K-band-based close-range end-to-end sparse MIMO FMCW radar system

Abstract

The invention discloses a K-band-based close-range end-to-end sparse MIMO FMCW radar system. The method comprises the following steps: performing range-fast Fourier transform on an intermediate frequency signal acquired by a sparse MIMO FMCW millimeter wave radar system to obtain a frequency domain signal; after strong leakage interference targets in the frequency domain signals are filtered through a frequency domain leakage filtering technology and intermediate frequency signals of interested targets are reserved, speed-fast Fourier transform is carried out to obtain a distance-Doppler graph, and distance and speed information of the interested targets is extracted; performing sparse array traditional beam forming on the obtained target signal to estimate angle information of the target, and matching angle and distance information by using a digital beam forming technology to obtain the distance, the speed and the angle of each interested target so as to form a spatial point cloud; and finally, for multi-frame acquisition of the spatial point clouds formed by the above steps, carrying out angle averaging on the point clouds at the same distance to smooth the trajectory, and forming the motion point cloud trajectory of the target. The method is suitable for synchronous and asynchronous systems, solves the existing multipath leakage problem, and is high in robustness.

Description

Technical Field

[0001] The present invention relates to an accurate point cloud-level multi-target trajectory tracking technology, specifically a K-band short-range end-to-end sparse multiple-input multiple-output (MIMO) frequency-modulated continuous-wave (FMCW) radar system. Background Art

[0002] Existing multiple-input multiple-output (MIMO) frequency-modulated continuous-wave (FMCW) radar systems usually exhibit performance degradation in short-range scenarios, mainly due to the limitations of existing hardware and signal processing links. The existing uniform array arrangement results in a low spatial resolution of the radar system, making it impossible to distinguish targets at short distances. In addition, when the existing signal processing link is applied to short-range scenarios, it will be affected by the transceiver coupling signal, resulting in strong interference targets and unable to accurately detect targets. Summary of the Invention

[0003] Aiming at the deficiency that the prior art cannot achieve multi-target point cloud-level trajectory tracking in short-range scenarios based on millimeter-wave radar, the present invention proposes a K-band short-range end-to-end sparse MIMO FMCW radar system, which eliminates false targets caused by the leakage of the transmitter (Tx) to the receiver (Rx) through frequency-domain leakage cancellation technology, so that the distance, speed, and angle of human targets can be accurately obtained, and then a trajectory point cloud is formed. The invention is applicable to synchronous and asynchronous systems, and solves the existing multi-path leakage problem, with strong robustness.

[0004] The present invention is realized through the following technical solutions:

[0005] The present invention relates to a method for indoor multi-target point cloud trajectory tracking of a sparse MIMO FMCW millimeter-wave radar, including:

[0006] Step 1) Perform range-fast Fourier transform on the intermediate-frequency signal collected by the sparse MIMO FMCW millimeter-wave radar system to obtain a frequency-domain signal;

[0007] Step 2) Use frequency-domain leakage filtering technology to filter out strong leakage interference targets in the frequency-domain signal and retain the intermediate-frequency signal of the target of interest;

[0008] Step 3) Perform velocity-fast Fourier transform on the intermediate-frequency signal of the target of interest to obtain a range-Doppler map and extract the range and velocity information of the target of interest;

[0009] Step 4) Perform sparse array conventional beamforming on the obtained target signal to estimate the angle information of the target, and use digital beamforming technology to match the angle and range information, so as to obtain the range, velocity, and angle of each target of interest, and then form a spatial point cloud;

[0010] Step 5) For the spatial point cloud formed through the above steps in multi-frame acquisition, the angles of the point clouds at the same distance are averaged to smooth the trajectory, forming the moving point cloud trajectory of the target. Technical Effects

[0011] The present invention pre-collects against leakage signals and processes the actual signals using the frequency domain characteristics of intermediate frequency signals, solving the problem that existing leakage filtering technologies cannot simultaneously solve the multi-path leakage problems of synchronous and asynchronous systems, and being applicable to dynamic and static targets. Furthermore, accurate human spatial signals are obtained to generate human trajectory point clouds. Compared with the prior art, the present invention effectively reduces leakage, and the signal-to-interference ratio and signal-to-noise ratio are increased by 25 dB and 15 dB respectively. It is applicable to synchronous and asynchronous systems, and has high robustness and generality, capable of solving the multi-path leakage problem. The 3 dB beamwidth of the system is 8.97°, and the measured angular resolution is 13.8°, which are increased by 29.8% and 32.8% respectively compared with existing uniform arrays. At the same time, it reaches a sidelobe level of -10 dB and a field of view of 100°. Compared with the existing system framework, the system is more accurate in the multi-target trajectory point cloud level detection performance, especially significant in short-distance applications. Description of the Drawings

[0012] Figure 1 is a flowchart of the present invention;

[0013] Figure 2 is a schematic diagram of the sparse MIMO FMCW radar system of the present invention;

[0014] Figure 3 is a comparison diagram of the spatial resolution between the system of the present invention and the existing system;

[0015] Figure 4 is a diagram of the point cloud experimental results of the present invention;

[0016] In the figure, (a)-(d) represent different experimental groups, (a-1), (b-1), (b-4), (b-6), (c-1) and (c-4) represent different test scenarios, (a-2), (b-2) and (c-2) represent the original point clouds, and other images show the trajectory results obtained by averaging the point clouds at the same distance. Detailed Embodiments

[0017] Such as Figure 2As shown in the figure, a sparse MIMO FMCW millimeter-wave radar indoor multi-target point cloud trajectory tracking system designed in this embodiment includes: a voltage-controlled oscillator chip 1 (VCO), a phase-locked loop chip 2 (PLL), a receiving chip 3 (Receiver), a micro-control unit 4 (MCU), baseband amplifiers 5, an analog-to-digital converter 6, and a signal processing module 7. Among them: The voltage-controlled oscillator chip 1 and the phase-locked loop chip 2 work together under the configuration of the micro-control unit 4 to generate a chirp signal (Chirp), which is transmitted into space by the transmitting antenna in the sparse array. When the signal encounters a target, it is reflected and received by the receiving antenna in the sparse array and enters the receiving chip 3. An intermediate frequency signal is generated through mixing and a low-pass filter. After passing through the baseband amplifiers 5, the signal is fully amplified and collected by the analog-to-digital converter 6 and transmitted to a computer for subsequent processing in the signal processing module 7.

[0018] The signal processing module 7 described above includes: a distance processing unit, a frequency-domain leakage filtering unit, a distance-Doppler processing unit, a sparse array angle-solving unit, an angle-distance matching unit, and a point cloud trajectory output unit. Among them: The distance processing unit is used to process the original signal collected by the ADC to obtain spectral information; the frequency leakage filtering unit is used to filter out strong leakage interference targets in the frequency domain and retain the targets of interest. The distance-Doppler processing unit is used to solve the distance-Doppler map of the retained targets of interest to obtain distance and velocity information. The sparse array angle-solving unit is used to solve the spatial angle information of the targets of interest. The angle-distance matching unit is used to match the distance and angle information of the targets to achieve distance-velocity-angle matching alignment. The point cloud trajectory output unit is used to output the trajectory point clouds of each target.

[0019] The sparse array is arranged as follows: The distance between TX1 and TX2 is 2.77λ, and the distances between RX2, RX3, RX4 and RX1 are 1.06λ, 1.68λ, and 2.23λ respectively.

[0020] As Figure 1 shown, this embodiment designs a multi-target point cloud trajectory generation method based on the above system, including:

[0021] Step 1: Perform a distance-fast Fourier transform on the signals collected by the sparse MIMO FMCW radar, specifically including: Considering the MIMO FMCW radar signal model from single-channel to multi-channel, while maintaining theoretical rigor, simplifies the expression of the equation. The modulation signal transmitted by the radar where: t is the fast time, A is the amplitude of the signal, f c is the center frequency, the modulation slope γ = B / T c , B is the bandwidth, T c is the pulse repetition time (PRT), is the initial phase. After the signal target is scattered, the received echo signal S Rx (t) = σS Tx (t - Δt), where: σ is the attenuation factor, related to factors such as the propagation medium and the radar cross section (RCS), Δt is the time delay, given by Δt = 2R(τ) / c, R(τ) = R + vτ is the radial distance between the target and the radar, τ is the "slow time", and v and c are the target velocity and the speed of light respectively. By mixing the received signal with the local oscillator (LO) signal and passing it through a low-pass filter (LPF), an intermediate frequency (IF) signal is obtained where: 4πγR 2 (τ) / c 2 is the residual video phase (RVP), which can be ignored, and λ is the wavelength.

[0022] The intermediate frequency (IF) signal obtained by the described FMCW radar is usually stored in a two-dimensional data matrix, where: the number of rows corresponds to the total number N of the transmitted chirp signals chirp , and the number of columns is the number of sampling points N of each IF signal sampling , and these sampling points are determined by the sampling rate and the pulse repetition time (PRT) of the system. Performing a one-dimensional fast Fourier transform (1DFFT) along the fast time dimension can extract the distance of the target where: sinc(x) = sin(x) / x, f r is the frequency parameter. For each row of S 1DFFT , the spectrum of the processed intermediate frequency (IF) signal can be obtained. When f r = 2γR / c, a peak will appear, corresponding to the frequency of the IF signal. It can be observed that this value is related to the distance of the target, and the distance R = cf r / 2γ is determined.

[0023] In an actual scenario, the radar system will inevitably acquire signals not only from ideal targets but also strong leakage signals. These leakage signals are directly coupled from the transmitter (TX) to the receiver (RX), with extremely small energy loss and usually much stronger than the signals reflected by the targets.

[0024] Step 2: Use the frequency domain leakage filtering technique to filter out strong leakage interference targets and retain the targets of interest, specifically including:

[0025] 2.1 Model the actual signal containing strong leakage signals and target signals as: where: α and β represent the amplitudes of the leakage signal and the target signal respectively, and M and N represent the number of leakage signals and targets respectively.

[0026] 2.2 Collect the anti-leakage signal S in a darkroom environment in advance Anti-leakage , the anti-leakage signal only contains the leakage signal because it is collected in a dark room, and then the leakage is filtered out in the frequency domain, that is, the anti-leakage signal S Anti-leakage and the actual signal S ALL Process it and convert it into the frequency domain. It should be emphasized that the velocity associated with the leakage signal is neglected because it does not exist. In addition, the amplitude coefficient α m Usually it remains constant, and in the case of slight fluctuations in energy, the amplitude can be easily adjusted to achieve alignment. In addition, the above two equations are divided in the frequency domain and applied to each frequency point to obtain After analysis, it is found that for the frequency point corresponding to the leakage signal, only the asynchronous phase component remains, and the target signal is the sum of the anti-leakage signal divided by . In order to simplify the subsequent derivation to remove interference, the denominator is combined into A Total (f r )exp(jζ Total (f r )).

[0027] 2.3 For each frequency point, extract the phase and Division Subtract from, specifically: For synchronous systems, it is natural to observe that after the division, only the amplitude remains, so simply subtracting 1 eliminates the leakage signal. The result contains only the signal corresponding to the target. It is well known that while the change in amplitude is irrelevant to the extraction of the target spatial parameters, the change in phase has a significant impact. Therefore, in order to accurately compensate for these phase differences, S Subrtaction Must be with S Anti-leakage Multiplying together can be expressed as: The amplitude components are combined into A Final (f r ), replacing the overall amplitude expression. Any fluctuations in this component have no effect on subsequent processing, ensuring that the integrity of the spatial parameters is maintained. At this point, the leakage signal is effectively eliminated while retaining the phase of the target signal, with only a slight change in amplitude, which has little impact on information extraction. More importantly, this technique is applicable to both synchronous and asynchronous systems and fully solves the problem of multipath leakage. After the leakage is eliminated, the target signal becomes more significant, making it clearly visible in the range spectrum. In order to obtain the trajectory point cloud of the target, it is necessary to further process S Final , extract speed and orientation, thereby achieving multi-dimensional point cloud imaging.

[0028] Step 3: Perform a velocity - Fast Fourier Transform on the target signal to obtain the distance and velocity of the target, specifically including:

[0029] 3.1 Perform a Fast Fourier Transform on S Final in the slow - time dimension to obtain the Doppler frequency domain at different distances, form a range - Doppler map, and use the Constant False Alarm Rate (CFAR) technique to extract the target, obtaining the distance and velocity of the target.

[0030] 3.2 Achieve azimuth angle estimation by analyzing the phase relationship between virtual channels: When there are P transmitting antennas and Q receiving antennas, and they are all horizontally arranged, and the first antenna in each array is designated as the reference element. The spacings between subsequent elements and the reference element are represented as d tp and d rq . Therefore, the steering vectors of the transmitting - end and receiving - end arrays can be reformulated as Next, perform azimuth angle estimation. The steering vector of the virtual array is derived based on the sparse array arrangement, specifically as where: The symbol represents the Kronecker product. Fully considering the phase change caused by the switching delay of the Time - Division Multiplexing (TDM) model and the velocity of the target, the sparse MIMO signal model can be expressed as: In the system of the present invention, before each transmitting antenna (TX) switches to the next antenna, it transmits all the pulse signals (chirps) within a single frame. Therefore, a compensation factor N chirp is required in the phase. By multiplying each Doppler cell by exp(-j(4πv n / λ)T c (P tp - 1)N chirp ), the phase change is compensated. Based on the detected target, the corresponding velocity v n is selected to extract the vector corresponding to 2v n / λ of the range - Doppler matrix. This is a two - dimensional matrix containing the virtual channel dimension and the distance dimension, with dimensions P×Q and N sampling , respectively. It should be noted that this matrix may contain multiple targets sharing the same velocity. Therefore, it is necessary to match the angles and distances of these targets to ensure accurate estimation. Finally, the snapshot matrix is combined with the steering vector and processed using the existing Capon Beamforming (CBF), specifically as: This process generates an angular spectrum to estimate the azimuth angle, which is crucial because the spatial spectrum cannot be directly obtained by FFT for a sparse array. Here, H represents the conjugate transpose.

[0031] Step 4: Solve the angular solution of the sparse array for the target and match the range-angle to form the target point cloud, specifically including: On the selected velocity dimension, use digital beamforming (DBF) to match the ranges and angles of multiple targets. The snapshot matrix is weighted to point to the angles of different targets, thereby obtaining the range spectrum for each angle. By matching the ranges detected by different angle-range spectra and the ranges detected at the selected velocity in the range-Doppler map, the velocity-angle-range matching of each target is achieved.

[0032] Step 5: Take the mean of the point cloud information obtained from multiple-frame acquisitions at the same range to form the target point cloud trajectory, specifically including: Process multiple-frame data to obtain a large amount of well-matched point cloud information of multiple targets. The point clouds of each target at the same range are averaged in the angular dimension to represent the point cloud at that range. Since the human body is a multi-scattering target, different angular solutions will be generated in different frames at the same range. By taking the average, the generation of high-quality point clouds can be maintained, and the clarity and smoothness of the trajectory can be enhanced.

[0033] As Figure 3 shown, it is the performance comparison between the designed sparse MIMO system and the existing uniform MIMO system. In this evaluation, the proposed 2T4R sparse MIMO FMCW radar system is compared with the 2T4R uniform MIMO FMCW radar system based on the commercial Iclegend Micro ICL1122 radar module. Both systems are used to detect two corner reflector targets near 0° to evaluate the resolution and sidelobe level (SLL). The comparison of the existing beamforming (CBF) results shows that the uniform array only achieves a resolution of 20.7°, while the proposed system shows an improved resolution of 13.8° and maintains a sidelobe level of -10 dB. Although the actual resolution is slightly lower than the 3 dB beamwidth, this is understandable because factors such as manufacturing tolerances and chip performance have an impact. However, the system is significantly superior to the uniform array radar system in terms of performance.

[0034] As Figure 4 shown, it is the actual experimental results of the end-to-end sparse MIMO FMCW radar system framework. Specifically: To verify the effectiveness of leakage interference removal, a single-target person walks towards the radar at an oblique angle relative to the radar's radial axis from 2 meters away, as Figure 4 (a-1) shown. Figure 4(a-2) and (a-3) show the spatial parameter identification of the target and its historical trajectory, confirming the system's ability to accurately track and identify the target in a spatial environment. In contrast, when the leakage is not removed, only the interfering target can be detected, and all the point clouds gather at the same position, as shown in Figure 4 (a-4). Additionally, it is worth noting that due to the phase shift introduced by the target angle offset, the interfering target also has an angular component, which significantly increases the possibility of misinterpretation. These results highlight the crucial role of the proposed system in short-distance, point-cloud-level applications, with the processed signal-to-interference ratio and signal-to-noise ratio improved by 25 dB and 15 dB respectively. To verify the advantage of the proposed system over existing uniform arrays in point-cloud-level applications, experiments were conducted, including closely spaced targets such as straight gait and crossing gait patterns. For comparison, a commercial radar module with a uniform array configuration was used, as shown in Figure 4 (b-1), (b-4), and (b-6). To ensure a fair comparison, special attention was paid to having the two radar systems collect data along nearly the same trajectory. In the first experiment, two targets walked towards the radar along a straight line with a small angular separation, located on both sides of the radar's radial axis, with an angular deviation of approximately 13.8°. Considering the width of the human body, this angle is very close for two human targets. Data were collected using the proposed system and processed through the signal processing chain to obtain the point-cloud trajectories shown in Figure 4 (b-2) and (b-3). It can be clearly seen from the gait trajectories that both targets move at a constant angle close to 14° to 17°, and the system can accurately track their historical gait patterns, including distance, speed, and azimuth angle. Additionally, the complete point-cloud data clearly show that two human targets moving at such a small angle almost overlap, verifying the system's effectiveness in detecting closely spaced human targets at short distances without being interfered by leakage signals. In contrast, when the angular separation between the two targets is approximately 21°, one person moves away while the other remains stationary. Data were collected using the uniform array system and the proposed signal processing chain was applied to the data, as shown in Figure 4 (b-6). Although the movement trajectories of both people are clearly identified, this larger separation angle reflects the limitation of the existing system in detecting closely spaced targets, as a smaller angular separation cannot be accurately resolved. Subsequently, when the two targets approach each other but maintain a fixed angular offset, the existing radar system can only detect a merged target in the angular spectrum due to its limited angular resolution, as shown in Figure 4(as shown in (b-5)). The generated point cloud is a combined representation of the two targets. In contrast, the proposed system exhibits superior performance, accurately identifying two closely spaced targets even in the case of small angular separation, and is not affected by leakage interference. The experimental results verify the effectiveness of the system in detecting closely spaced human targets, demonstrating its excellent performance in solving small-angle scenarios, while successfully eliminating leakage interference. In the second experiment, the targets move in a crossed gait pattern, as Figure 4 (shown in (c-1) and (c-4)). Both systems are used, and the data is processed through the proposed signal processing chain. The generated point cloud trajectories are as Figure 4 (shown in (c-2), (c-3), and (c-4)), and the two trajectories maintain almost the same angular separation throughout the process. The point cloud trajectories obtained from the two systems show that during the crossing process with limited resolution, one target partially occludes the other target, resulting in the loss of information of the occluded target. For the proposed system, the lost point cloud data is limited to less than 0.5 meters, while the data loss of the uniform array system exceeds 1.2 meters. The proposed system reduces the lost point cloud data by nearly three times, making the point cloud information at the crossing point significantly richer. This shows the superior performance of the system in dealing with target occlusion, and its ability to retain more detailed target information is particularly prominent. Through these experiments, the effectiveness and robustness of the proposed system and signal processing chain are verified, demonstrating the excellent performance of the system. This highlights the key significance and practical application value of the system in short-range MIMO FMCW millimeter-wave radar systems. A series of experiments verify the performance of the proposed system and signal processing link, demonstrating its ability to obtain high-quality point clouds and accurate gait trajectories. However, in scenarios involving target occlusion, such as crossed paths, some point cloud data may still be lost due to the occlusion effect. In addition, when the target is out of the detection range, further trajectory information is sometimes required. The point cloud generated by the proposed system can be used as the input of a neural network to predict point cloud data in these challenging scenarios. Supplementary verification is carried out to demonstrate the feasibility of using a neural network for point cloud prediction. A simple long short-term memory (LSTM) model is adopted to predict the additional point cloud data at the end of the trajectory and fill in the missing point cloud at the connection of the crossed trajectories. The results are as Figure 4 (shown in (d)), comprehensively demonstrating that high-quality point clouds can be effectively used for AI-driven processing. The system framework significantly enhances the capabilities of millimeter-wave radar in short-range sensing applications.

[0035] Compared with the prior art, the proposed method utilizes the frequency-domain characteristics of intermediate-frequency signals, effectively reducing leakage. It is applicable to both synchronous and asynchronous systems, and has high robustness and generality. It can solve the multipath leakage problem, thereby achieving precise point cloud processing. The signal-to-interference ratio and signal-to-noise ratio are increased by 25 dB and 15 dB respectively. The designed K-band 2T4R sparse MIMO FMCW radar system has a 3 dB beamwidth of 8.97° and a measured angular resolution of 13.8°, which are increased by 29.8% and 32.8% respectively compared with existing uniform arrays. At the same time, it reaches a sidelobe level of -10 dB and a field of view of 100°, laying a solid foundation for obtaining high-quality point cloud data. Compared with existing systems, this system is more precise in multi-target point cloud-level detection performance, especially significant in short-distance applications.

[0036] In summary, the present invention comprehensively considers the hardware system and the signal processing link, combines and designs the two, and completes the pioneering millimeter-wave radar indoor multi-target point cloud trajectory tracking.

[0037] Those skilled in the art can make local adjustments to the above specific embodiments in different ways without departing from the principles and purposes of the present invention. The protection scope of the present invention is defined by the claims and is not limited by the above specific embodiments. All implementation solutions within its scope are subject to the present invention.

Claims

1. A sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking method, characterized in that: include: Step 1) performing a range-fast Fourier transform on the intermediate frequency signal collected by the sparse MIMO FMCW millimeter wave radar system to obtain a frequency domain signal; Step 2) using frequency domain leakage filtering technology to filter out strong leakage interference targets in the frequency domain signal and retain the intermediate frequency signal of the target of interest; Step 3) performing a velocity-fast Fourier transform on the intermediate frequency signal of the target of interest to obtain a range-Doppler map and extract the range and velocity information of the target of interest; Step 4) sparse array traditional beamforming is performed on the acquired target signal to estimate the angle information of the target, and digital beamforming technology is used to match the angle and distance information, thereby obtaining the distance, speed and angle of each target of interest, and then forming a spatial point cloud; Step 5) For the spatial point cloud formed by the above steps through multi-frame acquisition, the angles of the point clouds at the same distance are averaged to smooth the trajectory to form the moving point cloud trajectory of the target.

2. The sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking method according to claim 1 is characterized in that: The distance-fast Fourier transform specifically includes: (1.1) Modulated signal transmitted by MIMO FMCW radar Where: t is the fast time, A is the amplitude of the signal, f c is the center frequency, the modulation slope γ=B / T c , B is the bandwidth, T c is the pulse repetition time (PRT), is the initial phase. When the signal target is scattered, the echo signal S received by the radar Rx (t) = σS Tx (t-Δt), where: σ is the attenuation factor, which is related to factors such as the propagation medium and radar cross-section (RCS), Δt is the time delay, given as Δt = 2R(τ) / c, R(τ) = R + vτ is the radial distance between the target and the radar, τ is the "slow time", v and c are the target speed and the speed of light respectively; (1.2) The received signal is mixed with the local oscillator (LO) signal and passed through a low-pass filter (LPF) to obtain the intermediate frequency (IF) signal. Where: 4πγR 2 (τ) / c 2 is the residual video phase (RVP), which is negligible, and λ is the wavelength; (1.3) Perform one-dimensional fast Fourier transform (1DFFT) along the fast time dimension to extract the distance of the target Where: sinc(x) = sin(x) / x, f r is the frequency parameter, for S 1DFFT Each row of the processed intermediate frequency (IF) signal is obtained. r =2γR / c, a peak value will appear, corresponding to the frequency of the IF signal, which is related to the distance of the target, distance R = cf r / 2γ.

3. The sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking method according to claim 1 is characterized in that: The frequency domain leakage filtering technology specifically includes: (2.1) The actual signal containing strong leakage signal and target signal is modeled as: Where: α and β represent the amplitude of leakage signal and target signal respectively, M and N represent the number of leakage signal and target respectively; (2.2) Collect the anti-leakage signal S in a darkroom environment in advance Anti-leakage , the anti-leakage signal only contains the leakage signal because it is collected in a dark room, and then the leakage is filtered out in the frequency domain, that is, the anti-leakage signal S Anti-leakage and the actual signal S ALL Process it and convert it into the frequency domain. It should be emphasized that the velocity associated with the leakage signal is neglected because it does not exist. In addition, the amplitude coefficient α m Usually it remains constant. In the case of slight fluctuations in energy, the amplitude can be easily adjusted to achieve alignment. In addition, the above two equations are divided in the frequency domain and applied to each frequency point to obtain After analysis, it is found that for the frequency point corresponding to the leakage signal, only the asynchronous phase component remains, and the target signal is the sum of the anti-leakage signal divided by. In order to simplify the subsequent derivation to remove interference, the denominator is combined into A Total (f r )exp(jζ Total (f r )); (2.3) For each frequency point, extract the phase and get the value from S Division Subtract from, specifically: For synchronous systems, it is natural to observe that after the division, only the amplitude remains, so simply subtracting 1 eliminates the leakage signal, and the result contains only the signal corresponding to the target. It is well known that although the change in amplitude is irrelevant to the extraction of the target spatial parameters, the phase change has a significant impact. Therefore, in order to accurately compensate for these phase differences, S Subrtaction Must be with S Anti-leakage Multiplying together can be expressed as: The amplitude components are combined into A Final (f r ), replaces the overall amplitude expression. Any fluctuation in this component has no effect on subsequent processing, thus ensuring that the integrity of the spatial parameters is maintained. At this point, the leakage signal is effectively eliminated while retaining the phase of the target signal. The amplitude has only changed slightly, which has little impact on information extraction. More importantly, this technology is applicable to both synchronous and asynchronous systems and comprehensively solves the problem of multipath leakage. After the leakage is eliminated, the target signal becomes more significant, making it clearly visible in the range spectrum. In order to obtain the target trajectory point cloud, it is necessary to further process S Final , extract speed and orientation, thereby achieving multi-dimensional point cloud imaging.

4. The sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking method according to claim 1 is characterized in that: The distance and speed of the target are obtained by: (3.1) for S Final Fast Fourier transform is performed in the slow time dimension to obtain the Doppler frequency domain at different distances, forming a range-Doppler map, and the constant false alarm rate (CFAR) technology is used to extract the target to obtain the target's distance and speed; (3.2) Azimuth estimation is achieved by analyzing the phase relationship between virtual channels: When there are P transmitting antennas and Q receiving antennas, and they are all arranged horizontally, the first antenna in each array is designated as the reference element, and the spacing between the subsequent elements and the reference element is represented by d tp and d rq , the steering vectors of the transmitter and receiver arrays are reformulated as Next, the azimuth angle is estimated. The steering vector of the virtual array is derived based on the sparse array arrangement, specifically: in: symbol represents the Kronecker product. Taking into account the switching delay of the time division multiplexing (TDM) model and the phase change caused by the speed of the target, the sparse MIMO signal model can be expressed as: Each transmit antenna (TX) transmits all the chirps in a single frame before switching to the next antenna, so a compensation factor N is required in the phase chirp , by multiplying each Doppler bin by exp(-j(4πv n / λ)T c (P tp -1)N chirp ) to compensate for the phase change and select the corresponding speed v based on the detected target n To extract the range-Doppler matrix 2v n / λ corresponds to a vector, which is a two-dimensional matrix containing the virtual channel dimension and the distance dimension, with dimensions P×Q and N respectively sampling ,It should be noted that the matrix may contain multiple targets that share the same velocity. Therefore, the angles and distances of these targets need to be matched to ensure accurate estimation. Finally, the snapshot matrix is ​​combined with the steering vector and processed using the existing beamforming (CBF), specifically: This process generates an angular spectrum to estimate the azimuth, which is a critical step because the spatial spectrum cannot be directly obtained through FFT for sparse arrays. Where: H represents the conjugate transpose.

5. The sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking method according to claim 1 is characterized in that: The matching specifically includes: using digital beam forming (DBF) to match the distances and angles of multiple targets in a selected speed dimension, the snapshot matrix is ​​weighted to point to the angles of different targets, so as to obtain the distance spectrum of each angle, and the speed-angle-distance matching of each target is achieved by matching the distances detected by the distance spectra of different angles and the distances detected at the selected speed on the range-Doppler map.

6. The sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking method according to claim 1 is characterized in that: The target point cloud trajectory is obtained in the following way: multiple frames of data are processed to obtain a large amount of matched point cloud information of multiple targets, and the point cloud of each target at the same distance is averaged in the angle dimension to represent the point cloud at the distance. Since the human body is a multi-scattering target, different angle solutions will be generated in different frames at the same distance. By taking the average, the generation of high-quality point clouds is maintained, and the clarity and smoothness of the trajectory are enhanced.

7. A sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking system implementing any of the methods described in claims 1-6, characterized in that: include: A voltage-controlled oscillator chip, a phase-locked loop chip, a receiving chip, a microcontroller unit, a baseband amplifier, an analog-to-digital converter and a signal processing module, wherein: the voltage-controlled oscillator chip and the phase-locked loop chip work together under the configuration of the microcontroller unit to generate a chirp signal, which is transmitted into space by the transmitting antenna in the sparse array; when the signal encounters a target, it is transmitted and received by the receiving antenna in the sparse array and enters the receiving chip; an intermediate frequency signal is generated through mixing and a low-pass filter; after passing through the baseband amplifier, the signal is fully amplified and collected by the analog-to-digital converter and transmitted to a computer in the signal processing module for subsequent processing.

8. The sparse MIMO FMCW millimeter wave radar indoor multi-target point cloud trajectory tracking system according to claim 7 is characterized in that: The signal processing module includes: a distance processing unit, a frequency domain leakage filtering unit, a distance-Doppler processing unit, a sparse array angle solving unit, an angle-distance matching unit and a point cloud trajectory output unit, wherein: the distance processing unit is used to process the collected original signal to obtain spectrum information; the frequency leakage filtering unit is used to filter out the strong leakage interference target in the frequency domain to retain the target of interest, the range-Doppler processing unit is used to solve the range Doppler map of the retained target of interest to obtain the distance and speed information, the sparse array angle solving unit is used to solve the spatial angle information of the target of interest, the angle-distance matching unit is used to match the distance and angle information of the target to achieve distance-speed-angle matching alignment, and the point cloud trajectory output unit realizes the output of the trajectory point cloud of each target.

Images