Method for forming frequency selective surface filter and recording medium recording the same

Abstract

A frequency selection filter generation method performed by a manager and multiple clients comprises: a step in which the manager generates a trial solution; a first local optimization step in which each of the multiple clients locally optimizes the trial solution to generate a local optimal solution and incorporates it into a candidate solution by computing an objective function; a genetic change step in which the manager genetically changes the candidate solution; a second local optimization step in which each of the multiple clients performs local optimization on the genetically changed candidate solution to generate a second local optimal solution; and a frequency selection filter generation method in which the manager determines whether the second local optimal solution will replace the candidate solution or decides to discard the second local optimal solution.

Description

Technology Field

[0001] The present disclosure generally relates to a method for generating a frequency-selective filter and a recording medium recording the same. Background Technology

[0002] Since the frequency response characteristics of FSS filters vary significantly depending not only on the geometric shape of the structure selected as the unit cell but also on the arrangement and period of the unit cells, as well as the material properties of the dielectric and conductor materials used as substrates supporting these unit cells, various methods have been studied and proposed to accurately obtain the frequency characteristics desired by the user.

[0003] Meanwhile, genetic algorithms are algorithms based on the fundamental theories of biological genetics in the natural world, taking Darwin's theory of survival of the fittest as their basic concept. Genetic algorithms represent possible solutions to a problem in a defined data structure and then generate increasingly better solutions by gradually modifying them. Here, the data structure representing the solutions is the gene, and the process of generating increasingly better solutions by modifying them can be expressed as evolution.

[0004] These genetic algorithms may include crossover and mutation. Crossover operations generally involve selecting multiple solutions and performing crossover operations between them; the resulting solution then forms a new gene by receiving genetic factors at non-overlapping positions through crossover operations of each parent solution. Mutation operations are operations in which the order or value of genetic factors within a given solution's gene is arbitrarily changed to transform it into a different solution. The problem to be solved

[0005] In conventional technology, FSS filters were designed by gradually adjusting frequency characteristics through changes and modifications to the arrangement of unit cells to achieve a target frequency response set by the designer. This involved performing time-consuming simulations on the design results and repeatedly modifying the design based on the simulation findings. Furthermore, genetic algorithms require high computing power because they evaluate frequency characteristics by forming numerous variations to approximate the desired result. Consequently, a challenge arose in that designing an FSS filter with the desired frequency response required significant time and manpower.

[0006] One of the problems to be solved by this embodiment is to resolve the difficulties of the aforementioned prior art. means of solving the problem

[0007] One aspect of the present disclosure relates to a frequency selection filter generation method performed by a manager and a plurality of clients, wherein the frequency selection filter generation method comprises: a step in which the manager generates a trial solution; a first local optimization step in which each of the plurality of clients locally optimizes the trial solution to generate a local optimal solution and computes an objective function to incorporate it into a candidate solution; a genetic change step in which the manager genetically changes the candidate solution; a second local optimization step in which each of the plurality of clients performs local optimization on the genetically changed candidate solution to generate a second local optimal solution; and the manager determines whether the second local optimal solution will replace the candidate solution or decides to discard the second local optimal solution to generate a frequency selection filter.

[0008] Another aspect of the present disclosure is a recording medium readable by an electronic device, wherein a frequency selection filter generation method performed by a manager and a plurality of clients is recorded thereon, said frequency selection filter generation method comprises: a step in which the manager generates a trial solution; a first local optimization step in which each of the plurality of clients locally optimizes the trial solution to generate a local optimal solution and computes an objective function to incorporate it into a candidate solution; a genetic change step in which the manager genetically changes the candidate solution; a second local optimization step in which each of the plurality of clients performs local optimization on the genetically changed candidate solution to generate a second local optimal solution; and a manager determines whether the second local optimal solution will replace the candidate solution or decides to discard the second local optimal solution to generate a frequency selection filter. In any aspect of the present embodiment, each of the manager and the plurality of clients is a computer program executed in a computing device comprising a CPU and memory.

[0009] In one aspect of the present embodiment, the first local optimization step and the second local optimization step include the step of each of the plurality of clients modifying the sequence of the attempted solution, the step of calculating the objective function value of the attempted solution having the modified sequence, and the step of setting the attempted solution with the modified sequence as the local optimal solution if the frequency response of the attempted solution with the modified sequence corresponds more to the frequency response provided by the user than to the frequency response of the attempted solution from the objective function value. In this aspect, the step of modifying the sequence of the attempted solution is performed using any one of the Nelder-Mead algorithm, the annealing simulation algorithm, the formational space annealing algorithm, the conjugated direction method, and the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm.

[0010] In one aspect of the present embodiment, the genetic change step is performed as either a mutation that changes one or more digits of the local optimal solutions or a crossover that replaces the sequences of two or more local optimal solutions.

[0011] In any aspect of the present embodiment, the frequency selection filter generation method further comprises the step of the manager setting a cutoff distance from the candidate solution generated in the first local optimization step, wherein the step of setting the cutoff distance includes: the step of calculating Hamming distances between the candidate solutions, the step of calculating the average of the Hamming distances, and the step of setting the average value of the Hamming distances as the cutoff distance. In this regard, the step of the manager deciding to discard and replace the second local optimal solution comprises: determining the nearest candidate solution, which is the candidate solution closest to the second local optimal solution, from the Hamming distance calculation result of the candidate solutions; and if the distance between the second local optimal solution and the nearest candidate solution is smaller than the current cutoff distance value, comparing the objective function value of the second local optimal solution with the objective function value among the nearest candidate solutions, discarding the second local optimal solution when the frequency response characteristics of the nearest candidate solution are more similar to the target frequency response characteristics than the frequency response characteristics of the second local optimal solution, and discarding the nearest candidate solution when the frequency response of the second local optimal solution is more similar to the target frequency response than the frequency response of the nearest candidate solution, and replacing the second local optimal solution with the nearest candidate solution.Additionally, in another aspect, the step of the manager deciding to discard and replace the second local optimal solution comprises: selecting a candidate solution having frequency response characteristics that are least similar to the target frequency response characteristics from the objective function; comparing the objective function of the second local optimal solution; and if, based on the objective function value of the second local optimal solution, the frequency response characteristics of the second local optimal solution are more similar to the target frequency response characteristics than the frequency response characteristics of the candidate solution, the second local optimal solution replaces the candidate solution, and if, based on the objective function value of the candidate solution, the frequency response characteristics of the candidate solution are more similar to the target frequency response characteristics than the second local optimal solution, the second local optimal solution is discarded. In another aspect, after the step of the manager deciding to replace or discard the second local optimal solution, the manager further includes the step of reducing the cutoff distance. Effects of the invention

[0012] According to the present embodiment, it is possible to design an FSS filter having a target general frequency response using a highly efficient method and computer calculation for combinatorial pattern generation, which is virtually impossible with conventional technology. Brief explanation of the drawing

[0013] FIG. 1(a) is a schematic plan view of an FSS filter, and FIG. 1(b) is a schematic cross-sectional view of a part of an FSS filter. FIG. 2 is a block diagram schematically illustrating the entities performing the FSS filter design method of the present embodiment. FIG. 3 is a flowchart illustrating an overview of the FSS filter design method of the present embodiment. Figure 4 is a diagram illustrating the objective function. Figure 5 is a diagram illustrating the results of local optimization. Figure 6 is a diagram illustrating the steps for a manager to generate a solution by attempting a mutation. Referring to Figure 7, this is a diagram illustrating the steps for a manager to generate an intersection attempt solution. Figure 8 is a diagram showing the speed at which a frequency selectable surface filter is formed while varying the number of processors running on the client from 1 to 50. FIG. 9(a) is a diagram illustrating the shape of a frequency selective surface filter according to the present embodiment. FIG. 9(b) is a frequency response characteristic calculated from the FSS filter shape obtained through an optimization process. FIG. 10(a) is a diagram illustrating the shape of a frequency selective surface (FSS) filter according to the present embodiment. FIG. 10(b) is a frequency response characteristic calculated from the FSS filter shape obtained through an optimization process. FIG. 11(a) is a diagram illustrating the shape of a frequency selective surface filter according to the present embodiment. FIG. 11(b) is a frequency response characteristic calculated from the FSS filter shape obtained through an optimization process. Specific details for implementing the invention

[0014] Hereinafter, a method for designing a frequency selective surface (FSS) filter according to the present embodiment will be described with reference to the attached drawings. FIG. 1(a) is a plan view schematically illustrating an FSS filter (10), and FIG. 1(b) is a cross-sectional view schematically illustrating a cross-section of a part of the FSS filter. The FSS filter (10) of FIG. 1 can be formed by arranging a plurality of unit FSS filters in an array form. Referring to FIG. 1(a) and FIG. 1(b), the FSS filter (10) includes a plurality of unit cells (120) arranged in a square. For example, the unit cells (120) may include a unit cell (120b) filled with a metal film and an empty unit cell (120a). A dielectric layer (140) may be located below the metal film.

[0015] The frequency selection characteristics of the FSS filter (10) vary depending on how the unit cells (120) are filled with metal. In the illustrated example, the FSS filter (10) may be formed by arranging 20 square-shaped unit cells (120) horizontally and 20 vertically. However, this is merely an example, and the number, shape, and form of the unit cells (120) may vary.

[0016] The outer periphery (130) of the FSS filter (10) may be filled with metal. Since the FSS filter (10) assumes infinite periodicity, the outer periphery (130) can be filled with metal to cover it, thereby obtaining the infinite periodicity required by the FSS filter (10). In the example illustrated in FIG. 1, the unit cells (130) of the outer periphery are not included in the irreducible region and remain filled with a conductor.

[0017] The outer periphery (130) of the FSS filter (10) may be formed of metal (120b) and empty unit cells (120a). In this case, the irreducible region includes the unit cells of the outer periphery, and the FSS filter (10) can obtain the infinite periodicity required in the FSS filter (10) by covering the outer periphery of the irreducible region with metal and empty unit cells.

[0018] An irreducible zone (110) is extracted from the surface of the FSS filter (10). The irreducible zone (110) refers to a minimum size representation area that can cover the entire surface of the FSS filter (10) using symmetry. The irreducible zone (110) is illustrated in FIG. 1(a). Each unit cell belonging to the irreducible zone (110) may be designated in a zigzag manner, such as 1–9, 10–17, 18–24, 25–30, 31–35, 36–39, 40–42, 43–44, and 45, as exemplified. However, this is merely an example for distinguishing and designating unit cells (120) in a specific area of ​​the irreducible zone (110), and is not intended to limit the scope of the invention by restricting the method of designating unit cells within the irreducible zone.

[0019] In this way, unit cells within the reduced area (110) can be referred to as a one-dimensional sequence, and any unit cell (120) within the reduced area (110) can be represented as a single digit within the sequence. Here, the digit is indicated as 1 or 0. 1 refers to a unit cell filled with metal. 0 refers to an empty unit cell. Through such representation, intersection and mutation operations can be processed concisely.

[0020] Since electromagnetic waves can be incident on the installed filter at various angles, the FSS filter needs to have rotational symmetry. To ensure the same filter performance for various incident angles, the filter itself must be designed to have rotational symmetry. A reduced region is introduced in consideration of these physical conditions. Therefore, by determining only the reduced region through an optimization process, the shape of the filter having rotational symmetry can be determined. The reduced region, expressed in this way in a one-dimensional array format, is extended to the entire surface of the FSS filter (10) by applying the symmetry required by the FSS filter.

[0021] In one embodiment, a target frequency response is set, which is a frequency response to be implemented by the FSS filter (10). The target frequency response can determine characteristics such as the passband frequency [Hz], cutband frequency [Hz], signal magnitude [dB] of the passband, and signal magnitude [dB] of the cutband, which are to be formed by the FSS filter (10).

[0022] FIG. 2 is a block diagram schematically illustrating entities that perform the FSS filter design method of the present embodiment. Referring to FIG. 2, the FSS filter design method of the present embodiment is performed by a manager (100) that performs FSS filter design calculations and manages the design method, and a plurality of clients (200a, 200b, ..., 200n) that perform FSS filter design calculations.

[0023] The manager (100) may be a program executed by an arithmetic unit including a CPU and memory. Clients (200a, 200b, ..., 200n) may be programs implemented by an arithmetic unit including a CPU and memory. In one embodiment, the manager (100) and clients (200a, 200b, ..., 200n) may be Python, Julia, C, C ++ It may be written in computer languages ​​such as, but these are merely examples and are not intended to limit the manager (100) and clients (200a, 200b, ..., 200n).

[0024] FIG. 3 is a flowchart illustrating an overview of the FSS filter design method of the present embodiment. Referring to FIGS. 1 to 3, the FSS filter design method of the present embodiment comprises the steps of: a manager generating a trial solution (S100); each client (200a, 200b, ..., 200n) forming a candidate solution from the trial solution (S200); the manager (100) calculating the distance between the candidate solutions and setting a blocking distance (S300); a step of selecting a mutation and an intersection (S400); if a mutation is selected, the manager (100) generating a mutated trial solution (S500a); if an intersection is selected, the manager (100) generating an intersection trial solution (S500b); each client (200a, 200b, ..., 200n) locally optimizing the generated trial solution to generate a local optimal solution (S600); and the manager (100) comparing the generated local optimal solution with a candidate solution to replace or discard it (S700). The manager (100) includes a step (S800) of reducing the blocking distance and a step (S900) of the manager (100) determining whether to continue.

[0025] When the FSS filter design method is executed, the manager (100) provides data capable of performing automated operations to each of the computing devices where the clients (200a, 200b, ..., 200n) are executed. In one embodiment, the data for performing automated operations may be a batch script program located in a folder within each client (200a, 200b, ..., 200n) that performs a predetermined operation. The batch script may include code for performing local optimization, which will be described later.

[0026] The manager (100) generates (S100) a trial solution corresponding to a sequence corresponding to the arrangement of unit cells within the promised area (110) and provides it to a folder formed in each client (200a, 200b, ..., 200n). For example, the sequence of trial solutions generated by the manager (100) may be formed from random values. For another example, the sequence of trial solutions initially generated may be a sequence provided by a user.

[0027] Each of the clients (200a, 200b, ..., 200n) performs local optimization with the provided trial solution to obtain a local optimal solution and generates candidate solutions based on it (S200). The local optimization process is a process of modifying a part of the sequence included in the trial solution so that the frequency response of the trial solution approaches the target frequency response provided by the user. The number of candidate solutions may be equal to the number of clients, and may also be greater or less than the number of clients.

[0028] FIG. 4 is a diagram illustrating the objective function. Referring to FIG. 4, the solid line represents the target frequency response provided by the user, and the dashed line represents the frequency response of the FSS filter formed according to a particular experimental solution. The objective function is calculated from the numerical difference between the target frequency response and the frequency response of the FSS filter formed by the experimental solution. The target frequency response may include information such as the passband nodal frequency (a, b), the cut-off band frequency (a0, b0), the passband signal magnitude (Y1), and the cut-off band signal magnitude (Y2). In FIG. 4, d1, d2, and d3 represent the difference between the target frequency response and the frequency response of the filter formed by the experimental solution.

[0029] The objective function may, for example, be a function obtained by integrating the difference between the target frequency response—expressed as a numerical function directly specified by the user representing the frequency response characteristics desired by the user in a frequency band—and the frequency response characteristic function calculated from the trial solution, and the frequency band may be any one of all bands or the frequency band of interest. In one embodiment, the target frequency response may therefore be set as a frequency response function corresponding to any filter set as a target, such as a bandpass filter or a band-stop filter.

[0030] As another example, the objective function may be a function obtained by squaring the difference between the objective function corresponding to a preset target frequency response and the frequency response characteristic function calculated from the candidate solution over all frequency intervals and summing them. As yet another example, the objective function may be a function obtained by calculating the absolute values ​​of the differences (d1, d2, d3) in the frequency response and summing them. By calculating the objective function in this way, it is possible to prevent the difference values ​​from being canceled out by the signs of the frequency response differences (d1, d2, d3).

[0031] The embodiment illustrated in FIG. 4 simply compares the amplitude of the target frequency response at three frequencies with the amplitude of the frequency response of the filter formed by the trial solution, but this is merely a schematic example for illustrative purposes, and the objective function compares the amplitude of the target frequency response with the amplitude of the frequency response of the filter formed by the trial solution at a resolution of every 1 kHz to 100 MHz. In one embodiment, the objective function can be implemented in either a CST or Julia-based simulation tool, and the frequency response for the corresponding unit cell structure can be obtained by inputting a unit cell sequence. These are merely examples and are not intended to limit the electromagnetic wave analysis simulation tool.

[0032] Local optimization can be performed using the Nelder-Mead algorithm or the Simulated Annealing algorithm. The Nelder-Mead algorithm uses a simplex, a special polyhedron with n+1 vertices in an n-dimensional space. The Nelder-Mead algorithm evaluates function values ​​at the vertices of the simplex and transforms the simplex by replacing the worst vertex with a new vertex.

[0033] The simulated solution algorithm is an algorithm that accepts a modified solution as the latest solution whenever the objective function value decreases among randomly modified solutions. In addition to this, it can be implemented using the conjugated direction method, the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm, etc.

[0034] Simulated annealing is a probabilistic optimization algorithm that mimics the process of solid crystallization by slowly lowering the temperature to find the lowest energy state. It utilizes the movement of atoms, which can move relatively freely at high temperatures. In simulated annealing, optimization performance can be controlled by adjusting the initial temperature and cooling rate. The initial temperature is generally set not to be too high, and the cooling rate is slow to ensure sufficient time to find an approximate solution. Simulated annealing is useful for solving global optimization problems, and this algorithm is utilized in various fields.

[0035] The formational space annealing algorithm is a global optimization method that appropriately combines genetic algorithms, local optimization, and simulation of annealing. It actively seeks diverse and distinct solutions, processing multiple solutions simultaneously rather than handling just a single one. It defines and actively utilizes the distance between different solutions. This can be viewed as a mechanism to ensure diversity among the obtained solutions, and as a probabilistic approach, it is applicable when there are a very large number of independent variables. In particular, when local energy minimization and objective function calculations are time-consuming, the problem can be solved through algorithm parallelization. It is an algorithm with excellent parallel efficiency and very simple parallelization.

[0036] In the embodiment illustrated in FIG. 1, 45 digits of the irreducible region are set as vertices of a simplex, and transformations of sequence digits are performed on them and repeated until the objective function value is improved. The transformation evaluates the function value at the current vertex of the simplex to determine the worst, second worst, and best points. Subsequently, the process may be carried out in the steps of reflection, which creates a new point by reflecting the worst point through the midpoint of the other two points; expansion, which creates a new point by moving further in that direction if the reflected point is better; and contraction, which creates a new point by contracting to the midpoint if the function value is not improved by the reflected point.

[0037] If the objective function value calculated as a result of performing the aforementioned reflection, expansion, and contraction steps is not improved, the attempted solution is discarded, the client performing local optimization modifies the attempted solution, and applies the local optimization algorithm to the new attempted solution.

[0038] FIG. 5 is a diagram illustrating the results of performing local optimization. Referring to FIG. 5, when local optimization is performed, the values ​​assigned to some digits of the randomly generated trial solution change, and the objective function value for the changed trial solution also decreases, allowing it to approach the target frequency response more closely. In one embodiment, local optimization can be performed on multiple digits, such as LO1 and LO3, or applied to a single digit, such as LO2.

[0039] Each client (200a, 200b, ..., 200n) forms an end flag indicating that the local optimization process for the corresponding attempt solution has ended when the local optimization step is completed. For example, the manager (100) can determine that the local optimization process for the corresponding attempt solution on the client has ended by monitoring the end flag in the operation folder formed on each client (200a, 200b, ..., 200n). For example, the end flag may be a file indicating that the local optimization process for the corresponding attempt solution has ended. For another example, the end flag may be a notification containing a specific bit sequence provided by the client to the manager.

[0040] The manager (100) monitors the client's termination flag. The local optimal solution formed when the local optimization process is completed becomes a candidate solution, and the manager (100) brings the formed candidate solution and the objective function value calculated with the candidate solution to a computing device where the manager (100) program is executed. The process of forming the above-described multiple candidate solutions (S200) is performed in parallel by each of the multiple clients (200a, 200b, ..., 200n).

[0041] The manager (100) forms candidate solutions in all clients (200a, 200b, ..., 200n), collects objective function values ​​calculated from the candidate solutions, and calculates the distance between the candidate solutions (S300). In one embodiment, the distance refers to a value obtained by comparing the sequence of one candidate solution with the sequence of another candidate solution in order, digit-wise, assigning a value of 0 to identical digits and a value of 1 to different digits as a result of the comparison, and summing the values. That is, the distance between the candidate solutions is equal to the Hamming distance of the candidate solutions.

[0042] For example, the fact that the distance between one candidate solution 1 and another candidate solution 2 is 3 means that there are 3 different digits in each candidate solution. The distance between candidate solutions can also be called the similarity (or dissimilarity) between candidate solutions.

[0043] In one embodiment, the manager (100) sets a cut-off distance. For example, the cut-off distance may be set to half of the average distance between the calculated candidate solutions. However, the set cut-off distance may be adjusted in a subsequent process (S800). For example, the cut-off distance may be adjusted so that it decreases to a value 0.97 times smaller each time the design method according to the present embodiment is repeated and an intersection or variation is attempted 1 to 100 times. However, the cut-off distance is set so that it does not become smaller than a specific value. For example, the cut-off distance is set so that it does not become smaller than 1 / 5 of the average distance between the calculated candidate solutions.

[0044] Next, the candidate solutions are genetically modified using a genetic algorithm (S400). In one embodiment, the genetic modification of the candidate solutions may be performed by performing a mutation on the candidate solutions to form a mutated attempt solution (S500a), or by performing a crossover to form a crossover attempt solution (S500b).

[0045] In one embodiment, whether to perform a crossover or a mutation on a candidate solution can be determined by a random number. For example, a random number generator (not shown) can output a value between 0 and 1, and either a crossover or a mutation can be selected based on the relationship with a threshold value of 0.5.

[0046] Referring to FIG. 6, the step (S500a) of generating a trial solution as a mutation by the manager (100) is described. FIG. 6 is an operation that inverts the values ​​assigned to a finite number of digits in a candidate solution (C3) selected as a mutation. Here, the inversion of the state for the digits means from 1 to 0 or from 0 to 1. The embodiment illustrated in FIG. 6 illustrates that mutations occur in three consecutive unit cells among the candidate solutions (C3). The location of the unit cells where mutations occur and the number of unit cells are both randomly determined by a random number generator (not shown).

[0047] Referring to FIG. 7, the step (S500b) of the manager (100) generating a crossover solution is described. The embodiment illustrated in FIG. 7 illustrates candidate solutions (C1, C2) and a crossover solution (T1) represented as a sequence containing 13 digits. The manager (100) generates a new crossover solution (T1) by crossing the unit cells of the shaded candidate solutions (C1, C2). The position of the digits and the number of digits at which the crossover occurs are both randomly determined by a random number generator (not shown).

[0048] In an embodiment for selecting candidate solutions for crossover or mutation, the manager (100) may randomly select different candidate solutions C1 and C2 for crossover or candidate solution C3 for mutation from among a plurality of candidate solutions. In another embodiment for selecting candidate solutions, the manager (100) selects candidate solutions for mutation or crossover with a higher probability the more similar they are to the target frequency response from among a plurality of candidate solutions. That is, the lower the objective function value due to similarity to the target frequency response, the higher the probability of being subject to crossover or mutation operations.

[0049] The manager (100) selects candidate solutions for crossover or mutation by choosing either a single tournament method or a Poisson's distribution method. In the single tournament method, two different candidate solutions are selected randomly first, and the candidate solution with the smaller objective function value is finally selected. In the Poisson's distribution method, when sorted in ascending order based on the objective function value, a Poisson distribution function is created using the average rank and rank deviation to finally select the candidate solutions. The selected rank probabilistically follows the Poisson distribution. The best solution has the smallest objective function value and its rank is 1st.

[0050] The number of trial solutions formed by the manager (100) performing crossover or mutation may correspond to the number of clients (200a, 200b, ..., 200n). The manager (100) provides the trial solutions generated by performing crossover or mutation to the clients (200a, 200b, ..., 200n). Each client (200a, 200b, ..., 200n) performs the local optimization described above with the provided trial solutions (S600) and calculates the objective function value for each trial solution.

[0051] In one embodiment, the step of performing local optimization (S600) may be the same as the step of each client (200a, 200b, ..., 200n) generating a candidate solution using a trial solution generated by the manager (100). In another embodiment, the step of each client (200a, 200b, ..., 200n) generating a local optimal solution using a trial solution generated by the manager (100) may be performed using the Nelder-Mead algorithm or Simulated Annealing.

[0052] In conventional technology, local optimization is performed by computing an optimization algorithm for multiple trial solutions one by one, computing an objective function based on the calculated result, and if the objective function value increases, discarding the trial solution and generating another trial solution to compute the optimization algorithm, which consumes a long time. Furthermore, as illustrated in FIG. 3, this local optimization is included in a repeating loop, which consumes a long time.

[0053] However, in this embodiment, multiple clients are used to perform local optimization in parallel. The local optimization process is the most time-consuming computational process and requires a large number of local optimization processes. Therefore, it can be observed that the computation time associated with optimization is reduced as the number of clients increases.

[0054] The manager (100) collects local optimal solutions formed by completing local optimization from each client (200a, 200b, ..., 200n) and decides whether to discard or replace them (S700). In one embodiment, the distance between the local optimal solution and candidate solutions is calculated after local optimization of the trial solution generated through variation or intersection. From the result of calculating the distance between the local optimal solution and candidate solutions, the nearest candidate solution, which is the candidate solution closest (similar) to the local optimal solution, is determined.

[0055] If the distance between the local optimum and the nearest candidate solution is smaller than the current cutoff distance, the objective function value of the local optimum is compared with that of the nearest candidate solution. If the objective function value of the local optimum is greater than that of the nearest candidate solution (i.e., if the frequency response characteristics of the nearest candidate solution are more similar to the target frequency response characteristics than those of the attempted solution), the local optimum is discarded.

[0056] On the other hand, if the objective function value of the local optimum is smaller than that of the nearest candidate solution (i.e., if the frequency response of the local optimum is more similar to the target frequency response than the frequency response of the nearest candidate solution), the local optimum replaces the nearest candidate solution, and the existing nearest candidate solution is discarded.

[0057] If the distance between the local optimum and the nearest candidate solution is equal to or greater than the current cutoff distance value, the objective function value of the local optimum is compared with the objective function value of the candidate solution having the largest objective function value among the existing candidate solutions (i.e., the candidate solution having the frequency response characteristics least similar to the desired frequency response characteristics).

[0058] If the objective function value of a local optimum is smaller than the largest objective function value among the candidate solutions (i.e., the frequency response characteristics of the local optimum are more similar to the target frequency response characteristics than those of the candidate solution being compared), the local optimum replaces the candidate solution being compared, and the candidate solution being compared is discarded. Conversely, if the objective function value of a local optimum is larger than the largest objective function value among the candidate solutions (i.e., the frequency response characteristics of the candidate solution being compared are more similar to the target frequency response than those of the local optimum), the local optimum is discarded.

[0059] Through these processes, local optimal solutions are included in the group formed by existing candidate solutions and become new candidate solutions, and among the existing candidate solutions, those that are not similar to the target frequency response characteristics are discarded. Therefore, the frequency response characteristics of the FSS filter (10) formed by the candidate solutions belonging to the candidate solution group gradually approach the target frequency response characteristics.

[0060] The cut-off distance value is reduced (S800). The cut-off distance value is the criterion for determining when a local optimum replaces a candidate solution. Therefore, since different types of local optimums can be replaced by a candidate solution group, diversity can be secured in the candidate solution group. This replacement method ensures diversity among candidate solutions and is an operation that cannot be found in existing genetic algorithms.

[0061] It is determined whether to continue the above-mentioned process (S900). In one embodiment, the decision to continue may be based on the variation of the objective function value. If iterations are performed several times based on the objective function value but the objective function value no longer decreases, or if the variation of the objective function value is small, the process is terminated. A plurality of unit FSS filters (10) designed in this manner can be arranged in an array form to form an FSS filter.

[0062] Simulation example

[0063] The following describes a simulation example with reference to the attached drawings. The manager computer program for executing the frequency-selective surface filter design method according to the present embodiment and the client's local optimization program are written in the Python language, and the creation of files indicating the start and end notifications of the client's local optimization program is written in a shell script.

[0064] Frequency response characteristic calculations for the trial solutions are performed using PSSFSS (polarization selective surfaces frequency selective surfaces), a Julia-language-based package for polarization and frequency selective surface analysis. The calculation results using the PSSFSS package are saved in file format. These files are read by a Python program.

[0065] The manager computer program monitors termination notifications generated by the client's shell script and collects local optimal solutions and objective function results. Subsequently, the manager program generates a new attempt solution and inputs it into the client, and the client starts a local optimization program for the new attempt solution via the shell script.

[0066] A simulation example is described below with reference to FIG. 8. FIG. 8 is a diagram showing the speed of forming a frequency-selective surface filter while varying the number of processors running on the client from 1 to 50. As illustrated in FIG. 8, it can be seen that the speed of forming the frequency-selective surface filter increases linearly as the number of processors running on the client increases. This is because more computations can be performed simultaneously with more computer resources in the same amount of time.

[0067] This confirms that the local optimization process is the step that consumes the most time in the process of forming frequency-selective surface filters, and that a larger number of frequency-selective surface filters are formed by a large number of processors performing local optimization in parallel.

[0068] In the following, a frequency-selective surface filter design method according to the present embodiment is described for 0.1×0.1 mm 2 Unit cell size, total size including 27×27 unit cells 2.7×2.7 mm 2 An FSS filter was designed. The number of unit cells corresponding to the reduced region is 105, including the outer periphery (130) of the FSS filter (10), and the number of possible combinations from this is 2 105 It is a branch.

[0069] FIG. 9(a) is a diagram illustrating the shape of a frequency selective surface (FSS) filter according to the present embodiment. FIG. 9(b) is a frequency response characteristic calculated from the FSS filter shape obtained through an optimization process. This is close to the target wideband response characteristic, with a center frequency of 60 GHz and a passband from 59.5 GHz to 60.5 GHz (based on transmission loss of 1 dB or less).

[0070] FIG. 10(a) is a diagram illustrating the shape of a frequency selective surface (FSS) filter according to the present embodiment. FIG. 10(b) is a frequency response characteristic calculated from the FSS filter shape obtained through an optimization process. This is close to the target wideband response characteristic, with a center frequency of 62.5 GHz and a passband of 62 GHz to 63 GHz (based on transmission loss of 10 dB or more).

[0071] FIG. 11(a) is a diagram illustrating the shape of a frequency selective surface (FSS) filter according to the present embodiment. FIG. 11(b) is a frequency response characteristic calculated from the FSS filter shape obtained through an optimization process. This is close to the target dual-band response characteristics, with a first center frequency of 60 GHz and a passband from 59.5 GHz to 60.5 GHz (based on transmission loss of 1 dB or less), and a second center frequency of 78 GHz and a passband from 76.5 GHz to 79.5 GHz (based on transmission loss of 1 dB or less).

[0072] To aid in understanding the present invention, the embodiments illustrated in the drawings have been described with reference to the examples shown; however, these are merely illustrative examples for implementation, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the present invention should be determined by the appended claims. Explanation of the symbols

[0073] 10: FSS filter 100: Manager 110:Irreducible domain 120: Unit cell 130: Outsourcing 140: Genome layer 200: Client

Claims

Claim 1 A frequency selection filter generation method performed by a manager and multiple clients, wherein the frequency selection filter generation method comprises: a step in which the manager generates a trial solution; a first local optimization step in which each of the multiple clients locally optimizes the trial solution to generate a local optimal solution and calculates an objective function to incorporate it into a candidate solution; a genetic change step in which the manager genetically changes the candidate solution; a second local optimization step in which each of the multiple clients performs local optimization on the genetically changed candidate solution to generate a second local optimal solution; and a frequency selection filter generation method in which the manager determines whether the second local optimal solution will replace the candidate solution or decides to discard the second local optimal solution. Claim 2 A method for generating a frequency selection filter according to claim 1, wherein each of the manager and the plurality of clients is a computer program executed in a computing device including a CPU and memory. Claim 3 A method for generating a frequency selection filter according to claim 1, wherein the first local optimization step and the second local optimization step comprise the steps of each of the plurality of clients modifying the sequence of the attempt solution, calculating the objective function value of the attempt solution having the modified sequence, and setting the attempt solution with the modified sequence as the local optimal solution if the frequency response of the attempt solution with the modified sequence corresponds more to the frequency response provided by the user than to the frequency response of the attempt solution from the objective function value. Claim 4 In paragraph 3, the step of modifying the sequence of the above-mentioned attempts is a frequency-selective filter generation method performed using any one of the Nelder-Mead algorithm, the annealing simulation algorithm, the conformational space annealing algorithm, the conjugate direction method, and the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm. Claim 5 A method for generating a frequency-selective filter according to claim 1, wherein the genetic change step is performed as either a mutation that changes one or more digits of the local optimal solutions or a crossover that replaces the sequences of two or more local optimal solutions. Claim 6 The method for generating a frequency selection filter according to claim 1 further comprises the step of the manager setting a cutoff distance from the candidate solution generated in the first local optimization step, wherein the step of setting the cutoff distance includes: the step of calculating Hamming distances between the candidate solutions, the step of calculating the average of the Hamming distances, and the step of setting the average value of the Hamming distances as the cutoff distance. Claim 7 In claim 6, the step of the manager determining the discarding and replacement of the second local optimal solution comprises: determining the nearest candidate solution, which is the candidate solution closest to the second local optimal solution, from the Hamming distance calculation result of the candidate solutions; and if the distance between the second local optimal solution and the nearest candidate solution is smaller than the current cutoff distance value, comparing the objective function value of the second local optimal solution with the objective function value among the nearest candidate solutions, discarding the second local optimal solution when the frequency response characteristic of the nearest candidate solution is more similar to the target frequency response characteristic than the frequency response characteristic of the second local optimal solution, and discarding the nearest candidate solution when the frequency response of the second local optimal solution is more similar to the target frequency response than the frequency response of the nearest candidate solution, and the second local optimal solution is a method for generating a frequency selection filter that replaces the nearest candidate solution. Claim 8 A method for generating a frequency selection filter according to claim 6, wherein the step of the manager determining to discard and replace the second local optimal solution comprises: selecting a candidate solution having a frequency response characteristic that is least similar to the target frequency response characteristic from an objective function; comparing the objective function of the second local optimal solution; and if, from the objective function value of the second local optimal solution, the frequency response characteristic of the second local optimal solution is more similar to the target frequency response characteristic than the frequency response characteristic of the candidate solution, the second local optimal solution replaces the candidate solution, and if, from the objective function value of the candidate solution, the frequency response characteristic of the candidate solution is more similar to the target frequency response characteristic than the second local optimal solution, the second local optimal solution is discarded. Claim 9 A method for generating a frequency selector filter according to claim 6, further comprising the step of the manager reducing the cutoff distance after the step of the manager determining to replace or discard the second local optimal solution. Claim 10 A recording medium readable by an electronic device, wherein a method for generating a frequency-selective filter performed by a manager and multiple clients is recorded and the method for generating the frequency-selective filter is recorded as a program and readable by an electronic device, wherein the method comprises: a step in which the manager generates a trial solution; a first local optimization step in which each of the multiple clients locally optimizes the trial solution to generate a local optimal solution and calculates an objective function to incorporate it into a candidate solution; a genetic change step in which the manager genetically changes the candidate solution; a second local optimization step in which each of the multiple clients performs local optimization on the genetically changed candidate solution to generate a second local optimal solution; and a recording medium readable by an electronic device such that a method for designing a frequency-selective surface filter is executed by the manager determining whether the second local optimal solution will replace the candidate solution or whether to decide to discard the second local optimal solution. Claim 11 In paragraph 10, each of the above manager and the above plurality of clients is a recording medium readable by an electronic device, which is recorded as a program to execute a method for designing a frequency selective surface filter, which is a computer program executed in a computing device including a CPU and memory. Claim 12 A recording medium readable by an electronic device and programmed to perform a method for designing a frequency-selective surface filter, wherein the first local optimization step and the second local optimization step each include the step of each of the plurality of clients modifying the sequence of the attempt solution, the step of calculating the objective function value of the attempt solution having the modified sequence, and the step of setting the attempt solution with the modified sequence as the local optimal solution if the frequency response of the attempt solution with the modified sequence corresponds more to the frequency response provided by the user than to the frequency response of the attempt solution from the objective function value. Claim 13 In claim 12, the step of modifying the sequence of the above-mentioned attempts is a design method for a frequency-selective surface filter, which is programmed and readable on an electronic device to be performed using any one of the Nelder-Mead algorithm, the annealing simulation algorithm, the conformational space annealing algorithm, the conjugated direction method, and the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm. Claim 14 A recording medium readable by an electronic device and programmed to perform a method for designing a frequency-selective surface filter, wherein the genetic change step is performed as either a mutation that changes one or more digits of any one of the local optimal solutions or a crossover that replaces the sequence of two or more of the local optimal solutions. Claim 15 A recording medium readable by an electronic device and recorded as a program to perform a method for designing a frequency-selective surface filter according to claim 10, wherein the method for generating the frequency-selective filter further comprises the step of the manager setting a cutoff distance from the candidate solution generated in the first local optimization step, wherein the step of setting the cutoff distance comprises: the step of calculating Hamming distances between the candidate solutions, the step of calculating the average of the Hamming distances, and the step of setting the average value of the Hamming distances as the cutoff distance. Claim 16 In claim 15, the step of the manager determining the discarding and replacement of the second local optimal solution comprises: determining the nearest candidate solution, which is the candidate solution closest to the second local optimal solution, from the Hamming distance calculation result of the candidate solutions; and if the distance between the second local optimal solution and the nearest candidate solution is smaller than the current cutoff distance value, comparing the objective function value of the second local optimal solution with the objective function value among the nearest candidate solutions, discarding the second local optimal solution when the frequency response characteristics of the nearest candidate solution are more similar to the target frequency response characteristics than the frequency response characteristics of the second local optimal solution, discarding the nearest candidate solution when the frequency response of the second local optimal solution is more similar to the target frequency response than the frequency response of the nearest candidate solution, and the second local optimal solution is recorded as a program and is a reading medium in an electronic device to perform a design method of a frequency-selective surface filter that replaces the nearest candidate solution. Claim 17 A recording medium readable by an electronic device that is programmed to perform a design method for a frequency-selective surface filter, wherein, in paragraph 15, the step of the manager determining to discard and replace the second local optimal solution comprises: selecting a candidate solution having frequency response characteristics that are least similar to the target frequency response characteristics from the objective function, comparing the objective function of the second local optimal solution, and if the frequency response characteristics of the second local optimal solution from the objective function value of the second local optimal solution are more similar to the target frequency response characteristics than the frequency response characteristics of the candidate solution, the second local optimal solution replaces the candidate solution, and if the frequency response characteristics of the candidate solution from the objective function value of the candidate solution are more similar to the target frequency response characteristics than the second local optimal solution, the second local optimal solution is discarded. Claim 18 A recording medium readable by an electronic device and programmed to perform a design method of a frequency-selective surface filter according to claim 15, wherein the manager further includes the step of reducing the cutoff distance after the step of the manager determining to replace or discard the second local optimal solution.

Images