A low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar

Abstract

The present application provides a low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar, which solves the technical problems of low angle estimation resolution, poor angle resolution and high computational complexity of existing radars. It includes: using DDM-MIMO radar to obtain discrete intermediate frequency signals, preprocessing them in the frequency domain to obtain an angle data matrix; using an angle super-resolution algorithm to calculate the angle data matrix to obtain a three-dimensional parameter estimation result; wherein, the frequency domain preprocessing includes 2D-FFT preprocessing, target detection and deambiguation by adding additional empty bands; the angle super-resolution algorithm refers to obtaining a covariance matrix through beam space conversion and real-valued transformation, and then using a multi-stage Wiener filter to obtain a signal subspace, constructing and solving the shift-invariant equation of the ESPRIT algorithm, and obtaining a three-dimensional parameter estimation result. The present application can be widely used in the technical field of millimeter-wave MIMO radar.

Description

Technical Field

[0001] The present application belongs to the technical field of millimeter wave MIMO radar, and more specifically, to a low-complexity super-resolution 3D estimation method based on millimeter wave MIMO radar. Background Art

[0002] In recent years, the emerging millimeter-wave MIMO radar has the ability of single-shot rapid estimation. Combined with broadband and virtual array element synthesis technology, it can achieve high-resolution angle estimation of multiple radar targets.

[0003] The transmitting end of the millimeter-wave MIMO radar emits mutually orthogonal waveforms, and the radar echo is separated into each waveform through the matching filter of each receiving channel.

[0004] The current mainstream MIMO waveform diversity technology solutions include TDMA, FDMA, CDMA, DDMA, etc. Among them, TDMA uses different single antennas to transmit signals alternately, which fails to fully utilize the hardware resources of the transmitting antenna, resulting in low transmission power and large losses. It is only suitable for close-range detection of low-power radar scenarios. FDMA needs to generate multiple FMCW waveforms of different frequencies within the chirp signal cycle, and CDMA needs to change the transmission phase within the chirp signal cycle, which is difficult to achieve on current millimeter-wave radar chips.

[0005] When the millimeter-wave MIMO radar uses DDMA waveform diversity technology to detect the target's velocity parameters, each transmitting antenna sends a signal at the same time, and each antenna will have a detection result on the range Doppler map (RDM) after target detection. However, it is impossible to distinguish the detection results corresponding to each antenna, resulting in incorrect target estimation and incorrect selection of the target angle data matrix, resulting in low resolution of the estimation algorithm. Summary of the invention

[0006] The purpose of the embodiments of the present application is to provide a low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar to solve the technical problems of low resolution and high complexity of the estimation algorithm in the prior art.

[0007] To achieve the above object, the embodiment of the present application provides a low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar, comprising the following steps:

[0008] Use DDM-MIMO radar to obtain discrete intermediate frequency signals, and perform frequency domain preprocessing to obtain the angle data matrix;

[0009] The angle data matrix is ​​calculated using the angle super-resolution algorithm to obtain the three-dimensional parameter estimation result;

[0010] Among them, frequency domain preprocessing includes 2D-FFT preprocessing, target detection and deambiguation by adding additional empty bands; the angle super-resolution algorithm refers to obtaining the covariance matrix through beam space conversion and real value transformation, and then using a multi-stage Wiener filter to obtain the signal subspace, constructing the shift-invariant equation of the ESPRIT algorithm and solving it to obtain the three-dimensional parameter estimation result;

[0011] The shift-invariant equation for constructing the ESPRIT algorithm is:

[0012] ;

[0013] in, E S is the signal subspace, , , b is the angle feature information, Ω b is a real-valued diagonal matrix, b1 is a real-valued diagonal matrix Ω b The first angle feature information, b2 is the real-valued diagonal matrix Ω b The second angle feature information, b p is a real-valued diagonal matrix Ω b The Pth angle feature information of , Г1 and Г2 are selection matrices, and θ is the azimuth angle.

[0014] Preferably, deambiguation refers to cyclically adding adjacent sub-intervals of the range-Doppler matrix obtained by 2D-FFT preprocessing, determining in which interval the empty band with the lowest energy value is located, and separating the range parameter and speed parameter corresponding to each transmitting antenna.

[0015] Preferably, the three-dimensional parameter estimation result is:

[0016] ;

[0017] In the formula, θ i is the three-dimensional parameter estimation result, b p is a real-valued diagonal matrix Ω b The p-th eigenvalue of , where P is the number of eigenvalues.

[0018] Preferably, the formula for beam space conversion is:

[0019] ;

[0020] ;

[0021] ;

[0022] Where Z is the data matrix, W H is the beam projection matrix, w i His the spatial filter of the ith array element, j is a complex number, N v is the number of virtual array elements, i is the i-th array element, X A is the angle data matrix, Z (μ p ) is the projected data matrix, W0 is the spatial filter of the first array element, For the The spatial domain filter of the array elements, μ p is the pth target.

[0023] Preferably, the formula for real value conversion is:

[0024] ;

[0025] Where, X Y is a real-valued data matrix, Re(Z) is the real part of the data matrix, and Im(Z) is the imaginary part of the data matrix.

[0026] Preferably, the formula of the covariance matrix is,

[0027] ;

[0028] Where, X Y is a real-valued data matrix, X Y H is the conjugate transpose of the real-valued data matrix, L is the number of sampling points.

[0029] Preferably, a multi-stage Wiener filter is used to perform multi-stage Wiener filtering on the covariance matrix to obtain a signal subspace. The specific formula is as follows:

[0030] ;

[0031] ;

[0032] E S =[ h 1, h 2,…, h p ];

[0033] In the formula, h i is the i-th matched filter, E S is the signal subspace, E is the mathematical expectation, d i-1 (t) is the i-1th reference signal, h i H is the conjugate transpose of the i-th matched filter, X i-1 (t) is the i-1th data vector, d i-1* (t) is the conjugate complex number of the i-1th reference signal, d i (t) is the i-th reference signal, h1 is the first matched filter, h2 is the second matched filter, h p is the pth matched filter, and t is the number of targets.

[0034] Preferably, the formula for the discrete intermediate frequency signal is:

[0035] ;

[0036] In the formula, S IF [k , , n v ] is a discrete intermediate frequency signal, a p is the reflection intensity of the discrete intermediate frequency signal corresponding to the pth target, j is a complex number, μ is the frequency modulation slope of the transmitted frequency modulated continuous wave signal, k is the kth chirp signal, T c is the modulation period, R p is the distance to the pth target, θ p is the azimuth of the pth target, V p is the speed of the pth target; c is the speed of light, λ is the wavelength, f s is the sampling frequency of the analog-to-digital converter, n is the nth receiving array element, N is the number of receiving arrays, N[ k, l, n v ] is discrete additive Gaussian white noise, For the sampling points, n v For nth v virtual array elements, m is the mth antenna, M is the number of antennas, M e is the number of empty bands, d t is the distance between transmitting array elements, d r is the receiving array element spacing.

[0037] Preferably, target detection refers to performing 2D CA-CFAR detection on the range Doppler matrix;

[0038] 2D CA-CFAR detection refers to the use of a two-dimensional constant false alarm rate detector to filter out noise and clutter interference in the range Doppler matrix.

[0039] Preferably, the formulas of Γ1 and Γ2 are:

[0040] ;

[0041] ;

[0042] In the formula, is the number of beams, N v is the number of virtual array elements.

[0043] The beneficial effects of the present invention are as follows: first, the present invention uses DDM-MIMO radar to obtain discrete intermediate frequency signals, and the DDMA waveform used by the present invention has higher transmission power, longer detection distance and is more stable in high-speed target speed measurement. Secondly, the frequency domain preprocessing of the present invention is used to process the discrete intermediate frequency signal, and then the angle data matrix is ​​obtained. The frequency domain preprocessing includes 2D-FFT preprocessing, target detection and deambiguation by adding additional empty bands. The present invention adds additional empty bands on the basis of the original transmitting antenna for deambiguation, accurately distinguishes the echo signals of different transmitting antennas, and improves the accuracy of the transmitting antenna and the separation speed sub-interval. Finally, the angle data matrix is ​​calculated using the angle super-resolution algorithm to obtain a three-dimensional parameter estimation result. The angle super-resolution algorithm designed by the present invention obtains the signal subspace through beam space conversion, real value transformation, and multi-stage Wiener filter, which cleverly reduces the calculation dimension of the matrix in the algorithm, avoids complex number operations, and enhances the algorithm's decoherence ability. After performance analysis, it can be seen that the root mean square error curve and the complexity of the algorithm of the angle super-resolution algorithm are lower than those of the traditional super-resolution algorithm, which effectively reduces the calculation complexity of the traditional super-resolution algorithm while maintaining high resolution. Therefore, the present invention can detect targets faster and at a longer distance, obtain angle data matrices with higher accuracy, with less noise and clutter interference, and design an angle super-resolution algorithm with lower complexity and higher angle estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 A schematic diagram of the overall process provided for an embodiment of the present application;

[0046] Figure 2 A frequency deviation diagram corresponding to a transmitting antenna provided in an embodiment of the present application;

[0047] Figure 3 A schematic diagram of target detection provided in an embodiment of the present application;

[0048] Figure 4 A schematic diagram of 2D CA-CFAR detection provided in an embodiment of the present application;

[0049] Figure 5 A speed interval diagram with two empty bands provided in an embodiment of the present application;

[0050] Figure 6 A schematic diagram of frequency domain preprocessing provided in an embodiment of the present application;

[0051] Figure 7 A beam space projection diagram provided by an embodiment of the present application;

[0052] Figure 8 A multi-stage Wiener filter forward recursive flow chart provided in an embodiment of the present application;

[0053] Fig. 9 A 2D-FFT preprocessing diagram provided in an embodiment of the present application;

[0054] Fig.10 2D CA-CFAR detection results provided by an embodiment of the present application;

[0055] Fig.11 A schematic diagram of deambiguation provided in an embodiment of the present application;

[0056] Fig.12 A diagram of 3D parameter estimation simulation results provided in an embodiment of the present application;

[0057] Fig.13 A target detection result diagram provided in an embodiment of the present application;

[0058] Fig.14 A diagram showing the measured results of 3D parameter estimation provided in an embodiment of the present application;

[0059] Fig.15 A diagram showing the change of the root mean square error with the signal to noise ratio provided in an embodiment of the present application;

[0060] Fig.16 A diagram showing how the root mean square error varies with the number of snapshots provided in an embodiment of the present application;

[0061] Fig.17 A graph showing the relationship between the algorithm complexity and the number of array elements provided in an embodiment of the present application;

[0062] Fig.18 A graph showing the relationship between algorithm complexity and the number of snapshots provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] See also Figure 1, an overall flow chart of a low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar provided in an embodiment of the present application, for the sake of convenience of explanation, only the part related to the present embodiment is shown, which is described in detail as follows:

[0065] S1: Use DDM-MIMO radar to obtain discrete intermediate frequency signals, and perform frequency domain preprocessing on them to obtain an angle data matrix.

[0066] 1) Use multiple transmitting antennas of DDM-MIMO radar to send frequency-modulated continuous waveforms. When the frequency-modulated continuous waveform encounters the detection target, it will be reflected to form an echo signal. The echo signal is sampled by the analog-to-digital converter (ADC) to obtain a discrete intermediate frequency signal. The specific steps are:

[0067] ①Build a transmission signal model:

[0068] The multiple transmitting antennas of DDM-MIMO radar (a type of millimeter wave MIMO radar) transmit frequency modulated continuous waveform signals to detect targets. The frequency modulated continuous waveform is a sawtooth wave with a modulation period of T. c The bandwidth of the transmitted frequency modulated continuous waveform signal is B, and the frequency modulation slope of the transmitted frequency modulated continuous waveform signal is μ =B / T c , the complex exponential form of the frequency modulated continuous waveform signal transmitted within one modulation cycle is:

[0069] ;

[0070] In the formula, f c is the carrier frequency of the transmitted frequency modulated continuous waveform signal, A is the complex amplitude of the transmitted frequency modulated continuous waveform signal, j is a complex number, and t is one of the modulation cycles.

[0071] Under the DDMA (Doppler diversity transmission) waveform in the frequency modulation continuous waveform, the speed values ​​detected by a target in the echo signals reflected by different transmitting antennas are different. By utilizing this speed difference, the frequency modulation continuous waveform signals transmitted by different transmitting antennas can be separated at the receiving end.

[0072] ②Add empty band:

[0073] To account for velocity ambiguity, additional null bands are added.

[0074] The chirp signal is a time-varying signal. Its phase shift is used to distinguish echo signals from different transmitting antennas, thereby improving spatial resolution.

[0075] See also Figure 2, the frequency offset on different transmit antennas (Tx1, Tx2, Tx3, Tx4) is achieved by applying different phase rotations between chirp signals (slow time), while the phase within the chirp signal (fast time) is unchanged.

[0076] If there are M transmitting antennas, the number of empty bands is M e , then the complex exponential form of the DDMA waveform after adding the empty band is:

[0077] ;

[0078] In the formula, f c is the carrier frequency of the transmitted frequency modulated continuous waveform signal, μ is the frequency modulation slope of the transmitted frequency modulated continuous waveform signal, t is one of the modulation cycles, m is the mth transmitting antenna, A is the complex amplitude of the transmitted frequency modulated continuous waveform signal, k is the kth chirp signal, k = 1,2, … ,K, K is the number of transmitted chirp signals, and j is a complex number.

[0079] For example, by using four transmitting antennas and adding two null bands, the phase offset of each transmitting antenna on different chirp signals can be obtained.

[0080] ③ Discrete intermediate frequency signal:

[0081] When the DDM-MIMO radar detects the echo signals of the distance, speed and angle parameters of P targets, if the sampling frequency of the analog-to-digital converter (ADC) is f s , then the discrete intermediate frequency signal after sampling by the analog-to-digital converter (ADC) is:

[0082] ;

[0083] In the formula, S IF [k , , n v ] is a discrete intermediate frequency signal, a p is the reflection intensity of the discrete intermediate frequency signal corresponding to the pth target, μ is the frequency modulation slope of the transmitted frequency modulated continuous wave signal, k is the kth chirp signal, T c is the modulation period, R p is the distance to the pth target, θ p is the azimuth of the pth target, V p is the speed of the pth target; c is the speed of light, λ is the wavelength, fs is the sampling frequency, n is the nth receiving array element, N is the number of receiving arrays, N[ k, , n v ] is discrete additive Gaussian white noise, For the sampling points, n v For nth v virtual array elements, j is a complex number, m is the mth antenna, M is the number of antennas, M e is the number of empty bands, d t is the distance between transmitting array elements, d r is the receiving array element spacing.

[0084] 2) Preprocess the discrete intermediate frequency signal in the frequency domain to obtain the angle data matrix. The specific steps of frequency domain preprocessing are as follows:

[0085] ①2D-FFT preprocessing:

[0086] The discrete intermediate frequency signal is transformed from the time domain to the range Doppler domain by performing a two-dimensional fast Fourier transform on the discrete intermediate frequency signal to generate a range Doppler matrix.

[0087] ②Target detection:

[0088] The range-Doppler matrix is ​​used for target detection.

[0089] The transmitted signal with the null band added is used to detect the target. The final range Doppler matrix has noise and clutter interference. Target detection is introduced to filter out noise and clutter interference.

[0090] The principle of target detection is as follows Figure 3 As shown in the figure: when there is a target at the index point in the range Doppler matrix, its energy value will be greater than the index point where there is no target. The energy value is determined by color. The warmer the color, the higher the energy. Therefore, this detection method can accurately detect the target and filter out excess noise and clutter interference.

[0091] See also Figure 4 The target detection of the present invention adopts a two-dimensional constant false alarm rate detector (2D CA-CFAR) to filter out these noise points and clutter interference, and determines the detection threshold by calculating the average power of the reference unit around the set protection unit and training unit, locks the index point where the target exists, and obtains the range Doppler matrix that removes noise and clutter interference.

[0092] ③ Defuzzification:

[0093] The target sub-interval of the range-Doppler matrix after removing noise and clutter is determined by defuzzification. The radar’s unambiguous velocity interval [−Vmax ,V max ] is divided into six sub-intervals of equal length (the maximum unambiguous speed range corresponding to each sub-interval becomes V max / 3).

[0094] For example, using four transmitting antennas and two empty bands, for a target, the discrete intermediate frequency signals of the transmitting antennas Tx1, Tx2, Tx3 and Tx4 will fall into four cyclically continuous sub-intervals in sequence, and no discrete intermediate frequency signals of the transmitting antennas fall into the remaining two sub-intervals. In this case, we only need to judge the energy values ​​of the two empty bands and find out which two intervals the empty bands are located. This will lead to the determination of the sub-interval in which the real target falls, and the resolution of the ambiguity.

[0095] The defuzzification principle is as follows Figure 5 As shown, if the discrete intermediate frequency signal falls into the four sub-intervals 1, 2, 3 and 4 in sequence, then 5 and 6 are the sub-intervals in which the null bands fall.

[0096] Therefore, as long as the position of the empty band is determined, the sub-interval next to it is the sub-interval where the real target falls. The order in which the transmitting antenna falls into the sub-interval can be correctly separated to obtain the angle data matrix of the real target.

[0097] ④ Overall frequency domain preprocessing:

[0098] The overall frequency domain preprocessing is as follows Figure 6 As shown, assuming that X is the matrix expression of the received discrete intermediate frequency signal, X RD It is the range-Doppler matrix after 2D-FFT preprocessing (Range-FTT and Doppler-FTT), and the target detection (2D CA-CTAR) and deambiguation are performed on it to obtain the range-Doppler matrix X of the real target. C Assuming there are three targets, the peak positions of the distances and velocities of the three targets are (x1, y1), (x2, y2) and (x3, y3) in the range-Doppler matrix respectively. The selected data blocks are named B1, B2 and B3, and they are combined to form the angle data matrix X A .

[0099] S2: Use the angle super-resolution algorithm to calculate the angle data matrix to obtain the three-dimensional parameter estimation result.

[0100] 1) Please refer to Figure 7 , beam space projection is performed on the angle data matrix to concentrate the signal in a fixed area to reduce the operational dimension of the angle data matrix.

[0101] The present invention uses beams, the angle data matrix after projection is:

[0102] ;

[0103] ;

[0104] ;

[0105] Where Z is the data matrix, W H is the beam projection matrix, w i H is the spatial filter of the ith array element, j is a complex number, N v is the number of virtual array elements, i is the i-th array element, X A is the angle data matrix, Z (μ p ) is the projected data matrix, W0 is the spatial filter of the first array element, For the The spatial domain filter of the array elements, μ p is the pth target.

[0106] 2) Obtain the data matrix Z after beam projection, and convert the data matrix Z from complex value space to real value space to avoid complex number operations and reduce computational complexity.

[0107] The real-valued data matrix is ​​obtained by the following formula:

[0108] ;

[0109] Where, X Y is a real-valued data matrix, Re(Z) is the real part of the data matrix, and Im(Z) is the imaginary part of the data matrix.

[0110] After obtaining the real-valued data matrix, the covariance matrix is ​​derived R :

[0111] ;

[0112] Where, X Y is a real-valued data matrix, X Y H is the conjugate transpose of the real-valued data matrix, L is the number of sampling points.

[0113] 3) Use a step-by-step forward multi-stage Wiener filter to obtain the estimated subspace E S .

[0114] The multi-stage Wiener filter is completed by the Wiener filter (a kind of iterative filter) and the matched filter (a kind of linear filter).

[0115] The multidimensional filter decomposes the observation vector through spatial projection of many orthogonal sub-vectors to obtain the signal subspace. The forward recursive process is as follows: Figure 8 shown.

[0116] Specifically, the matched filter is initialized, and d 0= R (1,:) and X 0= R ,in d 0 is the reference signal, X 0 is the data vector, B1, B2 and B3 are the blocking matrices in the process, R is the covariance matrix, extract Fig. 9 The first row of the covariance matrix R The first row is used as the reference signal.

[0117] The recursive formula of the matched filter is shown as follows:

[0118] ;

[0119] ;

[0120] In the formula, h i is the i-th matched filter, E is the mathematical expectation, d i-1 (t) is the i-1th reference signal, h i H is the conjugate transpose of the i-th matched filter, X i-1 (t) is the i-1th data vector, d i-1 * (t) is the conjugate complex number of the i-1th reference signal, d i (t) is the i-th reference signal.

[0121] The Wiener filter performs filtering in various orthogonal subspaces to adapt to different types of signal and noise characteristics and improve the performance of the multi-Wiener filter.

[0122] Under the prior condition that the number of transmitted signals is known, P The first matched filters have the same span as the signal subspace. In addition, the remaining matched filters have the same span as the noise subspace (all possible noise combinations). Therefore, the composition of the signal subspace is given by:

[0123] E S =[ h 1, h 2,…, h p ];

[0124] In the formula, E S is the signal subspace, h1 is the first matched filter, h2 is the second matched filter, h p is the pth matched filter.

[0125] 4) After obtaining the signal subspace, the shift invariant equation of the ESPRIT algorithm can be constructed as shown below:

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] In the formula, E S is the parameter of the signal subspace, Г1 and Г2 are selection matrices, Ω b is a real-valued diagonal matrix whose diagonal elements contain the desired DOA information, and b1 is a real-valued diagonal matrix Ω b The first angle feature information, b2 is the real-valued diagonal matrix Ω b The second angle feature information, b p is a real-valued diagonal matrix Ω b The Pth angle feature information, is the number of beams, N v is the number of virtual array elements, and θ is the azimuth angle.

[0132] 5) Solve Ω from the signal subspace using the least squares method b , and calculate the real-valued diagonal matrix Ω b of P eigenvalues ​​b1, b2, ..., b p , and finally through P The three-dimensional parameter estimation result θ of the target is calculated by using the eigenvalues i , as shown below:

[0133] ;

[0134] In the formula, θ i is the three-dimensional parameter estimation result of the target, b p is a real-valued diagonal matrix Ω b There are P eigenvalues ​​of , where p is the number of eigenvalues.

[0135] The three-dimensional parameter estimation result of the target can be obtained through the above-mentioned angle super-resolution algorithm.

[0136] Specific embodiment 1: Estimation method simulation and correctness verification.

[0137] A low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar is proposed and verified by simulation. The carrier frequency of the DDM-MIMO radar is 77 GHz, and it uses four transmitting antennas and four receiving antennas, and adds two additional empty bands.

[0138] The discrete intermediate frequency signal is preprocessed by 2D-FFT in the distance dimension and the speed dimension to obtain Fig. 9 The range-Doppler plot of (a visual representation of the range-Doppler matrix).

[0139] Target detection Fig.10 As shown in the figure, the protection unit in the distance dimension is set to 8, the training unit is set to 32, the protection unit in the speed dimension is set to 6, and the training unit is set to 12. The detection threshold is determined by calculating the average power of the reference units around the set protection units and training units, and the index point where the target exists is locked. The target is effectively detected under various noise and clutter interference, and the range Doppler map with noise and clutter interference removed is obtained.

[0140] The results after target detection are defuzzified, and the adjacent sub-intervals of the range Doppler map are cyclically added to determine which two intervals the empty bands with lower energy are located in. The velocity parameters corresponding to each transmitting antenna are separated, and the angle data matrix of the real target is obtained according to the detection results of the corresponding sub-intervals of TX1.

[0141] from Fig.11 It can be seen that Tx1 of target 1 falls in the third sub-interval, and the empty band falls in the first and second sub-intervals. The two empty bands of target 2 fall in the second and third sub-intervals, and Tx1 falls in the fourth sub-interval. Finally, the empty band of target 3 falls in the third and fourth sub-intervals, and Tx1 falls in the fifth sub-interval. The parameters corresponding to the three targets are analyzed in this way.

[0142] Substituting the angle data matrix of the real target into the angle super-resolution algorithm, the three-dimensional parameter estimation result of the real target is obtained. The distance, speed and azimuth estimation results of the real value and estimated value of the target are shown in Table 1:

[0143] Table 1 Three-dimensional parameter estimation results

[0144]

[0145] It can be seen from the above table that the difference between the estimated value and the true value is only about 0.1, and the error is extremely small, which shows that the present invention achieves super-resolution estimation of the target. Fig.12 The target three-dimensional parameter estimation results.

[0146] Specific embodiment 2: actual measurement and verification of the estimation method.

[0147] The DDM-MIMO radar was used for actual measurement verification. Three transmitting antennas and four receiving antennas were used, and an empty band was added. The DDM-MIMO radar was fixed, and two people walked one in front of the other 6 meters in front of the DDM-MIMO radar, respectively at the azimuths of -13º and 14º of the DDM-MIMO radar. The discrete intermediate frequency signal obtained by the DDM-MIMO radar was preprocessed by 2D-FFT to obtain the range Doppler map (an intuitive expression of the range Doppler matrix).

[0148] See also Fig.13 , target detection is performed on the range Doppler map to obtain the range Doppler map with noise and clutter interference removed.

[0149] The range Doppler image without noise and clutter interference is defuzzified, and it is separated that the Tx1 of target 1 is located in the third subinterval, and the Tx1 of target 2 is located in the second subinterval.

[0150] The angle data matrix of the target is obtained, and the angle super-resolution algorithm is used to obtain the three-dimensional parameter estimation results of the measured estimate. The distance, speed and azimuth estimation results of the target real value and estimated value are shown in Table 2.

[0151] Table 2 Three-dimensional parameter estimation results

[0152]

[0153] Comparing the estimated value of the actual measurement with the estimated value of the angle super-resolution algorithm of this method, it can be seen that the difference between the two is only about 0.1, and the error is extremely small and can be ignored. Fig.14 Target three-dimensional parameter estimation results diagram.

[0154] Specific embodiment 3: Estimation method performance analysis.

[0155] 1. Analysis of signal-to-noise ratio and snapshot number.

[0156] The signal-to-noise ratio is the ratio of signal power to noise power. It is a key parameter for measuring signal quality and is used to describe the strength of the signal relative to the size of the noise.

[0157] In signal processing, snapshot number refers to the number of times a signal is sampled within a certain period of time.

[0158] The angle super-resolution algorithm of the present invention is compared with the traditional super-resolution algorithm under different signal-to-noise ratios and snapshot numbers, and the root mean square error (RMSE) is used to evaluate the accuracy of the algorithm.

[0159] ;

[0160] In the formula, θ RMSE is the parameter of the root mean square error, P is the target number, J is the number of Monte Carlo experiments, θ j For the j The angle estimation parameters for this experiment are, θ p For the p Parameters for target setting.

[0161] In this experiment, the number of Monte Carlo experiments for each target in different scenarios is J is 100, and a total of 6 targets are tested (LS-ESPRIT algorithm, PM algorithm, MUSIC algorithm, Unitary ESPRIT algorithm, Capon algorithm and ESPRIT-Like algorithm). Fig.15 RMSE as the signal-to-noise ratio changes.

[0162] Fig.16 The relationship between RMSE and the number of target snapshots is given, where the signal-to-noise ratio of the six targets is 10dB and the number of Monte Carlo experiments is J is 100.

[0163] Under the same experimental environment, the RMSE curve of the angle super-resolution algorithm is lower than that of the other five algorithms, and is only slightly higher than the MUSIC algorithm under low signal-to-noise ratio conditions. However, as the signal-to-noise ratio increases, the algorithm performance approaches that of the MUSIC and Capon algorithms. The MUSIC algorithm requires spectral peak search, which is extremely complex and not conducive to engineering practice. Therefore, the angle super-resolution algorithm is better.

[0164] 2. Complexity analysis.

[0165] Since the complexity of traditional estimation methods is mainly concentrated on angle estimation, the complexity brought by the angle estimation algorithm is mainly analyzed. Here, the computational complexity of real number multiplication is mainly considered. The number of clocks consumed by one complex number multiplication is approximately the same as the number of clocks consumed by four real number multiplications.

[0166] in, W is the order of multi-stage Wiener filter, n g is the global search scope, L is the number of iterations, M is the number of transmitting array elements, N is the number of receiving array elements, P is the number of targets, is the number of beams.

[0167] First, the theoretical analysis of the algorithm complexity is shown in Table 3.

[0168] Table 3 Theoretical analysis of algorithm complexity

[0169]

[0170] Secondly, we analyze the relationship between the algorithm complexity and the number of array elements and snapshots, such as Figure 17-18 .

[0171] As the number of array elements and snapshots increases, the complexity of the algorithm also increases. Among them, the FFT algorithm has the lowest complexity, but the resolution of the FFT algorithm is limited by the number of FFT points, and it cannot distinguish multiple targets with very close angles. At the same time, it has a grid effect, which makes the target angle estimation inaccurate.

[0172] Finally, the angle super-resolution algorithm has the lowest complexity among the other super-resolution algorithms and is more suitable for angle estimation in large-scale MIMO arrays.

[0173] Through the above performance analysis, the angle super-resolution algorithm effectively reduces the computational complexity of the 3D parameter estimation method while maintaining high resolution, and has wider application potential.

[0174] The present invention provides a low-complexity super-resolution 3D estimation method based on millimeter wave MIMO radar. The present invention first uses DDM-MIMO radar to obtain discrete intermediate frequency signals. The DDMA wave used by the present invention has higher transmission power, longer detection distance and is more stable in high-speed target speed measurement. Secondly, the discrete intermediate frequency signal is processed by the frequency domain preprocessing of the present invention to obtain an angle data matrix. The frequency domain preprocessing includes 2D-FFT preprocessing, target detection and defuzzification by adding additional empty bands. The present invention adds additional empty bands on the basis of the original transmitting antenna for defuzzification, accurately distinguishes the echo signals of different transmitting antennas, and improves the accuracy of the transmitting antenna and the separation speed sub-interval. Finally, the angle data matrix is ​​calculated using an angle super-resolution algorithm to obtain a three-dimensional parameter estimation result. The angle super-resolution algorithm designed by the present invention obtains the signal subspace through beam space conversion, real value transformation, and multi-stage Wiener filter, which cleverly reduces the operation dimension of the matrix in the algorithm, avoids complex number operations, and enhances the decorrelation ability of the algorithm. After performance analysis, it can be seen that the root mean square error curve and algorithm complexity of the angle super-resolution algorithm are lower than those of the traditional super-resolution algorithm. While maintaining high resolution, it effectively reduces the computational complexity of the traditional super-resolution algorithm. Therefore, the invention detects targets faster and farther, obtains angle data matrix with higher accuracy, less noise and clutter interference, and the designed angle super-resolution algorithm has lower complexity and higher accuracy of angle estimation.

[0175] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0176] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar, characterized in that: The following steps are involved: Use DDM-MIMO radar to obtain discrete intermediate frequency signals, and perform frequency domain preprocessing to obtain the angle data matrix; The angle data matrix is ​​calculated using the angle super-resolution algorithm to obtain the three-dimensional parameter estimation result; The frequency domain preprocessing includes 2D-FFT preprocessing, target detection and deambiguation by adding additional empty bands; the angle super-resolution algorithm refers to obtaining the covariance matrix through beam space conversion and real value transformation, and then using a multi-stage Wiener filter to obtain the signal subspace, constructing the shift-invariant equation of the ESPRIT algorithm and solving it to obtain the three-dimensional parameter estimation result; The shift invariant equation for constructing the ESPRIT algorithm is: ； in, E S is the signal subspace, , , b is the angle feature information, Ω b is a real-valued diagonal matrix, b1 is a real-valued diagonal matrix Ω b The first angle feature information, b2 is the real-valued diagonal matrix Ω b The second angle feature information, b p is a real-valued diagonal matrix Ω b The Pth angle feature information of , Г1 and Г2 are selection matrices, and θ is the azimuth angle.

2. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The deambiguation refers to cyclically adding adjacent sub-intervals of the range-Doppler matrix obtained by 2D-FFT preprocessing, determining in which interval the empty band with the lowest energy value is located, and separating the range parameter and speed parameter corresponding to each transmitting antenna.

3. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The three-dimensional parameter estimation result is: ； In the formula, θ i is the three-dimensional parameter estimation result, b p is a real-valued diagonal matrix Ω b The p-th eigenvalue of , where P is the number of eigenvalues.

4. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The formula for the beam space conversion is: ； ； ； Where Z is the data matrix, W H is the beam projection matrix, w i H is the spatial filter of the ith array element, j is a complex number, N v is the number of virtual array elements, i is the i-th array element, X A is the angle data matrix, Z (μ p ) is the projected data matrix, W0 is the spatial filter of the first array element, For the The spatial domain filter of the array elements, μ p is the pth target.

5. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The formula for the real value conversion is: ； Where, X Y is a real-valued data matrix, Re(Z) is the real part of the data matrix, and Im(Z) is the imaginary part of the data matrix.

6. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The formula of the covariance matrix is, ； Where, X Y is a real-valued data matrix, X Y H is the conjugate transpose of the real-valued data matrix, L is the number of sampling points.

7. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The covariance matrix R is subjected to multi-stage Wiener filtering using the multi-stage Wiener filter to obtain a signal subspace. The specific formula is as follows: ； ； E S =[ h 1, h 2,…, h p ]； In the formula, h i is the i-th matched filter, E S is the signal subspace, E is the mathematical expectation, d i-1 (t) is the i-1th reference signal, h i H is the conjugate transpose of the i-th matched filter, X i-1 (t) is the i-1th data vector, d i-1 * (t) is the conjugate complex number of the i-1th reference signal, d i (t) is the i-th reference signal, h1 is the first matched filter, h2 is the second matched filter, h p is the pth matched filter, and t is the number of targets.

8. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The formula of the discrete intermediate frequency signal is: ； In the formula, S IF [k , , n v ] is a discrete intermediate frequency signal, a p is the reflection intensity of the discrete intermediate frequency signal corresponding to the pth target, j is a complex number, μ is the frequency modulation slope of the transmitted frequency modulated continuous wave signal, k is the kth chirp signal, T c is the modulation period, R p is the distance to the pth target, θ p is the azimuth of the pth target, V p is the speed of the pth target; c is the speed of light, λ is the wavelength, f s is the sampling frequency of the analog-to-digital converter, n is the nth receiving array element, N is the number of receiving arrays, N[ k, , n v ] is discrete additive Gaussian white noise, For the sampling points, n v For nth v virtual array elements, m is the mth antenna, M is the number of antennas, M e is the number of empty bands, d t is the distance between the transmitting array elements, d r is the receiving array element spacing.

9. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar as claimed in claim 1, characterized in that: The target detection refers to performing 2D CA-CFAR detection on the range Doppler matrix; The 2D CA-CFAR detection refers to using a two-dimensional constant false alarm rate detector to filter out noise and clutter interference in the range Doppler matrix.

10. The low-complexity super-resolution 3D estimation method based on millimeter-wave MIMO radar according to claim 1, characterized in that: The formulas of Г1 and Г2 are: ； ； In the formula, is the number of beams, N v is the number of virtual array elements.

Images