Millimeter wave radar DML algorithm realization method and system easy for engineering realization

Abstract

The invention discloses a millimeter wave radar DML algorithm realization method and system easy for engineering realization, and the method comprises the steps: firstly obtaining detection unit channel data and a steering vector matrix, and calculating digital beam forming spectrums of all search angles through the steering vector matrix; k angles with the strongest energy are screened out from a DBF power spectrum to serve as the basis of a preliminary search interval. Performing coarse search on each preliminary search interval to determine a candidate target angle, and if the candidate angle exceeds a radar view field, outputting a single target angle; and otherwise, carrying out fine search to refine the target angle. And calculating a double-target judgment coefficient, and finally outputting a single-target or double-target angle according to a judgment result. According to the method, the double-stage search strategy based on the DBF spectrum result is introduced, complex matrix inversion operation is avoided, the calculation complexity is reduced, the overall search time is shortened by reasonably setting the search range and the step value, and the real-time processing capacity is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of millimeter wave radars, and in particular relates to a method and system for realizing a millimeter wave radar DML algorithm that is easy to realize in engineering. Background Art

[0002] In modern radar systems, super-resolution angle measurement algorithms are crucial to improving target recognition accuracy and tracking capabilities. Common super-resolution angle measurement algorithms include CAPON, MUSIC, ESPRIT, and DML. These algorithms each have different advantages and disadvantages. Among them, the DML (Deterministic Maximum Likelihood) algorithm has received widespread attention in practical applications due to its low requirements on the number of snapshots and its ability to maintain good resolution under low signal-to-noise ratio conditions.

[0003] However, the traditional DML algorithm has some significant technical challenges in the engineering implementation process. First, the DML algorithm requires complex matrix inversion operations, which not only increases the computational complexity, but also places high demands on hardware resources, limiting its application in low-cost or resource-constrained environments. Secondly, in order to ensure sufficient angle search accuracy, the traditional DML algorithm usually needs to perform a comprehensive search process, which is time-consuming and is an important bottleneck for application scenarios with high real-time requirements. In addition, when there are multiple coherent targets in the scene, how to accurately and efficiently identify these targets is also a problem that needs to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for implementing a millimeter-wave radar DML algorithm that is easy to implement in engineering, which greatly reduces the time consumption of the algorithm, determines the number of targets based on prior information and statistical radar angle measurement characteristics, and achieves super-resolution to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides a method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering, comprising the following steps:

[0006] Get the current detection unit channel data X r and the steering vector matrix W;

[0007] Calculating digital beamforming DBF spectra of all search angles according to the steering vector matrix, and saving DBF complex spectrum and DBF power spectrum;

[0008] From the DBF power spectrum, K angles with the strongest energy are selected by sorting them from large to small according to the spectrum peak intensity, and a preliminary search interval is determined based on these angles;

[0009] For each preliminary search interval, a coarse search is performed to determine candidate target angles;

[0010] Determine whether the candidate target angle is within the radar field of view FOV. If not, end the process and output the single target angle; if yes, perform a precise search to refine the target angle.

[0011] Based on the refined search results and pre-trained weight coefficients, the dual-target decision coefficient is calculated, and the single-target or dual-target angle is output according to the decision result.

[0012] Preferably, the calculation formula for calculating the digital beamforming DBF spectrum of all search angles according to the steering vector matrix is:

[0013] Yr(θ)=s(θ) H *Xr

[0014]

[0015] Where Y r is the complex spectrum, Yr(θ) is the DBF complex value corresponding to the search angle θ, s(θ) is the steering vector corresponding to the search angle θ, H is the matrix transpose conjugate, X r is the channel data of the current detection unit, P r is the power spectrum, Pr(θ) is the DBF power corresponding to the search angle θ, and N is the number of channels.

[0016] Preferably, for each preliminary search interval, performing a coarse search to determine a candidate target angle comprises:

[0017] Set the coarse search angle step δ°, δ<BeamWidth0, and is an integer;

[0018] Traverse and select any different interval combinations and angle combinations of the K angles with the strongest energy, calculate the DML spectra of all angle combinations, and search for the angle with the strongest power (A1, B1):

[0019] R xx =Xr*Xr H

[0020]

[0021] N=s(θ) H *s(θ)

[0022] M=s(φ) H *s(θ)

[0023] Yr(θ)=s(θ) H *Xr

[0024] Yr(φ)=s(φ)H *Xr

[0025]

[0026] P A (θ,φ)=w(θ,φ)*(w(θ,φ) H *w(θ,φ)) -1 *w(θ,φ) H

[0027] tr(s(θ)*Xr H )=Xr H *s(θ)=s H *Xr H =Yr(θ)

[0028]

[0029] Where T is the matrix transpose; H is the matrix transpose conjugate; θ, φ are the search angles; s(θ) is the steering vector corresponding to the search angle θ; Yr(θ) is the DBF complex value corresponding to the search angle θ; Pr(θ) is the DBF power corresponding to the search angle θ; P A (θ, φ) is the energy projection matrix corresponding to the search angles θ, φ; A, B are the two target angles obtained by DML search; λ is the radar operating wavelength; dk ​​is the array element spacing of the kth channel; N is the number of channels; s(φ) is the steering vector corresponding to the search angle φ; Yr(φ is the DBF complex value corresponding to the search angle φ; Pr(φ) is the DBF power corresponding to the search angle φ; tr is the trace of the matrix.

[0030] Preferably, the determining whether the candidate target angle is within the radar field of view FOV, if not, ending the process and outputting the single target angle, comprises:

[0031] Determine whether the angles (A1, B1) are both within the radar FOV at the current distance. If they exceed the radar detection capability, it can be determined that there are no dual targets at present, thereby terminating the DML angle measurement process in advance.

[0032] Preferably, the determining whether the candidate target angle is within the radar field of view FOV, and if so, performing a precise search to refine the target angle, comprises:

[0033] Calculate the fine search interval [Search3, Search4] = [A1-0.5*α*BeamWidth θ ,A1+0.5*α*BeamWidth θ ],

[0034] [Search 5,Search 6]=[B1-0.5*α*BeamWidthθ ,B1+0.5*α*BeamWidth θ ], BeamWidth0 is the array 3dB beam width (0°);

[0035] In the fine search areas [Search3, Search4] and [Search5, Search6], the DML spectra of all angle combinations are calculated again to search for the angle (A2, B2) with the strongest power.

[0036] Preferably, the step of calculating the dual-objective decision coefficient based on the refined search results and the pre-trained weight coefficients includes:

[0037] According to the feature vector U of the angle (A2, B2) and the pre-trained weight coefficient W, the dual-target decision coefficient γ = U is calculated T *W;

[0038] According to the judgment result, single target output or dual target output is selected. If γ<=0, it is judged as a single target, and the angle corresponding to the strongest power is still output as the target angle according to the result of the DBF spectrum; if γ>0, it is judged as a dual target, then (A2, B2) is output as the target angle, and the DML angle measurement process ends.

[0039] On the other hand, the present invention proposes an implementation system of a millimeter-wave radar DML algorithm that is easy to implement in engineering. The system includes: a data acquisition module for acquiring radar echo data; a processing module for executing all calculation steps in the method; a storage module for storing data and parameters required for calculation; and an output module for outputting final target angle information.

[0040] Preferably, the processing module adopts a two-stage search mechanism, firstly performing a rough search based on the DBF spectrum, and then performing a fine search based on the result of the rough search.

[0041] Technical effects and advantages of the present invention: The present invention proposes a method and system for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering, which has the following advantages over the prior art:

[0042] The present invention introduces a two-stage search strategy based on digital beamforming (DBF) spectrum results, avoiding the direct complex matrix inversion operation, thereby greatly reducing the computational complexity. At the same time, by reasonably setting the range and step value of coarse search and fine search, the overall search time is effectively reduced, making the algorithm more suitable for real-time processing scenarios. In addition, statistical features and pre-trained weight coefficients are combined to assist multi-target judgment, further improving the accuracy of target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1The present invention is a flow chart of a method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering. DETAILED DESCRIPTION

[0044] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0045] This embodiment provides an implementation method of the millimeter wave radar DML algorithm that is easy to implement in engineering, aiming to solve the two main problems of the traditional DML algorithm in engineering implementation: one is that the computational complexity is too high, and the other is that the search time is too long. These two problems directly affect the actual application effect and efficiency of the algorithm. Especially in an environment with limited resources, how to reduce the computational cost and shorten the processing time while ensuring the angle measurement accuracy has become a key technical problem.

[0046] Before describing this plan in detail, it is necessary to explain:

[0047] The DML algorithm is a deterministic likelihood algorithm that requires a maximum number of detectable targets to be set in advance. In this embodiment, it is set to 2. Some formula derivations are listed below.

[0048]

[0049] (1) Search for the beam width corresponding to the angle θ

[0050] (2) Echo data Xr = [x0, x1, x2, ..., x n-1 ] T

[0051] (3) Echo data covariance matrix R xx =Xr*Xr H

[0052] (4) DML Steering Vector:

[0053]

[0054] (5)N=s(θ) H *s(θ)

[0055] (6)

[0056] (7)Yr(θ)=s(θ)H *Xr

[0057] (8)

[0058] (9)

[0059] (10)

[0060] (11)

[0061] (12)

[0062] (13)tr(s(θ)*Xr H )=Xr H *s(θ)=s H *Xr H =Yr(θ)

[0063] (14)

[0064] DML estimation principle formula After the formula is split and simplified, Yr(θ) and It can be obtained from the DBF spectrum, so the calculation results of DML can be directly obtained from the calculation results of DBF, without the need for matrix inversion calculation, reducing the complexity of the operation.

[0065] The DML spectrum search mode is changed to a two-level search. First, K angles are selected from the DBF spectrum (for example, the first K with the strongest energy are selected). The first-level search (coarse search) is performed on the adjacent areas of the K angles. The search angle step value must be smaller than the beam width of the array, otherwise it will affect the accuracy of the DML angle output. In addition, the size of the angle area can be determined according to the beam width widening formula to reduce unnecessary searches while maintaining accuracy as much as possible. If the angle and distance output by the first-level search exceed the radar detection range, there is no need to perform a second-level search (fine search), and the DML process is terminated to save time. If the second-level search conditions are met, the angle range of the second-level search is determined according to the results of the first-level search, and the second-level search step value is set to the DBF weight coefficient angle step value. The angle area can also be determined according to the beam width widening formula, and the search of the entire radar FOV is finally completed.

[0066] Since the above DML algorithm is based on a dual-target model, most of the actual scenes are single-target, so it is necessary to collect some features to make target decisions. Through some simulation and measured data, we found that: when other external conditions are the same, the two features of 2D-FFT angle measurement data channel consistency and DBF sidelobe level are inconsistent for dual targets and single targets. The following is a simple simulation explanation:

[0067] Table 1. Single target simulation parameters

[0068] Target Angle Number of channels Beam width signal to noise ratio 3.1° 12 4° 15dB

[0069] Table 2. Dual-objective simulation parameters

[0070] Target Angle strength Number of channels Beam width signal to noise ratio [3.1°,10.5°] 1:1 12 4° 15dB

[0071] Under the same simulation conditions, the performance of single target and dual target is inconsistent, so it can be used as a feature. In addition, the DBF spectrum energy and DML energy corresponding to the dual targets solved by DML, and the signal energy are all related to the judgment and can also be used as features. These are listed as feature vectors U. The following describes how to obtain the weight coefficient W.

[0072] Collect raw ADC data in different scenarios (including darkroom environment and outdoor environment, dual-target scene and single-target scene, including different individual radars, to maintain sample diversity).

[0073] Use MATLAB to calculate 2D-FFT and generate the detection matrix, and then detect the target through the constant false alarm (cfar) algorithm.

[0074] Calculate the DBF spectrum and DML spectrum of the target, save the target 2D-FFT angle measurement data of all samples, channel consistency, DBF sidelobe level, echo signal energy, 2 DBF spectrum peak energies corresponding to dual angles, and DML spectrum peak energies corresponding to dual angles.

[0075] Different labels are given to single targets and dual targets. Through the classification learner tool of Matlab, the linear SVM model is selected for parameter training, and then the training results are exported as the weight coefficient W.

[0076] After completing the above steps, the dual-objective decision coefficient γ in the DML algorithm is T *W, γ>0 is judged as a dual target, γ<=0 is judged as a single target.

[0077] Illustratively, the present invention provides Figure 1 The method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering includes the following steps:

[0078] Get the current detection unit channel data Xr And the steering vector matrix W.

[0079] Call the above formula (7) and formula (9) to calculate the DBF spectrum of all angles and save the DBF complex spectrum Y r and power spectrum P r .

[0080] Search the DBF power spectrum, sort it from large to small according to the spectrum peak intensity, take the first K angles, assume that the i-th angle is θ, and the corresponding beam width is BeamWidth θ , calculate the corresponding search angle interval [Search1, Search2] = [θ-0.5*α*BeamWidth θ ,θ+0.5*α*BeamWidth θ ], α is the expansion factor (the recommended value is between 1 and 2) to obtain the search interval corresponding to K angles.

[0081] Set the coarse search angle step δ° (δ<BeamWidth0, and is an integer), traverse any different interval combinations and angle combinations in step 3, call the above formula (3) to formula (4), calculate the DML spectra of all angle combinations, and search for the angle (A1, B1) where the strongest power is located.

[0082] Determine whether the angles (A1, B1) are within the radar FOV at the current distance. If they exceed the radar detection capability, it can be determined that there are no dual targets at present, thus terminating the DML angle measurement process in advance. Then directly search the DBF power spectrum P r , get the angle corresponding to the strongest power, that is, the target angle, and the steps are completed here.

[0083] According to the rough search angle (A1, B1) obtained in step 4, calculate the fine search interval:

[0084] [Search 3,Search 4]=[A1-0.5*α*BeamWidth θ ,A1+0.5*α*BeamWidth θ ];

[0085] [Search 5,Search 6]=[B1-0.5*α*BeamWidth θ ,B1+0.5*α*BeamWidth θ ];

[0086] In the fine search area [Search3, Search4] and [Search5, Search6], the above formulas (3) to (4) are called to calculate the DML spectra of all angle combinations, and the angle (A2, B2) with the strongest power is searched.

[0087] According to the characteristic vector U of the angle (A2, B2) (the channel consistency mentioned above, DBF sidelobe level, echo signal energy, 2 DBF spectrum peak energies corresponding to the dual angles, DML spectrum peak energy corresponding to the dual angles) and the pre-trained weight coefficient W, the dual-target decision coefficient γ = U is calculated. T *W, select single target output or dual target output according to the judgment result. If γ<=0, it is judged as a single target, and the angle corresponding to the strongest power is still output as the target angle according to the result of the DBF spectrum; if γ>0, it is judged as a dual target, then (A2, B2) is output as the target angle, and the DML angle measurement process ends.

[0088] On the other hand, the present invention proposes an implementation system of a millimeter-wave radar DML algorithm that is easy to implement in engineering, including: a data acquisition module for acquiring radar echo data; a processing module for executing all calculation steps in the method; a storage module for storing data and parameters required for calculation; and an output module for outputting final target angle information.

[0089] Furthermore, the processing module adopts a two-stage search mechanism, firstly performing a rough search based on the DBF spectrum, and then performing a fine search based on the result of the rough search.

[0090] The above-mentioned data acquisition module, processing module, storage module, and output module are also used to implement other steps of the above-mentioned method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering, as follows:

[0091] Get the current detection unit channel data X r and the steering vector matrix W;

[0092] Calculating digital beamforming DBF spectra of all search angles according to the steering vector matrix, and saving DBF complex spectrum and DBF power spectrum;

[0093] From the DBF power spectrum, K angles with the strongest energy are selected by sorting them from large to small according to the spectrum peak intensity, and a preliminary search interval is determined based on these angles;

[0094] For each preliminary search interval, a coarse search is performed to determine candidate target angles;

[0095] Determine whether the candidate target angle is within the radar field of view FOV. If not, end the process and output the single target angle; if yes, perform a precise search to refine the target angle.

[0096] Based on the refined search results and pre-trained weight coefficients, the dual-target decision coefficient is calculated, and the single-target or dual-target angle is output according to the decision result.

[0097] Preferably, the calculation formula for calculating the digital beamforming DBF spectrum of all search angles according to the steering vector matrix is:

[0098] Yr(θ)=s(θ) H *Xr

[0099]

[0100] Where Y r is the complex spectrum, Yr(θ) is the DBF complex value corresponding to the search angle θ, s(θ) is the steering vector corresponding to the search angle θ, H is the matrix transpose conjugate, X r is the channel data of the current detection unit, P r is the power spectrum, Pr(θ) is the DBF power corresponding to the search angle θ, and N is the number of channels.

[0101] Preferably, for each preliminary search interval, performing a coarse search to determine a candidate target angle comprises:

[0102] Set the coarse search angle step δ°, δ<BeamWidth0, and is an integer;

[0103] Traverse and select any different interval combinations and angle combinations of the K angles with the strongest energy, calculate the DML spectra of all angle combinations, and search for the angle with the strongest power (A1, B1):

[0104] R Xx =Xr*Xr H

[0105]

[0106] N=s(θ) H *s(θ)

[0107] M=s(φ) H *s(θ)

[0108] Yr(θ)=s(θ) H *Xr

[0109] Yr(φ)=s(φ) H *Xr

[0110]

[0111] P A(θ,φ)=w(θ,φ)*(w(θ,φ) H *w(θ,φ)) -1 *w(θ,φ) H

[0112] tr(s(θ)*Xr H )=Xr H *s(θ)=s H *Xr H =Yr(θ)

[0113]

[0114] Wherein, T is the matrix transpose; H is the matrix transpose conjugate; θ, φ are the search angles; s(θ) is the steering vector corresponding to the search angle θ; Yr(θ) is the DBF complex value corresponding to the search angle θ; Pr(θ) is the DBF power corresponding to the search angle θ; PA(θ, φ) is the energy projection matrix corresponding to the search angles θ, φ; A, B are the two target angles obtained by DML search; λ is the radar operating wavelength; dk ​​is the array element spacing of the kth channel; N is the number of channels; s(φ) is the steering vector corresponding to the search angle φ; Yr(φ) is the DBF complex value corresponding to the search angle φ; Pr(φ) is the DBF power corresponding to the search angle φ; tr is the trace of the matrix.

[0115] Preferably, the determining whether the candidate target angle is within the radar field of view FOV, if not, ending the process and outputting the single target angle, comprises:

[0116] Determine whether the angles (A1, B1) are both within the radar FOV at the current distance. If they exceed the radar detection capability, it can be determined that there are no dual targets at present, thereby terminating the DML angle measurement process in advance.

[0117] Preferably, the determining whether the candidate target angle is within the radar field of view FOV, and if so, performing a precise search to refine the target angle, comprises:

[0118] Calculate the fine search interval [Search 3, Searck 4] = [A1-0.5*α*BeamWidthθ, A1+0.5*α*BeamWidthθ],

[0119] [Search 5,Search 6]=[B1-0.5*α*BeamWidth θ ,B1+0.5*α*BeamWidth θ ], BeamWidth0 is the array 3dB beam width (0°);

[0120] In the fine search areas [Search 3, Search 4] and [Search 5, Search 6], the DML spectra of all angle combinations are calculated again to search for the angle (A2, B2) with the strongest power.

[0121] Preferably, the step of calculating the dual-objective decision coefficient based on the refined search results and the pre-trained weight coefficients includes:

[0122] According to the feature vector U of the angle (A2, B2) and the pre-trained weight coefficient W, the dual-target decision coefficient γ = U is calculated T *W;

[0123] According to the judgment result, single target output or dual target output is selected. If γ<=0, it is judged as a single target, and the angle corresponding to the strongest power is still output as the target angle according to the result of the DBF spectrum; if γ>0, it is judged as a dual target, then (A2, B2) is output as the target angle, and the DML angle measurement process ends.

[0124] In summary, the present invention introduces a two-stage search strategy based on digital beamforming (DBF) spectrum results, which avoids directly performing complex matrix inversion operations, thereby greatly reducing the computational complexity. At the same time, by reasonably setting the range and step value of coarse search and fine search, the overall search time is effectively reduced, making the algorithm more suitable for real-time processing scenarios. In addition, the combination of statistical features and pre-trained weight coefficients to assist multi-target judgment further improves the accuracy of target recognition.

[0125] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering, characterized in that: It includes the following steps: Get the current detection unit channel data X r and the steering vector matrix W; Calculate the digital beamforming (DBF) spectrum for all search angles according to the steering vector matrix, and save the DBF complex spectrum and the DBF power spectrum; Sort the DBF power spectrum from largest to smallest according to the peak intensity, and screen out K angles with the strongest energy. Based on these angles, determine the preliminary search range; For each preliminary search range, perform a coarse search to determine the candidate target angles; Judge whether the candidate target angles are within the radar field of view (FOV). If not, end the process and output the single target angle; If so, perform a fine search to refine the target angles; Based on the fine search results and the pre-trained weight coefficients, calculate the dual-target decision coefficient, and output the single target or dual-target angle according to the decision result.

2. The method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering according to claim 1, characterized in that: The formula for calculating the digital beamforming (DBF) spectrum for all search angles according to the steering vector matrix is: Yr(θ)=s(θ) H *Xr Where Y r is the complex spectrum, Yr(θ) is the DBF complex value corresponding to the search angle θ, s(θ) is the steering vector corresponding to the search angle θ, H is the matrix transpose conjugate, X r is the channel data of the current detection unit, P r is the power spectrum, Pr(θ) is the DBF power corresponding to the search angle θ, and N is the number of channels.

3. The method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering according to claim 2, characterized in that: For each preliminary search range, performing a coarse search to determine the candidate target angles includes: Set the coarse search angle step δ°, where δ < BeamWidth0 and is an integer; Traverse and screen out any different interval combinations and angle combinations of the K angles with the strongest energy, calculate the DML spectrum for all angle combinations, and search for the angle (A1, B1) where the strongest power is located: R xx =Xr*Xr H N=s(θ) H *s(θ) M=s(φ) H *s(θ) Yr(θ)=s(θ) H *Xr Yr(φ)=s(φ) H *Xr P A (θ,φ)=w(θ,φ)*(w(θ,φ) H *w(θ,φ)) -1 *w(θ,φ) H tr(s(θ)*Xr H )=Xr H *s(θ)=s H *Xr H =Yr(θ) Where T is the matrix transpose; H is the matrix transpose conjugate; θ, φ are the search angles; s(θ) is the steering vector corresponding to the search angle θ; Yr(θ) is the DBF complex value corresponding to the search angle θ; Pr(θ) is the DBF power corresponding to the search angle θ; P A (θ, φ) is the energy projection matrix corresponding to the search angles θ, φ; A, B are the two target angles obtained by DML search; λ is the radar operating wavelength; dk ​​is the array element spacing of the kth channel; N is the number of channels; s(φ) is the steering vector corresponding to the search angle φ; Yr(φ) is the DBF complex value corresponding to the search angle φ; Pr(φ) is the DBF power corresponding to the search angle φ; tr is the trace of the matrix.

4. The method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering according to claim 3, characterized in that: Judging whether the candidate target angles are within the radar field of view (FOV). If not, ending the process and outputting the single target angle includes: Judge whether the angles (A1, B1) are both within the radar FOV at the current distance. If it exceeds the radar detection ability, it can be determined that there is no dual target at present, and the DML angle measurement process can be ended in advance.

5. The method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering according to claim 3, characterized in that: Judging whether the candidate target angles are within the radar field of view (FOV). If so, performing a fine search to refine the target angles includes: Calculate the fine search interval [Search3, Search4] = [A1-0.5*α*BeamWidth θ ,A1+0.5*α*BeamWidth θ ], [Search5,Search6]=[B1-0.5*α*BeamWidth θ ,B1+0.5*α*BeamWidth θ ], BeamWidth0 is the array 3dB beam width (0°); Within the fine search regions [Search3, Search4] and [Search5, Search6], then calculate the DML spectrum for all angle combinations again, and search for the angle (A2, B2) where the strongest power is located.

6. The method for implementing a millimeter wave radar DML algorithm that is easy to implement in engineering according to claim 5, characterized in that: Based on the fine search results and the pre-trained weight coefficients, calculating the dual-target decision coefficient includes: According to the feature vector U of the angle (A2, B2) and the pre-trained weight coefficient W, the dual-target decision coefficient γ = U is calculated T *W; Select single target output or dual target output according to the decision result. If γ <= 0, it is judged as a single target, and still output the angle corresponding to the strongest power in the DBF spectrum as the target angle according to the result; if γ > 0, it is judged as a dual target, then output (A2, B2) as the target angle, and the DML angle measurement process ends.

7. A system for implementing a millimeter wave radar DML algorithm according to any one of claims 1 to 6, characterized in that the system It includes: A data acquisition module for acquiring radar echo data; A processing module for executing all calculation steps in the method; A storage module for storing the data and parameters required for calculation; An output module for outputting the final target angle information.

8. The system according to claim 7, characterized in that The processing module adopts a two-stage search mechanism. First, a coarse search is performed based on the DBF spectrum, and then a fine search is performed based on the results of the coarse search.

Images