Radar signal waveform generator

Abstract

This invention provides a radar signal waveform generator that can efficiently generate radar signal waveforms corresponding to verification conditions, thereby significantly reducing the number of verifications and verification time in desktop development of radar equipment. [Solution] The radar signal waveform generation device 1 includes an image classifier 4 that receives a verification condition image as input, and a generation AI 5 that generates a radar signal waveform corresponding to the feature quantities of the verification condition image extracted by the image classifier 4. The generation AI 5 includes a generator 6 that generates a radar signal waveform from the feature quantities of the verification condition image, and a discriminator 7 that evaluates the output from the generator 6 by comparing it with a training radar signal waveform.

Description

Technical Field

[0001] The present invention relates to a radar signal waveform generation device capable of generating a radar signal waveform corresponding to verification conditions in the performance verification (in-aircraft development) of a radar device.

Background Art

[0002] In vehicles such as automobiles, in-vehicle radar devices have come to be widely used. As verification methods for the performance of such in-vehicle radar devices, there are verification by field experiments and verification by simulation. Among these, verification by field experiments is a method of acquiring data while actually running a vehicle equipped with an in-vehicle radar device in a verification environment, and it is possible to directly verify the actual in-vehicle radar device. However, in such verification by field experiments, it is actually difficult to align the verification conditions, and there is a problem that elucidating the factors causing performance degradation and reproducing malfunctions do not proceed as expected. That is, in field experiments, even when trying to reproduce specific verification conditions, the actual verification environment will be different each time, so there were problems with the robustness and reproducibility of the verification. Also, when a problem occurs in the verification, there is a problem that it is difficult to grasp what is the main factor of the problem only by experiments.

[0003] On the other hand, verification by simulation is in-aircraft verification based on simulation on a computer (radio wave propagation CAE and radar signal processing), and since verification conditions can be set as an image on a computer, there is no problem with the reproducibility of verification like in field experiments, and data in situations that are difficult to obtain in field experiments can also be obtained.

[0004] For this reason, verification by simulation has been widely used, and various related technologies have been proposed. For example, as a technology related to the construction of radio wave propagation models used in simulations (models that reproduce radio waves that hit a target and bounce back), Patent Document 1 (Japanese Patent Application Publication No. 2023-060988) proposes an invention that constructs a radio wave propagation model using machine learning, with spatial and object image data and the radio wave propagation characteristics within that space as training data. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2023-060988 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] However, conventional simulation-based verification had the following problems. Specifically, in radio wave propagation simulations that trace radar propagation paths using methods such as ray tracing, it is necessary to calculate the radar wave propagation path for each verification condition, which requires an enormous amount of computation time. For this reason, it is difficult to perform verification under various verification conditions in real time, and it was sometimes impossible to complete all the necessary verifications during the development period.

[0007] This invention was made in consideration of the above circumstances, and aims to provide a radar signal waveform generation device that can efficiently generate radar signal waveforms corresponding to verification conditions, and can significantly reduce the number of verifications and verification time in desktop development of radar equipment. [Means for solving the problem]

[0008] To achieve the above objective, the present invention adopts the following solution. That is, as described in claim 1, a radar signal waveform generation device that generates a radar signal waveform corresponding to a verification condition image based on a verification condition image representing verification conditions for a radar device comprises a classifier into which the verification condition image is input, and a generation AI into which feature quantities of the verification condition image extracted by the classifier are input, and which outputs the radar signal waveform corresponding to the feature quantities.

[0009] According to the above solution, the radar signal generation device (for example, radar signal generation device 1) is configured by combining a classifier (for example, image classifier 4) and a generation AI (for example, generation AI 5), so that radar signal waveforms that appropriately correspond to the verification conditions can be generated all at once.

[0010] In this case, the classifier extracts features (numerical vectors) from the large amount of information contained in the verification condition images, and the generating AI learns based on these features and generates radar signal waveforms. This significantly reduces the computation time required by the computer during the AI's learning process and the verification of the radar device. Furthermore, by using only a portion of the numerous verification condition images as training data, the radar signal waveform generator can generate the radar signal waveforms corresponding to the remaining verification condition images. Compared to generating radar signal waveforms using the ray tracing method for all verification condition images, this significantly reduces the number of verifications and the verification time required for radar device verification. Therefore, the performance of the radar device under arbitrary verification conditions can be efficiently verified, allowing for rapid and efficient development of the radar device on paper.

[0011] A preferred embodiment based on the above solution method is as described in claim 2 and subsequent claims of the patent. That is, the generating AI may be a conditional adversarial generative network in which the feature quantities are input as condition labels (corresponding to claim 2). In this case, the generating AI, which is a conditional adversarial generative network, can generate a radar signal waveform that appropriately corresponds to the verification condition image from the feature quantities of the verification condition image.

[0012] The generating AI includes a generator that generates radar signal waveforms, and the generator may be trained using a classifier that evaluates the output from the generator by comparing it with the training radar signal waveform so that the radar signal waveform generated by the generator matches the training radar signal waveform corresponding to the training verification condition image based on the features of the training verification condition image (corresponding to claim 3). In this case, the generator (e.g., generator 6) is appropriately trained by the classifier (e.g., classifier 7).

[0013] The radar signal waveform may also be an impulse response waveform (corresponding to claim 4). In this case, the impulse response waveform obtained from the radar signal generation device can be used to perform appropriate performance verification of the radar device.

[0014] The classifier may be a classifier that classifies the input validation condition image by the received power value (corresponding to claim 5). In this case, the classifier can appropriately extract features from the validation condition image.

[0015] The radar device may be an in-vehicle radar device (corresponding to claim 6). In this case, since the radar signal waveform generator can generate radar signal waveforms all at once from a large number of verification condition images generated as the vehicle equipped with the in-vehicle radar device moves, efficient verification can be performed. [Effects of the Invention]

[0016] According to the present invention, a radar signal waveform generation device is configured by combining two types of machine learning models (a classifier and a generative AI). This allows for the simultaneous generation of radar signal waveforms from verification condition images, significantly reducing the verification time (computer calculation time) and the number of verifications required for performance verification of radar equipment. Therefore, it becomes possible to efficiently perform desktop development of radar equipment in a short amount of time. [Brief explanation of the drawing]

[0017] [Figure 1] Block diagram showing a configuration example of a radar signal waveform generation device according to an embodiment of the present invention. [Figure 2] Diagram showing an example of a verification condition image in free space. [Figure 3] Diagram showing an example of a verification condition image in a two-wave environment. [Figure 4] Graph showing the comparison result of the radar signal waveform (power-distance characteristic) generated by the radar signal waveform generation device and the ray tracing method from the verification condition image in free space. [Figure 5] Graph showing the comparison result of the radar signal waveform (power-distance characteristic) generated by the radar signal waveform generation device and the ray tracing method from the verification condition image in a two-wave environment.

Embodiments for Carrying Out the Invention

[0018] Hereinafter, embodiments of the present invention will be described based on the accompanying drawings. FIG. 1 shows a block configuration diagram of a radar signal waveform generation device 1 according to an embodiment of the present invention. As shown in the figure, the radar signal waveform generation device 1 includes a verification condition image database 2, a radar signal database 3, an image classifier 4, and a generation AI 5.

[0019] The verification condition image database 2 is a database in which verification condition images representing verification conditions (for example, in-vehicle radar devices) to be verified (performance evaluation) of a radar device are stored. Here, the verification condition is a condition indicating a situation in which the performance of the radar device, such as the distance between the radar device and the target (object) to be detected and the height of the radar device, should be verified (evaluated), and is set for each situation to be verified.

[0020] The verification condition image database 2 stores a large number of verification condition images representing various verification conditions to be verified. For example, in the case of an in-vehicle radar device, as the vehicle moves, the positional relationship between the radar device and the target changes every moment, so the verification conditions also change every moment. Therefore, in the verification condition image database 2, there exist a large number of verification condition images corresponding to the change in the traveling position of the vehicle as the images to be verified.

[0021] The verification condition images are generated, for example, as graphics on a computer by CAD or the like. FIGS. 2 and 3 schematically show very simple examples of the verification condition images. FIGS. 2(A) to (C) show verification condition images 10a to 10c as an example of the verification condition images when only the radar device 11 and the target 12 are included in the image and the surrounding environment such as the road surface is not considered (in the case of free space).

[0022] In this example, a situation where the radar device 11 (the vehicle equipped with the radar device 11) approaches the target 12 from the state shown in the verification condition image 10a to the state shown in the verification condition image 10c is shown, and the distance L between the radar device 11 and the target 12 changes from 30 m in the verification condition image 10a, to 20 m in the verification condition image 10b, and to 10 m in the verification condition image 10c. In this example, the distance L between the radar device 11 and the target 12 is the main parameter defining the verification conditions. Although FIGS. 2(A) to (C) show an example where the change in the distance L is at 10 m intervals, actually, even in the case of such simple verification condition images, a large number of verification condition images in which the distance L changes at finer intervals are prepared and considered.

[0023] FIG. 3 shows a verification condition image 20 in the case where the road surface 23 is included in the verification condition image in addition to the radar device 21 and the target 22 (in the case of a two-wave environment where it is necessary to consider the radar wave reflected by the road surface).

[0024] In this example, in addition to the direct reflected waves 24 from the target 22, radar waves 25 reflected by the road surface 23 must also be considered, making the verification conditions more complex than in the case of free space. In this case, in addition to the distance L between the radar device 21 and the target 22, the height H of the radar device 21 from the road surface 23 is also a major parameter that defines the verification conditions.

[0025] While Figures 2 and 3 show very simple verification condition images as examples, in reality, numerous images are prepared that reproduce in detail complex verification conditions (including not only position but also size and shape) that are in line with the actual environment in which the radar equipment is used, and radar signal waveforms corresponding to each verification condition image are calculated.

[0026] Radar signal database 3 is a database that stores radar signal waveforms of radar equipment. The radar signal waveforms are the response waveforms (e.g., impulse response waveforms) of the radar equipment obtained when the verification conditions are set, and are stored in a one-to-one pairing with the corresponding verification condition images.

[0027] In the validation condition image database 2 and radar signal database 3, a portion of the paired validation condition images and radar signal waveforms stored become training data for the image classifier 4 and the generating AI 5. In this case, the training radar signal waveform is calculated from the training validation condition images by, for example, a radio wave propagation simulation using the ray tracing method (a simulation method that treats radio waves as light and calculates the main propagation path).

[0028] Image classifier 4 is a classifier that classifies input validation condition images based on the received power value of the radar device (the strength of the reflected radio waves). For example, it is an image classifier that utilizes a convolutional neural network (CNN) (e.g., ResNet). To enable appropriate classification, image classifier 4 is pre-trained using training validation condition images and training radar signal waveforms as training data.

[0029] Thus, while the image classifier 4 is a classifier that performs classification, in this invention, it is not the classification performed by the image classifier 4 itself that is utilized, but rather the features of the validation condition image extracted as intermediate values ​​during the classification process. That is, the image classifier 4 is used as a feature extraction model for the validation condition image, and the features output from the image classifier 4 are input to the subsequent generation AI 5. Here, the features of the validation condition image are, for example, feature vectors of the intermediate layers or fully connected layers of a CNN, and the numerical values ​​(vectors) extracted as features are input to the generation AI 5.

[0030] Generative AI 5 is a generative AI that generates (calculates) radar signal waveforms (e.g., impulse response waveforms) corresponding to validation condition images based on the features of the validation condition images input from the image classifier 4. Generative AI 5 is composed of, for example, a conditional generative adversarial network (CGAN), where the features of the validation condition images are given as condition labels and conditional generation is performed.

[0031] To explain in more detail, the Generator AI 5 comprises a generator 6 and a classifier 7. The generator 6 is trained via the classifier 7 based on a training validation condition image and a training radar signal waveform so that it generates a radar signal waveform that conforms to the input validation condition image from that validation condition image. That is, during the model construction of the radar signal waveform generator 1 (the training process of the Generator AI 5), the features of the training validation condition image are input from the image classifier 4 to the generator 6, the generator 6 generates an output based on the features of the training validation condition image, and this output is input to the classifier 7.

[0032] Classifier 7 is a classifier that evaluates the data generated by generator 6 by comparing it with training data. Specifically, classifier 7 compares the output from generator 6 with the training radar signal waveform corresponding to the verification condition image stored in radar signal waveform database 3, determines whether the output from generator 6 is true (whether it is true that the output from generator 6 appropriately reproduces the training radar signal, or false that it does not appropriately reproduce the training radar signal), and provides the determination result to generator 6. After receiving the truth / false determination from classifier 7, generator 6 continues to learn by changing its parameters so that it can output an appropriate radar signal waveform that is as close as possible to the training data.

[0033] Furthermore, during the learning process of the Generating AI 5, appropriate signal transformation processing (e.g., Fourier transform processing) may be performed as needed to reduce the size of the training data. For example, the training impulse response waveform from the radar signal waveform database 3 may be converted into a received signal waveform to be used as training data for the Generating AI 5, and the received signal waveform, which is the output of the Generating AI 5 (generator 6) after training, may be inversely transformed to obtain an impulse response waveform as the output of the radar signal waveform generator 1.

[0034] Once the generator 6 has completed its training (i.e., it can output an appropriate radar signal waveform that reproduces the training data), the discriminator 7 has finished its role and, when using the radar signal waveform generator 1 (when verifying the radar device), it is removed from the radar signal waveform generator 1 as needed.

[0035] When using the radar signal waveform generator 1 for verifying a radar system, an unverified verification condition image (a verification condition image for which the corresponding radar signal waveform is unknown) is input to the radar signal waveform generator 1, and a radar signal waveform (e.g., an impulse response waveform) corresponding to the verification condition image is output from the radar signal waveform generator 1 (generator 6). The output radar signal waveform is then processed as needed and used for verifying the radar system (desktop development).

[0036] Figures 4 and 5 are graphs showing a comparison of radar signal waveforms generated by the radar signal waveform generator 1 and those generated by the ray tracing method. These graphs show the power distance characteristics obtained by processing the generated impulse response waveforms. Figure 4 shows the comparison results in a verification condition image in free space (see Figure 2), and Figure 5 shows the comparison results in a verification condition image in a two-wave environment (see Figure 3). In each graph, the results from the radar signal waveform generator 1 are plotted as white circles, and the results from the ray tracing method are plotted as black circles.

[0037] As can be seen from Figures 4 and 5, in both free space and two-wave environments, the power distance characteristics obtained by the radar signal waveform generator 1 and the power distance characteristics calculated by the ray tracing method show good agreement (the trend of distance attenuation of received power is reproduced). Therefore, it is considered that the radar signal waveform generator 1 of the present invention can appropriately reproduce radar signal waveforms with accuracy comparable to those calculated by the ray tracing method.

[0038] As described above, the radar signal waveform generation device 1 of this embodiment is configured by combining an image classifier 4 that receives a verification condition image as input and outputs the feature quantities (numerical vectors) of the verification condition image, and a generation AI 5 (generator 6 and classifier 7) that receives the feature quantities of the verification condition image output from the image classifier 4 as input and outputs a radar signal waveform (e.g., an impulse response waveform) corresponding to the feature quantities of the verification condition image. Therefore, it is possible to generate radar signal waveforms that appropriately correspond to the verification conditions all at once.

[0039] In this case, the image classifier 4 automatically extracts numerical vectors, which are the features of the verification condition image, from the large amount of information contained in the verification condition image (by first digitizing the image data), and the generating AI 5 (generator 6) learns based on these features (numerical vectors that express the differences in the verification conditions) and generates radar signal waveforms. Therefore, the computation time by the computer can be greatly reduced during the learning process of the generating AI 5 and during the verification of the radar device after learning is complete. In addition, in this case, the user of the radar signal waveform generator 1 only needs to input the verification condition image to the radar signal waveform generator 1, and does not need to consider the input to the generating AI 5 (generator 6).

[0040] Furthermore, if there are a large number of verification condition images to be verified, the radar signal waveforms can be prepared for only a portion of them (for example, by calculating them using the ray tracing method) and used as training data for the generating AI 5. For the remaining verification condition images (verification condition images for which the corresponding radar signal waveforms are unknown), the radar signal waveform generation device 1 can generate the radar signal waveforms all at once. For example, by training with verification condition images 10a and 10b in Figure 2 as training data, the radar signal waveform corresponding to verification condition image 10c can be generated.

[0041] Therefore, compared to generating radar signal waveforms using the ray tracing method for all verification condition images, the computer calculation time is significantly reduced, and the number of verifications and verification time in radar device performance verification can be greatly reduced. Thus, the performance (in-vehicle performance) of the radar device can be efficiently verified under any verification condition, and the desktop development of the radar device can be advanced efficiently in a short amount of time.

[0042] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above, and appropriate modifications can be made within the scope of the claims. [Industrial applicability]

[0043] This invention can be used in the desktop development of in-vehicle radar systems. [Explanation of Symbols]

[0044] 1. Radar signal waveform generator 2. Verification Conditions Image Database 3. Radar signal waveform database 4 Image classifier 5 Generation AI 6 generator 7. Classifier 10a~10c Verification Condition Images 11 Radar equipment 12 Targets 20 Verification Condition Images 21 Radar equipment 22 Targets 23 Road surface 24. Direct reflected waves from the target 25. Radar waves reflected from the road surface L: Distance between radar device and target H Radar device height from the road surface

Claims

1. In a radar signal waveform generation device that generates a radar signal waveform corresponding to a verification condition image based on a verification condition image that expresses the verification conditions for a radar device, A classifier into which the aforementioned verification condition image is input, The feature quantities of the verification condition image extracted by the classifier are input to a generating AI that outputs the radar signal waveform corresponding to the feature quantities. A radar signal waveform generator equipped with [a specific feature / feature].

2. In the radar signal waveform generation device according to claim 1, The aforementioned generating AI is a radar signal waveform generator which is a conditional generative adversarial network in which the aforementioned feature quantities are input as conditional labels.

3. The aforementioned generating AI includes a generator that generates radar signal waveforms, The generator is a radar signal waveform generator that is trained using a discriminator that compares the output from the generator with the training radar signal waveform so that the radar signal waveform generated by the generator matches the training radar signal waveform corresponding to the training verification condition image based on the features of the training verification condition image.

4. In the radar signal waveform generation device according to claim 1, The radar signal waveform generator is an impulse response waveform.

5. In the radar signal waveform generation device according to claim 1, The classifier is a radar signal waveform generator that classifies the input verification condition image into classes based on the received power value.

6. In the radar signal waveform generation device according to claim 1, The radar device is a radar signal waveform generator, which is an in-vehicle radar device.

Images