Artificial intelligence aided design and performance prediction method of radio frequency coaxial connector
By establishing the mapping relationship between key dimensional parameters and electrical performance of RF coaxial connectors through an ANN model, and employing an inverse optimization method, the problems of long design cycle and poor reusability of RF coaxial connectors were solved, achieving rapid and reliable design and optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-13
- Publication Date
- 2026-05-12
AI Technical Summary
Existing design methods for RF coaxial connectors suffer from long design cycles, poor reusability, and inability to meet increasingly stringent design requirements. Furthermore, when frequency bands or interface standards change, repeated parameter scanning and electromagnetic simulations are required.
An artificial neural network (ANN) model is used to establish the mapping relationship between key size parameters and electrical performance. The optimization method is used to directly find the optimal solution, avoiding manual iteration and realizing automated design. Furthermore, there is no need to re-simulate when design requirements change.
It significantly shortens the design cycle of RF coaxial connectors, improves design reusability, reduces reliance on electromagnetic simulation, and enables rapid and reliable design.
Smart Images
Figure CN122021353A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radio frequency and microwave device simulation technology, specifically to an artificial intelligence-aided design and performance prediction method for radio frequency coaxial connectors. Background Technology
[0002] Radio frequency (RF) coaxial connectors, as a core component for achieving efficient and low-loss transmission of high-frequency signals, are widely used in communication base stations, phased array radars, and high-precision testing equipment. With the development of RF systems, the design requirements for RF coaxial connectors are becoming increasingly stringent, and their demand is growing rapidly. Therefore, the rapid and reliable design of RF coaxial connector structures has become a critical aspect of RF system engineering.
[0003] Current designs for RF coaxial connectors primarily rely on traditional methods such as theoretical estimation, electromagnetic simulation, or manual tuning. Designers first determine the basic radial dimensions based on the characteristic impedance formula for coaxial transmission lines. Then, they use 3D electromagnetic simulation software to perform parametric scanning (hereinafter referred to as "parameter scanning") on structures such as impedance matching sections and insulating media, iterating continuously to find the combination of dimensional parameters that meets the electrical performance specifications. The electrical performance specifications of RF coaxial connectors typically refer to scattering parameters, such as voltage standing wave ratio (VSWR) and insertion loss, which measure the matching and transmission efficiency of the RF coaxial connector.
[0004] However, the aforementioned traditional design methods have limitations. First, there is a non-linear relationship between the critical dimensional parameters and electrical performance of RF coaxial connectors, requiring designers to repeatedly scan these parameters and manually iterate, resulting in a long design cycle and failing to meet the growing demand for RF coaxial connectors. Second, when the operating frequency band or interface standard changes, repeated parameter scans and electromagnetic simulations are required, leading to poor reusability. Therefore, a design method that can automatically predict structural dimensions, reduce reliance on electromagnetic simulation, and offer good design reusability is crucial. Summary of the Invention
[0005] The purpose of this invention is to provide an artificial intelligence-aided design and performance prediction method for radio frequency coaxial connectors. Artificial Neural Networks (ANNs) are an important technical means for realizing artificial intelligence. This invention establishes a mapping relationship between key dimensional parameters and electrical performance through an ANN model, and directly optimizes the electrical performance indicators using a reverse optimization method, avoiding manual iteration. When design requirements change, there is no need to re-simulate, resulting in good design reusability and significantly shortening the overall design cycle.
[0006] To achieve the above objectives, this application provides the following solution: On the one hand, the present invention provides an artificial intelligence-aided design and performance prediction method for radio frequency coaxial connectors, specifically including the following steps: S1. Establish a parametric model of the RF coaxial connector, set the scanning range of key dimension parameters in the parametric model and perform parameter scanning to generate corresponding electrical performance simulation data. S2. Based on the key dimension parameters and their corresponding electrical performance simulation data, construct an ANN model and train it to obtain a trained ANN model, which is used to characterize the mapping relationship between the key dimension parameters and their corresponding electrical performance simulation data. S3. Obtain the target electrical performance and select the initial values of the key size parameters. Input the initial values of the key size parameters into the trained ANN model to obtain the initial electrical performance prediction value. If the deviation between the initial electrical performance prediction value and the target electrical performance does not meet the preset requirements, then execute S4. S4. Call multiple optimization algorithms and execute a reverse iterative optimization method with the target electrical performance as the objective. In each iteration, the current key dimension parameters are input into the trained ANN model to obtain the predicted electrical performance value. The prediction loss function is called to calculate the deviation between the predicted electrical performance value and the target electrical performance. With minimizing the deviation as the optimization objective, iterative search is performed within the scanning range of the key dimension parameters, and the optimal key dimension parameters that make the deviation meet the preset requirements are output.
[0007] In some specific implementations, the following step is included before step S2: The electrical performance simulation data is interpolated and converted into a serialized archive file format for storage to obtain a training data file. The training data file is then used to train the ANN model.
[0008] In some specific implementations, the ANN model includes an input layer, a multi-level neural network layer, and an output layer. The number of neurons in the input layer corresponds to the number of key dimensional parameters of the RF coaxial connector, and the number of neurons in the output layer corresponds to the number of frequency points of the electrical performance curve output by the ANN model.
[0009] In some specific implementation schemes, the specific process of step S2 is as follows: The key dimension parameters are linearly normalized and mapped to a preset range to obtain normalized key dimension parameters. The activation function is applied to the neurons of each layer of the ANN model to perform nonlinear processing, so that the ANN model learns the nonlinear relationship between normalized key size parameters and electrical performance simulation data; The ANN model is trained using electrical performance simulation data, and the trained ANN model is obtained.
[0010] In some specific implementations, applying activation functions to perform nonlinear processing on each layer of neurons in the ANN model includes: The sigmoid activation function is applied to each neuron in the input layer of the ANN model. The Leaky ReLU activation function is applied to process the neurons in each layer of the ANN model.
[0011] In some specific implementation schemes, the training process of the ANN model is as follows: Define the training loss function, set the number of iterations, and use the key size parameters and their corresponding electrical performance simulation data as the training data file, dividing the training data file into multiple training batches; In multiple training batches, an optimizer is used to iteratively train the ANN model to obtain the training results corresponding to each training batch. The training results for each training batch are compared with their corresponding training data files to verify the training accuracy of the ANN model. Determine if the training accuracy meets the preset threshold. If it does, stop training and output the currently trained ANN model. Otherwise, adjust the training parameters or continue iterating until the set number of iterations is reached, and output the final trained ANN model.
[0012] In some specific implementations, the prediction loss function also includes the step of using a one-sided penalty mechanism to retain only the portion of the predicted electrical performance that is inferior to the target electrical performance for weighted calculation, while setting the remaining values to 0.
[0013] In some specific implementations, the reverse iterative optimization method performs the following operations in each iteration: S41. Call the target preprocessing function to generate a frequency index based on the target electrical performance; S42. Input the current key dimension parameters into the trained ANN model to obtain the corresponding electrical performance prediction values, which include the S-parameter array. S43. Call the array truncation function, pass the S-parameter array to the array truncation function, use the frequency index generated by the target preprocessing function to truncate the subarray from zero frequency to the frequency index, and obtain the truncated electrical performance prediction value. S44. Call the prediction loss function to calculate the deviation between the truncated predicted electrical performance and the target electrical performance. S45. Based on the deviation, call multiple optimization algorithms in sequence, and start the optimization iteration with the same initial value and boundary conditions respectively; S46. During the optimization iteration process, repeatedly call the trained ANN model and the prediction loss function, repeat steps S42-S45, and continuously update the key size parameters until the maximum number of iterations is reached or the deviation between adjacent iteration results is less than the preset error, and then output the optimal key size parameters.
[0014] In some specific implementations, the target electrical performance includes the target operating frequency band and the target standing wave ratio (VSWR). The target preprocessing function performs the following operations: Based on the target operating frequency band, the position of the target operating frequency band in the discrete spectrum is determined by the argmin operator, and a frequency point index is generated. Convert the target standing wave VSWR to the target return loss S. 11 : ; Among them, the frequency index is used as the boundary of the array truncation function to truncate the S-parameter array and obtain the predicted electrical performance value after truncation within the target operating frequency band; The target return loss is used to calculate the deviation between the target loss and the truncated electrical performance prediction value, which is then input into the prediction loss function.
[0015] In some specific implementations, multiple optimization algorithms, including at least two of the Nelder-Mead algorithm, Powell algorithm, Differential Evolution algorithm, and COBYLA algorithm, are used to select the optimal key size parameters by running them in parallel and comparing the predicted loss function values of the optimal solutions returned by each optimization algorithm.
[0016] The beneficial effects of this invention are as follows: This invention utilizes a pre-trained ANN model, fully integrating parametric modeling and ANN model training. When the design requirements of the RF coaxial connector change, only the new target electrical performance (target operating frequency band and target VSWR) needs to be input into the interactive interface. Then, through the introduction of various optimization algorithms, the optimal balance can be automatically found among multi-dimensional indicators, overcoming the limitations of traditional manual iterative optimization. It also supports the automatic export of data files containing frequency correspondences, facilitating subsequent verification. The ANN model can automatically generate the optimal key dimension parameters that meet the new design requirements without the need for re-scanning parameters, automating manual iterative optimization and significantly shortening the design cycle of the RF coaxial connector. Attached Figure Description
[0017] Figure 1 A flowchart of an AI-aided design and performance prediction method for an RF coaxial connector provided in an embodiment of the present invention; Figure 2 This is a structural model diagram of an RF coaxial connector provided in an embodiment of the present invention; Figure 3 Based on Figure 2 Key dimensional parameters are labeled in the structural model diagram of the RF coaxial connector. Figure 4 A schematic diagram of an ANN model provided in an embodiment of the present invention; Figure 5 The standing wave ratio (SWR) curves are shown for comparing the training results of the ANN model provided in this embodiment with the target data. Figure 6 The curve showing the insertion loss of the ANN model training results provided in this embodiment of the invention compared with the original data. Figure 7 A schematic diagram of the interactive interface provided in an embodiment of the present invention; Figure 8 The predicted standing wave ratio (SWR) curve of the RF coaxial connector with optimal critical dimensions provided for embodiments of the present invention; Figure 9 This is the predicted insertion loss curve of the RF coaxial connector with optimal critical dimensions provided in the embodiments of the present invention; Numbering on the map: 1-Inner conductor part, 2-Outer conductor part, 3-Air filling part, 4-Insulator, 5-Slotted part of female connector. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the invention.
[0020] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.
[0021] Furthermore, for clarity and brevity, descriptions of well-known structures, functions, and configurations may have been omitted. Those skilled in the art will recognize that various changes and modifications can be made to the examples described herein without departing from the spirit and scope of this disclosure.
[0022] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.
[0023] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.
[0024] Example 1 like Figure 1 As shown, this embodiment provides an AI-aided design and performance prediction method for RF coaxial connectors, specifically including the following steps: S1. Establish a parametric model of the RF coaxial connector, set the scanning range of key dimension parameters in the parametric model and perform parameter scanning to generate corresponding electrical performance simulation data. To facilitate efficient reading of the subsequent ANN model, the electrical performance simulation data needs to be preprocessed, interpolated, and converted into a serialized archive file format for storage to obtain the training data file. This training data file is then used to train the ANN model. Interpolation methods, such as cubic spline interpolation, can be used for each interval... x i , x i+1 On the above, cubic spline interpolation function S i ( x ) is a cubic polynomial, that is: S i ( x )= a i + b i ( x-x i ) +c i ( x-x i ) 2 +d i ( x-x i ) 3 ,in i =0,1,…, n -1, x i , x i+1 Indicates two adjacent (the first) i and i +1) Electrical performance simulation data, where x represents the electrical performance simulation data obtained after interpolation. ai , b i , c i and d i is a coefficient.
[0025] S2. Based on the key dimension parameters and their corresponding electrical performance simulation data, construct an ANN model and train it to obtain a trained ANN model, which is used to characterize the mapping relationship between the key dimension parameters and their corresponding electrical performance simulation data. In this embodiment, the ANN model used is an artificial neural network model. The ANN can establish a mapping relationship between the structural parameters and electrical performance indicators of the RF coaxial connector, thereby enabling rapid assisted design of the RF coaxial connector and prediction of its electrical performance. Specifically, the ANN model includes an input layer, a multi-level neural network layer, and an output layer. The number of neurons in the input layer corresponds to the number of key dimensional parameters of the RF coaxial connector, and the number of neurons in the output layer corresponds to the number of frequency points of the electrical performance curve output by the ANN model.
[0026] Specifically, the process of step S2 is as follows: The key dimension parameters are linearly normalized and mapped to a preset range to obtain normalized key dimension parameters. The activation function is applied to perform nonlinear processing on each neuron of the ANN model. In some implementations, the Sigmoid activation function is applied to process each neuron in the input layer of the ANN model, and the Leaky ReLU activation function is applied to process the neurons in each level of the neural network of the ANN model; so that the ANN model learns the nonlinear relationship between normalized key size parameters and electrical performance simulation data. The ANN model is trained using electrical performance simulation data, and the trained ANN model is obtained.
[0027] Specifically, the training process of an ANN model is as follows: Define the training loss function, set the number of iterations, and use the key size parameters and their corresponding electrical performance simulation data as the training data file, dividing the training data file into multiple training batches; In multiple training batches, an optimizer is used to iteratively train the ANN model to obtain the training results corresponding to each training batch. The training results for each training batch are compared with their corresponding training data files to verify the training accuracy of the ANN model. Determine if the training accuracy meets the preset threshold. If it does, stop training and output the currently trained ANN model. Otherwise, adjust the training parameters or continue iterating until the set number of iterations is reached, and output the final trained ANN model.
[0028] S3. Obtain the target electrical performance and select the initial values of the key size parameters. Input the initial values of the key size parameters into the trained ANN model to obtain the initial electrical performance prediction value. If the deviation between the initial electrical performance prediction value and the target electrical performance does not meet the preset requirements, then execute S4. In order to optimize the direction of electrical performance, the prediction loss function includes the following steps when calculating the deviation between the predicted electrical performance and the target electrical performance: adopting a one-sided penalty mechanism, only retaining the part of the predicted electrical performance that is worse than the target electrical performance for weighted calculation, and setting the rest of the values to 0.
[0029] S4. Call multiple optimization algorithms and execute a reverse iterative optimization method with the target electrical performance as the objective. In each iteration, the current key dimension parameters are input into the trained ANN model to obtain the predicted electrical performance value. The prediction loss function is called to calculate the deviation between the predicted electrical performance value and the target electrical performance. With minimizing the deviation as the optimization objective, iterative search is performed within the scanning range of the key dimension parameters, and the optimal key dimension parameters that make the deviation meet the preset requirements are output.
[0030] Multiple optimization algorithms, including at least two of Nelder-Mead, Powell, Differential Evolution, and COBYLA, are used to select the optimal key size parameters by running them in parallel and comparing the predicted loss function values of the optimal solutions returned by each optimization algorithm.
[0031] Specifically, each iteration of the inverse iterative optimization method performs the following operations: S41. Call the target preprocessing function to generate a frequency index based on the target electrical performance; S42. Input the current key dimension parameters into the trained ANN model to obtain the corresponding electrical performance prediction values, which include the S-parameter array. S43. Call the array truncation function, pass the S-parameter array to the array truncation function, use the frequency index generated by the target preprocessing function to truncate the subarray from zero frequency to the frequency index, and obtain the truncated electrical performance prediction value. S44. Call the prediction loss function to calculate the deviation between the truncated predicted electrical performance and the target electrical performance. S45. Based on the deviation, call multiple optimization algorithms in sequence, and start the optimization iteration with the same initial value and boundary conditions respectively; S46. During the optimization iteration process, repeatedly call the trained ANN model and the prediction loss function, repeat steps S42-S45, and continuously update the key size parameters until the maximum number of iterations is reached or the deviation between adjacent iteration results is less than the preset error, and then output the optimal key size parameters.
[0032] In some specific implementations, the target electrical performance includes the target operating frequency band and the target standing wave ratio (VSWR). The target preprocessing function performs the following operations: Based on the target operating frequency band, the position of the target operating frequency band in the discrete spectrum is determined by the argmin operator, and a frequency point index is generated. Convert the target standing wave VSWR to the target return loss S. 11 : ; Among them, the frequency index is used as the boundary of the array truncation function to truncate the S-parameter array and obtain the predicted electrical performance value after truncation within the target operating frequency band; The target return loss is used to calculate the deviation between the target loss and the truncated electrical performance prediction value, which is then input into the prediction loss function.
[0033] To better illustrate the implementation process of this application, specific examples are provided below: 1. Establish a parametric model of the RF coaxial connector, set the scanning range of key dimensional parameters and perform parameter scanning to obtain electrical performance simulation data of the RF coaxial connector; Among them, establishing a parametric model of the RF coaxial connector includes parametric modeling of the RF coaxial connector in three-dimensional electromagnetic simulation software and parametric processing of key dimensional parameters (hereinafter referred to as "key dimensional parameters") that affect impedance matching.
[0034] Specifically, a parametric model of the RF coaxial connector can be created in 3D electromagnetic simulation software, such as... Figure 2 As shown, this embodiment provides a parametrically modeled RF coaxial connector structure model. The RF coaxial connector structure model includes an inner conductor portion 1 and an outer conductor portion 2 sleeved outside the inner conductor portion 1. There is a gap between the inner conductor portion 1 and the outer conductor portion 2, which serves as an air-filling portion 3. A female head slotted portion 5 is provided at the lower end of the inner conductor portion 1, and an insulator 4 is provided between the inner conductor portion 1 and the female head slotted portion 5. The insulator 4 is made of Teflon with a dielectric constant of 2.02. It should be noted that this application... Figure 2 The RF coaxial connector structural model given is for illustrative purposes only and does not include the external mechanical structure.
[0035] And regarding Figure 2 The key dimensional parameters of the RF coaxial connector structure model are as follows: Figure 3As shown, key dimensional parameters include, but are not limited to: the inner diameter of the coaxial portion of the air-filled section 3. d 1 and outer diameter D 1. Inner diameter of the coaxial portion of the filler insulator 4 d 2 and outer diameter D 2, and the slot width of the slotted portion 5 of the female head. d and thickness s wait.
[0036] Furthermore, the scanning range of key dimensional parameters of the RF coaxial connector is set in the 3D electromagnetic simulation software. Within the scanning range, different key dimensional parameters of the RF coaxial connector are scanned to generate electrical performance simulation data corresponding to each set of key dimensional parameters. In this embodiment, the electrical performance simulation data includes standing wave simulation data and insertion loss simulation data. After the scanning process is completed, a large amount of electrical performance simulation data of the RF coaxial connector is obtained. By running a script, the electrical performance simulation data is automatically extracted from the 3D electromagnetic simulation software, and the automated script reads and exports the electrical performance simulation data in batches from the 3D electromagnetic simulation software. When reading the electrical performance simulation data, the automated script preprocesses the electrical performance simulation data, including: performing cubic spline interpolation on the electrical performance simulation data to obtain uniformly distributed frequency sampling points within the target operating frequency band; converting the data into a serialized archive file format for storage to obtain training data files, which facilitates efficient reading and training of the subsequent ANN model.
[0037] 2. Construct an ANN model and define neural network layers. Use simulation data of the electrical performance of RF coaxial connectors for training to establish the mapping relationship between key dimensional parameters and electrical performance. Understandably, this step aims to construct, train, and validate an ANN model, establishing a mapping relationship between key dimensional parameters and electrical performance in this embodiment. The ANN model constructed in this embodiment is as follows: Figure 4 As shown, the ANN model includes an input layer, multi-level neural network layers, and an output layer. The number of neurons in the input layer of the ANN model corresponds to the number of key size parameters in the embodiment. Assuming that the number of key size parameters selected in this application is 6, the 6 key size parameters are the inner diameter of the coaxial portion of the air-filled portion 3. d 1 and outer diameter D 1. Inner diameter of the coaxial portion of the filler insulator 4 d 2 and outer diameter D 2, and the slot width of the slotted portion 5 of the female head. d and thickness sThe input layer of the ANN model has 6 neurons. The number of neurons in the output layer corresponds to the number of frequency points of the electrical performance curves output by the ANN model. For example, if the required number of frequency points for the ANN output is 101, then the output layer of the ANN model has 101 neurons, and the output layer outputs standing waves and insertion losses at different frequencies. The intermediate layers are three-level neural network layers with 101, 201, and 401 neurons respectively. These neuron numbers were determined by comparing the training accuracy of ANN models with different network structures.
[0038] Before training the ANN model using electrical performance simulation data of the RF coaxial connector, the six key dimensional parameters were first linearly normalized and mapped to the dimensionless interval [0, 1]. Then, the Sigmoid activation function was applied to each neuron in the input layer of the ANN model to enhance the nonlinearity near the boundary. Finally, the Leaky ReLU activation function was applied to the neurons in each of the remaining neural network layers to establish the nonlinear relationship between the key dimensional parameters and electrical performance of the embodiment, and to ensure that the neurons can be stably transmitted to the deep network during training, thereby improving training stability.
[0039] Specifically, the mathematical formula for linear normalization is: a = ,in a These are the normalized dimensionless parameters (critical dimension parameters). a ∈[0,1]; θ These are the original values for the key dimensional parameters. θ ∈[ θ min , θ max ]; θ min This represents the lower limit value of the critical dimensional parameter. θ max This represents the upper limit of the critical dimension parameter, and 0 < 0. θ min < θ max .
[0040] Next, the ANN model is trained using electrical performance simulation data. The serialized archive file obtained above is used as the training data file. The number of iterations is set (assumed to be 1500), and the training data file is divided into multiple training batches (for example, the training data file is divided into 450 training batches of 10). A training loss function is used, where the mean squared error (MSE) function can be adopted to quantify the deviation between the training result and the training data file. Its mathematical formula is: .in,N This refers to the number of samples, i.e., the number of training data files used in a single training session. y i These are the actual values, i.e., the values in the training data file; These are the values from the training results. An optimizer (such as the AdamW optimizer) is used to iteratively train the ANN model. The training results of the ANN model are compared with the training data file to verify and evaluate the training accuracy of the ANN model. The standing wave ratio comparison curve is shown below. Figure 5 As shown in the figure (horizontal axis is frequency (GHz), vertical axis is standing wave ratio), the insertion loss comparison curve is as follows: Figure 6 As shown in the figure (horizontal axis represents frequency (GHz), and vertical axis represents insertion loss (dB)), the comparison curves show that the ANN model training results fit the training data file well. The mapping relationship between the key dimensional parameters and electrical performance of the RF coaxial connector established by the ANN model can accurately predict the standing wave ratio and insertion loss of the RF coaxial connector.
[0041] In addition, an efficient logging system was built during training to record the entire training process. Based on the training requirements of the ANN model for the RF coaxial connector, a one-time global configuration was completed using `logging.basicConfig`, writing key node information during training to a text file in a fixed path, recording complete training information for the RF coaxial connector. Through script execution, the logging system automatically records batch information, the current S-parameter type, and its final convergence value, constituting complete training information for the RF coaxial connector.
[0042] In addition, global parameter settings for the ANN model are required. The storage path for the ANN model dictionary set is set; this dictionary set stores the network weights, neuron biases, and optimizer states of the trained ANN model. The storage path for the normalized dictionary file is also set; this normalized dictionary file stores the boundary values, normalized values, and frequency sampling point information of key size parameters. After setting the storage paths, a path index dictionary is created based on these paths to ensure that the ANN model can quickly retrieve the required data when predicting electrical performance.
[0043] Understandably, by establishing a mapping relationship between key dimensional parameters and electrical performance through an ANN model, the ANN model learns the correspondence between key dimensional parameters and electrical performance of RF coaxial connectors, establishes a nonlinear mapping relationship between key dimensional parameters and electrical performance indicators, and thus forms the positive predictive capability of the ANN model from key dimensional parameters to electrical performance of RF coaxial connectors.
[0044] 3. Configure multiple optimization algorithms and define the core functional functions related to the prediction of the electrical performance of RF coaxial connectors using the ANN model; Multiple optimization algorithms are configured, including defining a main optimization function and introducing various optimization algorithms, to achieve automatic optimization of key dimensional parameters of the RF coaxial connector and selection of the optimal solution. This embodiment employs four optimization algorithms: Nelder-Mead, Powell, Differential Evolution, and COBYLA algorithms to achieve automatic optimization of key dimensional parameters and selection of the optimal solution.
[0045] The core functionalities are defined, including a prediction function, a prediction loss function, and a plotting function. The prediction function receives the normalized key dimensional parameters of the RF coaxial connector and feeds them into the trained ANN model to calculate the predicted standing wave ratio (SWR) and insertion loss at each frequency point. The prediction loss function is used during the optimization phase to calculate the deviation between the electrical performance predicted by the ANN model and the target electrical performance (specifically, to calculate the deviation between the SWR and insertion loss predicted by the ANN model and the target SWR and insertion loss). The plotting function is used to visualize the SWR and insertion loss curves corresponding to the optimal structural parameters after optimization. The prediction loss function... Loss The mathematical formula is: Loss= ,in For weighted norm; This is the predicted value for return loss. S 11 The target value for return loss; This is the predicted value of insertion loss. S 21 The target value for insertion loss; w As the weight, it is generally set to 10. Loss During the function's operation, a one-sided penalty mechanism is used for the deviation between the predicted value and the target value, that is, only the positive deviation value is retained for weighted calculation, and the rest are set to 0.
[0046] 4. Based on the target electrical performance, the ANN model and various optimization algorithms are invoked, and the key dimensional parameters of the RF coaxial connector are optimized using a reverse optimization method with electrical performance indicators as the target. Understandably, this step aims to call the trained ANN model and adopt a reverse optimization method with electrical performance indicators as the target. Through the joint operation of core function and multiple optimization algorithms, it realizes automatic optimization from target standing wave and target insertion loss to key size parameters of the embodiment. This is different from the traditional forward design method of RF coaxial connectors, which relies on experience and iterative adjustment of size parameters.
[0047] Furthermore, to facilitate the input of target electrical performance, this embodiment obtains the target electrical performance by constructing an interactive interface that allows users to input the target operating frequency, target VSWR, and target insertion loss of the RF coaxial connector, such as... Figure 7 As shown, the interactive interface includes a cutoff frequency (GHz) input control, a target standing wave (VSWR) input control, a target insertion loss (dB) input control, and a prediction button. For example, when the user sets the cutoff frequency to 18 GHz (the target operating frequency band is from DC to the cutoff frequency, i.e., the target operating frequency band is DC-18 GHz), the target VSWR is 1.05, and the target insertion loss is 0.04 dB, after the input is completed, clicking the prediction button will execute the following actions.
[0048] Then, a target preprocessing function is defined to preprocess the input target electrical performance. This function consists of two parts. The first part generates the frequency index. k : via the argmin operator k= argmin| f-f target |; Determine the target operating frequency band f target In discrete spectrum f Frequency index in k Secondly, it addresses the target electrical performance by processing the target standing wave ratio (VSWR), converting the input VSWR into return loss S. 11 The conversion formula is: .
[0049] Define an array truncation function, using the frequency index generated by the target preprocessing function. k Using the boundary as a limit, the S-parameter array is truncated, retaining the S-parameter array within the range from zero frequency to the index point.
[0050] Next, the trained ANN model is invoked, taking the target electrical performance of the RF coaxial connector as the optimization direction. Multiple optimization algorithms are then used to automatically search for the optimal solution that satisfies the target electrical performance within the space of critical size parameters, achieving reverse design from electrical performance indicators to critical size parameters. The ANN model is invoked, and initial values for the critical size parameters are selected from the normalized critical size parameter interval [0,1]. In this embodiment, the initial value is set to 0.5. These initial values are input into the ANN model to calculate the initial VSWR prediction and insertion loss prediction values of the RF coaxial connector at each frequency point. The predicted values are compared with the target values, and the deviation between the two is calculated using a prediction loss function, providing the optimization direction for the optimization algorithm.
[0051] Based on the deviation value, the four optimization algorithms set in this embodiment are sequentially called, each starting the optimization iteration with the same initial value and boundary conditions. During the optimization iteration process, the prediction function and prediction loss function are repeatedly called, continuously updating the key dimension parameters until the maximum number of iterations is reached or the deviation between adjacent iteration results is less than a preset threshold, at which point the optimal solution is returned. The boundary conditions are the normalized range of values for the scan range of the key dimension parameters, determined after the scan range is normalized. Multiple optimization algorithms run independently and in parallel. By comparing the prediction loss function values of the optimal solutions returned by each algorithm, the optimal combination of structural parameters that makes the RF coaxial connector meet the target is automatically selected, achieving reverse optimization design with electrical performance indicators as the objective.
[0052] 5. Based on the optimal key dimension parameters, predict the electrical performance of the RF coaxial connector, generate the corresponding curves, and complete the auxiliary design of the RF coaxial connector.
[0053] Understandably, predicting the electrical performance of an RF coaxial connector based on optimal critical dimension parameters involves inputting the optimized critical dimension parameters of the RF coaxial connector into a trained ANN model to obtain the predicted electrical performance value of the RF coaxial connector under those critical dimension parameters, thereby verifying whether the design results meet the performance requirements. Furthermore, plotting functions are called to generate the corresponding standing wave ratio (SWR) curves and insertion loss curves as a function of frequency, completing the auxiliary design of this embodiment. In this embodiment, the final generated SWR curve is as follows: Figure 8 As shown, at DC-18 GHz, the standing wave ratio (SWR) is less than or equal to 1.05, satisfying the target SWR requirement; the final generated insertion loss curve is shown below. Figure 9 As shown, at DC-18 GHz, the insertion loss is less than or equal to 0.04 dB, which meets the target insertion loss of the input.
[0054] It is understandable that, compared to traditional design methods for RF coaxial connectors, the AI-aided design and performance prediction method for RF coaxial connectors described in this invention has significant advantages. Traditional methods rely on repeated iterations of electromagnetic simulation, wasting a large amount of computer resources, and design changes require re-scanning parameters. The method proposed in this invention first performs parametric modeling of the RF coaxial connector, then performs parameter scanning to obtain a large amount of simulation data of the RF coaxial connector structure, and then trains the ANN model to establish the mapping relationship between the structural parameters and electrical performance of the RF coaxial connector. A single prediction takes only seconds. Finally, by setting optimization objectives and utilizing core functions and optimization algorithms, automatic optimization is performed. The optimal structural parameters of the RF coaxial connector are obtained directly through a reverse optimization method targeting electrical performance indicators. Based on the optimal structural parameters, performance prediction is performed, completing the AI-aided design of the RF coaxial connector. This avoids manual iteration, eliminates the need for re-simulation when design requirements change, has good design reusability, and significantly shortens the overall design cycle.
[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.
Claims
1. An AI-aided design and performance prediction method for radio frequency coaxial connectors, characterized in that, Specifically, the following steps are included: S1. Establish a parametric model of the RF coaxial connector, set the scanning range of key dimension parameters in the parametric model and perform parameter scanning to generate corresponding electrical performance simulation data. S2. Based on the key dimension parameters and their corresponding electrical performance simulation data, construct an ANN model and train it to obtain a trained ANN model, which is used to characterize the mapping relationship between the key dimension parameters and their corresponding electrical performance simulation data. S3. Obtain the target electrical performance and select the initial values of the key size parameters. Input the initial values of the key size parameters into the trained ANN model to obtain the initial electrical performance prediction value. If the deviation between the initial electrical performance prediction value and the target electrical performance does not meet the preset requirements, then execute S4. S4. Call multiple optimization algorithms and execute a reverse iterative optimization method with the target electrical performance as the objective. In each iteration, the current key dimension parameters are input into the trained ANN model to obtain the predicted electrical performance value. The prediction loss function is called to calculate the deviation between the predicted electrical performance value and the target electrical performance. With minimizing the deviation as the optimization objective, iterative search is performed within the scanning range of the key dimension parameters, and the optimal key dimension parameters that make the deviation meet the preset requirements are output.
2. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 1, characterized in that, The steps preceding step S2 include: The electrical performance simulation data is interpolated and converted into a serialized archive file format for storage to obtain a training data file. The training data file is then used to train the ANN model.
3. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 1, characterized in that, The ANN model consists of an input layer, multiple neural network layers, and an output layer. The number of neurons in the input layer corresponds to the number of key size parameters of the RF coaxial connector, and the number of neurons in the output layer corresponds to the number of frequency points of the electrical performance curve output by the ANN model.
4. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 3, characterized in that, The specific process of step S2 is as follows: The key dimension parameters are linearly normalized and mapped to a preset range to obtain normalized key dimension parameters. The activation function is applied to the neurons of each layer of the ANN model to perform nonlinear processing, so that the ANN model learns the nonlinear relationship between normalized key size parameters and electrical performance simulation data; The ANN model is trained using electrical performance simulation data, and the trained ANN model is obtained.
5. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 4, characterized in that, Applying activation functions to perform nonlinear processing on neurons in each layer of an ANN model includes: The sigmoid activation function is applied to each neuron in the input layer of the ANN model. The Leaky ReLU activation function is applied to process the neurons in each layer of the ANN model.
6. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 1, characterized in that, The training process of an ANN model is as follows: Define the training loss function, set the number of iterations, and use the key size parameters and their corresponding electrical performance simulation data as the training data file, dividing the training data file into multiple training batches; In multiple training batches, an optimizer is used to iteratively train the ANN model to obtain the training results corresponding to each training batch. The training results for each training batch are compared with their corresponding training data files to verify the training accuracy of the ANN model. Determine if the training accuracy meets the preset threshold. If it does, stop training and output the currently trained ANN model. Otherwise, adjust the training parameters or continue iterating until the set number of iterations is reached, and output the final trained ANN model.
7. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 1, characterized in that, The prediction loss function also includes the following steps when calculating the deviation between the predicted electrical performance and the target electrical performance: using a one-sided penalty mechanism, only retaining the part of the predicted electrical performance that is inferior to the target electrical performance for weighted calculation, and setting the rest of the values to 0.
8. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 1, characterized in that, In the reverse iterative optimization method, the following operations are performed in each iteration: S41. Call the target preprocessing function to generate a frequency index based on the target electrical performance; S42. Input the current key dimension parameters into the trained ANN model to obtain the corresponding electrical performance prediction values, which include the S-parameter array. S43. Call the array truncation function, pass the S-parameter array to the array truncation function, use the frequency index generated by the target preprocessing function to truncate the subarray from zero frequency to the frequency index, and obtain the truncated electrical performance prediction value. S44. Call the prediction loss function to calculate the deviation between the truncated predicted electrical performance and the target electrical performance. S45. Based on the deviation, call multiple optimization algorithms in sequence, and start the optimization iteration with the same initial value and boundary conditions respectively; S46. During the optimization iteration process, repeatedly call the trained ANN model and the prediction loss function, repeat steps S42-S45, and continuously update the key size parameters until the maximum number of iterations is reached or the deviation between adjacent iteration results is less than the preset error, and then output the optimal key size parameters.
9. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 8, characterized in that, The target electrical performance includes the target operating frequency band and the target standing wave ratio (VSWR). The target preprocessing function performs the following operations: Based on the target operating frequency band, the position of the target operating frequency band in the discrete spectrum is determined by the argmin operator, and a frequency point index is generated. Convert the target standing wave VSWR to the target return loss S. 11 : ； Among them, the frequency index is used as the boundary of the array truncation function to truncate the S-parameter array and obtain the predicted electrical performance value after truncation within the target operating frequency band; The target return loss is used to calculate the deviation between the target loss and the truncated electrical performance prediction value, which is then input into the prediction loss function.
10. The method for AI-aided design and performance prediction of an RF coaxial connector according to claim 1, characterized in that, Multiple optimization algorithms, including at least two of Nelder-Mead, Powell, Differential Evolution, and COBYLA, are used to select the optimal key size parameters by running them in parallel and comparing the predicted loss function values of the optimal solutions returned by each optimization algorithm.