EP-4738179-A1 - ANTENNA DESIGN METHOD AND APPARATUS

EP4738179A1EP 4738179 A1EP4738179 A1EP 4738179A1EP-4738179-A1

Abstract

This application pertains to the field of artificial intelligence AI technologies, and discloses an antenna design method and apparatus. The method includes: determining a design parameter of an antenna through a plurality of iterations, where an i th iteration process is as follows: obtaining a plurality of vectors from optimization space, where M elements in each vector respectively represent M design parameters of the antenna; and inputting input distributions of the M design parameters and the plurality of vectors to a surrogate model of Bayesian optimization BO, to obtain a plurality of groups of predicted values of a performance parameter of the antenna, where a processing process of the surrogate model includes calculating a covariance matrix between a plurality of random vectors, and a covariance between any two vectors is obtained by performing calculation on sample sets obtained by sampling input distributions of the vectors. During calculation, an acceleration algorithm based on Nystrom approximation is used to reduce complexity, to quickly determine the design parameter of the antenna. The foregoing method is applicable to a design parameter input of any distribution type, and can greatly improve inference efficiency of a design parameter optimization process.

Inventors

YANG, LIN
LYU, Junlong
LV, WENLONG
CHEN, Zhitang

Assignees

Huawei Technologies Co., Ltd.

Dates

Publication Date: 20260506
Application Date: 20240403

Claims (19)

An antenna design method, wherein the method comprises: determining a design parameter of the antenna through a plurality of iterations, wherein an i th iteration process is as follows: obtaining a plurality of vectors from optimization space, wherein each of the plurality of vectors comprises M elements, the M elements respectively represent M design parameters of the antenna, the optimization space is used to describe a value range of each of the M design parameters, and M and i are positive integers; obtaining input distributions of the M design parameters, and inputting the input distributions of the M design parameters and the plurality of vectors to a surrogate model of Bayesian optimization BO, to obtain a plurality of groups of predicted values of a performance parameter of the antenna and a plurality of degrees of certainty respectively corresponding to the plurality of groups of predicted values, wherein the plurality of groups of predicted values respectively correspond to the plurality of vectors, the plurality of degrees of certainty are respectively used to describe confidences of the plurality of groups of predicted values, a processing process of the surrogate model comprises calculating a covariance between any two of the plurality of vectors, and the covariance between any two vectors is obtained by performing calculation on a sample set obtained by sampling the input distributions; processing the plurality of groups of predicted values and the plurality of degrees of certainty based on an acquisition function of the BO, to determine a first vector from the plurality of vectors, wherein an output of the acquisition function that corresponds to the first vector is the largest; and inputting the first vector to an objective function to obtain a group of performance parameters of the antenna; and when the group of performance parameters of the antenna meets a preset condition or i is equal to a preset quantity of times, determining M design parameters comprised in the first vector as design parameters of the antenna.
The method according to claim 1, wherein the preset condition comprises a group of reference performance parameters of the antenna, and when a difference between the group of performance parameters of the antenna and the group of reference performance parameters is less than a preset threshold, the group of performance parameters of the antenna meets the preset condition.
The method according to claim 1 or 2, wherein the method further comprises: when the group of performance parameters of the antenna does not meet the preset condition and i is not equal to the preset quantity of times, adding the first vector and the group of performance parameters of the antenna, as one piece of training data, to a dataset used for training the surrogate model, and starting a next iteration.
The method according to any one of claims 1 to 3, wherein the plurality of vectors comprise a second vector and a third vector, and a covariance between the second vector and the third vector is obtained by inputting a first sample set and a second sample set to a maximum mean discrepancy MMD kernel function for processing, wherein m vectors comprised in the first sample set are obtained through sampling based on the second vector and the input distribution, m vectors comprised in the second sample set are obtained through sampling based on the third vector and the input distribution, each vector in the first sample set and the second sample set comprises M elements, and the M elements respectively represent the M design parameters.
The method according to claim 4, wherein a process of processing the first sample set and the second sample set by the MMD kernel function comprises: adding a first matrix to a second matrix, and subtracting a third matrix multiplied by 2, wherein the first matrix is obtained by multiplying a fourth matrix, a transposed matrix of the fourth matrix, and a pseudo-inverse matrix of the fourth matrix, the fourth matrix is obtained by processing h vectors in the first sample set by using the MMD kernel function, and h is a positive integer less than m; the second matrix is obtained by multiplying a fifth matrix, a transposed matrix of the fifth matrix, and a pseudo-inverse matrix of the fifth matrix, and the fifth matrix is obtained by processing h vectors in the second sample set by using the MMD kernel function; and the third matrix is obtained by multiplying a sixth matrix, a transposed matrix of the sixth matrix, and a pseudo-inverse matrix of the sixth matrix, and the sixth matrix is obtained by processing h vectors in the first sample set and h vectors in the second sample set by using the MMD kernel function.
The method according to any one of claims 1 to 5, wherein the method further comprises: before obtaining the plurality of vectors from the optimization space, training the surrogate model by using the dataset of the surrogate model.
The method according to any one of claims 1 to 6, wherein the input distribution is obtained by collecting statistics on production data, and an input distribution of each design parameter is represented by a probability distribution or a Gaussian distribution.
The method according to any one of claims 1 to 7, wherein the M design parameters comprise one or more of the following parameters of the antenna: a thickness, a dielectric coefficient of a material, and a shape parameter; and the performance parameter comprises one or more of the following parameters of the antenna: a scattering parameter, a radiation pattern, and a sidelobe width.
An antenna design apparatus, wherein the apparatus determines a design parameter of the antenna through a plurality of iterations, and the apparatus comprises: an obtaining unit, configured to: during an i th iteration process, obtain a plurality of vectors from optimization space, wherein each of the plurality of vectors comprises M elements, the M elements respectively represent M design parameters of the antenna, the optimization space is used to describe a value range of each of the M design parameters, and M and i are positive integers; and obtain input distributions of the M design parameters; a processing unit, configured to: obtain input distributions of the M design parameters, and input the input distributions of the M design parameters and the plurality of vectors to a surrogate model of Bayesian optimization BO, to obtain a plurality of groups of predicted values of a performance parameter of the antenna and a plurality of degrees of certainty respectively corresponding to the plurality of groups of predicted values, wherein the plurality of groups of predicted values respectively correspond to the plurality of vectors, the plurality of degrees of certainty are respectively used to describe confidences of the plurality of groups of predicted values, a processing process of the surrogate model comprises calculating a covariance between any two of the plurality of vectors, and the covariance between any two vectors is obtained by performing calculation on a sample set obtained by sampling the input distributions; process the plurality of groups of predicted values and the plurality of degrees of certainty based on an acquisition function of the BO, to determine a first vector from the plurality of vectors, wherein an output of the acquisition function that corresponds to the first vector is the largest; and input the first vector to an objective function to obtain a group of performance parameters of the antenna; and a determining unit, configured to: when the group of performance parameters of the antenna meets a preset condition or i is equal to a preset quantity of times, determine M design parameters comprised in the first vector as design parameters of the antenna.
The apparatus according to claim 9, wherein the preset condition comprises a group of reference performance parameters of the antenna, and when a difference between the group of performance parameters of the antenna and the group of reference performance parameters is less than a preset threshold, the group of performance parameters of the antenna meets the preset condition.
The apparatus according to claim 9 or 10, wherein the processing unit is further configured to: when the group of performance parameters of the antenna does not meet the preset condition and i is not equal to the preset quantity of times, add the first vector and the group of performance parameters of the antenna, as one piece of training data, to a dataset used for training the surrogate model, and start a next iteration.
The apparatus according to any one of claims 9 to 11, wherein the plurality of vectors comprise a second vector and a third vector, and a covariance between the second vector and the third vector is obtained by inputting a first sample set and a second sample set to a maximum mean discrepancy MMD kernel function for processing, wherein m vectors comprised in the first sample set are obtained through sampling based on the second vector and the input distribution, m vectors comprised in the second sample set are obtained through sampling based on the third vector and the input distribution, each vector in the first sample set and the second sample set comprises M elements, and the M elements respectively represent the M design parameters.
The apparatus according to claim 12, wherein a process of processing the first sample set and the second sample set by the MMD kernel function comprises: adding a first matrix to a second matrix, and subtracting a third matrix multiplied by 2, wherein the first matrix is obtained by multiplying a fourth matrix, a transposed matrix of the fourth matrix, and a pseudo-inverse matrix of the fourth matrix, the fourth matrix is obtained by processing h vectors in the first sample set by using the MMD kernel function, and h is a positive integer less than m; the second matrix is obtained by multiplying a fifth matrix, a transposed matrix of the fifth matrix, and a pseudo-inverse matrix of the fifth matrix, and the fifth matrix is obtained by processing h vectors in the second sample set by using the MMD kernel function; and the third matrix is obtained by multiplying a sixth matrix, a transposed matrix of the sixth matrix, and a pseudo-inverse matrix of the sixth matrix, and the sixth matrix is obtained by processing h vectors in the first sample set and h vectors in the second sample set by using the MMD kernel function.
The apparatus according to any one of claims 9 to 13, wherein the processing unit is further configured to: before obtaining the plurality of vectors from the optimization space, train the surrogate model by using the dataset of the surrogate model.
The apparatus according to claim 14, wherein the input distribution is obtained by collecting statistics on production data, and an input distribution of each design parameter is represented by a probability distribution or a Gaussian distribution.
The apparatus according to any one of claims 9 to 15, wherein the M design parameters comprise one or more of the following parameters of the antenna: a thickness, a dielectric coefficient of a material, and a shape parameter; and the performance parameter comprises one or more of the following parameters of the antenna: a scattering parameter, a radiation pattern, and a sidelobe width.
A computer device, wherein the computer device comprises at least one processor, a memory, and an interface circuit, the memory, the interface circuit, and the at least one processor are interconnected through a line, the at least one memory stores instructions, and when the instructions are executed by the processor, the method according to any one of claims 1 to 8 is implemented.
A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 8 is implemented.
A computer program product, wherein the computer program product comprises instructions, and when the instructions are executed, the method according to any one of claims 1 to 8 is implemented.

Description

This application claims priority to Chinese Patent Application No. 202310940699.9, filed with the China National Intellectual Property Administration on July 28, 2023 and entitled "ANTENNA DESIGN METHOD AND APPARATUS", which is incorporated herein by reference in its entirety. TECHNICAL FIELD This application relates to the field of artificial intelligence (Artificial Intelligence, AI) technologies in big data, and in particular, to an antenna design method and apparatus. BACKGROUND Bayesian optimization (Bayesian Optimization, BO) is a sample-efficient global optimization algorithm, and is usually applied to various verification-expensive black-box optimization issues. Each iteration of the algorithm mainly includes two important steps: (1) A probability-based surrogate model (Surrogate Model) is constructed based on existing observation data to describe a relationship between an input x and an output y of a to-be-optimized objective function. (2) An acquisition function (Acquisition Function) is designed to use a modeling result of the surrogate model to provide guidance for optimization decision-making and select a next query point (Query Point). Iteration is repeatedly performed until an optimal value is found or a maximum quantity of iterations is reached. Currently, the BO has been successfully applied to many optimization tasks, such as antenna design, neural network structure search, algorithm hyperparameter optimization, wireless network performance optimization, and robot control. During optimization for some complex issues, randomness inevitably occurs in an optimization process to some extent, for example, a manufacturing tolerance and a process limitation during production of a product like an antenna, or environment fluctuation and an execution error in a control optimization process. Consequently, random noise is introduced into a design parameter x during verification, and the design parameter x is changed to x', leading to fluctuation of an optimization result. In the conventional technology, to cope with uncertainty introduced during optimization, modeling is directly performed on an uncertain input. To be specific, a surrogate model is constructed between the uncertain input and an input for an objective function, to resolve the foregoing problem of uncertainty. However, the surrogate model constructed in the conventional technology is applicable only to a case in which a perturbation type follows a Gaussian-like distribution. In addition, a calculation process during optimization is complex, and inference efficiency is low. SUMMARY Embodiments of this application provide an antenna design method and apparatus, to find, under perturbation of an input of any distribution type, a design parameter that enables an antenna to obtain an optimal average performance parameter. An application scope is wide, and calculation complexity can be effectively reduced through accelerated processing, to greatly improve inference efficiency of a surrogate model during design parameter optimization. According to a first aspect, this application provides an antenna design method. The method includes: determining a design parameter of the antenna through a plurality of iterations, where an ith iteration process is as follows: obtaining a plurality of vectors from optimization space, where each of the plurality of vectors includes M elements, the M elements respectively represent M design parameters of the antenna, the optimization space is used to describe a value range of each of the M design parameters, and M and i are positive integers; obtaining input distributions of the M design parameters, and inputting the input distributions of the M design parameters and the plurality of vectors to a surrogate model of Bayesian optimization BO, to obtain a plurality of groups of predicted values of a performance parameter of the antenna and a plurality of degrees of certainty respectively corresponding to the plurality of groups of predicted values, where the plurality of groups of predicted values respectively correspond to the plurality of vectors, the plurality of degrees of certainty are respectively used to describe confidences of the plurality of groups of predicted values, a processing process of the surrogate model includes calculating a covariance between any two of the plurality of vectors, and the covariance between any two vectors is obtained by performing calculation on a sample set obtained by sampling the input distributions; processing the plurality of groups of predicted values and the plurality of degrees of certainty based on an acquisition function of the BO, to determine a first vector from the plurality of vectors, where an output of the acquisition function that corresponds to the first vector is the largest; inputting the first vector to an objective function to obtain a group of performance parameters of the antenna; and when the group of performance parameters of the antenna meets a preset c