Fast joint super-resolution estimation method for multi-parameters of millimeter-wave radar frequency domain beamforming

Abstract

The present invention relates to the technical field of millimeter wave radar signal processing for security, and specifically to a method for fast joint super-resolution estimation of multi-parameters of a millimeter wave radar in frequency domain beam, comprising: obtaining a mixed frequency beat signal of a security millimeter wave radar; transforming space-time multi-parameter domain joint data into a frequency domain by fast Fourier transform; selecting frequency domain data to realize data dimension reduction; constructing a covariance matrix and performing eigenvalue decomposition; obtaining a signal subspace or a noise subspace according to the number of information sources, and obtaining target parameter information. The present invention realizes fast joint estimation of multi-dimensional parameters of the security millimeter wave radar by utilizing frequency domain beam space dimension reduction, solves the problem of high computational complexity of a traditional multi-dimensional joint super-resolution algorithm, and provides technical support for the engineering realization of multi-dimensional parameter joint super-resolution of the security millimeter wave radar.

Description

Technical Field

[0001] The present invention relates to the technical field of millimeter wave radar signal processing for security, and in particular to a millimeter wave radar frequency domain beam multi-parameter fast joint super-resolution estimation method. Background Art

[0002] In the target parameter estimation algorithm of security millimeter-wave radar, the traditional fast Fourier transform algorithm can quickly realize the multi-dimensional parameter estimation of the target, but due to the existence of fence effect and spectrum leakage problems, the estimation accuracy of target information is low, the resolution is insufficient, and the stability is poor. Especially in terms of angle estimation, limited by the number of antennas, the resolution is extremely low in multi-target scenarios. In order to improve the accuracy and resolution of target parameter estimation in view of the bottleneck of the traditional FFT algorithm, super-resolution algorithms are gradually applied to the target parameter estimation of security millimeter-wave radar. Although super-resolution algorithms represented by multiple signal classification algorithms and rotationally invariant subspace algorithms have the improvement of target information estimation accuracy and resolution, the huge amount of data calculation seriously affects the process of engineering application. Similarly, the high efficiency and low complexity of the space-time multi-parameter joint super-resolution algorithm is a problem that needs to be solved urgently and is also a problem that cannot be circumvented in engineering.

[0003] At the same time, in the multi-target scene application of security millimeter-wave radar, parameter estimation usually includes angle measurement corresponding to the array element domain, distance and speed measurement corresponding to the fast and slow time domains, and the original sampled multi-parameter space-time joint data is very large, and the storage and calculation of the original data are difficult problems for engineering implementation. In engineering operations, the parameter area of ​​interest is generally determined based on the target information or data preprocessing results, and then super-resolution is performed. Therefore, it is of great research significance to study a frequency domain beam multi-parameter fast joint super-resolution estimation algorithm. Summary of the invention

[0004] In view of the high computational complexity of traditional super-resolution algorithms, the present invention proposes a method for fast joint super-resolution estimation of millimeter-wave radar frequency domain beam multi-parameters, which greatly reduces the dimension of beam domain data and makes joint super-resolution engineering possible, thereby realizing fast joint estimation of multiple parameters of security millimeter-wave radar.

[0005] The present invention is achieved by the following measures:

[0006] A millimeter wave radar frequency domain beam multi-parameter fast joint super-resolution estimation method, characterized by comprising the following steps:

[0007] Step 1: Obtain the mixed frequency beat signal of the security millimeter wave radar, specifically:

[0008] Step 1-1: Establish sawtooth wave transmission signal model s(t):

[0009]

[0010] Among them, A mpTs Indicates the corresponding amplitude of the transmitted signal; f c is the signal carrier frequency; μ=B / T m represents the frequency modulation slope, T m is the signal repetition period, B is the signal bandwidth;

[0011] Step 1-2: The receiving end obtains the received signal r(t),

[0012]

[0013] Among them, A mpRs is the corresponding amplitude of the received signal, τ = 2R(t) / c is the time delay between the received signal reaching the circulator and the transmitted signal, when the target has speed, R(t) = R0 + vt, R0 is the initial distance when stationary;

[0014] Step 1-3: The local oscillator signal in the mixer is mixed with the received signal to obtain a mixed beat signal. After ignoring its high order and extending its repetition period, the mixed beat signal S IF (t) is

[0015]

[0016] Where m = 0, 1...., M-1 represents the repetition cycle number;

[0017] Step 1-4: Use TDM-MIMO radar to expand the virtual aperture and set the number of transmitting antennas of TDM-MIMO radar to L TX , the number of receiving antenna elements is L RX , assuming that the receiving antenna array element spacing is d r =λ / 2, the distance between transmitting antenna elements is d t , and satisfies d t =L RX ×d r , assuming that an arbitrary receiving antenna l RX The distance between the first receiving antenna in the receiving antenna array is d rlRX =(l RX -1)d r , then the number of virtual array elements generated is L = L TX ×L RX The radar system structure can be compared with the array aperture (L TX L RX -1)×λ / 2 virtual uniform array equivalent, for the mixing beat signal model described in step 1-3, the TDM-MIMO LFMCW radar sawtooth beat discrete signal model X(n,l,m) is obtained:

[0018]

[0019] Among them l tx =1,2,…L TX , l rx =1,2,…L RX are the number of transmitting and receiving antennas respectively, l=0,1....,L-1 represents the virtual array receiving antenna element number, n is the sampling number in a single frequency modulated continuous wave cycle, f s is the sampling frequency, G is the additive Gaussian white noise;

[0020] Step 2: transform the space-time multi-parameter domain joint data into the frequency domain through fast Fourier transform;

[0021] Step 3: Select frequency domain data to achieve data dimensionality reduction;

[0022] Step 4: Construct the covariance matrix and perform eigenvalue decomposition;

[0023] Step 5: Obtain the signal subspace or noise subspace according to the number of information sources and obtain the target parameter information.

[0024] The step 2 of the present invention specifically includes:

[0025] Step 2-1: Arrange the TDM-MIMO LFMCW radar sawtooth wave beat discrete signal to obtain N×L×M dimensional three-dimensional data X 3dim , X 3dim Each N×L plane data is called a frame of data, then X 3dim There are M frames of data in total;

[0026] Step 2-2: The one-dimensional discrete Fourier transform formula is known to be

[0027]

[0028] The two-dimensional Fourier transform formula is

[0029]

[0030] X 3dimThe three-dimensional Fourier transform is operated by combining the one-dimensional Fourier transform with the two-dimensional Fourier transform to realize the transformation of the space-time multi-parameter domain joint data to the frequency domain: first, the two-dimensional Fourier transform is performed on the M-frame distance-angle data plane to complete the frequency domain transformation of the distance-angle dimension data, and then the data of each frame after the two-dimensional Fourier transform can be vectorized to obtain (N×L)×M two-dimensional data, and then the two-dimensional data is subjected to (N×L) times M-point Fourier transform to realize the speed dimension Fourier transform; after M times (N×L) two-dimensional Fourier transform and (N×L) times M-point one-dimensional Fourier transform, the converted frequency domain data Y NL×M for

[0031]

[0032] in, represents the Kronecker product.

[0033] The step three of the present invention specifically includes:

[0034] Step 3-1: In order to reduce the amount of data calculation, select the frequency domain data: NL×M For vectorization, we have

[0035]

[0036] The left side of the above formula is the vectorization of the 3D data after Fourier transformation, and the right side is the 3D beam space transformation expression of the original data. It can be explained that directly selecting 3D frequency domain data and 3D beam space dimensionality reduction are equivalent; Step 3-2: Directly selecting 3D frequency domain data and 3D beam space dimensionality reduction are equivalent. The process of directly selecting 3D frequency domain data can be realized through beam space transformation: Define the angle-dimensional beam transformation matrix Where L b Select the number of effective beams in the spatial domain, that is, the length of the data after dimensionality reduction, W θ The first line of the

[0037]

[0038] Where l,0≤l≤(L b -1),L b ≤L, when L b = L, that is W θ is the angle-dimensional full beam transformation matrix, and the distance-dimensional beam transformation matrix is ​​defined as N b Select the number of effective beams in the range dimension, and its nth line is:

[0039]

[0040] Where 0≤n≤(Nb -1),N b ≤N, when N b =N, that is, W R is the distance dimension full beam transformation matrix, and the velocity dimension beam transformation matrix is ​​defined as M b Select the number of effective beams in the velocity dimension, whose mth line is:

[0041]

[0042] Where m,0≤m≤(M b -1),M b ≤M, when M b =M, that is, W v is the velocity dimension full beam transformation matrix;

[0043] Step 3-3: Determine the selection of L according to the target parameter area b 、M b 、N b The corresponding area of ​​interest is selected in the frequency domain data by the linear correspondence relationship to obtain the three-dimensional frequency domain data after dimensionality reduction. The starting values ​​i, j, and k of the target data in the three dimensions are determined by the distribution area of ​​the target;

[0044] The three-dimensional frequency domain data after dimension reduction After rearrangement, we can get the data vector Z

[0045]

[0046] The step 4 of the present invention specifically includes:

[0047] Step 4-1: Construct the covariance matrix R ZZ

[0048]

[0049] Step 4-2: R ZZ Perform eigenvalue decomposition to obtain the noise subspace U n-beam-3MUSIC .

[0050] The step five of the present invention specifically includes:

[0051] Step 5-1: Define the three-dimensional search guide vector as

[0052]

[0053] where b R It is represented as the steering vector after distance dimension transformation, b θ Expressed as the steering vector after velocity dimension transformation, b vIt is represented as the steering vector after velocity dimension transformation;

[0054] Step 5-2: Through the noise subspace U n-beam-3MUSIC Establish the peak search function P beam-3MUSIC

[0055]

[0056] P beam-3MUSIC The parameter information of each dimension corresponding to the maximum point of is the speed, distance and angle information of the target.

[0057] The present invention utilizes beam space transformation and a frequency domain beam multi-parameter fast joint super-resolution estimation algorithm to ensure the target estimation accuracy while solving the high computational complexity problem of the traditional super-resolution algorithm, effectively reducing the computational complexity of the super-resolution algorithm, and providing a theoretical reference for the actual engineering application of target parameter super-resolution estimation based on security millimeter-wave radar. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Attached Figure 1 It is a flow chart of the present invention.

[0059] Attached Figure 2 : This is a target speed-distance information validity estimation diagram of the present invention and the traditional 3DFFT algorithm, where Figure 2 (a) The speed-distance information estimation spectrum peak diagram of the algorithm of the present invention, Figure 2 (b) is an effective estimation diagram of speed-distance information of the algorithm of the present invention, Figure 2 (c) is the spectral peak diagram of speed-distance information estimation by 3DFFT algorithm. Figure 2 (d) is the effective estimation of speed-distance information by 3DFFT algorithm.

[0060] Attached Figure 3 This is a comparison diagram of target information distribution between the present invention and the traditional 3DFFT algorithm, where Figure 3 (a) Four-dimensional target information distribution map, Figure 3 (b) is the target information distribution diagram in speed dimension. Figure 3 (c) is the target information distribution diagram in angle dimension.

[0061] Attached Figure 4 This is a comparison chart of the target estimation performance in each dimension between the present invention and the traditional 3DFFT algorithm, where the signal-to-noise ratio changes from -30dB to 20dB in steps of 10dB, the number of Monte Carlo times is 50, and the estimation error of Target1 is analyzed by using the traditional 3DFFT algorithm and the frequency domain beam dimensionality reduction 3DMUSIC algorithm proposed in the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0063] Aiming at the problems of low estimation accuracy, insufficient resolution and poor stability of target information in traditional fast Fourier transform algorithm and high computational complexity of traditional super-resolution algorithm, which cannot be applied in engineering, the present invention proposes a frequency domain beam multi-parameter fast joint super-resolution estimation algorithm, which transforms the target information data into frequency domain, obtains part of the data of interest through beam transformation by the equivalence of direct selection of beam space transformation and frequency domain data, and performs super-resolution estimation. The selected beam space dimension is usually much smaller than the original array data dimension, which greatly reduces the complexity of the super-resolution algorithm. Specifically, the main process of the present invention is as follows: Figure 1 As shown:

[0064] The first step is to obtain the mixed frequency beat signal of the security millimeter wave radar, wherein the first step includes the following steps:

[0065] (1) Establish sawtooth wave transmission signal model s(t)

[0066]

[0067] Among them, A mpTs Indicates the corresponding amplitude of the transmitted signal; f c is the signal carrier frequency; μ=B / T m represents the frequency modulation slope, T m is the signal repetition period, and B is the signal bandwidth.

[0068] (2) The receiving end obtains the received signal r(t)

[0069]

[0070] Among them, A mpRs is the corresponding amplitude of the received signal, τ = 2R(t) / c is the time delay between the received signal reaching the circulator and the transmitted signal, and when the target has speed, R(t) = R0 + vt, and R0 is the initial distance when stationary.

[0071] (3) The local oscillator signal in the mixer is mixed with the received signal to obtain a mixing beat signal. After ignoring its high order and extending its repetition period, the mixing beat signal S IF (t) is

[0072]

[0073] Where m=0,1....,M-1 represents the repetition cycle number.

[0074] (4) In order to improve the angle estimation accuracy, the TDM-MIMO radar is used to expand the virtual aperture and the number of transmitting antennas of the TDM-MIMO radar is set to L. TX , the number of receiving antenna elements is LRX , assuming that the receiving antenna array element spacing is d r =λ2, the spacing between transmitting antenna elements is d t , and satisfies d t =L RX ×d r , assuming that an arbitrary receiving antenna l RX The distance between the first receiving antenna in the receiving antenna array is d rlRX =(l RX -1)d r . Then the number of virtual array elements that can be generated is L = L TX ×L RX The radar system structure can be compared with the array aperture (L TX L RX -1)×λ2 virtual uniform array. For the above mixing beat signal model, the TDM-MIMO LFMCW radar sawtooth beat signal model x(t) is obtained as follows:

[0075]

[0076] Among them l tx =1,2,…L TX , l rx =1,2,…L RX are the number of transmitting and receiving antennas respectively, l=0,1....,L-1 represents the array element number of the virtual array receiving antenna, and G(t) is additive white Gaussian noise.

[0077] The second step is to transform the space-time multi-parameter domain joint data into the frequency domain by fast Fourier transform, and the second step includes the following steps:

[0078] (1) The TDM-MIMO LFMCW radar sawtooth wave beat discrete signal is sorted to obtain N×L×M dimensional three-dimensional data X 3dim , X 3dim Each N×L plane data is called a frame of data, then X 3dim There are M frames of data in total.

[0079] (2) It is known that the one-dimensional discrete Fourier transform formula is

[0080]

[0081] The two-dimensional Fourier transform formula is

[0082]

[0083] X 3dimThe three-dimensional Fourier transform is operated by combining the one-dimensional Fourier transform with the two-dimensional Fourier transform to realize the transformation of the space-time multi-parameter domain joint data to the frequency domain. First, the M-frame distance-angle data plane is subjected to a two-dimensional Fourier transform, and the frequency domain transformation of the distance-angle dimension data is completed. Then, the data of each frame after the two-dimensional Fourier transform can be vectorized to obtain (N×L)×M two-dimensional data, and then the two-dimensional data is subjected to (N×L) times M-point Fourier transform to realize the speed dimension Fourier transform. After M times (N×L) two-dimensional Fourier transform and (N×L) times M-point one-dimensional Fourier transform, the converted frequency domain data Y can be obtained. NL×M for

[0084]

[0085] in, represents the Kronecker product.

[0086] The third step is to select frequency domain data to achieve data dimensionality reduction, and the third step includes the following steps:

[0087] (1) In order to reduce the amount of data calculation, the present invention selects the huge frequency domain data. The study found that directly selecting data in the frequency domain is equivalent to beam space transformation data, that is, super-resolution of part of the frequency domain data is equivalent to beam space dimension reduction super-resolution of the original data. To illustrate the equivalence relationship, we will use the frequency domain data Y NL×M For vectorization, we have

[0088]

[0089] The left side of the above equation is the vectorization of the 3D data after Fourier transformation, and the right side is the 3D beam space transformation expression of the original data. This shows that directly selecting 3D frequency domain data and 3D beam space dimensionality reduction are equivalent. The proof process of this theorem is as follows.

[0090] (2) Theorem 1: Define the matrix F N 、F L 、F M 、F LN , X NL×M , where F N is an N×N dimensional matrix, F L is an L×L dimensional matrix, F M is an M×M dimensional matrix, F LN is an L×N dimensional matrix and X NL×M is a NL×M dimensional matrix, then

[0091]

[0092] The proof process is as follows: If x1,…,xM Yes X NL×M Column vector, then X NL×M It can be written as Transformation, where e i is the identity matrix I M The i-th column of

[0093]

[0094] Will Substitution is

[0095] The above proof is complete.

[0096] (3) Directly selecting three-dimensional frequency domain data and three-dimensional beam space dimensionality reduction are equivalent, so the process of directly selecting three-dimensional frequency domain data can be achieved through beam space transformation.

[0097] Define the angle-dimensional beam transformation matrix Where L b Select the number of effective beams in the spatial domain, that is, the length of the data after dimensionality reduction, W θ The first line of the

[0098]

[0099] Where l,0≤l≤(L b -1),L b ≤L, when L b = L, that is W θ is the angle-dimensional full beam transformation matrix. The distance-dimensional beam transformation matrix is ​​defined as N b Select the number of effective beams in the range dimension, and its nth line is:

[0100]

[0101] Where 0≤n≤(N b -1),N b ≤N, when N b =N, that is, W R is the distance dimension full beam transformation matrix. The velocity dimension beam transformation matrix is ​​defined as M b Select the number of effective beams in the velocity dimension, whose mth line is:

[0102]

[0103] Where m,0≤m≤(M b -1),M b ≤M, when M b =M, that is, W v is the velocity-dimensional full-beam transformation matrix.

[0104] (4) Determine the selection of L according to the target parameter area of ​​interest b 、M b 、N b The corresponding area of ​​interest is selected in the frequency domain data by the linear correspondence relationship to obtain the three-dimensional frequency domain data after dimensionality reduction. The starting values ​​i, j, and k of the target data in the three dimensions are determined by the distribution area of ​​the target.

[0105] The three-dimensional frequency domain data after dimension reduction After rearrangement, we can get the data vector Z

[0106]

[0107] The fourth step is to construct a covariance matrix and perform eigenvalue decomposition. The fourth step includes the following steps:

[0108] (1) Construct the covariance matrix R ZZ

[0109]

[0110] (2) R ZZ Perform eigenvalue decomposition to obtain the noise subspace U n-beam-3MUSIC

[0111] The fifth step is to obtain a signal subspace or a noise subspace according to the number of information sources, and obtain target parameter information. The fifth step includes the following steps:

[0112] (1) Through the above parts, the three-dimensional search guide vector can be defined as

[0113]

[0114] where b R It is represented as the steering vector after distance dimension transformation, b θ Expressed as the steering vector after velocity dimension transformation, b v Represented as the steering vector after velocity dimension transformation.

[0115] (2) Through the noise subspace U n-beam-3MUSIC Establish the peak search function P beam-3MUSIC

[0116]

[0117] P beam-3MUSIC The parameter information of each dimension corresponding to the maximum point of is the speed, distance and angle information of the target.

[0118] Example:

[0119] The performance of the present invention can be illustrated by the following complexity analysis and simulation:

[0120] 1. Simulation conditions:

[0121] This example uses a 24GHz millimeter wave TDM-MIMO radar platform, with a bandwidth of 150MHz for transmitting FMCW signals, a chirp period of 16us, 200 snapshots, 2*128 chirps, 2 transmitting array elements, 15 receiving array elements, and 30 extended virtual array elements. According to the traditional FFT method, the distance resolution R res =1m, velocity resolution V res =1.53m / s, angular resolution is θ res =3.8°. The two target parameters are set as Target1: distance is 150m, angle is 5°, speed is 3.0m / s; Target2: distance is 150.5m, angle is 8°, speed is 4.0m / s. In order to further evaluate the performance of the present invention, the number of Monte Carlo experiments is set to 50, and the root mean square error (RMSE) is used as the evaluation index.

[0122]

[0123]

[0124]

[0125] Where: ment is the number of Monte Carlo experiments, and are respectively the incident angle θ, distance R and estimated velocity v of the real target signal during the i-th Monte Carlo experiment.

[0126] 2. Complexity analysis:

[0127] In this example, it is assumed that the number of antenna array elements is L = 30, the number of fast time domain sampling snapshots in the distance dimension is N = 200, and the number of slow time domain snapshots in the speed dimension is M = 256. The number of angle dimension searches is defined as l = 236, the number of distance dimension searches is n = 89, and the number of speed dimension searches is m = 118. After the frequency domain dimensionality reduction is realized by the algorithm proposed in the present invention, L is taken in the angle dimension frequency domain beam. b =8, the range dimension frequency domain beam takes N b =13, velocity dimension frequency domain beam takes M b = 13. For the traditional 3DMUSIC algorithm, the amount of data storage after acquiring data is The algorithm proposed in this invention can reduce the amount of data storage to That is, the data storage capacity can be reduced by (30×256×200) / (8×13×13)=1.1361e3, that is, the storage capacity can be reduced by three orders of magnitude in the data storage part.

[0128] The computational complexity of the joint super-resolution algorithm based on MUSIC is mainly concentrated in two aspects: the first is the eigenvalue decomposition (EVD) processing part, the computational complexity is The second is the peak search (PS) part, the computational complexity is The total complexity of the algorithm is The traditional 3D MUSIC joint super-resolution algorithm, with the above system parameters, has a total computational complexity of 3.6239e18. The frequency domain beam dimensionality reduction algorithm proposed in the present invention reduces the complexity of the EVD and PS parts to Substituting the corresponding parameters, the total computational complexity is reduced to 5.8222e9. Compared with the frequency domain beam dimensionality reduction, the computational complexity is reduced by at least 9 orders of magnitude. When the selected target area is small, the number of beams can be even smaller, achieving further reduction in complexity.

[0129] 3. Simulation content and results

[0130] Simulation 1, target speed-distance information validity estimation diagram of the present invention and the traditional 3DFFT algorithm. The three-dimensional spectrum peak data is not easy to display directly. The two-dimensional data is regarded as a frame of data, and coherent accumulation is performed in the other dimension to draw a two-dimensional image, which intuitively shows the estimation of target resolution and parameters. Taking speed-distance as an example, the distance-speed dimension validity estimation diagram of the frequency domain beam reduction 3DMUSIC algorithm is drawn when SNR=0dB, and compared with the validity estimation diagram of the traditional 3DFFT algorithm, as shown in the figure. Figure 2 shown.

[0131] Figure 2 It can be seen from the observation that the algorithm proposed in the present invention can effectively realize the effective estimation of target information. Under the simulation condition, 3DFFT cannot distinguish the two targets, while the algorithm proposed in the present invention can realize the effective distinction of the two targets and more accurate parameter estimation.

[0132] Simulation 2, compare the target information distribution comparison diagram of the present invention and the traditional 3DFFT algorithm, wherein, in this experiment, SNR=0dB is set, and 50 Monte Carlo experiments are performed to compare the estimated values ​​of each dimension of the two targets with the target true values ​​by using the frequency domain beam dimensionality reduction 3DMUSIC algorithm and the 3DFFT algorithm proposed in the present invention, and obtain the distance dimension target information distribution diagram, the angle dimension target information distribution diagram and the speed dimension target information distribution diagram, such as Figure 3 shown.

[0133] from Figure 3 It can be seen that under the current simulation conditions, the traditional 3DFFT algorithm cannot complete the target resolution of Target1 and Target2, while the algorithm proposed in the present invention can successfully resolve the two targets, and the 3D parameter estimation results of the two targets are close to the true values. The 3DFFT algorithm cannot break through the Rayleigh limit, and its resolution depends on the array size. It cannot resolve multiple targets with similar parameters. The algorithm proposed in the present invention achieves super resolution in all three dimensions.

[0134] Simulation 3, comparison of the target estimation performance of each dimension between the present invention and the traditional 3DFFT algorithm, the signal-to-noise ratio changes from -30dB to 20dB with a step size of 10dB, the number of Monte Carlo is 50 times, and the traditional 3DFFT algorithm and the frequency domain beam dimensionality reduction 3DMUSIC algorithm proposed by the present invention are used to analyze the estimation error of Target1. The results are as follows Figure 4 shown.

[0135] from Figure 4 It can be seen that under different signal-to-noise ratios, the frequency domain beam dimensionality reduction 3DMUSIC algorithm proposed in the present invention has higher estimation accuracy in target distance, speed and angle dimensions than the traditional 3DFFT algorithm, and its estimation performance in the three dimensions improves with the increase of signal-to-noise ratio, and the estimation accuracy of speed dimension and distance dimension is slightly higher than that of angle dimension.

[0136] The above theoretical analysis and experimental results show that the present invention solves the problem of high computational complexity of traditional super-resolution algorithms, effectively reduces the computational complexity of super-resolution algorithms, and provides the possibility for the practical engineering application of super-resolution estimation of target parameters based on security millimeter-wave radar.

Claims

1. A millimeter-wave radar frequency domain beam multi-parameter fast joint super-resolution estimation method, characterized in that: The following steps are involved: Step 1: Obtain the mixed frequency beat signal of the security millimeter wave radar; Step 2: transform the space-time multi-parameter domain joint data into the frequency domain through fast Fourier transform; Step 3: Select frequency domain data to achieve data dimensionality reduction, including: Step 3-1: In order to reduce the amount of data calculation, select the frequency domain data: NL×M For vectorization, we have The left side of the above formula is the vectorization of the 3D data after Fourier transformation, and the right side is the 3D beam space transformation expression of the original data. This shows that directly selecting 3D frequency domain data and 3D beam space dimensionality reduction are equivalent. Step 3-2: Directly selecting 3D frequency domain data and 3D beam space dimensionality reduction are equivalent. The process of directly selecting 3D frequency domain data can be achieved through beam space transformation: Define the angle-dimensional beam transformation matrix Where L b Select the number of effective beams in the spatial domain, that is, the length of the data after dimensionality reduction, W θ The first line of the Where l,0≤l≤(L b -1),L b ≤L, when L b = L, that is W θ is the angle-dimensional full-beam transformation matrix, and the distance-dimensional beam transformation matrix is ​​defined as N b Select the number of effective beams in the distance dimension, and its nth line is: Where 0≤n≤(N b -1),N b ≤N, when N b =N, that is, W R is the distance dimension full beam transformation matrix, and the velocity dimension beam transformation matrix is ​​defined as M b Select the number of effective beams in the velocity dimension, whose mth line is: Where m,0≤m≤(M b -1),M b ≤M, when M b =M, that is, W v is the velocity dimension full beam transformation matrix; Step 3-3: Determine the selection of L according to the target parameter area b 、M b 、N b The corresponding area of ​​interest is selected in the frequency domain data by the linear correspondence relationship to obtain the three-dimensional frequency domain data after dimensionality reduction. The starting values ​​i, j, and k of the target data in the three dimensions are determined by the distribution area of ​​the target; The three-dimensional frequency domain data after dimension reduction After rearrangement, we can get the data vector Z Step 4: Construct the covariance matrix and perform eigenvalue decomposition; Step 5: Obtain the signal subspace or noise subspace according to the number of information sources and obtain the target parameter information.

2. The method for fast joint super-resolution estimation of multi-parameters of millimeter-wave radar frequency domain beam according to claim 1 is characterized in that: The step 2 specifically includes: Step 2-1: Arrange the TDM-MIMO LFMCW radar sawtooth wave beat discrete signal to obtain N×L×M dimensional three-dimensional data X 3dim , X 3dim Each N×L plane data is called a frame of data, then X 3dim There are M frames of data in total; Step 2-2: The one-dimensional discrete Fourier transform formula is known to be The two-dimensional Fourier transform formula is X 3dim The three-dimensional Fourier transform is operated by combining the one-dimensional Fourier transform with the two-dimensional Fourier transform to realize the transformation of the space-time multi-parameter domain joint data to the frequency domain: first, the two-dimensional Fourier transform is performed on the M-frame distance-angle data plane to complete the frequency domain transformation of the distance-angle dimension data, and then the data of each frame after the two-dimensional Fourier transform can be vectorized to obtain (N×L)×M two-dimensional data, and then the two-dimensional data is subjected to (N×L) times M-point Fourier transform to realize the speed dimension Fourier transform; after M times (N×L) two-dimensional Fourier transform and (N×L) times M-point one-dimensional Fourier transform, the converted frequency domain data Y NL×M for in, represents the Kronecker product.

3. The method for fast joint super-resolution estimation of multi-parameters of millimeter-wave radar frequency domain beam according to claim 1 is characterized in that: The step 1 is specifically as follows: Step 1-1: Establish sawtooth wave transmission signal model s(t): Among them, A mpTs Indicates the corresponding amplitude of the transmitted signal; f c is the signal carrier frequency; μ=B / T m represents the frequency modulation slope, T m is the signal repetition period, B is the signal bandwidth; Step 1-2: The receiving end obtains the received signal r(t), Among them, A mpRs is the corresponding amplitude of the received signal, τ = 2R(t) / c is the time delay between the received signal reaching the circulator and the transmitted signal, when the target has speed, R(t) = R0 + vt, R0 is the initial distance when stationary; Step 1-3: The local oscillator signal in the mixer is mixed with the received signal to obtain a mixed beat signal. After ignoring its high order and extending its repetition period, the mixed beat signal S IF (t) is Where m = 0, 1...., M-1 represents the repetition cycle number; Step 1-4: Use TDM-MIMO radar to expand the virtual aperture and set the number of transmitting antennas of TDM-MIMO radar to L TX , the number of receiving antenna elements is L RX , assuming that the receiving antenna array element spacing is d r =λ2, the spacing between transmitting antenna elements is d t , and satisfies d t =L RX ×d r , assuming that an arbitrary receiving antenna l RX The distance between the first receiving antenna in the receiving antenna array is d rlRX =(l RX -1)d r , then the number of virtual array elements generated is L = L TX ×L RX The radar system structure can be compared with the array aperture (L TX L RX -1)×λ2 virtual uniform array equivalent, for the mixing beat signal model described in step 1-3, the TDM-MIMO L FMCW radar sawtooth wave beat discrete signal model X(n,l,m) is obtained as Among them l tx =1,2,...L TX , l rx =1,2,...L RX are the number of transmitting and receiving antennas respectively, l=0,1....,L-1 represents the virtual array receiving antenna element number, n is the sampling number in a single frequency modulated continuous wave cycle, f s is the sampling frequency, and G is the additive white Gaussian noise.

4. The method for fast joint super-resolution estimation of multi-parameters of millimeter-wave radar frequency domain beam according to claim 1 is characterized in that: The step 4 specifically includes: Step 4-1: Construct the covariance matrix R ZZ Step 4-2: R ZZ Perform eigenvalue decomposition to obtain the noise subspace U n-beam-3MUSIC .

5. The method for fast joint super-resolution estimation of multi-parameters of millimeter-wave radar frequency domain beam according to claim 1 is characterized in that: The step five specifically includes: Step 5-1: Define the three-dimensional search guide vector as where b R It is represented as the steering vector after distance dimension transformation, b θ Expressed as the steering vector after velocity dimension transformation, b v It is represented as the steering vector after velocity dimension transformation; Step 5-2: Through the noise subspace U n-beam-3MUSIC Establish the peak search function P beam-3MUSIC P beam-3MUSIC The parameter information of each dimension corresponding to the maximum point of is the speed, distance and angle information of the target.

Images