Design method of alternative model for Fresnel zone patch antenna integrating physical mechanism

By integrating physical mechanisms and deep neural network models, the radiation characteristics of Fresnel waveband chip antennas are quickly and accurately predicted, solving the problems of long and low accuracy in the existing technology, and achieving efficient Fresnel waveband chip antenna design.

CN120297165BActive Publication Date: 2025-08-12NANJING UNIV OF INFORMATION SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510782408.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-06-12
Publication Date
2025-08-12
Estimated Expiration
2045-06-12

AI Technical Summary

Technical Problem

In the Fresnel waveband antenna design, the analytical method has low accuracy, full-wave simulation takes a long time and high computing resources, and the training cost of traditional deep learning alternative models is high and the accuracy is insufficient.

Method used

By integrating the physical mechanism, the physical parameters of the Fresnel waveband antenna are calculated, the deep neural network model is constructed, and the theoretical gain data and simulated gain data are used to train the model to predict the gain situation at the radiation angle.

Benefits of technology

It realizes rapid and accurate prediction of the radiation characteristics of Fresnel waveband antennas, reducing training costs, and improving accuracy and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120297165B_ABST
    Figure CN120297165B_ABST
Patent Text Reader

Abstract

The present invention discloses a design method for a replacement model of a Fresnel zone patch antenna that integrates physical mechanisms, and belongs to the field of antenna design technology in artificial intelligence technology and communication technology. The method includes: calculating the physical parameters of the Fresnel zone patch antenna; calculating the estimated gain data and theoretical gain data of the Fresnel zone patch antenna with different physical parameters within a preset radiation angle range; constructing a deep neural network model, and training the deep neural network model using data in a training set and label data; using the trained deep neural network model as a replacement model for the Fresnel zone patch antenna, and using the replacement model to predict the gain at different radiation angles. The replacement model designed by the present invention can quickly and accurately predict the gain within the main radiation angle of the FZPA. Compared with traditional replacement models, it has high accuracy, fast speed, low training cost, and significantly higher efficiency than traditional FZPA design methods.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to antenna design technology in artificial intelligence technology and communication technology, and in particular to a design method for a Fresnel zone plate antenna replacement model integrating physical mechanisms. Background Art

[0002] In the design of Fresnel zone plate antennas (FZPAs), analytical methods and full-wave simulation are the primary approaches used to determine the FZPA's radiation characteristics (including radiation pattern, gain, sidelobes, etc.). The analytical method, based on Maxwell's equations, theoretically calculates the radiation characteristics of a Fresnel zone plate antenna under certain approximate conditions. This method is useful for roughly evaluating the antenna's overall performance, but due to the introduction of approximations, its accuracy is relatively low. Full-wave simulation, performed using numerical simulation methods on high-performance computers, offers high accuracy. However, Fresnel zone plate antennas are typically large, and full-wave simulation requires significant time and computing resources. FZPAs have numerous parameters, and the optimization process requires numerous full-wave simulations, making this method time-consuming and inefficient.

[0003] A surrogate model is a simplified model used to replace a more complex or computationally expensive original model. Most surrogate models are data-driven, using machine learning, deep learning, and other methods to build a surrogate model based on data. While capturing the essential characteristics of the original model, they also provide a faster and more efficient approximation of its behavior. Establishing a surrogate model for FZPA can significantly improve FZPA optimization efficiency.

[0004] Traditional deep learning alternative models directly use antenna physical parameters as input features for training, but they require large data sets for training and their accuracy still needs to be further improved. Summary of the Invention

[0005] Purpose of the invention: In response to the above problems, the purpose of the present invention is to provide a design method for an alternative model of a Fresnel zone patch antenna that integrates physical mechanisms. The alternative model can quickly and accurately predict the gain of the Fresnel zone patch antenna within the main radiation angle.

[0006] Technical solution: The design method of the Fresnel zone patch antenna replacement model integrating physical mechanisms of the present invention comprises the following steps:

[0007] Step 1: Calculate the physical parameters of the Fresnel zone patch antenna according to the design principle of the Fresnel zone patch antenna;

[0008] Step 2: setting a range of physical parameters, and obtaining physical parameters of a preset number of Fresnel zone plate antennas by hypercube sampling within the set range;

[0009] Step 3: Calculate theoretical gain data of Fresnel zone plate antennas with different physical parameters within a preset radiation angle range based on physical theory;

[0010] Step 4: Calculate the simulated gain data of Fresnel zone plate antennas with different physical parameters within a preset radiation angle range based on simulation software;

[0011] Step 5: construct a data set using the physical parameters and the theoretical gain data of the Fresnel zone patch antenna within a preset angle range, use the simulated gain data as label data, and divide the data set into a training set and a test set in proportion;

[0012] Step 6: Build a deep neural network model and train the deep neural network model using the data in the training set and the labeled data. The input of the deep neural network model is the physical parameters and theoretical gain data of the Fresnel zone patch antenna, and the output is the gain of the Fresnel zone patch antenna at a certain radiation angle.

[0013] In step 7, the trained deep neural network model is used as a replacement model for the Fresnel zone patch antenna, and the replacement model is used to predict the gain at different radiation angles.

[0014] Furthermore, the physical parameters include the radius of each area of the Fresnel zone patch antenna from b1 to b N , and focal length F, Indicates the number of the Fresnel zone.

[0015] Furthermore, step 3 includes:

[0016] Use the cosine function to construct the feed equivalent model, the formula is:

[0017] ,

[0018] Where, represents the gain function of the feed source, T represents the maximum gain of the patch antenna as the feed source, represents the incident angle, n represents the exponential parameter of the feed gain function, which characterizes the shape of the feed pattern;

[0019] The formula for calculating the electric field after the feed passes through the Fresnel zone plate antenna is:

[0020] ,

[0021] ,

[0022] Where, represents the radial distance from the observation point to the antenna, represents the elevation angle between the observation point and the antenna axis in the spherical coordinate system, represents the azimuth between the observation point and the antenna axis in the spherical coordinate system, The decomposition of the electric field intensity in the spherical coordinate system is Quantity, represents the normalized complex constant factor, is a discrete parameter, represents the boundary angle corresponding to the m'th band, The decomposition of the electric field intensity in the spherical coordinate system is Quantity; represents the aperture field amplitude factor in the open area, represents the complex phase exponential term, is the phase parameter, Indicates far field The integral kernel of the electric field component in the direction, Indicates far field The integral kernel of the electric field component in the direction is:

[0023] ,

[0024] ,

[0025] ,

[0026] ,

[0027] ,

[0028] Where, and are the zero-order and second-kind Bessel functions, j represents the imaginary unit, k represents the wave number, F represents the focal length, d Represents the geometric thickness of a single dielectric plate along the propagation direction of electromagnetic waves;

[0029] To calculate the gain of a Fresnel zone patch antenna, the formula is:

[0030] ,

[0031] Where, represents the free space impedance, Indicates the transmit power, It represents the modulus of electric field strength related to distance r and mode parameter n.

[0032] Furthermore, the radius b of the Fresnel zone patch antenna is N The following formula must be satisfied:

[0033] ,

[0034] Where, Indicates the number of the Fresnel zone, represents the focal length, represents the wavelength, Indicates the FZPA order.

[0035] Furthermore, the Fresnel zone plate antenna is composed of a dielectric substrate, the thickness of the odd-numbered rings of the dielectric substrate is denoted as t, and the thickness of the even-numbered rings of the dielectric substrate is denoted as tw, where w satisfies the following formula:

[0036] ,

[0037] Where, Represents the relative dielectric constant.

[0038] Furthermore, the preset radiation angle range of the Fresnel zone patch antenna is 0 to 20 degrees.

[0039] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0040] 1. Compared with the analytical method, the present invention can obtain the gain result quickly while greatly improving the accuracy;

[0041] 2. Compared with the full-wave simulation method, the present invention maintains accuracy while greatly improving the speed;

[0042] 3. While introducing deep learning, the present invention does not directly use antenna physical parameters as input features for training, but instead incorporates physical principles for training. The training cost is lower than that of traditional models, the accuracy is higher, and it has higher efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of the design method for an alternative model of a Fresnel zone patch antenna that incorporates physical mechanisms;

[0044] Figure 2 It is the structural diagram of the second-order FZPA;

[0045] Figure 3 It is a schematic diagram of the structure of artificial neural network;

[0046] Figure 4 The following is the operation flow chart of each model;

[0047] Figure 5 The MAE comparison chart of model A and model B;

[0048] Figure 6 This is the MSE comparison chart of model A and model B;

[0049] Figure 7 This is a comparison chart of the predicted and true values of 5 groups of samples randomly selected from the test set after model A training is completed;

[0050] Figure 8 This is a comparison chart of the predicted and true values of 5 groups of samples randomly selected from the test set after model B training is completed. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of this application more clear, this application is further described in detail below with reference to the accompanying drawings and embodiments.

[0052] The design method of the Fresnel zone patch antenna replacement model integrating physical mechanisms described in this embodiment is as follows: Figure 1 As shown, the method includes the following steps:

[0053] Step 1: According to the design principle of the Fresnel zone patch antenna, the physical parameters of the Fresnel zone patch antenna are calculated.

[0054] Fresnel Zone Plate Antenna (FZPA) is a lens antenna with a relative dielectric constant ε r The dielectric substrate can be regarded as a dielectric substrate with a radius from b1 to b N The physical parameters include the radius of each area of the Fresnel zone patch antenna from b1 to b N , and focal length F.

[0055] like Figure 2 The diagram shows the structure of the second-order FZPA, i.e. Q=2, with a relative dielectric constant of ε r The dielectric substrate is composed of three circular rings with radii b1, b2, and b3, and a focal length of F. In the figure, P is the focus. The thickness of the odd-numbered rings is t, the thickness of the even-numbered rings is tw, and the difference in thickness between the odd-numbered and even-numbered rings is w. represents the angle between the focus and the outer edge of the first torus, Represents the angle between the focus and the outer edge of the third torus.

[0056] The focal length F is selected to maximize the FZPA aperture efficiency. Without considering other loss factors, the aperture efficiency Irradiation efficiency and spillover efficiency The product of the irradiation efficiency measures the uniformity and matching degree of the energy radiated by the feed source on the reflecting surface, and the overflow efficiency measures the proportion of the feed source radiated energy that is effectively intercepted by the reflecting surface, avoiding the energy "overflowing" to the area outside the reflecting surface.

[0057] Irradiation efficiency and spillover efficiency The expressions are:

[0058] ,

[0059] ,

[0060] ,

[0061] Where, is the physical area of the reflecting surface, is the integral variable on the reflection surface, representing the coordinates of any point on the reflection surface, is the field intensity distribution at each point on the reflecting surface, Indicates the location The power density at represents the area vector element of the reflecting surface, is the second kind of surface integral on the sphere, A is the surface area corresponding to the reflection surface, Represents the surface area integral corresponding to the reflection surface.

[0062] Furthermore, according to the design principle of the Fresnel zone patch antenna, the radius b of the Fresnel zone patch antenna is N The following formula must be satisfied:

[0063] ,

[0064] Where, Indicates the number of the Fresnel zone, represents the focal length, represents the wavelength, Indicates the FZPA order.

[0065] Furthermore, the Fresnel zone plate antenna is composed of a dielectric substrate, the thickness of the odd-numbered rings of the dielectric substrate is denoted as t, and the thickness of the even-numbered rings of the dielectric substrate is denoted as tw, where w satisfies the following formula:

[0066] ,

[0067] Where, Represents the relative dielectric constant.

[0068] The feed is at focal length F. Since electromagnetic waves travel different distances to different locations on the FZPA, this creates a path difference, causing interference after the waves pass through the FZPA. The above formula calculates a suitable w, compensating the phase at the corresponding location, transforming the originally destructive superposition into an enhanced superposition, thereby achieving high antenna gain.

[0069] First, according to the aperture efficiency When the focal length is at its maximum, the focal diameter ratio F / D is calculated to be 0.31, where F is the focal length and D is the diameter. At this point, Q is 2, and the aperture efficiency reaches a maximum of 76.01%. Based on the design principles of the Fresnel zone patch antenna, the initial physical dimension parameters b1 to b9 and the focal length F are calculated, as shown in Table 1.

[0070] Step 2: setting a range of physical parameters, and obtaining a preset number of physical parameters of Fresnel zone plate antennas by hypercube sampling within the set range.

[0071] Step 3: Calculate theoretical gain data of Fresnel zone plate antennas with different physical parameters within a preset radiation angle range based on physical theory.

[0072] Furthermore, the preset radiation angle range of the Fresnel zone patch antenna is 0 to 20 degrees.

[0073] Furthermore, step 3 includes:

[0074] Use the cosine function to construct the feed equivalent model, the formula is:

[0075] ,

[0076] Where, represents the gain function of the feed source, T represents the maximum gain of the patch antenna as the feed source, represents the incident angle, n represents the exponential parameter of the feed gain function, which characterizes the shape of the feed pattern;

[0077] The formula for calculating the electric field after the feed passes through the Fresnel zone plate antenna is:

[0078] ,

[0079] ,

[0080] Where, represents the radial distance from the observation point to the antenna, represents the elevation angle between the observation point and the antenna axis in the spherical coordinate system, represents the azimuth between the observation point and the antenna axis in the spherical coordinate system, The decomposition of the electric field intensity in the spherical coordinate system is Quantity, represents the normalized complex constant factor, is a discrete parameter, represents the boundary angle corresponding to the m'th band, The decomposition of the electric field intensity in the spherical coordinate system is Quantity; represents the aperture field amplitude factor in the open area, represents the complex phase exponential term, is the phase parameter, Indicates far field The integral kernel of the electric field component in the direction, Indicates far field The integral kernel of the electric field component in the direction is:

[0081] ,

[0082] ,

[0083] ,

[0084] ,

[0085] ,

[0086] Where, and are the zero-order and second-kind Bessel functions, j represents the imaginary unit, k represents the wave number, F represents the focal length, d Represents the geometric thickness of a single dielectric plate along the propagation direction of electromagnetic waves;

[0087] To calculate the gain of a Fresnel zone patch antenna, the formula is:

[0088] ,

[0089] Where, represents the free space impedance, Indicates the transmit power, It represents the modulus of electric field strength related to distance r and mode parameter n.

[0090] Step 4: Calculate the simulated gain data of Fresnel zone plate antennas with different physical parameters within a preset radiation angle range based on simulation software.

[0091] Based on the initial values, we set the focal length and radius ranges. This range ensures that the deep learning model can learn as many cases as possible while preventing the inner radius from being larger than the outer radius, or the radii of two adjacent circles being too close, which would render the simulation impossible. Hypercube sampling was performed within the set range, resulting in 1000 different sets of FZPA physical dimensions, as shown in Table 1 (units in millimeters).

[0092] Table 1

[0093]

[0094] In this example, the primary radiation angle of the Fresnel zone patch antenna is set between 0 and 20 degrees, with a single point selected for each degree. Using physical theory, the gain of 1,000 FZPAs with different physical parameters from 0 to 20 degrees is calculated as theoretical gain data. This theoretical gain data and the physical parameters are used together as input features. Matlab and CST simulations are then used to generate the gain of 1,000 FZPAs with different physical parameters from 0 to 20 degrees, which are then used as simulated gain data.

[0095] Step 5: Construct a data set using the physical parameters and the theoretical gain data of the Fresnel zone plate antenna within a preset angle range, use the simulated gain data as label data, and divide the data set into a training set and a test set in proportion.

[0096] In one example, a data set was constructed using physical parameters and theoretical gain data of Fresnel zone patch antennas within 0 to 20 degrees. After the data set was constructed, the training set and test set were divided into a ratio of 7:3, and finally a deep neural network was trained.

[0097] Step 6: Build a deep neural network model and train the deep neural network model using the data in the training set and the labeled data; the input of the deep neural network model is the physical parameters and theoretical gain data of the Fresnel zone patch antenna, and the output is the gain of the Fresnel zone patch antenna at a certain radiation angle.

[0098] In one example, a deep neural network model uses an artificial neural network with a structure such as Figure 3 As shown, the network consists of two convolutional layers and three linear layers. The first convolutional layer uses 30 3×3 filters with a stride of 1 to maintain the input size. The second layer increases the number of filters to 60, using the same filter size, to help capture more complex features. Each convolutional layer is followed by a ReLU activation function, and finally three linear layers, with the number of neurons adjusted to match the output size.

[0099] The input items of the artificial neural network are the physical parameters of the antenna and the gain of each group of FZPA within the main radiation angle calculated by physical mechanism. The physical parameters of the antenna include the radius of each area b1 to b9 of the FZPA and the focal length F. The main radiation angle for gain calculation by physical mechanism is 0 degrees to 20 degrees, with one point for each degree. In this example, a second-order FZPA is used, that is, Q=2, the number of areas is 9, and the FZPA working at 30GHz is used for principle verification. The FZPA uses the relative dielectric constant r The dielectric substrate is 2.1 mm thick, with a thickness difference w set to 5.2679 mm based on theoretical calculations. The feed uses a circular patch antenna with a maximum gain of 4.95 dBi. The true value of the artificial neural network is the gain of the FZPA from 0 to 20 degrees obtained by CST simulation, with one point taken for each degree.

[0100] This example uses the Adam optimizer with a learning rate of 0.005 and defines the network’s loss function as the mean squared error (MSE).

[0101] In step 7, the trained deep neural network model is used as a replacement model for the Fresnel zone patch antenna, and the replacement model is used to predict the gain at different radiation angles.

[0102] In order to highlight the superiority of the performance of the surrogate model constructed in the present invention, the traditional deep learning surrogate model is selected as the comparison model, denoted as Model A, and the surrogate model obtained in the present invention is designated as Model B. The workflows of the two models are as follows: Figure 4 As shown in the figure, initial physical parameters of the FZPA are calculated based on its design principles. These parameters are then hypercube sampled within a given range, resulting in 1,000 sets of FZPA physical parameters. CST simulation and physical principle calculations are then performed on these 1,000 sets of FZPAs with different parameters, yielding both the actual FZPA gain from 0 to 20 degrees and the theoretically estimated gain. This data is used for subsequent model training. Model A directly uses the antenna physical parameters—the FZPA's radius b1 to b9 and focal length F—as input features. Model B's input features consist of the antenna physical parameters and the gain of each FZPA at its primary radiation angle (one point per degree from 0 to 20 degrees) calculated from physical theory. Both models utilize the same neural network architecture, and the actual values of the models are the FZPA gain from 0 to 20 degrees obtained from CST simulation. After training, the desired FZPA gain from 0 to 20 degrees can be obtained using Models A and B, respectively.

[0103] Combine Figure 5 and Figure 6,It can be seen that the mean absolute error (MAE) and MSE of each angle of the model are much lower than those of model A, indicating that the prediction effect of model B is more accurate and efficient.

[0104] Combine Figures 7 and 8 After the training of each model is completed, the comparison of the prediction and true value of 5 groups of samples randomly selected from the test set shows that the prediction effect of model B is much better than that of model A.

[0105] In summary, the high-precision Fresnel zone patch antenna alternative model proposed in the present invention, which integrates physical mechanisms, is trained based on the physical size parameters of the FZPA and the results of physical theory calculations. It can accurately, quickly and efficiently obtain the radiation characteristics of the FZPA, thereby solving the problem of difficulty in quickly and accurately obtaining the radiation characteristics of the FZPA.

Claims

1. A design method for a Fresnel zone patch antenna replacement model that integrates physical mechanisms, characterized by: The steps include: Step 1: Calculate the physical parameters of the Fresnel zone patch antenna according to the design principle of the Fresnel zone patch antenna; Step 2: setting a range of physical parameters, and obtaining physical parameters of a preset number of Fresnel zone plate antennas by hypercube sampling within the set range; Step 3: Calculate theoretical gain data of Fresnel zone plate antennas with different physical parameters within a preset radiation angle range based on physical theory; Step 4: Calculate the simulated gain data of Fresnel zone plate antennas with different physical parameters within a preset radiation angle range based on simulation software; Step 5: construct a data set using the physical parameters and the theoretical gain data of the Fresnel zone patch antenna within a preset angle range, use the simulated gain data as label data, and divide the data set into a training set and a test set in proportion; Step 6: Build a deep neural network model and train the deep neural network model using the data in the training set and the labeled data. The input of the deep neural network model is the physical parameters and theoretical gain data of the Fresnel zone patch antenna, and the output is the gain of the Fresnel zone patch antenna at a certain radiation angle. In step 7, the trained deep neural network model is used as a replacement model for the Fresnel zone patch antenna, and the replacement model is used to predict the gain at different radiation angles.

2. The design method of a Fresnel zone plate antenna alternative model integrating physical mechanisms according to claim 1 is characterized in that: The physical parameters include the radius of each area of the Fresnel zone patch antenna from b1 to b N , and focal length F, Indicates the number of the Fresnel zone.

3. The design method of a Fresnel zone plate antenna alternative model integrating physical mechanisms according to claim 2, characterized in that: Step 3 includes: Use the cosine function to construct the feed equivalent model, the formula is: , Where, represents the gain function of the feed source, T represents the maximum gain of the patch antenna as the feed source, represents the incident angle, n represents the exponential parameter of the feed gain function, which characterizes the shape of the feed pattern; The formula for calculating the electric field after the feed passes through the Fresnel zone plate antenna is: , , Where, represents the radial distance from the observation point to the antenna, represents the elevation angle between the observation point and the antenna axis in the spherical coordinate system, represents the azimuth between the observation point and the antenna axis in the spherical coordinate system, The decomposition of the electric field intensity in the spherical coordinate system is Quantity, represents the normalized complex constant factor, is a discrete parameter, represents the boundary angle corresponding to the m'th band, The decomposition of the electric field intensity in the spherical coordinate system is Quantity; represents the aperture field amplitude factor in the open area, represents the complex phase exponential term, is the phase parameter, Indicates far field The integral kernel of the electric field component in the direction, Indicates far field The integral kernel of the electric field component in the direction is: , , , , , Where, and are the zero-order and second-kind Bessel functions, j represents the imaginary unit, k represents the wave number, F represents the focal length, d Represents the geometric thickness of a single dielectric plate along the propagation direction of electromagnetic waves; To calculate the gain of a Fresnel zone patch antenna, the formula is: , Where, represents the free space impedance, Indicates the transmit power, It represents the modulus of electric field strength related to distance r and mode parameter n.

4. The design method of a Fresnel zone plate antenna alternative model integrating physical mechanisms according to claim 2 or 3, characterized in that: The radius b of the Fresnel zone patch antenna N The following formula must be satisfied: , Where, Indicates the number of the Fresnel zone, represents the focal length, represents the wavelength, Indicates the FZPA order.

5. The design method of a Fresnel zone plate antenna alternative model integrating physical mechanisms according to any one of claims 1 to 3, characterized in that: The Fresnel zone patch antenna is composed of a dielectric substrate. The thickness of the odd-numbered rings of the dielectric substrate is denoted as t, and the thickness of the even-numbered rings is denoted as tw. w satisfies the following formula: , Where, Represents the relative dielectric constant.

6. The design method of a Fresnel zone plate antenna alternative model integrating physical mechanisms according to any one of claims 1 to 3, characterized in that: The preset radiation angle range of the Fresnel zone patch antenna is 0~20 degrees.

Citation Information

Patent Citations

  • Antenna design method based on neural network

    CN109086531A

  • Multi-target antenna design method based on neural network

    CN109271695A