A patch antenna parameter optimization method based on feature analysis

Abstract

This invention relates to the fields of antenna design and electromagnetic simulation technology, specifically a patch antenna parameter optimization method based on feature analysis. The method includes: a physical benchmark construction step, acquiring geometric and material data and using a feature mode engine for source-free solving; a sensitivity coupling step, screening dominant modes and applying geometric perturbations to establish a parameter-mode sensitivity matrix; an objective function construction step, combining the mode quality factor and the sensitivity matrix to construct a composite objective function including a sensitivity penalty term; and a global optimization verification step, iteratively obtaining the optimal parameter combination and generating a robustness score report based on process tolerances. This invention achieves a shift from numerical optimization to physically robust optimization, significantly reducing yield losses caused by manufacturing errors.

Description

Technical Field

[0001] This invention relates to the fields of antenna design and electromagnetic simulation technology, specifically a method for optimizing patch antenna parameters based on feature analysis. Background Technology

[0002] In the current design and mass production process of patch antennas, traditional solutions generally rely on electromagnetic simulation software to perform black-box numerical optimization of apparent indicators such as S-parameters; existing design modes are usually based on ideal geometric models for parameter iteration, failing to incorporate PCB etching errors and substrate material property deviations that inevitably exist in the actual production environment into the optimization closed loop.

[0003] Although such solutions can quickly obtain design solutions that meet the specifications in a simulation environment, the lack of quantitative evaluation of the intrinsic stability of the physical structure makes the final antenna solution extremely sensitive to manufacturing tolerances. This disconnect between design and manufacturing leads to poor consistency and low yield in the mass production stage. Furthermore, since robustness defects can only be discovered after trial production, it significantly increases the cost of hardware trial and error and prolongs the product development cycle.

[0004] Therefore, how to build an optimization mechanism that integrates physical sensitivity analysis at the design source to output a highly robust antenna design scheme that can effectively resist manufacturing errors has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for optimizing patch antenna parameters based on feature analysis. Specifically, the technical solution of this invention includes: Step 1: Obtain the initial geometric model data, substrate material parameters, and preset process tolerance range data of the patch antenna to be optimized, and construct the antenna electromagnetic simulation model; based on the initial geometric model data, use the characteristic mode analysis engine to perform source-free solution, and extract the eigenvalues ​​and characteristic current distribution data of the first N characteristic modes; Step 2: Based on the eigenvalues, calculate the modal significance of each eigenmode and select the dominant mode according to the preset significance threshold; construct a sensitivity analysis model for the dominant mode, and calculate the partial derivatives of the eigenvalues ​​with respect to the geometric parameters by applying perturbation to the initial geometric model data, and establish the parameter-modal sensitivity matrix. Step 3: Calculate the modal quality factor of each feature mode based on the eigenvalues, and construct a composite objective function containing frequency alignment term, saliency maximization term and sensitivity penalty term by combining modal saliency and parameter-modal sensitivity matrix; Step 4: Use a multi-objective optimization algorithm to iterate and optimize the composite objective function in the parameter space to obtain the globally optimal combination of geometric parameters. Based on the response distribution of the globally optimal combination of geometric parameters within the process tolerance range, generate a robustness score report to output an antenna design scheme that can resist processing errors.

[0006] Preferably, step one includes: S11. Collect the length, width, feed point position, and chamfer size of the patch antenna to be optimized as the initial geometric parameter vector; collect the dielectric constant, loss tangent, and thickness of the dielectric substrate as material property data. S12. Collect the etching error range and dielectric constant deviation range existing in the production process, and define them as the process tolerance vector; S13. Import the initial geometric parameter vector and material property data into the electromagnetic simulation solver, establish a discretized mesh model, solve the generalized eigenvalue equation, and output the eigenvalues ​​of the first N orders and the corresponding characteristic current distribution.

[0007] Preferably, step two includes: S21. Based on the eigenvalues, calculate the modal significance of the nth eigenmode; S22. Compare the modality saliency with a preset saliency threshold. When it is greater than the saliency threshold, mark the corresponding feature mode as the dominant mode. S23. For each dominant mode, based on the initial geometric parameter vector, the components in the process tolerance vector are superimposed to perform geometric perturbation to generate perturbed eigenvalues. S24. Calculate the rate of change of eigenvalues ​​before and after the perturbation, obtain the partial derivatives of the eigenvalues ​​with respect to each geometric parameter, and construct a parameter-modal sensitivity matrix from all the partial derivatives, where the matrix elements characterize the influence weight of the size change of a specific physical part on the resonant frequency of a specific mode.

[0008] Preferably, step three includes: S31. Based on the eigenvalues ​​and their derivatives with respect to frequency, calculate the modal quality factor of each characteristic mode; S32. Set a preset quality factor threshold, remove narrowband modes with a modal quality factor higher than the preset quality factor threshold, and retain broadband potential modes. S33. Construct a composite objective function, which consists of three weighted parts: the first part is the absolute value of the difference between the resonant frequency and the target frequency; the second part is the negative value of the modal significance; and the third part is the total sensitivity value calculated based on the parameter-modal sensitivity matrix.

[0009] Preferably, step four includes: S41. Initialize the optimization algorithm population, using the initial geometric parameter vector P as the starting point; S42. In each iteration, calculate the value of the composite objective function corresponding to the current parameter combination; S43. Determine whether the total sensitivity value is less than or equal to the preset robustness baseline. If yes, retain the parameter combination in the next generation population; otherwise, apply a penalty weight to the parameter combination so that it is eliminated in subsequent iterations. S44. When the number of iterations reaches the preset value or the composite objective function converges, output the combination of geometric parameters that minimizes the composite objective function as the globally optimal combination of geometric parameters.

[0010] Preferably, step four also includes: S45. Based on the globally optimal combination of geometric parameters and combined with the process tolerance vector, a virtual mass production sample set is generated using the Monte Carlo analysis method. S46. Calculate the proportion of samples in the virtual mass production sample set that meet the design indicators, and generate a robustness score report. S47. Based on the robustness rating report, classify and evaluate the antenna design scheme and output the classification results.

[0011] Preferably, in step S47, a first preset ratio threshold and a second preset ratio threshold are preset, wherein the first preset ratio threshold is greater than the second preset ratio threshold, and the graded evaluation includes: When the sample ratio is greater than or equal to the first preset ratio threshold, the design scheme is determined to be highly robust, and a direct mass production verification signal is generated. When the sample ratio is less than the first preset ratio threshold and greater than or equal to the second preset ratio threshold, the design scheme is determined to be conditionally available, and a process data package containing suggestions for improving PCB etching accuracy is generated. When the sample ratio is less than the second preset ratio threshold, the design scheme is determined to be highly sensitive and risky, and a warning signal is generated to prompt the antenna topology to be readjusted.

[0012] Preferably, the application of the parameter-modal sensitivity matrix also includes: Identify the geometric region corresponding to the element with the largest value in the parameter-modal sensitivity matrix and mark it as the sensitive source region; In the electromagnetic simulation model of the antenna, the sensitive source region is subjected to mesh refinement, or stricter constraints are imposed on the size parameters of the sensitive source region during the optimization process.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a physical benchmark model based on characteristic mode analysis and performs source-free solving. The system can identify the inherent resonant modes of the antenna structure, overcoming the limitations of traditional design processes that only focus on apparent indicators such as S-parameters. By establishing a digital mapping between physical parameters and electromagnetic characteristic modes, it realizes the transformation from black-box numerical optimization to white-box physical robust optimization, ensuring that the optimization process is based on the intrinsic stability of the structure, and providing a solid physical foundation for solving the problem of design schemes being sensitive to manufacturing tolerances. 2. This invention constructs a parameter-modal sensitivity matrix and introduces a composite objective function containing a sensitivity penalty term, which can accurately quantify the influence weight of the size change of a specific physical part on the modal resonant frequency. By combining process tolerance vector with perturbation analysis in the optimization iteration, the algorithm is forced to avoid those fragile modal regions with good radiation performance but extremely sensitive to size changes, guiding the search to converge to a broadband flat solution space that is insensitive to etching errors and dielectric constant deviations, thereby ensuring the manufacturability of the design from a mathematical perspective. 3. This invention implements an adaptive correction mechanism for the feed point during the global optimization process, and performs a deterministic search for the best matching point using the characteristic current amplitude distribution, thereby decoupling the geometric size optimization from the feed position matching. This mechanism effectively eliminates impedance mismatch noise caused by random walks of the feed position, prevents potential high-performance geometric individuals from being misjudged and eliminated, significantly improves the algorithm's ability to identify the quality of geometric parameters, and ensures that the final output of the globally optimal parameter combination has good impedance matching characteristics in physics. 4. This invention establishes a clear mapping relationship between tolerance and statistical distribution through virtual mass production verification and robustness classification evaluation based on Monte Carlo analysis. It can predict the mass production yield of the scheme under the normal distribution boundary at the simulation design stage. By generating a classification report containing direct mass production, process improvement suggestions or redesign warnings, yield problems that originally needed to be discovered in the trial production stage can be solved in advance, which significantly reduces the hardware trial and error cost caused by PCB etching errors and realizes deep collaboration between design and manufacturing. Attached Figure Description

[0014] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0016] Example 1: Please see Figure 1 A method for optimizing patch antenna parameters based on feature analysis, the specific steps of which include: Step 1: Obtain the initial geometric model data, substrate material parameters, and preset process tolerance range data of the patch antenna to be optimized, and construct the antenna electromagnetic simulation model; based on the initial geometric model data, use the characteristic mode analysis engine to perform source-free solution, and extract the eigenvalues ​​and characteristic current distribution data of the first N characteristic modes; Step 2: Based on the eigenvalues, calculate the modal significance of each eigenmode and select the dominant mode according to the preset significance threshold; construct a sensitivity analysis model for the dominant mode, and calculate the partial derivatives of the eigenvalues ​​with respect to the geometric parameters by applying perturbation to the initial geometric model data, and establish the parameter-modal sensitivity matrix. Step 3: Calculate the modal quality factor of each feature mode based on the eigenvalues, and construct a composite objective function containing frequency alignment term, saliency maximization term and sensitivity penalty term by combining modal saliency and parameter-modal sensitivity matrix; Step 4: Use a multi-objective optimization algorithm to iterate and optimize the composite objective function in the parameter space to obtain the globally optimal combination of geometric parameters. Based on the response distribution of the globally optimal combination of geometric parameters within the process tolerance range, generate a robustness score report to output an antenna design scheme that can resist processing errors.

[0017] This embodiment elaborates on the physical layer execution logic of the above-mentioned feature analysis-based patch antenna parameter optimization method, aiming to solve the technical problem that the existing antenna design process only focuses on apparent indicators such as S-parameters, resulting in the design scheme being sensitive to manufacturing tolerances and having low mass production yield. This method realizes the transformation from black-box numerical optimization to white-box physical robust optimization by constructing a physical benchmark model and a sensitivity coupling mechanism. The system executes step one, constructing a physical baseline model based on characteristic modes. During this process, the system acquires initial geometric model data of the patch antenna to be optimized, such as the copper plating shape and size, as well as substrate material parameters and preset process tolerance ranges. To ensure that subsequent frequency scanning and sensitivity analysis have clear computational boundaries, the system must simultaneously define and input the target operating frequency in this step. and frequency step , which serves as the core configuration parameter for the electromagnetic simulation model; The system constructs an electromagnetic simulation model of the antenna, specifically using an electromagnetic solution kernel based on the method of moments, configuring open radiation boundary conditions, and employing a characteristic mode analysis engine to perform a source-free solution with all physical excitation ports removed. This solution process aims to identify the inherent resonant modes of the antenna structure, which objectively exist regardless of the feed point location, thus providing a physical benchmark for subsequent analysis. The system executes step two, constructing a parameter-modal sensitivity analysis model. Based on this, the system calculates the modal significance of each characteristic mode based on eigenvalues ​​and filters out the dominant modes according to a preset significance threshold. For the dominant modes, the system constructs a sensitivity analysis model by applying perturbations to the initial geometric model data, calculating the partial derivatives of the eigenvalues ​​with respect to the geometric parameters, and establishing a parameter-modal sensitivity matrix. This matrix acts as a quantization tool for tolerance amplifiers, enabling accurate identification of vulnerable modes that have good radiation performance but are extremely sensitive to size changes. The system executes step three, constructing a robustness-oriented composite objective function. The system calculates the modal quality factor of each feature mode based on the eigenvalues, and constructs a composite objective function by combining modal saliency and parameter-modal sensitivity matrix. This function, by introducing a sensitivity penalty term, forces the optimization algorithm to avoid parameter regions sensitive to processing errors, guiding the search to converge to a wide and stable solution space. The system executes step four, which involves global optimization and robustness verification. A multi-objective optimization algorithm is used to iterate the optimization of the composite objective function in the parameter space to obtain the globally optimal combination of geometric parameters. Based on the response distribution of this combination within the process tolerance range, a robustness score report is generated. This embodiment introduces a physical layer optimization framework that couples characteristic mode analysis and sensitivity analysis, taking into account the intrinsic stability of the physical structure from the design source. The antenna design scheme output by this method not only has excellent performance in simulation, but also has extremely high mass production consistency, significantly reducing yield loss caused by PCB etching errors, and realizing deep collaboration between design and manufacturing.

[0018] Example 2: Step 1 includes: S11, collecting the length, width, feed point position, and chamfer size of the patch antenna to be optimized as the initial geometric parameter vector P; collecting the dielectric constant, loss tangent, and thickness of the dielectric substrate as material property data; S12. Collect the etching error range and dielectric constant deviation range existing in the production process, and define them as the process tolerance vector; S13. Import the initial geometric parameter vector P and material property data into the electromagnetic simulation solver, establish a discretized mesh model, solve the generalized eigenvalue equation, and output the eigenvalues ​​of the first N orders and the corresponding characteristic current distribution.

[0019] This embodiment is a further specification of the steps for constructing the physical baseline model in Embodiment 1, with a focus on the vectorization definition of data and the initialization of the solver; The system executes step S11 to collect vectorized geometric and material parameters. To facilitate algorithm processing, this embodiment divides the parameters of the patch antenna to be optimized into structural parameter vectors. With feed parameter vector Define the initial geometric parameter vector. It consists of structural parameters and power supply parameters, i.e. The calculation formula is as follows: in, The length of the patch antenna. The width of the patch antenna. These are the dimensions of the chip cut corner; all units are in mm. in, and The coordinates of the feed point in a Cartesian coordinate system with the geometric center of the patch as the origin are shown in mm. Simultaneously, the system acquires property data of the dielectric substrate, including the dielectric constant. Loss tangent and thickness Furthermore, in order to support the construction of the sensitivity matrix and the calculation of the objective function in subsequent steps, the system must also explicitly acquire the target operating frequency in the design specifications at this stage. For example, 1.575 GHz in the GPS L1 band, and frequency scan steps used for numerical differentiation calculations. It is recommended to take 0.1% to 0.5%, for example, 2MHz, and store both in the global configuration parameter set; The system executes step S12 to define the process tolerance vector; in order to simulate a real production environment, this embodiment defines the process tolerance vector. This vector contains the etching error range and dielectric constant deviation range that exist in the manufacturing process; the etching error range comes from the process capability specifications provided by the PCB manufacturer, and the dielectric constant deviation range comes from the datasheet of the board supplier. The system executes step S13 to solve the generalized eigenvalue equations; the above data is imported into the electromagnetic simulation solver to establish a discretized mesh model; in this process, to overcome the ambiguity of the simulation environment configuration in traditional descriptions, this embodiment explicitly adopts the method of moments solver; for the boundary conditions, it is set as an open boundary based on the Green's function, without the need to set up absorbing walls; for the excitation source, a zero-port configuration is performed, that is, no voltage or current source is defined, and only the intrinsic impedance matrix of the structure is solved; in order to ensure the numerical stability and convergence of the solution of the generalized eigenvalue equations, this embodiment strictly limits the maximum size of the mesh to 1 / 20 of the wavelength corresponding to the highest operating frequency; it should be clarified here that the antenna electromagnetic simulation model and the sensitivity analysis model are both deterministic physical and mathematical models based on numerical solutions of computational electromagnetics; the construction of the antenna electromagnetic simulation model and the sensitivity analysis model is directly discretized by the computer-aided design system based on the input initial geometric parameter vector and material property data, and its solution process completely follows the physical laws of Maxwell's equations, without the need for a data-driven training stage in the sense of machine learning; In this step, the system automatically generates a discretized scanning frequency band vector based on the previously acquired frequency parameters: in, ;against Each frequency point The solver independently runs the following generalized eigenvalue equations. It should be noted that although standard eigenmode analysis is often performed at a single frequency, this embodiment aims to accurately obtain the gradient information of eigenvalues ​​as a function of frequency. This type of broadband scan must be performed; the dense discrete data points generated in this step provide the necessary sample support for subsequent modal tracking, ensuring the accuracy of numerical differentiation and thus avoiding modal aliasing or derivative calculation distortion caused by excessively large frequency steps. The calculation formula is as follows: in, and All 3D matrix This represents the total number of basis functions after grid discretization. for 3D characteristic current vector; Derived from solving equations, its physical meaning is the first... The characteristic current distribution vector of order 1, dimensionless; Derived from solving equations, its physical meaning is the first... The first eigenvalue represents the ratio of stored energy to radiated energy in this mode, and is dimensionless. This embodiment establishes a digital mapping between physical parameters and electromagnetic characteristic modes, and in particular clarifies the discretization definition of the frequency axis, providing accurate data support for subsequent differential-based sensitivity analysis.

[0020] Example 3: Step two includes: S21, calculating the modal significance of the nth feature mode based on the eigenvalues; S22. Compare the modality saliency with a preset saliency threshold. When it is greater than the saliency threshold, mark the corresponding feature mode as the dominant mode. S23. For each dominant mode, in the structural parameter vector Based on this, the components in the process tolerance vector are superimposed to perform geometric perturbation, generating perturbed eigenvalues; S24. Calculate the rate of change of eigenvalues ​​before and after the perturbation, obtain the partial derivatives of the eigenvalues ​​with respect to each geometric parameter, and construct a parameter-modal sensitivity matrix from all the partial derivatives, where the matrix elements characterize the influence weight of the size change of a specific physical part on the resonant frequency of a specific mode.

[0021] This embodiment is a further specification of the steps for constructing the sensitivity analysis model in Embodiment 1. It elaborates on the construction process of the sensitivity matrix and provides a detailed explanation of the tolerance mapping for multiple physical dimensions. The system executes steps S21 to S22 to perform dominant mode selection; the system calculates the eigenvalues ​​based on the eigenvalues ​​obtained in step S13. Calculate the first Modal saliency of each characteristic mode The calculation formula is: This formula clearly states... The dimensionless real-valued eigenvalues ​​output by the solver characterize the ratio of modal stored energy to radiated energy. The imaginary unit; To better understand the physical meaning of this significance index, it should be noted that the formula extracts the magnitude of the modal weighting coefficients: complex terms It describes the complex response characteristics of the mode at a specific frequency, where the real part reflects radiation damping and the imaginary part reflects reactive energy storage; Taking the modulus of this complex number, i.e., calculating the Euclidean distance from the origin of the complex plane to this complex number point, aims to filter out phase information and simply extract the modal response intensity to external excitation; this operation utilizes a nonlinear compression mechanism to compress the range of values... The unbounded real eigenvalue mapping is The bounded significance index of the interval intuitively quantifies the degree to which the mode approaches the resonance state, i.e. hour This eliminates the ambiguity regarding the data type and physical meaning of variables; in response to modal saliency Greater than the preset significance threshold The system will mark the corresponding feature mode as the dominant mode; The system executes steps S23 to S24 to construct the parameter-modal sensitivity matrix. In step S23, for each dominant mode, the system performs a tolerance-geometric mapping perturbation operation; specifically, the system starts from the process tolerance vector... Extracting the initial geometric parameter vector The geometric tolerance components that correspond one-to-one with each element in the formula; The system targets the structural parameter vector. Perform a perturbation operation; since eigenmode analysis (CMA) is a passive solution, the eigenvalues ​​and mode distributions depend only on the conductor geometry and are independent of the feed location. Therefore, the feed parameter vector is not introduced when constructing the sensitivity matrix. The components in the equation are used to avoid generating invalid zero gradient calculations; in Based on this, geometric tolerance components are superimposed to perform perturbation; The system executes step S24; to accurately respond to the definition of the weights representing the influence of changes in the size of a specific physical part on the resonant frequency of a specific mode in the embodiment, this embodiment not only calculates the partial derivatives of the eigenvalues ​​with respect to the geometric parameters, but also combines the partial derivatives of the eigenvalues ​​with respect to the frequency; in this process, to prevent errors in derivative calculation due to mode order swapping during frequency scanning, the system introduces a mode tracking mechanism; before performing mode tracking, the initial state must be defined: for the starting frequency point of the frequency scan. The system establishes the initial modal order based on the physical properties of the characteristic modal analysis and the modal significance ranking criterion. Specifically, in Calculate all Significance of each mode and in accordance with The modes are indexed and numbered in descending order of size. This initial sort defines the tracking algorithm. Time reference benchmark This solves the cold start problem of recursive algorithms; after establishing the starting point, the current frequency point is calculated. and adjacent frequency points The modal confidence (MAC) between the characteristic current vectors is calculated using the following formula: in, Indicates frequency The first Characteristic current vectors of each mode Indicates frequency The first Characteristic current vectors of each mode; system traversal , search for The modal index that is the largest and greater than 0.9 is used to ensure the eigenvalue curve. The physical continuity; where the superscript This represents the conjugate transpose of a vector, used to calculate the inner product of two complex vectors. After ensuring modal alignment, the MAC algorithm is used to track the target frequency. Then, the system performs a secondary verification of the dominant mode: calculating the tracked mode in Modal saliency at the location Only retain Greater than the preset threshold The modes are included in the sensitivity matrix construction process to prevent modes that dominate at low frequencies but are already in a non-resonant state at the operating frequency from being included in the optimization objective; the parameter-mode sensitivity matrix is ​​constructed using implicit function differentiation rules. Its elements The calculation is as follows: Calculate the partial derivatives of the eigenvalues ​​with respect to the geometric parameters, using the following formula: in, In order to target the Geometric parameters The applied perturbation step size is taken as 1 / 10 to 1 / 5 of the corresponding process tolerance; the partial derivative of the eigenvalue with respect to frequency is calculated using the following formula: Calculate frequency sensitivity The calculation formula is as follows: in, The physical meaning of is the first The change of the first geometric parameter causes the first The drift of each dominant mode resonant frequency, in units of This embodiment eliminates the uncertainty in numerical calculation by introducing a MAC tracking mechanism, thus ensuring the mathematical rigor of the sensitivity matrix.

[0022] Example 4: Step 3 includes: S31, calculating the modal quality factor of each characteristic mode based on the eigenvalues ​​and their derivatives with respect to frequency; S32. Set a preset quality factor threshold, remove narrowband modes with a modal quality factor higher than the preset quality factor threshold, and retain broadband potential modes. S33. Construct a composite objective function, which consists of three weighted parts: the first part is the absolute value of the difference between the resonant frequency and the target frequency; the second part is the negative value of the modal significance; and the third part is the total sensitivity value calculated based on the parameter-modal sensitivity matrix.

[0023] This embodiment is a further specification of the steps for constructing the composite objective function in Embodiment 1. It describes in detail the mathematical construction logic of the multi-dimensional optimization objective and focuses on correcting the physical weighting method of the sensitivity term to conform to engineering practice. The system executes steps S31 to S32 to assess broadband potential; based on eigenvalues ​​and their derivatives with respect to frequency, the system calculates the modal quality factor. To ensure that the calculation results conform to the principles of electromagnetic field physics and are feasible, this embodiment adopts the recognized Q-value estimation method based on eigenvalue gradients in characteristic modal analysis; especially in the resonant region of the dominant mode, the calculation formula is as follows, utilizing the physical correspondence between the slope of the eigenvalue curve and energy storage: Among them, due to the eigenvalues ​​of the dominant mode near the resonance point Approaching 0, this formula, based on the Taylor expansion approximation, accurately describes the energy storage-emission ratio of the mode at the resonant point and serves as a physical benchmark for evaluating the bandwidth potential of an antenna. Let be the angular frequency, satisfying the relation , The frequency of electromagnetic waves; This is the partial derivative of the eigenvalue with respect to the angular frequency, specifically at the resonant frequency of the dominant mode. Place, that is The calculation is performed at the point of intersection, and the result is obtained through the central difference method. Compared to unverified approximate formulas, this calculation method ensures the robustness of the physical foundation of the optimization process; in response to the modal quality factor Higher than the preset quality factor threshold The settings are determined by working backward from the target bandwidth, such as... The system removes the narrowband mode and retains only the broadband potential mode; The system executes step S33 to construct the composite objective function. To ensure the uniformity of the physical dimensions in the objective function and to address the sensitivity aggregation problem when multiple dominant modes coexist, this embodiment normalizes and performs nonlinear aggregation on the sensitivity term, as shown in the following formula: in, This represents the set of geometric parameter indices to be optimized, i.e. , geometric parameter vector dimensionality; The resonant frequency of the dominant mode; For the target frequency; For the first The upper limit of manufacturing tolerance corresponding to each geometric parameter; This is the weighting coefficient; a recommended value is... To strengthen robust constraints; In this step, the physical definitions and dimensional treatments for each part are as follows: Frequency alignment term: To normalize the frequency error, ensuring the magnitude is between 0 and 1; significance maximization term: Using a negative sign transforms maximizing significance into minimizing the objective function value. The value range is [0,1]; the third part is the sensitivity penalty term, the calculation logic of which is: summing the squares of the normalized frequency offset drift; where, It represents the relative frequency drift under the maximum tolerance and is dimensionless. By introducing the square operation, the weight of the high-sensitivity term is nonlinearly amplified, so that the high sensitivity risk of any single parameter will lead to a sharp increase in the value of the objective function. By introducing squaring operations, the system nonlinearly amplifies the weight of large numerical sensitivity terms, so that the high sensitivity risk of any single parameter will lead to a sharp increase in the objective function value, thereby forcing the optimization algorithm to prioritize avoiding these areas with potential single-point failures. in, The frequency sensitivity calculated in step S24 is expressed in units of... , This is derived from the process tolerance vector. The absolute tolerance radius, in units of The product of the two Characterized by the maximum frequency drift under worst-case manufacturing error, in units of In order to add this product to the first dimensionless frequency error, the product is multiplied by the target frequency. This is divided, thus transforming into the normalized maximum potential frequency offset; This embodiment introduces a sensitivity penalty term based on absolute tolerance weighting and normalization, which forces the optimization algorithm to find flat regions in the search space where the frequency drift ratio is still within an acceptable range even if the maximum allowable processing error occurs, thereby mathematically guaranteeing the manufacturability of the design.

[0024] Example 5: Step four includes: S41, initializing the optimization algorithm population, using the initial geometric parameter vector P as the starting point; S42. In each iteration, calculate the value of the composite objective function corresponding to the current parameter combination; S43. Determine whether the total sensitivity value is less than or equal to the preset robustness baseline. If yes, retain the parameter combination in the next generation population; otherwise, apply a penalty weight to the parameter combination so that it is eliminated in subsequent iterations. S44. When the number of iterations reaches the preset value or the composite objective function converges, output the combination of geometric parameters that minimizes the composite objective function as the globally optimal combination of geometric parameters.

[0025] This embodiment is a further specification of the global optimization steps in Embodiment 1, describes the execution flow of multi-objective optimization, and supplements the special processing logic for the feed point parameters. The system executes step S41 to initialize the optimization algorithm population; an initial population is generated using a genetic algorithm or particle swarm optimization algorithm; the system enters an iterative loop, executing steps S42 to S43; in each iteration, the composite objective function corresponding to the current parameter combination is calculated. The value of the multi-objective optimization algorithm is as follows: The system maps the antenna geometry and feed coordinates in the computer-aided design model into multi-dimensional position vectors of individual algorithms, and under the given process tolerance boundary constraints, substitutes the electromagnetic response physical index output by the solver into the composite objective function as the fitness evaluation standard, and directly searches for the optimal solution that satisfies the physical matching through mathematical iteration; Specifically, before calculating the objective function in step S42, the system executes an adaptive correction subroutine for the feed point; to ensure the feasibility of the preset impedance matching criterion in this embodiment, the following deterministic search algorithm based on grid traversal is used to update the algorithm. Extract the characteristic current magnitude of the dominant mode at all discrete mesh vertices under the current geometric parameters. ; For the vertex indices of the discretized mesh model, a matching target current value is set, and its calculation formula is as follows: The physical basis for setting this target value is: according to the power-current square law relationship in antenna theory. The current amplitude decays to its peak value. The half-value point of the modal radiative power is approximately 0.707. In a typical microstrip patch structure, the real part of the input impedance at this location is statistically closest to... The power supply standard is used, therefore, for typical microstrip patch structures, the best physically matching anchor point is selected; the set of mesh vertices is traversed to filter out those that meet the standard. A subset of candidate nodes; where The tolerance threshold is defined as follows: That is, allow The search tolerance; this parameter is set based on the grid discretization density and is designed to cover numerical jumps caused by grid discontinuities, ensuring that the algorithm can always capture effective feed candidate points and avoid empty set errors; Within the subset of candidate nodes, a neighborhood consistency screening is performed: for the first iteration of population initialization, nodes are selected based on the initial geometric parameter vector. The node with the smallest initial Euclidean distance to the feed coordinates is selected. For subsequent iterations, the node with the smallest Euclidean distance to the corresponding parent individual in the previous generation is selected, and its physical coordinates are assigned to the vector. In The components are used to ensure the physical continuity of the optimization trajectory; from the perspective of optimization algorithm principles, this adaptive correction step of the feed point constitutes the individual repair operator or Lamarck evolution mechanism in the meme algorithm. In traditional stochastic optimization, the feed point coordinates With geometric dimensions Strong coupling often leads to a large number of geometric individuals being misjudged as having low fitness due to poor randomness in the feed position. This embodiment achieves decoupling between geometric optimization and feed matching by performing deterministic physical corrections before fitness evaluation and writing the corrected genes, i.e., the optimal feed point, back into the population. This mechanism is equivalent to implicitly eliminating the influence of impedance mismatch while optimizing the geometric structure, forming a Lamarck-like evolutionary mechanism. This not only does not interfere with the evolutionary mechanism, but also greatly improves the algorithm's ability to distinguish the merits of geometric parameters by eliminating random noise introduced by impedance mismatch, ensuring that the optimization converges to the physically realizable optimal design. Through this subroutine, the algorithm ensures that the individuals evaluated in each iteration are physically approximately impedance matched, thus avoiding invalid solutions caused by random walks at the feed position. The system executes step S43 to perform sensitivity gating; the system then determines the total sensitivity value. Is it less than or equal to the preset robustness baseline? The baseline was set at the 40th percentile of the population sensitivity distribution; in response to The system applies a penalty weight to this parameter combination. If not, then a penalty weight is applied to the parameter combination. ,set up Much greater than the maximum value of the current population objective function, i.e. , here This refers to the maximum value of the objective function for all individuals in the current iteration population, for example... This will cause it to be phased out in subsequent iterations; The system executes step S44. When the number of iterations reaches a preset value or the composite objective function converges, the specific convergence criterion is: continuity. The optimal objective function value of the population over generations, for example, 20 generations. relative change Less than the preset convergence threshold ,For example Or, the fitness variance of individuals in the population is lower than a preset genetic diversity threshold, for example, taken as 1% of the initial population fitness variance, calculated using the following formula: To prevent the algorithm from falling into premature convergence, the output contains the globally optimal combination of parameters, including the best geometry and the best feed location. ; It should be noted that the above convergence threshold and iterative algebra These are merely exemplary values, and those skilled in the art can make adaptive adjustments based on the convergence speed of the optimization algorithm and the availability of computing resources.

[0026] Example 6: Step four also includes: S45, generating a virtual mass production sample set based on the globally optimal combination of geometric parameters and the process tolerance vector using Monte Carlo analysis; S46. Calculate the proportion of samples in the virtual mass production sample set that meet the design indicators, and generate a robustness score report. S47. Based on the robustness rating report, classify and evaluate the antenna design scheme and output the classification results; In step S47, a first preset ratio threshold and a second preset ratio threshold are preset, with the first preset ratio threshold being greater than the second preset ratio threshold. The graded evaluation includes: when the sample ratio is greater than or equal to the first preset ratio threshold, the design scheme is determined to be highly robust, and a direct mass production verification signal is generated; when the sample ratio is less than the first preset ratio threshold but greater than or equal to the second preset ratio threshold, the design scheme is determined to be conditionally usable, and a process data package containing suggestions for improving PCB etching accuracy is generated; when the sample ratio is less than the second preset ratio threshold, the design scheme is determined to be highly sensitive and risky, and a warning signal prompting a readjustment of the antenna topology is generated.

[0027] This embodiment is a further specification of the robustness verification steps in Embodiment 1, describing the logic of post-verification and hierarchical evaluation of the optimization results; The system executes step S45 to generate a virtual mass production sample set; based on the globally optimal combination of geometric parameters. Combined with process tolerance vector The system employs Monte Carlo analysis to establish a clear mapping relationship between tolerance and statistical distribution: the process tolerance vector... Each component in Defined as a normal distribution The boundary, corresponding to the process capability range with a confidence level of 99.7%, is based on the process capability index in general electronic manufacturing processes. Based on the assumptions made, the standard deviations of each parameter are calculated. ; Using a stochastic process to generate The formula for calculating a virtual mass production sample is as follows: in, Follows a multidimensional normal distribution , Represents a multidimensional normal distribution, with the covariance matrix... It is a diagonal matrix, with the diagonal elements representing the variances of the parameters. That is, assuming that the machining errors of each geometric parameter are independent of each other, it is recommended ; The system executes steps S46 to S47 to perform robustness classification evaluation; in this step, it is necessary to explain the switching of the simulation model; since the previous steps S1-S44 use eigenmode analysis without excitation source (CMA), it cannot directly output S-parameters. Therefore, in step S46, the system utilizes the feed point location acquired in step S11 but not yet used. Data, constructing an active-driven model; the specific operation is: in the mesh model Discrete ports or waveguide ports are added at the coordinates, and the electromagnetic solver is switched from the characteristic mode solver to the frequency domain method of moments solver or the finite element solver, thereby performing full-wave simulation for each virtual sample; during this process, the characteristic impedance of the discrete port is set to... , directly calculate its normalized reflectance coefficient S11; The system statistically analyzes the proportion of virtual mass production samples that meet design specifications, such as S11 < -10dB. ; in response Greater than or equal to the first preset ratio threshold For example, if the system determines the design scheme to be highly robust (e.g., 98%), it generates a direct mass production verification signal; in response to... Less than And greater than or equal to the second preset ratio threshold For example, if the system determines the design is feasible at 90%, it generates a process data package containing suggestions to improve PCB etching accuracy; in response to Less than The system determines that the design scheme is highly sensitive and risky, and generates a warning signal prompting the antenna topology to be readjusted. This embodiment constructs a virtual mass production verification stage, bringing yield issues that would normally only be discovered during trial production to the simulation design stage. This tiered evaluation mechanism not only provides a pass / fail conclusion for the design scheme but also offers targeted process improvement suggestions, significantly reducing hardware trial-and-error costs and shortening the product development cycle. The antenna design scheme output using this method achieves the following specific improvements: First, by constructing a composite objective function including a sensitivity penalty term, the optimization algorithm is forced to avoid vulnerable mode regions with single-point failure risks, ensuring that the output geometric parameter combination maintains the stability of the resonant frequency response even under superimposed manufacturing error disturbances, guaranteeing the manufacturability of the design. Second, by introducing Monte Carlo analysis to predict mass production yield under the normal distribution boundary during the simulation design stage, robustness defects that would normally only be discovered during trial production are addressed upfront, overcoming yield losses caused by traditional blind numerical optimization and significantly reducing hardware trial-and-error costs.

[0028] Example 7: Applications of the parameter-modal sensitivity matrix also include: identifying the geometric part corresponding to the element with the largest value in the parameter-modal sensitivity matrix and marking it as the sensitive source region; in the electromagnetic simulation model of the antenna, performing mesh refinement processing on the sensitive source region, or applying stricter constraints on the size parameters of the sensitive source region during the optimization process.

[0029] This embodiment is a further extension of the application of the sensitivity matrix in Embodiment 3, describing a method for local fine-tuning using sensitivity information; The system performs sensitive source region identification; the system traverses the parameter-modal sensitivity matrix. Identify the element with the largest value ; Assumption This corresponds to the chamfer size of the patch, and the system marks this chamfer area as a sensitive source area; Based on this, the system implements targeted processing measures. On the one hand, in the antenna electromagnetic simulation model, the system performs local mesh refinement processing on the sensitive source area, setting the mesh size to 1 / 50 of the wavelength or smaller to improve the calculation accuracy of the area. On the other hand, during the optimization process, the system applies stricter constraints on the size parameters of the sensitive source area, or marks stricter tolerance requirements specifically for the area in the output PCB processing drawings. This embodiment achieves a non-uniform distribution of computing and manufacturing resources by identifying and locking sensitive source regions; it concentrates high-precision meshes and precision machining tolerances on key parts that have the greatest impact on performance, effectively controlling simulation calculation time and mass production costs while ensuring overall antenna performance, thus embodying the concept of lean design.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for optimizing patch antenna parameters based on feature analysis, characterized in that, The specific steps include: Step 1: Obtain the initial geometric model data, substrate material parameters, and preset process tolerance range data of the patch antenna to be optimized, and construct the antenna electromagnetic simulation model; based on the initial geometric model data, use the characteristic mode analysis engine to perform source-free solution, and extract the eigenvalues ​​and characteristic current distribution data of the first N characteristic modes; Step 2: Based on the eigenvalues, calculate the modal saliency of each eigenmode, and select the dominant mode according to the preset saliency threshold; construct a sensitivity analysis model for the dominant mode, and calculate the partial derivatives of the eigenvalues ​​with respect to the geometric parameters by applying perturbation to the initial geometric model data, and establish a parameter-modal sensitivity matrix; Step 3: Calculate the modal quality factor of each feature mode based on the eigenvalues, and construct a composite objective function containing a frequency alignment term, a saliency maximization term, and a sensitivity penalty term by combining the modal saliency and the parameter-modal sensitivity matrix. Step 4: Use a multi-objective optimization algorithm to iterate and optimize the composite objective function in the parameter space to obtain the globally optimal combination of geometric parameters. Based on the response distribution of the globally optimal combination of geometric parameters within the process tolerance range, generate a robustness score report to output an antenna design scheme that can resist processing errors.

2. The method for optimizing patch antenna parameters based on feature analysis according to claim 1, characterized in that, Step one includes: S11. Collect the length, width, feed point position, and chamfer size of the patch antenna to be optimized as the initial geometric parameter vector; collect the dielectric constant, loss tangent, and thickness of the dielectric substrate as material property data. S12. Collect the etching error range and dielectric constant deviation range existing in the production process, and define them as the process tolerance vector; S13. Import the initial geometric parameter vector and the material property data into the electromagnetic simulation solver, establish a discretized mesh model, solve the generalized eigenvalue equation, and output the eigenvalues ​​of the first N orders and the corresponding characteristic current distribution.

3. The method for optimizing patch antenna parameters based on feature analysis according to claim 2, characterized in that, Step two includes: S21. Calculate the modal saliency of the nth feature mode based on the eigenvalues. S22. Compare the modality saliency with a preset saliency threshold. When it is greater than the saliency threshold, mark the corresponding feature mode as the dominant mode. S23. For each dominant mode, based on the initial geometric parameter vector, the components in the process tolerance vector are superimposed to perform geometric perturbation to generate perturbed eigenvalues. S24. Calculate the rate of change of eigenvalues ​​before and after the perturbation, obtain the partial derivatives of the eigenvalues ​​with respect to each geometric parameter, and construct a parameter-modal sensitivity matrix from the set of all partial derivatives, where the matrix elements characterize the influence weight of the size change of a specific physical part on the resonant frequency of a specific mode.

4. The method for optimizing patch antenna parameters based on feature analysis according to claim 3, characterized in that, Step three includes: S31. Based on the eigenvalues ​​and their derivatives with respect to frequency, calculate the modal quality factor of each characteristic mode; S32. Set a preset quality factor threshold, remove narrowband modes whose modal quality factor is higher than the preset quality factor threshold, and retain broadband potential modes. S33. Construct a composite objective function, which consists of three weighted parts: the first part is the absolute value of the difference between the resonant frequency and the target frequency; the second part is the negative value of the modal significance; and the third part is the total sensitivity value calculated based on the parameter-modal sensitivity matrix.

5. The method for optimizing patch antenna parameters based on feature analysis according to claim 4, characterized in that, Step four includes: S41. Initialize the optimization algorithm population, using the initial geometric parameter vector P as the starting point; S42. In each iteration, calculate the value of the composite objective function corresponding to the current parameter combination; S43. Determine whether the total sensitivity value is less than or equal to the preset robustness baseline. If yes, retain the parameter combination for the next generation of the population; otherwise, apply a penalty weight to the parameter combination so that it is eliminated in subsequent iterations. S44. When the number of iterations reaches a preset value or the composite objective function converges, output the geometric parameter combination that minimizes the composite objective function as the globally optimal geometric parameter combination.

6. The method for optimizing patch antenna parameters based on feature analysis according to claim 5, characterized in that, Step four also includes: S45. Based on the globally optimal combination of geometric parameters and the process tolerance vector, a virtual mass production sample set is generated using Monte Carlo analysis. S46. Calculate the proportion of samples in the virtual mass production sample set that meet the design indicators, and generate a robustness score report; S47. Based on the robustness rating report, the antenna design scheme is graded and evaluated, and the grading results are output.

7. The method for optimizing patch antenna parameters based on feature analysis according to claim 6, characterized in that, In step S47, a first preset ratio threshold and a second preset ratio threshold are preset, wherein the first preset ratio threshold is greater than the second preset ratio threshold. The graded evaluation includes: When the sample ratio is greater than or equal to the first preset ratio threshold, the design scheme is determined to be highly robust, and a direct mass production verification signal is generated. When the sample ratio is less than the first preset ratio threshold and greater than or equal to the second preset ratio threshold, the design scheme is determined to be conditionally available, and a process data package containing suggestions for improving PCB etching accuracy is generated. When the sample ratio is less than the second preset ratio threshold, the design scheme is determined to be highly sensitive and risky, and a warning signal is generated to prompt the antenna topology to be readjusted.

8. The method for optimizing patch antenna parameters based on feature analysis according to claim 3, characterized in that, The application of the parameter-modal sensitivity matrix also includes: Identify the geometric location corresponding to the element with the largest value in the parameter-modal sensitivity matrix and mark it as the sensitive source region; In the electromagnetic simulation model of the antenna, the sensitive source region is subjected to mesh refinement processing, or a more stringent constraint range is applied to the size parameters of the sensitive source region during the optimization process.

Images