A method and apparatus for FMCW radar super-resolution range imaging

Abstract

The application provides a kind of FMCW radar super-resolution distance imaging method and device, it is related to FMCW radar distance imaging technical field, the distance imaging method, comprising: spectrum peak rough estimation, frequency refinement in interest interval, interest interval frequency spectrum initialization, filter weight optimization update, interest interval frequency spectrum update, iteration end condition decision and frequency distance conversion;By constructing the manifold matrix of the frequency domain in interest interval, the frequency estimation of FMCW radar beat frequency signal is carried out using iterative adaptive super-resolution algorithm, a higher resolution range image can be obtained, and the size of the frequency domain in interest interval is reduced by spectrum rough estimation, the algorithm model is simplified, and the operation amount is greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of FMCW radar range imaging technology, and is suitable for FMCW system vehicle-mounted radar, radar liquid level meter, etc., and particularly relates to a FMCW radar super-resolution range imaging method and device. BACKGROUND

[0002] FMCW radar has a wide range of applications in automatic driving vehicle-mounted millimeter wave radar, radar liquid level meter, etc. due to its high integration, low cost, all-weather, all-day work, etc. FMCW radar transmits a continuous wave signal of linear frequency modulation, and the beat signal can be obtained after mixing the target echo with the transmitted signal. The beat signal is a single-frequency signal related to the target distance and speed, and the target distance can be calculated according to the frequency of the beat signal.

[0003] The range resolution of FMCW radar is related to the transmitted signal bandwidth, specifically c / 2B, where c is the electromagnetic wave propagation speed and B is the transmitted signal bandwidth. The wider the bandwidth of FMCW radar, the better the range resolution, and it is easier to distinguish closely adjacent targets. However, due to the shortage of electromagnetic spectrum resources, the available bandwidth of radar is often limited. The FCC certification and CE certification issued by the United States both stipulate that the available frequency range of 24G millimeter wave radar is 24GHz-24.25GHz, and the limit resolution is only 0.6m, which cannot meet the needs of adjacent target resolution in ADAS and automatic driving. Although zero padding on the sampling sequence can refine the spectrum and improve the distance measurement accuracy to some extent, this method is still based on the traditional Fourier transform for spectrum estimation and cannot improve the resolution.

[0004] Literature 1 "Adaptive pulse compression via MMSE estimation" of IEEE Transactions on Aerospace and Electronic System and literature 2 Chinese patent "Range imaging method and range imaging system based on pulse compression technology" with publication number CN106371095A both describe a distance super-resolution method based on iterative adaptive method, but these methods are all for pulse system radar, and the filter weight vector in imaging is the transmitted signal sample. Since FMCW radar does not directly collect and store the transmitted waveform, but mixes the transmitted signal and the echo signal to produce the beat signal, the above method is not applicable to FMCW radar.

[0005] Literature 3 proposes RISR, which can obtain a super-resolution result of spatial spectrum, but there is no literature that applies this method to spectrum analysis, and the basis of traditional RISR super-resolution is that the spectrum line is fine enough, which inevitably increases the great amount of calculation. SUMMARY

[0006] The present application aims to provide a FMCW radar super-resolution range imaging method and device to solve the problems raised in the background.

[0007] To solve the above technical problems, the present application provides the following technical solutions: a FMCW radar super-resolution range imaging method, comprising the following steps:

[0008] Step 1: spectral peak coarse estimation, obtaining the spectral peak position of the beat signal sampling sequence; the beat signal is the signal obtained by mixing the FMCW radar target echo and the transmitted signal, and the beat signal sampling sequence is denoted by s, N is the number of samples, represents a complex matrix of N x 1 dimension, and is a mathematical symbol representing the matrix dimension and type;

[0009] Step 2: frequency refinement in the interest interval, refining the frequencies within a certain range around the spectral peak position obtained by spectral peak coarse estimation, and the frequency points after refinement in the interest interval are L is the number of frequency points after refinement in the interest interval, and forms the popular matrix of the interest interval frequency domain as shown in the following formula:

[0010] A = exp(j2πf xh T t) T (1)

[0011] In formula (1), t is the sampling time sequence, which is a 1 x N vector, j is the complex symbol, and T represents the matrix transpose operator;

[0012] Step 3: interest interval spectrum initialization, using the popular matrix A in step 2 as the filter weight matrix initial value to obtain the refined spectrum initialization result as shown in the following formula:

[0013] y0 = A H s (2)

[0014] In formula (2), H represents the matrix conjugate transpose operator;

[0015] The frequency in the interest interval is the frequency of interest. After spectral peak coarse estimation, it can be roughly known that the target is at what frequency / distance (distance and frequency are one-to-one corresponding), and these potential target frequency / distance ranges are the focus of attention, so the frequency is refined for fine imaging. This step discards the frequencies not of interest, i.e. the frequencies / distance without target, saving computing power.

[0016] Step 4: Filter weight optimization update, update the filter weight based on MMSE criterion, the cost function of MMSE criterion is

[0017] J = E{||y - W H s|| 2} (3)

[0018] In formula (3), E{} represents the expectation operator, y is the spectrum of the beat signal, and W is obtained by deriving formula (3) and setting W equal to zero, and the optimal weight vector is as follows:

[0019] W = (APA H +R v )AP (4)

[0020] In formula (4), P = [yy H ]⊙I L×L ,⊙ represents Hadamard product, I LxL represents an LxL identity matrix, P is obtained from the spectrum y initialized or obtained in the last iteration, and R v is the noise covariance matrix;

[0021] Step 5: Interest interval spectrum update, update the spectrum estimation result using the updated filter weight, as shown in the following formula:

[0022] y = W H s (5)

[0023] Step 6: Iteration end condition judgment, repeat iteration steps 6-7 until the iteration number or end condition is met;

[0024] The imaging effect will be better after each iteration. The iteration end condition is a compromise between effect and real-time performance. If it is considered that several iterations are enough, a fixed iteration number can be set. According to the improvement after iteration, it can be judged whether to terminate the loop iteration. If the iteration this time and the last time only improve a little, it is considered that the performance has basically converged, and there is no need to iterate again.

[0025] Step 7: Frequency distance conversion, according to the corresponding relationship between frequency and distance, the FMCW radar range super-resolution result is obtained, as shown in the following formula:

[0026]

[0027] In formula (6), c is the electromagnetic wave propagation speed, f r is the frequency of FMCW radar, f1, f2,..., f L are different refined frequency points.

[0028] The FMCW radar frequency and distance are one-to-one corresponding, and the corresponding relationship is that the beat signal frequency f corresponds to the target distance After the foregoing iteration, the spectrum, that is, the spectrum amplitude y of different f (refined frequency points f1, f2,..., f L ) is obtained, which is equivalent to the target size at the distance

[0029] Further, the fast Fourier transform (FFT) is used in step 1 to perform spectrum peak rough estimation.

[0030] The present application provides an FMCW radar super-resolution distance imaging device, which comprises a spectrum peak rough estimation module, an interesting interval frequency refinement module, an interesting interval spectrum initialization module, a filter weight optimization update module, an interesting interval spectrum update module, an iteration end condition judgment module and a frequency-distance conversion module.

[0031] The spectrum peak rough estimation module is used to obtain the spectrum peak position of the beat signal sampling sequence; the beat signal is a signal obtained by mixing the FMCW radar target echo and the transmitted signal, and the beat signal sampling sequence is denoted by s, N is the sampling number, represents an N×1-dimensional complex matrix, and is a mathematical symbol representing the matrix dimension and type;

[0032] The interesting interval frequency refinement module is used to refine the frequency within a certain range around the spectrum peak position obtained by the spectrum peak rough estimation, and the refined interesting interval frequency frequency point is L is the number of refined interesting interval frequency frequency points, and forms the popular matrix of the interesting interval frequency domain as shown in the following formula:

[0033] A = exp (j2πf xh T t) T (1)

[0034] In formula (1), t is the sampling time sequence, which is a 1×N vector, j is a complex symbol, and T represents the matrix transposition operator;

[0035] The interesting interval spectrum initialization module uses the popular matrix as the filter weight matrix initial value to obtain the refined spectrum initialization result as shown in the following formula:

[0036] y0 = A H s (2)

[0037] In formula (2), H represents the matrix conjugate transposition operator;

[0038] ​The frequency of the interest interval is the frequency of interest. After rough estimation of the spectrum peak, it can be known approximately that the target is at what frequency / distance (distance and frequency are one-to-one correspondence). The potential target frequency / distance range is the focus, so fine imaging is performed for these frequencies. This step discards the frequencies of no interest, i.e. the frequencies / distance of no target, thereby saving computing power.

[0039] The filter weight optimization updating module updates the filter weight based on the MMSE criterion. The cost function of the MMSE criterion is

[0040] J=E{||y-W H s|| 2} (3)

[0041] In formula (3), E{} represents an expectation operator, y is the frequency spectrum of the beat signal, W is derived using formula (3), and W is equal to zero, to obtain an optimal weight vector, as shown in the following formula:

[0042] W=(APA H +R v )AP (4)

[0043] In formula (4), P=[yy H ]⊙I L×L ,⊙ represents a Hadamard product, I L×L represents an L×L unit matrix, P is obtained from the spectrum y obtained by initialization or the last iteration, and R v is a noise covariance matrix.

[0044] The interest interval spectrum updating module is used to update the spectrum estimation result using the updated filter weight, as shown in the following formula:

[0045] y=W H s (5)

[0046] The iteration end condition judgment module judges whether the iteration number or the iteration end condition is met. If yes, the iteration is stopped, otherwise the filter weight and the spectrum estimation value are updated.

[0047] The frequency distance conversion module is used to obtain the FMCW radar distance super-resolution result according to the corresponding relationship between the frequency and the distance, as shown in the following formula:

[0048]

[0049] In formula (6), c is the electromagnetic wave propagation speed, f r is the frequency of the FMCW radar, f1, f2, …, f L are different refined frequency points.

[0050] The FMCW radar frequency and distance are one-to-one correspondence, and the correspondence relation is that the beat signal frequency f corresponds to the target distance After the foregoing iteration, the spectrum is obtained, that is, the spectrum amplitude y of different f (refined frequency points f1, f2,..., f L ) is obtained, which is equivalent to the target size at the distance

[0051] The application provides a computer readable access medium, and a computer program is stored on the computer readable access medium.

[0052] Further, the spectrum peak rough estimation module performs spectrum peak rough estimation by using fast Fourier transform (FFT).

[0053] Compared with the prior art, the application has the beneficial effects that:

[0054] 1. The application applies the iterative adaptive super-resolution algorithm to the frequency estimation of the FMCW radar beat signal by constructing the frequency domain manifold matrix, and in combination with the refined frequency interval, can effectively suppress the estimated value of the non-target distance unit / frequency unit, reduce the influence of the strong target distance sidelobe on the adjacent small target, and obtain a higher resolution distance image.

[0055] 2. The application determines the rough position of the target through spectrum rough estimation, performs frequency refinement only near the rough position, greatly reduces the size of the frequency domain of the interest interval, reduces the matrix dimension of the inverse operation and the multiplication and addition operation, and greatly reduces the operation amount.

[0056] The accompanying drawings further illustrate the concept, specific structure and technical effects of the application, and fully understand the purpose, features and effects of the application. DETAILED DESCRIPTION

[0057] Figure 1 A step flow chart of the FMCW radar super-resolution distance imaging method provided by the application is provided.

[0058] Figure 2 The distance image result of the point target (distance 30 m, amplitude 1, and SNR=10 dB) in embodiment one is shown in the following table.

[0059] Figure 3 The distance image result of the two adjacent targets (distances 29.7 m and 30 m, scattering point amplitudes 0.1 and 1, and SNR=10 dB) in embodiment two is shown in the following table.

[0060] Figure 4 ​The range image result of the approximately continuous target in a certain distance (30-45m, there is a strong scattering point every 0.6m, the amplitude of the scattering points is 1, and the SNR is 10dB) in Example 3. DETAILED DESCRIPTION

[0061] The application will be further described below in connection with specific embodiments. It should be understood that these embodiments are only used for illustrating the application and not used for limiting the scope of the application. Furthermore, it should be understood that after reading the content taught by the application, those skilled in the art can make various modifications or changes to the application, and these equivalent forms also fall within the scope defined by the appended claims.

[0062] In the drawings, components of the same structure are denoted by the same reference numerals, and components similar in structure or function are denoted by similar reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the application is not limited to the size and thickness of each component. In order to make the drawing clearer, the thickness of some components is appropriately exaggerated in some places in the drawing.

[0063] As shown in FIG. 1, the FMCW radar provided by the application comprises a signal generator 1, a signal transmitter 2, a signal receiver 3, a signal mixer 4, a signal processor 5 and a display device 6. Figure 1 FIG. 2 shows a step flowchart of the FMCW radar super-resolution distance imaging method provided by the application. The embodiments provided by the application are all performed according to the step flow shown in FIG. 2. Figure 1

[0064] The application provides a process for performing super-resolution processing on the beat signal of the FMCW radar. The FMCW radar transmits a linear frequency modulation signal, the frequency range is 77.0GHz-77.3GHz, the frequency modulation rate f is 10MHz / us, the bandwidth is 300MHz, the single transmission signal pulse width is 30us, the radar receives the signal and mixes the transmission signal to obtain the beat signal, the sampling rate of the beat signal is 30Msps, and the sampling sequence of the beat signal is r y is the actual spectrum of the beat signal, and is specifically as follows:

[0065]

[0066] In formula (7), y(f) is the signal amplitude of the frequency component f, and L0 is the number of all frequency components constituting the beat signal. l l

[0067] The prevalence matrix of the spectrum of the region of interest is defined as follows:

[0068]

[0069] t is a sampling period; s

[0070] ​​​​​The beat signal sampling sequence is denoted as s as follows:

[0071] s=Ay+v (8)

[0072] In formula (8), v is an N x 1 noise sequence.

[0073] The FMCW radar super-resolution distance imaging method steps are as follows:

[0074] Step 1: spectrum peak coarse estimation, using the mip core of the FPGA to realize the FFT of the above beat signal sampling sequence s, to obtain the coarse spectrum of the beat signal The CFAR detection is used to search the spectrum peak positions of the spectrum of the beat signal, f1, f2,..., f K , K is the number of spectrum peaks;

[0075] Step 2: frequency refinement in the interested interval, the frequencies within a certain range around the spectrum peak positions obtained by the spectrum peak coarse estimation are refined by 10 times, and the frequency points of the refined frequency in the interested interval are L is the number of frequency points of the refined frequency in the interested interval,

[0076] f xh =[f1-3Δf:Δf / 10:f1+3Δf,f2-3Δf:Δf / 10:f2+3Δf,...,f K -3Δf:Δf / 10:f K +3Δf] (9)

[0077] In formula (9), Δf is the discrete frequency interval in the coarse spectrum, Δf=1 / T r , T r is the single transmission signal pulse width.

[0078] In this embodiment, when the frequencies within a certain range around the spectrum peak positions obtained by the spectrum peak coarse estimation are refined, the frequencies of 3Δf on the left and right of the spectrum peak corresponding to the coarse estimation are refined by 10 times, or the frequencies of 1Δf, 2Δf or 4Δf on the left and right of the spectrum peak corresponding to the coarse estimation can also be refined.

[0079] The frequency points in f xh are processed to remove the overlapping frequency points. The flow matrix of the frequency domain in the interested interval is formed as follows:

[0080] A=exp(j2πf xh T t) T (1)

[0081] In formula (1), t is the sampling time sequence, which is a 1 x N vector, j is a complex symbol, and T represents the matrix transposition operator.​

[0082] Step 3: Interest interval spectrum initialization, using the popular matrix A in step 2 as the filter weight matrix initial value to obtain the refined spectrum initialization result As shown in the following formula:

[0083] y0=A H s (2)

[0084] In formula (2), H represents the matrix conjugate transpose operator;

[0085] Step 4: Filter weight optimization update, using the initialization result and the noise covariance matrix, the filter weight is updated based on the MMSE criterion, and the cost function of the MMSE criterion is

[0086] J=E{||y-W H s|| 2} (3)

[0087] In formula (3), E{} represents the expectation operator, y is the spectrum of the beat signal, and the optimal weight vector is obtained by deriving W using formula (3) and setting W equal to zero, as shown in the following formula:

[0088] W=(E{ss H}) -1 E{sy H} (9)

[0089] Bring formula (8) s=Ay+v into formula (9), and after simplification, we get

[0090] W=(APA H +R v )AP (4)

[0091] In formula (4), P=[yy H ]⊙I L×L ,⊙ represents Hadamard product, I L×L represents an L×L unit matrix, P is obtained from the spectrum y obtained by initialization or the last iteration, R v is the noise covariance matrix, R v =var·I N×N ;

[0092] Step 5: Interest interval spectrum update, using the updated filter weight to update the spectrum estimation result, as shown in the following formula:

[0093] y=W H s (5)

[0094] Step 6: Iteration end condition judgment, repeat steps 6-7 until the iteration number or end condition is met;

[0095] Step 7: Frequency-distance conversion: According to the corresponding relationship between frequency and distance, the FMCW radar range super-resolution result is obtained, as shown in the following formula:

[0096]

[0097] In formula (6), c is the propagation speed of electromagnetic waves, f r is the frequency modulation of the FMCW radar, f1, f2, ..., f L For different refined frequency points.

[0098] The FMCW radar frequency and distance are one-to-one corresponding. The corresponding relationship is: the beat signal frequency f corresponds to the target distance After the previous iteration, the spectrum is obtained, which is different f (refined frequency points f1, f2, ..., f L )’s spectrum amplitude y, which is equivalent to the distance The target size at .

[0099] Example 1

[0100] like Figure 2 As shown in FIG, using the above-mentioned FMCW radar super-resolution range imaging method, the point target range image result is given under the conditions of a distance of 30m, a scattering point amplitude of 1, and an SNR of 10dB.

[0101] Example 2

[0102] like Figure 3 As shown in FIG, using the above-mentioned FMCW radar super-resolution range imaging method, the range image results of two adjacent targets (the distances are 29.7m and 30m respectively, the scattering point amplitudes are 0.1 and 1, and the SNR is 10dB) are given.

[0103] Example 3

[0104] like Figure 4 As shown in FIG, using the above-mentioned FMCW radar super-resolution range imaging method, the range image results of an approximately continuous target within a range (distance 30 to 45 m, there is a strong scattering point every 0.6 m, the scattering point amplitude is 1, and SNR = 10 dB) are given.

[0105] Figures 2-4In the method, the line corresponding to the FFT result is the output result of step 1, the line corresponding to the initialization result is the output result of step 3, and the line corresponding to the iterative adaptive result is the result after the iteration of step 6 is completed; the narrower and sharper the peak at the target distance (abscissa) is, the better the resolution is; when two targets are very close, two independent peaks can be distinguished, which indicates that the resolution is good; therefore, it can be concluded that the iterative adaptive distance imaging method used in the application for frequency estimation of the beat frequency signal of the FMCW radar can obtain a distance image with higher resolution.

[0106] In the first to third embodiments, the number of frequency points in the full frequency domain before refinement is 900, which is refined to 9000 according to 10 times; after the spectrum rough estimation, the number of frequency refinement points in the interest region of the first, second and third embodiments is 60, 60 and 360 respectively. It can be seen that after the spectrum rough estimation, the matrix dimension involved in the operation in the iterative adaptive process can be greatly reduced, the operation burden is reduced, and the calculation speed of the algorithm is improved.

[0107] Example 4

[0108] The application provides an FMCW radar super-resolution distance imaging device, which comprises a spectrum peak rough estimation module, an interest region frequency refinement module, an interest region spectrum initialization module, a filter weight optimization updating module, an interest region spectrum updating module, an iteration end condition judgment module and a frequency distance conversion module.

[0109] The spectrum peak rough estimation module is based on an fft IP core and a CFAR module provided by FPGA, and is used for performing FFT on the beat signal to obtain a rough spectrum of the beat signal, and searching for a spectrum peak position of the spectrum of the beat signal by using CFAR detection. N is a sampling number, represents a complex matrix of N*1 dimensions, and is a mathematical symbol representing matrix dimensions and types.

[0110] The interest region frequency refinement module is used for refining frequencies within a certain range around the spectrum peak position obtained by the spectrum peak rough estimation, and the frequency points of the refined interest region frequency are L is the number of frequency points of the refined interest region frequency, and a popular matrix of the interest region frequency domain is formed. As shown in the following formula:

[0111] A = exp (j2pi f xh T t) T (1)

[0112] In formula (1), t is a sampling time sequence, is a 1*N vector, j is a complex symbol, and T represents a matrix transposition operator.

[0113] The interest interval spectrum initialization module uses the popular matrix as an initial value of a filter weight matrix to obtain a refined spectrum initialization result As shown in the following formula:

[0114] y0=A H s (2)

[0115] In formula (2), H represents a matrix conjugate transpose operator;

[0116] The interest interval frequency is a frequency of interest. After rough estimation of the spectrum peak, it can be known approximately that the target is at what frequency / distance (distance and frequency are one-to-one correspondence). The potential target frequency / distance range is the focus, so the frequency is refined for fine imaging. This step discards the uninterested frequency, i.e. the target-free frequency / distance, and saves the computing power.

[0117] The filter weight optimization updating module updates the filter weight based on the MMSE criterion. The cost function of the MMSE criterion is

[0118] J=E{||y-W H s|| 2} (3)

[0119] In formula (3), E{} represents an expectation operator, y is the spectrum of the beat signal, W is derived using formula (3), and W is equal to zero, to obtain an optimal weight vector, as shown in the following formula:

[0120] W=(APA H +R v )AP (4)

[0121] In formula (4), P=[yy H ]⊙I L×L , ⊙ represents a Hadamard product, I L×L represents an L×L unit matrix, P is obtained from the spectrum y obtained by initialization or the last iteration, and R v is a noise covariance matrix;

[0122] The interest interval spectrum updating module is used to update the spectrum estimation result using the updated filter weight, as shown in the following formula:

[0123] y=W H s (5)

[0124] The iteration end condition judgment module judges whether the iteration number or the iteration end condition is satisfied. If yes, the iteration is stopped, otherwise the filter weight and the spectrum estimation value are updated iteratively;

[0125] A frequency distance conversion module is configured to obtain a FMCW radar distance super-resolution result according to a corresponding relationship between the frequency and the distance, as shown in the following formula:

[0126]

[0127] In formula (6), c is the electromagnetic wave propagation speed, f r is the frequency modulation of the FMCW radar, f1, f2, …, f L are different refined frequency points.

[0128] The FMCW radar frequency and the distance are one-to-one corresponding, and the corresponding relationship is that the beat signal frequency f corresponds to the target distance After the foregoing iteration, the frequency spectrum is obtained, that is, the frequency spectrum amplitude y of different f (the refined frequency points f1, f2, …, f L ) is obtained, which is equivalent to the target size at the distance .

[0129] The application further provides a computer readable access medium, and the computer readable access medium stores a computer program. Figure 1 The computer program is executed by a processor to implement the FMCW radar super-resolution distance imaging method shown in the application.

[0130] The above-described embodiments only express several embodiments of the application, and the description is relatively specific and detailed, but it cannot be understood as a limitation on the patent scope of the application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which all belong to the protection scope of the application. Therefore, the protection scope of the patent of the application should be subject to the appended claims.

Claims

1. A FMCW radar super-resolution range imaging method, characterized in that: The following steps are involved: Step 1: Roughly estimate the spectrum peak to obtain the spectrum peak position of the beat signal sampling sequence; the beat signal is the signal obtained by mixing the FMCW radar target echo and the transmission signal. The beat signal sampling sequence is represented by s. N is the number of samples, represents an N×1 dimensional complex matrix; Step 2: Refine the frequency of the interest interval. Refine the frequency within a certain range around the peak position obtained by rough estimation of the peak. The refined frequency points of the interest interval are L is the number of frequency points in the refined interval of interest, and forms the popular matrix of the frequency domain of the interval of interest As shown in the following formula: A=exp(j2πf xh T t) T (1) In formula (1), t is the sampling time sequence, which is a 1×N vector, j is a complex symbol, and T represents the matrix transpose operator; Step 3: Initialize the spectrum of the interval of interest, using the popular matrix A in step 2 as the initial value of the filter weight matrix to obtain the refined spectrum initialization result As shown in the following formula: y0=A H s (2) In formula (2), H represents the matrix conjugate transpose operator; Step 4: Optimize and update the filter weights. Update the filter weights based on the MMSE criterion. The cost function of the MMSE criterion is: J=E{||y-W H s|| 2 } (3) In formula (3), E{} represents the expectation operator, y is the spectrum of the beat signal, and the optimal weight vector is obtained by taking the derivative of W using formula (3) and setting W equal to zero, as shown in the following formula: W=(WHAT H +R v )AP (4) In formula (4), P = [yy H ]⊙I L×L , ⊙ represents the Hadamard product, I L×L represents the L×L identity matrix, P is obtained by initialization or the spectrum y obtained in the previous iteration, R v is the noise covariance matrix; Step 5: Update the spectrum of the interval of interest and use the updated filter weights to update the spectrum estimation result, as shown in the following formula: y=W H s (5) Step 6: Determine the end condition of the iteration and repeat steps 6 to 7 until the number of iterations or the end condition is met; Step 7: Frequency-distance conversion: According to the corresponding relationship between frequency and distance, the FMCW radar range super-resolution result is obtained, as shown in the following formula: In formula (6), c is the propagation speed of electromagnetic waves, f r is the frequency modulation of the FMCW radar, f1, f2, ..., f L For different refined frequency points.

2. The FMCW radar super-resolution range imaging method according to claim 1, wherein: In step 1, the fast Fourier transform (FFT) is used to perform rough estimation of the spectrum peak.

3. The FMCW radar super-resolution range imaging method according to claim 1, wherein: The beat signal sampling sequence is expressed as s as follows: s=Ay+v (8) In formula (8), v is an N×1 noise sequence.

4. The FMCW radar super-resolution range imaging method according to claim 3, wherein: In step 4, use formula (3) to derive W and set W equal to zero to obtain the optimal weight vector, as shown below: W=(E{ss H }) -1 Easy H } (9) Substituting equation (8) s=Ay+v into equation (9), we can obtain W=(WHAT H +R v )AP (4) R v is the noise covariance matrix, R v =var·I N×N .

5. An FMCW radar super-resolution range imaging device, characterized in that: It includes spectrum peak rough estimation module, interest interval frequency refinement module, interest interval spectrum initialization module, filter weight optimization and update module, interest interval spectrum update module, iteration end condition judgment module and frequency distance conversion module; The spectrum peak rough estimation module is used to obtain the spectrum peak position of the beat signal sampling sequence; the beat signal is the signal obtained by mixing the FMCW radar target echo and the transmission signal, and the beat signal sampling sequence is represented by s. N is the number of samples, represents an N×1 dimensional complex matrix; The frequency refinement module of interest interval is used to refine the frequency within a certain range around the peak position obtained by rough estimation of the peak. The refined frequency points of the interest interval are L is the number of frequency points in the refined interval of interest, and forms the popular matrix of the frequency domain of the interval of interest As shown in the following formula: A=exp(j2πf xh T t) T (1) In formula (1), t is the sampling time sequence, which is a 1×N vector, j is a complex symbol, and T represents the matrix transpose operator; The spectrum initialization module of the interest interval uses the popular matrix as the initial value of the filter weight matrix to obtain the refined spectrum initialization result As shown in the following formula: y0=A H s (2) In formula (2), H represents the matrix conjugate transpose operator; The filter weight optimization update module updates the filter weight based on the MMSE criterion. The cost function of the MMSE criterion is J=E{||y-W H s|| 2 } (3) In formula (3), E{} represents the expectation operator, y is the spectrum of the beat signal, and the optimal weight vector is obtained by taking the derivative of W using formula (3) and setting W equal to zero, as shown in the following formula: W=(WHAT H +R v )AP (4) In formula (4), P = [yy H ]⊙I L×L , ⊙ represents the Hadamard product, I L×L represents the L×L identity matrix, P is obtained by initialization or the spectrum y obtained in the previous iteration, R v is the noise covariance matrix; The spectrum update module of the interest interval is used to update the spectrum estimation result using the updated filter weights, as shown in the following formula: y=W H s (5) The iteration end condition judgment module determines whether the number of iterations or the iteration end condition is met. If yes, the iteration is stopped; otherwise, the iteration continues to update the filter weights and spectrum estimation values. The frequency-distance conversion module is used to obtain the FMCW radar range super-resolution result based on the correspondence between frequency and distance, as shown in the following formula: In formula (6), c is the propagation speed of electromagnetic waves, f r is the frequency modulation of the FMCW radar, f1, f2, ..., f L For different refined frequency points.

6. The FMCW radar super-resolution range imaging device according to claim 5, wherein: The spectrum peak rough estimation module uses fast Fourier transform FFT to perform spectrum peak rough estimation.

7. The FMCW radar super-resolution range imaging device according to claim 5, wherein: The beat signal sampling sequence is expressed as s as follows: s=Ay+v (8) In formula (8), v is an N×1 noise sequence.

8. The FMCW radar super-resolution range imaging device according to claim 7, wherein: The filter weight optimization update module uses formula (3) to derive W and sets W equal to zero to obtain the optimal weight vector, as shown below: W=(E{ss H }) -1 Easy H } (9) Substituting equation (8) s=Ay+v into equation (9), we can obtain W=(What H +R v )AP (4) R v is the noise covariance matrix, R v =var·I N×N .

9. A computer-readable storage medium, characterized in that: A computer program is stored on a computer-readable medium, and when the computer program is executed by a processor, the FMCW radar super-resolution range imaging method according to any one of claims 1 to 4 is implemented.

Images