Apparatus and method for calibrating a multiple-input-multiple-output radar sensor

Abstract

Device (14) for calibrating a multiple-input-multiple-output, MIMO, radar sensor (12), comprising: an input interface (20) for receiving a target list of the MIMO radar sensor with angle data and channel data for a plurality of targets in an environment of a vehicle, wherein the angle data comprises information about a target angle at which a target is located and the channel data comprises information about received reflection signals of the target in individual channels of the MIMO radar sensor; a modeling unit (22) for generating model data for each target of the plurality of targets with information on expected reflection signals for an assumed detection of a single target at the target angle in a far field of the MIMO radar sensor based on the angle data; a processor unit (24) for determining a model error for each target of the plurality of targets with information on a deviation between the channel data and the model data of the target; a selection unit (28) for selecting one of the plurality of targets based on the determined model error, wherein the model error of the selected target is small compared to the model errors of the other targets; and an adaptation unit (30) for determining calibration coefficients with information for adapting channel outputs of the channels of the MIMO radar sensor based on the model data and the channel data, wherein the calibration coefficients compensate for the deviation of the channel data from the model data.

Description

The present invention relates to a device for calibrating a multiple-input multiple-output (MIMO) radar sensor and a corresponding method and to a system having a MIMO radar sensor and a device.Modern vehicles (cars, transporters, trucks, motor cycles etc.) comprise a multiplicity of sensors which provide information to the driver and control individual functions of the vehicle in a partially or fully automated manner. An important prerequisite here is the detection, recognition and modelling of the environment of the host vehicle. By means of environmental sensors, such as radar, lidar, ultrasonic and camera sensors, sensor data with information about the environment is captured. Based on the captured data and, if appropriate, with additional consideration of data available in the vehicle, objects in the environment of the vehicle can then be identified and classified. Based on this, for example, a behavior of an autonomous or semi-autonomous vehicle can be adapted to a current situation or additional information can be made available to the driver.A widely used sensor principle is radar technology. Radar sensors for vehicles typically include multiple transmitting and receiving elements and are operated as multiple-input multiple-output (MIMO) radars. The transmitting and receiving elements form virtual channels of the radar sensor (Rx / Tx antenna pairs), each channel representing a different position. By evaluating the phase and amplitudes of the radar signals on the different channels (channel responses), a target can be detected (angular position in azimuth and elevation, Doppler or relative speed and distance).During the production and processing of radar sensors, production-related differences and deviations occur. To improve the signal quality, it is therefore customary to measure each individual radar sensor after production in order to achieve calibration of the channel responses. The calibration often takes place only when the radar sensor is already installed in the vehicle. For this purpose, a target is usually generated in a measuring chamber at the angle of zero degrees azimuth and elevation and the channel response is measured. This consists of n complex values of the n available channels which are extracted from the range / doppler map for the corresponding target at a known position. A target at zero degrees should ideally produce the same response in amplitude and phase on all channels. The measured values therefore directly represent the deviation for the corresponding sensor and can be used for normalization. This approach yields (complex) calibration coefficients for the channels. These calibration coefficients are then applied by dividing by the range / doppler map after it has been calculated.One challenge is that the behavior of a radar sensor may change over its lifetime. Due to the effect of factors such as temperature changes, humidity, vibrations, etc., a high load and a degradation of the radar sensor may occur, in particular in the automotive sector. A calibration carried out once can deteriorate considerably and after a certain time only have a limited accuracy. Changes within the radar sensor may occur both in the long term, for example over several years, and in the short term, for example over a few minutes. Therefore, repeated (complicated) calibration or also (cost-intensive) replacement of the radar sensor is often required.DE 10 2014 208 899 A1 discloses a method for calibrating a MIMO radar sensor for motor vehicles. US 2018 / 0151958 A1 discloses a radar apparatus and an error correction method thereof.On the basis of this, the present invention is based on the object of improving the reliability and accuracy when using a radar sensor. In particular, the effort or costs for a new calibration should be reduced. Robustness with respect to degradation of the radar sensor is intended to be achieved.To achieve this object, in a first aspect, the invention relates to a device for calibrating a MIMO radar sensor, having:an input interface for receiving a target list of the MIMO radar sensor including angle data and channel data for a plurality of targets in a vicinity of a vehicle, wherein the angle data comprises information about a target angle at which a target is located and the channel data comprises information about received reflection signals of the target in individual channels of the MIMO radar sensor;a modeling unit for generating model data for each target of the plurality of targets with expected reflection signal information for assumed detection of a single target at the target angle in a far field of the MIMO radar sensor based on the angle data;a processing unit for determining a model error for each target of the plurality of targets with information on a deviation between the channel data and the model data of the target;a selection unit for selecting a target of the plurality of targets based on the determined model error, wherein the model error of the selected target is small compared to the model errors of the other targets; andan adjustment unit for determining calibration coefficients with information for adjusting channel outputs of the channels of the MIMO radar sensor based on the model data and the channel data, wherein the calibration coefficients compensate the deviation of the channel data from the model data.In a further aspect, the present invention relates to a system having a MIMO radar sensor and a device as described above.Further aspects of the invention relate to a method embodied according to the apparatus and to a computer program product having program code for carrying out the steps of the method when the program code is executed on a computer, and to a storage medium on which a computer program is stored which, when it is executed on a computer, brings about execution of the method described herein.Preferred embodiments of the invention are described in the dependent claims. It is understood that the features mentioned above and those still to be explained below can be used not only in the respectively specified combination, but also in other combinations or alone, without departing from the scope of the present invention. In particular, the apparatus, the system, the method and the computer program product can be embodied according to the embodiments described for the apparatus and the system in the dependent claims.According to the present invention, the sensor data of a MIMO radar sensor are received. The received target list includes, in particular, the azimuth and elevation angles of a target in a surrounding environment of a vehicle (angle data) and the channel responses of the various channels of the MIMO radar sensor (channel data). The sensor data are preferably received by the radar sensor after preprocessing. Based on the received data, a model of an assumed single target (single target model) perceived at the target angle of the real target in a far field of the radar sensor is first generated. A model is thus generated which comprises model data corresponding to the channel data. In other words, simulated ideal channel responses for an assumed target at the (real) target angle are generated as model data. A model error is calculated for this assumed single target. This model error represents a deviation between the real target or the real data (channel data) and the simulated individual target or the ideal data (model data). In the next step, at least one target is selected based on the ascertained model error. Specifically, a target having a small model error compared to the other targets is selected. For this selected target, the assumption is made that the real target is actually a single target. This assumption means that the deviation between channel data and model data corresponds to inaccurate calibration. Calibration coefficients are determined to adjust the outputs of the channels for future measurements to compensate for this deviation. In other words, the radar sensor is therefore calibrated to the effect that the channel responses or the channel outputs of the channels of the MIMO radar sensor are modified in such a way that the real data better correspond to the model for a selected target.The device according to the present invention thus enables a calibration of a radar sensor during the operation of the radar sensor. In particular, it is possible for the calibration to be carried out continuously and for the calibration to be updated continuously. The calibration coefficients for the channels are adjusted after each measurement. The channel data determined in the following time step are then determined or preprocessed on the basis of the new calibration coefficients. As a result, changes in the sensor can be immediately detected and compensated. The calibration coefficients correspond to the initial calibration coefficients determined during production and are preferably applied to the channel outputs in addition to these. In this respect, an additional calibration results in the sense of a fine adjustment of the channels. Both short and long term effects can be compensated. This results in improved detection and a more accurate assignment of the azimuth and elevation angles to a target. A higher accuracy is achieved.In a preferred embodiment, the modeling unit is designed to ascertain a scaling vector having average scaling factors for the individual channels of the MIMO radar sensor on the basis of the channel data. Furthermore, the modeling unit is designed to ascertain a control matrix based on the arrangement of the channels of the MIMO radar sensor and the angle data. In addition, the modeling unit is configured to generate the model data based on a multiplication of the control matrix with the scaling vector. First, in particular, an average phase and amplitude are estimated based on the channel data and a corresponding scaling vector is determined. Then a control matrix (also referred to as a steering vector) is calculated which reflects the arrangement of the (virtual) channels as well as the angle at which a target was perceived. The control matrix is multiplied by the scaling vector to determine the model data. A correct mapping of the sensor topology and of the current measured values is generated. In addition, an efficiently calculable approach results which can be executed with a high update frequency even with limited computing power.In a preferred embodiment, the processor unit is designed to ascertain a mean square error over the channels of the MIMO radar sensor. The model error is calculated as a mean square error. An error is determined and squared for each channel. The square root of the mean value then corresponds to the square mean value. This root mean square value serves as a measure of the deviation between the channel data and the model data of the target. This results in an efficiently calculable possibility for ascertaining the model error. A simple assessment of the deviation between model and reality is achieved.In an advantageous embodiment, the selection unit is configured to select a target having a mean square error that lies below a threshold value. The threshold value is preferably predefined. A comparison with a threshold value represents an efficient possibility for an evaluation. Only targets having a mean square error below a threshold will be further considered. A simple calculability is achieved.In an advantageous embodiment, the selection unit is configured to select a target based on a signal-to-noise ratio of the received reflection signals. Preferably, a target is formed having a signal-to-noise ratio above a threshold. It is possible to additionally take account of a signal-to-noise ratio when selecting the targets. Only strong signals or relatively strong signals are then included in the further processing. This results in an improvement in the accuracy of the method, since the perception of strong targets, i.e. targets with a high signal-to-noise ratio, is improved or more accurate.In an advantageous embodiment, the adaptation unit is designed to determine the calibration coefficients on the basis of a moving average. By using a moving average, smoothing over time can be achieved. It is constantly responding in small steps to changes in the channel responses of the radar sensor. By using a moving average, an error or incorrect calibration due to one-time effects is avoided.In an advantageous embodiment, the apparatus comprises a displacement compensation unit for determining an angular displacement based on the calibration coefficients and for calculating and applying a compensation factor to the calibration coefficients if the angular displacement is above a threshold value. The threshold value is again preferably predefined. In other words, the displacement compensation unit is configured to compensate for a drift that can result when the calibration coefficients are applied. A method according to the present invention may result in the zero point of the radar sensor being shifted over time. If such a displacement is detected, a compensation is carried out. The system is again directed toward the original set zero point. This has the advantage of ensuring long-term accuracy.The displacement compensation unit is preferably designed to perform beamforming for a zero point and for points in a vicinity of the zero point on the basis of the calibration coefficients. Preferably, all points in the vicinity of the zero point have the same angular distance from the zero point. Furthermore, the displacement compensation unit is configured to calculate and apply the compensation factor if the magnitude response of the beamforming for a point in the vicinity of the zero point is greater than the magnitude response of the beamforming for the zero point. A direct vicinity can be defined, for example, by a 0.02 degree distance in azimuth and / or elevation. If it is determined that beamforming at one of these points in the vicinity of the zero point delivers a greater response in terms of amount than the zero point itself, an adjustment is carried out. In particular, it is possible that the values of the determined direction with the greatest channel response in terms of amount is used as a basis for the calculation of a quotient for applying to the calibration coefficients. Long-term accuracy is ensured. A deviation in the sense of a bias is avoided.In a further advantageous embodiment, the displacement compensation unit is designed to calculate the compensation factor based on one half of an angle between the zero point and the point in the vicinity of the zero point with the highest magnitude response of beamforming. Preferably, a center between the zero point and the determined point with the highest response in terms of amount can be used as a basis for the readjustment or fine calibration. Long term accuracy is maintained.In a further advantageous embodiment, the adaptation unit is designed to determine the calibration coefficients of a previous time step if it is determined based on the target list that the vehicle is not moving. It may lead to errors whenever the same goals are considered. Therefore, it is possible that the fine calibration according to the invention is temporarily deactivated if the same targets are always detected over a long time when the vehicle is stationary. This ensures that no constant errors propagate in the determination of the calibration coefficients. Long term accuracy and targeted calibration are ensured.In a further advantageous embodiment, the modeling unit is configured to generate the model data based on an influence of the individual channels on one another. It is possible that in modelling no unique relationship between a channel response and a single calibration coefficient is used, but a more general modelling is established. In particular, the influence of adjacent channels on one another can be taken into account. This results in further improved accuracy in the generation of the model data (modeling).In an advantageous configuration of the system, the MIMO radar sensor is configured to emit and receive radar signals in a frequency range from 70 GHz to 80 GHz, preferably from 77 GHz. These frequencies have been found to be particularly advantageous for use in automotive applications.A radar sensor emits a radar signal and receives reflections of the radar signal on objects within a range of view of the radar sensor. The field of view refers to an area within which objects can be detected. A surroundings of a vehicle include, in particular, a range visible from a radar sensor mounted on the vehicle in the surroundings of the vehicle. A radar sensor may also include a plurality of individual sensors, which enable a 360 degree surround view, for example, and may thus record a complete image of the surroundings of the vehicle. The sensor data of a radar signal include, in particular, a distance, a point velocity corresponding to micro-Doppler information, an elevation angle and an azimuth angle for different detections of the radar sensor. Below a scanning point, an individual point, i.e. an individual detection with the above-mentioned information, is detected. Information is understood to mean information. Usually, a plurality of scan points are generated during a measurement cycle of the radar sensor. A measurement cycle is understood to mean a single passage through the visible region. The scan points recorded in a measurement cycle can be referred to as a target list (radar target list). The values and coefficients referred to herein may each be vector sizes comprising multiple values. It is possible that the coefficients and values herein are complex values.The invention will be described and explained in more detail below with reference to some selected exemplary embodiments in conjunction with the accompanying drawings. The following are shown: FIG. 1 shows a schematic illustration of a system according to the invention in a vehicle in an environment; FIG. 2 shows a schematic illustration of an apparatus according to the invention; FIGS. 3 a, 3 b show a schematic representation of a data processing according to the invention; FIG. 4 is a schematic illustration of a displacement compensation according to the invention; and FIG. 5 shows a schematic illustration of a method according to the invention.FIG. 1 schematically illustrates a system 10 according to the present invention having a MIMO radar sensor 12 and a device 14 for calibrating MIMO radar sensor 12. In the exemplary embodiment shown, the system 10 is integrated into a vehicle 16. Objects 18 in the environment of vehicle 16 are detected by MIMO radar sensor 12. For this purpose, the MIMO radar sensor 12 comprises a plurality of transmitting elements and a plurality of receiving elements. The transmitting elements transmit radar signals into the environment of the vehicle 16. reflections of the radar signals at objects 18 in the environment of the vehicle 16 are received via the receiving elements. The combination of the transmission elements with the reception elements thus results in a number of virtual reception channels, each reception channel corresponding to a combination of a transmission element with a reception element. For each channel, a channel response having an amplitude and a phase is received.When starting up such a MIMO radar sensor 12, a calibration of the individual channels is normally carried out in order to enable accurate measurements. For this purpose, a radar target (for example a corner reflector) is usually used in an absorber hall. At a position of the target at zero degrees, the same channel response should be generated on all channels. During operation of the vehicle, this calibration deteriorates over time, so that a new calibration must be carried out. According to the present invention, a device for calibrating the MIMO radar sensor is now proposed, which device makes possible a continuous (inline) calibration based on the current measurement data.FIG. 2 schematically illustrates a device 14 according to the invention. The device 14 can be integrated, for example, in a vehicle control device of a vehicle. It is also possible for the device 14 to be integrated with the MIMO radar sensor 12. It is likewise possible for the device 14 to be executed as a software product in a processor of a vehicle controller or of a radar sensor.Device 14 includes an input interface 20 via which a target list of the MIMO radar sensor is received. The target list is a radar target list which comprises at least angle data, in particular elevation and azimuth angles, and channel data corresponding to the individual channel responses of the virtual channels of the MIMO radar sensor 12 for a plurality of targets. The input interface 20 may be connected to a vehicle bus system, for example.Furthermore, the device 14 according to the invention comprises a modelling unit 22, in which an individual target model is determined or model data is generated for each target of the target list. In particular, channel responses are calculated as model data, which would be expected for the detection of an individual target in a far field of the radar sensor (for example at a distance of 20 meters) at the real angle with optimum calibration. To generate the model data, the angle data (azimuth and elevation) of the detected target is used to determine an expected channel response for a single target (single target model).Furthermore, the device 14 comprises a processor unit 24. In particular, a deviation between the real data, i.e. the channel data, and the previously generated model data, i.e. the modeled data for an assumed individual target, is calculated. For this purpose, an error metric is preferably determined based on the mean square error (MSE) between the measured signal and the individual target model. The ratio of the MSE to the signal power represents a measure for the deviation of the signal from an ideal individual target with the same signal strength.For the derivation of the MSE, it is assumed here that a vector x of the signal channel responses results for a target:Here, A nm is the control matrix (steering matrix) prepared from the solid angle δ. The dimension n corresponds to the number of channels, the dimension m corresponds to the number of (point) targets that contribute to the signal. The unknown vector s of complex scaling factors (scaling vector) has a length m. The individual scaling factors correspond to the amplitude and phase shift of the m point targets. In addition, a noise component σ results. The model error (single target error) considered for calibration and the corresponding models (single target models) are apparent from the equations for m=1.The scaling vector s can be estimated as mean scaling ≅from the measured signal according toIn this case, A + denotes the pseudoinversary of the control matrix A. Starting from this, the result is a model signal for the individual target (model data) when normalization is effected to the input signal (channel data)The model error may then be determined as a mean square error between channel data and model data according to.The approach is schematically illustrated in FIGS. 3 aand 3 b. For example, for an assumed linear antenna array 26, in an ideal case for a target being perceived at an angle, the combination of amplitude (upper diagram) and phase (lower diagram) shown in FIG. 3 aresults. An amount of 1 is assumed here for the amplitude. The phase is assumed to start at zero. The representation corresponds to a control matrix of an individual target which is perceived at the solid angle δ. FIG. 3 a thus illustrates an approach for modeling.In FIG. 3 b, the solid lines schematically show exemplary real measured values or channel data. In order to determine the model error for the target, the estimated complex scaling vector in the absolute value ≅ is then applied to the steering vector (cf. FIG. 3 a). The difference in amplitude and phase corresponds to the model error for this target. The dashed lines correspond to the model or the model data.Returning to FIG. 2, the apparatus 14 further comprises a selection unit 28. In particular, a target is selected that has only a small model error. In other words, only those detections are considered which have a small deviation from the individual target model. For these targets, it is assumed that they are actually real individual targets or targets well separated in all radar dimensions.In this selection of the target in the selection unit 28, it is possible to take into account additional criteria such as a high signal-to-noise ratio and / or a high signal power in order to further restrict the selection of the suitable targets for the calibration and thus achieve a more efficient calibration.Calibration coefficients are now determined in an adaptation unit 30 of the device 14 in order to be able to carry out a calibration of the channel outputs of the channels of the MIMO radar sensor on the basis thereof. Assuming that a selected target is indeed a single or well separated target in all radar dimensions, a relationship between the complex signal (channel data) and the single target model (model data) is determined and assuming that this relationship corresponds to a calibration error for the target under consideration. In other words, a calibration error is thus determined, which can then be compensated by calibration coefficients.Since the assumptions made for a single target can be fulfilled only to a limited extent (real targets can represent an ideal single target only in exceptional cases, for example, on account of their extent), the adaptation unit 30 is preferably designed to perform long-term observation. For this purpose, a large number of different targets are observed. For fine calibration estimation or calibration coefficients determination, a moving average filter is applied for each channel. Per cycle of signal processing, the error determined in this time step is therefore only partially taken into account in the determination of the calibration coefficients. In signal processing in the MIMO radar sensor, an improvement can be achieved by applying the calibration coefficients together with the compensation based on the calibration carried out during the production calibration. After a settling time, the quality of the radar signal with respect to the angular dimensions is significantly increased.The device 14 according to the invention optionally also comprises a displacement compensation unit 32. Due to the separate filtering for the individual channels, an error can occur in the signal processing in the determination of the angles for a target. As a result, the zero point of the angles already set during the production or the initial calibration can shift. In order to prevent this error, it is advantageous that the calibration effect arising from the application of the calibration coefficients to the channel outputs of the channels of the MIMO radar sensor is continuously monitored. If an angle-dependent shift results over all channels, this can then be compensated. For this purpose, it is preferably provided that a compensation factor is calculated and applied when the angle shift is above a threshold value.FIG. 4 schematically shows an approach for calculating the compensation factor. The azimuth is shown on the horizontal axis. The elevation is plotted on the vertical axis. Using the values of the complex fine calibration (calibration coefficients), beamforming is carried out for a zero point N at zero degree azimuth and zero degree elevation. Further, beamforming is performed for points P in a direct vicinity of the zero point N. For example, a direct neighborhood may be defined by a deviation of 0.02 degrees in both directions, respectively. In the illustration, the vicinity of the zero point N therefore comprises nine points P. The absolute response of the beamforming is determined for the zero point and the points P. If the absolute-value response is the greatest at zero degrees, no relevant angular displacement has taken place. If it is determined that the absolute response of the beamforming ends in the case of an angle combination outside the zero point N, i.e. is greater for a point P in the vicinity of the zero point N, a compensation factor is determined. In particular, the corresponding control vector (control matrix) can be divided at half an angle in the direction of the point P with maximum magnitude response of the beamforming complex. In the example shown, the point S, marked with the cross, between the zero point N and the point P is obtained in the vicinity of the zero point N. By using the compensation factor determined, a more continuous angular drift can be actively compensated for continuously.The adaptation unit 30 can be configured to re-output the calibration coefficients determined in the previous time step if the same targets are always detected over a longer time when the vehicle is stationary. In other words, the fine calibration is temporarily deactivated and no new calibration coefficients are calculated. This can increase reliability because multiple consideration of the same targets can cause bias.Optionally, it is possible for an estimation of the antenna position to be carried out. If the calibration error is compensated only by a complex factor, the deviation of the feed lines can be compensated for with a corresponding amplitude and phase (attenuation and length). It is also possible to estimate a displacement of the virtual antenna array directly. For example, the section of the incident single-target wavefront with the antenna plane can be considered as a modification. A remaining fine calibration error can then be compensated by displacement perpendicular to the section. Note that these approaches require targets in different directions to implement compensation of the antenna structure in all directions.Optionally, it is also possible to include an estimate of the influence of adjacent leads. Instead of establishing a unique relationship between the respective channel response and a calibration coefficient, a more general modeling of the channels may also be established. In particular, the influence of a channel on an adjacent channel (supply lines) can be taken into account in the modeling. For this purpose, a plurality of measurements are then required in order to solve an equation system with a plurality of unknown (complex) scaling factors between the channels. In principle, however, the previously described approach remains unchanged.FIG. 5 schematically illustrates a method according to the invention. The method comprises steps of receiving S 10 a target list, generating S 12 model data, determining S 14 a model error, selecting S 16 a target, and determining S 18 calibration coefficients. The method may be implemented, for example, as software which is executed on a processor of a radar sensor or on a vehicle control device. The method according to the present invention is preferably used during normal operation of a radar sensor. It is thus continuously checked whether the current driving situation is suitable for fine calibration.The invention has been fully described and explained with reference to the drawings and the specification. The description and explanation are to be taken by way of example and not limitation. The invention is not limited to the disclosed embodiments. Other embodiments or variations will become apparent to those skilled in the art upon use of the present invention, as well as upon a detailed analysis of the drawings, disclosure and appended claims.In the claims, the words "comprise" and "with" do not exclude the presence of further elements or steps. The undefined article "a" or "an" does not exclude the presence of a plurality. A single element or unit may perform the functions of several of the units recited in the claims. An element, a unit, an interface, a device and a system can be partially or completely implemented in hardware and / or in software. The mere naming of some measures in several different dependent claims is not to be understood as meaning that a combination of these measures cannot likewise be used advantageously. A computer program can be stored / distributed on a non-volatile data carrier, for example on an optical memory or on a solid state drive (SSD). A computer program can be distributed together with hardware and / or as part of hardware, for example by means of the Internet or by means of wired or wireless communication systems. Reference signs in the patent claims should be understood to be non-limiting.Reference numerals denote reference numerals10 System 12 MIMO radar sensor 14 Device 16 Vehicle 18 Object 20 Input interface 22 Modeling unit 24 Processor unit 26 Antenna array 28 Selection unit 30 Adaptation unit 32 Displacement compensation unit

Claims

An apparatus (14) for calibrating a multiple-input multiple-output, MIMO, radar sensor (12), comprising: an input interface (20) for receiving a target list of the MIMO radar sensor with angle data and channel data for a plurality of targets in an environment of a vehicle, wherein the angle data comprises information about a target angle at which a target is located and the channel data comprises information about received reflection signals of the target in individual channels of the MIMO radar sensor; a modeling unit (22) for generating model data for each target of the plurality of targets with information about expected reflection signals for an assumed detection of an individual target at the target angle in a far field of the MIMO radar sensor based on the angle data; a processor unit (24) for determining a model error for each target of the plurality of targets with information on a deviation between the channel data and the model data of the target; a selection unit (28) for selecting a target of the plurality of targets based on the determined model error, wherein the model error of the selected target is small compared to the model errors of the other targets; and an adjustment unit (30) for determining calibration coefficients with information on an adjustment of channel outputs of the channels of the MIMO radar sensor based on the model data and the channel data, wherein the calibration coefficients compensate the deviation of the channel data from the model data.The apparatus (14) according to claim 1, wherein the modeling unit (22) is configured to determine a scaling vector with mean scaling factors for the individual channels of the MIMO radar sensor (12) based on the channel data; to determine a control matrix based on the arrangement of the channels of the MIMO radar sensor and the angle data; and to generate the model data based on a multiplication of the control matrix with the scaling vector.The apparatus (14) according to any one of the preceding claims, wherein the processor unit (24) is configured to determine a mean square error across the channels of the MIMO radar sensor (12).The apparatus (14) according to claim 3, wherein the selection unit (28) is configured to select a target having a mean square error that is below a threshold; and the threshold is preferably predefined.The device (14) according to any one of the preceding claims, wherein the selection unit (28) is configured to select a target based on a signal-to-noise ratio of the received reflection signals, wherein preferably a target having a signal-to-noise ratio above a threshold value is selected.The apparatus (14) according to any of the preceding claims, wherein the adjustment unit (30) is configured to determine the calibration coefficients based on a moving average.The apparatus (14) according to any of the preceding claims, comprising a displacement compensation unit (32) for determining an angular displacement based on the calibration coefficients and for calculating and applying a compensation factor to the calibration coefficients if the angular displacement is above a threshold value, wherein the threshold value is preferably predefined.The apparatus (14) according to claim 7, wherein the displacement compensation unit (32) is configured to perform beamforming for a zero point and for points in a vicinity of the zero point based on the calibration coefficients, wherein preferably all points in the vicinity of the zero point have the same angular distance from the zero point; and to calculate and apply the compensation factor if the magnitude response of the beamforming for a point in the vicinity of the zero point is greater than the magnitude response of the beamforming for the zero point.The apparatus (14) according to claim 8, wherein the displacement compensation unit (32) is configured to calculate the compensation factor based on a half of an angle between the zero point and the point in the vicinity of the zero point having the highest magnitude response of beamforming.The device (14) according to any one of the preceding claims, wherein the adjustment unit (30) is configured to determine the calibration coefficients of a previous time step if it is determined based on the target list that the vehicle is not moving.The device (14) according to claim 1, wherein the modelling unit (22) is configured to generate the model data based on an influence of the individual channels on one another.System (10) having a MIMO radar sensor (12) and a device (14) according to one of Claims 1 to 11.The system (10) according to claim 12, wherein the MIMO radar sensor (12) is configured to emit and receive radar signals in a frequency range of 70 GHz to 80 GHz, preferably of 77 GHz.A method for calibrating a multiple-input multiple-output, MIMO, radar sensor (12), comprising the steps of: receiving (S10) a target list of the MIMO radar sensor with angle data and channel data for a plurality of targets in a vicinity of a vehicle, wherein the angle data comprises information about a target angle at which a target is located and the channel data comprises information about received reflection signals of the target in individual channels of the MIMO radar sensor; generating (S12) model data for each target of the plurality of targets with information about expected reflection signals for an assumed detection of an individual target at the target angle in a far field of the MIMO radar sensor based on the angle data; determining (S14) a model error for each target of the plurality of targets with information on a deviation between the channel data and the model data of the target; selecting (S16) a target of the plurality of targets based on the determined model error, wherein the model error of the selected target is small compared to the model errors of the other targets; and determining (S18) calibration coefficients with information on an adjustment of channel outputs of the channels of the MIMO radar sensor based on the model data and the channel data, wherein the calibration coefficients compensate the deviation of the channel data from the model data.A computer program product having program code for performing the steps of the method of claim 14, when the program code is executed on a computer.

Images