Object detection method and angle-resolved FMCW radar sensor system

Abstract

1. A method for detecting an object using an angle-resolved FMCW radar sensor, comprising: generating an at least three-dimensional spectrum (56) having a distance dimension (d), a Doppler dimension representing the relative velocity (v) of an object, and angle dimensions (α, θ) based on received signals from multiple receive channels of the radar sensor; the radar sensor being undersampled in the Doppler dimension, resulting in an incomplete and ambiguous spectrum (56) in this dimension; and using a frequency modulation scheme with multiple interleaved sequences of frequency ramps equidistant in time to resolve the ambiguity; and for each of multiple velocity hypotheses, a velocity-dependent phase shift between received signals obtained for the different sequences is modeled based on known time shifts between the frequency ramps belonging to the different sequences and compared with the measured phase shift; wherein, during the phase comparison, the angle-dependent phase shift of the different receive channels is analyzed together with the velocity-dependent phase shift to reconstruct a complete and unambiguous spectrum (60); and object detection (60) is performed based on the reconstructed spectrum (60).

Description

[Technical Field]

[0001] The present invention relates to a method for object detection using an angle-resolved FMCW radar sensor, in which an at least three-dimensional spectrum having a distance dimension, a Doppler dimension indicating the relative velocity of the object, and an angle dimension is generated based on received signals from multiple receive channels of the radar sensor, the radar sensor being undersampled in the Doppler dimension, as a result of which the spectrum is incomplete and ambiguous in this dimension, and in which a frequency modulation scheme with multiple mutually interleaved sequences of frequency ramps equidistant in time is used to resolve this ambiguity, and for each of multiple velocity hypotheses a velocity-dependent phase shift between the received signals obtained for the different sequences is modeled based on known time shifts between the frequency ramps belonging to the different sequences and compared with the measured phase shift.

[0002] More particularly, the present invention relates to a method and radar sensor system for a motor vehicle. [Background technology]

[0003] Radar sensors are often used for surrounding environment monitoring in driver assistance systems because they allow direct measurement of an object's distance (d), relative velocity (v), and azimuth and elevation angles (α, θ).

[0004] Radar sensors are known that operate using a sequence of identical, relatively short frequency ramps, so-called "rapid chirps." These ramps have a high frequency shift compared to their duration and are therefore steep. Therefore, in the baseband signal, the distance-dependent component of the frequency shift predominates, while the Doppler shift is sampled by the sequence of ramps. Therefore, a sufficiently high repetition rate of the short ramps is required to enable an unambiguous determination of the relative velocity within the desired measurement range of the relative velocity. In particular, the time offset between successive short ramps must be less than half the period of the Doppler frequency. Such short time offsets place high demands on analog hardware. To achieve a good signal-to-noise ratio, a correspondingly large number of such short ramps would be required. This results in relatively high memory requirements and computational costs when analyzing the data.

[0005] To further improve the functionality of driver assistance systems and increase the installation rate in vehicles, sensors with increasingly better performance and longer range are required, but at the same time, the sensors should be inexpensive.In order to enable accurate speed estimation and distance estimation of radar objects with as low hardware and calculation costs as possible, it has been proposed to use multiple sequences of frequency-modulated ramps with time intervals between them, in which Doppler shift is undersampled through the ramp sequence, and the information obtained about relative speed is therefore ambiguous.

[0006] From DE 10 2014 212 280 A1, a method is known using ramp sequences interleaved with one another in time, which allows unambiguous velocity estimation by analyzing the relative phase between identical peak positions in the two-dimensional spectrum based on a two-dimensional fast Fourier transform (2D-FFT) of each ramp sequence and comparing it with a model equation of the ambiguity hypothesis. The resolution of the ambiguity is hereinafter referred to as "Velocity Ambiguity Resolution (VAR)". In DE 10 2014 212 284 A1 and DE 10 2017 2003 17 A1, this method was extended to multiple transmit antennas operating in time division multiplexing or in code division multiplexing with periodic codes via chirps.

[0007] However, due to underpulling of the Doppler dimension, targets with different relative velocities may overlap. If this overlap does not prevent target signal detection, these targets can be separated using the method described in DE 10 2014 223 990 A1, although this requires high computational effort. Furthermore, detection may be hindered if a target is masked by a second target located in an adjacent range-Doppler cell.

[0008] DE102021213495 describes a method in which, on the one hand, the range or sensitivity of a radar sensor can be increased, thereby enabling earlier recognition of weaker targets, and, on the other hand, the negative effects of target spectral overlap can be avoided. In this case, modulation sequences are coherently added by a "velocity beamformer" already before detection. In this case, reception performance can be increased for "desired" velocities and decreased for "undesired" velocities.

[0009] The above mentioned approaches do not analyze velocity and angle information together, and therefore the resolution of multiple targets is determined by their individual resolution in velocity and angle, respectively.

[0010] DE102020202498A1, DE102020202499A1, and DE102020202500A1 describe methods for MIMO radar systems (Multiple Input Multiple Output), in which a high-resolution but ambiguous spectrum in both the Doppler and angular dimensions is analyzed in a first detection stage. Spectra obtained for different ramp sequences and different combinations of transmit and receive antennas are non-coherently added, and object detection is performed based on the resulting sum spectrum. Ambiguity hypotheses are then tested for the detected objects to resolve the ambiguity. After correcting for velocity-dependent and angle-dependent phase shifts, the phase-corrected complex amplitudes of the detections are coherently added, resulting in a lower-resolution but unambiguous spectrum in a second detection stage. The relative velocity and azimuth and elevation angles of the detected objects can then be determined unambiguously and with high resolution by fusing the results obtained in both detection stages. Summary of the Invention [Problem to be solved by the invention]

[0011] The objective of the present invention is to further increase the sensitivity of detection and improve the robustness to multi-target scenarios. [Means for solving the problem]

[0012] This problem is solved by the present invention in that, during phase adjustment, the angle-dependent phase shifts of the different receiving channels are analyzed together with the velocity-dependent phase shifts to reconstruct a complete and unambiguous spectrum, and object detection is performed on the basis of the reconstructed spectrum.

[0013] The method according to the present invention makes extensive use of information obtained in multiple multidimensional spectra to reconstruct the "ideal" spectrum that would have been obtained in a given situation if the radar sensor had not undersampled in either the Doppler or angular dimensions. Since actual object detection is only performed based on this reconstructed spectrum, which is both high-resolution and unambiguous, improved object separation is achieved. In addition, the joint processing of velocity and angular information significantly reduces the probability of peak overlap and object masking. Furthermore, increased sensitivity is achieved by analyzing information from multiple individual spectra.

[0014] Advantageous embodiments of the method are set out in the dependent claims. Various methods can be used to reconstruct an unambiguous spectrum. One of these methods consists in integrating the already mentioned Velocity Ambiguity Resolution (VAR), i.e., velocity ambiguity resolution, with known methods for angle estimation (ANG), using a common single-target or multi-target model for velocity ambiguity and angle information.

[0015] In the case of a single-target model, such joint processing can be achieved by coherent velocity and angle beamforming. Angular beamforming is a known method in which the phases of signals received in different (real or virtual) receiving channels are corrected, specifically to compensate for phase shifts that depend on the positioning angle and are caused by misalignments between different transmit and receive antennas, thereby making the sensor primarily sensitive to “beams” coming from a specific direction. Similarly, velocity beamforming corrects the relative velocity-dependent phase shifts from lamp to lamp within a lamp sequence, thereby increasing sensitivity to objects with a specific relative velocity. In coherent velocity and angle beamforming, both phase corrections are combined with each other. In this way, sensitivity to specific critical points in the multidimensional spectrum can be targeted, thereby better resolving overlapping peaks that are close together. Conversely, signals from distracting objects, such as roadside delineators, can also be targeted and suppressed.

[0016] Processing together results in stronger sidelobe suppression than when processing separately or sequentially, thereby increasing the dynamic range for targets with different velocities and angles and allowing weaker targets to be recognized. In addition to simultaneous 2D (velocity and angle) analysis, joint processing also contemplates an iterative approach in which estimates for both dimensions (velocity and angle) are alternately optimized.

[0017] F. Marvasti et al., "A Unified Approach to Sparse Signal Processing," EURASIP Journal on Adv. in Sig. Proc., Volume 2012, No. 1, February 2012, describes a sparse reconstruction algorithm that can be used as an alternative method to reconstruct an unambiguous spectrum.

[0018] Another alternative is the use of deep learning or artificial intelligence. In this case, a neural network is trained to reconstruct the corresponding unambiguous spectrum from the ambiguous spectrum obtained by undersampling. The necessary training data for the neural network can be generated, for example, by capturing the same surrounding environment with a radar sensor working with undersampling on the one hand and with a radar sensor working without undersampling but with otherwise equivalent specifications on the other hand. Alternatively, the training data can be generated by simulation using suitable simulation methods (ray tracing, channel simulation, ...).

[0019] In principle, the method described here can resolve ambiguities in the Doppler dimension as well as ambiguities in the azimuth and / or elevation angles, although the method is also advantageous for radar sensors with a full antenna array and thus providing an unambiguous spectrum in the angle dimension.

[0020] The subject of the invention is also a radar sensor system in which the method according to the invention is implemented. Automotive radar sensors typically have an internal processor unit that is responsible for analyzing the digital received signal and calculating the range, relative velocity, and positioning angle of the located object. However, in another embodiment, part of this process can be outsourced to an external processor unit, which can provide more processing capacity. For example, it is conceivable that the internal processor unit of the radar sensor only performs a Fourier transform in the range dimension, while further Fourier transforms as well as unambiguous spectral reconstruction and object detection are performed by an external processor unit.

[0021] If sufficient computational resources are available, conventional analysis, such as that described in the above cited references, may be performed in parallel with the improved analysis according to the present invention and may be used as a second, algorithmically independent path in ambient environment capture. In this case, the conventional analysis and the improved analysis may optionally be implemented in different control devices, i.e., the conventional analysis may be implemented in, for example, a radar sensor, while the improved analysis is implemented in an external processor unit.

[0022] A further possibility consists in carrying out the detailed processing according to the invention only for the range of the two-dimensional range and velocity spectrum that is of particular interest in a given situation.

[0023] In the following, exemplary embodiments are explained in more detail on the basis of the drawings. [Brief explanation of the drawings]

[0024] [Figure 1] FIG. 1 is a block diagram of an FMCW radar sensor. [Figure 2] FIG. 10 shows the modulation pattern of two sequences of identical ramps transmitted with a time offset of T12. [Figure 3] FIG. 1 illustrates an antenna array of an angle-resolved radar sensor. [Figure 4] FIG. 2 is a block diagram of an internal processor unit and an external processor unit of a radar sensor that together comprise a radar sensor system according to the present invention. [Figure 5] 1 is a schematic diagram of an input data set for a processing procedure in the method according to the invention; [Figure 6] FIG. 2 is a flow diagram of the method. DETAILED DESCRIPTION OF THE INVENTION

[0025] FIG. 1 shows a simplified block diagram of an FMCW radar sensor 10 mounted in a vehicle, e.g., at the front, for measuring the distance d, relative velocity v, and positioning angle of objects 12, 14, e.g., vehicles traveling ahead. The radar sensor 10 includes a voltage-controlled oscillator 16 that provides a frequency-modulated transmit signal via a mixer 18 to a transceiver 20, which transmits the signal in the direction of the objects 12, 14. Signals reflected by the objects are received by the transceiver 20 and mixed with a portion of the transmit signal in the mixer 18. Baseband signals b1, b2 are thus obtained, digitized, and further analyzed in a processor unit 22. The processor unit 22 includes a controller 24 that controls the function of the oscillator 16. The frequency of the transmit signal provided by the oscillator is modulated by a rising or falling ramp sequence during a radar measurement.

[0026] 2 shows the transmission frequency f of the transmitted signal 28 as a function of time t. During the measurement, one transmitting antenna transmits two sequences of lamps with identical lamp parameters, which are interleaved in time with each other. A first sequence 30 of lamps 32 is shown in FIG. 2 with a solid line, whereas a second sequence 34 of lamps 36 is shown with a dashed line. The number i of the sequence to which the lamp belongs and the lamp index j of each lamp within a sequence are provided.

[0027] Each ramp 36 of the second sequence 34 is offset by a time offset T12 with respect to the ramp 32 of the first sequence 30 having the same ramp index j. Within each sequence 30, 34, successive ramps 32 or 36 are offset from each other by a time distance Tr2r, i.e., the time distance Tr2r is the same in both sequences. Furthermore, there is an interval P between each two successive ramps of one sequence.

[0028] In the example shown in Figure 2, the difference in lamp center frequency between successive lamps 32 or 36 within a sequence 30, 34 is 0. Thus, all lamps have the same frequency progression.

[0029] Within the baseband signal b1, b2, component b1, which originates from ramp 32 of the first sequence 30, alternates with component b2, which originates from ramp 36 of sequence 34. Due to the time offset T12 between ramps 32 and 36, components b1 and b2 have a relative phase shift that, for each positioned object, depends on the relative velocity of that object. This phase shift can be calculated and allows for the resolution of ambiguities that arise due to undersampling in the Doppler dimension.

[0030] 1 only two interleaved ramp sequences 30 and 34 are provided, however in practice the number of frequency ramps interleaved with one another can be significantly higher, which may allow for more accurate resolution of ambiguities or allow for a greater degree of undersampling.

[0031] 3 shows a schematic representation of the antenna arrays of the transceiver arrangement 20. A plurality of transmit antennas 38 form a transmit array 40 (TX), and a plurality of receive antennas 42 form a receive array 44 (RX). In the example shown, both arrays are two-dimensional, thus allowing MIMO angle measurements in both azimuth and elevation.

[0032] Within the receive array 44, the receive antennas 42 are arranged at uniform distances in one angular resolution direction y, e.g., azimuth. In this regard, the distance between the individual receive antennas is large enough that a large aperture and correspondingly high angular resolution can already be achieved with fewer antennas. However, the Nyquist unambiguity criterion is not met because the antenna-to-antenna distance is greater than half the wavelength of the radar radiation.

[0033] In the example shown, the receive antennas 42 are also spaced at uniform distances in elevation (angular resolution direction z), and even in this direction the antenna distance is large enough to result in non-unique undersampling.

[0034] The transmit antennas 38 of the transmit array 44 are spaced at non-uniform distances in azimuth, but these distances are selected to provide unambiguous angle measurements, although the aperture is significantly smaller than that of the receive array 44, and therefore the angular resolution is lower. In elevation, the transmit array 40 is also designed for unambiguous angle measurements with a small aperture.

[0035] 3 also shows a composite array 46, which is obtained when each of the receive antennas 42 is combined with each of the transmit antennas 38, thereby adding up the time-of-flight differences of the signals from the transmit antennas to the object and from the object to the receive antenna. It is the aperture of this virtual array 46 that determines the resolution of the radar sensor. However, for the receive array to be disambiguated, the signal components originating from the different transmit antennas 38 must be separated from one another in the received signal. This signal separation can be achieved in known manner by time-division multiplexing, frequency-division multiplexing, or code-division multiplexing.

[0036] Furthermore, in the example shown in Figure 3, for each y position of a receive antenna, there is also a receive antenna at every z position, so that both angularly resolved directions y and z in the receive array 38 are decoupled from each other. In contrast, the transmit array 40 is an example of a non-decoupled array, where for some y positions (the right two positions in Figure 3), not all z positions are occupied. In general, a decoupled array facilitates data analysis, while a non-decoupled array requires fewer antenna elements. The decision to use a decoupled or non-decoupled array can be made differently for both the transmit and receive sides depending on their respective requirements.

[0037] An equidistant arrangement of the antenna elements (in azimuth and / or elevation) also facilitates the analysis of the data, as this arrangement allows for example the use of a Fast Fourier Transform (FFT), whereas a non-equidistant arrangement of the antennas, as here is the case for the transmit antenna 38, has the advantage that, for a given aperture, the unambiguity angular range can be optimized.

[0038] In general, for the radar systems described herein, all combinations of equidistant and non-equidistant, as well as decoupled and non-decoupled, geometries are contemplated. Similarly, embodiments are possible in which the transmit array is designed for ambiguous, high-resolution angle measurements, while the receive array is designed for unambiguous angle measurements with lower angular resolution.

[0039] FIG. 4 shows as a block diagram the processor unit 22 of the radar sensor shown in FIG. 1 and an external processor unit 48, which is used for further processing of signals preprocessed in the processor unit 22 of the radar sensor.

[0040] In the internal processor unit 22, a first transformation stage 50 performs a one-dimensional fast Fourier transform (FFT) on the baseband signals b1, b2, ... obtained from a single frequency ramp 32 or 36 to form a one-dimensional spectrum 52. Each object located by the radar sensor appears in this spectrum as a signal peak at a frequency indicating the object's distance d. Digital data representing the resulting spectrum 52 is sent to the external processor unit 48 for further processing. In the external processor unit, the spectra 52 resulting from the same ramps of the same ramp sequence 30 or 34 are each subjected to a further Fourier transform. The frequency axis of each one-dimensional spectrum 52 is divided into many range cells, and the Fourier transform in the transformation stage 54 is performed separately for each range cell. The signal obtained from successive ramps of a single sequence has a periodically varying frequency, which depends on the relative velocity v of the respective located object. In this way, the transformation stage 54 provides a two-dimensional spectrum 56 for each ramp sequence, with a distance dimension d and a velocity or Doppler dimension v.

[0041] In the case of a MIMO radar with an antenna array as shown in FIG. 3, each of the transmit antennas 38 can in principle be associated with a respective receive antenna 42. Each receive antenna 32 provides one input signal to one receive channel of the radar sensor. The analysis of this signal depends on which of the transmit antennas 38 was active on that frequency ramp. For multiple receive antennas 42 with different distances in the (horizontal) angular resolution direction y relative to the active transmit antenna 38, their signals have a phase shift that depends on the distance between these receive antennas 42 in the direction y and the azimuth angle α of the located object. Correspondingly, the signals of receive antennas 42 with different distances in the direction z relative to the active transmit antenna 38 have a phase shift that depends on the distance between these receive antennas in this direction and the elevation angle θ of the located object.

[0042] If all combinations of transmit and receive antennas are utilized in time division multiplexing, frequency division multiplexing, or code division multiplexing, a different two-dimensional spectrum 56 is obtained for each of these combinations and for each ramp sequence 30 or 34. seq is the number of lamp sequences 30, 34, and N TX is the number of transmit antennas, and N RX is the number of receiving antennas, the total number of analyzable spectra 56 is N seq ×N TX ×N RX Each of these spectra 56 by itself is ambiguous due to undersampling in the velocity dimension v. However, N seq This ambiguity can be resolved by using information from the spectra 56 obtained for different frequency ramps. In principle, it would be sufficient to analyze one of the spectra obtained for the same ramp sequence but for different combinations of transmit and receive antennas, since these sequences are redundant. Independently, angular estimation in azimuth and elevation can be performed based on the spectra obtained for different combinations of transmit and receive antennas. For this purpose, for example, the spectrum 56 can be organized into two sequences: one in which the spectra are arranged according to increasing distances between the transmit and receive antennas in the direction y, and the other in which the spectra are arranged according to increasing distances in the direction z. In this case, both sequences of two-dimensional spectra 56 together form a single four-dimensional spectrum.

[0043] Based on the phase difference in these spectra for each sequence, angle estimation in azimuth or elevation can then be performed using known methods. In principle, for these angle estimations, spectra 56 that correspond to the same combination of transmit and receive antennas but belong to different ramp sequences are redundant. For example, when the antenna distances form a uniform mesh, as in the case of receive antenna 42 in FIG. 3, angle estimation can be efficiently performed using a fast Fourier transform, resulting in a four-dimensional spectrum with dimensions d, velocity v, azimuth angle α, and elevation angle θ. In this regard, ambiguities may arise in the case of a partially packed antenna array, and these ambiguities must be resolved using known methods.

[0044] However, combining velocity ambiguity resolution (VAR) with angle estimation (ANG) can prove useful. That is, the above-mentioned redundancy between different groups of spectra is incomplete, especially when multiple radar objects are simultaneously present. For example, when two objects with at least approximately the same range and the same relative velocity are simultaneously located, overlapping peaks occur in the two-dimensional spectrum 56, which makes object separation and disambiguation difficult. However, when these objects differ in azimuth and / or elevation angles, no overlapping peaks occur in the three-dimensional or four-dimensional spectrum containing angular information. Therefore, a single-target model is often sufficient for velocity and / or angle ambiguity resolution, whereas more complex multi-target models are rarely required.

[0045] In any case, by using a common single-target or multi-target model to model both velocity-dependent and angle-dependent phase shifts, it is possible to more fully utilize the information contained in the multiple spectra 56 and to reconstruct an "ideal," unambiguous four-dimensional spectrum, whereby phasing between spectra obtained for different frequency ramps and different antenna combinations at least partially recovers the information lost due to undersampling.

[0046] Therefore, in the external processor unit 48, a reconstruction stage 58 is provided to jointly analyze these spectra 56 using a common single or multi-target model for velocity-dependent and angle-dependent phase shifts to combine and resolve velocity ambiguity and possibly angle ambiguity, thereby reconstructing an unambiguous four-dimensional spectrum 60, with each signal peak uniquely indicating the range d, relative velocity v, azimuth angle α, and elevation angle θ of the object in question.

[0047] The spectral values ​​in the unambiguous spectrum 60 are then phase-corrected for various combinations of relative velocity v, azimuth angle α, and elevation angle θ using a correction function a(v,α,θ), and these spectral values ​​are coherently added along the dimensions v, α, and θ. This coherent addition suppresses sub-maxima in the spectrum and increases the signal-to-noise ratio, thereby enabling object detection with high sensitivity and resolution in the detection stage 62. That is, the processor unit 48 provides unambiguous values ​​for the distance d, relative velocity v, azimuth angle α, and elevation angle θ for each object detected with high sensitivity.

[0048] For the processing process in the reconstruction stage 58, various data structures can be used, as explained with reference to Figure 5. This figure shows a single two-dimensional (ambiguous) spectrum 56, which is partitioned into range cells 66 in the range dimension d and into Doppler cells 68 in the Doppler or velocity dimension v.

[0049] In one embodiment, the input data for the reconstruction stage 58 is a data structure 70 called a "range gate." This data structure contains the spectral values ​​belonging to a single range cell 66 for the entire spectrum 56. Thus, this data structure contains the spectral values ​​for all Doppler cells in the spectrum 56 for all ramp sequences and all combinations of transmit and receive antennas.

[0050] In another embodiment, the data structure 72 of the input data is a group of adjacent range gates so that changes in spectral values ​​in the distance dimension can also be considered when analyzing the data.

[0051] In a further embodiment, the input data is a data structure 74 consisting of only one range and Doppler cell in all spectra 56 for different lamp sequences and antenna combinations. Alternatively, the input data may be a "patch" 76 containing multiple adjacent range and Doppler cells.

[0052] In either case, the input data is converted into a reconstructed range-Doppler-azimuth and elevation spectrum 60 for that range gate or range and Doppler cell of that data structure 72, 74, or 76 using an algorithm based on a common single or multiple target model.

[0053] Figure 6 summarizes the essential steps of the method once again, according to which the method comprises the following steps: S1: performing a one-dimensional Fourier transform in the distance dimension in the transform stage 50 S2 Fourier transform in the velocity dimension in a transformation stage 54 to generate an ambiguous spectrum 56 S3. Reconstructing the unambiguous spectrum 60 without undersampling, phase correcting and coherently adding all range-velocity subspectra in the reconstruction stage 58. S4. Detecting objects on the basis of the sum spectrum in the detection stage 62 Includes.

Claims

1. A method for object detection using an angle-resolved FMCW radar sensor, comprising: generating an at least three-dimensional spectrum (56) having a distance dimension (d), a Doppler dimension indicating the relative velocity (v) of the object, and an angular dimension (α, θ) based on received signals from multiple receive channels of the radar sensor; the radar sensor being undersampled in the Doppler dimension, such that the spectrum (56) is incomplete and ambiguous in that dimension; and generating multiple, interleaved sequences of time-equidistant frequency ramps (32, 36) to resolve the ambiguity.

1. A method for detecting an object using a frequency modulation scheme with frequency ramps (30, 34), wherein for each of a plurality of velocity hypotheses a velocity-dependent phase shift between the received signals obtained for the different sequences is modeled based on known time shifts between frequency ramps belonging to the different sequences and compared with a measured phase shift, characterized in that during the phase comparison an angle-dependent phase shift of the different receive channels is analyzed together with the velocity-dependent phase shift to reconstruct a complete and unambiguous spectrum (60), and the object detection is performed on the basis of the reconstructed spectrum (60).

2. 2. The method of claim 1, wherein the reconstruction of the unambiguous spectrum (60) is performed by velocity disambiguation combined with angle estimation based on a common single or multi-target model of velocity and angle dependence of phase.

3. The method of claim 1 , wherein the reconstruction of the unambiguous spectrum (60) is performed using a sparse reconstruction algorithm or a neural network.

4. The method of claim 3 , wherein the training data is generated using a simulation algorithm.

5. 5. The method according to claim 1, wherein undersampling is also performed in at least one angular dimension (α, θ) and angular ambiguity is resolved during the reconstruction of the unambiguous spectrum.

6. 6. The method according to claim 1, wherein the input data for the reconstruction of the unambiguous spectrum (60) is a multidimensional data structure (70; 72; 74; 76) containing spectral values ​​for at least one range-Doppler cell of the at least three-dimensional ambiguous spectrum (56).

7. 7. The method of claim 6, wherein the data structure (70) includes the spectral values ​​for all Doppler cells (68) within at least one range cell (66).

8. The method of claim 1 , wherein a combined velocity and angle beamforming algorithm is used in the disambiguation.

9. 9. The method according to claim 1, wherein the spectral values ​​of the unambiguous spectrum (60) are phase corrected and coherently summed for the Doppler dimension and at least one angular dimension, and the object detection is performed based on the coherent sum.

10. A radar sensor comprising a digital processor unit (22) in which the method according to any one of claims 1 to 9 is implemented.

11. 10. A radar sensor system comprising a radar sensor having an internal processor unit (22) and an external processor unit (48), characterized in that the method according to any one of claims 1 to 9 is implemented in part in the internal processor unit (22) and in part in the external processor unit (48).

12. A radar sensor system comprising a radar sensor and an external processor unit (48), characterized in that the method according to any one of claims 1 to 9 is implemented in the external processor unit (48).

Images