4d millimeter wave radar point cloud feedback composite time-sharing waveform generation method and system

Abstract

This invention discloses a method and system for generating composite time-division waveforms from 4D millimeter-wave radar point cloud feedback. The method reads multiple frames of 4D point cloud and their associated track states, divides them into distance intervals, extracts the demand features for each interval based on the point cloud, track states, and associated results, and forms interval demand values. Using a complete multi-input multi-output measurement group as the smallest allocation unit, the method allocates frame duration budgets according to a preset lower limit and the ratio of the relative occupancy time for demand-satisfying gain, and generates composite time-division waveform configurations. The waveform configuration identifier simultaneously associates the configuration with the generation, acquisition, and processing parameters, taking effect at frame boundaries. Then, each subframe is calculated according to the identifier, and the point cloud is compensated to the same reference time. The system includes modules for partition demand assessment, complete multi-input multi-output measurement group allocation, waveform configuration identifier version submission, and point cloud generation. Using this scheme, composite time-division resources can be adapted to different scenarios, avoiding truncating complete measurement groups and ensuring parameter consistency and point cloud time alignment.

Description

Technical Field

[0001] This invention relates to the field of vehicle-mounted and traffic-sensing millimeter-wave radar technology, specifically to a method for generating composite time-division waveforms from 4D millimeter-wave radar point cloud feedback, and also to a system for generating composite time-division waveforms from 4D millimeter-wave radar point cloud feedback. Background Technology

[0002] 4D imaging millimeter-wave radar typically operates in the 77GHz or 79GHz frequency band, employing frequency-modulated continuous wave (FM) and a multiple-input multiple-output (MIMO) system. Within a measurement frame, the radar transmits multiple linearly modulated (LFM) signals. After mixing, analog-to-digital conversion, range processing, Doppler processing, and two-dimensional angle measurement, a 4D point cloud is formed, containing range, radial velocity, azimuth, elevation, amplitude, and confidence level. The MIMO system requires that each transmitted or coded state appear as a complete set; otherwise, the virtual array channels will be incomplete, and angle calculations will lose their consistent array manifold.

[0003] Composite time-division waveforms can be configured with multiple subframes within the same measurement frame to serve different distance ranges; however, if the subframe parameters and resource share are determined only by a fixed template or a single target number, they still cannot adapt to changes in the range requirements reflected by the actual point cloud, track status, and the correlation results between the two.

[0004] In existing technologies, a type of fixed waveform or multi-mode preset composite frame scheme has been proposed: the radar determines the bandwidth, slope, sweep duration, linear frequency modulation repetition period, frame period, and intra-frame linear frequency modulation pulse number during the calibration or manufacturing stage, or presets different linear frequency modulation configurations such as long-range, medium-range, and short-range, and transmits them in a fixed ratio or fixed order. This type of scheme can accommodate different detection ranges, but it usually does not adjust the subframe resource ratio in real time according to the demand-related characteristics obtained from point cloud and track feedback for each range interval. The resource allocation of each interval is fixed in the design stage; if the continuous resource ratio is directly converted into the number of linear frequency modulation pulses, it may also truncate the complete multi-input multi-output measurement group and destroy the virtual array or multi-input multi-output decoding matrix.

[0005] In existing technologies, an adaptive scheme driven by vehicle state or single target state has been proposed: waveform parameters are adjusted based on vehicle speed, heading angular velocity, single target distance, signal-to-noise ratio, interference intensity, or number of targets. This type of scheme can improve detection capabilities under specific working conditions. However, in scenarios with the same number of targets, waveform resource requirements may be completely different due to differences in space occupation, correlation quality, track uncertainty, and velocity dispersion. It is difficult to accurately characterize the interval requirements based solely on the total number of targets or a single state quantity, and measurement resources are easily allocated to low-demand intervals.

[0006] In existing technologies, another type of point cloud post-processing enhancement scheme has been proposed: under a given waveform, point cloud quality is improved through constant false alarm rate (CFAR) improvement, super-resolution angle estimation, clustering, multi-frame accumulation, or track filtering. However, the processed 4D point cloud and track status are not fed back to the waveform generation stage of the next measurement frame, and the waveform and processing result do not form a closed loop. Furthermore, if the center times of multiple subframes are different, directly merging the point clouds without unified reference time compensation will introduce position misalignment; if different versions of parameters are used at the waveform generation end, analog-to-digital conversion acquisition end, and signal processing end, distance mapping errors and virtual channel mismatches will occur; if the generation configuration lacks hardware executability verification, it may not be executable on hardware; frequent switching near the threshold will also affect the stability of subsequent correlation.

[0007] In summary, existing technologies still have the following shortcomings: they use fixed multi-mode switching or only switch waveforms based on vehicle speed and total number of targets, failing to differentiate interval requirements based on point cloud, track status, and associated results; direct conversion of continuous resource ratios may truncate complete multi-input multi-output measurement groups; the versions of generation, acquisition, and processing parameters may be inconsistent; composite subframe point clouds lack unified reference time compensation; and there is a lack of hardware executability verification and smooth switching mechanisms to support closed-loop adaptation. Therefore, how to integrate multi-frame 4D point clouds with track feedback, discrete allocation of complete multi-input multi-output measurement groups, consistent submission of waveform configuration identifiers, and unified reference time compensation into a feasible closed loop is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to provide a composite time-division waveform generation method and system for 4D millimeter-wave radar point cloud feedback. This method extracts relevant features of interval requirements based on point cloud, track status, and their correlation, forming interval requirement values. Using a complete multi-input multi-output measurement group as the smallest allocation unit, it allocates frame duration budgets according to the ratio of a preset lower limit quantity to the relative duration of gain required to meet the requirements. Finally, it establishes a closed loop through consistent waveform configuration identification and unified reference time compensation. This solves the problems of fixed or multi-mode waveforms being difficult to adapt to scene perception features, continuous proportional conversion easily truncating complete measurement groups, and composite subframe parameter mismatch and point cloud time misalignment.

[0009] To achieve the above objectives, the technical solution provided by this invention is as follows: This invention provides a method for generating composite time-division waveforms based on 4D millimeter-wave radar point cloud feedback, comprising: reading 4D millimeter-wave radar point clouds and their associated track states from multiple recent measurement frames; dividing the detection space into at least two distance intervals; extracting demand-related features for each distance interval based on the point cloud, the track state, and the correlation results between the point cloud and the track, forming interval demand values; allocating no less than a preset lower limit number of complete multiple-input multiple-output measurement groups to each distance interval based on the interval demand values, using complete multiple-input multiple-output measurement groups as the minimum allocation unit; allocating the remaining frame duration budget group by group according to the ratio of the demand satisfaction gain brought by each additional complete multiple-input multiple-output measurement group to the duration occupied by that measurement group, generating composite time-division waveform configurations; using waveform configuration identifiers to simultaneously associate the composite time-division waveform configurations with waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters, and switching to the currently used configuration at frame boundaries; calculating the acquired data of each subframe according to the waveform configuration identifiers, and compensating the point clouds of each subframe to the same reference time. In the above technical solution, point cloud and track feedback can drive the limited frame budget to tilt towards the high demand range, the discrete allocation of complete measurement groups ensures the integrity of the virtual array channels, and the consistent submission of waveform configuration identifiers and unified time compensation ensure parameter consistency and point cloud time alignment.

[0010] As a preferred embodiment, the extraction of demand-related features for each distance interval and the formation of interval demand values ​​specifically includes: extracting space occupancy-related features based on the point cloud; extracting track stability and uncertainty-related features based on the track status; extracting association quality-related features based on the association results between the point cloud and the track; extracting velocity dispersion-related features; normalizing and weighting the above features to obtain the interval demand value; and introducing the historical track existence quantity obtained from the track status as a safety net term to obtain a normalized demand component for resource allocation. Specifically, space occupancy-related features may include the proportion of occupied grid cells; track stability and uncertainty-related features may include the number of stable tracks, historical missed detection rate, and uncertainty characterized by the track status covariance; association quality-related features may include the proportion of unassociated valid points and the proportion of newly generated tracks. In the above technical solution, the various features clearly correspond to their data sources, enabling the differentiation of scenarios with similar point counts but different spatial distributions, association states, and degrees of uncertainty.

[0011] As a preferred embodiment, the complete multiple-input multiple-output (MIMO) measurement group is the minimum and complete set of transmit states or encoded states necessary to recover a predetermined virtual array or decoding matrix. Its specific composition is determined by the adopted MIMO separation system. The separation system includes parallel systems such as time-division multiplexing, binary phase modulation, and Doppler diversity multiple access: under the time-division multiplexing system, each transmit channel transmits in turn, and a complete MIMO measurement group includes one transmission from each of all active transmit channels; under the binary phase modulation system, each transmit channel is distinguished by binary phase coding; under the Doppler diversity multiple access system, each transmit channel is distinguished by different Doppler biases. In the latter two systems, a complete MIMO measurement group is the minimum complete set of encoded states or transmit states required to recover all virtual channels. Because the complete MIMO measurement group is used as the minimum allocation unit, resource allocation does not truncate the complete measurement group with non-integer quantum truncation; the number of output linear frequency modulated pulses and the number of transmit states are naturally integers, preventing incomplete measurement cycles due to proportional rounding, thus ensuring the integrity of the virtual array channels.

[0012] As a preferred approach, before generating the composite time-division waveform configuration, for distance intervals that meet the activation conditions, candidate parameters that simultaneously satisfy distance resolution, beat frequency, speed ambiguity, and frame-level hardware executable constraints are selected from the set of waveform parameters supported by the hardware. Before the composite time-division waveform configuration takes effect, hardware executableness verification is performed on the entire frame configuration. If the verification fails, the following steps are taken: replacing candidate parameters for the current interval, reducing the measurement groups for the current interval, closing the lowest priority unsafe interval, and reverting to safe waveforms. The candidate screening process verifies the effective sampling bandwidth and distance resolution, maximum beat frequency, linear frequency modulation repetition period and speed ambiguity for each enabled interval, as well as frame-level constraints such as frame duration, duty cycle, storage, processing throughput, and thermal properties. The effective sampling bandwidth can be determined by the product of the frequency modulation slope and the sampling duration. The maximum beat frequency can be calculated based on the worst-case scenario where the distance and Doppler terms have the same sign, and must not exceed the upper limit allowed by both the intermediate frequency passband and the sampling rate. Candidate parameters that do not meet the product's speed range requirements can be deleted, or a multi-repetition-period deambiguation template already implemented at the processing end can be used instead. Full-frame verification items may include RF sweep range, phase-locked loop slope, sampling rate, number of sampling points, sampling start point, intermediate frequency passband, transmission sequence, duty cycle, frame period, streaming buffer write bandwidth, digital signal processing cycle, and temperature rise. In the above technical solution, the pre-generation screening and pre-effective full-frame verification are interconnected, reducing the risk of unexecutable configurations entering the hardware.

[0013] As a preferred embodiment, the method of simultaneously associating the composite time-division waveform configuration with waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters specifically includes: writing the waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters into the spare parameter area of ​​each module, all three using the same version number and checksum; after all relevant modules have confirmed the writing is complete, switching the spare parameter area to the currently used configuration at the frame boundary; if any module fails to write or has an inconsistent version, maintaining the currently used configuration; if the processing end detects an inconsistency in parameter version or checksum, marking the corresponding subframe as invalid. The frame boundary can be a subframe boundary or a measurement frame boundary; the information used by the processing end to determine inconsistency may include waveform configuration identifier, linear frequency modulation pulse index, number of sampling points, transmission order, or checksum. This technical solution avoids distance mapping errors and virtual channel mismatches caused by mismatched versions of acquisition and processing parameters.

[0014] As a preferred solution, the compensation of point clouds in each subframe to the same reference time specifically includes: for points already associated with a track, compensating the points to the reference time by combining the vehicle's pose and track motion; when reliable track motion information is unavailable, compensation is performed only based on the vehicle's pose; for points not associated with a track, approximate extrapolation is performed along the line of sight using radial velocity, and their fusion weight is reduced; after time alignment is completed, duplicate points are merged. Specifically, for already associated points, a method can be adopted whereby the coordinates are first transformed to global coordinates, then extrapolated using track velocity, and finally transformed back to the radar coordinates of the reference time; when merging duplicate points, thresholds can be set based on spatial distance, velocity difference, angle difference, and whether the waveform configuration identifier is valid. In the above technical solution, motion misalignment introduced by different center times of composite subframes can be reduced, lowering the probability of misassociation of tracks.

[0015] As a preferred approach, smoothing constraints are applied to interval start / stop and parameter switching. Specifically, interval start / stop requires multiple consecutive frames to meet the corresponding threshold conditions before switching; after waveform configuration switching, a minimum preset number of frames must be maintained before switching again, and this preset number of frames is not less than the number of consecutive frames required for start / stop determination; between adjacent measurement frames, the changes in waveform parameters and the number of measurement groups in each interval do not exceed their respective preset upper limits. Start / stop can employ dual thresholds, with the number of consecutive determination frames denoted as K and the minimum number of frames remaining as L, satisfying L≥K; the upper limits for continuous parameter changes and the upper limits for changes in the number of measurement groups can be calibrated separately. In the above technical solution, dual thresholds, consecutive frame determination, minimum remaining time, and change rate limits reduce the risk of waveform jitter entering the hardware.

[0016] As a preferred solution, when one of the following situations occurs—preset lower limit allocation being infeasible, feedback loss, or hardware malfunction—the system reverts to a pre-verified safe waveform. In cases requiring emergency switching, the system is allowed to switch directly to the safe waveform without being constrained by the upper limit of parameter changes in adjacent measurement frames. The infeasibility of preset lower limit allocation may include the total occupancy time corresponding to the preset lower limit quantity exceeding the frame duration budget; situations requiring emergency switching may include safety alarms, collision risks, or hardware failures. In the above technical solution, basic detection capabilities are maintained even when allocation is infeasible or the system malfunctions.

[0017] This invention also provides a composite time-division waveform generation system for 4D millimeter-wave radar point cloud feedback, comprising: an antenna array and transceiver front-end 10; a waveform generator 20; an analog-to-digital conversion sampling and preprocessing module 30; a partitioned demand assessment module 40, used to read the 4D millimeter-wave radar point cloud and its associated track status from multiple recent measurement frames, divide the detection space into at least two distance intervals, extract demand-related features for each distance interval based on the point cloud, the track status, and the correlation results between the point cloud and the track, and form an interval demand value; and a complete multi-input multi-output measurement group allocation module 60, used to allocate no less than a preset lower limit number of complete multi-input multi-output measurement groups to each distance interval based on the interval demand value, using complete multi-input multi-output measurement groups as the minimum allocation unit, and to allocate the remaining... The frame duration budget is allocated group by group according to the ratio of the gain required for each additional complete MIMO measurement group to the duration occupied by that measurement group, generating a composite time-division waveform configuration. The waveform configuration identifier version submission module 80 uses the waveform configuration identifier to simultaneously associate the composite time-division waveform configuration with waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters, and switches to the currently used configuration at frame boundaries. The 4D millimeter-wave radar point cloud generation module 100 calculates the acquired data of each subframe according to the waveform configuration identifier and compensates the point clouds of each subframe to the same reference time. The track management module 110 outputs the track status and feeds it back to the partition requirement assessment module 40. The radar controller 120 coordinates the above modules to complete the composite time-division waveform generation closed loop. In the above technical solution, the system modules and method steps correspond, supporting a composite time-division waveform generation closed loop driven by point cloud feedback and with consistent parameters and time alignment.

[0018] As a preferred embodiment, the system further includes a candidate waveform filtering module 50, a hardware boundary verification module 70, and a smooth switching module 90. The candidate waveform filtering module 50 and the hardware boundary verification module 70 are used to filter candidate parameters and perform full-frame hardware verification. The smooth switching module 90 is used to implement smooth switching constraints and trigger a safe waveform rollback in cases of allocation infeasibility, feedback loss, hardware anomalies, or security alarms. In the above technical solution, the module responsibilities are clearly defined, facilitating engineering deployment and fault tracing.

[0019] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 The diagram shown is a flowchart of an embodiment of a composite time-division waveform generation method for 4D millimeter-wave radar point cloud feedback provided by the present invention. Figure 2 The diagram shown is a schematic representation of an embodiment of a composite time-division waveform generation system for 4D millimeter-wave radar point cloud feedback provided by the present invention. Figure 3 The diagram shown is a flowchart of an embodiment of the interval demand assessment provided by the present invention. Figure 4 The diagram shown is a flowchart illustrating an embodiment of the complete multiple-input multiple-output measurement group allocation provided by the present invention. Figure 5 The diagram shown is a flowchart illustrating an embodiment of the waveform configuration identifier version submission provided by the present invention. Figure 6 The diagram shown is a flowchart of an embodiment of the unified reference time compensation provided by the present invention.

[0022] Figure labeling: 10, Antenna array and transceiver front end; 20, Waveform generator; 30, Analog-to-digital conversion sampling and preprocessing module; 40, Partition requirement assessment module; 50, Candidate waveform screening module; 60, Complete multi-input multi-output measurement group allocation module; 70, Hardware boundary verification module; 80, Waveform configuration identification version submission module; 90, Smooth switching module; 100, 4D millimeter-wave radar point cloud generation module; 110, Track management module; 120, Radar controller. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] In the description of this invention, it should be noted that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. Unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly; for example, it can refer to a fixed connection, a detachable connection, or an integral connection; it can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0025] It should be noted that the 4D millimeter-wave radar point cloud described in this invention refers to a point cloud that includes at least three-dimensional position information and radial velocity information; the complete output may further include attributes such as amplitude, confidence level, azimuth angle, and pitch angle. The track management module 110 only provides this invention with predicted position, velocity, probability of existence, covariance, and number of consecutive missed detections, without limiting its internal tracking algorithm. The safety waveform refers to a fixed backoff waveform that has been verified offline, covers the basic detection range, and satisfies all hardware boundaries, denoted as... P safe .

[0026] In this embodiment of the invention, the meanings of each symbol are as follows: z — Distance interval index, z∈Z, Z contains at least the near field interval and the far field interval; p i —The i-th point in the point cloud, including distance r i ,position i i , up and down f i radial velocity v i Amplitude a i Confidence level c i and time t i ; x j , P j —The state vector and state covariance of the j-th track; D z , R z —Interval demand values ​​and normalized resource demand components; n z —The number of complete multi-input multi-output measurement groups assigned to interval z; Q z—A complete multi-input multi-output measurement group in interval z; Waveform configuration identifier —A configuration identifier that binds subframe acquisition data with a unique waveform and processing parameter version; B eff,z —Effective sampling bandwidth of interval z; S z —The frequency modulation slope of the z-interval; F s,z , N s,z —The sampling rate and number of sampling points in interval z; T budget —Duration budget for composite time-division frames; T Q,z —The duration of a complete multi-input multi-output measurement set in interval z; M—The most recent measurement frame involved in the feedback decision; P safe —Pre-verified security waveform configuration.

[0027] Example 1 This embodiment provides an overall flowchart of a composite time-division waveform generation method for 4D millimeter-wave radar point cloud feedback. See [link to relevant documentation]. Figure 1 .

[0028] Step 101, read valid feedback. Radar power-on, waveform configuration identifier version verification failed, continuous... M loss When there is no valid feedback or a hardware alarm occurs, the radar controller 120 directly calls the pre-verified safety waveform. P safe Under normal conditions, the partition requirement assessment module 40 reads the point cloud and track data from the most recent M measurement frames; M ranges from 4 to 20, with 8 to 16 being preferred. Feedback exceeding the allowable delay is not included in the current decision to avoid outdated point clouds and tracks misleading waveform switching.

[0029] Step 102: Divide the detection space and statistically analyze its features. Divide the detection space into near-field, mid-field, and far-field regions based on distance thresholds. Each distance interval is then gridded according to azimuth and elevation angles. Isolated points are first eliminated based on detection thresholds, static clutter suppression, and neighborhood consistency. The grid occupancy ratio is calculated based on point cloud statistics. O z ;Statistical stable track count based on track status H z Historical false negative rate F miss,z Uncertainty characterized by track position covarianceU cov,z and the amount of historical flight tracks E z The proportion of unassociated valid points is calculated based on the correlation results between point clouds and flight paths. A z and the proportion of new flight paths N new,z And statistically analyze the velocity dispersion. V disp,z Each statistic employs an exponential moving average or a sliding window of length M to avoid single-frame abrupt changes directly altering the waveform.

[0030] Step 103: Calculate the interval demand and form the activation set. O z , H z , U z , V disp,z and F miss,z Normalized according to interval design capacity or moving quantile respectively o z , h z , u z , v z and f z All values ​​are truncated to [0,1]. Interval activation uses... i on and i off Dual threshold, i on > i off The start / stop state is switched only after K consecutive frames meet the conditions, thus obtaining the set of enabled intervals. Z on .

[0031] Step 104, Filter candidate parameters. The candidate waveform filtering module 50 selects candidate parameters from the set of waveform parameters supported by the hardware for each z∈ Z on Candidate parameters that meet the constraints of distance resolution, beat frequency, speed ambiguity, and frame-level hardware executableness are selected, as detailed in Example 3.

[0032] Step 105, assign according to complete MIMO measurement groups. The complete MIMO measurement group assignment module 60 assigns complete MIMO measurement groups. Q z To allocate the minimum resource quantum, first allocate a preset lower limit. nfloor,z Then, the demand-satisfaction gain is allocated to each group according to the ratio of demand satisfaction gain to occupation time; the demand satisfaction gain can be calculated using the marginal utility model, see Example 4 for details.

[0033] Step 106: Perform full-frame hardware verification. The hardware boundary verification module 70 verifies the RF sweep range, phase-locked loop slope, sampling rate, number of sampling points, sampling start point, intermediate frequency passband, transmission sequence, duty cycle, frame period, streaming buffer write bandwidth, digital signal processing cycle, and temperature rise; in case of failure, it processes the data in a fixed order and outputs a status code.

[0034] Step 107: Submit waveform configuration identifier version. The waveform configuration identifier version submission module 80 writes the composite time-division waveform configuration into the spare parameter area after binding it with the waveform configuration identifier. After the three modules have completed writing and the versions are consistent, the configuration is switched to the currently used configuration at the frame boundary. See Example 5 for details.

[0035] Step 108: Solve and unify time compensation according to waveform configuration identifiers. The 4D millimeter-wave radar point cloud generation module 100 uses the slope, sampling time, repetition period, and transmission sequence corresponding to each waveform configuration identifier to complete range, velocity, azimuth, and elevation estimations. It does not mix data from different waveform configuration identifiers before range fast Fourier transform or virtual array formation, and compensates the point cloud to a unified reference time. t ref See Example 6 for details.

[0036] Step 109, Track Management and Smooth Handover. The track management module 110 outputs track status feedback to the partition demand assessment module 40; the smooth handover module 90 applies a minimum number of resident frames L and a parameter change rate limit. Δx max and Δn max And rollback in case of an anomaly. P safe See Example 7 for details.

[0037] In this embodiment, the method forms a closed loop of "point cloud and track feedback → interval requirements → candidate screening → complete measurement group allocation → hardware verification → waveform configuration identifier submission → separate calculation and time compensation → track feedback", which enables the composite time-division waveform to dynamically adapt to the scene and ensures hardware execution.

[0038] Example 2 This embodiment details the interval demand assessment of the partitioned demand assessment module 40. See [link to relevant documentation]. Figure 3 .

[0039] In a specific numerical example, the near-field interval z=1 corresponds to 0–30 m, the mid-field interval z=2 corresponds to 30–80 m, and the far-field interval z=3 corresponds to 80–200 m. Each interval is divided into a Δθ×Δφ grid within an azimuth ±60° and an elevation ±15° range. The near-field grid size is 2°×2°, the mid-field grid size is 3°×3°, and the far-field grid size is 5°×5°. M=12, and the exponential moving average coefficient α=0.15.

[0040] Track covariance uncertainty U cov,z Take the track position covariance trace within the interval Tr(P j ) Divide by the interval feature scale L z 2 The mean was then truncated to [0,1]. To align with the five normalized components in the interval demand values, the associated quality correlation features and track covariance uncertainty were combined into a single value. U z (Historical false negative rate is not included in this formula to avoid confusion with other formulas.) D z In f z (item weighting) U z =clip( b 1 A z + b 2 N new,z + b 3 U cov,z (0,1) where, b 1 = 0.30 b 2 = 0.30 b 3 = 0.40 b 1~ b 3 represents a non-negative weight with a total weight of 1. When there is no current point cloud but a historical track with a high probability of existence exists, F miss,z and E z The need for detection in this area remains.

[0041] The interval demand value and the normalized resource demand components are calculated as follows: D z =clip( w o o z + w h h z + wu u z + w v v z + w f f z ,0,1) R z =( D z + d 0 E z ) / Σk∈Z on (D k +d 0 E k ) in, o z , h z , u z , v z , f z respectively by O z , H z , U z , V disp,z , F miss,z Normalization yields the result; w o =0.20、 w h =0.15、 w u =0.30、 w v =0.15、 w f =0.20, w o , w h , w u , w v , w f The weights are non-negative and their sum is 1; d 0 = 0.10 is the historical track safety factor. If the denominator is 0, call... P safe .

[0042] Interval activation uses dual-threshold hysteresis: i on =0.35、 i off =0.20. When D z K=3 consecutive frames higher than i on Enabled when set; D z Three consecutive frames below i off Enabled by clearing the time frame. This mechanism prevents frequent starts and stops within a range due to single-frame jitter near the threshold.

[0043] In a congested near-field example, O 1 = 0.72 H 1=6 (The design capacity of the near-field stable track is taken as 8, therefore...) h 1 = min(6 / 8, 1) = 0.75 A 1 = 0.18 N new,1 =0.25、 F miss,1 =0.05、 U cov,1 =0.40、 V disp,1 =0.55. First, get... U 1 = 0.30 × 0.18 + 0.30 × 0.25 + 0.40 × 0.40 = 0.289, then take... o 1 = 0.72 u 1 = 0.289 v 1 = 0.55 f 1 = 0.05, weighted average D 1 = 0.20 × 0.72 + 0.15 × 0.75 + 0.30 × 0.289 + 0.15 × 0.55 + 0.20 × 0.05 ≈ 0.44. (Far-field value) O 3 = 0.12 D 3 = 0.15. The total number of targets is similar for both, but near-field crowding and velocity dispersion are higher. R 1 is significantly greater than R 3. Subsequent measurement group allocation will be biased towards the near field. This example illustrates that the present invention can distinguish between scenarios with similar numbers of points but different spatial occupancy and uncertainties.

[0044] Example 3 This embodiment details the parameter constraint verification of the candidate waveform screening module 50.

[0045] Candidate parameters should include at least the initial frequency. f0,z FM slope S z Sampling rate F s,z Number of sampling points N s,z Analog-to-digital conversion sampling start time, repetitive periodic sequence, number of transmission channels N tx,z Launch sequence and upper limit of measurement groups n max,z .make T s,z = N s,z / F s,z The effective sampling bandwidth and distance resolution constraints are: B eff,z =| S z | T s,z ΔR z ≈ k w c / (2 B eff,z )≤ ΔR req,z in, k w ≥1 indicates main lobe broadening caused by the window function, where c is the speed of light. Near field values ​​are... ΔR req,1 =0.25m k w =1.2; Midfield take ΔR req,2 =0.50m; far field is taken ΔR req,3 =1.00m.

[0046] The maximum beat frequency is checked under the worst-case scenario where the distance and Doppler terms have the same sign: f IF,max,z =2| S z | D req,z / c+2 v max,z / λ+ f margin f IF,max,z ≤min( B IF,usable , or s F s,z )in, B IF,usableFor the passband allowed by both analog and digital intermediate frequency links, or s Calibrated by the sampling structure and filter; 0 < 0 during real number sampling or s <1 / 2; When using a single-sided analytical spectrum for complex orthogonal sampling, 0< can be taken. or s <1, but must not exceed the actual simulated passband. Near field D req,1 =30m v max,1 =35m / s; far field D req,3 =200m v max,3 =55m / s; f margin =2MHz.

[0047] The timing constraints for linear frequency modulation are: T r,z ≥ T pre,z + T s,z + T post,z + T idle,z For time-division multiplexed multiple-input multiple-output (MIMO) systems with a fixed repetition period, the equivalent sampling interval of the same transmit channel is... T eff,z = N tx,z T r,z The initial verification is v amb,z ≈λ / (4 T eff,z ), Δv z ≈ k v λ / (2 N rep,z T eff,z If the product speed range is not met, the candidate parameter is deleted, or a multi-repetition periodic defuzzification template already implemented in the processing end is used instead; the multi-repetition periodic template enumerates the fuzzy order within the target speed range, and is only retained when the optimal candidate and the second-best candidate reach a preset uniqueness margin.

[0048] In a near-field candidate example, S 1 = 120MHz / μs N s,1 =512、 F s,1 =20MHz, thereforeT s,1 =25.6μs B eff,1 =3.072GHz ΔR 1 ≈ 0.058m, which satisfies ΔR req,1 ;according to D req,1 =30m v max,1 =35m / s (77GHz, λ≈3.9mm) and f margin =2MHz, f IF,max,1 The value is approximately 26.02MHz, which is greater than min(25MHz, 0.45×20MHz)=9MHz. Therefore, this candidate is discarded and replaced with a candidate with a lower slope. Candidate screening is performed before the measurement group allocation to ensure that all parameters entering the allocation stage meet the RF and IF boundaries.

[0049] Example 4 This embodiment details the complete multiple-input multiple-output measurement group allocation of the complete multiple-input multiple-output measurement group allocation module 60. (See also...) Figure 4 .

[0050] For time-division multiplexing systems, each transmission channel transmits in turn during the same time period, forming a complete multi-input multi-output measurement group. Q z This includes one transmission from each of the enabled transmission channels; if a double-repetition-cycle template is used, it also includes complete transmission sequences under two repetition cycles. For binary phase modulation systems, each transmission channel is distinguished by binary phase coding and can transmit simultaneously or overlappingly; the complete measurement set is the minimum complete coded state set required to recover all virtual channels. For Doppler diversity multiple access systems, each transmission channel is distinguished by different Doppler biases; the complete measurement set is the minimum complete transmission state set or coded state set required to recover all virtual channels. The above systems can be used individually or in combination when product permits; regardless of the system used, the allocation is based on the complete measurement set under the corresponding system as the quantum, and must not be truncated. Let the duration of a complete measurement set in interval z be denoted as . T Q,z .

[0051] The frame duration budget constraint is: Σz∈Z on n z T Q,z ≤ T budget ,and n floor,z ≤ n z ≤n max,z The allocation process is as follows: shilling n z = n floor,z , n floor,z This is the preset lower limit quantity. If the preset lower limit total duration Σ n floor,z T Q,z It has exceeded T budget If the preset lower limit is not lowered, the current adaptive composite frame will be canceled and the frame will be rolled back. P safe When the preset lower limit is feasible, the marginal utility models that satisfy the gain requirements are added one by one: J z (n) = R z [1-exp(- k z n)] G z (n) =[ J z (n+1) - J z (n) ] / T Q,z in, G z (n) This is the ratio of demand satisfaction gain to occupancy time resulting from adding a complete measurement group. k z >0 is calibrated offline, near field k 1=0.35, Midfield k 2=0.25, far field k 3 = 0.20. Each time, a range is selected from those that still satisfy the constraints of frame duration, storage, processing, transmit duty cycle, and temperature rise. G z (n z ) The largest interval makes n z Increase by 1; repeat until no interval can be included in the complete measurement set. The exponential model can be replaced by a monotonically decreasing calibration gain table.

[0052] In a numerical example, T budget =48ms Z on ={1,2,3}T Q,1 =2.8ms T Q,2 =2.2ms T Q,3 =1.8ms, n floor ={1,1,1} n max ={8,6,4}, R={0.52,0.28,0.20}. A preset lower limit of 6.8ms for the total duration is feasible; all intervals are taken as... n max The total duration is 8×2.8+6×2.2+4×1.8=42.8ms, which is still less than [the specified duration]. T budget Distribute the marginal utility ratios group by group according to the above ratios until all intervals have been reached. n max Therefore, n={8,6,4}, with a total duration of 42.8ms. The remaining budget cannot be included in a complete measurement group due to the upper limit constraint of the measurement group. If the continuous ratio 0.52×48 / 2.8≈8.9 is mistakenly rounded directly to 9 groups, it will exceed the near field. n max,1 =8 and may truncate the complete emission cycle; this embodiment uses... Q z For quantum and subject n max,z Constraints are necessary to fundamentally avoid this problem.

[0053] Example 5 This embodiment details the waveform configuration identifier version submission of the waveform configuration identifier version submission module 80. See [link to documentation]. Figure 5 .

[0054] Controller 120 decomposes the selected composite time-division waveform configuration into a set of waveform generation parameters. F tx Analog-to-digital conversion acquisition parameter set F adc With signal processing parameter set F dsp All three modules share the same version number and checksum. Each module maintains its current configuration and a backup parameter area: new configurations are first written to the backup parameter area, and the waveform generator 20, analog-to-digital conversion sampling and preprocessing module 30, and 4D millimeter-wave radar point cloud generation module 100 respectively return a confirmation of completion and a copy of their local version number and checksum. Only when the version numbers and checksums of all three modules are completely identical will the radar controller 120 switch the backup parameter area to the current configuration at the next subframe or measurement frame boundary. If any module fails to write or has inconsistent versions, the current configuration remains unchanged.

[0055] The acquired data is recorded and transmitted synchronously with the analog-to-digital conversion data, along with the waveform configuration identifier index. The complete parameter table is stored in the controller 120 or the data descriptor. If the processing end detects inconsistencies in the waveform configuration identifier, linear frequency modulation pulse index, number of sampling points, transmission order, or checksum, it marks the subframe as invalid and does not use speculative parameter calculations to avoid distance mapping errors and virtual channel mismatches.

[0056] If the hardware does not support a spare parameter area within a single frame, all modules can be written and verified within the stop-transmission protection zone. Transmission can only resume after successful verification. The waveform configuration identifier can also be transmitted via the acquisition descriptor, control bus metadata, or data packet header, without requiring the waveform configuration identifier to be modulated into the RF signal. This mechanism ensures that the same version of the parameters for generation, acquisition, and processing are effective, eliminating systematic deviations caused by inconsistent slopes or sampling start points.

[0057] Example 6 This embodiment details the unified reference time compensation of the 4D millimeter-wave radar point cloud generation module 100. See [link to documentation]. Figure 6 .

[0058] Center time of each subframe within the composite time-division frame t z The difference lies in how the processing unit estimates range, velocity, azimuth, and elevation using the slope, sampling time, repetition period, and transmission sequence corresponding to each waveform configuration identifier. Data from different waveform configuration identifiers are not mixed before the range fast Fourier transform or virtual array is formed. A unified reference time is used. t ref Typically, the measurement frame start time or frame geometric center time is taken.

[0059] For points with associated tracks, the compensation process is as follows: first, based on the subframe time... t z Radar coordinates transformed to global coordinates p g (t z ) = R gs (t z )p s (t z ) + t gs (t z ) Then according to the trajectory speed v g Extrapolation to reference time p g (tref ) = p g (t z ) + v g (t ref -t z ) Finally, the radar coordinates are transformed back to the reference time. p s (t ref ) = R sg (t ref )p g (t ref ) + t sg (t ref ) When no reliable track velocity is available, velocity extrapolation is omitted, and only vehicle pose compensation is performed. For unassociated points, radial velocity is used for first-order approximate extrapolation along the current line of sight, and the fusion weights are reduced. w fuse ,For example w fuse =0.5. After completing time alignment, duplicate points are merged based on spatial distance, velocity difference, angle difference, and waveform configuration flag validity; the merging threshold example is a distance of 0.3m, a velocity of 0.5m / s, and an azimuth of 2°.

[0060] In a pair of interval example subframes, near-field subframes t 1 = Frame start + 6ms, far-field subframe t 3 = frame start + 28ms t ref =Frame start. For a correlated point approaching radially at 15m / s, without compensation, the equivalent distance deviation is approximately 0.33m; after compensation using the above three formulas, the near-field and far-field point clouds are on the same... t ref With the integration of data, the correlation threshold statistics of the trajectory management module 110 remain stable, reducing false correlations caused by waveform switching near the threshold.

[0061] Example 7 This embodiment details the constraints and safe rollback of the smooth switching module 90.

[0062] Interval start / stop requires K=3 consecutive frames to meet the threshold condition; after template switching, it must remain stationary for at least L=5 frames, where L≥K. Continuous parameter x∈{ S z , F s,z , T s,z , T r,z The condition |x(t) - x(t-1)| ≤ Δx max Example ΔS max =10MHz / μs ΔF s,max =2MHz ΔT s,max =2μs; the number of measurement groups satisfies | n z (t) - n z (t-1) |≤ Δn max =2. Safety warnings, collision risks, or hardware malfunctions allow direct bypassing of the normal limit and switching to... P safe .

[0063] Radar power-on, waveform configuration identifier version verification failed, continuous M loss =When there is no effective feedback for 4 frames or the hardware boundary check module 70 fails continuously, the smooth switching module 90 is triggered. P safe . P safe The waveform is a factory-calibrated fixed waveform, covering basic detection from 0 to 180m, meeting all hardware boundaries, and is suitable for near-field applications. n floor,1 =2, Far Field n floor,3 =1, adaptive composite frame is not enabled. After feedback is restored and two consecutive frames pass verification, adaptive mode is gradually restored according to K and L constraints.

[0064] This embodiment, in conjunction with embodiments one through six, enables the closed loop to smoothly adapt to changes in the scene and quickly revert to its previous state in case of anomalies, thus balancing detection performance and system stability.

[0065] Example 8 This embodiment provides a hardware deployment example of a composite time-division waveform generation system for 4D millimeter-wave radar point cloud feedback. See [link to relevant documentation]. Figure 2 .

[0066] The system includes an antenna array and transceiver front-end 10, a waveform generator 20, an analog-to-digital conversion sampling and preprocessing module 30, a partition requirement assessment module 40, a candidate waveform screening module 50, a complete MIMO measurement group allocation module 60, a hardware boundary verification module 70, a waveform configuration identification version submission module 80, a smooth switching module 90, a 4D millimeter-wave radar point cloud generation module 100, a track management module 110, and a radar controller 120. The radar controller 120 can be a microcontroller core within a dedicated radar chip, a soft core embedded in a field-programmable gate array (FPGA), or a radar process within a domain controller. The partition requirement assessment module 40, candidate waveform screening module 50, complete MIMO measurement group allocation module 60, hardware boundary verification module 70, waveform configuration identification version submission module 80, and smooth switching module 90 can be deployed in software form within the radar controller 120 or partially hard-coded into the FPGA.

[0067] The data flow is as follows: the antenna array and transceiver front-end 10 transmit frequency-modulated continuous waves under the drive of waveform generator 20; the analog-to-digital conversion sampling and preprocessing module 30 collects intermediate frequency data and records it along with waveform configuration identifiers; the 4D millimeter-wave radar point cloud generation module 100 calculates and compensates the waveform configuration identifiers according to the waveform configuration identifiers. t ref The track management module 110 outputs track feedback to the partition demand assessment module 40; the partition demand assessment module 40, the candidate waveform screening module 50, and the complete multi-input multi-output measurement group allocation module 60 generate a composite time-division waveform configuration; after the hardware boundary verification module 70 verifies the configuration, the waveform configuration identifier version submission module 80 submits consistent parameters to the waveform generator 20, the analog-to-digital conversion sampling and preprocessing module 30, and the 4D millimeter-wave radar point cloud generation module 100; the smooth switching module 90 monitors the switching process and rolls back in case of an anomaly. P safe .

[0068] In a 77GHz quad-transmit quad-receive time-division multiplexed multiple-input multiple-output example, the number of virtual channels is 16, and near-field is enabled. B eff,1 ≥2GHz, far-field enabled B eff,3 ≥0.4GHz, frame period 50ms T budget =46ms, leaving 4ms for processing and communication. The system completes feedback reading, demand assessment, candidate screening, measurement group allocation, verification, and waveform configuration identifier submission in 10ms cycles to meet vehicle real-time requirements. Chips that do not support multiple waveform configuration files in a single frame can allocate different intervals to adjacent measurement frames, using the complete frame as the resource quantum, while the waveform configuration identifier binding and unified reference time compensation remain unchanged.

[0069] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating composite time-division waveforms from 4D millimeter-wave radar point cloud feedback, characterized in that, include: Read the 4D millimeter-wave radar point cloud and its associated track status from multiple recent measurement frames, divide the detection space into at least two distance intervals, and extract the demand-related features of each distance interval based on the point cloud, the track status, and the correlation results between the point cloud and the track to form the interval demand value. Based on the interval demand value, a complete multiple input multiple output measurement group is used as the smallest allocation unit, and no less than a preset lower limit number of complete multiple input multiple output measurement groups are allocated to each distance interval; for the remaining frame duration budget, the remaining time is allocated group by group according to the ratio of the demand satisfaction gain brought by each additional complete multiple input multiple output measurement group to the duration occupied by that measurement group, thereby generating a composite time-division waveform configuration; Using a waveform configuration identifier, the composite time-division waveform configuration is simultaneously associated with waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters, and the configuration is switched to the currently used configuration at frame boundaries. The acquired data of each subframe is calculated according to the waveform configuration identifier, and the point cloud of each subframe is compensated to the same reference time.

2. The method according to claim 1, characterized in that, The extraction of demand-related features for each distance interval and the formation of interval demand values ​​specifically includes: Based on the point cloud, spatial occupancy-related features are extracted; Based on the trajectory state, extract trajectory stability and uncertainty related features; Extract association quality-related features based on the association results of point cloud and track; Extracting velocity dispersion-related features; After normalizing and weighting the above features, the interval demand value is obtained; and the historical track existence quantity obtained from the track status is introduced as a safety net term to obtain the normalized demand component used for resource allocation.

3. The method according to claim 1, characterized in that, The complete multiple-input multiple-output measurement group is the minimum and complete set of transmit states or encoded states necessary to recover the predetermined virtual array or decoding matrix, and its specific composition is determined by the multiple-input multiple-output separation system adopted.

4. The method according to claim 1, characterized in that, Before generating the composite time-division waveform configuration, for distance intervals that meet the activation conditions, candidate parameters that simultaneously satisfy distance resolution, beat frequency, speed ambiguity, and frame-level hardware executable constraints are selected from the set of waveform parameters supported by the hardware. Before the composite time-division waveform configuration takes effect, hardware executableness verification is performed on the entire frame configuration. If the verification fails, the following steps are taken: replace the candidate parameters for this interval, reduce the measurement groups for this interval, close the lowest priority unsafe interval, and roll back the safe waveform.

5. The method according to claim 1, characterized in that, The configuration of the composite time-division waveform, which is simultaneously associated with waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters, specifically includes: The waveform generation parameters, analog-to-digital conversion acquisition parameters, and signal processing parameters are written into the spare parameter area of ​​each module, and all three use the same version number and checksum. After all relevant modules have confirmed that the writing is completed, the spare parameter area is switched to the currently used configuration at the frame boundary. If any module fails to write or the version is inconsistent, the currently used configuration remains unchanged. If the processing end finds that the parameter version or checksum is inconsistent, the corresponding subframe is marked as invalid.

6. The method according to claim 1, characterized in that, The process of compensating the point clouds of each subframe to the same reference time specifically includes: For points already associated with the track, the points are compensated to the reference time by combining the vehicle's position and attitude with the track motion; when reliable track motion information is unavailable, compensation is performed only based on the vehicle's position and attitude. For points not associated with a track, they are approximately extrapolated along the line of sight using radial velocity, and their fusion weights are reduced. After completing the time alignment, merge the duplicate points.

7. The method according to claim 1 or 2, characterized in that, Smooth constraints are also applied to interval start / stop and parameter switching, specifically including: Interval start / stop requires multiple consecutive frames to meet the corresponding threshold conditions before switching; After the waveform configuration is switched, it shall remain in place for at least a preset number of frames without switching again, and the preset number of frames shall not be less than the number of consecutive frames required for start / stop determination. Between adjacent measurement frames, the changes in waveform parameters and the number of measurement groups in each interval do not exceed their respective preset upper limits.

8. The method according to claim 1, characterized in that, If any of the following situations occur: the preset lower limit allocation is not feasible, feedback is lost, or hardware malfunction occurs, the system will fall back to the pre-verified safe waveform. In cases requiring emergency switching, the system is allowed to switch directly to the safe waveform without being constrained by the upper limit of parameter changes in adjacent measurement frames.

9. A composite time-division waveform generation system for 4D millimeter-wave radar point cloud feedback, characterized in that, include: Antenna array and transceiver front end; Waveform generator; Analog-to-digital conversion sampling and preprocessing module; The partitioned demand assessment module is used to read the 4D millimeter-wave radar point cloud and its associated track status from multiple recent measurement frames, divide the detection space into at least two distance intervals, and extract demand-related features for each distance interval based on the point cloud, the track status, and the correlation results between the point cloud and the track to form interval demand values. The complete multiple input multiple output measurement group allocation module is used to allocate no less than a preset lower limit number of complete multiple input multiple output measurement groups to each distance interval based on the interval demand value, using complete multiple input multiple output measurement groups as the minimum allocation unit, and allocate the remaining frame duration budget group by group according to the ratio of the demand satisfaction gain brought by each additional complete multiple input multiple output measurement group to the duration occupied by that measurement group, thereby generating a composite time-division waveform configuration. The waveform configuration identifier version submission module is used to use the waveform configuration identifier to simultaneously associate the composite time-division waveform configuration with waveform generation parameters, analog-to-digital conversion acquisition parameters and signal processing parameters, and switch to the currently used configuration at the frame boundary; The 4D millimeter-wave radar point cloud generation module is used to calculate the acquired data of each subframe according to the waveform configuration identifier, and compensate the point cloud of each subframe to the same reference time. The track management module is used to output track status and feed it back to the partition demand assessment module; The radar controller coordinates the above modules to complete the closed loop of composite time-division waveform generation.

10. The system according to claim 9, characterized in that, It also includes a candidate waveform screening module, a hardware boundary verification module, and a smooth switching module; the candidate waveform screening module and the hardware boundary verification module are used to perform the candidate parameter screening and whole-frame hardware executability verification as described in claim 4; the smooth switching module is used to perform the smooth constraint as described in claim 7, and trigger a fallback to a safe waveform in the case described in claim 8.

Images