Antenna performance rapid prediction system based on plane equivalent source
By constructing a fast antenna performance prediction system based on planar equivalent sources, and using machine learning models and Green's functions to calculate electromagnetic field distribution, the system solves the problems of high computational resource consumption and insufficient model generalization ability in complex antenna design. It achieves fast and accurate prediction of three-dimensional radiation performance and near-field distribution, and supports rapid topology optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-30
- Publication Date
- 2026-05-15
AI Technical Summary
Existing antenna performance prediction methods consume large amounts of computational resources and take a long time when dealing with complex structures or large-scale array antennas, making it difficult to meet the needs of rapid iterative design. Furthermore, they cannot fully reflect the radiation characteristics and near-field distribution in three-dimensional space, and the interpretability and generalization ability of the models are insufficient.
A rapid antenna performance prediction system based on planar equivalent sources is constructed. A dataset is generated through full-wave simulation, and a machine learning model is trained to predict equivalent current sources and equivalent magnetic current sources. The electromagnetic field distribution is calculated by combining free-space Green's function, thereby achieving rapid prediction of three-dimensional radiation performance and near-field distribution.
It enables rapid and accurate prediction of antenna performance across the entire space, reduces computational resources and time costs, improves the model's generalization ability and interpretability, and supports rapid topology optimization design.
Smart Images
Figure CN122046953A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of antenna performance prediction technology, and more specifically to a fast antenna performance prediction system based on planar equivalent sources. Background Technology
[0002] As a key component in wireless communication systems, the performance of antennas directly affects the communication quality and efficiency of the entire system. Conventional antenna design processes primarily rely on full-wave simulation to obtain antenna performance parameters, a method that provides relatively accurate results when dealing with simple antenna structures. However, with the continuous advancement of antenna technology and increasingly complex design requirements, the applicability of full-wave simulation methods decreases significantly when designing complex antenna structures or large-scale array antennas.
[0003] Full-wave simulation requires precise calculations of the electromagnetic field around the antenna, involving complex numerical computation processes, including numerous matrix operations and iterative solutions. For complex antenna structures, irregular geometry and diverse material properties lead to extremely complex electromagnetic field distributions, resulting in exponentially increasing computational complexity. For large-scale array antennas, the sheer number of elements and significant coupling between them further complicates the calculations. In such cases, full-wave simulation not only consumes substantial computational resources, such as high-performance computer clusters, but also takes a long time, typically several hours or even days to complete a single simulation. This severely limits the efficiency of antenna design, lengthens the design cycle, and makes it difficult to meet the demands of rapid iterative design.
[0004] To improve antenna design efficiency, machine learning-based methods have become a research hotspot and are being applied to antenna performance prediction. Machine learning possesses powerful data processing and pattern recognition capabilities, enabling it to learn complex relationships between inputs and outputs from large amounts of data, thus achieving rapid prediction of antenna performance. For example, the paper "C. Cui, WT Li, XT Ye, P. Rocca, YQ Hei, and XW Shi, 'An Effective Artificial Neural Network-Based Method for Linear Array BeampatternSynthesis' IEEE Transactions on Antennas and Propagation, vol. 69, no. 10, pp. 6431–6443, Oct. 2021" discloses a machine learning-based array antenna design framework. This framework uses an Encoder-Decoder architecture to design array antennas, where the Decoder part can quickly predict the antenna's radiation performance based on the array's topology. This method improves antenna design efficiency to some extent and provides a new approach to antenna performance prediction.
[0005] However, existing machine learning-based antenna performance prediction methods still have many limitations. Most current methods directly predict antenna performance parameters such as radiation patterns based on the antenna's topology, resulting in predictions that typically only cover a single plane of the radiation pattern and cannot comprehensively reflect the antenna's radiation characteristics in three-dimensional space. Furthermore, near-field distribution is crucial for antenna performance analysis and electromagnetic compatibility assessment with other devices, but existing methods struggle to predict this near-field distribution.
[0006] Furthermore, existing methods suffer from significant shortcomings in interpretability and generalization ability. Machine learning models are often considered black-box models, with their internal decision-making processes difficult to understand, resulting in poor interpretability of predictions and hindering targeted optimization of antenna designs based on these predictions. Simultaneously, existing models exhibit limited generalization ability when faced with different types and sizes of antennas, generally requiring model retraining, which further increases the complexity and cost of antenna design. Summary of the Invention
[0007] The purpose of this invention is to propose a fast antenna performance prediction system based on planar equivalent sources, which can improve the effectiveness and efficiency of antenna performance prediction.
[0008] To achieve the above objectives, this invention proposes a fast antenna performance prediction system based on a planar equivalent source, including a construction and training module, which performs the following steps: Determine the scope of topology optimization and the plane containing the equivalent source. S u The size and shape of the antenna are randomly generated within an optimized range, and the antenna's planar shape is obtained through full-wave simulation. S u electric field and magnetic field ; Construct the dataset into input and output parts. The input part of the dataset is the antenna topology. The dataset output is based on the electric field. and magnetic field Computational plane S u Equivalent current source on and equivalent magnetic current source Equivalent current source and equivalent magnetic current source It is represented as real and imaginary components, and the processed dataset is divided into training and validation sets; Establish a first machine learning prediction model and a second machine learning prediction model, based on antenna topology respectively. Predict equivalent current sources and equivalent magnetic current sources, train two models using data from the training set, and validate the accuracy of the trained machine learning models using data from the validation set.
[0009] Beneficial effects of the basic scheme: This scheme uses planar equivalent sources as the core link. By predicting the equivalent current source and equivalent magnetic current source on the plane above the antenna aperture, the three-dimensional radiation performance and near-field performance at any location of the antenna can be further derived. Compared with the limitations of traditional machine learning models that can only predict specific far-field parameters or near-field parameters at fixed locations, this system achieves full-coverage prediction of antenna performance across the entire space, making it applicable to a wider range of scenarios.
[0010] Traditional antenna performance prediction relies on full-wave simulation, which consumes significant computational resources and time, especially for complex antenna topologies or multi-condition testing scenarios, resulting in extremely low efficiency. This solution constructs a machine learning prediction model that correlates the antenna topology with planar equivalent sources during training. After training, only the antenna topology needs to be input to quickly output equivalent current and magnetic current sources, eliminating the need to repeatedly execute the full-wave simulation process. This significantly shortens the antenna performance prediction cycle and substantially reduces computational resource investment and simulation costs.
[0011] In the dataset construction phase, this scheme calculates equivalent sources based on real electric and magnetic field data obtained from full-wave simulation, ensuring the accuracy and reliability of the dataset. At the same time, it adopts a mode of separating training and validation sets, and trains and verifies the accuracy of two machine learning prediction models separately, which can effectively avoid the risk of model overfitting and ensure the accuracy of equivalent source prediction results.
[0012] This approach generates antenna models randomly within a preset topology optimization range and constructs the dataset during dataset construction, enabling the trained machine learning model to adapt to diverse antenna topologies. For novel antenna topologies or intermediate solutions during topology optimization, no additional model architecture adjustments are required to quickly predict equivalent sources and performance, demonstrating strong generalization ability and engineering practicality.
[0013] The real and imaginary component representations of equivalent current and magnetic current sources can be directly applied to subsequent electromagnetic field calculations. Based on the predicted equivalent source data, the near-field and far-field performance parameters of the target location can be easily derived. This characteristic directly supports rapid antenna topology optimization design. During the antenna development phase, by quickly predicting the performance of different topology schemes, the iterative selection process for the optimal antenna solution can be accelerated.
[0014] As a feasible and preferred solution, the equivalent current source and equivalent magnetic current source Based on the principle of equivalence, it is determined by the following formula:
[0015] in, Perpendicular to the equivalent source plane S u The normal vector.
[0016] As a feasible and preferred option, the equivalent current source and equivalent magnetic current source Represented as real and imaginary parts, specifically:
[0017]
[0018] Here, re(·) represents the real part, and im(·) represents the imaginary part.
[0019] As a feasible preferred embodiment, the system also includes a prediction model usage module, which performs the following steps: Input the antenna topology to be predicted into the trained model to obtain a planar antenna. S uThe equivalent current source and equivalent magnetic current source are calculated; based on the free space Green's function, the electromagnetic field distribution generated by the equivalent current source and equivalent magnetic current source is calculated.
[0020] As a feasible preferred embodiment, the field calculation based on the free-space Green's function includes: The electric field is calculated using the electric dextral Green's function and the magnetic dextral Green's function, through the following formula. and magnetic field :
[0021]
[0022] in, and These are the electric dyadic Green's function and the magnetic dyadic Green's function, respectively.
[0023] As a feasible preferred embodiment, the electric dyadic Green's function and the magnetic dyadic Green's function are expressed as follows:
[0024]
[0025] in, For unit vector, For wave number, For free space Green's functions.
[0026] As a feasible and preferred option, the free-space Green's function The expression is:
[0027] in, and These are the position vectors of the source point and the field point, respectively.
[0028] As a feasible and preferred approach, the electric field expression for near-field calculations is: .
[0029] As a feasible and preferred approach, the electric field expression for far-field calculations is: . Attached Figure Description
[0030] Figure 1 This is a schematic diagram of the execution logic of the module for constructing and training a fast antenna performance prediction based on a planar equivalent source.
[0031] Figure 2 This is a schematic diagram of a millimeter-wave series-fed microstrip array topology.
[0032] Figure 3 A schematic diagram comparing machine learning predictions of current sources with full-wave simulation results.
[0033] Figure 4 A schematic diagram comparing machine learning predictions of magnetic flux sources with full-wave simulation results.
[0034] Figure 5 A schematic diagram comparing the far-field 3D radiation pattern predicted by machine learning with the full-wave simulation results.
[0035] Figure 6 A comparison of near-field distribution predictions from machine learning and full-wave simulation results. Detailed Implementation
[0036] To make the technical solution and advantages of this application clearer, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are only some embodiments of the present invention, and are only used to explain this application, not to limit it. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated; they can be combined with each other to achieve better technical effects. The same reference numerals appearing in the accompanying drawings of the following embodiments represent the same features or components, and can be applied to different embodiments.
[0037] Furthermore, unless otherwise defined, the technical or scientific terms used in this invention description shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains.
[0038] The present invention will now be described in further detail with reference to the accompanying drawings.
[0039] Reference Figure 1 This disclosure provides a fast antenna performance prediction system based on a planar equivalent source, including a module for constructing and training a fast antenna prediction model and a module for using the prediction model.
[0040] The construction and training module performs the following steps.
[0041] Step S1: Determine the antenna type and element model, and determine the topology optimization range and the plane containing the planar equivalent source. S u The size and shape of the antenna are determined. Within the optimization range, an antenna model is randomly generated, and the electric field of the antenna on the plane is obtained through full-wave simulation. and magnetic field .
[0042] Step S2, dataset preprocessing, constructing the dataset input part, i.e., antenna topology. Construct the dataset output section, which is the equivalent current source of the antenna on the plane. and equivalent magnetic current source The expression is:
[0043] in, Perpendicular to the equivalent source plane S u The normal vector.
[0044] Equivalent current source and equivalent magnetic current source Represented as real and imaginary parts, this allows conventional machine learning models to handle complex numbers, i.e.:
[0045]
[0046] Where re(·) represents the real part and im(·) represents the imaginary part, since and z-components and normal vector Collinear, with a value of 0, is omitted from the dataset.
[0047] The processed dataset is divided into a training set and a validation set.
[0048] Step S3: Build a machine learning prediction model and Based on the antenna topology t, the equivalent current source and equivalent magnetic current source are predicted respectively, and the two models are trained using the data in the training set using appropriate machine learning algorithms.
[0049] Step S4: Use the data in the validation set to validate the accuracy of the trained machine learning model.
[0050] The prediction model uses modules to perform the following steps.
[0051] Step T1: Input the antenna topology to be predicted into the trained model to obtain the planar topology. S u The equivalent current source and equivalent magnetic current source on the surface.
[0052] Step T2: Based on the free-space Green's function, calculate the field generated by the equivalent source:
[0053]
[0054] in, and These are the electric dyadic Green's function and the magnetic dyadic Green's function, respectively.
[0055]
[0056]
[0057] in, Green's function in free space:
[0058] In near-field calculations, the electric field can be expressed as:
[0059] When calculating the electric field in the far field, it can be simplified to:
[0060] Finally, the near-field and far-field performance of the antenna is calculated based on the equivalent current source and magnetic current source.
[0061] This disclosure provides examples of applying the system to predict the far-field performance of microstrip millimeter-wave antennas.
[0062] The topology of millimeter-wave microstrip array antennas is as follows: Figure 2 As shown, the border represents the edge processing of the antenna, which operates at a frequency of 5 GHz and contains 12 microstrip elements. The width of each element is adjusted... w i Amplitude and phase control are achieved, enabling pattern synthesis. Due to the left-right symmetry of the array element structure, the antenna topology... =[ w 1, w 2, w 3, w 4, w 5, w 6).
[0063] The distance between the plane of the predicted equivalent source and the array aperture plane needs to be set to account for the rapid attenuation of the fall wave and drastic near-field changes when the distance is too close, and the need for a larger plane to ensure the accuracy of the results when the distance is too far. After verification, this distance is set to... In this embodiment, the distance is 10mm, and the distance interval of the data on the plane is set to... In this embodiment, the diameter is 6mm, and the length and width of the plane are set to 80mm and 540mm respectively. The model construction and training process is as follows.
[0064] S1-1. Within the scope of antenna topology optimization, randomly generate 600 sets of antenna topologies, and perform full-wave simulation on the randomly generated antenna models; S2-1. Extract the electric and magnetic fields on the plane, calculate the equivalent current source and equivalent magnetic current source, construct the dataset, and divide it into training set and validation set of size 500 and 100 respectively. S3-1. Using two training sets of 500, construct equivalent current source and equivalent magnetic current source prediction networks based on generalized regression neural network (GRNN) respectively. and Select smoothing factor ; S4-1. Use the data in the validation set to validate the accuracy of the trained machine learning model.
[0065] The steps for predicting the far-field performance of the model in the validation set are as follows: Step T1-1: Input the antenna topology to be predicted into the trained model to obtain the equivalent plane. S u Equivalent current sources and equivalent magnetic current sources on the surface; Step T2-2: Calculate the far field generated by the equivalent source based on the free-space Green's function.
[0066] Figure 3 and Figure 4 The results show a comparison between the model's predicted current source and magnetic current source and the full-wave simulation results. Figure 5 The comparison between the far-field three-dimensional radiation pattern calculated based on the predicted equivalent source and the full-wave simulation results is shown, and it can be seen that the radiation pattern matches well in the upper half space.
[0067] This disclosure provides examples of applying the system to predict the near-field distribution of microstrip millimeter-wave antennas.
[0068] Using the model from the previous example to predict the equivalent source distribution, and then calculating the near-field distribution within a rectangular region 2.5m from the aperture surface of the millimeter-wave microstrip array with dimensions of 2m, the steps are as follows: Step S1-2: Input the antenna topology to be predicted into the trained model to obtain the equivalent plane. S u Equivalent current sources and equivalent magnetic current sources on the surface; Step S2-2: Calculate the near field generated by the equivalent source based on the free-space Green's function.
[0069] Figure 6 The results show a comparison between the predicted near-field distribution and the full-wave simulation results, and the results agree well.
[0070] The above content is merely an embodiment of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can improve and implement this solution based on the guidance provided in this application and their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. A fast antenna performance prediction system based on planar equivalent sources, characterized in that, It includes a construction and training module, which performs the following steps: Determine the scope of topology optimization and the plane containing the equivalent source. S u The size and shape of the antenna are randomly generated within an optimized range, and the antenna's planar shape is obtained through full-wave simulation. S u Electric and magnetic fields ; Construct the dataset into input and output parts. The input part of the dataset is the antenna topology. The dataset output is based on the electric field. and magnetic field Computational plane S u Equivalent current source on and equivalent magnetic current source Equivalent current source and equivalent magnetic current source It is represented as real and imaginary components, and the processed dataset is divided into training and validation sets; Establish a first machine learning prediction model and a second machine learning prediction model, based on antenna topology respectively. Predict equivalent current sources and equivalent magnetic current sources, train two models using data from the training set, and validate the accuracy of the trained machine learning models using data from the validation set.
2. The fast antenna performance prediction system based on planar equivalent sources according to claim 1, characterized in that, Equivalent current source and equivalent magnetic current source Based on the principle of equivalence, it is determined by the following formula: in, Perpendicular to the equivalent source plane S u The normal vector.
3. The fast antenna performance prediction system based on planar equivalent sources according to claim 1, characterized in that, Equivalent current source and equivalent magnetic current source Represented as real and imaginary parts, specifically: Here, re(·) represents the real part, and im(·) represents the imaginary part.
4. The fast antenna performance prediction system based on planar equivalent sources according to claim 1, characterized in that, It also includes a prediction model usage module, which performs the following steps: Input the antenna topology to be predicted into the trained model to obtain a planar antenna. S u The equivalent current source and equivalent magnetic current source are calculated; based on the free space Green's function, the electromagnetic field distribution generated by the equivalent current source and equivalent magnetic current source is calculated.
5. The fast antenna performance prediction system based on planar equivalent sources according to claim 1, characterized in that, The field calculation based on the free-space Green's function includes: The electric field is calculated using the electric dextral Green's function and the magnetic dextral Green's function, through the following formula. and magnetic field : in, and These are the electric dyadic Green's function and the magnetic dyadic Green's function, respectively.
6. The fast antenna performance prediction system based on planar equivalent sources according to claim 5, characterized in that, The electric dyadic Green's function and the magnetic dyadic Green's function are expressed as follows: in, For unit vector, For wave number, For free space Green's functions.
7. The fast antenna performance prediction system based on planar equivalent sources according to claim 6, characterized in that, Green's function in free space The expression is: in, and These are the position vectors of the source point and the field point, respectively.
8. The fast antenna performance prediction system based on planar equivalent sources according to claim 1, characterized in that, For near-field calculations, the electric field expression is: 。 9. The fast antenna performance prediction system based on planar equivalent sources according to claim 1, characterized in that, In far-field calculations, the electric field expression is: 。