Radar target distance speed estimation method

Abstract

The invention relates to a radar target distance speed estimation method. The method comprises the following steps: acquiring a target difference frequency signal; extracting a plurality of IMF components of the target difference frequency signal; screening the IMF component with high correlation with the target difference frequency signal, performing denoising, and performing signal reconstruction to obtain a first processing signal; and analyzing and obtaining a distance estimation value and a speed estimation value of the radar target based on the first processing signal. According to the invention, joint estimation of the distance and the speed can be accurately and rapidly carried out in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing, and in particular to a radar target distance and speed estimation method. Background Art

[0002] With the development of radar technology, FMCW radar has been widely used in automobiles, drones, industrial inspection and other fields due to its high resolution, low power consumption and relatively simple hardware implementation. FMCW radar can obtain the distance information of the target by modulating the frequency, and measure the speed of the target by the Doppler effect. The accuracy and speed of frequency estimation in this process are directly related to the accuracy and real-time performance of radar speed and distance measurement. Therefore, how to accurately and quickly estimate the distance and speed in complex environments has become a hot topic in radar measurement research.

[0003] The main goal of FMCW radar signal processing is to estimate the distance and speed of the target by analyzing the frequency shift caused by the Doppler effect and the time delay of the reflected signal. FFT is usually used to convert time domain signals into frequency domain signals and extract distance and speed information in the frequency domain. However, when FFT is applied to FMCW radar signal processing, especially in low signal-to-noise ratio environments, there are several limitations: signal stationarity and linearity assumptions, resolution limitations, and noise sensitivity. The above reasons make it difficult for traditional distance and speed joint estimation algorithms to achieve ideal results in complex environments and low signal-to-noise ratio conditions. It is urgent to explore a method that can comprehensively solve the above problems and improve the estimation resolution and estimation accuracy of distance and speed while maintaining low computational complexity. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a radar target distance and speed estimation method, which improves the estimation resolution and estimation accuracy while maintaining low calculation complexity.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a radar target distance and speed estimation method, comprising the following steps:

[0006] Acquire a target difference frequency signal;

[0007] Extracting multiple IMF components of the target difference frequency signal;

[0008] After selecting the IMF component with high correlation with the target difference frequency signal for denoising, a first processed signal is obtained by signal reconstruction;

[0009] An estimated distance value and an estimated speed value of the radar target are obtained based on the first processed signal analysis.

[0010] Further, the screening of the IMF components having a high correlation with the target difference frequency signal for denoising includes:

[0011] Calculating the correlation coefficient between each of the IMF components and the target difference frequency signal;

[0012] Acquire the IMF component whose correlation coefficient is greater than a correlation threshold as a first target component;

[0013] Acquire the IMF component whose correlation coefficient is not greater than a correlation threshold as a second target component;

[0014] The first target component is denoised.

[0015] Furthermore, the denoising of the first target component is achieved by performing singular value decomposition on each of the first target components and removing singular values ​​that are less than a singular value threshold.

[0016] Further, the multiple IMF components include all IMF components except the first-order IMF components.

[0017] Further, the extracting multiple IMF components of the target difference frequency signal includes:

[0018] Decomposing the target difference frequency signal by using the CEEMD method to obtain multiple IMF components and remainders;

[0019] Extract all IMF components except the first-order IMF components.

[0020] Furthermore, the target difference frequency signal is decomposed by the CEEMD method to obtain multiple IMF components and remainders, including:

[0021] Adding a plurality of white noises of equal amplitude and in pairs of positive and negative values ​​to the target difference frequency signal to obtain a plurality of second processed signals corresponding to different white noises;

[0022] Decomposing each of the second processed signals by using an EMD method to obtain a multi-order IMF noise-added component and a remainder of each of the second processed signals;

[0023] The summed average of the IMF noise-added components of any order corresponding to different white noises is calculated as the IMF component of the target difference frequency signal of the order.

[0024] Furthermore, the first processed signal is obtained by signal reconstruction by using the first target component, the second target component and the remainder after denoising to perform signal reconstruction.

[0025] Further, obtaining a distance estimation value and a speed estimation value of the radar target based on the first processed signal analysis includes:

[0026] Converting the first processed signal from the time domain to the frequency domain to obtain a frequency domain signal;

[0027] Performing peak detection on the frequency domain signal to obtain a target frequency peak index and a distance frequency domain signal matrix;

[0028] Calculate the target main frequency corresponding to the radar target according to the frequency peak index;

[0029] The Doppler estimation is performed using the range-frequency domain signal matrix to obtain a frequency estimation value corresponding to the radar target; and the range estimation value and the speed estimation value corresponding to the radar target are calculated using the target main frequency and the frequency estimation value respectively.

[0030] Further, the Doppler estimation comprises the following steps:

[0031] Extract the FFT bands corresponding to the distance and construct the covariance matrix;

[0032] Performing singular value decomposition on the covariance matrix after forward-backward averaging;

[0033] Extract the noise subspace of all targets obtained after singular value decomposition;

[0034] Construct the spectral function matrix and solve the roots of the characteristic polynomial, select the root with the largest amplitude, and obtain the peak rate estimate of the target.

[0035] Beneficial Effects

[0036] Due to the adoption of the above technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention introduces the CEEMD method and uses several IMF components to characterize signal characteristics, thereby avoiding the limitations of FFT when applied to FMCW radar signal processing, solving the non-stationary and nonlinear characteristics caused by target motion, multipath effect, radar hardware or dynamic influence, and avoiding the loss of signal details in a low-noise environment; in addition, by introducing the SVD method, the first P larger singular values ​​are selected through SVD decomposition, and the denoised high-correlation IMF components, unprocessed IMF components and residuals are used for signal reconstruction, and the effective signal is protected while denoising, thereby solving the problem of residual noise of the CEEMD method affecting the signal; the present invention performs denoising through the CEEMD-SVD-joint method, and based on the denoised reconstructed signal, the joint estimation of distance and speed is realized by using FFT and Root-MUSIC fusion, thereby improving the accuracy and robustness of the joint estimation of target distance and speed in a low signal-to-noise ratio (SNR) environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of an embodiment of the present invention;

[0038] Figure 2 This is a comparison chart of the RMSE of single target velocity estimation of the algorithm under different SNRs in the implementation mode of the present invention;

[0039] Figure 3 This is a comparison chart of the RMSE of the multi-target velocity estimation algorithm for target one under different SNRs according to the implementation mode of the present invention;

[0040] Figure 4 This is a comparison chart of the RMSE of the multi-target velocity estimation algorithm for target two under different SNRs according to the implementation mode of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0042] The embodiment of the present invention relates to a FMCW radar target CEEMD-SVD-FFT-Root-MUSIC distance and velocity estimation method, such as Figure 1 As shown, the following steps are included:

[0043] Step 1: Obtain FMCW radar difference frequency signal

[0044] Assume f t (t) and f e (t) are the frequency variation functions of the transmitting signal and receiving signal of the FMCW radar, respectively. The radar transmits M cycles of Chirps in each frame. Each Chirps has N sampling points, n represents the nth sampling point of the Chirps signal, m represents the mth Chirps signal of the Chirps signal sequence, and assumes that the sweep bandwidth is B, the delay is τ (the time from signal transmission to reception), the sweep period is T, and f b is the difference-frequency signal (Difference-FrequencySignal, DFS, i.e. the frequency difference between the transmitted signal and the received signal) when measuring a stationary object, c is the electromagnetic wave speed, and its scanning frequency K can be expressed as

[0045]

[0046] On the rising edge of the signal, the following relationship exists

[0047] f t (t) = f0 + Kt#(2)

[0048] f r (t) = f t (t-τ)#(3)

[0049] Then the function of the difference frequency signal is

[0050] f b (t) = f b (t)-f r (t) = Kτ#(4)

[0051] Assuming that the transmission signal of the FMCW radar is a cosine wave, the time domain changes of its transmission and reception signals are as follows:

[0052]

[0053] Among them, A tx and A rx are the amplitude gains of the transmitted and received signals, respectively, and are the phases of the transmitted and received signals respectively.

[0054] The transmitted signal is mixed with the received signal to obtain the in-phase component, and then passed through a low-pass filter to obtain equation (7).

[0055]

[0056] By sorting,

[0057]

[0058] Similarly, the orthogonal components are

[0059]

[0060] The in-phase component and the quadrature-phase component are combined into a complex signal, namely the intermediate frequency signal S(t). Since the value of τ is small, τ can be ignored. 2 , and the additional phase does not contain distance information, so the difference frequency signal is

[0061] S(t)=A mix W R (t)·exp(j(2π·f b (t)·t+2πf0τ))=A mix W R (t)e jψ(t) #(10)

[0062] Among them, ψ(t) = 2π·Kτ·t+2πf0τ is the phase of the intermediate frequency signal, W R (t) is a rectangular window.

[0063] Step (2) Perform CEEMD decomposition on the difference frequency signal x(t)

[0064] The new signal is constructed by using the difference frequency signal x(t) and the positive and negative paired additive Gaussian white noise n(t) with amplitude ξ0.

[0065]

[0066] Wherein, p is a coefficient for controlling the positive and negative of noise, i is the number of times Gaussian white noise is added, i=1, 2, ..., M / 2.

[0067] Using EMD to analyze the new signal Decompose and obtain several IMF components imf i,j (t) and the remainder res i (t):

[0068]

[0069] Where j represents the signal x i (t) is the number of IMF components decomposed, and N is the total number of IMF components decomposed.

[0070] Then, for each input signal Decompose until the remainder cannot be decomposed. i,j ±(t) Perform the summed average calculation to obtain the jth IMF component of the complete integrated EMD (CEEMD) decomposition

[0071]

[0072] For the N IMF components obtained by decomposition, it can be expressed as

[0073]

[0074] Step (iii) Determine the appropriate autocorrelation coefficient threshold and filter out the intrinsic mode functions (IMF) components with high correlation

[0075] Since the main component of the first-order IMF component is noise and only contains a small amount of effective signal, the present invention chooses to directly remove it.

[0076] According to the characteristics of EMD approximate orthogonality, among the IMF components obtained by EMD decomposition, some IMFs have a high degree of orthogonality with the original signal and a strong correlation, while other IMF components have a low degree of orthogonality with the original signal and a weak correlation. Among them, the IMF components with strong correlation contain most of the effective information and noise of the original signal, while the IMF components with weak correlation mainly contain a small amount of effective information and noise. Therefore, in the present invention, after obtaining the IMF components and remainders through CEEMD decomposition, the correlation between each IMF component and the original signal is used to represent the closeness of the relationship between the two variables. The mathematical expression of the correlation coefficient is as follows:

[0077]

[0078] Where X and Y represent two random variables, Cov(X,Y) represents the covariance between random variables X and Y, σ X σ Y and represent their standard deviations, and represent their means, respectively. The range of the correlation coefficient is [-1,1]. The larger the absolute value of the correlation coefficient, the higher the correlation between the variables.

[0079] In the present invention, the IMF component obtained by CEEMD decomposition is expressed as Y i (T), then the correlation coefficient between the IMF component and the original signal is expressed as:

[0080]

[0081] Step (iv) De-noise the selected high-correlation IMF components by combining SVD decomposition, and reconstruct the signal by combining the denoised IMF components, the undenoised IMF and the signal residual

[0082] The correlation coefficient of each IMF component is calculated by formula (16), and the SVD method is continued to be used for further denoising of the IMF component with a large correlation coefficient, so that the IMF component with high correlation can remove the noise while retaining the effective information as much as possible, and obtain a new IMF component. After obtaining the signal after the new IMF component, the new IMF component after SVD denoising, the unprocessed IMF component and the residual signal (remainder) are reconstructed to obtain the first processed signal Y(t).

[0083] In this embodiment, denoising is performed by using the SVD method to reduce the influence of residual noise on the signal.

[0084] For the IMF components with high correlation with the original signal after CEEMD decomposition, they can be represented by a two-dimensional matrix of order K×N, where K represents K IMF components and N represents the number of sampling points, which can be recorded as Y K×N , then the norm of Y is:

[0085]

[0086] The total signal energy can be expressed as:

[0087] ||Y|| F 2 =trace{YY T}=trace{Y T Y}#(18)

[0088] Where trace{} represents the trace of the matrix, which is equal to the sum of the diagonal elements of the matrix. Take the first P singular values ​​to reconstruct the matrix and record the reconstructed matrix as Y P , which can be expressed as:

[0089] Y P =U∑ P V T #(19)

[0090] Among them, ∑ P =diag(σ1,σ2,…,σ P ,0,……,0).

[0091] From equations (18) and (19), we can obtain:

[0092]

[0093] Among them, r is the rank of the matrix, σ 2 is the noise variance. For formula (20), if Y P =0, then

[0094]

[0095] As can be seen from the above formula, the total energy of the signal is equal to the sum of the diagonal elements of the matrix, that is, the sum of the non-zero singular values. When reconstructing the signal, the larger the singular value, the stronger the signal energy and the greater the contribution to signal reconstruction. Therefore, when reconstructing the signal, we only need to select the first P larger singular values.

[0096] Step 5: FFT frequency estimation and peak detection

[0097] The first processed signal Y(n,m) of M Chirps is subjected to N-point FFT processing to obtain the frequency domain signal Y(i,m) and perform peak detection to obtain the distance peak index value I of l targets (l=1, 2, ..., L).

[0098] By the peak index value I of the corresponding target l And the frequency resolution Δf to obtain the peak frequency of the corresponding target:

[0099] f peak,l =I l ·Δf#(22)

[0100] For all distance units corresponding to each target (range FFT results), that is, distance bins

[0101]

[0102] Then, a target is selected as the input signal of the Doppler MUSIC algorithm. In this embodiment, the range FFT result of the first target is Set as the input signal to the Doppler algorithm:

[0103]

[0104] Since the signal subspace G S and the noise subspace G N Orthogonal, we can get R x G N =σ 2 G N , then we know that A H G N =Ο. Let A=[a(ω1),…,a(ω D )] into the equation, we can get

[0105] a H (ω)E N =0,ω=ω1,ω2,…,ω D #(25)

[0106] Then we can construct the polynomial

[0107] f(ω)=a H (ω)E N E N H a(ω)#(26)

[0108] Among them, a(ω)=[1,exp(-jω),…,a(-jω(M-1))].

[0109] Since the signal vector and the noise subspace are orthogonal, for the signal frequency ω i ,have

[0110] f(ω i )=0#(27)

[0111] Finally, the Root-MUSIC spectrum for Doppler estimation is obtained

[0112]

[0113] By solving the peak value of the above spectrum S(ω), we can get the root ω with the largest amplitude. i , we can get the Doppler frequency estimate

[0114]

[0115] Wherein, PRF is the pulse repetition frequency, which is the inverse of the pulse repetition period (PRI), that is, PRF=1 / PRI.

[0116] Step 6: Distance and speed estimation

[0117] The frequency estimate f according to the distance dimension peak,l and the frequency estimate f in the Doppler dimension i To estimate the distance and speed of the target respectively:

[0118]

[0119] In order to verify the performance of the CEEMD-SVD-FFT-Root-MUSIC ranging and speed measurement algorithm, MATLAB was used to perform simulation, and the performance of the CEEMD-SVD-FFT-Root-MUSIC algorithm proposed in the present invention was verified. The simulation parameters are shown in Table 1.

[0120] Table 1 Simulation parameters

[0121]

[0122] Simulation experiments were conducted on environmental experimental parameters such as rivers, manholes, and sedimentation ponds in cities. Based on the parameters in Table 1, MATLAB was used to implement 2D-FFT, FFT-MUSIC, FFT-Root-MUSIC, and the simulation of the joint distance and speed estimation of the measurement target in this embodiment. Multiple Monte Carlo experiments were performed to calculate the root mean square error RMSE of the distance and speed estimation results. The specific experimental results are as follows:

[0123] (1) Simulation estimation

[0124] Based on the parameter settings, the maximum distance is 29.9076m, the maximum speed is 3.1565m / s, the distance resolution is 0.1499m, and the speed resolution is 0.1973m / s. For the single target simulation at 1.8m / s at 5m, 100 Monte Carlo simulation experiments were conducted and the root mean square error RMSE was calculated. The performance of each algorithm is shown by comparing the RMSE values ​​of 2D-FFT, FFT-MUSIC and CEEMD-SVD-FFT-Root-MUSIC algorithms under different SNRs, such as Figure 1 As shown. Under different SNRs, in the simulation setting of a single target at 1.8m / s at a distance of 5m, the distance estimation value is 5.0235m. For the estimation of a single target, the performance of 2D-FTT, FFT-MUSIC and CEEMD-SVD-FFT-Root-MUSIC algorithms for single target speed estimation under different SNRs is compared. Due to the fence effect, the RMSE values ​​of speed estimation of 2DFFT and FFT-MUSIC algorithms are basically stable. In the figure, the CEEMD-SVD-FFT-Root-MUSIC algorithm proposed in this paper is significantly better than the first two algorithms. Figure 3 It can be seen that when SNR>-2dB, the estimation accuracy of the CEEMD-SVD-FFT-Root-MUSIC algorithm can reach 10 -4 Order of magnitude.

[0125] For the dual targets at 2m / s and 1.8m / s at an initial distance of 5m, 100 Monte Carlo simulations were performed and the root mean square error (RMSE) was calculated. The results are as follows: Figure 3 and Figure 4 As shown in Figure 2, for the multi-target simulation experiment, the estimated distance is 5.0235m. Figure 3 and Figure 4It can be seen that under low SNR values, the FFT-MUSIC and FFT-Root-MUSIC algorithms have large errors in the estimation of the signal subspace and the noise subspace because the MUSIC and its improved Root-MUSIC algorithm are affected by noise, and their RMSE values ​​also fluctuate. The CEEMD-SVD-FFT-Root-MUSIC algorithm proposed in this paper combines the CEEMD decomposition algorithm on the basis of the Root-MUSIC algorithm. Then the residual noise will be transferred from the high-frequency IMF component to the low-frequency IMF component. The estimation error will accumulate with the increase in the number of experiments, and the experimental results are not very stable. However, it can be seen from the figure that the estimation performance of the CEEMD-SVD-FFT-Root-MUSIC algorithm is significantly better than that of the 2D-FFT, FFT-MUSIC and FFT-Root-MUSIC algorithms. When the SNR is greater than or equal to -2dB, the accuracy of the CEEMD-SVD-FFT-Root-MUSIC algorithm can reach 10 -3 Order of magnitude.

Claims

1. A radar target distance and speed estimation method, characterized in that: The following steps are involved: Acquire a target difference frequency signal; Extracting multiple IMF components of the target difference frequency signal; After selecting the IMF component with high correlation with the target difference frequency signal for denoising, a first processed signal is obtained by signal reconstruction; An estimated distance value and an estimated speed value of the radar target are obtained based on the first processed signal analysis.

2. The method according to claim 1, characterized in that The screening of the IMF component having a high correlation with the target difference frequency signal for denoising comprises: Calculating the correlation coefficient between each of the IMF components and the target difference frequency signal; Acquire the IMF component whose correlation coefficient is greater than a correlation threshold as a first target component; Acquire the IMF component whose correlation coefficient is not greater than a correlation threshold as a second target component; The first target component is denoised.

3. The method according to claim 2, characterized in that The denoising of the first target component is achieved by performing singular value decomposition on each of the first target components and removing singular values ​​that are less than a singular value threshold.

4. The method according to claim 2, characterized in that: The multiple IMF components include all IMF components except the first-order IMF component.

5. The method according to claim 4, characterized in that The step of extracting a plurality of IMF components of the target difference frequency signal comprises: Decomposing the target difference frequency signal by using the CEEMD method to obtain multiple IMF components and remainders; Extract all IMF components except the first-order IMF components.

6. The method according to claim 5, characterized in that The first processed signal is obtained by signal reconstruction, which is obtained by performing signal reconstruction using the first target component, the second target component and the remainder after denoising.

7. The method according to claim 5, characterized in that The target difference frequency signal is decomposed by the CEEMD method to obtain multiple IMF components and remainders, including: Adding a plurality of white noises of equal amplitude and in pairs of positive and negative values ​​to the target difference frequency signal to obtain a plurality of second processed signals corresponding to different white noises; Decomposing each of the second processed signals by using an EMD method to obtain a multi-order IMF noise-added component and a remainder of each of the second processed signals; The summed average of the IMF noise-added components of any order corresponding to different white noises is calculated as the IMF component of the target difference frequency signal of the order.

8. The method according to claim 1, characterized in that The obtaining of a distance estimation value and a speed estimation value of the radar target based on the first processed signal analysis comprises: Converting the first processed signal from the time domain to the frequency domain to obtain a frequency domain signal; Performing peak detection on the frequency domain signal to obtain a target frequency peak index and a distance frequency domain signal matrix; Calculate the target main frequency corresponding to the radar target according to the frequency peak index; The Doppler estimation is performed using the range-frequency domain signal matrix to obtain a frequency estimation value corresponding to the radar target; and the range estimation value and the speed estimation value corresponding to the radar target are calculated using the target main frequency and the frequency estimation value respectively.

9. The method according to claim 8, characterized in that The Doppler estimation comprises the following steps: Extract the FFT bands corresponding to the distance and construct the covariance matrix; Performing singular value decomposition on the covariance matrix after forward-backward averaging; Extract the noise subspace of all targets obtained after singular value decomposition; Construct the spectral function matrix and solve the roots of the characteristic polynomial, select the root with the largest amplitude, and obtain the peak rate estimate of the target.

Images