Design Method and Filter Antenna Based on Glass Integrated Packaging

Abstract

This invention discloses a design method and antenna for a glass-integrated packaged filter antenna. The method includes: determining the parameters influencing the filtering characteristics and using them as input excitations; generating an initial dataset using Latin hypercube sampling; training a deep neural network to obtain a first surrogate model; optimizing using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set; selecting new sample points for full-wave simulation to form new data pairs; updating the first surrogate model to obtain a second surrogate model; and using the second surrogate model to identify new input excitations. This invention has the advantages of efficient optimization process, low-cost design investment, and achieving globally optimal performance.

Description

Technical Field

[0001] This invention relates to a design method and a filter antenna based on glass integrated packaging, belonging to the fields of millimeter-wave communication and artificial intelligence technology. Background Technology

[0002] In recent years, integrated antennas based on through-glass vias (TGV) have attracted much attention due to their low loss, high integration, and compatibility with chip manufacturing processes. However, filter antennas, which integrate the filter circuit and the antenna on the same substrate, are more difficult to design than ordinary antennas.

[0003] The design of glass-integrated filter antennas faces numerous challenges: First, their structure is highly complex, being a three-dimensional multilayer structure containing radiating patches, parasitic elements, feed probes, and numerous metallized vias (TGVs) and slots for filtering, with highly nonlinear coupling relationships between structural parameters. Second, there are multi-objective optimization conflicts; the antenna design must simultaneously satisfy multiple, even conflicting, performance indicators. For example, there are significant design conflicts between achieving good impedance matching (reflection coefficient S11 < -10dB) and obtaining high radiation gain over a wide operating frequency band, and achieving deep suppression (i.e., forming multiple radiation nulls) outside the passband. Finally, traditional design methods are inefficient, relying on engineers' experience and three-dimensional electromagnetic simulation software (such as HFSS) for repeated "manual trial and error" tuning. This process not only consumes a large amount of computational resources and time, but also, due to the huge parameter space, it is difficult to find a globally optimal solution, often only obtaining locally optimal or suboptimal solutions. Summary of the Invention

[0004] In view of this, the present invention provides a design method, a filter antenna, a device, a computer device, and a storage medium based on glass integrated packaging for filter antennas. By establishing a proxy model of antenna structure and performance and using optimization algorithms to automatically optimize the design, it can significantly improve design efficiency.

[0005] The first objective of this invention is to provide a design method for a filter antenna based on glass integrated packaging.

[0006] The second objective of this invention is to provide a filter antenna based on glass integrated packaging.

[0007] The third objective of this invention is to provide a filter antenna design device based on glass integrated packaging.

[0008] The fourth object of the present invention is to provide a computer device.

[0009] The fifth object of the present invention is to provide a storage medium.

[0010] The first objective of this invention can be achieved by adopting the following technical solution:

[0011] A method for designing a filter antenna based on glass integrated packaging, comprising:

[0012] The parameters affecting the filtering characteristics are determined based on the structure of the filtering antenna and used as input excitation.

[0013] Based on the range of values ​​of the input stimulus, an initial dataset is generated using the Latin hypercube sampling method. The initial dataset includes sample points and the performance response of the sample points.

[0014] A deep neural network is trained based on the initial dataset to obtain the first surrogate model;

[0015] Using the first agent model as a performance evaluation tool, a non-dominated sorting genetic algorithm is used for multi-objective optimization to obtain the Pareto optimal solution set.

[0016] From the Pareto optimal solution set, one or more new sample points are selected according to a preset criterion for full-wave simulation to obtain the performance response of the new sample points and form a new data pair.

[0017] The new data pair is merged with the initial dataset, and the first proxy model is retrained or updated to obtain the second proxy model;

[0018] The second surrogate model is used to identify new input stimuli.

[0019] The second objective of this invention can be achieved by adopting the following technical solution:

[0020] A glass-integrated packaged filter antenna, comprising:

[0021] A dielectric substrate;

[0022] A main radiating patch is disposed on the surface of the dielectric substrate;

[0023] A metal floor, wherein the metal floor is disposed within the dielectric substrate;

[0024] A low-frequency radiation null point generation module, the low-frequency radiation null point generation module includes one or more sets of metallized through holes, the metallized through holes penetrating the dielectric substrate in the vertical direction, for electrically connecting the main radiating patch to the metal ground plane;

[0025] A high-frequency radiation null point generation module, the high-frequency radiation null point generation module including multiple sets of gaps etched on the main radiation patch;

[0026] A full-band out-of-band suppression enhancement module, wherein the full-band out-of-band suppression enhancement module includes at least one of the following:

[0027] A parasitic patch, the parasitic patch being disposed on at least one side of the main radiating patch;

[0028] A planar inverted-F antenna element, wherein the planar inverted-F antenna element is disposed on at least one side of the main radiating patch;

[0029] Among them, the spacing parameters of the metallized through-hole array, the length parameters of the slots, and the size parameters of the parasitic patch or planar inverted-F antenna element are intelligently optimized by the above-mentioned filtered antenna design method.

[0030] The third objective of this invention can be achieved by adopting the following technical solution:

[0031] A filter antenna design device based on glass integrated packaging, comprising:

[0032] The determination module is used to determine the parameters affecting the filtering characteristics based on the filtering antenna structure, and to use them as input excitation.

[0033] The generation module is used to generate an initial dataset by combining the range of values ​​of the input stimulus and using the Latin hypercube sampling method. The initial dataset includes sample points and the performance response of the sample points.

[0034] The first training module is used to train a deep neural network based on the initial dataset to obtain a first surrogate model;

[0035] The optimization module is used to use the first agent model as a performance evaluation tool and employ a non-dominated sorting genetic algorithm to perform multi-objective optimization to obtain a Pareto optimal solution set.

[0036] The selection module is used to select one or more new sample points from the Pareto optimal solution set according to a preset criterion for full-wave simulation, obtain the performance response of the new sample points, and form a new data pair.

[0037] The second training module is used to merge the new data pair with the initial dataset, retrain or update the first proxy model, and obtain the second proxy model.

[0038] The recognition module is used to identify new input stimuli using the second surrogate model.

[0039] The fourth objective of this invention can be achieved by adopting the following technical solution:

[0040] A computer device includes a processor and a memory for storing a processor-executable program, wherein when the processor executes the program stored in the memory, it implements the above-described filtered antenna design method.

[0041] The fifth objective of this invention can be achieved by adopting the following technical solution:

[0042] A storage medium storing a program that, when executed by a processor, implements the above-described filtered antenna design method.

[0043] The present invention has the following advantages over the prior art:

[0044] (1) Extremely high design efficiency and significantly reduced cost: This invention uses a fast surrogate model to replace time-consuming full-wave simulation software for iterative optimization. Once the model is trained, thousands of optimization evaluations can be completed in minutes, shortening the traditional design cycle of several weeks to several hours, greatly saving computing resources and manpower costs.

[0045] (2) Excellent optimization effect and global optimal performance: Genetic algorithm is a global optimization algorithm that can effectively avoid getting trapped in local optima. For the complex parameter space of TGV antenna, this invention can explore excellent design schemes that are not intuitive and surpass human engineers' experience, thereby achieving performance breakthroughs in multiple dimensions such as bandwidth, gain, and out-of-band suppression.

[0046] (3) Effectively handle multi-objective conflict problems: Antenna design is essentially a process of balancing multiple objectives. This invention, by constructing a multi-output surrogate model and defining a comprehensive fitness function, can systematically and simultaneously optimize multiple conflicting performance indicators, find a Pareto optimal solution set that balances various performances, and provide designers with diverse choices.

[0047] (4) Advanced Technology and Broad Application Prospects: This invention is the first to systematically apply advanced artificial intelligence optimization methods to the emerging field of millimeter-wave TGV filter antenna design, solving the core design efficiency and performance bottlenecks in this field. This invention has good scalability and can be extended to the design of other TGV-based three-dimensional integrated passive devices (such as filters and couplers). Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0049] Figure 1 This is a perspective view of a glass-integrated packaged filter antenna according to an embodiment of the present invention.

[0050] Figure 2 This is a top view of a glass-integrated packaged filter antenna according to an embodiment of the present invention.

[0051] Figure 3 This is a side view of a glass-integrated packaged filter antenna according to an embodiment of the present invention.

[0052] Figure 4 This is a flowchart illustrating a glass-integrated packaging-based filter antenna design method according to an embodiment of the present invention.

[0053] Figure 5 This is a structural diagram of a glass-integrated packaged filter antenna design device according to an embodiment of the present invention.

[0054] Figure 6 This is a schematic diagram of simulation data according to an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0056] The core technical problem this invention aims to solve is how to overcome the shortcomings of complex TGV filter antenna structures, multi-target performance conflicts, and low efficiency of traditional design methods. It proposes a novel design method that can systematically and automatically find the optimal antenna structure parameters. Furthermore, by collaboratively optimizing the physical dimensions and relative positions of key structures, low-frequency nulls, high-frequency nulls, and suppression enhancement effects are coupled together to jointly solve the problem of insufficient out-of-band suppression capability of traditional filter antennas, while simultaneously satisfying a wider operating bandwidth and better out-of-band suppression performance.

[0057] This invention aims to design and optimize a high-performance TGV filtering antenna operating in the 5G millimeter-wave n257 / n258 frequency band (approximately 24-30GHz). The specific implementation scheme is as follows:

[0058] like Figures 1 to 3As shown, this embodiment provides a glass-integrated packaged filter antenna, which includes: a dielectric substrate; a main radiating patch 2 disposed on the surface of the dielectric substrate; a metal ground plane 3 disposed within the dielectric substrate; a low-frequency radiation null generation module, the low-frequency radiation null generation module including one or more sets of metallized vias 5, the metallized vias 5 penetrating the dielectric substrate in the vertical direction for electrically connecting the main radiating patch 2 and the metal ground plane 3; and a high-frequency radiation null generation module, the high-frequency... The radiation null generation module includes multiple sets of slots etched on the main radiating patch 2; a full-band out-of-band suppression enhancement module, the full-band out-of-band suppression enhancement module including at least one of: a parasitic patch 1, the parasitic patch 1 being disposed on at least one side of the main radiating patch 2; and a planar inverted-F antenna element 4, the planar inverted-F antenna element 4 being disposed on at least one side of the main radiating patch 2; wherein, the spacing parameters of the metallized via array 5, the length parameters of the slots, and the size parameters of the parasitic patch 1 or the planar inverted-F antenna element 4 are intelligently optimized by the filtering antenna design method of this embodiment.

[0059] Furthermore, the filtered antenna also includes: a microstrip line 6; a feed port 7 connected to the microstrip line 6; and an L-shaped probe 8, one end of which is connected horizontally to the main radiating patch 2, and the other end of which is connected vertically to the microstrip line 6.

[0060] Furthermore, the dielectric substrate comprises a first substrate 9, a second substrate 10, a third substrate 11, a first metal layer 12, a second metal layer 13, and a third metal layer 14. The third metal layer 14 is located at the top layer and is used to mount radiating and parasitic patches; the second metal layer 13 is located in the middle layer and serves as a reference ground; the first metal layer 12 is located at the bottom layer and is used to fabricate the horizontal feed line portion for L-shaped probe feeding. A wider bandwidth can be achieved through L-shaped probe coupling feeding.

[0061] Specifically, the antenna entity used in this embodiment is a highly integrated structure based on a multilayer glass substrate. Its excellent filtering performance is not determined by a single component, but is achieved through the synergistic effect and coupling of the following three core functional modules:

[0062] 1) Low-frequency radiation null generation module: short-circuit TGV array

[0063] This module consists of one or more sets of through-metallized vias (TGVs) that penetrate the dielectric substrate vertically, directly connecting the top-layer main radiating patch to the internal metal ground plane. At specific frequencies near the lower limit of the operating band (e.g., ~21 GHz), current is concentrated primarily on these short-circuited TGVs, forming an antiphase distribution with the current on the main radiating patch. The far-field radiation generated by this antiphase current cancels each other out, creating a deep radiation null in the antenna's gain response, constituting the steep roll-off edge on the low-frequency side of the filtering characteristics.

[0064] 2) High-frequency radiation null point generation module: Surface gaps of the radiation patch

[0065] This module consists of two or more sets of specifically shaped slots etched onto the metal surface of the main radiating patch. These slots alter the high-frequency current path on the patch surface. At two different frequency points within the upper limit of the operating frequency band (e.g., 33 GHz and 35 GHz), these slots excite different higher-order resonant modes, causing the current on the patch to again form an antiphase or orthogonal canceling distribution. The result is the continuous generation of two independent radiation nulls on the high-frequency side, ensuring excellent out-of-band rejection capabilities on the high-frequency side as well.

[0066] 3) Full-band out-of-band suppression enhancement module: parasitic and / or PIFA cooperative unit

[0067] The module consists of parasitic patches located around the main radiating patch that are not directly fed, and planar inverted-F antenna (PIFA) elements. They harvest energy from the main radiating patch via near-field coupling. The dimensions of these elements are precisely designed so that the induced current throughout the out-of-band region (including the low-frequency and high-frequency stopbands) is opposite in direction to the residual radiated current of the main radiating patch. This active anti-phase radiation cancellation effect, much like noise-canceling headphones, further reduces the gain level across the entire stopband, significantly improving the overall out-of-band rejection performance of the antenna.

[0068] The three modules described above are not simply stacked, but rather highly coupled and work in concert. The location and size of the short-circuited TGV (module 1) affect the current distribution on the patch, thereby changing the optimal location and size of the null point generated by the surface slot (module 2). Simultaneously, the combined effect of these two modules determines the magnitude and phase of the energy coupled to the parasitic unit (module 3). Therefore, only through global and coordinated optimization of all relevant parameters can the low-frequency null point, high-frequency null point, and suppression enhancement effect be perfectly integrated to form a wide-coverage, deeply suppressed ultra-wide stopband. This complex coordinated optimization is precisely the core problem that the filtered antenna design method subsequently employed in this invention aims to solve.

[0069] It is worth mentioning that the technical advantages of the filter antenna structure used in this invention are as follows:

[0070] Novel Integrated Filtering Mechanism: This antenna achieves highly efficient integrated filtering through a clever composite structure. It utilizes a short-circuited TGV array to precisely generate a radiation null on the low-frequency side of the passband, while simultaneously generating two additional radiation nulls on the high-frequency side of the passband using specific slots etched into the main radiating patch. Furthermore, the introduction of a parasitic patch and a planar inverted-F antenna (PIFA) further enhances the overall out-of-band suppression level. This multi-mechanism fusion design results in a steep passband roll-off characteristic, achieving both deep and wide stopband suppression without the need for any external independent filtering circuitry.

[0071] High miniaturization and integration: Because the filtering function is fully embedded inside the antenna radiating element, rather than using the traditional "antenna + filter" cascade method, circuit area is greatly saved. Combined with multilayer glass substrates and TGV three-dimensional vertical interconnect technology, the antenna structure can be compactly stacked in the vertical direction, further compressing the overall size of the antenna to achieve an ultra-compact volume of 0.7λc × 0.56λc × 0.065λc, making it ideal for modern antenna-in-package (AiP) solutions with stringent space requirements.

[0072] Combining wide bandwidth and high efficiency: This structure achieves excellent filtering performance without sacrificing the antenna's operating bandwidth and efficiency. Its L-shaped probe-coupled feeding method enables wide-band impedance matching, fully covering key 5G millimeter-wave frequency bands (such as n257 / n258). Simultaneously, thanks to the low-loss characteristics of the glass substrate in the millimeter-wave band and the high reliability of the TGV interconnect, the antenna maintains high radiation efficiency.

[0073] To achieve optimal synergy among the aforementioned functional modules and overcome the limitations of high data acquisition costs and subjective optimization target setting in traditional AI optimization methods, this embodiment employs an innovative adaptive closed-loop optimization process. This process combines active learning with the non-dominated sorting genetic algorithm (NSGA-II), and the specific implementation plan is as follows:

[0074] like Figure 4 As shown, this embodiment provides a method for designing a filter antenna based on glass integrated packaging. This method includes:

[0075] S101 determines the parameters affecting the filtering characteristics based on the filter antenna structure and uses them as input excitation.

[0076] In this step, the antenna structure is parametrically modeled. Based on the analysis, the following six parameters that have the most critical impact on the filtering characteristics are selected as the input excitation X=[d,d1,L s1 ,L s2 Wp ,L p ], and set a reasonable range of values ​​for it, that is, determine the optimization parameter space, where: d represents the spacing between short-circuited TGV arrays; d1 represents the spacing of vias within the same TGV array; L s1 L represents the length of the first set of surface gaps. s2 Indicates the length of the second set of surface gaps; W p Indicates the width of the parasitic patch; L p This indicates the length of the parasitic patch. For example, d ranges from [0.5, 1.5] mm, d1 ranges from [0.1, 0.3] mm, and L... s1 The range is [1.2, 1.8] mm, L s2 The range is [0.8, 1.2] mm, W p The range is [1.5, 2.5] mm, L p The range is [3.0, 4.0] mm.

[0077] S102 combines the range of values ​​of the input stimulus and uses the Latin hypercube sampling method to generate an initial dataset, which includes sample points and the performance response of the sample points.

[0078] In this step, the Latin hypercube sampling (LHS) method is used to generate only a small initial training set (e.g., 100 sample points) in the parameter space. These 100 sets of parameters are then subjected to full-wave simulation, and their performance metrics are extracted to form the initial training dataset T_initial.

[0079] M antenna performance metrics (also known as performance responses) are defined, with the average reflection coefficient within the operating frequency band, the in-band peak gain, and the gain values ​​at multiple out-of-band radiation nulls serving as the output response. Through parametric scanning and electromagnetic simulation, a large number of "input excitation-output response" data pairs are generated, forming the training dataset.

[0080] S103 trains a deep neural network based on the initial dataset to obtain the first surrogate model.

[0081] In this step, using the initial dataset described above, a preliminary deep neural network (DNN) surrogate model M_initial is trained.

[0082] It's important to note that the training objective at this stage is not to pursue accuracy, but rather to build a surrogate model with basic predictive capabilities and good generalization performance on a small dataset. This provides a reliable starting point for subsequent iterations, and the core challenge lies in preventing overfitting. For small datasets, the initial model uses a relatively shallow and narrow deep neural network. For example, it can be configured with a single input layer containing 6 nodes, corresponding to 6 input parameters, followed by two fully connected hidden layers with 32 and 64 nodes respectively. Each hidden layer is then followed by a ReLU activation layer. A regression layer is then added. Since preventing overfitting on small datasets is crucial, a Dropout layer is added after each hidden layer in the above network structure. In each training iteration, this layer randomly sets the output of its neurons to zero with a certain probability (e.g., p = 0.3, or 30%). This prevents the network from over-relying on any single neuron and forces it to learn multiple independent feature representations, effectively preventing overfitting. In addition, some common settings include: using mean squared error as the loss function, using the Adam optimizer, and setting the batch size to 16 or 32.

[0083] It is worth noting that not all neural networks are suitable as surrogate models for this invention. This invention chooses DNNs because they are well-suited to the present invention. There is a highly nonlinear physical relationship between antenna structural parameters and electromagnetic performance (such as S-parameters and gain). DNNs, through their multi-layered structure and nonlinear activation functions (such as ReLU), can effectively analyze complex continuous functions. Furthermore, this invention requires the simultaneous prediction of multiple performance metrics, and the output layer of a DNN can flexibly be configured with multiple neurons to support multi-output tasks. In contrast, convolutional neural networks, recurrent neural networks, and shallow neural networks cannot accomplish these tasks. Therefore, this invention selects a DNN with at least two or more hidden layers as the surrogate model to ensure it has sufficient model capacity to learn and characterize the complex electromagnetic behavior of antennas.

[0084] S104 uses the first surrogate model as a performance evaluation tool and employs a non-dominated sorting genetic algorithm for multi-objective optimization to obtain a Pareto optimal solution set.

[0085] In this step, the surrogate model is used as the fitness function of the genetic algorithm. By simulating the selection, crossover and mutation operations in biological evolution, a global search is performed in the N-dimensional parameter space to find one or more sets (Pareto fronts) of structural parameters that enable M performance indicators to reach the optimal or balanced state simultaneously.

[0086] S105 selects one or more new sample points from the Pareto optimal solution set according to a preset criterion to perform full-wave simulation, obtains the performance response of the new sample points, and forms a new data pair.

[0087] S106 The new data pair is merged with the initial dataset, and the first proxy model is retrained or updated to obtain the second proxy model.

[0088] In this embodiment, the "active learning-optimization" iterative loop (S104-S106) is entered, and a set number of iterations (e.g., 20) is established. In each iteration, the following operations are performed:

[0089] A) Multi-objective optimization:

[0090] A non-dominated sorting genetic algorithm (NSGA-II) is used, with the current surrogate model (the first surrogate model, M_k in the k-th iteration) as the performance evaluation tool, for multi-objective optimization. The goal of this optimization is to simultaneously minimize S11_avg, G_null1, G_null2, and G_null3, and maximize Gain_peak.

[0091] After several generations of evolution, NSGA-II will output a current Pareto optimal solution set P_k.

[0092] Wherein, S11_avg: average reflection coefficient; G_null1: gain of transmission zero 1; G_null2: gain of transmission zero 2; G_null3: gain of transmission zero 3; Gain_peak: peak gain.

[0093] It should be noted that NSGA-II is a conventional genetic algorithm, but it has been improved compared to ordinary genetic algorithms by introducing non-dominated sorting and crowding distance to enhance its multi-objective optimization capabilities. This algorithm can not only find the optimal solution for a single objective, but also provide a multi-objective optimal solution with trade-offs among multiple objectives, which is crucial in the multi-objective optimization that is strongly required in antenna design.

[0094] B) Intelligent sample point selection (core of active learning):

[0095] From the Pareto optimal solution set P_k, select one or more new sample points X_new for accurate full-wave simulation. The criteria could be:

[0096] Expected Improvement Criterion: Select the point that is expected to bring the greatest improvement to the current best performance.

[0097] Uncertainty Sampling Criterion: If the surrogate model can output prediction uncertainty (such as using Gaussian process regression or Bayesian neural network), then select the point where the model prediction is most uncertain in order to most efficiently improve the model's accuracy in key areas.

[0098] In this embodiment, the solution with the largest crowding distance (most representative) on the Pareto front is selected as the new sample point X_new.

[0099] It should be noted that new sample points refer to candidate solutions in the optimization process, i.e., the positions of new input variables or design parameters in the solution space. In multi-objective optimization, each candidate solution corresponds to a specific set of structural variables for the antenna design. Expected improvement criterion: The expected performance improvement at a given point relative to the current optimal solution. Assuming x is a candidate solution, f(x) is the predicted value of the surrogate model at point x, and f* is the objective function value corresponding to the currently known optimal solution, the goal is to find the maximum improvement of f(x) relative to f*. Uncertainty criterion: Selecting the points with the most uncertain predictions for more accurate simulations to improve model accuracy. By selecting those points with the highest uncertainty, the surrogate model can acquire more information in these regions, thereby improving the model's accuracy in critical areas.

[0100] C) Full-wave simulation and model update:

[0101] Input X_new into HFSS for accurate full-wave simulation to obtain its true performance response Y_new. Add the new data pair (X_new, Y_new) to the existing training set to form the updated training set T_k+1. Use T_k+1 to retrain or incrementally update the surrogate model to obtain a more accurate model M_k+1, i.e., the second surrogate model.

[0102] In other embodiments, instead of the number of iterations, a convergence criterion is set. Specifically, based on the convergence criterion of the Pareto front mean, after each iteration, the average of the objective functions of all solutions in the current Pareto optimal solution set is calculated. If the position of this average solution no longer moves significantly in the objective space, or the relative rate of change is less than a set threshold, the algorithm is considered to have converged. For example, the threshold δ = 0.5% can be set.

[0103] The loop terminates when the number of iterations reaches a preset value, or when the Pareto front no longer changes significantly in several consecutive iterations. The final output is no longer a single optimal solution, but the complete Pareto optimal front obtained in the last iteration.

[0104] S107 uses the second surrogate model to identify new input stimuli.

[0105] In this step, six parameters that have the most critical impact on the filtering characteristics are selected as input excitations, and their reasonable value ranges are set. The optimal solution for the output value is sought, that is, simultaneously minimizing S11_avg, G_null1, G_null2, and G_null3, and maximizing Gain_peak. After a series of operations, the final optimization result is: W = 8.8mm, L = 6mm, W1 = 3.7mm, L1 = 2.1mm, W2 = 1.8mm, L2 = 3.3mm, L... s1 =1.4mm,L s2 =1.05mm,W f1 =0.2mm,W f2 =0.17mm, G=0.1mm, G1=0.2mm, G2=0.2mm, G3=0.35mm, L w1 =1.6mm,L w2 =0.6mm, d=0.94mm, d1=0.17mm, R f =0.6mm, R1=0.125mm, R2=0.08mm, H1=0.09mm, H2=0.7mm, H3=0.02mm. S 11 Simulation data of gain, such as Figure 6 As shown in the simulation data, the antenna achieves a relative bandwidth of 26.5% and a maximum gain of 6.5 dBi within the passband, exhibiting typical bandpass filtering characteristics in its gain curve. Specifically, there are three radiating nulls on both sides of the passband, with a low-frequency radiating null located at 17.8 GHz, achieving a stopband gain suppression level of 16 dB; and two radiating nulls appearing at 32 GHz and 33.6 GHz respectively in the high-frequency band, increasing the gain suppression depth to 19.5 dB. These radiating nulls significantly improve the passband sideband characteristics without affecting the radiation performance within the passband, demonstrating excellent out-of-band suppression capability. Based on the above simulation results, the designed filter antenna successfully achieves synergistic optimization of out-of-band suppression and wide bandwidth. While ensuring high-gain radiation in the center frequency band, the out-of-band roll-off characteristic constructed through multiple radiating nulls fully verifies that the structure possesses both filtering frequency selection and directional radiation functions, meeting the design requirements.

[0106] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0107] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0108] like Figure 5 As shown, this embodiment provides a filter antenna design device based on glass integrated packaging. The filter antenna design device includes:

[0109] The determination module 201 is used to determine the parameters affecting the filtering characteristics based on the structure of the filtering antenna, and to use them as input excitation.

[0110] The generation module 202 is used to generate an initial dataset by combining the range of values ​​of the input stimulus and using the Latin hypercube sampling method. The initial dataset includes sample points and the performance response of the sample points.

[0111] The first training module 203 is used to train a deep neural network based on the initial dataset to obtain a first surrogate model.

[0112] The optimization module 204 is used to use the first agent model as a performance evaluation tool and employ a non-dominated sorting genetic algorithm to perform multi-objective optimization to obtain a Pareto optimal solution set.

[0113] Selection module 205 is used to select one or more new sample points from the Pareto optimal solution set according to a preset criterion for full-wave simulation, obtain the performance response of the new sample points, and form a new data pair.

[0114] The second training module 206 is used to merge the new data pair with the initial dataset, retrain or update the first proxy model, and obtain the second proxy model.

[0115] The recognition module 207 is used to recognize new input stimuli using the second surrogate model.

[0116] This embodiment provides a computer device including a processor, a memory, an input device, a display device, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the computer programs are executed by the processor, they implement the filtered antenna design method described in the above embodiment.

[0117] This embodiment provides a storage medium, which is a computer-readable storage medium, storing a computer program. When the computer program is executed by a processor, it implements the filter antenna design method of the above embodiment.

[0118] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0119] In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this embodiment, the computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0120] The computer-readable storage medium described above can be used to write computer programs for executing this embodiment in one or more programming languages ​​or combinations thereof. These programming languages ​​include object-oriented programming languages—such as Java, Python, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0121] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for designing a filter antenna based on glass integrated packaging, characterized in that, include: The parameters affecting the filtering characteristics are determined based on the structure of the filtering antenna and used as input excitation. Based on the range of values ​​of the input stimulus, an initial dataset is generated using the Latin hypercube sampling method. The initial dataset includes sample points and the performance response of the sample points. A deep neural network is trained based on the initial dataset to obtain the first surrogate model; Using the first agent model as a performance evaluation tool, a non-dominated sorting genetic algorithm is used for multi-objective optimization to obtain the Pareto optimal solution set. From the Pareto optimal solution set, one or more new sample points are selected according to a preset criterion for full-wave simulation to obtain the performance response of the new sample points and form a new data pair. The new data pair is merged with the initial dataset, and the first proxy model is retrained or updated to obtain the second proxy model; Using the second surrogate model, new input stimuli are identified; X =[ d,d 1 ,L s1 ,L s2 ,W p ,L p ],in: X Indicates the input stimulus; d Indicates the spacing between short-circuited TGV arrays; d 1 Indicates the spacing of vias within the same TGV array; L s1 Indicates the length of the first set of surface gaps; L s2 Indicates the length of the second set of surface gaps; W p Indicates the width of the parasitic patch; L p Indicates the length of the parasitic patch; The deep neural network includes: an input layer with 6 nodes corresponding to 6 input parameters; a first fully connected hidden layer with 32 nodes; a first ReLU activation layer connected after the first fully connected hidden layer; a first Dropout layer connected after the first ReLU activation layer; a second fully connected hidden layer with 64 nodes; a second ReLU activation layer connected after the second fully connected hidden layer; a second Dropout layer connected after the second ReLU activation layer; and a regression layer as the output layer, used to output the prediction performance metrics.

2. The filter antenna design method according to claim 1, characterized in that, The objectives of the multi-objective optimization search include: the average reflection coefficient; the gain value of multiple transmission zeros; and the peak gain value.

3. The filter antenna design method according to claim 1, characterized in that, The preset criteria are criteria used in active learning, including the expectation enhancement criterion and the uncertainty criterion.

4. A filter antenna based on glass integrated packaging, characterized in that, include: A dielectric substrate; A main radiating patch is disposed on the surface of the dielectric substrate; A metal floor, the metal floor being disposed within the dielectric substrate; A low-frequency radiation null point generation module, the low-frequency radiation null point generation module includes one or more sets of metallized through holes, the metallized through holes penetrating the dielectric substrate in the vertical direction, for electrically connecting the main radiating patch to the metal ground plane; A high-frequency radiation null point generation module, the high-frequency radiation null point generation module including multiple sets of gaps etched on the main radiation patch; A full-band out-of-band suppression enhancement module, wherein the full-band out-of-band suppression enhancement module includes at least one of the following: A parasitic patch, the parasitic patch being disposed on at least one side of the main radiating patch; A planar inverted-F antenna element, wherein the planar inverted-F antenna element is disposed on at least one side of the main radiating patch; The spacing parameters of the metallized through-hole array, the length parameters of the slots, and the size parameters of the parasitic patch or planar inverted-F antenna element are intelligently optimized by the filter antenna design method described in any one of claims 1-3.

5. The filter antenna according to claim 4, characterized in that, Also includes: A microstrip line; A power supply port, the power supply port being connected to the microstrip line; An L-shaped probe, one end of which is connected horizontally to the main radiating patch and the other end of which is connected vertically to the microstrip line.

6. A filter antenna design device based on glass integrated packaging, characterized in that, include: The determination module is used to determine the parameters affecting the filtering characteristics based on the filtering antenna structure, and to use them as input excitation. The generation module is used to generate an initial dataset by combining the range of values ​​of the input stimulus and using the Latin hypercube sampling method. The initial dataset includes sample points and the performance response of the sample points. The first training module is used to train a deep neural network based on the initial dataset to obtain a first surrogate model; The optimization module is used to use the first agent model as a performance evaluation tool and employ a non-dominated sorting genetic algorithm to perform multi-objective optimization to obtain a Pareto optimal solution set. The selection module is used to select one or more new sample points from the Pareto optimal solution set according to a preset criterion for full-wave simulation, obtain the performance response of the new sample points, and form a new data pair. The second training module is used to merge the new data pair with the initial dataset, retrain or update the first proxy model, and obtain the second proxy model. The recognition module is used to identify new input stimuli using the second agent model; X =[ d,d 1 ,L s1 ,L s2 ,W p ,L p ],in: X Indicates the input stimulus; d Indicates the spacing between short-circuited TGV arrays; d 1 Indicates the spacing of vias within the same TGV array; L s1 Indicates the length of the first set of surface gaps; L s2 Indicates the length of the second set of surface gaps; W p Indicates the width of the parasitic patch; L p Indicates the length of the parasitic patch; The deep neural network includes: an input layer with 6 nodes corresponding to 6 input parameters; a first fully connected hidden layer with 32 nodes; a first ReLU activation layer connected after the first fully connected hidden layer; a first Dropout layer connected after the first ReLU activation layer; a second fully connected hidden layer with 64 nodes; a second ReLU activation layer connected after the second fully connected hidden layer; a second Dropout layer connected after the second ReLU activation layer; and a regression layer as the output layer, used to output the prediction performance metrics.

7. A computer device comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the filter antenna design method according to any one of claims 1-3.

8. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the filter antenna design method according to any one of claims 1-3.

Images