Method for providing calibration data for antenna diagrams - Patents.com

Abstract

1. A method for providing calibration data for calibrating antenna diagrams of devices configured to transmit and / or receive electromagnetic radiation and belonging to a common manufacturing model, the method comprising: a) selecting a set of devices from the manufacturing model to create a training file; b) for each selected device, measuring a target diagram (34) using a first number of measurement points (24, 26) and creating an associated source diagram (28) using a second number of measurement points (24) that is less than the first number; and storing the target and source diagrams (34, 28) in the training file; c) using the training file to train a neural network (30) to determine associated target diagrams (34) based on the source diagrams (28); d) measuring the source diagrams (28) of remaining devices from the manufacturing model; e) creating associated target diagrams (34) using the neural network (30); and f) using the target diagrams (34) as calibration data for the remaining devices.

Description

[Technical Field]

[0001] The present invention relates to a method for providing calibration data for calibrating antenna diagrams of devices configured to transmit and / or receive electromagnetic radiation and belonging to a common manufacturing model.

[0002] In particular, the invention relates to the calibration of the antenna diagram of radar sensors used to detect the traffic environment, for example in motor vehicles. [Background technology]

[0003] Angle-resolved radar sensors for automobiles typically include transmit and receive antennas with multiple antenna elements positioned offset from one another. Angular information about the radar target being located is encoded in the amplitude and phase relationships between the signals received by the different antenna elements and can be extracted from the received data by matching the amplitude and phase of the received signals with an antenna diagram that describes the angular dependence of amplitude and phase. Similarly, matching with the antenna diagram can determine the radar cross section of the target being located, and thus a measure of the extent of the object being located.

[0004] Ideally, radar sensors belonging to the same manufacturing model and therefore identical in structure would all have the same antenna diagram. However, due to unavoidable manufacturing tolerances and other interferences, the actual antenna diagrams of radar sensors differ slightly from each other. Therefore, to obtain accurate positioning data, these deviations must be corrected by individually calibrating the antenna diagram for each radar sensor. For this purpose, calibration data is required. This is obtained by measuring the amplitude and phase of the received signal under standardized conditions. For this purpose, radar echoes from a standardized reflector positioned at a known angle relative to the radar sensor are evaluated. Each reflector position represents a measurement point where individual measurements must be performed. To obtain calibration data that shows the angular dependence of amplitude and phase for a given sensor with high angular resolution, the angular distance between individual measurement points should be as small as possible. However, this means that a relatively large number of measurements must be performed for each individual sensor. The number of measurements increases as the angular range covered by the calibration data increases, significantly increasing for radar sensors with two-dimensional (azimuth and elevation) angular resolution. Accurate calibration of sensors incurs significant costs due to the time and effort required for the calibration measurements. Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present invention is to minimize the number of calibration measurements required for a given quality of calibration data. [Means for solving the problem]

[0006] This object, according to the present invention, a) selecting a set of devices from a manufacturing model to create a training file; b) for each selected device, measuring a target diagram using a first number of measurement points and creating an associated source diagram using a second number of measurement points that is less than the first number, and storing the target and source diagrams in a training file; c) using the training file to train the neural network to determine each associated target diagram based on the source diagram; d) measuring source diagrams of the remaining devices from the manufacturing model; e) generating an associated target diagram using a neural network; f) using the target diagram as calibration data for the remaining devices; This is achieved by a method comprising:

[0007] The above-mentioned "device" may be a radar sensor or another system capable of transmitting and / or receiving electromagnetic radiation. "Target diagram" shall mean an antenna diagram that reproduces the angular dependence of the amplitude and phase of the received signal with the required angular resolution. "Source diagram" shall mean an antenna diagram based on a smaller number of measurement points compared to the target diagram.

[0008] The present invention is based on the finding that for sensors (devices) of the same manufacturing model, there are characteristic correlations between different parts of the antenna diagram, and that based on measurement results obtained for measurement points in one part of the diagram, the appearance of other parts of the diagram can be inferred without actually measuring the other parts of the diagram. According to the present invention, a neural network is trained to recognize these correlations based on appropriate training data. Once the neural network is trained, it is sufficient to measure only a source diagram with a relatively small number of measurement points for a single sensor and input this data into the neural network, which then generates a complete target diagram that can then be used to calibrate the sensor.

[0009] In this way, the costs associated with calibrating the sensors can be significantly reduced. Advantageous embodiments and developments of the invention are defined in the dependent claims. A source diagram is a diagram that covers the same angular range as its respective associated target diagram, but has a lower angular resolution due to a smaller number of measurement points.

[0010] On the other hand, the source diagram may cover a smaller angular range than that of the target diagram, in which case the neural network is used to extrapolate the calibration data for a larger angular range.

[0011] For sensors with angular resolution in two dimensions, the source diagram may represent one-dimensional measurement line (e.g., azimuth) through the two-dimensional measurement field, while the target diagram determined by the neural network further includes calibration data for at least one measurement line in the second dimension (elevation).

[0012] A complete target diagram is measured with the sensors (devices) selected to create the training file. The target diagram is simply reduced to a smaller set of measurement points, allowing the associated source diagram to be generated without additional measurements. Here, when training the network, the source diagram forms the input data, and the associated target diagram provides feedback based on which the weights of the neural connections in the network are adjusted, for example, by backpropagation.

[0013] In one embodiment, calibration data is stored for each sensor, this calibration data including both a source diagram and a target diagram, and when a source diagram is obtained by reducing measurement points, the source diagram is included in the target diagram anyway.

[0014] In another embodiment, the measured source diagram is stored in each sensor to be calibrated, but instead of the target diagram, a trained neural network is stored in the sensor, which then provides the necessary calibration data during operation of the sensor. This embodiment is advantageous when the amount of data for the parameters of the trained neural network is smaller than the amount of data for the complete target diagram.

[0015] The subject of the invention is also a neural network trained to generate calibration data according to the method described above, and a radar sensor in which such a neural network for generating calibration data is stored.

[0016] Exemplary embodiments are explained in more detail below on the basis of the drawings. [Brief explanation of the drawings]

[0017] [Figure 1] FIG. 2 is a diagram illustrating an example of a measurement field of a radar sensor. [Figure 2]2A and 2B are diagrams showing examples of four amplitude antenna diagrams for a measurement line passing through the measurement field along line II-II in FIG. 1; [Figure 3] 3 shows an example of an amplitude antenna diagram of the same four radar sensors as in FIG. 2, corresponding to a measurement line through the measurement field along the line III-III in FIG. 1; [Figure 4] FIG. 1 shows a measurement field with different measurement lines. [Figure 5] 5 shows an example of an amplitude and phase antenna diagram corresponding to a measurement line passing through the measurement field along line VV in FIG. 4. FIG. [Figure 6] 6 shows amplitude and phase antenna diagrams corresponding to measurement lines passing through the measurement field along line VI-VI in FIG. 4. [Figure 7] FIG. 1 shows a diagram for explaining a method according to an exemplary embodiment of the present invention. [Figure 8] 1 is a flow diagram of a method for training a neural network. [Figure 9] 3 is a flow chart for a first exemplary embodiment of a method for providing calibration data according to this invention. [Figure 10] 4 is a flow chart of a method according to a second exemplary embodiment of the present invention. [Figure 11] FIG. 8 shows a diagram similar to FIG. 7 for a method according to a modified exemplary embodiment of the invention. [Figure 12] FIG. 8 shows a diagram similar to FIG. 7 for a method according to a modified exemplary embodiment of the invention. DETAILED DESCRIPTION OF THE INVENTION

[0018] 1 shows the measurement field 10 of a radar sensor with angular resolution in two dimensions, i.e., in both azimuth (horizontal axis in the diagram) and elevation (vertical axis). The amplitudes (complex amplitude magnitudes) of the signals received from different directions are shown by contour lines 12.

[0019] In a conventional method for providing calibration data for a radar sensor, horizontal measurement lines, e.g., corresponding to line II in FIG. 1, and vertical measurement lines, e.g., corresponding to line III in FIG. 1, are set through the measurement field 10, and on each measurement line the amplitude of the received signal and the phase difference between the different antenna elements of the radar sensor are measured with as high an angular resolution as possible.

[0020] Figure 2 shows the results of amplitude measurements on line II of Figure 1 for four identically configured radar sensors. Four curves 14a, 16a, 18a, and 20a can be seen in Figure 2, showing the amplitude as a function of azimuth angle for the four sensors, respectively. Each curve is a portion of the antenna diagram for the radar sensor in question. It can be seen that all four curves have similar profiles, but due to manufacturing tolerances, material properties, etc., there are some deviations between the curves and from the ideal antenna diagram that might theoretically be expected for this manufactured model sensor.

[0021] FIG. 3 shows the corresponding curves 14b, 16b, 18b, and 20b measured along line III of FIG. The examples shown in Figures 2 and 3 illustrate a phenomenon exploited by the invention described herein: deviations between sensors in one portion of the antenna diagram are strongly correlated with deviations in other portions of the antenna diagram. For example, Figure 2 shows that curves 14a and 16a have very similar profiles and differ only slightly from one another. This small deviation between the curves correlates with the similarly small deviation between corresponding curves 14b and 16b in Figure 3, i.e., in the portion of the antenna diagram corresponding to the measurement line in the elevation angle direction. In particular, in this example, curve 14b has a noticeable anomaly in the elevation angle range between 12° and 15°, which is indicated by an arrow in Figure 3. This anomaly is that the curve extends more flatly in this range than in adjacent angle ranges. The same anomaly is also observed in curve 16b.

[0022] Such observations suggest that sensors that exhibit very similar behavior in one part of the antenna diagram (here, the measurement line along line II) will also exhibit similar behavior in other parts of the diagram (here, the measurement line along line III).

[0023] Therefore, in accordance with the present invention, a neural network is used to recognize such regularities or patterns in the antenna diagrams of different sensors and then, based on these regularities, predict the shape of the antenna diagram in areas where no measurements have been performed.

[0024] First, based on Figures 4 to 6, a conventional method for providing calibration data by measuring antenna diagrams is illustrated. Figure 4 shows a radar sensor measurement field 22 with measurement lines V and VI extending through the center of the measurement field, along which the antenna diagram is measured. Figure 5 shows amplitude and phase antenna diagrams taken along measurement line V. For this purpose, multiple measurements were performed at measurement points 24 distributed along measurement line V. Each measurement point 24 represents a measurement in which a reflector is positioned on measurement line V, the radar echo of the reflector is measured, and the azimuth angle of the reflector corresponds to the position of the measurement point on the horizontal axis in Figure 5. The upper part of Figure 5 shows the amplitude antenna diagram for an individual radar sensor, i.e., the measured amplitude (dB) as a function of azimuth angle (°). The lower part of Figure 5 shows the phase antenna diagram, i.e., the phase difference between adjacent antenna elements as a function of azimuth angle. This phase difference varies within the range of -π to +π, where the values ​​-π and +π can be considered identical since they both correspond to a phase angle of 180°. It can be seen that the phase difference was measured at the same measurement point 24 as the amplitude.

[0025] 6 shows the corresponding amplitude and phase antenna diagrams taken on measurement line IV. Again, amplitude and phase measurements were performed at a number of measurement points 26 evenly distributed on measurement line IV.

[0026] Even if limited to measuring the antenna diagram on measurement lines V and VI, this method allows a large number of individual measurements to be carried out for each sensor (corresponding to the sum of measurement points 24 and 26).

[0027] 7 shows how the number of individual measurements required can be significantly reduced: for each sensor to be calibrated, actual measurements are performed at only a smaller number of measurement points, and thus a source diagram 28 is obtained that represents only a portion of the complete antenna diagram.

[0028] In the example shown in Figure 7, the source diagram 28 is an amplitude and phase antenna diagram for measurement line V in Figure 4. Therefore, the source diagram 28 represents only the azimuth measurement line through the measurement field; no measurements need to be performed on measurement line VI extending in the elevation direction. This means that the number of individual measurements required corresponds only to the number of measurement points 24 on measurement line V.

[0029] A neural network 30 specially trained for this application is then used to generate a composite diagram 32 showing the amplitude and phase differences on measurement line VI in Figure 4, thereby completing the antenna diagram. The source diagram 28 and the composite diagram 32 together form a so-called target diagram 34, which provides calibration data for the calibration of the radar sensor.

[0030] The neural network 30 has an input stage IN that supplies the real and imaginary parts of the complex amplitude for each measurement point 24 of the source diagram 28 to neurons 38 of a first layer 40 of the neural network. The information of the neurons 38 of the first layer 40 is then further processed in stages in hidden layers 42, until finally, in the output layer 44, for each neuron, a predicted value of the real or imaginary part of the complex amplitude is obtained, which corresponds to the measurement point 26 on the measurement line VI in Figure 4. The amplitude and phase values ​​are then output via an output stage OUT of the neural network, which together form the synthetic antenna diagram 32.

[0031] Neural network 30 can have any architecture known for neural networks. In the illustrated example, at least the lower layers 40, 42 form a fully interconnected network, with each neuron in the first layer 40 influencing the state of each neuron in the subsequent layer 42. Optionally, neural network 30 may be a convolutional neural network, such as those often used in pattern recognition algorithms.

[0032] However, the neural network must be trained on appropriate training data until it is able to fulfill the functions shown in Figure 7. The essential steps of the method for training the network are shown as a flow diagram in Figure 8.

[0033] In step S1, a certain number N of sensors are selected from the production model of radar sensors for which calibration data is provided, and for each of these sensors, a source diagram 28 and a target diagram 34 are measured in the conventional manner according to the method shown in Figures 4 to 6. However, data regarding the source diagram 28 does not need to be measured separately, since the measurement results are obtained in any case when the complete target diagram 34 is measured. The sum of the source diagrams 28 and target diagrams 34 for all N sensors thus obtained in step S1 forms a training file 36, which provides training data for the neural network 30.

[0034] Next, in step S2, actual training of the neural network 30 occurs. Before training, the network 30 is in an initial configuration. In this initial configuration, each neural connection between a neuron 38 in one layer and a neuron in a downstream layer has a weight that determines how strong and in which direction the state of the upstream neuron changes to the state of the downstream neuron. Then, in a first training step, a first source diagram 28 is input to the network via the input stage IN. The result obtained from the output stage OUT is compared with the associated target diagram 34 in the training file 36. Based on the deviation between the result and the target diagram 34, the weights of the neural connections are modified so that when the same input is input again, the result will be closer to the desired target diagram 34. Then, in the next training step, a source diagram 28 for a different sensor is input to the network, and the weights are again modified based on the result. When the number N of source file and target file pairs in the training file 36 is sufficiently large, the weights gradually converge to a configuration that allows the network to relatively accurately predict the associated target diagram 34 for each source diagram 28. The sum of the weights then defines a transfer function 38, which assigns to each source diagram an associated target diagram.

[0035] In practice, a validation phase will likely follow, in which an additional number of sensors are selected from the production model to test the performance of the neural network 30. If the tests pass, the network is ready for use, and calibration data (i.e., target files 34) can be generated for each sensor in the production model using the method shown in FIG.

[0036] 9 is a flow diagram of a possible method for calibrating a sensor. In this method, in a first step S11, a source diagram 28 for the sensor to be calibrated is measured. In step S12, this source file is input into a neural network 30, and a target diagram 34 is generated using or a transfer function 38 defined by the neural network. Then, in step S13, the target diagram 34 is stored, and also includes the original source diagram 28 as a partial data set. This target diagram 34 then provides the calibration data with which the sensor is calibrated before start-up.

[0037] An alternative method is shown in Figure 10. In this method, the first step S11 is identical to step S11 in Figure 9, i.e., a source diagram for the sensor to be calibrated is measured. In a next step 12', the source diagram 28 and the neural network 30 are stored in the digital memory of the sensor. Storing the neural network here means storing the network pattern of neurons (connectome) and the weights of all neural connections.

[0038] After step S12', the sensor is already ready for use. Once the sensor is operational, in a further step S13', a neural network 30 is executed in the sensor's electronics and, using a transfer function 38, a target diagram 34 and therefore calibration data is generated. Here, the calibration can be limited to the part of the antenna diagram for which there is current positioning data.

[0039] An alternative method for providing calibration data is shown in Figure 11, in a diagram similar to Figure 7. For simplicity, we will assume that the sensor to be calibrated has angular resolution in azimuth only.

[0040] In this case, the source diagram 28' includes only the measurement points 24' within a limited azimuth angle range. Based on the source diagram 28, the neural network 30 extrapolates the amplitude and phase difference to the full angular range, thereby providing a composite diagram 32'. The composite diagram 32' spans the radar sensor's full detection angle range and includes the source diagram 28' as a partial data set. The amplitude profile outside the angular range of the source diagram 28' can be represented, for example, by pairs of values ​​for discrete "measurement points" (or more appropriately, "support points"). Here, the number of support points may be greater than the number of measurement points 24' in the original source diagram 28'. In this case, the neural network 30' has a larger number of neurons in the output layer than in the first layer.

[0041] A further variation of the method is shown in FIG. The source diagram 28'' has measurement points 24'', which are evenly distributed across the entire azimuth angle range. However, the number of these measurement points is relatively small, and therefore the diagram has only a small angular resolution. In this case, the neural network 30'' is trained to interpolate intermediate values ​​between the measurement points 24'', thereby resulting in a composite diagram 32'' with higher angular resolution.

[0042] It should be understood that the methods described above in connection with Figures 7, 11 and 12 can be combined with each other as desired.

Claims

1. 1. A method for providing calibration data for calibrating antenna diagrams of devices configured to transmit and / or receive electromagnetic radiation and belonging to a common manufacturing model, comprising: a) selecting a set of devices from said manufacturing models to create a training file (36); b) for each selected device, measuring a target diagram (34) using a first number of measurement points (24, 26; 24'; 24'') and creating an associated source diagram (28; 28'; 28'') using a second number of measurement points (24; 24'; 24'') that is smaller than the first number, and storing the target and source diagrams (34, 28; 28'; 28'') in the training file (36); c) using said training file (36) to train a neural network (30; 30'; 30'') to determine said respective associated target diagrams (34) based on said source diagrams (28; 28'; 28''); d) measuring the source diagrams (28; 28'; 28'') of the remaining devices from the manufacturing model; e) using said neural network (30; 30'; 30'') to generate said associated target diagram (34); f) using said target diagram (34) as calibration data for the remaining devices; A method characterized by:

2. 2. The method of claim 1, wherein the neural network (30; 30'; 30'') is trained to generate a composite diagram (32; 32'; 32''), which together with the source diagram (28; 28'; 28'') forms the target diagram (34).

3. 3. The method of claim 1, wherein the source diagram (28') has a smaller value range than the target diagram (34), and the neural network (30') is trained to determine the target diagram (34) by extrapolating the values ​​from the source diagram (28').

4. 4. The method of claim 1, wherein the neural network is trained to interpolate between measurement points of the source diagram during the creation of the target diagram.

5. 5. The method of claim 1 for a device configured for two-dimensional angle measurement, wherein the source diagram (28) contains data for only one dimension, and the neural network (30) is trained to generate data for a second dimension from the data of the source diagram (28).

6. 6. A method for calibrating antenna diagrams of devices configured to transmit and / or receive electromagnetic radiation and belonging to a common production model, comprising a method for providing calibration data as claimed in any one of claims 1 to 5, characterized in that the target diagram (34) is used to calibrate the devices before start-up.

7. 7. A method for calibrating antenna diagrams of devices configured to transmit and / or receive electromagnetic radiation and belonging to a common production model, comprising a method for providing calibration data as claimed in any one of claims 1 to 6, characterized in that for each device to be calibrated, the measured source diagram (28; 28'; 28'') together with parameters of the neural network (30; 30'; 30'') are stored in the device and used to provide the calibration data and calibrate the antenna diagram during operation of the device.

8. A neural network trained to perform step e) of the method of claim 1.

9. 10. A radar sensor for a motor vehicle, comprising a storage unit for the source diagram (28) measured in step (d) of the method of claim 1, and an electronic circuit in which the neural network (30; 30'; 30'') of claim 9 is implemented.

Images