A mobile target trajectory display method and system based on FMCW radar

Abstract

A mobile target trajectory display method and system based on FMCW radar belong to the technical field of radar signal processing. The method comprises: obtaining radar original echo data; by extracting the common spatial distribution characteristics of short-range leakage, a template signal is reconstructed, and the original echo data is suppressed to eliminate near-field interference; after suppressing the leakage, the adjacent Chirp difference frequency signals are subjected to time domain convolution operation to enhance the echo energy of weak moving targets and preliminarily suppress noise; based on the amplitude-phase double variable sensitivity, space-time connectivity and path length propagation mechanism, triple clutter detection is carried out, non-desired static clutter and transient interference are finely filtered out, and the target trajectory is reconstructed. The present application can effectively overcome the pain points of single-transmit-single-receive FMCW radar, such as serious near-field leakage, weak human echo and complex clutter interference in indoor environment, and significantly improve the target detection accuracy and robustness of low-cost radar system in complex environment.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and in particular to a method and system for displaying the trajectory of a moving target based on FMCW radar. Background Technology

[0002] Frequency modulated continuous wave (FMCW) radar is widely used in smart homes, security monitoring, human-computer interaction, and vital sign monitoring due to its significant advantages such as high range resolution, accurate velocity measurement, no privacy leakage risk, and low power consumption. Especially in indoor close-range detection scenarios, FMCW radar can detect the presence, location, and movement trajectory of people non-contactly. However, in practical applications, especially in low-cost single-input single-output (SISO) FMCW radar systems, achieving high signal-to-noise ratio and high-precision moving target detection still faces the following three major technical challenges.

[0003] First, short-range leakage and strong near-field reflections severely limit the detection blind zone. In highly integrated and miniaturized FMCW radar systems, due to limited isolation between the transmitting and receiving antennas and imperfect impedance matching at the antenna ports, some transmitted signals are directly coupled to the receiving channel, forming so-called "leakage signals" or "direct waves." Furthermore, strong reflected echoes from stationary objects near the radar (such as mounting brackets, walls, furniture, etc.) are also superimposed on the received signal. These leakage signals and near-field static reflections typically have extremely high energy, often several orders of magnitude higher than long-range or weak human body reflection signals. In the frequency domain, these strong static components are mainly concentrated in the low-frequency band of the difference frequency signal (corresponding to zero-range or near-zero-range), and their huge dynamic range easily leads to analog-to-digital converter (ADC) saturation or severe spectral sidelobe leakage in subsequent digital signal processing. This leakage can overwhelm weak moving target signals within adjacent range cells, creating a detection "blind zone" that makes it difficult for the radar to detect moving target signals. Traditional fixed filters or simple background cancellation methods are often unable to adapt to environmental changes. Once the static scene around the radar is fine-tuned or the temperature drifts, the fixed leakage model immediately fails, resulting in a significant decrease in the suppression effect, or even causing the target to be completely submerged.

[0004] Second, human body reflected signals are weak, and conventional denoising algorithms easily lead to feature loss. Unlike metallic targets or large vehicles, the radar cross section (RCS) of the human body is small and fluctuates dramatically with changes in posture, clothing material, and angle of movement. In complex indoor environments, multipath signals after multiple reflections further attenuate the effective echo energy. Especially when the human body is in a state of slight movement (such as slow walking or hand gestures), the signal-to-noise ratio (SNR) of its echo signal decreases. In existing radar signal processing, MTI (Moving Target Indication) single-delay-line cancellation or phasor mean cancellation algorithms are often used to filter out static backgrounds. However, the MTI algorithm not only weakens the absolute amplitude of the target but also retains extremely cluttered high-speed background noise; while the phasor mean cancellation algorithm, although retaining micro-Doppler information, has limited enhancement effect on human slight movements (such as breathing or slow walking) under extremely low SNR conditions. Conventional mean / median filtering often comes at the cost of sacrificing high-frequency details of the signal and cannot effectively utilize the phase coherence characteristics between adjacent chirps to actively improve the signal-to-noise ratio.

[0005] Third, removing static clutter and unwanted dynamic interference is challenging, leading to frequent false trajectories. When generating range-time maps or range-Doppler maps, in addition to the main static background, the environment contains numerous unwanted dynamic disturbances, such as fan rotation, curtain swaying, pet activity, or electromagnetic interference from electronic devices. These interference signals may exhibit characteristics similar to human motion in amplitude and phase, causing traditional detection algorithms based on a single threshold (such as amplitude threshold or Doppler threshold only) to generate a large number of false alarms. More problematic is that random noise often appears as isolated bright spots on the time-frequency map; without effective spatiotemporal correlation analysis, these isolated noise points are easily misidentified as transient targets. In contrast, genuine human motion trajectories exhibit significant continuity, smoothness, and persistence in both the time and range dimensions. Existing detection methods often lack in-depth analysis of this "spatiotemporal continuity" and "motion persistence," making it difficult to distinguish between genuine continuous human trajectories and discontinuous random interference. This results in fragmented trajectories or the inclusion of numerous false paths, severely impacting the accuracy of subsequent target tracking and behavior recognition.

[0006] In summary, current single-transmitter, single-receiver FMCW radar systems still have significant shortcomings in handling strong near-field leakage, amplification of weak human body echoes, and accurate trajectory extraction under complex backgrounds. There is an urgent need for a radar signal processing method and system that can adaptively extract and suppress short-range leakage, effectively enhance weak target echoes, and perform refined trajectory selection by combining multi-dimensional spatiotemporal characteristics (phase, amplitude, spatiotemporal connectivity, and path length) to overcome the aforementioned technical challenges and achieve highly robust target detection under complex backgrounds. Summary of the Invention

[0007] The purpose of this invention is to solve the problems of serious near-field short-range leakage, easy submersion of weak human body echoes by noise, and frequent false trajectories caused by static clutter and transient interference in the existing single-transmit single-receive (SISO) FMCW radar in complex indoor detection scenarios. The invention provides a moving target trajectory display method and system based on FMCW radar to achieve robust target detection and trajectory extraction with high signal-to-noise ratio and high accuracy.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for displaying the trajectory of a moving target based on FMCW radar includes the following steps:

[0010] 1) Acquire the echo signal received by the FMCW radar and convert it into the original difference frequency data matrix;

[0011] 2) Perform short-range leakage suppression processing on the original difference frequency data matrix to eliminate near-field static interference components and generate the first processed data;

[0012] 3) Perform temporal convolution operation on the signals of adjacent Chirp in the first processed data to enhance the echo energy of the moving target and suppress incoherent noise, and generate the second processed data;

[0013] 4) Perform multiple clutter detection on the second processed data to filter out unwanted static clutter and transient interference, and extract and output the trajectory of the moving target.

[0014] Step 2) includes the following: performing frequency domain filtering on the original difference frequency data matrix to extract a low-frequency data matrix corresponding to the short-range leakage distance range; calculating the covariance matrix of the low-frequency data matrix and performing eigenvalue decomposition on the covariance matrix to extract the principal eigenvector corresponding to the largest eigenvalue; using the principal eigenvector to perform linear mapping reconstruction on the low-frequency data matrix to generate a short-range leakage template signal; subtracting the short-range leakage template signal from the original difference frequency data matrix to obtain the first processed data.

[0015] Step 3) includes the following: extracting the nth chirp signal and the (n+1)th chirp signal from the first processed data; performing time-domain convolution on the nth chirp signal and the (n+1)th chirp signal, and using the phase coherence of the target echo between adjacent chirps to accumulate signal energy, thereby obtaining the second processed data.

[0016] Step 4) includes the following: transforming the second processed data to the frequency domain to obtain a frequency domain data matrix containing amplitude and phase information; performing bivariate change detection based on the frequency domain data matrix to generate an initial binarized trajectory map; performing spatiotemporal neighborhood connectivity filtering on the initial binarized trajectory map to remove isolated noise points and generate a denoised binarized trajectory map; performing path length propagation and filtering based on dynamic programming on the denoised binarized trajectory map to retain trajectory points with continuous path lengths greater than or equal to a preset threshold and generate a target trajectory mask; reconstructing the signal using the target trajectory mask and outputting the moving target trajectory.

[0017] The bivariate change detection includes: calculating the phase difference and amplitude difference of each distance unit between adjacent slow time intervals; if the phase difference exceeds a preset phase change threshold and the amplitude difference exceeds a preset amplitude change threshold, then the corresponding point is marked as a potential target point.

[0018] The execution of path length propagation and filtering based on dynamic programming includes:

[0019] Forward propagation phase: Traverse along the slow time dimension. If the current point is a valid point, find the point with the largest path length in the adjacent distance cells of its previous time step, and add 1 to its path length as the path length of the current point. If there is no adjacent valid predecessor point, reset the path length of the current point to 1.

[0020] Backtracking phase: Traverse backwards from the last column of slow time data to identify endpoints where the cumulative path length is greater than or equal to the preset minimum effective trajectory length; starting from the endpoints, backtrack backwards according to the decreasing path length relationship, and mark the points on the backtracking path as the final effective trajectory points.

[0021] A moving target trajectory display system based on FMCW radar includes:

[0022] The radio frequency hardware system is used to transmit FMCW signals and receive echo signals, and output raw difference frequency data.

[0023] A control system is used to configure and drive the radio frequency hardware system through a communication protocol and coordinate the timing synchronization of each module.

[0024] The signal processing system, which is communicatively connected to the radio frequency hardware system, is configured to execute the moving target trajectory display method based on FMCW radar as described in any one of claims 1 to 6.

[0025] The radio frequency hardware system adopts a single-transmit, single-receive architecture, including:

[0026] A frequency signal source is used to generate analog pre-transmission signals and status indication signals according to the configuration parameters of the control system;

[0027] A power divider is used to split the analog pre-transmit signal into a transmit signal and a local oscillator signal;

[0028] A transmitting antenna is used to radiate the transmitted signal after it has been amplified by a first power amplifier.

[0029] A receiving antenna, used to receive echo signals;

[0030] A mixer is used to mix the echo signal amplified by a low-noise amplifier with the local oscillator signal amplified by a second power amplifier.

[0031] An analog-to-digital converter (ADC) is used to convert the mixed and filtered difference frequency signal into a digital signal and output it to the signal processing system.

[0032] Compared with the prior art, the beneficial effects achieved by the technical solution of this invention are:

[0033] 1. Adaptive and precise suppression of short-range leakage, completely eliminating near-field detection blind zones. This invention overcomes the limitations of traditional fixed filters or simple background cancellation methods, which suffer from poor environmental adaptability. By extracting low-frequency data in the frequency domain and constructing a covariance matrix for eigenvalue decomposition, it extracts the principal eigenvectors characterizing the common spatial distribution features of short-range leakage to reconstruct the template signal. This method can adapt to environmental fine-tuning and temperature drift, accurately subtracting strong near-field direct waves and static reflections, greatly improving the signal-to-interference ratio (SIR), effectively preventing ADC saturation, and preserving the complete low-frequency space for subsequent weak target extraction.

[0034] 2. An innovative temporal convolution mechanism for adjacent chirps achieves a doubling of target energy under extremely low signal-to-noise ratios. Addressing the challenges of small radar cross-sections (RCS) and weak micro-motion signals in humans, this invention cleverly utilizes the high phase coherence of real moving target echoes between adjacent chirps, while random noise and residual clutter are incoherent. By performing temporal convolution on the difference frequency signals of adjacent chirps, a square-law accumulation of signal energy is achieved, approximately doubling the original signal-to-noise ratio on a decibel scale and significantly reducing the noise floor. Compared to traditional MTI single-delay-line cancellation or phasor mean cancellation algorithms, this method greatly enhances the detection sensitivity of weak targets without weakening the absolute target amplitude or losing micro-Doppler characteristics.

[0035] 3. A triple detection architecture based on multi-dimensional feature fusion achieves ultimate false alarm rejection and continuous trajectory reconstruction. This invention abandons the shortcomings of traditional single-threshold methods (such as simple CFAR) that easily lead to false trajectories, and constructs a triple progressive clutter filtering framework of "amplitude and phase dual-variable sensitivity detection, spatiotemporal connectivity filtering, path length propagation screening". It not only uses morphological methods (six-neighbor detection) to accurately remove isolated points caused by random noise spikes, but also introduces a dynamic programming algorithm to filter out unwanted transient interference such as fan rotation and curtain swaying by utilizing the "persistence" feature of human motion. The final extracted target trajectory is smooth, continuous and highly reliable, fundamentally solving the industry problem of high radar false alarm rates in complex indoor environments. Attached Figure Description

[0036] Figure 1 This is an architecture diagram of the radio frequency hardware part of the FMCW radar system in an embodiment of the present invention.

[0037] Figure 2 This is a diagram illustrating the architecture of the frequency signal source portion of the FMCW radar system in this embodiment of the invention.

[0038] Figure 3 This is a diagram showing the measured results of the linear frequency modulation signal output by the frequency signal source in an embodiment of the present invention (at the moment the frequency sweep begins).

[0039] Figure 4 This is a diagram showing the measured results of the first power amplifier and the second power amplifier in an embodiment of the present invention. (Note: 1. The first power amplifier and the second power amplifier are the same amplifier, referred to as first and second for ease of distinction; 2. In the figure, parameter S11 represents the input reflection coefficient, and S21 represents the forward transmission gain.)

[0040] Figure 5 This is a diagram showing the measured results of the receiving link amplifier in this embodiment of the invention (Note: 1. To avoid the vector network analyzer exceeding the test range, a 6dB attenuator is connected to the stage, and the actual gain is above 40dB; 2. In the figure, parameter S11 represents the input reflection coefficient, and S21 represents the forward transmission gain).

[0041] Figure 6 This is a diagram showing the measured results of the transceiver antenna in an embodiment of the present invention.

[0042] Figure 7 These are actual photographs of the radio frequency hardware system built in this embodiment of the invention.

[0043] Figure 8 These are real-world images of three test scenarios (corridor A, corridor B, and lawn) in this embodiment of the invention.

[0044] Figure 9 This is a flowchart illustrating the main signal processing steps in an embodiment of the present invention.

[0045] Figures 10-12 These are comparison images of the effects of short-range leakage suppression before and after in corridor A, corridor B, and lawn environments in the embodiments of the present invention (short-range leakage is at about 0.3m).

[0046] Figures 13-15 These are comparison images of the effects of enhancing moving targets before and after in corridor A, corridor B, and the lawn environment, respectively, in embodiments of the present invention.

[0047] Figure 16 These are the original data images under three environments without any static clutter filtering (from left to right: corridor A, corridor B, lawn).

[0048] Figure 17 These are the processing effects of the existing MTI first-order cancellation in three environments (from left to right: corridor A, corridor B, and lawn).

[0049] Figure 18 The images show the filtering effect of the phasor mean cancellation algorithm in the existing technology under three environments (from left to right: corridor A, corridor B, and lawn).

[0050] Figure 19 These are the effect diagrams of static clutter filtering algorithm processing in corridor A, corridor B and lawn environments in the embodiments of the present invention (from left to right: corridor A, corridor B, lawn). Detailed Implementation

[0051] To make the technical problems, solutions, and beneficial effects of this invention clearer and more understandable, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. This invention can also be implemented or applied through other different specific examples, and the details in this specification can be modified and changed based on different viewpoints and applications without departing from the spirit of the invention.

[0052] Example 1: System Architecture and Hardware Example

[0053] This embodiment provides a method and system for displaying the trajectory of a moving target based on FMCW radar. The system mainly includes a control system, a radio frequency (RF) hardware system, and a digital signal processing system. Both the control system and the digital signal processing system are based on the RF hardware circuitry. The RF hardware system is determined based on the FMCW radar architecture. Figure 1 As shown, the frequency signal source architecture in the hardware system is as follows: Figure 2 .

[0054] In practical applications, this embodiment determines the radio frequency hardware system based on a single-transmit, single-receive (SISO) frequency-modulated continuous wave radar architecture. To ensure that the radar can effectively detect weak human targets in complex indoor environments, this embodiment first performs rigorous link budgeting. Link budgeting is performed based on the principles of frequency-modulated continuous wave ranging and velocity measurement, the formula for calculating the receiving antenna acquisition power, the formula for calculating receiver sensitivity, and the formula for calculating free space loss, thus determining the operating scenario and system parameters.

[0055] The principle of frequency-modulated continuous wave ranging and velocity measurement is as follows:

[0056]

[0057]

[0058] Where R is the distance between the object under test and the radar, C is the speed of light, and BW is the bandwidth of the linear frequency modulated wave. Let R be the frequency of the intermediate frequency signal corresponding to the distance R. The sweep duration of the linear frequency modulated wave for a word. Let λ be the velocity of the object to be measured, λ be the wavelength corresponding to the harmonic frequency of the linear frequency modulated wave, and T be the duration of a single linear frequency modulated wave, which is equal to the sum of the sweep duration and the single frequency duration.

[0059] The formula for calculating the receiver antenna acquisition power is as follows:

[0060]

[0061] in, The amount of power that the receiving antenna can receive. The radar cross-section of the target. The signal power output from the input port of the transmitting antenna. For the gain of the transmitting antenna, For the gain of the receiving antenna, since the transmitting and receiving antennas used in this system are the same model, therefore... and Equal, λ is the wavelength corresponding to the harmonic frequency of the linear frequency modulated wave, and R is the distance between the object under test and the radar.

[0062] The formula for calculating receiver sensitivity is as follows:

[0063]

[0064] in, This represents the minimum acceptable input signal power for the receiver. is the thermal noise power, which is -174dBm at room temperature; BW is the bandwidth of the linear frequency modulated wave; NF is the noise figure; and SNR is the minimum detectable signal-to-noise ratio.

[0065] The formula for calculating free space loss is as follows:

[0066]

[0067] L represents the signal power lost over an output distance of length R, where R is the distance the electromagnetic wave travels. The harmonic frequency of the linear frequency modulated wave.

[0068] Based on the above formula, the link budget is performed to determine its working scenario and system parameters, as shown in Table 1:

[0069] Table 1

[0070]

[0071] in, Here, ΔR and Δv represent the range resolution and velocity resolution, respectively, and Power represents the system power.

[0072] Module selection and circuit design: Based on the above parameters, the RF hardware system in this embodiment has completed a fine selection of modules, as detailed below.

[0073] (1) Frequency signal source: As the clock and frequency source of the entire system, its performance directly determines the radar's range resolution, speed accuracy and phase noise level. In this embodiment, a fractional N-division phase-locked loop architecture is used to generate a high-resolution linear frequency modulated signal. Its workflow is as follows: the reference oscillator provides a high-stability reference clock, which is boosted to a higher frequency by a frequency multiplier and then reduced to the reference operating frequency of the phase detector by an R-division. The phase detector compares the phase difference between the reference signal from the R divider and the feedback signal from the N divider, and outputs an error current to drive the loop filter. The magnitude of the charge pump current affects the loop bandwidth and stability, and is typically set to 5~20μA. The N divider consists of an integer part and a fractional part. The latter dynamically adjusts the average division ratio through a Σ-Δ modulator, thereby achieving arbitrary frequency steps. The final output frequency formula of the voltage-controlled oscillator is: , where K is the number of bits for the fractional resolution (e.g., 24 bits can achieve subhertz-level steps). and These are the integer and decimal values ​​set for the N-divider, respectively; the voltage-controlled oscillator adjusts its output frequency according to the control voltage output by the charge pump, and its tuning range covers the required frequency band of the system (9.35GHz~10GHz). Its phase noise at an offset carrier of 10kHz is -110dBc / Hz, its normalized 1 / f noise is -120.55dBc / Hz, and its spurious rejection ratio is greater than 60dBc; the frequency of the reference oscillator... A 100MHz crystal oscillator is preferred to balance phase noise and frequency division flexibility, PFD frequency The higher the better, but it is limited by the upper limit of the phase detector. The VCO tuning range reserves ±10% margin to cope with temperature drift. The frequency resolution is determined by the fractional divider bit number. 24 bits can achieve a step of <0.01Hz, which is sufficient to meet the stringent frequency accuracy requirements of radar.

[0074] (2) Amplifier: The power amplifier should be selected to ensure that the output power is sufficient to cover the link budget required for the maximum detection distance. For indoor applications, the output power is usually required to be between 10dBm and 20dBm. At the same time, its 1dB compression point P1dB should be higher than the expected peak echo power to prevent distortion. The gain flatness should fluctuate less than ±3dB in the target frequency band, and the efficiency should be better than 30% to reduce heat generation. For low-noise amplifiers, the noise figure NF should be given priority. At the same time, its gain should be high enough to suppress the noise contribution of subsequent mixers and filters. In this example, a 40dB gain and a 1.8dB noise figure are selected, and the input third-order intercept point IIP3 is -13dBm to cope with strong interference scenarios. The gain adjustment range of the variable gain amplifier should cover at least 60dB, and the step accuracy should be ≤1dB to adapt to the echo intensity changes of targets at different distances. At the same time, its nonlinear distortion THD should be less than -50dBc to ensure signal fidelity.

[0075] (3) IQ mixer: As the core of downconversion, its isolation is crucial. The LO-to-RF and LO-to-IF isolation should both be greater than 30dB to prevent local oscillator leakage from polluting the receiving channel. The conversion loss should be controlled within 10dB. If a single-sideband structure is used, the image rejection ratio should be better than 40dB. It should also support zero IF or low IF architecture to simplify subsequent filtering design.

[0076] (4) Bandpass filter: The center frequency should match the expected intermediate frequency range, the bandwidth should be slightly larger than the maximum difference frequency, and the stopband attenuation should be greater than 40dB to suppress out-of-band spurious signals.

[0077] (5) Analog-to-digital converter (ADC): The sampling rate must satisfy the Nyquist criterion and leave a margin, that is, the sampling frequency fs ≥ 2 × ,in The frequency of the intermediate frequency signal corresponding to the maximum distance is selected as 500KSps to provide oversampling gain. In terms of resolution, in order to capture weak human body echoes and take into account strong leakage signals, a 16-bit resolution ADC is used, with an effective number of bits (ENOB) of not less than 12 bits, a signal-to-noise ratio (SNR) > 70dB, and a spurious-free dynamic range (SFDR) > 80dB. It also has external trigger input and multi-channel synchronization capabilities to support subsequent system upgrades and MIMO expansion.

[0078] In addition, the operating frequency bands of all RF components fully cover the carrier frequency range of the system design (9.35GHz~10GHz), the high-frequency signal path adopts an electromagnetic shield, the temperature stability meets industrial-grade standards (-40℃ to +85℃), and the impedance matching (50Ω), trace length consistency and shielding grounding measures are strictly controlled to reduce phase mismatch and electromagnetic interference.

[0079] The control system writes parameters such as the start frequency and stop frequency to the frequency signal source via the standard SPI communication protocol and configures the ADC trigger mode. The ADC converts the mixed, filtered, and amplified difference frequency signal into a digital signal and transmits it to the signal processing system. The measured performance of each hardware module is as follows: Figures 3 to 6 As shown, the actual hardware system and test environment are as follows: Figure 7 and Figure 8 As shown. Specifically, the system's signal flow begins at the frequency signal source. Under the instructions of the software control system, this module generates a chirp RF signal with precise slope and periodic characteristics. This signal is first sent to a power divider for power distribution. One path is amplified to the rated transmit power by a power amplifier and then radiated into space by the transmitting antenna. The other path is used as a local oscillator signal, which is conditioned by a buffer amplifier and then input to the local oscillator port of the IQ mixer. At the same time, the echo signal reflected from the target is captured by the receiving antenna and passes through a low-noise amplifier for initial gain boost and noise suppression. It then enters the IQ mixer and is orthogonally mixed with the local oscillator signal, outputting two baseband difference frequency signals: in-phase I and quadrature Q. This signal is then filtered by a bandpass filter to remove high-frequency harmonics and DC offset, retaining components within the effective intermediate frequency bandwidth. Next, a variable gain amplifier dynamically adjusts the amplitude according to the echo intensity to avoid saturation or quantization loss. Finally, an analog-to-digital converter completes the analog-to-digital conversion according to the set sampling rate and timing window, and transmits the data to the signal processing system for further analysis. Throughout the process, the software control system writes key parameters such as start frequency, stop frequency, chirp time, and repetition period to the frequency signal source via the standard communication protocol SPI. At the same time, it configures the ADC's trigger mode, number of sampling points, and resolution, and ensures that all modules are strictly synchronized within microseconds.

[0080] Example 2: Signal Processing Algorithm Example

[0081] After acquiring the raw difference frequency data from the ADC output, the signal processing system executes the core clutter suppression and target detection method of this invention, as follows: Figure 9 As shown. Specifically, it consists of three key steps:

[0082] Step A: Short-range leakage suppression

[0083] The data from the ADC is automatically stored as a CSV file by the host computer. The ADC sampling frequency fs is 500kHz, and the time for another frequency sweep is... Automatic sampling of 500 points is written to a CSV file. Therefore, the data is divided into 500-point segments. Once a column is filled with 500 points, the next 500 points are written to the next column. Each row is padded with 12 zeros to expand it to 512 rows, thus obtaining the original first data matrix. (M, N), where M is the number of fast-time sampling points (corresponding to the distance dimension), which is 512, and N is the number of intermediate frequency signals in one frame (corresponding to the time dimension), which is set to 1024 columns in this processing. Due to the coupling between the transmitting and receiving antennas and reflections from near-field stationary objects, the data contains short-range leakage components with extremely high energy. Its amplitude often far exceeds that of human body echoes. Therefore, the first step aims to construct a leakage template and subtract it from the original data to generate suppressed second data. Short-range leakage signals typically exhibit stable spatial distribution characteristics, meaning their waveform structure remains highly consistent across multiple chirp cycles, primarily characterized by low-frequency spectral components. Therefore, this embodiment employs a subspace projection method based on Principal Component Analysis (PCA) to accurately estimate and remove leakage. The covariance matrix reflects the correlation between different components in the data, and eigenvalue decomposition can extract the main components. Specifically:

[0084] For the original first data matrix Each column (i.e., each chirp) is subjected to a Fast Fourier Transform (FFT) to transform it to the frequency domain:

[0085]

[0086] Obtain the matrix Where k is the frequency index and corresponds to the distance cell. Short-range leakage is mainly concentrated in the low-frequency band near zero distance. Define the cutoff frequency index. Corresponding to the maximum leakage impact distance Construct a frequency domain filter H(k), and... Multiplying by H(k) yields a low-frequency data matrix that primarily contains leakage information. .

[0087] To extract the principal components of the leak, a low-frequency data matrix was calculated. The sample covariance matrix C(N,N) in the slow time dimension:

[0088]

[0089] Since the energy of the leakage signal is much greater than that of the noise and is highly correlated across different Chirps, the maximum eigenvalue of C will be significantly greater than that of other eigenvalues.

[0090] Eigenvalue decomposition (EVD) of C yields several eigenvalues ​​and corresponding eigenvectors:

[0091]

[0092] in For eigenvalues, This is the corresponding feature vector.

[0093] The magnitude of an eigenvalue indicates the contribution of the component represented by that eigenvector to the data; the larger the eigenvalue, the more important the information contained in the corresponding eigenvector. Compare all eigenvalues ​​and select the largest eigenvalue. The corresponding principal eigenvector is the standard waveform template of short-range leakage in the fast time dimension. This vector characterizes the common spatial (distance) distribution pattern of the leakage signal in all intermediate frequency signals, while noise and weak moving targets, due to their uncorrelatedness, have their energy dispersed in smaller eigenvalues.

[0094] Using principal feature vectors A linear mapping is performed on the original low-frequency data to reconstruct the leakage signal template. From the original first data Subtract the reconstructed leakage signal template from the middle To obtain pure second data This step eliminates short-range leakage, avoiding subsequent issues such as limited dynamic range and spectral leakage of the ADC, thus laying the foundation for weak target detection. Figures 10-12 The study demonstrated the short-range leakage cancellation effect with the human body as the test target under three environments, with the leakage signal suppressed by an average of about 20 dB, which greatly improved the signal-to-interference ratio (SIR).

[0095] Step B: Temporal convolution of adjacent chirps to enhance moving targets

[0096] In order to extract weak human body echoes at extremely low signal-to-noise ratios, this embodiment uses the second data S clean Perform temporal convolution operations on adjacent chirps. Leveraging the high phase coherence of human motion echoes between adjacent chirps, and the incoherence of noise and residual clutter, the convolution operation achieves square-law accumulation of signal energy, generating a third data S with a significantly improved signal-to-noise ratio. enhanced .

[0097] The modulated signal transmitted by the FMCW radar can be modeled as follows:

[0098]

[0099] The received signal is:

[0100]

[0101] By performing a down-conversion operation and ignoring the residual video phase, the difference frequency signal can be obtained (N is the target quantity):

[0102]

[0103] set up and Let be the difference frequency signals after removing short-range leakage from the nth and (n+1th)th Chirps respectively (taking a single object under test as an example), denoted as . for , This indicates that the expression follows a pattern with a mean of 0 and a variance of . The noise of a stationary Gaussian random process has a spectral expression that is a constant. Define the convolution operation:

[0104]

[0105] so, The signal-to-noise ratio is from 10log( Growth to (Assuming the target movement distance is extremely short and the amplitude remains approximately constant), this is equivalent to doubling the original signal-to-noise ratio on a decibel scale. It's important to note that convolution introduces phase accumulation into the target signal, amplifying the velocity information. Therefore, correction is needed during the subsequent Doppler Fourier transform. In this signal processing flow, the correction scheme is as follows: pre-store the original phase, extract the odd-numbered rows from the convolutional data; these extracted rows constitute the amplitude matrix. Combining these two matrices achieves a squared increase in signal energy while maintaining phase invariance. Figures 13-15 The study demonstrates target augmentation effects on humans in three different environments. By traversing adjacent chirp convolutions, the noise floor is significantly reduced, the signal-to-noise ratio of moving targets is improved, and third-party data is generated. .

[0106] Step C: Triple Clutter Detection and Trajectory Reconstruction

[0107] To further filter out complex static clutter and transient interference indoors, Perform triple clutter detection and trajectory reconstruction:

[0108] The first step is a bivariate sensitivity test: [The text abruptly ends here, likely due to an incomplete sentence or a The fast time dimension is then subjected to FFT again to obtain the frequency domain matrix Z(r,t) containing amplitude and phase information, where r is the distance index and t is the time index. The amplitude spectrum is denoted as A(r,t) = |Z(r,t)|, and the phase spectrum as Φ(r,t) = ∠Z(r,t). In real-world scenarios, subtle human movements (such as breathing or limb swaying) cause drastic changes in echo phase and fluctuations in amplitude, while static clutter (such as wall reflections) exhibits stable phase and amplitude, and random noise manifests as irregular, minute fluctuations. Therefore, thresholds can be set for amplitude and phase changes to distinguish between human motion and minute fluctuations within the range of random noise. The amplitude and phase differences of adjacent time points at the same distance cell are calculated, the phase is de-wrapped, and then compared with the specified thresholds to generate an initial binarized trajectory map. :

[0109]

[0110] The second layer is spatiotemporal connectivity detection, using morphological methods to remove isolated points: the fastest human reaction time is on the order of hundreds of milliseconds, so human movement will at least not occur during isolated chirps, and the initial graph... There may still be isolated points caused by random noise spikes. The reasonable movement in the chirp originates only from the r-1, r, r+1 of the previous chirp, and its direction is only from the r-1, r, r+1 of the next chirp. Therefore, for For each point with a value of 1, perform the above six-neighbor detection. If it is in the same row, the previous row, the next row, and the same row, the previous row, and the next row of the next column, all of which are 0, then it is considered isolated and its value is set to 0. Otherwise, it is retained.

[0111]

[0112] The third step is path length propagation filtering: based on a dynamic programming algorithm, only long paths with continuous motion characteristics are retained. After the second step of detection, what remains are some short-term interference fragments (such as the movement generated by non-isolated chirps, like an insect flying by or the momentary interference of an electrical switch). Real human motion has persistence, meaning the trajectory length should exceed a certain threshold. This step uses a dynamic programming algorithm to calculate the longest continuous path.

[0113] Initialize the path length matrix L(r,t)=0

[0114] Traverse from t=2 to t=N:

[0115] For each For a given point, find the marked point with the largest path length in the adjacent distance units {r-1, r, r+1} of the previous time t-1. The maximum path length is incremented by one and stored at the position of that point. If there are no adjacent valid points in the previous time, then that point is the starting point of the new path.

[0116] Starting from the last column t=N, scan backwards to find all columns that satisfy L(r,n)≥ The end point.

[0117] For each endpoint that meets the condition, backtrack forward according to the decreasing relationship of L until t=1 or the path is broken.

[0118] Mark all points on the backtracking path as 1 in Maskfinal(r,t) and the rest as 0 to obtain the final binary mask.

[0119] The signal is reconstructed using this mask, i.e., Z(r,t) is weighted to completely filter out unwanted static clutter and transient interference, thereby extracting a continuous and smooth human motion trajectory. Figure 19 The static clutter removal effect with the human body as the test target is demonstrated in three environments.

[0120] To fully demonstrate the superiority of the present invention, the method of this embodiment is compared with the traditional MTI single delay line cancellation method. Figure 17 ) and phasor mean cancellation algorithm ( Figure 18 (Compare)

[0121] Unfiltered raw radar data Figure 16 In the process, the target is overwhelmed by clutter; although the traditional MTI algorithm suppresses the DC component, the background noise is chaotic and the absolute amplitude of the target is severely weakened; although the phasor mean cancellation algorithm does not weaken the target amplitude, the suppression effect is limited and there are still a lot of false highlight interference.

[0122] In comparison, the results processed using the method of the embodiments of the present invention ( Figure 19 This indicates that static clutter interference is filtered out, weak moving targets are effectively enhanced, and the output range-time trajectory is clear, continuous, and uninterrupted, achieving highly robust moving target display in complex environments.

[0123] The embodiments described above are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for displaying the trajectory of a moving target based on FMCW radar, characterized in that, Includes the following steps: 1) Acquire the echo signal received by the FMCW radar and convert it into the original difference frequency data matrix; 2) Perform short-range leakage suppression processing on the original difference frequency data matrix to eliminate near-field static interference components and generate the first processed data; 3) Perform temporal convolution operation on the signals of adjacent Chirp in the first processed data to enhance the echo energy of the moving target and suppress incoherent noise, and generate the second processed data; 4) Perform multiple clutter detection on the second processed data to filter out unwanted static clutter and transient interference, and extract and output the trajectory of the moving target.

2. The method for displaying the trajectory of a moving target based on FMCW radar as described in claim 1, characterized in that, Step 2) includes the following: performing frequency domain filtering on the original difference frequency data matrix to extract a low-frequency data matrix corresponding to the short-range leakage distance range; calculating the covariance matrix of the low-frequency data matrix and performing eigenvalue decomposition on the covariance matrix to extract the principal eigenvector corresponding to the largest eigenvalue; using the principal eigenvector to perform linear mapping reconstruction on the low-frequency data matrix to generate a short-range leakage template signal; subtracting the short-range leakage template signal from the original difference frequency data matrix to obtain the first processed data.

3. The method for displaying the trajectory of a moving target based on FMCW radar as described in claim 1, characterized in that, Step 3) includes the following: extracting the nth chirp signal and the (n+1)th chirp signal from the first processed data; performing time-domain convolution on the nth chirp signal and the (n+1)th chirp signal, and using the phase coherence of the target echo between adjacent chirps to accumulate signal energy, thereby obtaining the second processed data.

4. The method for displaying the trajectory of a moving target based on FMCW radar as described in claim 1, characterized in that, Step 4) includes the following: transforming the second processed data to the frequency domain to obtain a frequency domain data matrix containing amplitude and phase information; based on the frequency domain data matrix, performing bivariate change detection to generate an initial binarized trajectory map; Spatiotemporal neighborhood connectivity filtering is performed on the initial binarized trajectory map to remove isolated noise points and generate a denoised binarized trajectory map; The denoised binary trajectory map is subjected to path length propagation and filtering based on dynamic programming, and trajectory points with continuous path lengths greater than or equal to a preset threshold are retained to generate a target trajectory mask. The target trajectory is reconstructed using the target trajectory mask, and the trajectory of the moving target is output.

5. The method for displaying the trajectory of a moving target based on FMCW radar as described in claim 4, characterized in that, The bivariate change detection includes: calculating the phase difference and amplitude difference of each distance unit between adjacent slow time intervals; if the phase difference exceeds a preset phase change threshold and the amplitude difference exceeds a preset amplitude change threshold, then the corresponding point is marked as a potential target point.

6. The method for displaying the trajectory of a moving target based on FMCW radar as described in claim 4, characterized in that, The execution of path length propagation and filtering based on dynamic programming includes: Forward propagation phase: Traverse along the slow time dimension. If the current point is a valid point, find the point with the largest path length in the adjacent distance cells of its previous time step, and add 1 to its path length as the path length of the current point. If there is no adjacent valid predecessor point, reset the path length of the current point to 1. Backtracking phase: Traverse backwards from the last column of slow time data to identify endpoints where the cumulative path length is greater than or equal to the preset minimum effective trajectory length; starting from the endpoints, backtrack backwards according to the decreasing path length relationship, and mark the points on the backtracking path as the final effective trajectory points.

7. A moving target trajectory display system based on FMCW radar, characterized in that, include: The radio frequency hardware system is used to transmit FMCW signals and receive echo signals, and output raw difference frequency data. A control system is used to configure and drive the radio frequency hardware system through a communication protocol and coordinate the timing synchronization of each module. The signal processing system, which is communicatively connected to the radio frequency hardware system, is configured to execute the moving target trajectory display method based on FMCW radar as described in any one of claims 1 to 6.

8. A moving target trajectory display system based on FMCW radar as described in claim 7, characterized in that: The radio frequency hardware system adopts a single-transmit, single-receive architecture, including: A frequency signal source is used to generate analog pre-transmission signals and status indication signals according to the configuration parameters of the control system; A power divider is used to split the analog pre-transmit signal into a transmit signal and a local oscillator signal; A transmitting antenna is used to radiate the transmitted signal after it has been amplified by a first power amplifier. A receiving antenna, used to receive echo signals; A mixer is used to mix the echo signal amplified by a low-noise amplifier with the local oscillator signal amplified by a second power amplifier. An analog-to-digital converter (ADC) is used to convert the mixed and filtered difference frequency signal into a digital signal and output it to the signal processing system.

Images