method of the target RDM geometric center point and peak point, the representative points of the three targets are obtained, which are respectively recorded as: representative point 1, representative point 2 and representative point 3. For each representative point, RDM provides its velocity information V cfar1 , V cfar2 , V cfar3 , and phase information

[0077] The HPC velocity ambiguity resolution algorithm first derives the hypothetical velocity ambiguity period N based on the number of radar transmitting antennas. Then, based on this period and the known velocity information, all possible hypothetical phase compensation (HPC) situations are enumerated for each velocity representative point.

[0078] Next, each hypothetical phase compensation value is applied to the target point information to calculate the angular power spectrum under various hypothetical conditions. By selecting the peak point hypothesis in the angular power spectrum, the true speed of each speed representative point is determined. The speed corresponding to the speed representative point is the true speed of the entire target.

[0079] Based on the velocity ambiguity periods N1, N2, and N3 corresponding to the peak points, the phase compensation value of each target is calculated. Phase compensation is performed on all target points according to their respective deambiguation periods. Then, DOA estimation (such as angle-dimensional FFT and DBF) is performed to accurately obtain the true azimuth angle value of each target.

[0080] Finally, the true distance of the target is calculated based on the relationship between the target point distance index and the radar distance resolution, and the coordinate transformation is performed to form a point cloud image with target distance and angle information.

[0081] Phase information Perform azimuth-elevation FFT to obtain the target's AEM matrix;

[0082] Afterwards, the AEM is subjected to a constant false alarm detection method based on the normalized zero-point angle power spectrum to obtain the target azimuth and pitch angle information;

[0083] Finally, the target's distance, azimuth, and pitch angle information are converted into coordinates to generate a three-dimensional point cloud image of the target space. In the target point cloud image, target points with different speeds will be represented by different colors, that is, the four-dimensional information of the target (distance, altitude, azimuth, and speed) is displayed in the same image.

[0084] The method is applied to intelligent transportation systems and 4D millimeter-wave radar imaging systems.

[0085] Beneficial effects of the present invention:

[0086] First, the representative point selection method proposed in the present invention improves the processing efficiency of the speed defuzzification algorithm by selecting representative points in the RDM target point cloud cluster to implement speed fuzzy periodic solution instead of global point enumeration processing.

[0087] Second, the velocity defuzzification method proposed in the present invention ensures that points belonging to the same target point cloud image have the same velocity, thereby improving the stability of the velocity defuzzification method.

[0088] Third, the velocity defuzzification method proposed in this invention can not only output the accurate velocity value of the target, but also output the accurate phase compensation value, ensuring that the target point cloud image position is consistent with the target's true position, thereby improving the quality of the point cloud image. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 Schematic diagram of the difference frequency signal of the present invention.

[0090] Figure 2 It is a system block diagram of the present invention.

[0091] Figure 3 It is a schematic diagram of representative point selection of the present invention.

[0092] Figure 4 Schematic diagram of the implementation process of the present invention.

[0093] Figure 5 This is the test scene graph of the present invention.

[0094] Figure 6 This is a test effect diagram of the present invention; Figure 6 (a) is a schematic diagram without velocity defuzzification; Figure 6 (b) is a schematic diagram of the original HPC velocity deblurred spatial point cloud image; Figure 6(c) is the top view of the original HPC velocity deblurred spatial stereo point cloud image; Figure 6 (d) Schematic diagram of the improved HPC speed deblurred spatial stereo point cloud image; Figure 6 (e) A top view of the spatial stereo point cloud image with improved HPC speed deblurring.

[0095] Figure 7 This is a comparison chart of the time complexity of the methods.

[0096] Figure 8 Schematic diagram of the two-dimensional FFT algorithm.

[0097] Figure 9 This is a schematic diagram of constant false alarm detection. DETAILED DESCRIPTION

[0098] The present invention will be further described in detail below with reference to the accompanying drawings.

[0099] like Figure 1 、 Figure 2 As shown, a velocity deambiguation method suitable for 4D millimeter-wave radar imaging;

[0100] (1) The range-Doppler (RDM) matrix detection part first performs a two-dimensional fast Fourier transform (2D-FFT) on the radar original sampling data to obtain the target RDM matrix.

[0101] Next, the target RDM is subjected to constant false alarm detection using the ordered statistics (OS-CFAR) detection algorithm to obtain the target point distance, velocity, and phase information.

[0102] According to the proposed representative point selection method based on the combination of the target RDM geometric center point and the peak point, the representative points of the three targets are obtained, which are respectively recorded as: representative point 1, representative point 2 and representative point 3. RDM provides its velocity information V cfar1 , V cfar2 , V cfar3 , and phase information The velocity and phase information of the target point are used as input for subsequent velocity defuzzification;

[0103] (2) The representative point selection part based on the combination of the target RDM geometric center point and the peak point first finds the midpoint of the line connecting the target point cloud peak point and the target RDM geometric center point, then traverses all target points and selects the point in the RDM closest to this midpoint as the representative point of the target;

[0104] The advantage of selecting the representative point based on the geometric center point of the target in the radar range-Doppler map (RDM) is that it can provide a point that can better represent the overall position of the target. When the radar position is close to both sides of the road to observe the vehicle target, the target points obtained may present an "L"-shaped arrangement pattern. In this layout, the white dot represents the geometric center point obtained from the target RDM analysis. The method of selecting the peak point as the representative point can reflect the highest energy point of the target, but may not reflect the exact position of the target. Therefore, this application proposes a method of selecting representative points based on the geometric center point and peak point of the target RDM to obtain representative points for each of the three targets.

[0105] (3) The hypothetical phase compensation velocity deambiguation part of the representative point,

[0106] |N|=(M TX / 2)-1 (1)

[0107] Use formula (1) to get the velocity blur period N, where M TX For the number of transmitting antennas, a 12-transmit, 16-receive antenna array is used, resulting in a velocity ambiguity period of 5. Then, based on the value of N, all hypothetical phase compensation scenarios are listed. Each hypothetical phase compensation value is used to compensate the radar data in turn. An FFT operation is then performed on each compensated data to obtain the angular power spectrum under each hypothetical condition.

[0108] The assumption that the peak point in the angular power spectrum is selected, and the corresponding velocity and the angle obtained are the true values ​​of the representative point;

[0109] (4) The phase compensation part of the representative point and the subordinate point compensates the phases of all target points by the same phase compensation value as the representative point to obtain the true azimuth angle value of the target point;

[0110] (5) The coordinate conversion part performs coordinate conversion based on the obtained target point speed and azimuth angle value to obtain the real position of the target in the three-dimensional coordinate system. The representative point speed represents the real speed of the target, thereby realizing three-dimensional imaging of the target.

[0111] Figure 3 This is a schematic diagram of representative point selection in the present invention. Selecting the peak point as the representative point can reflect the target's highest energy point, but may not reflect the target's exact position. Selecting the target RDM geometric center point is closer to the target's true geometric center point, but this point may also introduce errors because it is located in the radar sidelobe area.

[0112] Therefore, if Figure 3As shown in the figure, the red dot represents the actual geometric center of the target. Since the geometric center of different moving targets varies, for example, a car's geometric center is located inside the car and cannot be directly detected by radar. Therefore, the strategy adopted is to find an optimal representative point located near the midpoint of the line connecting the target's peak point and the target's RDM geometric center. This point is used as the representative point. This approach can address the shortcomings of selecting a single representative point. Although the peak point, as the target's highest energy point, does indicate the location of the target's strongest reflection, this location does not always coincide with the target's geometric center. The geometric center point is not always located in the target's highest energy region. MIMO radars have high range and velocity resolution, which causes target energy to spread, potentially resulting in the point being located within the radar's sidelobe region. Selecting such a geometric center point as the representative point may introduce errors in velocity deambiguation, affecting the accuracy of phase compensation and, consequently, the imaging results. Therefore, the strategy adopted in this paper is to find an optimal representative point located near the midpoint of the line connecting the target's peak point and the target's RDM geometric center. This point is used as the representative point.

[0113] The implementation process of the function of the present invention is as follows Figure 4 The specific steps are as follows:

[0114] Step 1: Perform constant false alarm detection on the target range-Doppler (RDM) matrix to obtain the range index, Doppler index, and power amplitude of the target point;

[0115] Step 2: Find the best representative point as the velocity representative point of the target, which is located near the midpoint of the line connecting the peak point of the target point cloud image and the geometric center point of the target RDM;

[0116] Step 3: Use the hypothetical phase compensation velocity defuzzification algorithm for the representative point of the target to obtain the velocity and phase compensation values ​​corresponding to the representative point;

[0117] Step 4: The speed of all target points is the speed of the representative point, and all target points are compensated with the phase compensation value calculated based on the representative point;

[0118] Step 5: Based on the obtained target point velocity and azimuth angle values, coordinate transformation is performed to obtain the true position of the target in the three-dimensional coordinate system. The representative point velocity represents the true velocity of the target.

[0119] The step 1 is specifically as follows:

[0120] First, the radar raw sampling data is processed by two-dimensional fast Fourier transform (2D-FFT) to obtain the target RDM matrix;

[0121] The two-dimensional FFT algorithm calculates the time delay of the echo through the spectrum of the difference frequency signal, thereby determining the target distance. The transmitted chirp signal is expressed by the formula;

[0122] S TX =Aexp(j×(2πf c t+πkt 2 )), t∈(0,T c )

[0123] Where B is the signal bandwidth, f c is the carrier frequency, T c is the period, A is the amplitude, is the slope of frequency versus time;

[0124] The difference frequency signal generated by mixing the target reflected signal with the reference signal is expressed by the formula:

[0125] S IF =Aexp(j×(2π·kτ·t+2π×f c τ-πkt 2 ))

[0126] in is the round-trip delay of the electromagnetic wave, c is the speed of light, and R represents the distance between the target and the radar. It can be seen that the frequency of the difference frequency signal is linearly related to the time delay τ. Therefore, an FFT is performed on the difference frequency signal to obtain its signal frequency and calculate the distance R of the target. The formula is:

[0127]

[0128] Since the intermediate frequency signal needs to be sampled in the actual processing, it is processed in the digital domain and the number of sampling points is N r , the sampling rate is F s Under the sampling condition of , the distance resolution R of the distance information obtained after digital signal processing is res , expressed as:

[0129]

[0130] The radar estimates the target speed by calculating the Doppler frequency shift between the difference frequency signals of at least two echoes. For the same moving target, due to the modulation period T c The difference frequency is very small, so the change in the target distance detected by the two adjacent difference frequency signals can be ignored. By performing FFT on the two adjacent difference frequency signals separately, different phase information can be obtained. The one-dimensional FFT peaks of the two difference frequency signals are the same, but the phases are different. The speed information of the target can be obtained based on this phase difference.

[0131] According to the above principle, the process of using the two-dimensional FFT algorithm to process a frame of target difference frequency signal is expressed as follows: The two-dimensional FFT algorithm is as follows: Figure 8 As shown;

[0132] First, the echo signals of multiple chirps in each frame are sampled and the sampling results are stored in the form of a matrix. Specifically, the sampling results of each chirp signal are stored in the matrix row by row;

[0133] Then, each row and column of the matrix is ​​transformed by FFT to obtain the range-Doppler spectrum. By analyzing the range-Doppler spectrum (RDM), the range and speed of the target are estimated.

[0134] Next, the target RDM is subjected to constant false alarm detection using the ordered statistics (OS-CFAR) detection algorithm;

[0135] like Figure 9 As shown, the first step of the algorithm is to detect the signal located at the unit under test. In order to prevent the energy of the target signal from leaking into the reference unit and affecting the accuracy of the detection result, it is necessary to set up several adjacent units as protection units, located in front and behind the target unit. Then, n training units in front and behind the protection unit are selected as the input of the CFAR algorithm. According to the calculation method of the threshold value designed by the CFAR processor, the design of the CFAR processor can be classified into different categories. Commonly used methods for calculating the threshold value include mean calculation, size sorting, etc., to obtain the distance, speed and phase information of the target point;

[0136] The target's velocity and phase information serve as input for subsequent velocity deambiguation. The range index corresponds to the target's range, and the Doppler index corresponds to the target's velocity. The target's velocity can be calculated using the target's Doppler shift. The target's power amplitude represents the target signal's echo strength.

[0137] The step 2 is specifically as follows:

[0138] Selecting the peak point as the representative point can reflect the target's highest energy point, but it may not reflect the target's exact location. Selecting the target RDM geometric center point is closer to the target's true geometric center point, but this point may also introduce errors because it is located in the radar sidelobe area.

[0139] Through the representative point selection strategy, the advantages of these two methods are combined;

[0140] First, find the point with the highest energy in the target point cloud, i.e., the peak point, and connect it to the midpoint of the target RDM geometric center. The target RDM geometric center is obtained by performing a 2D FFT on the RDM. After object detection, extract the target area and calculate the geometric center. Then, traverse all target points and select the point in the RDM closest to this midpoint as the target's representative point.

[0141] The geometric center point of the RDM is obtained by RDM calculation, and then the optimal point is found near the midpoint between the geometric center point of the RDM and the peak point (this optimal point belongs to the RDM and can be considered as one of the points of the RDM).

[0142] The step 3 is specifically as follows:

[0143] The velocity defuzzification algorithm of the target's velocity representative point is adopted as follows:

[0144] First, the detection phase shift of the CFAR detection output of the target point is Enumerate all possible phase values ​​that may need to be compensated:

[0145]

[0146] Secondly, according to formula (1), the radar data is compensated with each hypothetical phase compensation value in turn, and the compensated data is expressed as

[0147]

[0148] Among them, S i Represents the signal data in the i-th channel, i starts from 2, because only non-initial codes require Doppler phase compensation, Indicates the phase value that needs to be compensated. At the same time, extract data from the same index position in each channel RDM to form a channel data set as follows:

[0149]

[0150] The value of N depends on the number of transmitting antennas M TX To calculate, N satisfies:

[0151] |N|<=(M ТХ -1) / 2 (4)

[0152] When the number of transmitting antennas is odd;

[0153] |N|<=(M TX / 2)-1 (5)

[0154] When the number of transmitting antennas is even;

[0155] Use formulas (4) and (5) to get the velocity blur period N, where M TX is the number of transmitting antennas;

[0156] Then, according to the N value, all HPC situations are listed, and the radar data are compensated in turn with each hypothetical phase compensation value. H satisfies formula (2). Then, FFT operation is performed on each compensated data (2N+1 in total) to obtain the angular power spectrum under each hypothetical condition;

[0157] H=H -N ,...,H k ,...,H N (6)

[0158] By selecting the peak point in the angular power spectrum, the true speed of each speed representative point is determined, and the speed corresponding to the speed representative point is the true speed of the entire target.

[0159] This application derives the velocity blur period N, where M TX is the number of transmitting antennas. Using a 12-transmit and 16-receive antenna array, the velocity ambiguity period is 5;

[0160] Then, according to the N value, all the assumed phase compensation situations are listed, and the radar data is compensated in turn with each assumed phase compensation value. Then, FFT operation is performed on each compensated data to obtain the angular power spectrum under each assumed condition. The assumption of the peak point in the angular power spectrum is selected, and its corresponding speed and the obtained angle are the true values ​​of the representative point.

[0161] The step 4 is specifically as follows:

[0162] Calibrating the phase variation caused by target velocity through velocity deambiguation is crucial for radar imaging accuracy. After the RDM containing target velocity and distance information is processed by CFAR detection, an estimated value of the target velocity and Doppler phase difference can be obtained.

[0163] However, in actual applications, the speed of moving targets often exceeds the maximum detectable speed of the radar. TX =2,N RX =4) as an example, a virtual antenna array with 1 transmitter and 8 receivers can be formed. Assume that the result sequence passing through the CFAR target point is X (m , n) (0 <m<N TX ,0 <n<N RX ), then the phase compensation satisfies the formula:

[0164]

[0165] When using angular FFT for DOA estimation, the signal energy in each direction is converted to the frequency domain, so different angles are mapped to different positions in the frequency domain. When using FFT to process signals, the directional information of the target signal is "mapped" to the frequency domain, and the direction of arrival of the signal is determined by calculating the energy spectrum at different angles.

[0166] The formula is:

[0167]

[0168] Where M represents the number of receiving antennas; this application is based on an antenna array with 12 transmit and 16 receive antennas, so M is 16.

[0169] The FFT results in an angular spectrum P(θ), which represents the signal power at different angles. The target's angle of arrival is estimated by finding the peak position of the power spectrum. The angle corresponding to the peak in the calculated angular spectrum P(θ) is the target's angle of arrival. By analyzing these peaks, the arrival angles of multiple target sources can be estimated.

[0170] The step 5 is specifically as follows:

[0171] The improved hypothetical phase compensation velocity ambiguity resolution algorithm model is as follows: Figure 2 As shown in the figure, the algorithm input is the distance index, Doppler index and power amplitude of the target point after CFAR detection. Based on the representative point selection method of the target RDM geometric center point and peak point, the representative points of the three targets are obtained, which are respectively recorded as: representative point 1, representative point 2 and representative point 3. For each representative point, RDM provides its velocity information V cfar1 , V cfar2 , V cfar3 , and phase information The HPC velocity deambiguation algorithm first derives a hypothetical velocity ambiguity period N based on the number of radar transmit antennas. Based on this period and known velocity information, all possible hypothetical phase compensation (HPC) scenarios are enumerated for each representative point. Each hypothetical phase compensation value is then applied to the target point information, calculating the angular power spectrum under various hypothetical conditions. The true velocity of each representative point is determined by selecting the peak hypothesis in the angular power spectrum. The corresponding velocity of the representative point is the true velocity of the entire target. Based on the velocity ambiguity period corresponding to the peak point (N1, N2, and N3, respectively), the phase compensation value for each target is calculated. Phase compensation is then applied to all target points according to their respective deambiguation periods. Direction of Arrival (DOA) estimation (such as angular FFT or DBF) is then performed to accurately obtain the true azimuth angle of each target. Finally, the true target range is calculated based on the relationship between the target point range index and the radar range resolution. The target range and angle information are then transformed to generate a point cloud image of the target.

[0172] The measured effect of the present invention:

[0173] In order to verify the velocity deambiguation method suitable for 4D millimeter-wave radar imaging proposed in this invention, the accuracy and time complexity of velocity deambiguation and the final stereoscopic imaging effect are evaluated through experimental tests.

[0174] Figure 5 For the experimental test scenario, consider a small car with dimensions of 4.0m*1.6m*1.4m, a speed of approximately 27km / h, and a lane width of 3.75m. Assume that the radar is located to the left rear of the target and detects the target.

[0175] Figure 6 To test the renderings, the experimental analysis used point cloud image size, position, and color as evaluation indicators. Point cloud color was used to represent target speed, and point cloud position was used to represent target angle. Figure 6 (a) shows the point cloud image of the car when the velocity ambiguity is not resolved. As can be seen from the point cloud image, the speed corresponding to the color of the car point cloud is different from the actual speed of the car. In addition, there is a significant deviation between the point cloud position of the target and the actual position (blue box). Figure 6 (b) and (c) show the results of velocity deblurring tests using the original HPC algorithm. As can be seen from the point cloud image, after deblurring, the velocities of most target points are close to the target's true velocity. However, for a small number of points, deviations in the velocity estimation lead to errors in the target angle estimation, resulting in significant point cloud dispersion in the final point cloud image. Figure 6 (d) (e) are the target space stereo point cloud images obtained by using the velocity defuzzification method proposed in this invention. Figure 6 From the point cloud images (d) and (e), we can see that the speed corresponding to the color of the car's point cloud image is approximately -27 km / h, which is basically consistent with the car's actual speed. In addition, the point cloud image of car 1 is located in the interval [0,3.75], which is the same as the actual position of car 1. Figure 6 (e) is a top view of the three-dimensional point cloud of car 1. It can be seen from the figure that the length of the point cloud image is about 4m and the width is about 1.6m. The overall size of the point cloud image is close to the actual size of car 1. Figure 6 (d) It can be seen that the height of the point cloud image is about 1.6m, which is slightly higher than the actual height of the car of 1.4m. This is because the target expands in the angular spectrum. Because the target angular spectrum has more than just the main lobe, when the side lobes are relatively high, the target angular spectrum will spread, causing the target point cloud volume to be larger than the actual volume.

[0176] Figure 7The following is a time complexity comparison chart of the methods. Due to the low range and speed resolution, common traffic radars usually identify a small car as about 10 target points. In contrast, in MIMO radar imaging systems, due to the improved range, speed, and angle resolution and the shorter imaging distance, the size of the target can be presented as much as possible. A small car will be identified as about 400 target points. Setting the number of target points from 10 to 400 increases, such as Figure 5 The relationship between the number of target points and the algorithm's time complexity is intuitively represented. The time complexity of the HPC algorithm increases with the number of target points. Compared to the original HPC algorithm, whose time complexity is more sensitive to the number of target points, the improved HPC algorithm has significantly lower time complexity and exhibits a more gradual upward trend with increasing target points, demonstrating strong robustness. For MIMO radar imaging systems, high spatial resolution is required to achieve stereoscopic imaging of moving targets. This results in a single target being resolved into multiple points, increasing processing complexity. When processing high-resolution MIMO radar data using the original HPC algorithm, its time complexity increases rapidly with the number of target points. This results in low algorithm efficiency and excessive processing time in practical applications, especially when the number of target points is large. In contrast, the improved HPC algorithm can more effectively maintain algorithm performance and efficiency when processing high-resolution MIMO radar data. This is now suitable for the needs of stereoscopic imaging.

[0177] The key point of the present invention is to design a velocity deambiguation method suitable for 4D millimeter-wave radar, so that the 4D millimeter-wave radar has the ability to detect high-speed targets and optimize the point cloud imaging effect.

[0178] This invention is applicable to intelligent transportation systems. Traditional velocity deambiguation algorithms require multiple frames of data and suffer from poor real-time performance, failing to meet the requirements of intelligent transportation systems. However, the proposed method can enhance the spatial stereoscopic imaging capabilities of millimeter-wave radar for moving traffic targets, providing intelligent transportation systems with richer and more intuitive traffic target information.

[0179] The present invention is applicable to radar imaging systems. Radar imaging systems typically have high spatial resolution, which reduces the maximum unambiguous velocity range they can detect. The method proposed in this invention is applicable to radar imaging systems and can effectively reduce the time complexity of the velocity deambiguation algorithm.

Claims

1. A velocity deambiguation method suitable for 4D millimeter-wave radar imaging, characterized in that: The following steps are included: Step 1: Perform two-dimensional fast Fourier transform on the original target sampling data of the 4D millimeter-wave radar to obtain the target range-Doppler matrix; And through constant false alarm detection, the target point distance, target point speed and phase information are obtained; Step 2: Find the best representative point as the velocity representative point of the target, which is located near the midpoint of the line connecting the peak point of the target point cloud image and the geometric center point of the target RDM; Step 3: Use the hypothetical phase compensation velocity defuzzification algorithm on the velocity representative point of the target to obtain the velocity and phase compensation values ​​corresponding to the velocity representative point; Step 4: Select the velocity ambiguity period corresponding to the peak point of the angular power spectrum of the target point under the assumed conditions, and perform DOA estimation on the phase compensation value calculated based on the representative point to obtain the azimuth angle value; Step 5: Based on the obtained target point speed and azimuth angle values, the target's true distance is calculated according to the relationship between the target point distance index and the radar range resolution. The target distance and angle information are converted to obtain the target's true position in the three-dimensional coordinate system. The speed of the speed representative point represents the target's true speed. The step 5 is specifically as follows: The target point distance, velocity and phase information are used as the input of the improved hypothetical phase compensation velocity ambiguity resolution algorithm model. Based on the representative point selection method of the target RDM geometric center point and peak point, the representative points of the three targets are obtained, which are respectively recorded as: representative point 1, representative point 2 and representative point 3. For each representative point, RDM provides its velocity information V cfar1 , V cfar2 , V cfar3 , and phase information The velocity ambiguity resolution algorithm first derives the assumed velocity ambiguity period N based on the number of radar transmitting antennas. Then, based on this period and the known velocity information, all possible assumed phase compensation scenarios are enumerated for each representative point. Next, each hypothetical phase compensation value is applied to the target point information to calculate the angular power spectrum under various hypothetical conditions. By selecting the peak point hypothesis in the angular power spectrum, the true speed of each representative point is determined. The corresponding speed of the representative point is the true speed of the entire target. Based on the velocity ambiguity periods N1, N2, and N3 corresponding to the peak points, the phase compensation value of each target is calculated. Phase compensation is performed on all target points according to their respective deambiguation periods, and then DOA estimation is performed to accurately obtain the true azimuth angle value of each target. Finally, the target's true distance is calculated based on the relationship between the target point distance index and the radar range resolution, and coordinate transformation is performed to form a point cloud image with target distance and angle information; Phase information Perform azimuth-elevation FFT to obtain the target's AEM matrix; Afterwards, the AEM is subjected to a constant false alarm detection method based on the normalized zero-point angle power spectrum to obtain the target azimuth and pitch angle information; Finally, the target's distance, azimuth and pitch angle information are transformed into coordinates to generate a three-dimensional point cloud image of the target space.

2. The velocity deambiguation method for 4D millimeter-wave radar imaging according to claim 1, characterized in that: The step 1 is specifically as follows: Step (1): perform ADC sampling on the difference frequency signal generated by mixing the echo signals of multiple chirps and the local oscillator signal in each frame of the original target sampling data, and store the sampling results in the form of a matrix, and store each chirp signal sampling result in the matrix row by row; Step (2): performing FFT transformation on each row and each column of the matrix to obtain a target range-Doppler spectrum containing target speed and range information; Step (3): Perform constant false alarm detection on the target RDM; obtain the target point distance, target point speed and phase information.

3. The velocity deambiguation method for 4D millimeter-wave radar imaging according to claim 2, characterized in that: The step (2) is specifically as follows: The echo delay is calculated by the spectrum of the difference frequency signal to determine the target distance. The transmitted Chirp signal is expressed by the following formula: S TX =Aexp(j×(2πf c t+πkt 2 )),t∈(0,T c ) Among them, f c is the carrier frequency, T c is the period, A is the amplitude, is the slope of frequency versus time, B is the signal bandwidth, and t is the time coordinate of the signal frequency changing with time within one cycle; The difference frequency signal is generated by mixing the target reflected signal with the reference signal, and is expressed as follows: S IF =Aexp(j×(2π·kτ·t+2π·f c t-πkt 2 )) in is the round-trip delay of the electromagnetic wave, c is the speed of light, R represents the distance between the target and the radar, the frequency of the difference frequency signal is linearly related to the time delay τ, and an FFT is performed on the difference frequency signal to obtain the signal frequency of the difference frequency signal, and the distance R of the target is calculated. The formula is: In digital domain processing, the number of sampling points is N r , the sampling rate is F s Under the sampling condition of , the distance resolution R of the distance information obtained after digital signal processing is res , expressed as: For the same moving target, perform FFT on two adjacent difference frequency signals separately to obtain different phase information. The one-dimensional FFT peaks of the two difference frequency signals are the same, but the phases are different. The speed information of the target can be obtained based on this phase difference. The process of applying the two-dimensional FFT algorithm to process a frame of target difference frequency signal is expressed as: First, the echo signals of multiple chirps in each frame are sampled and stored in the form of a matrix. The sampling result of each chirp signal is stored in the matrix row by row. Then, each row and column of the matrix is ​​transformed by FFT to obtain the range-Doppler spectrum. By analyzing the range-Doppler spectrum RDM, the range and speed of the target are estimated.

4. The velocity deambiguation method for 4D millimeter-wave radar imaging according to claim 2, characterized in that: The step (3) is specifically as follows: 1) firstly, the threshold of the distance dimension of the measured unit in the RDM matrix is ​​solved. According to the principle of CACFAR algorithm, the detection threshold of the measured unit in the distance dimension is: Among them, M is the number of training units, rx i ,ry i , are the training units on the left and right of the tested unit, respectively, and K is the threshold factor; 2) Then, the threshold of the unit under test in the velocity dimension of the RDM matrix is ​​solved in the same way as the distance dimension. The threshold of the unit under test in the velocity dimension is dx i , dy are the training units above and below the tested unit respectively; 3) After calculating the distance dimension and speed dimension thresholds, the tested unit is tested for crossing the threshold. The crossing threshold test can be expressed as (A>T r )&(A>T d ), A is the power of the unit under test in the RDM. Only when it exceeds the distance dimension and speed dimension threshold at the same time, it is identified as a target; After CFAR, the distance, velocity and phase information of the target point are obtained; The velocity and phase information of the target point serve as the input for subsequent velocity deambiguation. The target point distance corresponds to the distance index, and the target point velocity corresponds to the Doppler index. The target velocity is calculated by the Doppler frequency shift of the target. The phase information and the echo intensity of the target signal are the power amplitude of the target point.

5. The velocity deambiguation method for 4D millimeter-wave radar imaging according to claim 1, characterized in that: The step 2 is specifically as follows: First, find the point with the strongest energy in the point cloud image of the target, that is, the peak point, and the midpoint of the line connecting the geometric center point of the target RDM. The geometric center point of the target RDM is obtained by 2D FFT, and then the target area is extracted after target detection, and the geometric center point is calculated. Then, all target points are traversed and the point closest to this midpoint in the RDM is selected as the representative point of the target, and this representative point is used as the speed representative point; the geometric center point of the RDM is obtained by RDM calculation, and then the optimal point is found near the midpoint between the geometric center point of the RDM and the peak point.

6. The velocity deambiguation method for 4D millimeter-wave radar imaging according to claim 1, characterized in that: In step 3, the velocity defuzzification algorithm using the assumed phase compensation is applied to the velocity representative point of the target. The method is as follows: First, the detection phase shift of the CFAR detection output of the target point is Enumerate all possible phase values ​​that may need to be compensated: Secondly, according to formula (1), the radar data is compensated with each hypothetical phase compensation value in turn, and the compensated data is expressed as: Among them, S i Represents the signal data in the i-th channel, i starts from 2, Indicates the phase value that needs to be compensated. At the same time, data is extracted from the same index position in each channel RDM to form a channel data set as follows: The value of N depends on the number of transmitting antennas M TX To calculate, N satisfies: ∣N∣<=(M ТX -1) / 2 (4) When the number of transmitting antennas is odd; ∣N∣<=(M TX / 2)-1 (5) When the number of transmitting antennas is even; Use formulas (4) and (5) to get the velocity blur period N, where M TX is the number of transmitting antennas; Then, according to the N value, all HPC situations are listed, and the radar data are compensated in turn with each hypothetical phase compensation value. H satisfies formula (2). Then, FFT operation is performed on each compensated data (2N+1 in total) to obtain the angular power spectrum under each hypothetical condition; H=H -N ,...,H k ,...,H N (6) By selecting the peak point in the angular power spectrum, the true speed of each speed representative point is determined, and the speed corresponding to the speed representative point is the true speed of the entire target.

7. The velocity deambiguation method for 4D millimeter-wave radar imaging according to claim 1, characterized in that: The step 4 is specifically as follows: When the RDM containing the target speed and distance information is processed by CFAR detection, the estimated value of the target speed and the Doppler phase difference are obtained; Set the result sequence passing through the CFAR target point to X (m，n) , 0 <m<N TX ,0 <n<N RX , then the phase compensation satisfies the formula: 0<m<N TX ,0<n<N RX ; Where X(m,n) is the original signal matrix, is the phase factor, m is the transmit antenna index, n is the receive antenna index, N TX is the number of transmitting antennas, N RX is the number of receiving antennas; Use angular FFT to estimate DOA. When using FFT to process signals, the direction information of the target signal is "mapped" to the frequency domain, and the direction of arrival of the signal is determined by calculating the energy spectrum at different angles. The formula is: Where M represents the number of receiving antennas; The result of FFT processing is an angle spectrum P(θ). The arrival angle of the target is estimated by finding the peak position of the power spectrum. In the calculated angle spectrum P(θ), the angle corresponding to the peak is the arrival angle of the target.

8. Application of the velocity deambiguation method for 4D millimeter-wave radar imaging according to any one of claims 1 to 7, characterized in that: The method is applied to intelligent transportation systems and radar imaging systems.

Images