Method and computing device for interpolating a SAR value for an antenna position of an antenna on a motor vehicle to be taken into account by a machine learning model

Abstract

The invention relates to a method and a computing device for interpolating a SAR value (4) for an antenna position of an antenna (7) on a motor vehicle (10) to be considered by a machine learning model. First, a predetermined antenna design with a predetermined antenna position of one of the antennas (7) on the motor vehicle (10) is determined, thereby generating antenna position data. Subsequently, near-field data is calculated from the far-field data and phase information. For at least one predetermined body position and distance of a simulation model (20), at least one SAR value (4) is determined from the near-field data, resulting in simulation data. This simulation data is combined with the far-field data and phase information, the antenna position data, and the near-field data to generate interpolation data.The interpolation data is then fed into the machine learning model and trained until a predefined training criterion is met. Once this criterion is met, newly determined far-field data and phase information for a previously unconsidered antenna position are fed in to interpolate SAR values ​​(4).

Description

[0001] The invention relates to a method for interpolating a SAR value for an antenna position of an antenna on a motor vehicle to be taken into account by a machine learning model, and to a computing device.

[0002] An antenna can form the interface between a radio system and its environment. However, investigating an antenna is challenging because modifications to the antenna design, its environment, and / or the matching network can affect its antenna characteristics, particularly its radiation pattern. Changes in the antenna's position or its environment necessitate re-simulation or re-measurement, which is particularly time-consuming.

[0003] WO 2023 / 187 447 A1 discloses a method for determining the correct placement of an antenna by predicting a radiation pattern.

[0004] US 2024 / 0147251A1 discloses a machine learning-based wireless planning method using antenna radiation patterns.

[0005] US patent 2024 / 0204816A1 discloses a method and a device for controlling a terminal antenna.

[0006] DE 10 2023 131 698 A1 discloses a method and a device for analyzing the radiation characteristics of a motor vehicle.

[0007] The current state of the art may only offer limited possibilities for a reliable replacement of costly measurements and / or simulations.

[0008] The present invention is based on the objective of determining a SAR value in a time-saving manner.

[0009] The problem is solved by the subject matter of the independent patent claims. Advantageous further developments of the invention are described by the dependent patent claims, the following description, and the figure(s).

[0010] Method for interpolating or estimating a SAR (specific absorption rate) value for an antenna position to be considered by a machine learning model (or a new or previously unknown position) of one or at least one antenna on a motor vehicle, comprising the steps: - Determining far-field data and phase information of the antenna, wherein the antenna comprises a predetermined antenna design and a predetermined or predefined antenna position on the motor vehicle, thereby generating antenna position data, - Calculating near-field data from the determined far-field data and phase information, - Determine at least one SAR value from the near-field data for at least one given body position and at least one given distance of a simulation model relative to the antenna, thereby generating simulation data, - Linking the simulation data with the far-field data and phase information, the antenna position data, the near-field data, thereby generating interpolation data, - Feeding the interpolation data into the machine learning model, and training the machine learning model until a predefined training criterion is met, - as soon as the specified training criterion is reached: feeding newly determined far-field data and phase information into the machine learning model of the previously unconsidered antenna position, whereby at least one SAR value is interpolated or determined for at least one specified body position and at least one specified distance of the simulation model relative to the antenna.

[0011] Determining the far-field data and phase information, as well as calculating at least one SAR value from the near-field data, can be achieved using state-of-the-art methods. The far-field data and phase information can be determined, for example, using near-to-far-field transformation (NF-FF transformation). Determining the SAR value from the near-field data can be achieved, for example, using the finite-difference time-domain (FDTD) method and / or the finite element method (FEM).

[0012] In other words, for various antenna placements on a vehicle (e.g., roof center and / or side mirror and / or bumper), the associated far- and near-field data, as well as the calculated SAR values, can be stored together in a database. This database can then form the basis for training the machine learning model.

[0013] The specified distance can be, for example, 5 millimeters to 1 or up to 3 meters.

[0014] The simulation model includes a virtual model of a human body, which is used in an (electromagnetic) simulation to calculate the SAR value.

[0015] The at least one predefined body position can, for example, include the simulation model sitting or positioned in the driver's seat and / or the passenger seat and / or on the rear seat. Furthermore, the at least one predefined body position can include standing and / or kneeling at and / or around the vehicle. The at least one predefined body position can include the simulation model leaning against the vehicle and / or supporting itself with one hand on the vehicle roof and / or door and / or standing next to the vehicle, particularly in front of the hood. The at least one predefined body position can enable realistic simulations of electromagnetic exposure in different usage situations in and / or on the vehicle.

[0016] The machine learning model can, for example, include an artificial neural network and / or a random forest. In particular, the machine learning model can be used to predict or interpolate one or more SAR values ​​for new combinations of body position and / or distance and / or antenna position.

[0017] The specified training criterion can include, for example, a specified prediction deviation, e.g., ±0.01 to ±0.05 W / kg (watts per kilogram), and / or a specified number of training cycles, e.g., 1000 to 50000 epochs.

[0018] Interpolation is used to estimate a SAR value for at least one given body position, distance, and / or antenna position, particularly without having to perform a simulation. Existing simulation data can be used to derive or interpolate a SAR value for an unmeasured or unsimulated antenna position.

[0019] The invention offers the advantage of reducing the number of physical measurements and / or simulations required. For example, this allows the vehicle's design to be modified or adapted so that, for instance, the SAR values ​​in the interior are lower or reduced.

[0020] The invention also includes further developments that result in additional advantages.

[0021] A further training program stipulates that the specified antenna design includes a predefined shape and / or frequency and / or band capability. The predefined shape could, for example, include a planar antenna, a whip antenna, a helical antenna, a microstrip antenna, and / or an internal antenna. The predefined frequency and / or band capability could include a narrowband antenna, a broadband antenna, and / or a multiband antenna. Consequently, it is advantageous that the machine learning model takes the predefined shape and / or frequency and / or band capability into account, allowing the (respective) SAR value to be interpolated (accurately).

[0022] Further training stipulates that at least one vehicle parameter of the motor vehicle is taken into account, whereby the at least one predefined vehicle parameter includes at least a predefined body shape and / or vehicle size and / or arrangement and / or design or structure of a vehicle part on which the antenna is positioned. In particular, at least one vehicle parameter that (potentially) influences the SAR value can be taken into account. The predefined body shape can, for example, include or consider the body style of an SUV (Sport Utility Vehicle) and / or a sedan and / or a coupé convertible. For example, the average dimensions of predefined vehicle models or only individual predefined representative vehicle models of a respective body style can be considered.

[0023] The specified vehicle size can include or consider vehicle classes such as microcars, compact cars, luxury cars, trucks, and vans. Here, the average dimensions or only specific, representative vehicle models from the respective vehicle classes can be considered.

[0024] Furthermore, predefined arrangements and / or designs or structures of predefined vehicle parts can be taken into account. A vehicle part can refer, for example, to the vehicle roof and / or the front bumper and / or the rear bumper and / or the hood. Different predefined designs or structures can be considered in this process.

[0025] This allows the respective SAR values ​​to be (accurately) interpolated, taking into account the arrangement and / or construction of the vehicle parts.

[0026] Further training stipulates that the simulation model includes a human model, which takes into account human geometries and tissue characteristics. This allows for a (realistic) simulation of how the human body absorbs radiation, enabling the provision of a SAR value.

[0027] A training course stipulates that at least one predetermined body position for the human model must be simulated inside or on the vehicle. Different body positions inside or on the vehicle affect radiation exposure and the resulting SAR value differently. This allows the identification of the body positions in which the SAR value(s) are highest or lowest.

[0028] Further training stipulates that the specified antenna position includes the antenna being positioned on the vehicle's roof, on the front bumper, on the rear bumper, inside the vehicle, and / or not attached to the vehicle but freestanding (independent and / or not connected to a vehicle). "Freestanding" can mean that the antenna is not attached to the vehicle but positioned independently. Using freestanding positioning, the effect of the vehicle or the specific position of the antenna on the vehicle can be determined. The freestanding position can serve as a reference or starting point because it provides an unaffected baseline against which the effects of the vehicle on the SAR values ​​can be compared.Furthermore, this allows for the simulation of multiple installation situations for at least one antenna, enabling the determination of different SAR values ​​for each and their assignment to the respective positions, antenna positions, or installation situations.

[0029] Further training stipulates that the machine learning model uses a nonlinear mapping, learned through training, between input parameters comprising far-field data, phase information, near-field data, antenna position data, simulation data, and at least one vehicle parameter. This nonlinear mapping is implemented using at least one artificial neural network, which includes an activation function to provide at least one interpolated SAR value as an output parameter for the antenna position to be considered. The activation function can, for example, include ReLU (Rectified Linear Unit) and / or Softmax. In other words, after training, the machine learning model is capable of predicting or interpolating a SAR value for an antenna position that has not yet been tested or considered.

[0030] Further training involves calculating near-field data using multipole expansion and / or the inverse Fourier transform and / or Green's function. Multipole expansion offers the advantage of reducing computational complexity by decomposing the determined near- and far-field data into (compact) terms. The inverse Fourier transform can be helpful for periodic and / or spectral signal processing. Green's function can be used to calculate and / or determine (complex) boundary conditions and / or irregular geometries.

[0031] Further training stipulates that the machine learning model must include at least a feedforward neural network (FNN) and / or a gradient boosting model. The gradient boosting model, in particular, is suitable for identifying structural dependencies between the input data and / or can support and / or improve the interpretability of the feedforward neural network. Ultimately, this allows for the interpolation of a SAR value for a new antenna position and / or distance.

[0032] For use cases or application situations that may arise during the procedure and are not explicitly described here, it may be provided that, according to the procedure, an error message and / or a request for user feedback is issued and / or a default setting and / or a predetermined initial state is set.

[0033] The invention also includes the computing device. The computing device can comprise a data processing device or a processor circuit configured to perform an embodiment of the method according to the invention. For this purpose, the processor circuit can comprise at least one microprocessor and / or at least one microcontroller and / or at least one FPGA (Field Programmable Gate Array) and / or at least one DSP (Digital Signal Processor). In particular, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or an NPU (Neural Processing Unit) can be used as the microprocessor. Furthermore, the processor circuit can comprise program code configured to perform the embodiment of the method according to the invention when executed by the processor circuit. The program code can be stored in a data memory of the processor circuit.The processor setup can be based on at least one circuit board and / or at least one SoC (System on Chip).

[0034] As a further solution, the invention also includes a computer-readable storage medium comprising program code which, when executed by a computer or a computer network, causes it to execute an embodiment of the method according to the invention. The storage medium can be provided at least partially as a non-volatile data storage medium (e.g., as flash memory and / or as an SSD - solid state drive) and / or at least partially as a volatile data storage medium (e.g., as RAM - random access memory). The storage medium can be located within the computer or computer network. However, the storage medium can also be operated, for example, as an app store server and / or cloud server on the internet. The computer or computer network can provide a processor circuit with, for example, at least one microprocessor.The program code can be provided as binary code, assembly code, source code in a programming language (e.g., C), or a program script (e.g., Python). Alternatively, the computer-readable storage medium can be implemented as a signal containing computer-readable data, such as a time-varying voltage signal or a radio signal.

[0035] The invention also includes combinations of the features of the described embodiments. The invention therefore also includes realizations that each exhibit a combination of the features of several of the described embodiments, provided that the embodiments have not been described as mutually exclusive.

[0036] The following are exemplary embodiments of the invention described. This is illustrated by: Fig. 1 Two diagrams illustrating a difference in SAR values ​​with only minimal deviation in the position of a human body; Fig. 2. A schematic representation of an antenna in different environments for determining SAR values ​​and Fig. 3 a flowchart according to one embodiment for determining SAR values.

[0037] In the figures, identical reference symbols denote functionally equivalent elements.

[0038] Fig. Figure 1 shows a (significant) difference in the SAR values ​​4 with only minimal deviation in the position or body position of a human body, to illustrate the relevance of the method. For example, the abscissa of the figure is given in the unit of frequency, expressed in gigahertz (GHz). It describes the range from 0.5 GHz to 5 GHz. The ordinate is given in the unit of specific absorption rate (SAR), expressed in watts per kilogram (W / kg). The value range extends from 0 to over 3.5 W / kg. A human model 20 is shown, with SAR values ​​4 determined for various predefined body positions. The x-axis shows the SAR limit for the head 2 and the SAR limit for the limbs 6. SAR values ​​4 can be determined according to the body positions and / or movements of the human model 20. Measurements or simulations of the in Fig. The values ​​shown in point 1 can be very time-consuming. The idea is that AI-supported analysis or a machine learning model can be used to measure only one value (e.g., near-field data and / or far-field data and / or phase information), and the AI ​​or machine learning model can then derive or interpolate the remaining or related values ​​or data (near-field data and / or far-field data and / or phase information), in particular a SAR value 4.

[0039] According to one embodiment, a SAR value 4 (specific absorption rate) for an antenna position 7 on a motor vehicle 10, to be taken into account by a machine learning model, can be interpolated. The following steps can be performed: - Determining far-field data and phase information of the antenna 7, wherein the antenna 7 comprises a predefined antenna design and a predefined antenna position on the motor vehicle 10, thereby generating antenna position data, - Calculating near-field data from the determined far-field data and phase information, - Determine at least one SAR value 4 from the near-field data for at least one given body position and at least one given distance of a simulation model relative to the antenna 7, thereby generating simulation data, - Linking the simulation data with the far-field data and phase information, the antenna position data, the near-field data, thereby generating interpolation data, - Feeding the interpolation data into the machine learning model, and training the machine learning model until a predefined training criterion is met, - as soon as the specified training criterion is reached: feeding newly determined far-field data and phase information into the machine learning model of the previously unconsidered antenna position of antenna 7, whereby at least one SAR value 4 is interpolated for at least one specified body position and at least one specified distance of the simulation model relative to antenna 7.

[0040] Fig. Figure 2 shows a schematic representation of an antenna 7 in different environments for determining respective SAR values ​​4. A SAR value 4 is determined in each case, with the antenna 7 positioned differently. In Figure 1, the antenna 7 is positioned independently of the vehicle 10, specifically to determine a SAR value 4 as a reference value. Subsequently, the antenna 7 is positioned differently on the vehicle 10, with SAR values ​​4 being determined in each case (see Figures 2 to 4). In Figure 5, for example, SAR values ​​4 can be determined or interpolated within a simulation using the machine learning model, e.g., for a new vehicle bumper 12, in particular another vehicle part, especially for a new vehicle 10.

[0041] The idea involves simulating or measuring antenna 7 with a predefined small portion of its environment and using AI or a machine learning model trained on at least similar antennas 7 to determine the influence of the environment. The similarity could relate, for example, to the antenna's design, position, frequency and / or band capability, and / or radiation pattern. If the AI ​​is trained with a sufficient or predefined amount of data, it can determine how the radiation, particularly the SAR value 4, changes with distance. It can also determine how the radiation changes with different and / or predefined objects near antenna 7. The AI ​​database or the machine learning model's database can be populated with antenna measurements from multiple locations and / or with different objects near or at a predefined distance from antenna 7.Additional points or data in the database can be populated, for example, through simulations. Since simulations are not 100% reliable, they can be used to fill in points that cannot be easily determined through measurements. The AI ​​can analyze the data and / or make assumptions for different antennas (7) placed in different and / or predetermined positions and / or with different car parts positioned around them, which is significantly faster than measuring or even simulating each scenario individually (see [reference]). Fig. 2) The specific absorption ratio or specific absorption rate (SAR) is a parameter usually evaluated for handheld devices, but recently SAR evaluation has also become mandatory in some countries for vehicle antennas or antennas for motor vehicles. If the antenna measurement database contains data on the signal phase (phase information), the near field can be reconstructed from the far field. The SAR is usually measured at a distance of less than 20 cm between the antenna and a person or human body. This is the near field. The near field (near electromagnetic field) is a vector whose amplitude decreases with and depending on the distance from the antenna. The amplitude cannot simply be predicted but must be calculated for each point.We can use AI to efficiently process and / or utilize (large) predefined amounts of near-field data, which is particularly necessary for this transformation. The SAR for the human body located at a specific or predetermined distance from antenna 7 can now be determined efficiently and / or quickly.

[0042] Fig. Figure 3 shows a flowchart according to an embodiment for determining SAR values ​​4. To determine or interpolate the SAR values ​​4 or a SAR value 4 using artificial intelligence (AI) or the machine learning model, according to Fig.The following three steps are performed. In step S1, the antenna 7 to be investigated can be defined (exactly), including its position or antenna position and / or the environment in which it is operated. Subsequently, in step S2, the AI ​​can check whether a similar or identical system already exists in its existing dataset. "Similar or identical system" means whether data already exists for a comparable antenna 7 that was tested under similar or identical conditions (e.g., on a similar or identical vehicle and / or in a similar or identical environment). In other words, a comparison of the parameters (e.g., distance and / or antenna design and / or antenna position and / or body position) can take place in order to interpolate, for example, a SAR value 4 for at least partially similar parameters. The at least partial similarity can be defined, for example, by means of a predefined Euclidean distance value.If insufficient data is available, measurements and / or simulations can be performed in step S3 to supplement missing information or data. In step S4, AI-assisted analysis can be carried out, in which a near-far-field transformation (NF2FF) can be performed if necessary, and any prediction error can be estimated. In other words, the machine learning model can be activated and / or started. This can involve performing an NF2FF and / or determining an error or deviation in the interpolation. The machine learning model can be trained until a predefined training criterion is met. Finally, in step S5, the result can be output, which includes the interpolated SAR values ​​for the investigated antenna position, particularly the one to be considered.

[0043] Overall, the examples show how AI-based antenna and SAR analysis can be provided and / or implemented. Reference symbol list 2 SAR limit for the limbs 4 SAR value 6 SAR limit for the head 7 Antenna 10 motor vehicle 12 new vehicle bumpers 20 Human Model

Claims

[1] Method for interpolating a specific absorption rate, hereinafter referred to as SAR value (4), for an antenna position of an antenna (7) on a motor vehicle (10) to be taken into account by a machine learning model, comprising the steps: - Determining far-field data and phase information of the antenna (7), wherein the antenna (7) comprises a predefined antenna design and a predefined antenna position on the motor vehicle (10), thereby generating antenna position data, - Calculating near-field data from the determined far-field data and phase information, - Determine at least one SAR value (4) from the near-field data for at least one given body position and at least one given distance of a simulation model (20) relative to the antenna (7), thereby generating simulation data, - Linking the simulation data with the far-field data and phase information, the antenna position data, the near-field data, thereby generating interpolation data, - Feeding the interpolation data into the machine learning model, and training the machine learning model until a predefined training criterion is met, and - as soon as the specified training criterion is reached: feeding newly determined far-field data and phase information into the machine learning model of the previously unconsidered antenna position of the antenna (7), whereby at least one SAR value (4) is interpolated for at least one specified body position and at least one specified distance of the simulation model (20) relative to the antenna (7). [2] Method according to claim 1, wherein the specified antenna design comprises a specified shape and / or frequency and / or band capability of the antenna (7). [3] Method according to one of the preceding claims, wherein at least one vehicle parameter of the motor vehicle (10) is taken into account, wherein the at least one predetermined vehicle parameter comprises at least one predetermined body shape and / or vehicle size and / or arrangement of a vehicle part on which the antenna (7) is positioned. [4] Method according to any of the preceding claims, wherein the simulation model comprises a Human Model (20) wherein the Human Model (20) takes into account human geometries and tissue characteristics. [5] Method according to claim 4, wherein the at least one predetermined body position for the human model (20) is located in or on the motor vehicle (10) in a simulation. [6] Method according to any of the preceding claims, wherein the specified antenna position comprises that the antenna (7) - on the roof of the motor vehicle (10) and / or - on the vehicle bumper at the front of the motor vehicle (10) and / or - on the vehicle bumper at the rear of the motor vehicle (10) and / or - inside the motor vehicle (10) and / or - is not attached to the motor vehicle (10), but is positioned freestanding. [7] Method according to one of the preceding claims, wherein the machine learning model uses a nonlinear mapping learned through training between input parameters comprising the far-field data, phase information, near-field data, antenna position data, simulation data and the at least one vehicle parameter, wherein the nonlinear mapping is realized by the use of at least one artificial neural network comprising an activation function to provide the at least one interpolated SAR value (4) as an output parameter for the antenna position of the antenna (7) to be taken into account. [8] Method according to one of the preceding claims, wherein the calculation of the near-field data is realized by means of a multipole expansion and / or inverse Fourier transform and / or a Green's function. [9] Method according to any of the preceding claims, wherein the machine learning model comprises at least a feedforward neural network (FNN) and / or gradient boosting model. [10] Computing device, wherein the computing device comprises a processor circuit which has program instructions which, when executed by the processor circuit, cause it to carry out a method according to one of the preceding method claims.

Images