Deep learning based close-spaced antenna decoupling method and device, and storage medium

Abstract

The application discloses a tight-spacing antenna decoupling method and device based on deep learning, and a storage medium. The method comprises the following steps: acquiring tight-spacing antenna data; inputting the tight-spacing antenna data into an antenna decoupling structure prediction model to obtain decoupling structure parameters; and performing tight-spacing antenna decoupling processing according to the decoupling structure parameters. The antenna decoupling structure prediction model is obtained through the following steps: acquiring rectangular microstrip patch antenna information; calculating patch length and patch width; establishing a binary antenna model; performing electromagnetic field simulation processing by using a preset script program and electromagnetic simulation software to obtain simulation result data sets; constructing an initial convolutional neural network model; inputting the simulation result data sets into the initial convolutional neural network model according to mean square error, so that the initial convolutional neural network model is trained to obtain the antenna decoupling structure prediction model. The application realizes antenna decoupling, improves efficiency and accuracy, and can be widely applied to the field of wireless communication technology.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a method, apparatus, and storage medium for decoupling closely spaced antennas based on deep learning. Background Technology

[0002] In the design of MIMO (Multiple-Input Multiple-Output) systems and array antenna architectures, electromagnetic coupling between adjacent antennas is a problem. Strong electromagnetic coupling between antennas leads to reduced gain, radiation pattern shift, and performance degradation. With the advancement and development of modern communication methods, portable and miniaturized communication devices with closely spaced antennas have gradually become mainstream. This results in even stronger electromagnetic coupling between antennas and more severe radiation pattern shifts. Traditional decoupling methods involve manually designing and optimizing the decoupling structure between antennas, which is time-consuming and inefficient. Furthermore, even decoupling structures designed using machine learning methods still suffer from large-angle radiation pattern shifts and low accuracy.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The embodiments of the present invention provide a method, apparatus and storage medium for decoupling closely spaced antennas based on deep learning, which effectively improves efficiency and accuracy.

[0005] On one hand, embodiments of the present invention provide a deep learning-based method for decoupling closely spaced antennas, comprising the following steps:

[0006] Acquire close-pitch antenna data;

[0007] The tight-pitch antenna data is input into the antenna decoupling structure prediction model to obtain the decoupling structure parameters.

[0008] Perform tight-pitch antenna decoupling processing based on the decoupling structure parameters;

[0009] The antenna decoupling structure prediction model is obtained through the following steps:

[0010] Obtain information about the rectangular microstrip patch antenna;

[0011] Calculate the patch length and patch width based on the rectangular microstrip patch antenna information;

[0012] A binary antenna model is established based on the patch length and the patch width;

[0013] Based on the binary antenna model, electromagnetic field simulation processing is performed using a preset script program and electromagnetic simulation software to obtain a simulation result dataset, which includes S21 coupling parameters and pattern parameters.

[0014] Construct the initial convolutional neural network model;

[0015] Based on the mean square error, the simulation result dataset is input into the initial convolutional neural network model to train the initial convolutional neural network model and obtain the antenna decoupling structure prediction model.

[0016] In some embodiments, calculating the patch length and patch width based on the rectangular microstrip patch antenna information includes:

[0017] Extract the relative permittivity and thickness of the dielectric substrate from the information of the rectangular microstrip patch antenna;

[0018] The patch width is calculated based on the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate;

[0019] Calculate the length difference based on the patch width, the dielectric substrate thickness, and the relative permittivity of the dielectric substrate;

[0020] The patch length is calculated based on the length difference, the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate.

[0021] In some embodiments, establishing the binary antenna model based on the patch length and the patch width includes:

[0022] Extract the dielectric substrate thickness from the rectangular microstrip patch antenna information;

[0023] Set the antenna spacing;

[0024] The binary antenna model is established based on the antenna spacing, patch length, patch width, dielectric substrate thickness, and radiating element size.

[0025] In some embodiments, the step of performing electromagnetic field simulation processing based on the binary antenna model using a preset script program and electromagnetic simulation software to obtain a simulation result dataset includes:

[0026] Based on the binary antenna model, set the simulation model parameters;

[0027] Create an electromagnetic simulation project;

[0028] Initialize the project settings for the electromagnetic simulation project;

[0029] Based on the simulation model parameters and the project settings, a simulation model is established;

[0030] The simulation model is used to perform the simulation, and the simulation result dataset is obtained.

[0031] In some embodiments, the process of constructing the initial convolutional neural network model includes:

[0032] Construct the first convolutional layer, with a kernel size of 32×32×32;

[0033] After the first convolutional layer, a first pooling layer is constructed, the size of which is 16×16×32;

[0034] After the first pooling layer, a second convolutional layer is constructed, and the kernel size of the second convolutional layer is 16×16×64.

[0035] After the second convolutional layer, a second pooling layer is constructed, the size of which is 5×5×64;

[0036] After the second pooling layer, a third convolutional layer is constructed, and the kernel size of the third convolutional layer is 5×5×128.

[0037] After the third convolutional layer, a third pooling layer is constructed, the size of which is 2×2×128;

[0038] After the third pooling layer, a flattening layer is constructed;

[0039] After the flattened layer, a first fully connected layer is constructed, wherein the number of neurons in the first fully connected layer is 128;

[0040] After the first fully connected layer, a second fully connected layer is constructed, and the second fully connected layer has 64 neurons.

[0041] After the second fully connected layer, a third fully connected layer is constructed, wherein the number of neurons in the third fully connected layer is 32;

[0042] After the third fully connected layer, an output layer is constructed, which has 14 neurons.

[0043] In some embodiments, calculating the patch width based on the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate includes:

[0044] Based on the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate, the patch width is calculated using the patch width calculation formula, which is:

[0045] ,

[0046] In the formula, The width of the patch, The speed of light, The operating resonant frequency is... is the relative permittivity of the dielectric substrate.

[0047] In some embodiments, calculating the patch length based on the length difference, the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate includes:

[0048] Based on the length difference, the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate, the patch length is calculated using the patch length calculation formula, which is:

[0049] ,

[0050] In the formula, The patch length is... The speed of light, The operating resonant frequency is... The relative permittivity of the dielectric substrate is _____. The length difference is mentioned.

[0051] On the other hand, embodiments of the present invention provide a deep learning-based close-pitch antenna decoupling device, comprising:

[0052] The first module is used to acquire data from closely spaced antennas;

[0053] The second module is used to input the close-pitch antenna data into the antenna decoupling structure prediction model to obtain the decoupling structure parameters.

[0054] The third module is used to perform tight-pitch antenna decoupling processing based on the decoupling structure parameters;

[0055] The antenna decoupling structure prediction model is obtained through the following steps:

[0056] Obtain information about the rectangular microstrip patch antenna;

[0057] Calculate the patch length and patch width based on the rectangular microstrip patch antenna information;

[0058] A binary antenna model is established based on the patch length and the patch width;

[0059] Based on the binary antenna model, electromagnetic field simulation processing is performed using a preset script program and electromagnetic simulation software to obtain a simulation result dataset, which includes S21 coupling parameters and pattern parameters.

[0060] Construct the initial convolutional neural network model;

[0061] Based on the mean square error, the simulation result dataset is input into the initial convolutional neural network model to train the initial convolutional neural network model and obtain the antenna decoupling structure prediction model.

[0062] On the other hand, embodiments of the present invention provide a computer device, including:

[0063] At least one processor;

[0064] At least one memory for storing at least one program;

[0065] When the at least one program is executed by the at least one processor, the at least one processor implements the method.

[0066] On the other hand, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described thereon.

[0067] The beneficial effects of this invention are as follows:

[0068] This invention first acquires closely spaced antenna data, then inputs it into an antenna decoupling structure prediction model to obtain decoupling structure parameters. Finally, it performs closely spaced antenna decoupling processing based on these parameters, thereby achieving antenna decoupling and improving efficiency and accuracy. Specifically, the antenna decoupling structure prediction model first acquires information about a rectangular microstrip patch antenna, calculates the patch length and width, then establishes a binary antenna model based on these parameters. Based on this model, it performs electromagnetic field simulation using a pre-defined script and electromagnetic simulation software to obtain a simulation result dataset. An initial convolutional neural network model is then constructed. Finally, based on the mean square error, the simulation result dataset is input into the initial convolutional neural network model for training, thus improving model accuracy.

[0069] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description and the drawings. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a flowchart of a deep learning-based close-pitch antenna decoupling method according to an embodiment of the present invention;

[0072] Figure 2 This is a flowchart illustrating an embodiment of the present invention for obtaining a prediction model of an antenna decoupling structure;

[0073] Figure 3 This is a schematic diagram of a binary antenna model according to an embodiment of the present invention;

[0074] Figure 4 This is a schematic diagram of a co-simulation process according to an embodiment of the present invention;

[0075] Figure 5 This is a schematic diagram of the front of a simulation model of a binary antenna array according to an embodiment of the present invention;

[0076] Figure 6 This is a schematic diagram of the back of a simulation model of a binary antenna array according to an embodiment of the present invention;

[0077] Figure 7 This is a schematic diagram of a convolutional neural network model structure according to an embodiment of the present invention;

[0078] Figure 8 This is a schematic diagram illustrating the comparison results of S-parameters according to an embodiment of the present invention;

[0079] Figure 9 This is a schematic diagram showing the comparison results of a radiation pattern according to an embodiment of the present invention;

[0080] Figure 10 This is a schematic diagram illustrating the overall process of decoupling structure design according to an embodiment of the present invention;

[0081] Figure 11 This is a schematic diagram of a tight-pitch antenna decoupling device based on deep learning, according to an embodiment of the present invention.

[0082] Figure 12 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0084] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0085] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0086] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0087] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0088] Antenna decoupling refers to reducing or eliminating mutual interference between antenna elements to improve antenna radiation efficiency and communication performance. In multi-antenna systems, mutual coupling between antenna elements can lead to reduced antenna efficiency and pattern distortion, thus affecting the performance of the entire system.

[0089] In related technologies, electromagnetic coupling between adjacent antennas is an unavoidable problem in the design of antenna arrays. Strong electromagnetic coupling between antennas leads to reduced antenna gain, radiation pattern shift, and performance degradation. With the advancement and development of modern communication methods, portable and miniaturized communication devices have gradually become mainstream, placing new demands on antenna miniaturization. This further intensifies electromagnetic coupling between antennas and exacerbates antenna radiation pattern shift, making decoupling for tight-pitch pattern correction an urgent problem to solve. Simultaneously, with the rapid development of wireless communication technology, antenna design is facing unprecedented performance challenges, particularly the urgent need for high gain and excellent directivity, making traditional single antennas insufficient for the complex requirements of the information age. Against this backdrop, array antennas have been widely adopted and applied in various fields as a solution. However, in MIMO (Multiple-Input Multiple-Output) systems and array antenna architectures, the mutual coupling between antenna elements inevitably leads to radiation pattern distortion, especially in mobile devices pursuing extreme miniaturization. Current research focuses on decoupling techniques for compact (pitch reduced to one-twentieth of a wavelength) array antennas, striving to overcome these technological bottlenecks. Despite significant progress, the problem of pattern tilt remains unresolved. Therefore, in-depth exploration and development of decoupling techniques for compact microstrip array antennas, particularly focusing on the accurate correction and optimization of patterns, is crucial. This will not only effectively improve the overall performance of the antenna system but also have profound implications for advancing engineering technologies such as miniaturization and integration of array antennas. Meanwhile, deep learning, as an emerging force, is profoundly changing the traditional landscape of electromagnetic field modeling and simulation. Compared to traditional computationally intensive methods such as the method of moments and the finite element method, deep learning, with its powerful data processing capabilities, can extract efficient and accurate prediction models from massive amounts of data, significantly accelerating the electromagnetic field solution process and improving computational accuracy. Furthermore, deep learning is increasingly widely used in electromagnetic optimization and design. It can not only automatically adjust design parameters to optimize device performance but also inject innovative design concepts and ways of thinking into the field of electromagnetics, greatly improving design efficiency and flexibility. Traditional decoupling methods rely on manual design and optimization of the decoupling structure between antennas, resulting in relatively large spacing between antennas, long processing times, and low efficiency. Meanwhile, even with decoupling structures designed using machine learning and other methods, the antenna still suffers from large-angle pattern shifts and low accuracy.

[0090] In view of this, the embodiments of the present invention are aimed at 0.05 To address the issue of pattern tilt in patch antennas with tight spacing, deep learning algorithms were introduced, and combined with script programs and electromagnetic simulation software for electromagnetic field simulation, a comprehensive technique was explored to achieve simultaneous decoupling and pattern correction.

[0091] The deep learning-based close-pitch antenna decoupling method provided in this application relates to the field of wireless communication technology. This deep learning-based close-pitch antenna decoupling method can be applied to terminals, servers, or software running on either terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the deep learning-based close-pitch antenna decoupling method, but is not limited to the above forms.

[0092] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0093] The embodiments of this application will be explained in detail below with reference to the accompanying drawings:

[0094] Figure 1 This is an optional flowchart of a deep learning-based close-pitch antenna decoupling method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S103.

[0095] Step S101: Obtain close-pitch antenna data;

[0096] Step S102: Input the close-pitch antenna data into the antenna decoupling structure prediction model to obtain the decoupling structure parameters;

[0097] Step S103: Perform tight-pitch antenna decoupling processing according to the decoupling structure parameters.

[0098] In some embodiments, the close-pitch antenna data can be acquired first, then the close-pitch antenna data can be input into the antenna decoupling structure prediction model to obtain the decoupling structure parameters, and then the close-pitch antenna decoupling processing can be performed according to the decoupling structure parameters, thereby correcting the radiation pattern, realizing antenna decoupling, and improving efficiency and accuracy.

[0099] In some embodiments, such as Figure 2 As shown, the specific implementation process of obtaining the antenna decoupling structure prediction model may include, but is not limited to, steps S201 to S206.

[0100] Step S201: Obtain information about the rectangular microstrip patch antenna;

[0101] Step S202: Calculate the patch length and patch width based on the rectangular microstrip patch antenna information;

[0102] Step S203: Establish a binary antenna model based on the patch length and patch width;

[0103] Step S204: Based on the binary antenna model, electromagnetic field simulation is performed using a preset script program and electromagnetic simulation software to obtain a simulation result dataset. The simulation result dataset includes S21 coupling parameters and pattern parameters.

[0104] Step S205: Construct the initial convolutional neural network model;

[0105] Step S206: Based on the mean square error, input the simulation result dataset into the initial convolutional neural network model to train the initial convolutional neural network model and obtain the antenna decoupling structure prediction model.

[0106] Steps S201 to S206, as shown in the embodiments of this application, obtain the antenna decoupling structure prediction model, thereby improving the model accuracy.

[0107] In step S201 of some embodiments, the rectangular microstrip patch antenna information can be obtained through an antenna database. The rectangular microstrip patch antenna information can also be obtained through other methods, and is not limited to these. It is understood that adding different metal strips between antennas or creating slots or grooves in the floor can achieve decoupling and pattern correction between antennas. For ease of study, this embodiment selects parallel, side-by-side binary microstrip antennas operating at 5.8 GHz as the research object, and arranges these binary microstrip antennas along the H-plane. The H-plane of the antenna, also called the magnetic plane, refers to a plane parallel to the magnetic field direction (or perpendicular to the electric field direction) and containing the antenna's maximum radiation direction.

[0108] In some embodiments, step S202, calculating the patch length and patch width based on the rectangular microstrip patch antenna information, may include, but is not limited to, the following steps:

[0109] Extract the relative permittivity and thickness of the dielectric substrate from the information of the rectangular microstrip patch antenna;

[0110] Calculate the patch width based on the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate;

[0111] Calculate the length difference based on the patch width, substrate thickness, and relative permittivity of the substrate.

[0112] The patch length is calculated based on the length difference, speed of light, operating resonant frequency, and relative permittivity of the dielectric substrate.

[0113] In some embodiments, the microstrip antenna used in this embodiment is an antenna formed by laying a conductive patch on a dielectric substrate with a ground plane. Depending on the shape of the conductive patch, microstrip patch antennas can be classified into ring, rectangular, and circular microstrip patch antennas, etc. The antenna designed in this embodiment is a rectangular microstrip patch antenna. This antenna is fed by a microstrip line, and a quarter-impedance matching transformer is designed to achieve impedance matching. The size of the rectangular microstrip patch antenna is designed based on the electrical parameters of the selected dielectric (such as the relative permittivity of the dielectric substrate and the loss tangent). The relative permittivity and thickness of the dielectric substrate can be extracted from the information of the rectangular microstrip patch antenna. Then, based on the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate, the patch width is calculated using the patch width calculation formula, where the patch width calculation formula is: In the formula, For patch width, At the speed of light, The operating resonant frequency, Let be the relative permittivity of the dielectric substrate. Then, based on the patch width, dielectric substrate thickness, and relative permittivity, calculate the length difference. The formula for calculating the length difference is: In the formula, For the length difference, The thickness of the dielectric substrate is given. Finally, based on the length difference, the speed of light, the operating resonant frequency, and the relative permittivity of the dielectric substrate, the patch length is calculated using the patch length calculation formula: In the formula, For patch length, At the speed of light, The operating resonant frequency, The relative permittivity of the dielectric substrate, For example, based on the antenna's operating resonant frequency of 5.8 GHz and a dielectric substrate thickness (a known condition) of 1.6 mm, the overall size of the dielectric substrate for the binary antenna can be designed to be 46.00 mm × 31.00 mm (patch length × patch width).

[0114] In some embodiments, step S203, establishing a binary antenna model based on the patch length and patch width, may include, but is not limited to, the following steps:

[0115] Extracting dielectric substrate thickness from rectangular microstrip patch antenna information;

[0116] Set the antenna spacing;

[0117] A binary antenna model is established based on the antenna spacing, patch length, patch width, dielectric substrate thickness, and radiating element size.

[0118] In some embodiments, the dielectric substrate thickness can be extracted first from the rectangular microstrip patch antenna information, then the antenna spacing can be set, and finally a binary antenna model can be established based on the antenna spacing, patch length, patch width, dielectric substrate thickness, and radiating element size. For example, the binary antenna model is as follows: Figure 3 As shown, the thickness of the extracted dielectric substrate is 1.6 mm, and the overall dimensions of the dielectric substrate are 46.00 mm × 31.00 mm (patch length × patch width). The spacing between the two antennas can be set to 2.52 mm, approximately one-twentieth of the wavelength, resulting in a binary antenna model. Simultaneously, the dimensions of the upper surface radiating element of a single antenna are 12.80 mm × 11.33 mm (length × width).

[0119] In some embodiments, in step S204, based on the binary antenna model, electromagnetic field simulation processing is performed using a preset script program and electromagnetic simulation software to obtain a simulation result dataset, which may include, but is not limited to, the following steps:

[0120] Based on the binary antenna model, set the simulation model parameters;

[0121] Create an electromagnetic simulation project;

[0122] Initialize the project settings for the electromagnetic simulation project;

[0123] Based on the simulation model parameters and project settings, a simulation model is established;

[0124] Simulation results datasets are obtained by using a simulation model.

[0125] In some embodiments, building a predictive model requires training on a large amount of data; however, acquiring such a massive dataset still requires significant time and effort. Therefore, a script written in MATLAB can be used to communicate with CST Studio Suite (electromagnetic simulation software) via an API interface, automating electromagnetic field simulations using CST Studio Suite. This automates the modeling and simulation process, greatly improving data acquisition efficiency. The pre-set script can be a MATLAB script, and the electromagnetic simulation software can be CST Studio Suite. The co-simulation process is as follows: Figure 4 As shown, in electromagnetic simulation software, the simulation model parameters can be set first based on the binary antenna model, then an electromagnetic simulation project can be created, the project settings of the electromagnetic simulation project can be initialized, and a simulation model can be built based on the simulation model parameters and project settings. Finally, the simulation model can be used to perform simulation and obtain a simulation result dataset. The simulation result dataset may include S21 decoupling parameters and radiation pattern parameters. For example, the front of the simulation model of the binary antenna array is shown below. Figure 5 As shown, it mainly consists of two patch antennas and a metal strip between them. The back of the simulation model of the binary antenna array is shown below. Figure 6 As shown, there is a strip-shaped defective ground structure between the patch antennas. It is understood that decoupling can be achieved using the metal strip and the defective ground structure.

[0126] In some embodiments, the process of constructing the initial convolutional neural network model in step S205 may include, but is not limited to, the following steps:

[0127] Construct the first convolutional layer, with a kernel size of 32×32×32;

[0128] After the first convolutional layer, a first pooling layer is constructed, with a size of 16×16×32.

[0129] After the first pooling layer, a second convolutional layer is constructed, with a kernel size of 16×16×64.

[0130] After the second convolutional layer, a second pooling layer is constructed, with a size of 5×5×64.

[0131] After the second pooling layer, a third convolutional layer is constructed, with a kernel size of 5×5×128.

[0132] After the third convolutional layer, a third pooling layer is constructed, with a size of 2×2×128.

[0133] After the third pooling layer, a flattening layer is constructed;

[0134] After the flattened layer, the first fully connected layer is constructed, with 128 neurons in the first fully connected layer;

[0135] After the first fully connected layer, a second fully connected layer is constructed, with 64 neurons in the second fully connected layer;

[0136] After the second fully connected layer, a third fully connected layer is constructed, with 32 neurons in the third fully connected layer;

[0137] After the third fully connected layer, an output layer is constructed, with 14 neurons.

[0138] In some embodiments, the convolutional neural network model structure is as follows: Figure 7 As shown, a first convolutional layer and a first pooling layer can be constructed first, followed by a second convolutional layer and a second pooling layer, and then a third convolutional layer and a third pooling layer. The kernel size of the first convolutional layer is 32×32×32, and the kernel size of the first pooling layer is 16×16×32. The kernel size of the second convolutional layer is 16×16×64, and the kernel size of the second pooling layer is 5×5×64. The kernel size of the third convolutional layer is 5×5×128, and the kernel size of the third pooling layer is 2×2×128. Furthermore, each pooling layer is followed by a Dropout layer, and LeakyReLU activation and same padding settings are applied. After the third pooling layer, a flattening layer is constructed, followed by a first fully connected layer, a second fully connected layer, and a third fully connected layer. The first fully connected layer has 128 neurons, the second fully connected layer has 64 neurons, and the third fully connected layer has 32 neurons. Furthermore, each fully connected layer uses L2 regularization and the LeakyReLU activation function. After the third fully connected layer, an output layer is constructed, which has 14 neurons and uses the ReLU activation function. Additionally, the input image size for the convolutional neural network model is 32×32×1 (length×width×height).

[0139] In some embodiments, in step S206, the simulation result dataset can be input into the initial convolutional neural network model based on the mean squared error (MSE) to train the initial convolutional neural network model and obtain the antenna decoupling structure prediction model. For example, a simulation result dataset with an image size of 32×32×1 is input into the initial convolutional neural network model. During training, the mean squared error (MSE) loss function is used to calculate the loss for parameter optimization. The formula for calculating the mean squared error is: In the formula, Mean square error, For the sample size, For the true value, The predicted values ​​are those of the model. It's understandable that the predicted values ​​output by the antenna decoupling structure prediction model represent the relevant parameters of the antenna decoupling structure, used for mechanical energy antenna decoupling. More importantly, the model's predicted values ​​can be compared with the actual values ​​to verify the model's accuracy. The actual values ​​are the parameters of the actual decoupling structure, obtained through CST simulation. To verify the accuracy of the network design, the S21 decoupling coefficients corresponding to the decoupling structure parameters output by the trained antenna decoupling structure prediction model are compared with the S21 decoupling coefficients of the actual structure, obtained through CST software simulation. Under the premise of meeting the expected design, a decoupling structure with parameters meeting actual requirements is designed using the antenna decoupling structure prediction model. It's understandable that CNN convolutional neural networks, through components such as convolutional layers, pooling layers, and activation functions, can effectively extract features from data such as images and audio. The decoupling structure is essentially a two-dimensional planar structure, equivalent to an image, so its features can be extracted using a convolutional neural network. In the comparison results between the actual and predicted values, the comparison results of the S-parameters are as follows: Figure 8 As shown, the comparison results of the radiation patterns are as follows: Figure 9 As shown in the figure, the prediction results show that, for both S-parameters and radiation pattern, the predicted response (predicted value) of the antenna decoupling structure prediction model matches the expected response (true value) well in waveform, and the resonant frequency is also predicted relatively accurately. By using deep learning to design the decoupling structure, the tilted radiation pattern was corrected.

[0140] In some embodiments, the overall process of decoupling structure design is as follows: Figure 10 As shown, we can first model a two-element antenna with a decoupling structure, then use MATLAB to control the CST simulation software for automatic modeling and simulation, then input a certain amount of simulation structure data into a convolutional neural network for training to obtain a prediction model, and finally verify the accuracy of the prediction model to design the decoupling structure.

[0141] In some embodiments, this embodiment can decouple the closely spaced antenna and correct its radiation pattern, achieving both simultaneously. This design not only reduces the antenna's size but also reduces weight and material usage, thus lowering manufacturing costs. The lightweight and portable design enables the new technology to be widely applied in various related fields such as airborne, spaceborne, shipborne, and vehicle-mounted systems. This embodiment utilizes deep learning for closely spaced decoupling and radiation pattern correction, combining metal strips and defective grounding to achieve its functionality. A prediction model based on a deep convolutional neural network is constructed to realize the design of the decoupling structure.

[0142] The beneficial effects of implementing the embodiments of the present invention include: The embodiments of the present invention first acquire closely spaced antenna data, then input the closely spaced antenna data into an antenna decoupling structure prediction model to obtain decoupling structure parameters, and finally perform closely spaced antenna decoupling processing based on the decoupling structure parameters, thereby achieving antenna decoupling and improving efficiency and accuracy. Specifically, the antenna decoupling structure prediction model first acquires rectangular microstrip patch antenna information, calculates the patch length and patch width, then establishes a binary antenna model based on the patch length and patch width, performs electromagnetic field simulation processing using a preset script program and electromagnetic simulation software based on the binary antenna model, obtains a simulation result dataset, then constructs an initial convolutional neural network model, and finally inputs the simulation result dataset into the initial convolutional neural network model based on the mean square error to train the initial convolutional neural network model, thereby improving model accuracy.

[0143] like Figure 11 As shown, embodiments of the present invention also provide a deep learning-based close-pitch antenna decoupling device, comprising:

[0144] The first module 801 is used to acquire close-pitch antenna data;

[0145] The second module 802 is used to input the close-pitch antenna data into the antenna decoupling structure prediction model to obtain the decoupling structure parameters.

[0146] The third module 803 is used to perform tight-pitch antenna decoupling based on the decoupling structure parameters;

[0147] The antenna decoupling structure prediction model is obtained through the following steps:

[0148] Obtain information about the rectangular microstrip patch antenna;

[0149] Calculate the patch length and patch width based on the rectangular microstrip patch antenna information;

[0150] A binary antenna model is established based on the patch length and patch width;

[0151] Based on the binary antenna model, electromagnetic field simulation is performed using a pre-set script program and electromagnetic simulation software to obtain a simulation result dataset, which includes S21 coupling parameters and pattern parameters.

[0152] Construct the initial convolutional neural network model;

[0153] Based on the mean square error, the simulation result dataset is input into the initial convolutional neural network model to train the initial convolutional neural network model and obtain the antenna decoupling structure prediction model.

[0154] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0155] like Figure 12 As shown, embodiments of the present invention also provide a computer device, including:

[0156] At least one processor 901;

[0157] At least one memory 902 is used to store at least one program;

[0158] When at least one program is executed by at least one processor, such that at least one processor achieves Figure 1 The method shown.

[0159] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0160] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements... Figure 1 The method shown.

[0161] The content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0162] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A deep learning based method for close-spaced antenna decoupling, characterized in that, The method comprises the following steps: obtaining close-spaced antenna data; inputting the close-spaced antenna data into an antenna decoupling structure prediction model to obtain decoupling structure parameters; performing close-spaced antenna decoupling processing according to the decoupling structure parameters; wherein the antenna decoupling structure prediction model is obtained by the following steps: obtaining rectangular microstrip patch antenna information; calculating patch length and patch width according to the rectangular microstrip patch antenna information; establishing a binary antenna model according to the patch length and the patch width; performing electromagnetic field simulation processing using a preset script program and electromagnetic simulation software according to the binary antenna model to obtain a simulation result dataset, the simulation result dataset comprising S21 coupling parameters and directional diagram parameters; constructing an initial convolutional neural network model; inputting the simulation result dataset into the initial convolutional neural network model according to mean square error to enable the initial convolutional neural network model to be trained to obtain the antenna decoupling structure prediction model.

2. The method of claim 1, wherein, The calculation of patch length and patch width according to the rectangular microstrip patch antenna information comprises: extracting dielectric plate relative permittivity and dielectric plate thickness from the rectangular microstrip patch antenna information; calculating the patch width according to light speed, working resonant frequency and the dielectric plate relative permittivity; calculating a length difference according to the patch width, the dielectric plate thickness and the dielectric plate relative permittivity; calculating the patch length according to the length difference, the light speed, the working resonant frequency and the dielectric plate relative permittivity.

3. The method of claim 1, wherein, The establishment of a binary antenna model according to the patch length and the patch width comprises: extracting dielectric plate thickness from the rectangular microstrip patch antenna information; setting an antenna spacing; establishing the binary antenna model according to the antenna spacing, the patch length, the patch width, the dielectric plate thickness and radiation element size.

4. The method of claim 1, wherein, The electromagnetic field simulation processing using a preset script program and electromagnetic simulation software according to the binary antenna model to obtain a simulation result dataset comprises: setting simulation model parameters according to the binary antenna model; creating an electromagnetic simulation project; initializing project settings of the electromagnetic simulation project; establishing a simulation model according to the simulation model parameters and the project settings; performing simulation using the simulation model to obtain the simulation result dataset.

5. The method of claim 1, wherein, The construction process of the initial convolutional neural network model comprises: constructing a first convolutional layer, the convolutional kernel size of the first convolutional layer being 32x32x32; constructing a first pooling layer after the first convolutional layer, the size of the first pooling layer being 16x16x32; constructing a second convolutional layer after the first pooling layer, the convolutional kernel size of the second convolutional layer being 16x16x64; constructing a second pooling layer after the second convolutional layer, the size of the second pooling layer being 5x5x64; constructing a third convolutional layer after the second pooling layer, the convolutional kernel size of the third convolutional layer being 5x5x128; constructing a third pooling layer after the third convolutional layer, the size of the third pooling layer being 2x2x128; After the third pooling layer, a flattening layer is constructed; After the flattening layer, a first fully connected layer is constructed, the number of neurons of the first fully connected layer being 128; After the first fully connected layer, a second fully connected layer is constructed, the number of neurons of the second fully connected layer being 64; After the second fully connected layer, a third fully connected layer is constructed, the number of neurons of the third fully connected layer being 32; After the third fully connected layer, an output layer is constructed, the number of neurons of the output layer being 14.

6. The method of claim 2, wherein, The calculation of the patch width according to the speed of light, the working resonant frequency and the relative dielectric constant of the dielectric plate comprises: The patch width is calculated according to the speed of light, the working resonant frequency and the relative dielectric constant of the dielectric plate through a patch width calculation formula, the patch width calculation formula being: ， wherein is the patch width, is the speed of light, is the operating resonant frequency, is the relative permittivity of the dielectric slab.

7. The method of claim 2, wherein, The calculation of the patch length according to the length difference, the speed of light, the working resonant frequency and the relative dielectric constant of the dielectric plate comprises: The patch length is calculated according to the length difference, the speed of light, the working resonant frequency and the relative dielectric constant of the dielectric plate through a patch length calculation formula, the patch length calculation formula being: ， wherein, is the patch length, is the speed of light, is the operating resonant frequency, is the relative permittivity of the dielectric slab, is the length difference.

8. A deep learning based compact antenna decoupling device, characterized in that, Comprise: A first module is configured to acquire tight-interval antenna data. A second module is configured to input the tight-interval antenna data into an antenna decoupling structure prediction model to obtain decoupling structure parameters. A third module is configured to perform tight-interval antenna decoupling processing according to the decoupling structure parameters. The antenna decoupling structure prediction model is obtained through the following steps: Obtain rectangular microstrip patch antenna information. According to the rectangular microstrip patch antenna information, calculate the patch length and the patch width. According to the patch length and the patch width, establish a binary antenna model. According to the binary antenna model, perform electromagnetic field simulation processing by using a preset script program and electromagnetic simulation software to obtain simulation result data sets, the simulation result data sets comprising S21 coupling parameters and directional diagram parameters. Construct an initial convolutional neural network model. According to the mean square error, input the simulation result data sets into the initial convolutional neural network model to enable the initial convolutional neural network model to be trained to obtain the antenna decoupling structure prediction model.

9. A computer apparatus, comprising: Comprise: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method of any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1-9. The computer program is executed by the processor to implement the method of any one of claims 1-7. The computer program is executed by the processor to implement the method of any one of claims 1-7.

Images