Laser radar ranging method and system based on mutual prime dual-channel ADC downsampling

By combining coprime dual-channel ADC sampling and a weighted compressed sensing model, the problems of high sampling difficulty and unstable ranging performance under low signal-to-noise ratio in existing lidar ranging methods are solved, and high-precision lidar ranging is achieved.

CN115220052BActive Publication Date: 2024-11-08XIDIAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210929945.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-05
Publication Date
2024-11-08
Estimated Expiration
2042-08-05

AI Technical Summary

Technical Problem

Existing lidar ranging methods based on compressed sensing theory suffer from problems such as high sampling difficulty, large time error, and inability to guarantee ranging performance under low signal-to-noise ratio.

Method used

A coprime dual-channel ADC sampling combination is used to downsample the lidar echo. By establishing a weighted compressed sensing model and optimization function, combined with a sparse reconstruction algorithm, the performance of sparse reconstruction is evaluated using a coprime dual-channel dictionary matrix to compensate for time delay errors and achieve high-precision ranging.

Benefits of technology

It achieves high-precision lidar ranging under low signal-to-noise ratio conditions, reduces the difficulty of hardware implementation, improves sampling accuracy and ranging accuracy, and can effectively reconstruct the difference frequency signal spectrum in low signal-to-noise ratio environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115220052B_ABST
    Figure CN115220052B_ABST
Patent Text Reader

Abstract

The present invention relates to a radar ranging method and system, and specifically to a laser radar ranging method and system based on coprime dual-channel ADC downsampling. The method overcomes the problems of large sampling difficulty, large time error and inability to guarantee ranging performance under low signal-to-noise ratio in the existing ranging method based on compressed sensing theory. First, the laser radar echo is downsampled using a coprime dual-channel ADC sampling combination; then, based on a reconstruction method, a sparsely reconstructed difference frequency signal spectrum S is obtained, and then the target distance value is obtained. Compared with random sampling, the coprime dual-channel ADC sampling proposed in the present invention is easy to implement in hardware and has a small system error, and it is easier to ensure the accuracy of sampling; at the same time, the influence of noise on the laser radar ranging is considered, and the influence of noise is considered in the sparse solution process, so as to realize a high-performance laser radar ranging technology under low signal-to-noise ratio.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a radar ranging method and system, and in particular to a radar ranging method and system based on mutually prime dual-channel ADC downsampling. Background Art

[0002] Since the laser radar was proposed in the 1960s, after years of research and development, its research directions and fields are numerous. Laser radar technology has very good directionality because the laser beam propagates in a straight line, and the beam is relatively narrow and can only be received on its propagation path. It has strong anti-interference ability and can work in many occasions. At the same time, because it can emit ultra-large bandwidth signals, has high distance resolution and good low-altitude detection performance, it is widely used in environmental perception, meteorological research, high-precision ranging and other fields. Among them, the use of frequency modulated continuous wave laser radar for laser ranging has been a hot topic of research in recent years. Frequency modulated continuous wave technology is an advanced laser ranging technology that can measure the absolute distance of the target. It has the characteristics of wide measurement range, high ranging accuracy, and can achieve non-contact ranging without cooperative targets. It has high real-time performance and is a hot research topic in the field of structural ranging.

[0003] Traditional pulse laser ranging technology measures distance by measuring the flight time (TOF) of the laser between the transmitting and receiving devices. Frequency modulated continuous wave is a continuous frequency modulated laser. Half of the emitted laser light stays in the local resonant cavity and mixes and interferes with the other half of the emitted light to obtain a difference frequency signal. The frequency of the difference frequency signal is estimated, and then the distance measurement is converted into the measurement of the difference frequency signal frequency according to the corresponding relationship between frequency and distance, and finally the distance information of the target is obtained.

[0004] At present, the research on improving the resolution of FMCW LiDAR ranging technology is mainly divided into two aspects. On the one hand, the effective bandwidth of FMCW ranging system is increased to improve the ranging accuracy of LiDAR by increasing the resolution in the distance direction. On the other hand, the ranging accuracy is improved by improving the spectrum estimation accuracy of the difference frequency signal in the FMCW LiDAR ranging system. These current studies are all based on the condition that LiDAR echo sampling satisfies the Nyquist sampling theorem. However, for LiDAR signals with high resolution and ultra-large bandwidth in the distance direction, the amount of echo sampling data is extremely large, and the hardware resources occupied are large. The existing sampling and signal processing equipment cannot process LiDAR signals with ultra-large bandwidth.

[0005] In order to overcome the above problems, He Jin et al. proposed a synthetic aperture laser radar imaging algorithm based on compressed sensing theory in their paper "Synthetic aperture laser radar imaging algorithm based on compressed sensing theory" (He Jin, Zhang Qun, Yang Xiaoyou, et al. Synthetic aperture laser radar imaging algorithm based on compressed sensing theory [J]. Journal of Astronautics, 2011, 32 (11): 2395-2402.). For laser radar signals with large bandwidth, sampling can be performed at conditions far lower than the Nyquist sampling theorem to achieve high-resolution laser radar imaging. First, the optical heterodyne method is used to detect the echo signal, and on this basis, the useful information in the signal is extracted by randomly sampling the heterodyne signal; then, the orthogonal matching algorithm is used to achieve high-resolution range reconstruction of the target; finally, the frequency scaling algorithm is combined to obtain a high-resolution two-dimensional image of the target. This method has the following shortcomings:

[0006] First, random sampling is performed when downsampling the target echo. However, random downsampling is difficult in practical applications, the time sampling error is large, and it is difficult to implement in hardware.

[0007] Second, the presence of noise is not considered, and the performance of sparse reconstruction under low signal-to-noise ratio conditions cannot be guaranteed. Summary of the invention

[0008] The purpose of the present invention is to provide a laser radar ranging method and system based on coprime dual-channel ADC downsampling, so as to overcome the problems of large sampling difficulty, large time error and the inability to guarantee ranging performance under low signal-to-noise ratio in the existing ranging method based on compressed sensing theory.

[0009] The technical solution of the present invention is to provide a radar ranging method based on compressed sensing and mutually prime dual-channel ADC sampling, which is special in that it includes the following steps:

[0010] Step 1: Downsample the laser radar echo using a mutually prime dual-channel ADC sampling combination;

[0011] Step 2: Reconstruct based on the downsampled data to obtain a sparsely reconstructed difference frequency signal spectrum S, and then obtain the target distance value;

[0012] Step 2.1: Establish a laser radar ranging model and optimization function based on weighted compressed sensing and coprime dual-channel ADC sampling:

[0013] The laser radar ranging model based on weighted compressed sensing and mutually prime dual-channel ADC sampling is as follows:

[0014] min(||WS||1), subject to||s r12 -U 12 FS||≤ξ

[0015] Where W represents the diagonal weight matrix; ξ is the noise level; s r12 is the dimension-reduced observation signal of the mutually prime dual-channel ADC sampling combination, U 12 is the measurement matrix of the mutually prime dual-channel ADC sampling combination, and F is the inverse Fourier basis matrix;

[0016] The optimization function is:

[0017]

[0018] in, is the sparsely reconstructed difference frequency signal spectrum estimate, μ is the sparse constraint coefficient;

[0019] Step 2.2, solving the Hessian matrix (Hessian matrix) and conjugate gradient function of the optimization function, and obtaining the sparsely reconstructed difference frequency signal spectrum S based on the weighted l1 norm minimization method;

[0020] Step 2.21, solve the initial Hessian matrix H(S 0 ):

[0021] H(S 0 )=2(U 12 F) H U 12 F+μU(S 0 )W 0

[0022] Where W 0 is the initial diagonal weight matrix, S 0 is the rough difference frequency signal spectrum, and the reduced-dimensional observation signal s is obtained by combining the mutually prime dual-channel ADC sampling r12 Perform Fourier transform to get; S 0 The first, second and nth elements in ; n is an integer greater than 2, and τ is a positive number

[0023] Step 2.22, based on the initial Hessian matrix H(S 0 ), get the initial conjugate gradient function of the optimization function

[0024]

[0025] Step 2.23: Based on the initial conjugate gradient function, obtain the sparsely reconstructed difference frequency signal spectrum estimate

[0026]

[0027] Step 2.24: Based on the judgment criteria, determine the estimated value of the sparsely reconstructed difference frequency signal spectrum Does it meet the requirements? If so, let the sparsely reconstructed difference frequency signal spectrum S equal to the sparsely reconstructed difference frequency signal spectrum estimate According to the sparsely reconstructed difference frequency signal spectrum, calculate the target distance information, otherwise, iterate and return to step 2.21, and update the difference frequency signal spectrum of the Hessian matrix in step 2.21 to the sparsely reconstructed difference frequency signal spectrum estimate obtained in step 2.23 Until the sparsely reconstructed difference frequency signal spectrum estimate Meet the requirements.

[0028] Furthermore, in step 2.1, the dimension-reduced observation signal s of the mutually prime dual-channel ADC sampling combination r12 , the observation matrix U of the mutually prime dual-channel ADC sampling combination 12 Determined by the following process:

[0029] According to the laser radar time domain echo data s of the first channel in the mutual prime dual-channel ADC sampling combination t1 , calculate the time domain difference frequency signal s of the first channel r1 ; According to the laser radar time domain echo data s of the second channel in the mutual prime dual-channel ADC sampling combination t2 , calculate the time domain difference frequency signal s of the second channel r2 ; will s r1 and r2 , arranged in a vector according to the sampling time, and the dimension-reduced observation signal s is obtained r12 ;

[0030] Arrange each row of the measurement matrix U1 of the first channel and the measurement matrix U2 of the second channel in the mutually prime dual-channel ADC sampling combination in a matrix according to the sampling time, and obtain the measurement matrix U 12 .

[0031] Furthermore, in step 2.21, the initial diagonal weight matrix W is calculated by the following formula: 0 ;

[0032]

[0033] in, is the initial diagonal weight matrix W 0 The diagonal elements in represent the spectrum S of the rough difference frequency signal 0 The nth element in The weighted coefficient of , σ is a constant to prevent the appearance of false targets in non-target areas due to excessive weights, and it is a positive number.

[0034] In order to further improve the reconstruction accuracy, in step 2.24, when returning to step 2.21, it is necessary to synchronously update the diagonal weight matrix in the Hessian matrix based on the following formula:

[0035]

[0036] Among them, S n The estimated value of the difference frequency signal spectrum sparsely reconstructed in step 2.23 The nth element in w n is the diagonal element in the updated diagonal weight matrix, which represents the spectrum estimate of the difference frequency signal after sparse reconstruction. The weighted coefficient of the nth element in .

[0037] Furthermore, in step 2.1, by estimating the noise variance σ 2 And the Laplace scaling factor γ, calculate the sparse constraint coefficient μ:

[0038] μ=σ 2 γ;

[0039] in:

[0040] σ 2 =E{(s N ) H s N};

[0041] γ=N / ||S 0 ||1;

[0042] Among them, s N is the noise sample, obtained from the rough difference frequency signal spectrum S 0 The noise unit that does not contain the target signal is taken out and vectorized to obtain; (s N ) H Indicates N The conjugate transpose of ; N is the number of sampling points that satisfies the Nyquist sampling theorem.

[0043] Further, in step 2.24, the judgment criteria are:

[0044]

[0045] If the number of iterations is 0, then For S 0 , if the number of iterations is greater than 0, then is the sparsely reconstructed difference frequency signal spectrum estimation value obtained in the previous iteration, and ρ represents the preset threshold.

[0046] Furthermore, in step 1, the optimal mutually prime dual-channel ADC sampling combination is obtained by screening, and the lidar echo is downsampled using the optimal mutually prime dual-channel ADC sampling combination.

[0047] Furthermore, in step 1, the specific screening process is as follows:

[0048] Step 1.1, calculate the RIP performance corresponding to the dictionary matrix of each group of mutually prime dual-channel ADC sampling combinations;

[0049] Step 1.2: Determine the optimal coprime dual-channel ADC sampling combination based on the RIP performance of the dictionary matrix.

[0050] In order to further improve the sampling accuracy, before using the optimal mutually prime dual-channel ADC sampling combination determined in step 1 to downsample the lidar echo, it is necessary to determine the optimal mutually prime dual-channel delay error and compensate the delay error to the mutually prime dual-channel ADC sampling combination, and then use the optimal mutually prime dual-channel ADC sampling combination after delay error compensation to downsample the lidar echo;

[0051] Specifically, the optimal mutually prime dual-channel delay error is determined through the following process:

[0052] The reference signal is sampled using the optimal mutually prime dual-channel ADC sampling combination, and the sampling time of the first channel is used as the reference, and the time delay error between the optimal mutually prime dual channels is obtained according to the offset position of the second channel difference frequency echo signal on the spectrum;

[0053] The bandwidth of the reference signal needs to satisfy the requirement of ensuring that any sampling rate in the optimal mutually prime dual-channel ADC sampling combination satisfies the Nyquist sampling rate of the difference frequency signal of the reference signal.

[0054] Furthermore, step 1.1 is specifically as follows:

[0055] Step 1.11: According to the frequency modulation period T of the laser radar p , Nyquist sampling rate F of the laser radar s , get the number of sampling points N under the Nyquist sampling rate, calculate the Nyquist sampling theorem observation matrix U and the inverse Fourier basis matrix F, where U is the N×N unit matrix, the number of rows of the U matrix represents the sampling point under the Nyquist sampling theorem and the time corresponding to the sampling point; F is the N×N inverse Fourier basis matrix;

[0056] Step 1.12: According to the frequency modulation period T of the laser radar p and the sampling rate of each set of mutually prime dual-channel ADC sampling combinations, to obtain the number of sampling points of each set of mutually prime dual-channel ADC sampling combinations;

[0057] According to the sampling time and the order of the sampling time of each set of mutually prime dual-channel ADC sampling combinations, the corresponding rows in the Nyquist sampling theorem measurement matrix U corresponding to the sampling time are retained to obtain the measurement matrix of the first channel in each set of mutually prime dual channels. Matrix size M1×N, the observation matrix of the second channel The matrix size is M2×N; M1 and M2 are the number of sampling points of the first channel and the second channel respectively;

[0058] The measurement matrix corresponding to each set of mutually prime dual channels and Each row of is arranged in each set of coprime dual-channel observation matrix according to the sampling time. In , if two channels have the same sampling time, only the row of the measurement matrix corresponding to one of the channels at the sampling time is retained; according to the measurement matrix of each set of mutually prime dual channels And the inverse Fourier basis matrix F, calculate the dictionary matrix corresponding to each set of mutually prime dual channels

[0059] Step 1.13: Calculate the RIP performance corresponding to the dictionary matrix. First calculate the matrix The characteristic value of Here, Ψ is the position of P strong scattering points in the difference frequency spectrum. 0 The matrix consisting of the columns in , for The conjugate transpose of The eigenvalue is in [1-δ p ,1+δ p ], the limiting isometry constant δ is obtained p The size of; P is a positive integer;

[0060] Step 1.14: Randomly select the positions of multiple groups of P strong scattering points in the difference frequency spectrum, repeat the operation of step 1.13, and obtain the corresponding δ of each group of mutually prime dual channels at different strong scattering point positions. p , for each set of mutually prime dual channels, p Taking the average value, we get

[0061] In step 1.2, select the value closest to 0 The corresponding mutually prime dual-channel combination is taken as the optimal mutually prime dual-channel combination.

[0062] The present invention also provides a radar ranging system based on compressed sensing and mutually prime dual-channel ADC sampling, comprising a memory and a processor, wherein the system is special in that a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the radar ranging method based on compressed sensing and mutually prime dual-channel ADC sampling are implemented.

[0063] The beneficial effects of the present invention are:

[0064] 1. The present invention adopts a mutually prime dual-channel ADC sampling combination to downsample the laser radar echo. Compared with random sampling, the mutually prime dual-channel ADC sampling proposed in the present invention has uniform sampling in each channel, is easy to implement in hardware, has small system error, and is easier to ensure the accuracy of sampling. The present invention also takes into account the influence of noise on the laser radar ranging, and takes into account the influence of noise in the sparse solution process, thereby realizing high-precision laser radar ranging technology under low signal-to-noise ratio.

[0065] 2. The present invention uses a coprime dual-channel dictionary matrix to evaluate the sparse reconstruction performance, selects a set of optimal coprime dual-channel combinations from the available coprime dual-channel combinations for lidar downsampling, and further improves the sampling accuracy.

[0066] 3. The present invention uses a reference signal to obtain the time delay error between the two channels in the optimal mutually prime dual-channel ADC sampling combination, and compensates the time delay error to the mutually prime dual-channel ADC sampling combination. Then, the optimal mutually prime dual-channel ADC sampling combination after delay error compensation is used to downsample the lidar echo, which can avoid the influence of the system error between the two channels on the reconstruction of the difference frequency signal spectrum, and further improve the sampling accuracy.

[0067] 4. The present invention utilizes the method of adaptive reweighted l1 norm minimization. In the compressed sensing solution, the weighted l1 term is closer to l0. The compressed sensing has better recovery accuracy and can effectively improve the super-resolution ranging accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The embodiment is a laser radar ranging method process based on mutual prime dual-channel ADC downsampling Figure 1 ;

[0069] Figure 2 The embodiment is a laser radar ranging method process based on mutual prime dual-channel ADC downsampling Figure 2 ;

[0070] Figure 3 The embodiment is a laser radar ranging method process based on mutual prime dual-channel ADC downsampling Figure 3 ;

[0071] Figure 4 In the embodiment Statistical average graph of maximum and minimum eigenvalues;

[0072] Figure 5 In the embodiment p Statistical mean graph;

[0073] Figure 6 It is a reference signal position diagram under different sampling channels in the embodiment;

[0074] Figure 7 : is a spectrum comparison diagram under ideal sampling and sparse reconstruction in the embodiment; wherein (a) is a spectrum diagram under ideal sampling, (b) is a spectrum diagram under downsampling 1, (c) is a spectrum diagram under downsampling 2, and (d) is a reconstructed spectrum diagram;

[0075] Figure 8(a) to Figure 8(i) 8(a) is the reconstructed spectrum of the OMP algorithm (SNR=20dB), FIG8(b) is the reconstructed spectrum of the fixed weight sparse reconstruction algorithm (SNR=20dB), and FIG8(c) is the reconstructed spectrum of the iterative weight sparse reconstruction algorithm (SNR=20dB); FIG8(d) is the reconstructed spectrum of the OMP algorithm (SNR=10dB), FIG8(e) is the reconstructed spectrum of the fixed weight sparse reconstruction algorithm (SNR=10dB), and FIG8(f) is the reconstructed spectrum of the iterative weight sparse reconstruction algorithm (SNR=10dB); FIG8(g) is the reconstructed spectrum of the OMP algorithm (SNR=5dB), FIG8(h) is the reconstructed spectrum of the fixed weight sparse reconstruction algorithm (SNR=5dB), and FIG8(i) is the reconstructed spectrum of the iterative weight sparse reconstruction algorithm (SNR=5dB). DETAILED DESCRIPTION

[0076] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0077] In order to reduce the difficulty of downsampling while ensuring the accuracy of downsampling and ranging accuracy under low signal-to-noise ratio, this embodiment uses a coprime dual-channel ADC sampling combination to downsample the laser radar echo, and then obtains a sparsely reconstructed difference frequency signal spectrum based on the weighted l1 norm minimization method, and then obtains the target distance value; it meets the requirements of high-performance difference frequency signal reconstruction under sampling below Nyquist and low signal-to-noise ratio, obtains a high-resolution target distance value, and can be applied to large-bandwidth laser radar ranging and high-performance laser radar ranging scenarios under low signal-to-noise ratio. It can be implemented using a radar ranging system based on compressed sensing coprime dual-channel ADC sampling, the system includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the steps of the radar ranging method based on coprime dual-channel ADC downsampling are implemented.

[0078] like Figure 1 As shown, the distance measurement method in this embodiment specifically includes the following steps:

[0079] Step 1. Input lidar parameters, reference signal and optional coprime dual-channel ADC sampling combination;

[0080] The laser radar parameters include the carrier frequency f of the laser radar c , wavelength λ, frequency modulation bandwidth B1, ranging range ΔR, difference frequency signal bandwidth B2, Nyquist sampling rate F s and frequency modulation period T p ; The reference signal needs to make any sampling rate in the mutually prime dual-channel ADC sampling combination meet the Nyquist sampling rate of the difference frequency signal of the reference signal; multiple selectable mutually prime dual-channel ADC sampling combinations; the mutually prime dual-channel ADC sampling combination refers to using a dual-channel ADC to sample the signal, the sampling intervals of the two channels are uniform and mutually prime, the signal is mutually primed, and the sampling intervals of the two channel ADCs are U / F respectively s and V / F s , U and V are prime numbers, and the sampling rates of the corresponding two channels are F s / U and F s / V.

[0081] If the optional mutually prime dual-channel ADC sampling combination is the mutually prime dual-channel ADC sampling combination required by the system, or the RIP performance difference of each mutually prime dual-channel ADC sampling combination is small, then any group of mutually prime dual-channel ADC sampling combinations can be directly used to downsample the lidar echo and execute step 5; otherwise, based on the process from step 2 to step 5, the optimal mutually prime dual-channel ADC sampling combination is selected to downsample the lidar echo.

[0082] Step 2. Calculate the RIP performance of the dictionary matrix corresponding to each set of mutually prime dual-channel ADC sampling combinations;

[0083] (2a) According to the frequency modulation period T of the laser radar p , Nyquist sampling rate F of the laser radar s , get the number of sampling points N under the Nyquist sampling rate; calculate the Nyquist sampling theorem observation matrix U and the inverse Fourier basis matrix F based on the number of sampling points N under the Nyquist sampling rate, where U is an N×N unit matrix, and the number of rows of the U matrix represents the sampling points under the Nyquist sampling theorem and the time corresponding to the sampling point; F is an N×N inverse Fourier basis matrix;

[0084] (2b) According to the frequency modulation period T of the laser radar p and the sampling rate of each mutually prime dual-channel ADC sampling combination, the number of sampling points of each channel of each mutually prime dual-channel ADC sampling combination is obtained, the number of sampling points of the first channel is M1, and the number of sampling points of the second channel is M2; according to the sampling time of each mutually prime dual-channel ADC sampling combination and the order of the sampling time, the corresponding row in the observation matrix U corresponding to the sampling time is reserved to obtain the observation matrix of the first channel in each mutually prime dual-channel ADC sampling combination Matrix size M1×N, the observation matrix of the second channel Matrix size M2×N; the observation matrix and Each row of is arranged in the observation matrix of each group of mutually prime dual-channel ADC sampling combination according to the sampling time. In the above example, if two channels have the same sampling time, only the row of the observation matrix corresponding to one of the channels at the sampling time is retained; according to the observation matrix of each group of mutually prime dual-channel ADC sampling combinations And the inverse Fourier basis matrix F, calculate the dictionary matrix corresponding to each group of coprime dual-channel ADC sampling combinations

[0085] (2c) Calculate the RIP performance corresponding to the dictionary matrix. First calculate the matrix The characteristic value of Here, Ψ is the position of P strong scattering points in the difference frequency spectrum. 0 The matrix consisting of the columns in , for The conjugate transpose of The eigenvalue is in [1-δ p ,1+δ p ], we can find δ p The size of; P is a positive integer, which can be determined according to the specific application scenario and requirements;

[0086] (2d) Randomly select the positions of multiple groups of P strong scattering points in the difference frequency spectrum, repeat step (2c), and obtain the corresponding δ of each group of mutually prime dual-channel ADC sampling combination at different strong scattering point positions. p , for each set of mutually prime dual-channel ADC sampling combinations, p Taking the average value, we get

[0087] Step 3. Determine the optimal mutually prime dual-channel ADC sampling combination;

[0088] δ p The closer to 0, the better the RIP performance of the dictionary matrix, so the one closest to 0 is selected The corresponding mutually prime dual-channel ADC sampling combination is used as the optimal mutually prime dual-channel ADC sampling combination.

[0089] Step 4. Using the reference signal to obtain the time delay error between two channels in the optimal mutually prime dual-channel ADC sampling combination;

[0090] The reference signal is sampled using the optimal mutually prime dual-channel ADC sampling combination, and the sampling time of the first channel is used as a reference. The time delay error between the mutually prime dual channels is obtained according to the offset position of the second channel difference frequency echo on the spectrum.

[0091] If the error between the mutually prime dual channels in the optimal mutually prime dual-channel ADC sampling combination is very small, much smaller than the error that the system needs to consider, this step can be omitted, such as Figure 2 As shown, the optimal mutually prime dual-channel ADC sampling combination is directly used to downsample the lidar echo. Accordingly, there is no need to compensate for the delay error in the next acquisition process.

[0092] Step 5. Downsample the lidar echo using the optimal mutually prime dual-channel ADC sampling combination;

[0093] The lidar echo is downsampled by using the first channel in the optimal mutually prime dual-channel ADC sampling combination to obtain the observation matrix U1 of the first channel and the lidar time domain echo data s of the first channel. t1 ; Compensate the delay error into the system, downsample the laser radar echo on the second channel after compensating the delay error, and obtain the observation matrix U2 of the second channel and the laser radar time domain echo data s of the second channel t2 .

[0094] Step 6. Calculate the time domain difference frequency signal of each channel in the optimal mutually prime dual-channel ADC sampling combination;

[0095] The laser radar time domain echo data s of the first channel obtained in step (5) t1 , calculate the time domain difference frequency signal s of the first channel r1 ; The radar time domain echo s of the second channel obtained according to step (5) t2 , calculate the time domain difference frequency signal s of the second channel r2 .

[0096] Step 7. Obtain the dimension-reduced observation signal, observation matrix and dictionary matrix of the optimal mutually prime dual-channel ADC sampling combination;

[0097] The s obtained in step (6) r1 and r2 , according to the position corresponding to the time, and before and after the sampling time, they are arranged in a vector, that is, the dimension reduction observation signal s of the optimal mutually prime dual-channel ADC sampling combination is obtained r12 , if the two channels have the same sampling time, only the time domain difference frequency signal data corresponding to the first channel at the sampling time is retained; at the same time, each row in the observation matrix U1 of the first channel and the observation matrix U2 of the second channel are also arranged in the same matrix according to the sampling time, and the observation matrix U of the optimal coprime dual-channel ADC sampling combination can be obtained. 12 , if two channels have the same sampling time, only the row of the measurement matrix corresponding to the first channel at that sampling time is retained;

[0098] like:

[0099] but

[0100] According to the observation matrix U 12 And the inverse Fourier basis matrix F, the dictionary matrix Ψ=U of the coprime dual-channel ADC sampling combination is obtained 12 F.

[0101] Step 8. Establish a weighted compressed sensing model and optimization function;

[0102] (8a) The measurement matrix U of the optimal mutually prime dual-channel ADC sampling combination obtained according to step (7) 12 , dictionary matrix Ψ, reduced dimension observation signal s r12 And the inverse Fourier basis matrix F, a lidar ranging model based on weighted compressed sensing and coprime dual-channel ADC sampling is established, as shown in formula (1):

[0103] min(||WS||1), subject to ||s r12 -U 12 FS||≤ξ (1)

[0104] Where W represents the diagonal weight matrix, and the diagonal elements of the diagonal weight matrix are represented by w n , w n Represents the weighted coefficient of the nth element in the sparsely reconstructed difference frequency signal spectrum S. ξ is the corresponding noise level. In order to restore the signal of the target area while suppressing the signal of the non-target area, a lower weight is applied to the target area and a larger weight is applied to the non-target area.

[0105] (8b) According to the lidar ranging compressed sensing model of coprime dual-channel ADC sampling in (8a), a solution optimization function is established, as shown in formula (2):

[0106]

[0107] in, is the sparsely reconstructed difference frequency signal spectrum estimate, μ is the sparse constraint coefficient, which depends on the statistical parameters of the noise and the target. In order to avoid the problem that S is not differentiable at 0, the following approximation is made: |S n |≈(|S n | 2 +τ) 1 / 2 , where τ is a tiny component, which is generally about 2 orders of magnitude smaller than the mean value of the difference frequency signal spectrum.

[0108] To solve the optimization function, formula (2) can be written as the following formula (3):

[0109]

[0110] By taking the derivative of formula (3), we obtain formula (4):

[0111]

[0112] Where U(S)=diag[1 / (|S1| 2 +τ) 1 / 2 ,1 / (|S2| 2 +τ) 1 / 2 ,…,1 / (|S n | 2 +τ) 1 / 2 ].

[0113] Step 9. Calculate the spectrum of the coarse difference frequency signal, estimate the noise variance and Laplace scaling factor, and calculate the sparse constraint coefficient;

[0114] (9a) The dimension-reduced observation signal s obtained in step (7) is r12 Perform Fourier transform to obtain the rough difference frequency signal spectrum result S 0 .

[0115] (9b) From the coarse difference frequency signal spectrum S 0 The noise unit that does not contain the target signal is taken out and vectorized using s N Indicates that, using the noise sample s N The noise variance can be accurately estimated as shown in formula (5):

[0116] σ 2 =E{(s N ) H s N} (5)

[0117] Among them (s N ) H Indicates N The conjugate transpose of .

[0118] (9c) Assuming that the elements in S follow the same distribution, each point has the same Laplace factor, as shown in formula (6):

[0119] γ=N / ||S 0 ||1 (6)

[0120] Where N is the number of sampling points that satisfy the Nyquist sampling theorem.

[0121] (9d) According to the noise variance σ estimated in (9b) 2 and the Laplace factor γ obtained in (9c), we can obtain the sparse constraint coefficient μ = σ2 γ.

[0122] Step 10. Calculate the initial diagonal weight matrix W 0 ;

[0123] According to step (9a), the rough difference frequency signal spectrum result S is obtained. 0 , let the initial diagonal weight matrix W 0 In Equal to S 0 The nth element in The initial weight coefficient can be set as:

[0124]

[0125] σ is a small constant. In order to prevent the phenomenon of false targets appearing in non-target areas due to excessive weights, σ ​​can be set to The mean of .

[0126] Step 11. Calculate the Hessian matrix and conjugate gradient function of the optimization function;

[0127] (11a) Based on the optimization function in step 8 and the rough difference frequency signal spectrum S obtained in step 9 0 And the initial weighted coefficient W obtained in step 10 0 , we can find the initial Hessian matrix H(S 0 ), as shown in formula (8):

[0128] H(S 0 )=2(U 12 F) H U 12 F+μU(S 0 )W 0 (8)

[0129] In the formula,

[0130] (11b) According to the optimization function in step (8) and the initial Hessian matrix in step (11a), the initial conjugate gradient function of the optimization function can be obtained, as shown in formula (9):

[0131]

[0132] Step 12. Using the weighted l1 norm minimization method, obtain the sparsely reconstructed difference frequency signal spectrum estimation value;

[0133] It is necessary to find the minimum value of the objective optimization function so that the conjugate gradient function of the objective function is 0. is the optimal solution, that is:

[0134]

[0135] The estimated value of the difference frequency signal spectrum after sparse reconstruction is:

[0136]

[0137] Step 13: Determine whether the spectrum estimation value of the difference frequency signal after sparse reconstruction meets the requirements;

[0138] (13a) According to the sparsely reconstructed difference frequency signal spectrum estimation value Determine whether the requirements are met:

[0139]

[0140] If the number of iterations is 0, then For S 0 ; If the number of iterations is greater than 0, then is the sparsely reconstructed difference frequency signal spectrum estimation value obtained in the previous iteration, ρ represents the preset threshold, and its value is a smaller threshold discrimination constant.

[0141] If the sparsely reconstructed difference frequency signal spectrum estimate If the requirements are met, then according to Calculate the target distance information and obtain a high-resolution target distance value;

[0142] If the sparsely reconstructed difference frequency signal spectrum estimate If the requirements are not met, iterate and return to step 11 to update the Hessian matrix. The difference frequency signal spectrum of the Hessian matrix in step 11 is updated to the sparsely reconstructed difference frequency signal spectrum estimation value obtained in step 12: until the sparsely reconstructed difference frequency signal spectrum estimation value is Satisfies the requirements. At the same time, the diagonal weight matrix in the Hessian matrix can also be updated synchronously based on the following formula:

[0143]

[0144] Among them, S n The spectrum estimate of the difference frequency signal sparsely reconstructed in step 12 The nth element in w n is the diagonal element in the updated diagonal weight matrix, which represents the spectrum estimate of the difference frequency signal after sparse reconstruction. Of course, when the accuracy of the difference frequency signal spectrum reconstruction is not high, the diagonal weight matrix can be directly used without updating the diagonal weight matrix to calculate the Hessian matrix, such as Figure 3 shown.

[0145] The beneficial effects of this embodiment are verified by simulation experiments as follows:

[0146] 1. Determine the basic parameters of the lidar, the reference signal and the mutually prime dual-channel ADC sampling combination;

[0147] In this embodiment, the laser wavelength λ is 1.5 μm, the frequency modulation bandwidth B1 is 1 GHz, the ranging range ΔR is 150 m, and the frequency modulation period T p is 1μs, the difference frequency signal bandwidth B2=2B1×ΔR / (cT p ) is taken as an example of a 1GHz laser ranging radar, the Nyquist sampling rate F s It is 1.25 times the bandwidth of the difference frequency signal. The specific parameters are shown in Table 1. There are 5 groups of mutually prime dual-channel combinations that can be selected. The sampling rate of each group of mutually prime dual-channel is shown in Table 2. The difference frequency signal bandwidth of the reference signal cannot exceed 0.85*F s / 15.

[0148] Table 1 Basic parameters of LiDAR

[0149]

[0150] Table 2 Mutually prime dual channel combinations

[0151]

[0152] 2. Calculate the RIP performance of the dictionary matrix corresponding to each set of mutually prime dual-channel ADC sampling combinations, and select the optimal mutually prime dual-channel ADC sampling combination;

[0153] According to the frequency modulation period T of the laser radar p , Nyquist sampling rate F of the laser radar s And the sampling rate of each set of mutually prime dual channels, calculate the dictionary matrix Ψ corresponding to each set of mutually prime dual channels 0 ; Under each combination of mutually prime dual channels, the number of strong scattering points P is traversed from 1 to 80, Here, Ψ is the position of P strong scattering points in the difference frequency spectrum. 0 The columns in the matrix are composed of 1000 random experiments under each P. The maximum eigenvalue and the minimum eigenvalue are statistically averaged to obtain the result as follows Figure 4 As shown, the δ of each set of mutually prime dual channels is calculated simultaneously. p , for δ p Taking the average value, we get like Figure 5 The specific results are shown in Table 3.

[0154] Table 3 RIP performance of mutually prime dual-channel combination

[0155]

[0156] Note:

[0157] The average value is the value corresponding to the strong scattering points 1 to 80 The average

[0158] δ p The closer to 0, the better the RIP performance of the dictionary matrix. Select the one closest to 0. The corresponding mutually prime dual-channel combination is taken as the optimal mutually prime dual-channel combination, from Table 3 and Figure 5 It can be seen that combination 2 The average value is the smallest, so combination 2 is the optimal coprime dual-channel combination in the entire ranging scenario, and the sampling rate of the first channel is obtained. s / 10, the second channel sampling rate F s / 11; If the number of targets in the laser radar ranging scene is determined, the corresponding Select mutually prime dual channels.

[0159] 3. Reconstruct the difference frequency signal spectrum based on the weighted compressed sensing model;

[0160] (3a) Use the optimal coprime dual channels to sample the reference signal and calculate the delay error between the channels. The result is Figure 6 As shown, the position of the target in different sampling channels can be obtained. In the first channel, the corresponding position of the target is 100m, and the corresponding position of the target in the second channel is 100.3m. According to the target position difference of 0.3m between the channels, the time error between the two channels is 2e-9s.

[0161] (3b) After compensating the channel error, the optimal mutually prime dual-channel is used to sample the lidar echo. According to the sampling result, the optimal mutually prime dual-channel observation matrix U is obtained. 12 , inverse Fourier basis matrix F, dictionary matrix Ψ and reduced dimension observation signal s r12 , calculate the reconstructed difference frequency signal spectrum:

[0162]

[0163] 4. Simulation results and analysis;

[0164] The optimal coprime dual-channel reconstructed difference frequency signal spectrum obtained by formula (14) is used to obtain the target distance information, which is compared with the target distance information obtained by the difference frequency signal spectrum under the Nyquist sampling theorem (ideal sampling), as shown in Figure 7As shown. (a) The figure shows the difference frequency signal spectrum obtained under the Nyquist sampling condition. It can be seen that there are 9 targets; (b) The figure shows the sampling rate of F s / 10, the difference frequency signal spectrum obtained under undersampling conditions; (c) is the spectrum of the difference frequency signal obtained under undersampling conditions with a sampling rate of F s / 11, the difference frequency signal spectrum obtained under undersampling conditions; (d) is the reconstructed spectrum obtained by reconstructing the difference frequency signal spectrum using formula (14). From (a) to (d), it can be seen that the proposed lidar ranging method based on coprime dual-channel ADC downsampling can accurately measure the distance between the target and the radar, while ensuring the accuracy of the target amplitude.

[0165] In the case of noise, according to the data sampled by the optimal mutually prime dual-channel ADC, the reconstructed difference frequency signal spectrum is obtained using formula (14), and the method used in the present invention is compared with the OMP algorithm and the fixed weight algorithm (the initial diagonal weight matrix constructed using the rough difference frequency signal spectrum). Figure 8(a) to Figure 8(i) shown.

[0166] The image recovered from the presence of noise Figure 8(a) to Figure 8(i) It can be seen that the OMP algorithm is very sensitive to noise. As the signal-to-noise ratio decreases, the estimation performance deteriorates and more noise points are introduced. The fixed weight method is more accurate in position estimation, but the amplitude difference is large compared with the ideal sampling. In contrast, the laser radar ranging method based on compressed sensing and coprime dual-channel ADC sampling proposed in the present invention can not only reduce the sampling rate, but also better guarantee the accuracy of the frequency amplitude and position under downsampling, and has good noise suppression ability.

Claims

1. A laser radar ranging method based on mutually prime dual-channel ADC downsampling, characterized in that: The following steps are involved: Step 1: Downsample the laser radar echo using a mutually prime dual-channel ADC sampling combination; Step 2: Reconstruct based on the downsampled data to obtain a sparsely reconstructed difference frequency signal spectrum S, and then obtain the target distance value; Step 2.1: Establish a laser radar ranging model and optimization function based on weighted compressed sensing and coprime dual-channel ADC sampling: The laser radar ranging model based on weighted compressed sensing and coprime dual-channel ADC sampling is as follows: min(||WS||1),subject to||s r12 -U 12 FS||≤ξ Where W represents the diagonal weight matrix, ξ is the noise level, s r12 is the dimension-reduced observation signal of the mutually prime dual-channel ADC sampling combination, U 12 is the measurement matrix of the mutually prime dual-channel ADC sampling combination, and F is the inverse Fourier basis matrix; The optimization function is: in, is the sparsely reconstructed difference frequency signal spectrum estimate, μ is the sparse constraint coefficient; Step 2.2, solving the Hessian matrix and conjugate gradient function of the optimization function, and obtaining the sparsely reconstructed difference frequency signal spectrum S based on the weighted l1 norm minimization method; Step 2.21, solve the initial Hessian matrix H(S 0 ): H(S 0 )=2(U 12 F) H U 12 F+μU(S 0 )W 0 Among them, W 0 is the initial diagonal weight matrix, S 0 is the rough difference frequency signal spectrum, and the reduced-dimensional observation signal s is obtained by combining the mutually prime dual-channel ADC sampling r12 Perform Fourier transform to get; S 0 The first, second and nth elements in ; n is an integer greater than 2, and τ is a positive number; Step 2.22, based on the initial Hessian matrix H(S 0 ), get the initial conjugate gradient function of the optimization function Step 2.23: Based on the initial conjugate gradient function, obtain the sparsely reconstructed difference frequency signal spectrum estimate Step 2.24: Based on the judgment criteria, determine the estimated value of the sparsely reconstructed difference frequency signal spectrum Does it meet the requirements? If so, let the sparsely reconstructed difference frequency signal spectrum S equal to the sparsely reconstructed difference frequency signal spectrum estimate The spectrum estimate of the difference frequency signal based on sparse reconstruction Calculate the target distance information, otherwise, iterate and return to step 2.21, and update the difference frequency signal spectrum of the Hessian matrix in step 2.21 to the sparsely reconstructed difference frequency signal spectrum estimate obtained in step 2.23 until the sparsely reconstructed difference frequency signal spectrum estimate is Meet the requirements.

2. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 1 is characterized in that: In step 2.1, the dimension-reduced observation signal s of the mutually prime dual-channel ADC sampling combination r12 , the observation matrix U of the mutually prime dual-channel ADC sampling combination 12 Determined by the following process: According to the laser radar time domain echo data s of the first channel in the mutual prime dual-channel ADC sampling combination t1 , calculate the time domain difference frequency signal s of the first channel r1 ; According to the laser radar time domain echo data s of the second channel in the mutual prime dual-channel ADC sampling combination t2 , calculate the time domain difference frequency signal s of the second channel r2 ; will s r1 and r2 , arranged in a vector according to the sampling time, and the dimension-reduced observation signal s of the mutually prime dual-channel ADC sampling combination is obtained r12 ; Arrange each row of the measurement matrix U1 of the first channel and the measurement matrix U2 of the second channel in the mutually prime dual-channel ADC sampling combination in a matrix according to the sampling time, and obtain the measurement matrix U 12 .

3. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 2 is characterized in that: In step 2.21, the initial diagonal weight matrix W is calculated by the following formula 0 ; in, is the initial diagonal weight matrix W 0 The diagonal elements in represent the spectrum S of the rough difference frequency signal 0 The nth element in The weighted coefficient of , σ is a constant to prevent the appearance of false targets in non-target areas due to excessive weights, and it is a positive number.

4. The laser radar ranging method based on coprime dual-channel ADC downsampling according to any one of claims 1 to 3, characterized in that: In step 2.24, when returning to step 2.21, the diagonal weight matrix in the Hessian matrix needs to be updated synchronously based on the following formula: Among them, S n The estimated value of the difference frequency signal spectrum sparsely reconstructed in step 2.23 The nth element in w n is the diagonal element in the updated diagonal weight matrix, representing the sparsely reconstructed difference frequency signal spectrum estimate The weighted coefficient of the nth element in , σ is a constant to prevent false targets from appearing in non-target areas due to excessive weights, and is a positive number.

5. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 4 is characterized in that: The sparse constraint coefficient μ in step 2.1 is calculated based on the following formula: μ = σ 2 c; in: σ 2 =And{(s N ) H s N }; γ=N / ||S 0 ||1; Among them, σ 2 is the noise variance, γ is the Laplace scaling factor, E{·} is the averaging operation, s N is the noise sample, obtained from the rough difference frequency signal spectrum S 0 The noise unit that does not contain the target signal is taken out and vectorized to obtain; (s N ) H Indicates N The conjugate transpose of ; N is the number of sampling points that satisfies the Nyquist sampling theorem.

6. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 5 is characterized in that: In step 2.24, the judgment criteria are: If the number of iterations is 0, then For S 0 ; If the number of iterations is greater than 0, then is the sparsely reconstructed difference frequency signal spectrum estimation value obtained in the previous iteration, and ρ represents the preset threshold.

7. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 6 is characterized in that: In step 1, the optimal mutually prime dual-channel ADC sampling combination is obtained through screening, and the lidar echo is downsampled using the optimal mutually prime dual-channel ADC sampling combination.

8. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 7 is characterized in that: The specific screening process in step 1 is as follows: Step 1.1, calculating the RIP performance corresponding to the dictionary matrix of multiple sets of mutually prime dual-channel ADC sampling combinations; Step 1.2: Determine the optimal coprime dual-channel ADC sampling combination based on the RIP performance of the dictionary matrix.

9. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 8 is characterized in that: In step 1, the optimal mutually prime dual-channel ADC sampling combination after delay error compensation is used to downsample the lidar echo; Specifically, the optimal mutually prime dual-channel delay error is determined through the following process: The reference signal is sampled using the optimal mutually prime dual-channel ADC sampling combination, and the sampling time of the first channel is used as the reference, and the time delay error between the optimal mutually prime dual channels is obtained according to the offset position of the second channel difference frequency echo signal on the spectrum; The bandwidth of the reference signal needs to satisfy the requirement of ensuring that any sampling rate in the optimal mutually prime dual-channel ADC sampling combination satisfies the Nyquist sampling rate of the difference frequency signal of the reference signal.

10. The laser radar ranging method based on mutually prime dual-channel ADC downsampling according to claim 9 is characterized in that: Step 1.1 is as follows: Step 1.11: According to the frequency modulation period T of the laser radar p The Nyquist sampling rate F of the laser radar s , obtain the number of sampling points N at the Nyquist sampling rate, and calculate the Nyquist sampling theorem observation matrix U and the inverse Fourier basis matrix F based on the number of sampling points N at the Nyquist sampling rate; Step 1.12: According to the frequency modulation period T of the laser radar p and the sampling rate of each set of mutually prime dual-channel ADC sampling combinations, to obtain the number of sampling points of each set of mutually prime dual-channel ADC sampling combinations; According to the sampling time and the order of the sampling time of each set of mutually prime dual-channel ADC sampling combinations, the corresponding rows in the Nyquist sampling theorem measurement matrix U corresponding to the sampling time are retained to obtain the measurement matrix of the first channel in each set of mutually prime dual-channel ADC sampling combinations. Matrix size M1×N, the observation matrix of the second channel The matrix size is M2×N; M1 and M2 are the number of sampling points of the first channel and the second channel respectively; The observation matrix and Each row of is arranged in the observation matrix according to the sampling time. middle, For each set of mutually prime dual-channel ADC sampling combination, if the two channels have the same sampling time, only the row of the observation matrix corresponding to one of the channels at the sampling time is retained; according to the observation matrix of each set of mutually prime dual-channel ADC sampling combination And the inverse Fourier basis matrix F, calculate the dictionary matrix corresponding to each group of coprime dual-channel ADC sampling combinations Step 1.13, calculate the RIP performance corresponding to the dictionary matrix; First, calculate the matrix The eigenvalues ​​of It refers to the Ψ corresponding to the position of P strong scattering points in the difference frequency spectrum. 0 The matrix consisting of the columns in , for The conjugate transpose of ; P is a positive integer; Afterwards, according to The eigenvalue of p ,1+δ p ], the limiting isometry constant δ is obtained p size; Step 1.14: Randomly select the positions of multiple groups of P strong scattering points in the difference frequency spectrum, repeat the operation of step 1.13, and obtain the corresponding δ of each group of mutually prime dual channels at different strong scattering point positions. p , for each set of mutually prime dual channels, p Taking the average value, we get In step 1.2, select the value closest to 0 The corresponding mutually prime dual-channel combination is taken as the optimal mutually prime dual-channel combination.

11. A laser radar ranging system based on coprime dual-channel ADC downsampling, comprising a memory and a processor, characterized in that: The memory stores a computer program, which, when executed by the processor, implements the steps of the laser radar ranging method based on coprime dual-channel ADC downsampling as described in any one of claims 1-10.