A vehicle-mounted millimeter wave radar signal processing method resistant to multipath interference

Abstract

The application relates to the technical field of radar signal processing, and discloses a vehicle-mounted millimeter wave radar signal processing method resisting multipath interference. The method acquires a radar echo signal and decomposes the radar echo signal into a plurality of overlapping subband signals in a frequency domain. A static environment Doppler manifold surface is constructed by using self-vehicle motion data, and a theoretical phase evolution model is calculated according to the static environment Doppler manifold surface. An actual measurement phase difference between the subbands is extracted, and a manifold consistency coherence coefficient is generated by coherently accumulating the actual measurement phase difference and the theoretical model. Direction of arrival estimation is performed on each subband signal, and a spatial spectrum drift statistical dispersion of an angle estimation value is calculated. A target multipath confidence score is generated according to the manifold consistency coherence coefficient and the spatial spectrum drift statistical dispersion, and a suppression decision is performed on a radar detection target. By introducing a kinematic physical constraint and a wideband spatial stability detection mechanism, the application effectively identifies and eliminates false targets generated by the multipath effect, and improves the environmental perception accuracy and robustness of the vehicle-mounted radar.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, specifically to a vehicle-mounted millimeter-wave radar signal processing method that resists multipath interference. Background Technology

[0002] Vehicle-mounted millimeter-wave radar, due to its all-weather capability, high range resolution, and speed measurement ability, has become a core sensor in Advanced Driver Assistance Systems (ADAS) and autonomous driving technologies. In complex road traffic environments, the electromagnetic waves emitted by radar often undergo one or more reflections from environmental features such as road surfaces, guardrails, tunnel walls, or sound barriers before being received by the receiver. This non-line-of-sight propagation phenomenon is known as multipath effect, and the resulting echo signals, after radar signal processing, can create false targets at locations where no real target exists.

[0003] False targets generated by multipath interference often closely resemble real targets in terms of signal amplitude, range, and velocity characteristics, especially in scenarios with tunnels or dense metal barriers, where the signal-to-noise ratio of multipath signals can even exceed that of direct reflection signals. Existing radar signal processing techniques typically rely on constant false alarm rate (CFAR) algorithms to filter based on signal amplitude, making it difficult to distinguish strong multipath interference from an energy perspective. While some techniques attempt to obtain elevation angle information by increasing the vertical aperture of the antenna to differentiate between ground reflections and high-altitude reflections, this significantly increases the size and cost of radar hardware. Furthermore, although traditional tracking logic can eliminate some unstable false targets by observing trajectory consistency over a long period, this post-processing method has significant lag, making it difficult to meet the stringent requirements of real-time performance and low false alarm rates for safety functions such as automatic emergency braking (AEB). More critically, most existing signal processing workflows treat broadband echoes as narrowband signals processed with a single carrier frequency, ignoring the fundamental physical difference that the scattering center of the real target is relatively stable with frequency, while multipath interference signals fluctuate dramatically with frequency. This limits the reliability of radar systems' environmental perception in highly dynamic and complex environments. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a vehicle-mounted millimeter-wave radar signal processing method that resists multipath interference, solving the problem of false targets generated by vehicle-mounted radar due to multipath effects in complex road environments.

[0005] To achieve the above objectives, the present invention provides a vehicle-mounted millimeter-wave radar signal processing method to resist multipath interference. This method aims to improve the radar's ability to distinguish between static real targets and false interference by introducing broadband frequency diversity and vehicle kinematic constraints.

[0006] The technical solution adopted in this invention is as follows: Acquire radar echo signals and decompose the broadband echo into multiple partially overlapping sub-band signals in the frequency domain using signal processing techniques. This sub-band decomposition introduces frequency diversity characteristics, enabling the system to observe subtle patterns in the change of target electromagnetic scattering characteristics with carrier frequency. Simultaneously, the system acquires the vehicle's motion state data, including velocity and angular velocity information, through the vehicle bus or sensor interface. Combined with the radar's installation geometry parameters, a static environment Doppler manifold surface is constructed. This manifold surface reveals the theoretical radial velocity distribution that an absolutely stationary object in any direction in space should possess relative to the radar under the current vehicle motion state, providing a rigorous physical benchmark for subsequent signal consistency verification.

[0007] Based on this, the present invention calculates a theoretical phase evolution model for the target to be detected based on the aforementioned manifold surface. This model, combining the target's distance information with radial relative velocity constraints, derives the ideal phase gradient that the echo signal should follow in both the fast-time frequency dimension and the slow-time pulse dimension. In the frequency dimension, the phase change is linearly related to the target's physical distance; in the pulse dimension, the phase change is locked with the Doppler frequency shift caused by the vehicle's motion. The system extracts the actual measured phase difference between adjacent sub-band signals and performs full-dimensional coherent accumulation of this measured value with the theoretical phase evolution model to generate a manifold-consistent coherence coefficient. For real targets that satisfy the single-reflection geometry, the phases of each sub-band and each pulse are superimposed in phase after compensating the theoretical model, resulting in a high coherence coefficient. However, for multipath interference, since its propagation path length and apparent velocity cannot simultaneously satisfy the theoretical geometric constraints, phase interference destructive occurs during accumulation, significantly reducing the coherence coefficient.

[0008] In parallel, this invention utilizes sub-band signals for spatial dimension stability detection. The system estimates the direction of arrival for each sub-band signal, obtaining a series of angle estimates that vary with frequency, and calculates the statistical dispersion of the spatial spectral drift of these angle values. The electromagnetic scattering center of a real point target is relatively stable within a finite bandwidth, and its angle estimates remain highly consistent across different sub-bands; while multipath false targets are typically formed by the interference of multiple path signals. As the frequency changes, the relative phase rotation between the paths causes the equivalent phase center after superposition to undergo drastic jumps or flickering in the angular domain.

[0009] Finally, this invention generates a target multipath confidence score based on the manifold consistency coherence coefficient and the spatial spectrum drift statistical dispersion, using a mapping function and weighted fusion logic. This score comprehensively reflects the physical authenticity of the target in terms of phase evolution and its frequency stability in spatial location. Based on this score, the system performs suppression decisions on targets in the detection list, generating invalid masks to eliminate false targets or suppress their signal strength.

[0010] This invention provides a method for processing vehicle-mounted millimeter-wave radar signals to resist multipath interference. It has the following beneficial effects: 1. This invention constructs a static environment Doppler manifold surface by acquiring the vehicle's motion state, and establishes a two-dimensional theoretical phase evolution model based on it. It then performs coherent accumulation verification with the actual measured signal, introducing the physical geometric constraints of radar echoes into the signal processing layer. By utilizing the essential differences between the real target and the multipath virtual image in terms of propagation path length and apparent velocity, false signals that do not conform to the single reflection geometry relationship undergo phase cancellation during the coherent accumulation process, thereby improving the ability to suppress multipath interference in the static environment.

[0011] 2. This invention employs frequency domain subband decomposition combined with direction-of-arrival estimation. By calculating the spatial spectral drift statistical dispersion between subbands, the spatial stability of the target is quantified. Utilizing the physical characteristics of multipath interference signals, which are formed by interference from multiple paths and whose equivalent phase center changes drastically with the carrier frequency, this invention can effectively identify false targets that are difficult to distinguish in conventional detection in the spatial dimension, thereby enhancing the robustness of radar in complex multipath scenarios such as tunnels and guardrails.

[0012] 3. This invention constructs a multi-dimensional fusion decision mechanism based on manifold consistency coherence coefficient and spatial spectrum drift characteristics. By generating a comprehensive multipath confidence score through a mapping function, it avoids the one-sidedness of single feature judgment. It can generate invalid masks to eliminate false targets while ensuring the probability of real obstacle detection. Without increasing additional hardware costs, it reduces the risk of false braking caused by radar multipath false alarms in autonomous driving systems. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the measurement phase difference extraction and differential phasor construction process of the present invention; Figure 3 This is a schematic diagram comparing the subband spatial spectrum drift in an application embodiment of the present invention; Figure 4 This is a schematic diagram comparing the point cloud effects before and after multipath suppression in an application embodiment of the present invention. Detailed Implementation

[0014] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] This invention provides a signal processing method for vehicle-mounted millimeter-wave radar that resists multipath interference. This method operates in the signal processing unit of a vehicle-mounted radar system.

[0016] The vehicle-mounted radar system is installed on the vehicle platform to detect the surrounding environment. The system includes a radio frequency (RF) front-end module, an analog-to-digital converter (ADC), a signal processing unit, and a vehicle communication interface. The RF front-end module includes a transmitting antenna array, a receiving antenna array, a waveform generator, a power amplifier, a low-noise amplifier, and a mixer.

[0017] The transmitting antenna array comprises multiple transmitting antenna elements configured to transmit frequency-modulated continuous wave signals toward a target area. A waveform generator, connected to the transmitting antenna array, is configured to generate linearly frequency-modulated pulse signals whose frequency varies linearly with time. A power amplifier, located between the waveform generator and the transmitting antenna array, amplifies the transmitted signal.

[0018] The receiving antenna array comprises multiple receiving antenna elements configured to receive electromagnetic wave echo signals reflected from a target area. The geometric layout of the receiving antenna array adopts a uniform linear array or sparse array structure, and the physical spacing between each receiving antenna element is a fixed value.

[0019] The mixer is connected to both the transmit and receive links. It is configured to mix the received echo signal with the currently transmitted signal, outputting an intermediate frequency (IF) signal. A low-noise amplifier, located between the receiving antenna array and the mixer, amplifies the weak echo signal.

[0020] The analog-to-digital converter (ADC) module is connected to the output of the mixer. The ADC module is configured to sample and quantize the analog intermediate frequency (IF) signal at a preset sampling rate, converting the continuous time signal into a discrete digital IF signal. This digital IF signal contains sample point data in the fast time dimension, pulse sequence data in the slow time dimension, and antenna channel data in the spatial dimension.

[0021] The signal processing unit is connected to the analog-to-digital converter module and receives digital intermediate frequency signals. The signal processing unit includes at least one of a digital signal processor, a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). The signal processing unit is configured to execute computer program instructions stored in a memory to implement the multipath interference suppression method of the present invention.

[0022] The vehicle communication interface is connected to the signal processing unit and also to the vehicle's control area network bus. The signal processing unit acquires the vehicle's motion status data in real time through the vehicle communication interface. The vehicle motion status data includes the vehicle's longitudinal velocity, lateral velocity, and yaw rate.

[0023] To describe the geometric relationship between the target, radar, and vehicle, a vehicle coordinate system and a radar coordinate system are defined.

[0024] The vehicle coordinate system has its origin at the center of the rear axle. The longitudinal axis is defined as the X-axis, the transverse axis as the Y-axis, and the direction perpendicular to the ground as the Z-axis. The direction of vehicle movement is the positive X-axis direction.

[0025] The radar coordinate system has its origin at the geometric center of the radar sensor plane. The radar line-of-sight direction is defined as the X-axis of the radar coordinate system, and the direction parallel to the radar array arrangement is defined as the Y-axis. The installation position of the radar sensor on the vehicle is represented by a translation vector, and the installation angle of the radar sensor is represented by the azimuth installation error and the elevation installation error.

[0026] The signal processing unit is logically divided into a data preprocessing module, a manifold construction module, a subband decomposition module, a joint calibration module, and a decision output module.

[0027] The data preprocessing module is configured to perform windowing and fast Fourier transform on the digital intermediate frequency signal, converting the time-domain signal into a frequency-domain signal.

[0028] The subband decomposition module is configured to divide the signal bandwidth in the frequency domain, generating multiple overlapping subband signal data.

[0029] The manifold construction module is configured to calculate the theoretical distribution constraints of static targets in the range-Doppler and angle domains based on the vehicle motion state data and radar installation parameters obtained through the vehicle communication interface, and generate static environment Doppler manifold surface data.

[0030] The joint calibration module is configured to calculate the theoretical phase gradient between sub-band signals based on static environmental Doppler manifold surface data, and to calculate the correlation index between the actual measured phase gradient and the theoretical phase gradient.

[0031] The decision output module is configured to generate a target confidence mask based on correlation indicators and spatial angle estimation differences between sub-bands, and to filter or weight the final output target point cloud data.

[0032] See attached document Figure 1 This invention provides a vehicle-mounted millimeter-wave radar signal processing method to resist multipath interference. The method mainly includes echo signal acquisition and decomposition steps, static environment manifold construction steps, phase gradient joint calibration steps, spatial spectrum consistency detection steps, and multidimensional feature fusion decision steps.

[0033] In the echo signal acquisition and decomposition step, the signal processing unit receives the raw digital intermediate frequency (IF) signal output from the analog-to-digital converter (ADC). This IF signal is structurally represented as a three-dimensional matrix, with the three dimensions corresponding to fast-time sampling points, slow-time pulse sequences, and the receiving antenna channel, respectively. The signal processing unit first performs a Fast Fourier Transform (FFT) on the IF signal in the fast-time dimension, converting the signal to the range-frequency domain. Subsequently, in the range-frequency domain, the signal processing unit uses a sliding window mechanism or a multi-channel bandpass filter bank to divide the full-bandwidth signal into multiple sub-band signals with different center frequencies and partially overlapping frequency bands. Each sub-band signal retains complete information about the original signal in both the slow-time and spatial dimensions.

[0034] In the static environment manifold construction step, the signal processing unit reads the vehicle's real-time motion state parameters, including longitudinal velocity, lateral velocity, and yaw rate, through the vehicle communication interface. Based on the radar sensor's installation position and angle parameters, the signal processing unit calculates the relative radial velocity distribution of the static scattering point in the radar coordinate system. For any azimuth and elevation angle within the radar field of view, the radial velocity of the real static target must satisfy the geometric projection relationship determined by the vehicle's own motion. This deterministic mapping relationship between angle and velocity constitutes the static environment Doppler manifold surface. This manifold surface serves as a benchmark reference for verifying the target's physical properties during data processing.

[0035] In the phase gradient joint calibration step, the signal processing unit utilizes the previously constructed static environmental Doppler manifold surface to perform coherence verification on the phase evolution of each sub-band signal. For any range cell to be detected, the signal processing unit derives the theoretical phase gradient that the point target at that location should possess in the fast-time frequency dimension and the slow-time Doppler dimension based on the theoretical velocity and distance information provided by the manifold surface. Simultaneously, the signal processing unit calculates the actual measured phase difference of adjacent sub-band signals at the corresponding range cell. The signal processing unit compares the actual measured phase difference with the theoretical phase gradient and calculates the manifold consistency coherence coefficient through coherent accumulation in the slow-time dimension. This coefficient characterizes whether the microscopic phase structure of the echo signal conforms to the physical constraints defined by macroscopic vehicle kinematics.

[0036] In the spatial spectrum consistency detection step, the signal processing unit independently performs direction-of-arrival estimation for each sub-band signal obtained from the decomposition. The signal processing unit uses a multi-signal classification algorithm or a digital beamforming algorithm to calculate the spatial spectrum distribution corresponding to each sub-band and extracts the spatial angle estimates for each sub-band. The signal processing unit constructs a spatial spectrum drift vector from all the sub-band angle estimates and calculates the statistical dispersion of this vector. This statistical dispersion reflects the spatial position stability of the target scattering center under frequency variation conditions.

[0037] In the multi-dimensional feature fusion decision step, the signal processing unit weightedly fuses the manifold consistency coherence coefficient and the spatial spectrum drift statistical dispersion to generate a false confidence score for each detected target. The signal processing unit generates an initial target point cloud list based on the constant false alarm rate (CFAR) detection algorithm and filters the targets in the point cloud list according to the false confidence scores. Targets with scores exceeding a preset threshold are identified as false targets caused by multipath interference and are removed; targets with scores below the preset threshold are identified as real targets and retained. Finally, the anti-interference processed target point cloud data is output.

[0038] In this embodiment, the signal processing unit first processes the raw digital intermediate frequency (IF) signal output by the analog-to-digital conversion module, and performs a frequency-domain-based subband decomposition operation during the data preprocessing stage. The raw digital IF signal received by the signal processing unit... It is a three-dimensional matrix structure, containing a fast-time dimension, a slow-time dimension, and a spatial dimension. For the... The receiving channel, the first The nth linear frequency modulated pulse, at the nth Discrete echo signal at each fast sampling time point It can be represented as: ; in, This represents the complex amplitude of the echo signal. The imaginary unit, Pi The starting carrier frequency of the frequency-modulated continuous wave. For fast sampling time, For frequency modulation slope, The sampling time interval for analog-to-digital conversion. For the target round-trip time delay, for a distance of... And the radial velocity is The goal, latency ,in The pulse repetition period, This represents the speed of electromagnetic wave propagation. This represents superimposed Gaussian white noise. The phase term in the formula includes a fast-time frequency component related to distance and a slow-time Doppler phase component related to velocity.

[0039] The data preprocessing module processes the above discrete signals. In the fast time dimension A time-domain window function is applied. The window function can be a Hamming window, Hanning window, or Blackman window, used to suppress spectral sidelobe leakage during the subsequent frequency-domain transformation. The windowed signal is then converted to a distance-frequency domain signal using a Fast Fourier Transform (FFT). ,in This is the distance frequency unit index, with a value range of 0 to... This transformation focuses the target echo energy at different distances onto different frequency units, with the frequency unit index... Distance from target They are directly proportional.

[0040] In obtaining the distance frequency domain signal Subsequently, the subband decomposition module performs spectrum segmentation. This embodiment does not directly use the full bandwidth data for subsequent angular dimension processing, but instead uses the full bandwidth... Divided into Sub-band. Define the first The bandwidth of each subband is ,and A preset frequency overlap rate is set between adjacent sub-bands to maintain the continuity of the signal in the frequency domain.

[0041] The subband decomposition module extracts subbands by applying a set of bandpass filters or a frequency-domain sliding window in the distance-frequency domain. For the first... Each sub-band is defined with its starting index in the distance frequency domain as... The terminating index is . No. The center frequency of each sub-band Relative to the starting carrier frequency A frequency offset is generated, which is determined by the sub-band's position within the full frequency band. The sub-band decomposition module extracts the index range. The data within is padded with zeros to the original length. Or maintain the truncation length and construct the first Data matrix of each sub-band .

[0042] After the above processing, the original broadband signal is decomposed into Each group consists of narrowband signal components with different center frequencies. Each sub-band data group... The data structure remains unchanged. The three-dimensional form (or truncated dimension) preserves the Doppler phase evolution information of the original signal in the slow-time dimension and the array phase difference information in the spatial dimension. This decomposition allows subsequent processing steps to independently observe the scattering response characteristics of the target at different carrier frequencies. For each sub-band Its corresponding effective wavelength These differences provide a data foundation for subsequent phase consistency verification based on wavelength differences.

[0043] In this embodiment, the manifold construction module establishes a rigorous mathematical mapping between the vehicle's kinematics and radar observation geometry, providing a physical benchmark for subsequent signal consistency verification.

[0044] The manifold construction module first obtains the time-synchronized vehicle motion state vector from the vehicle chassis bus or inertial measurement unit via the vehicle communication interface. This state vector includes the vehicle's longitudinal velocity in the vehicle coordinate system. lateral velocity and the yaw rate of rotation about the vertical axis Simultaneously, the module reads pre-calibrated radar installation parameters from the system memory, including the installation position vector of the radar center relative to the center of the vehicle's rear axle. And the installation rotation matrix of the radar coordinate system relative to the vehicle coordinate system .

[0045] Based on the principles of rigid body kinematics, the manifold building block calculates the instantaneous velocity vector of the radar sensor center in the world coordinate system. Because the radar's installation position is offset from the vehicle's center of rotation, the radar's actual velocity includes a translational component and a rotational tangential component caused by the vehicle's steering. This instantaneous velocity vector, expressed in the vehicle coordinate system, is the vector sum of the vehicle's velocity vector and its linear velocity term, i.e.: ; in, This represents the actual linear velocity vector of the radar sensor in the vehicle body coordinate system. Longitudinal velocity, Lateral velocity, Vertical velocity, Indicates transpose. It is the angular velocity vector. This represents the vector cross product operation. Subsequently, the transpose of the rotation matrix is ​​used. The velocity vector is then projected and transformed into the radar sensor's own coordinate system to obtain the velocity vector in the radar coordinate system. .

[0046] After obtaining the radar sensor's own motion vector, the manifold construction module establishes a Doppler mapping function for static environment points. For any spatial direction within the radar's detection field of view, a unit direction vector is defined. ,in It is the azimuth angle. The elevation angle is given. This unit direction vector is expressed in radar coordinates as... .

[0047] Based on the Doppler effect, the manifold building block calculates the theoretical radial relative velocity of an absolutely stationary object relative to the radar in that direction. The theoretical radial relative velocity is equal to the negative of the projection of the radar sensor's velocity vector onto the target's direction vector. Its calculation formula is expressed as: ; in, The projection of the forward motion onto the target direction. The projection of lateral motion onto the target direction. The projection of vertical motion (such as bumps or changes in slope) onto the target direction.

[0048] For two-dimensional road targets, which are the primary focus of vehicle-mounted radar, ignoring the pitch angle (i.e., In the case of [missing information], the above mapping relationship simplifies to a nonlinear curve in the angular velocity domain. This curve is the static environment Doppler manifold. Within the system's angular observation range, the manifold construction module discretizes the formula at a preset angular resolution, generating the corresponding manifold lookup table or constructing a real-time calculation kernel function.

[0049] This Doppler manifold surface not only describes the distribution patterns of stationary objects on the road surface (such as guardrails, streetlights, and stationary vehicles), but also constitutes the geometric boundary for determining whether a target is a multipath virtual image. For any static target formed by a single direct reflection, its detected radial velocity and angle must fall precisely on this manifold surface. Conversely, if the target's radial velocity deviates from the surface, it indicates that the target possesses a non-zero absolute velocity, or that the target is a false image generated by multipath effects. This manifold surface, constructed based on vehicle motion constraints, eliminates the limitations of traditional methods that rely solely on amplitude thresholds, introducing physical kinematic characteristics into the underlying logic of signal processing.

[0050] In the joint calibration module, to verify whether the echo signal conforms to the characteristics of real physical reflection, it is first necessary to establish a theoretical phase reference model based on manifold constraints. This model describes the phase change law that an ideal static point target conforming to vehicle kinematics should follow during broadband frequency scanning and long-term pulse observation.

[0051] The joint calibration module targets each range cell index to be processed. Based on the sampling rate and frequency modulation parameters of the radar system, the corresponding physical line-of-sight is determined. Simultaneously, this module traverses the assumed azimuth values ​​within a preset angle search space. For each assumed azimuth angle The joint calibration module calls the static environment Doppler manifold surface data generated in the aforementioned steps to obtain the radial relative velocity value that should theoretically exist at this angle. Based on a determined physical line-of-sight. and radial relative velocity value This module derives the differential characteristics of the phase in both the fast time-frequency dimension and the slow time-pulse dimension.

[0052] In the fast time-frequency dimension, phase changes are primarily caused by the physical distance to the target. Considering the frequency diversity effect resulting from subband decomposition, for a center frequency interval of... For two adjacent sub-bands, the difference in echo path length between the real target and the actual target will result in a fixed phase shift. The joint calibration module calculates the theoretical fast-time phase gradient. The calculation formula is as follows: ; in, The speed at which electromagnetic waves propagate in the air. Frequency difference. This gradient value characterizes the linear shift in the echo signal phase that should occur when the observed frequency changes while keeping the target position constant.

[0053] In the slow-time pulse dimension, phase changes are primarily caused by the relative motion between the target and the radar. The joint calibration module utilizes radial relative velocity values. To constrain this change. For pulse repetition intervals of... Two adjacent linear frequency modulated pulses, the small displacement of the real static target due to the vehicle's motion will cause a Doppler phase shift. The joint calibration module calculates the theoretical slow-time phase gradient. The calculation formula is as follows: ; in, The center carrier frequency of the radar. The pulse repetition period is represented by this gradient value, which characterizes the rate of phase rotation of the echo signal due to the passage of time, assuming the vehicle's current motion state.

[0054] The joint calibration module combines the gradient components of the two dimensions mentioned above to construct a two-dimensional theoretical phase evolution field. For any... Sub-band and first Each pulse, relative to the theoretical total phase of the reference subband and the reference pulse. It is modeled as a linear superposition of the two gradients mentioned above: ; in, Index of the sampling point subband Pulse index.

[0055] This theoretical phase model establishes a rigorous physical template. It requires that for any signal judged as a true static reflection, the phase transition between its subbands must be proportional to the distance. Locking, while the phase rotation between its pulses must be synchronized with the radial relative velocity value. Lock-on. These two conditions form a strong coupling constraint through radar waveform parameters. Any false target generated by multipath effect will not be able to simultaneously adapt to the evolution law of the two-dimensional phase gradient because its apparent range and apparent velocity do not satisfy the geometric relationship of a single reflection.

[0056] See attached document Figure 2 After the joint calibration module establishes the theoretical phase evolution model, it is necessary to extract the corresponding physical measurement values ​​from the actual received echo signals in order to perform subsequent consistency comparison.

[0057] The joint calibration module reads the subband signal set generated by the subband decomposition module. For each distance cell to be analyzed The joint calibration module does not directly process the absolute phase of a single sub-band because the absolute phase includes random initial phase and noise interference, making it difficult to use directly for precise calibration. This module is configured to perform conjugate difference operations between sub-bands to extract relative phase features.

[0058] Specifically, the joint calibration module selects data from two adjacent sub-bands. and (in The sub-bands are numbered from 1 to ). In each slow-time pulse index and the selected reference antenna channel Next, perform a complex conjugate multiplication operation to generate the measurement differential phasor. The mathematical expression for this operation is: ; in, Indicates the first The complex conjugate of the two sub-band signals. This operation mathematically eliminates the random initial phase and most of the systematic phase error shared by the two sub-band signals, while preserving the complex conjugate caused by the frequency difference. The resulting phase difference term.

[0059] The generated measurement differential phasor It is a complex numerical sequence. The phase angle of this sequence This represents the actual observed phase difference between subbands. This measured phase difference contains two parts of physical information: First, there is the geometric phase difference, which is determined by the actual scattering center position of the target. This component should be constant or linearly change with distance in the case of an ideal point target.

[0060] Secondly, there is the Doppler phase remnant caused by the target's motion in the slow time dimension. Because the center frequencies of different sub-bands are different, the Doppler frequency shift produced by the same radial velocity varies slightly across different sub-bands. This difference is retained after conjugate multiplication. In the phase evolution.

[0061] To capture the phase gradient in the slow time dimension, the joint calibration module maintains the above calculations across all pulse indices. To maintain integrity, no summation or compression of the pulse dimension is performed. Therefore, for each range cell, the system constructs a sub-band interferometric phase history vector that varies with time (pulse). This vector records the true phase response trajectory of the target echo under broadband frequency step and long-time pulse observations, serving as input data for subsequent coherent projection with the theoretical model. For multipath interference signals, since they are composed of superimposed signals from multiple paths, the phases of each path exhibit nonlinear destructive or constructive interference at different frequency sub-bands, leading to variations in the generated measurement differential phasor. The phase angle in the sequence It exhibits irregular jitter or drifting away from the theoretical trajectory.

[0062] After obtaining the measured differential phasor and the theoretical phase evolution model, the joint calibration module performs the core coherent accumulation operation to quantitatively evaluate the physical authenticity of the echo signal.

[0063] For each distance cell, the joint calibration module performs full-dimensional matched filtering between the measured phase difference sequence and the theoretical phase model based on manifold constraints. This process essentially involves coherent detection within the two-dimensional sub-band frequency pulse time domain. The joint calibration module constructs the manifold consistent coherence coefficient (MCC). The calculation of this coefficient traverses all possible assumed azimuth values. .

[0064] The specific calculation process is as follows: The joint calibration module first measures the differential phasor. Phase compensation is performed. This module utilizes the theoretical total phase. The complex conjugate form of the equation is used to weight the measurements to eliminate the theoretical phase rotation caused by real physical motion and distance. If the target is a real target and the angle is assumed... If correct, the compensated phasors will point in the same direction in the complex plane (i.e., phase aligned). Subsequently, the module will index all subbands. and pulse index The phasors after compensation are summed using complex numbers.

[0065] Manifold Consistency Coherence Coefficient The mathematical expression for it is defined as: ; in, This indicates the operation of taking the modulus of a complex number. The imaginary unit, The natural exponential function. The denominator is the sum of the amplitudes of all measured differential phasors, used for normalization to ensure the accuracy of the calculation results. It falls within the interval [0,1].

[0066] This calculation formula reveals the physical nature of the signal through double accumulation: The summation term in the numerator utilizes the frequency diversity characteristics between subbands and the time coherence characteristics between pulses.

[0067] For a direct reflection signal (real target) that satisfies the Doppler manifold constraint, the phase of each subband and each pulse after subtracting the theoretical phase... Afterwards, the residual phase approaches zero or a constant, and the complex vectors are superimposed in phase during accumulation, resulting in... Approaching 1.

[0068] For multipath interference signals (false targets), since their propagation path includes one or more non-line-of-sight reflections, their equivalent range and Doppler velocity cannot simultaneously satisfy the conditions set by... The defined linear coupling relationship. This leads to different subbands In addition, phase compensation not only fails to align the phases but also introduces a phase deflection that varies with frequency. During complex accumulation, these phase-mismatched vectors cancel each other out, causing a decrease in the magnitude, which in turn leads to... Approaching 0.

[0069] The joint calibration module has calculated all assumed angles. After the coherence coefficient is reduced, extract The maximum value in the angular dimension is taken as the final manifold consistency score for the target within that range cell. This score not only reflects the target's signal strength but also directly quantifies the physical stability of the target signal under wideband and long-term observation, forming the core criterion for subsequently distinguishing real targets from false interference.

[0070] After performing phase verification of the signal in the time-frequency dimension using the manifold consistency coherence coefficient, this embodiment further utilizes the geometric features in the spatial dimension to identify the physical properties of the target.

[0071] The spatial spectrum consistency detection module receives the output from the subband decomposition module. Sub-band data matrix This module is configured to perform direction-of-arrival estimation independently for each sub-band of data to observe the stability of the target's spatial angular response at different carrier frequencies.

[0072] Regarding the first Sub-band data Its corresponding center frequency is The spatial spectrum consistency detection module first constructs the spatial covariance matrix of this sub-band. For the selected distance unit Using the sampled data of this unit in the fast time dimension, the correlation between antenna channels is calculated: ; in, For the first The height of the belt is at a distance place 3D snapshot data vector, This indicates the conjugate transpose operation. This represents the expected value calculation, which is usually achieved by averaging multiple snapshots in practice.

[0073] Subsequently, the module uses either a digital beamforming (DBF) algorithm or a multiple signal classification (MUSIC) algorithm to calculate the first... Spatial spectral function of each sub-band To ensure the physical accuracy of the angle estimation, the array guide vector used in this step... It must be based on the specific center frequency of that sub-band. A design is implemented to compensate for the impact of frequency variations on the antenna phase center spacing (normalized to wavelength).

[0074] ; in, The speed of electromagnetic wave propagation. The imaginary unit, The physical spacing between the receiving antenna elements. Total number of receiving antennas. The spatial spectrum consistency detection module searches for the spatial spectrum function. The peak value was obtained. Estimated target angle under the band .

[0075] This module iterates through all Each sub-band generates a set of angle estimation sequences, which are then constructed into a spatial spectral drift vector. For a real physical point target, its electromagnetic scattering center is relatively stable within the radar's operating bandwidth; therefore, the estimated angle values ​​for each sub-band are... The signals should be highly consistent, with only minor statistical errors caused by signal-to-noise ratio limitations. However, for false targets generated by multipath effects, their echoes are typically formed by the interference and superposition of signals from multiple reflection paths at the receiver. Due to the different lengths of each path, the relative phase between the paths rotates with the change of subband frequency, causing the equivalent phase center after superposition to undergo abrupt jumps or flickering in the angular domain.

[0076] To quantify this physical phenomenon, the spatial spectrum consistency detection module calculates the statistical dispersion of the spatial spectrum drift vector. : ; in, The arithmetic mean of all sub-band angle estimates. No. The instantaneous value of the second observation or estimate. This statistical dispersion. It directly reflects the spatial stability of the target in the frequency domain, serving as an important dimension for determining whether a target is a multipath virtual image. Larger values... This indicates a conflict in the spatial position perception of the target at different frequencies, which is likely due to non-physical interference virtual images.

[0077] In the decision output module, the system maps the physical layer feature parameters calculated in the previous steps into a unified probability space, and generates a suppression mask for multipath interference accordingly.

[0078] The decision output module first receives the manifold consistency coherence coefficient calculated for each target (or distance cell) to be detected. Sub-band spatial spectrum drift statistical dispersion Since these two physical quantities have different dimensions and numerical ranges, the module first performs a normalization mapping operation to convert them into dimensionless multipath characteristic components.

[0079] For manifold consistency coherence coefficient Its value range is [0,1], and the closer the value is to 1, the more the target matches the true physical reflection characteristics. The decision output module constructs the first feature component. Characterizing phase inconsistency: ; Statistical dispersion of subband spatial spectral drift Its value is a non-negative real number, and the larger the value, the more unstable the spatial position of the target (i.e., the higher the probability of a multipath virtual image). The decision output module introduces an exponential decay mapping function to construct the second feature component. Characterizing spatial divergence: ; in, Natural exponential function, This is a preset sensitivity adjustment coefficient used to control the slope of the contribution of spatial drift to the multipath probability. This mapping function maps the unbounded variance value to the interval [0,1], ensuring the scale uniformity of the feature components.

[0080] Subsequently, the decision output module performs weighted fusion of the two feature components to calculate the comprehensive multipath confidence score of the target. The scoring uses a linear weighted model: ; in, and These are the phase feature weights and spatial feature weights, respectively, satisfying... The specific weight values ​​are set according to the radar's application scenario. In open road scenarios, phase consistency features are more reliable, and larger values ​​can be set. Value; In narrow tunnels or scenarios with dense guardrails, the spatial spectral drift characteristics are more significant, and can be appropriately increased. The percentage.

[0081] In obtaining the comprehensive multipath confidence score Then, the decision output module executes the threshold decision logic. The system has a preset multi-path decision threshold. .

[0082] The decision output module will and Perform point-by-point comparison to generate a binary validity mask. : ; The decision output module applies this validity mask to the radar's Constant False Alarm Rate (CFAR) output list. For targets with a mask value of 0, the system determines them to be false targets generated by multipath interference and performs suppression operations. Suppression operations include directly removing the point from the target list or forcibly setting its signal strength value to zero to prevent it from entering subsequent clustering and tracking algorithm modules.

[0083] For a target with a mask value of 1, the system determines it to be a real physical target. At this point, the decision output module utilizes the calculation... The optimal azimuth angle found during the process This process replaces the coarse angle estimates in the original CFAR detection, resulting in high-confidence point cloud data that has been validated by physical features and corrected for parameters. This mechanism ensures that the final output point cloud not only eliminates false targets but also improves the parameter accuracy of real targets based on broadband coherent accumulation.

[0084] Specific application example: Multipath interference suppression in tunnel scenarios Scenario Construction: This embodiment selects a typical urban tunnel driving scenario.

[0085] Radar parameters: carrier frequency 77GHz, bandwidth 2GHz, sampling period 50ms.

[0086] Vehicle status: The vehicle is traveling straight along the tunnel, with a longitudinal speed of... =60km / h≈16.7m / s.

[0087] Target settings: Target A: A disabled vehicle that is stationary ahead in the tunnel.

[0088] False Target (Target B): A multipath virtual image formed by radar signals reflected from the tunnel sidewalls (metal plates).

[0089] Implementation process and data analysis: Wideband subband decomposition: After receiving the echo signal, the system applies a sliding window in the range-frequency domain to decompose the total 2GHz bandwidth into subbands. =5 sub-bands (sub-band indices 1-5).

[0090] Manifold construction and model calculation: The system reads the vehicle body CAN bus data to construct a static environment Doppler manifold surface and calculates the theoretical phase evolution model.

[0091] Feature extraction reference appendix Figure 3 The system estimates the direction of arrival (DOA) for each sub-band signal and calculates the statistical dispersion of spatial spectral drift. In the figure, the horizontal axis represents the frequency sub-band index, and the vertical axis represents the estimated DOA value (in degrees).

[0092] Real target A ( Figure 3 (Solid line, circular marker): Since the electromagnetic scattering center of a real vehicle target is relatively stable within a 2GHz bandwidth, its estimated angle across the five sub-bands remains essentially on a straight line near 0° with minimal fluctuation. The calculated spatial spectrum drift statistical dispersion is only 0.05.

[0093] False target B ( Figure 3 (Dashed line, triangle marking): Because multipath signals are formed by the interference of multiple paths, their equivalent phase center changes drastically with frequency. For example... Figure 3 As shown, the estimated angle of the target between sub-band 1 and sub-band 5 exhibits abrupt oscillations (i.e., angular flicker). The calculated spatial spectrum drift statistical dispersion is as high as 0.88. This visual difference verifies the physical basis of this invention's method of distinguishing between real and false targets using spatial spectrum drift.

[0094] Fusion decision: Combining the manifold consistency coherence coefficient (MCC), the system generates the final multipath confidence score.

[0095] Real target A: High MCC, low drift, low score, and retention.

[0096] False target B: Low MCC, high drift, high score, generates invalid mask.

[0097] Experimental verification and effect comparison (combined) Figure 4 illustrate): To visually verify the beneficial effects of the technical solution of this invention, we conducted a real-vehicle test in a curved road scenario including metal guardrails, and plotted the processing results as follows. Figure 4 .

[0098] Comparative analysis: Please refer to Figure 4 In the diagram, the horizontal axis represents the horizontal distance (meters), the vertical axis represents the vertical distance (meters), the triangle icon represents the vehicle's position, and the circle icon represents the actual target.

[0099] Results of traditional methods ( Figure 4 (Left sub-image): When using the traditional CFAR (Constant False Alarm Detection) method, although the radar detects the real target (gray solid circle) 50m ahead, a large number of false point cloud targets appear on both sides of the road and ahead (as shown by the square markers). These false targets are mainly generated by multipath reflections from guardrails, and some virtual images are even projected onto the vehicle's driving path (e.g., points near coordinates (5,55)), which can easily cause the autonomous driving system to misjudge that there is an obstacle ahead and trigger false braking.

[0100] Detection results of the method of the present invention ( Figure 4 (Right subgraph): After applying the anti-multipath interference processing of this invention, the system performs suppression decisions based on the multipath confidence score. For example... Figure 4 As shown on the right, most of the original spurious interference (now marked with an "x" to indicate that it has been removed or suppressed) has been successfully filtered out.

[0101] Authenticity preserved: The real target (gray solid dot) 50m ahead was unaffected, and its signal strength and location information were fully preserved.

[0102] False alarm rate reduced: As can be seen from the comparison of the left and right images, the environmental point cloud processed by the method of this invention is cleaner and clearer, and ghost images on the road are effectively eliminated.

[0103] Combination Figure 3 Microscopic feature analysis and Figure 4 To improve the macroscopic scene effect, this invention introduces physical-level manifold constraints and broadband spatial stability detection, which reduces the false alarm rate while ensuring the real target detection rate, effectively solving the perception problem of vehicle radar in complex multipath environments.

Claims

1. A method for processing vehicle-mounted millimeter-wave radar signals to resist multipath interference, characterized in that, Includes the following steps: Acquire radar echo signals and decompose the radar echo signals into multiple sub-band signals that partially overlap in frequency band in the frequency domain; Acquire vehicle motion state data, construct a static environment Doppler manifold surface, and calculate a theoretical phase evolution model for the target to be detected based on the static environment Doppler manifold surface; The actual measured phase difference between adjacent sub-band signals is extracted, and the actual measured phase difference is coherently accumulated with the theoretical phase evolution model to generate a manifold consistency coherence coefficient. Direction of arrival (DOA) estimation is performed for each of the sub-band signals, and the spatial spectral drift statistical dispersion of the angle estimates is calculated. Based on the manifold consistency coherence coefficient and the spatial spectrum drift statistical dispersion, a target multipath confidence score is generated, and a suppression decision is performed on the target to be detected. The steps for calculating the theoretical phase evolution model for the target to be detected based on the static environment Doppler manifold surface include: The first phase gradient in the fast time frequency dimension is calculated based on the distance to the target to be detected and the center frequency difference between adjacent sub-band signals. Based on the theoretical radial relative velocity corresponding to the azimuth angle of the target to be detected on the Doppler manifold surface of the static environment, and the pulse repetition period, the second phase gradient of the slow-time pulse dimension is calculated. The first phase gradient and the second phase gradient are linearly combined to construct the theoretical phase evolution model that varies with the sub-band index and the pulse index. The step of extracting the actual measured phase difference between adjacent sub-band signals includes: Two adjacent sub-band signals in the frequency domain are selected, and complex conjugate multiplication is performed under the same distance cell, antenna channel, and pulse index. The sequence results of the complex conjugate multiplication operation at all pulse indices are retained to generate a sequence of measured differential phasors that varies with pulse time; The phase angle of the measured differential phasor sequence is extracted as the actual measured phase difference; The steps for calculating the spatial spectral drift statistical dispersion of the angle estimate include: For each of the sub-band signals, an array steering vector is constructed based on the center frequency of the sub-band signal; The spatial spectrum function of the sub-band signal is calculated using the array steering vector, and the peak value of the spatial spectrum function is searched to obtain the angle estimate value corresponding to the sub-band signal; Calculate the variance of the vector consisting of all the angle estimates, and determine the variance as the statistical dispersion of the spatial spectrum drift.

2. The vehicle-mounted millimeter-wave radar signal processing method for resisting multipath interference according to claim 1, characterized in that, The step of decomposing the radar echo signal into multiple sub-band signals that partially overlap in frequency band in the frequency domain includes: Perform a fast Fourier transform on the radar echo signal in the fast time dimension to generate a range frequency domain signal; Multiple frequency segments are extracted from the distance frequency domain signal by applying a frequency domain sliding window or bandpass filter bank with a preset bandwidth. Each frequency segment is padded with zeros or truncated to construct sub-band signals with different center frequencies, wherein the frequency coverage of adjacent sub-band signals overlaps.

3. The method for processing vehicle-mounted millimeter-wave radar signals to resist multipath interference according to claim 1, characterized in that, The steps for constructing the static environment Doppler manifold surface include: Acquire longitudinal velocity, lateral velocity, and yaw rate of the vehicle's motion state data; By combining the installation position vector and installation rotation matrix of the radar sensor, the instantaneous velocity vector of the radar sensor in the world coordinate system is calculated; For any azimuth and elevation angle within the radar field of view, the projection value of the instantaneous velocity vector in the corresponding direction is calculated, and the negative value of the projection value is determined as the theoretical radial relative velocity of the static target in the direction, thereby forming the Doppler manifold surface of the static environment.

4. The vehicle-mounted millimeter-wave radar signal processing method for resisting multipath interference according to claim 1, characterized in that, The step of coherently accumulating the actual measured phase difference with the theoretical phase evolution model to generate the manifold consistency coherence coefficient includes: The measured differential phasor sequence is weighted using the complex conjugate form of the theoretical phase evolution model to obtain a phase-compensated phasor sequence; Complex summation is performed on the phase-compensated phasor sequence across all subband and pulse index dimensions, and the modulus of the summation result is calculated. The sum of the amplitudes of all the measured differential phasor sequences is calculated, and the magnitude is normalized using the sum of the amplitudes to obtain the manifold consistency coherence coefficient.

5. The vehicle-mounted millimeter-wave radar signal processing method for resisting multipath interference according to claim 1, characterized in that, The steps for generating a target multipath confidence score based on the manifold consistency coherence coefficient and the spatial spectrum drift statistical dispersion include: The manifold consistency coherence coefficient is converted into a phase inconsistency eigenvalue using a first mapping function, wherein the higher the manifold consistency coherence coefficient, the lower the phase inconsistency eigenvalue. The spatial spectrum drift statistical dispersion is converted into a spatial divergence characteristic value using a second mapping function, wherein the higher the spatial spectrum drift statistical dispersion, the higher the spatial divergence characteristic value. The target multipath confidence score is obtained by weighted summation of the phase inconsistency eigenvalues ​​and the spatial divergence eigenvalues.

6. The vehicle-mounted millimeter-wave radar signal processing method for resisting multipath interference according to claim 1, characterized in that, The steps for performing suppression decisions on the target to be detected based on the target multipath confidence score include: Set a multipath determination threshold; If the target multipath confidence score is greater than the multipath determination threshold, an invalid mask is generated for the target to be detected. The invalid mask is applied to the target list output by the radar constant false alarm rate detection, and the target to be detected is removed or the signal strength of the target to be detected is set to zero.

7. The vehicle-mounted millimeter-wave radar signal processing method for resisting multipath interference according to claim 1, characterized in that, The radar echo signal is a frequency-modulated continuous wave signal. The data structure of the sub-band signal includes a fast time dimension, a slow time dimension, and a spatial array dimension. The theoretical phase evolution model is used to characterize the physical phase constraints of a real static point target under broadband frequency and long-term observation.

Images