JP2007074604A - Manufacturing method of antenna - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve an optimal efficiency by making a crossing-over which is an operator specific to genetic algorithm act effectively. <P>SOLUTION: A method comprises a simulation process for producing CAD data of an antenna having two-dimensional shape responding to a two-dimensional array, for performing electromagnetic simulation based on this CAD data, and for acquiring the fitness of the antenna from characteristic data after acquiring the characteristic data of the antenna, a determining process for determining whether or not a focusing qualification is to be satisfied based on the fitness, and an operation process for newly producing the CAD data by performing the crossing-over for exchanging a part of the CAD data with a part of the other CAD data corresponding to a part of this in a two-dimensional array state or by performing a mutation for converting a part of the CAD data in the two dimensional array state based on the fitness as an operation of the genetic algorithm when the focusing qualification is not satisfied. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method for manufacturing a planar antenna.

  By using a genetic algorithm to design a two-dimensional antenna, it is possible to design an antenna having a structure that humans cannot imagine. As a method using a genetic algorithm for such a two-dimensional antenna, a method shown in Non-Patent Document 1 is known. In this Non-Patent Document 1, an operation based on a genetic algorithm is performed by dividing a patch into a plurality of sections, replacing air with “0” and conductor with “1”, and replacing this with a one-dimensional array. .

As a method using a genetic algorithm for a three-dimensional antenna, as shown in Non-Patent Document 2, a method of applying a genetic algorithm to a two-dimensional antenna on the side of a prism is proposed. Has been. In addition, no optimum design method for an arbitrary three-dimensional antenna using a genetic algorithm has been reported so far.
“Electromagnetic Optimization by Genetic Algorithm, Yahya Rahmat-Samii and Eric Michielssen, pp249-278” "Automatic optimal design of multi-frequency small monopole antenna using GA chromosome construction method by applying maze generation algorithm, Tamami Maruyama et al., 2005 IEICE General Conference"

  “Crossover”, which is a representative operator of a genetic algorithm, includes the meaning of leaving excellent individual information (schema) of excellent individuals in many offspring.

  Here, in the antenna, since the current distribution on the conductor greatly affects the antenna characteristics, the two-dimensionally adjacent sections are strongly coupled and physically very meaningful. Is likely to be. However, as shown in FIG. 21, it is adjacent in the one-dimensional array (horizontal line section and dotted line section in (A)) but not adjacent when converted into a two-dimensional array (in (B)). The horizontal lines and the dotted lines are weakly connected and have no physical meaning, so the possibility of becoming a schema is low. Similarly, when expressed in a one-dimensional array, it is difficult to consider vertical coupling.

  For this reason, when a one-dimensional array is used, crossover, which is an operator specific to the genetic algorithm, does not work effectively, optimization efficiency is deteriorated, and optimization requires a large number of calculations. . In the optimal design of an antenna using simulation, generally, an electromagnetic field simulation must be performed every time in order to calculate the fitness. Therefore, as the number of calculations increases, the time required for optimization becomes enormous. It becomes impossible to perform the optimum design in a realistic calculation time.

  Therefore, an object of the present invention is to effectively perform crossover, which is an operator specific to a genetic algorithm, by using the antenna manufacturing method of the present invention, thereby improving the optimization efficiency.

  In order to achieve this object, the antenna manufacturing method of the present invention creates CAD data of a two-dimensional antenna corresponding to a two-dimensional array, performs electromagnetic field simulation based on the CAD data, and performs antenna characteristics. A simulation process for obtaining data and determining the fitness of the antenna from the characteristic data, a determination process for determining whether the convergence condition is satisfied based on the fitness after the simulation process, and after the determination process If the convergence condition is not satisfied, crossover is performed in which a part of CAD data is exchanged with a part of other CAD data corresponding to this part in a two-dimensional array state based on fitness as an operation of the genetic algorithm. Alternatively, a new CAD data is generated by performing a mutation that converts a part of the CAD data in a two-dimensional array state. And a step. Then, the simulation process is repeated for the CAD data newly generated in this operation process.

  According to the above antenna manufacturing method, crossover or mutation can be performed in consideration of the combination of sections in the vertical and horizontal directions, so that optimization efficiency is improved and calculation time required for optimization is greatly reduced. Is possible.

  Moreover, the method of inverting the bit of one section selected at random is used for the mutation method in the above-mentioned operation process. As a result, local search capability can be ensured, so that optimization efficiency can be improved and calculation time required for optimization can be greatly shortened.

  Further, in the first generation, a two-dimensional array other than the two-dimensional array converted by crossover or mutation is generated by random numbers. In addition to “random one-segment mutation” corresponding to local search and “crossover to exchange one random row or column” corresponding to mid-range search, “the remaining individuals are randomly selected” By including `` random search to generate '', it has a balanced solution search capability, can perform stable and efficient optimization, and can greatly reduce the calculation time required for optimization Become.

  Furthermore, the number of individuals to be mutated in one generation is dynamically changed according to the total number of calculations or the update status of the individual with the highest fitness. This reduces the number of individuals to be mutated at the beginning of a search that requires wide-area search capability, and can increase the wide-area search capability by increasing the number of individuals generated by random numbers instead. In the final stage of the search, the local search capability can be improved by reducing the number of individuals generated by random numbers and increasing the number of individuals to be mutated instead. If the individual with the highest fitness has not been updated for a long time, it is possible to make it easier to escape from the local solution by reducing the number of individuals to be mutated and increasing the number of individuals generated by random numbers instead. Thus, stable and efficient optimization can be performed, and the calculation time required for the optimization can be greatly shortened.

(Embodiment 1)
Hereinafter, Embodiment 1 of the present invention will be described with reference to FIG.

  With reference to FIG. 1, the overall steps of the optimal method for manufacturing a planar antenna are outlined below. The flowchart shown in FIG. 1 includes a first algorithm and a second algorithm.

  The first algorithm includes a first step (S1) for creating CAD data including an antenna having a two-dimensional shape to be optimized, and a second step for designating an object and the number of divisions to which the genetic algorithm is applied in the CAD data. (S2), a third step (S3) for dividing an object of CAD data based on the number of divisions, a fourth step (S4) for creating a two-dimensional array corresponding to the number of divisions, and a two-dimensional array A fifth step (S5) for associating the objects of the divided CAD data, a sixth step (S6) for generating a random two-dimensional array composed of binary variables, and the two-dimensional generated in the sixth step CAD data is created based on the arrangement, electromagnetic field simulation is performed based on this structure, and the seventh step (S7), which is a simulation step for determining the fitness, and the seventh step (S7). An eighth step (S8), which is a determination step for determining whether the convergence condition is satisfied based on the degree of response, and a step (S8) that ends when the convergence condition is satisfied (S8). F1).

  The second algorithm is an eighth step (S8) for determining whether the convergence condition is satisfied based on the fitness obtained in the seventh step (S7). This convergence condition is not satisfied. In some cases, taking into account the fitness as an operation of the genetic algorithm, a crossover that replaces a part of the two-dimensional array with a part of another two-dimensional array corresponding to this part, or a part of the two-dimensional array 9th step (S9), which is an operation step for performing a mutation to convert, and a predetermined operation in this ninth step (S9) is carried out, and whether the number of generations set in advance is exceeded The tenth step (S10) for determining whether or not and if this determination condition is satisfied, the seventh step (S7) to the tenth step (S10) are repeated, and the convergence shown in the seventh step (S7) A step of determining whether the condition is satisfied and terminating if the condition is satisfied If (F1) is included and the convergence condition is not satisfied, the ninth step (S9) is further performed. If the number of generations exceeds the ninth step (S9), the calculation is terminated. Process (F2).

  Next, each step will be described in detail below. Here, a case where the shape of a planar antenna attached on a bar-type substrate is optimized will be described as an example.

  First, in the first step (S1), as shown in FIG. 2, CAD data including an antenna to be optimized is created. Here, the bar type substrate 1, the dielectric 2, the antenna element 3, and the feed line 4 are modeled.

  Next, in the second step (S2), as shown in FIG. 3, a genetic algorithm is applied to select an object whose shape is to be optimized. Further, as shown in FIG. 4, the number of divisions of the selected object in the X direction and the Y direction is designated. Here, it is assumed that the X direction and the Y direction are divided into five parts and divided into a total of 25 sections.

  Next, in the third step (S3), the CAD data object is divided based on the number of divisions specified in the second step (S2), and the contents of the CAD data are updated. That is, as shown in FIG. 5, Object1 is divided according to the number of divisions and replaced with Object1-1, Object1-2,.

  In parallel with the third step (S3), in the fourth step (S4), a two-dimensional array is created based on the number of divisions specified in the second step (S2). Here, since the X direction and the Y direction are divided into five, a 5 × 5 array is created as shown in FIG.

  Next, in the fifth step (S5), the object divided in the third step (S3) is associated with the two-dimensional array created in the fourth step (S4) according to the arrangement. That is, as shown in FIG. 7, object1-1 is an array (1, 1), object1-2 is an array (1,2) ..., object1-25 is an array (5, 5), etc. Correlate as follows.

  Next, in a sixth step (S6), an initial generation individual in the genetic algorithm is generated. Specifically, as shown in FIG. 8, a random two-dimensional array composed of binary variables equal to the number of individuals per generation set in advance is generated.

  Next, in the seventh step (S7), CAD data corresponding to a two-dimensional antenna is created based on the two-dimensional array generated in the sixth step (S6). Here, when the array value set in the sixth step (S6) is “0”, it means air, and when the array value is “1”, it means metal. As shown in FIG. The material of the corresponding object is changed according to the array value. In addition, this time, we use binary variables as array values and optimize when there are only air and metal materials. However, if multiple values are used, multiple materials such as dielectrics are included. It is also possible to perform optimization. Further, based on the created CAD data, an electromagnetic field simulation is performed to calculate the antenna fitness. Here, the fitness is calculated from the target S-parameter / radiation characteristics and antenna characteristic data such as S-parameter / radiation characteristics obtained by electromagnetic field simulation. The smaller the difference from the target characteristics, the higher the fitness. , Will be recognized as an excellent individual.

Next, in the eighth step (S8), the calculation is terminated if the fitness satisfies the convergence condition set in advance, and if not, the process proceeds to the ninth step (S9). Here, when the calculation is terminated by this conditional branch (F1 in FIG. 1), it means that the optimum solution has been obtained. The evaluation criteria for the end of processing are as follows: • Maximum fitness in individual population> threshold value • Average fitness of individual population> threshold value. Here, the larger the threshold value, the higher the probability of obtaining an optimal solution, but the calculation time increases.

  Next, in the ninth step (S9), as an operation of the genetic algorithm, crossover is performed by exchanging any one row or one column between individuals selected and mutated based on fitness. Usually, in a genetic algorithm, a two-dimensional array is converted into a one-dimensional array and an operation is performed using the one-dimensional array. Here, the genetic algorithm is operated with the two-dimensional array as it is.

  Specifically, as shown in FIG. 10, two individuals are selected at random, and crossover is performed in a randomly selected vertical or horizontal row. As a result, the search can be performed while leaving the schema existing in the vertical direction, so that an efficient search can be performed. Here, as a crossover method other than the above, as shown in FIG. 11, a method of determining a crossover position, a crossover range, and the number of crossovers in one individual by a random number is conceivable, but the role in the crossover search is not determined by the random number. As a result, the optimization efficiency becomes very poor. Here, when the above-described method is used, the search can be performed within a certain range while leaving the schema of the crossover method, so that a stable mid-range search can be performed.

  The mutation method is shown below. As a mutation method, as shown in FIG. 12, a method is used in which one section is selected at random based on an excellent (high fitness) individual of the previous generation, and the bit of the selected section is inverted.

  Usually, in genetic algorithms, mutation is aimed at “avoiding limiting the range in the search space” and “escape from local solution”, but as in this time, the structure is directly used as bit information. In such a case, the mutation has no meaning as described above. When the structure like this time is converted into bit information as it is, the mutation changes a part of the current structure, and is considered to have the meaning of local search. Here, as a method of mutation, in addition to the method of “reversing the bit of only one section” as described above, the method of “fixing the number of sections to be reversed in one individual to a plurality” is conceivable. However, by increasing the number of sections to be reversed, the meaning of local search is reduced because the distance from the shape of the original individual is reduced, the local search capability is deteriorated, and the optimization efficiency is degraded. . In addition, a method such as “determining the number of sections to be inverted in one individual at random” is conceivable, but in this case, the meaning of the mutation search changes randomly, so that the optimization stability deteriorates. . For this reason as well, it can be said that the optimization efficiency is best when the above-described “method of randomly selecting one section and inverting the bits of the selected section” is used.

  When the number of individuals in one generation is larger than the sum of individuals caused by mutation and individuals caused by crossover, the number is generated by random numbers as in the case of individuals in the initial generation. As a result, “random one-segment mutation” corresponding to local search, “crossover that exchanges one random row or column” corresponding to mid-range search, and “remaining individuals randomly generated corresponding to wide-area search” By including all “random search to be performed”, it has a balanced solution search capability, can perform stable and efficient optimization, and can greatly reduce the calculation time required for optimization. Become.

  Next, in the tenth step (S10), if the number of generations set in advance is exceeded, the calculation is terminated. If not, the process returns to the seventh step (S7) again, and the same operation is repeated. Here, when the calculation is terminated by this conditional branch (F2 in FIG. 1), there is a possibility that the optimal solution may not have been obtained. Therefore, in order to obtain the optimal solution, the number of generations to be set is increased and the calculation is performed again. It is necessary to start over.

  According to the above manufacturing method, crossover or mutation can be performed in consideration of the combination of sections in the vertical and horizontal directions, so that optimization efficiency can be improved and calculation time required for optimization can be greatly shortened. It becomes.

(Embodiment 2)
Hereinafter, a method for manufacturing a three-dimensional antenna according to Embodiment 2 will be described with reference to the drawings.

  Unless otherwise specified, it is the same as in the first embodiment.

  The flowchart shown in FIG. 13 is a diagram illustrating the method for manufacturing the antenna according to the second embodiment. The main difference between FIG. 13 and FIG. 1 is that, in the seventh step, CAD data is created based on a three-dimensional array that is a set of two-dimensional arrays, an electromagnetic field simulation is performed based on this structure, and adaptation is performed. It is a point to find the degree.

  Next, each step will be described in detail below. Here, a case where an arbitrary three-dimensional antenna attached on a bar-type substrate is optimized will be described as an example.

  First, in the first step (S1), as shown in FIG. 14, CAD data including an antenna to be optimized is created. Here, the bar type substrate 7, the antenna element 8, and the feeder 9 are modeled.

  Next, in the second step (S2), as shown in FIG. 15, a genetic algorithm is applied to select an object whose shape is to be optimized. Further, as shown in FIG. 16, the number of divisions of the selected object in the X direction, the Y direction, and the Z direction is designated. Here, the X direction and the Y direction are divided into five, and the Z direction is divided into three, and divided into a total of 75 sections.

  Next, in the third step (S3), the CAD data object is divided based on the number of divisions specified in the second step (S2), and the contents of the CAD data are updated. That is, as shown in FIG. 17, Object1 is divided according to the number of divisions and replaced with Object1-1, Object1-2,.

  In parallel with the third step (S3), in the fourth step (S4), a three-dimensional array is created based on the number of divisions specified in the second step (S2). Specifically, a two-dimensional array composed of the number of divisions in the X direction and the number of divisions in the Y direction is prepared for the number of divisions in the Z direction. Therefore, since the X direction and the Y direction are divided into five and the Z direction is divided into three here, three 5 × 5 arrays are created as shown in FIG. Here, three arrays A, B, and C are prepared.

  Next, in the fifth step (S5), the object divided in the third step (S3) is associated with the two-dimensional array created in the fourth step (S4) according to the arrangement. That is, as shown in FIG. 19, object1-1 is array A (1,1), object1-2 is array A (1,2),..., Object1-25 is array A (5,5), Object1-26 is array B (1,1), ..., object1-50 is array B (5,5), object1-51 is array C (1,1), ..., object1-75 Corresponding to array C (5, 5).

  Next, in a sixth step (S6), an initial generation individual in the genetic algorithm is generated. Specifically, as in FIG. 8, a set of three two-dimensional arrays of random A, B, and C consisting of binary variables equal to the number of individuals per generation set in advance is generated.

  Next, in the seventh step (S7), CAD data corresponding to a three-dimensional antenna based on the set of two-dimensional arrays (array A, array B, array C) generated in the sixth step (S6). Create Further, based on the created CAD data, an electromagnetic field simulation is performed using a finite element method or a finite difference method to calculate the fitness of the antenna. Next, in the eighth step (S8), the calculation is terminated (F1) if the fitness satisfies the preset convergence condition, and if not, the process proceeds to the ninth step (S9). .

  Next, in the ninth step (S9), as an operation of the genetic algorithm, mutation and selection based on fitness, any one row in the same two-dimensional array of selected individuals or any one column or all in the same two-dimensional array Crossover is performed to exchange the same position in the two-dimensional array.

  Specifically, as shown in FIG. 20, two individuals are selected at random, and crossover is performed at the same position in a two-dimensional array of randomly selected vertical or horizontal rows of the same array or all arrays. As a result, the search can be performed while leaving the schemas existing in the vertical direction, the horizontal direction, and the height direction, so that an efficient search can be performed. When this method is used, the search can be performed within a certain range while leaving the schema, so that a stable mid-range search can be performed.

  According to the above manufacturing method, crossover or mutation can be performed in consideration of the combination of sections in the vertical, horizontal, and height directions, so that optimization efficiency is improved and calculation time required for optimization is greatly reduced. It becomes possible.

(Embodiment 3)
Hereinafter, a method for manufacturing the antenna according to the third embodiment will be described with reference to the drawings. Unless otherwise specified, this is the same as in the first or second embodiment.

  In the antenna manufacturing method according to the third embodiment, the number of individuals to be mutated in one generation in the ninth step (S9) shown in FIG. 1 or FIG. It is changed dynamically according to.

  When the number of individuals in the first generation is 30, the number of individuals to be mutated in the first generation is 10, the number of individuals generated by crossover in the first generation is 10, and the number of individuals generated by random numbers in the first generation is 10. Will be described. In order to avoid falling into a local solution at the beginning of the search, a wide-area search capability is required, but at the end of the search, a local search capability is required. Therefore, every 10 generations, the number of mutant individuals is increased by 1, and instead the number of individuals generated by random numbers is decreased by 1. In this method, mutation has local search capability, and generation of individuals by random numbers has wide search capability. Therefore, as the number of generations increases by this operation, the wide search capability is weakened and local search is performed. It becomes possible to strengthen ability.

  Also, the number of mutations may be changed even when the individual with the highest fitness has not been updated for a predetermined number of generations. For example, if an individual with the highest fitness level is not updated again for five generations after the individual with the highest fitness level has been updated, the number of mutant individuals is increased by 1, and instead the number of individuals generated by random numbers is increased to 1. Decrease the number. In this way, even if a local solution falls into the end of the search, it is possible to escape from the local solution by strengthening the global search capability, enabling efficient optimization and significantly reducing the number of calculations. It becomes possible.

  According to the antenna manufacturing method of the present invention, the time required for optimal design of a two-dimensional antenna is remarkably shortened, and the optimum design of an arbitrary three-dimensional antenna is enabled and the time required for optimal design is remarkably shortened. For example, it is useful as a method for manufacturing an antenna mounted on a mobile phone.

Process drawing explaining the optimal manufacturing method of the planar antenna in Embodiment 1 of this invention Conceptual diagram of a CAD diagram including an antenna element to be optimized in the first embodiment of the present invention FIG. 2 is a conceptual diagram for selecting an object to be optimized by applying the genetic algorithm according to the first embodiment of the present invention. The conceptual diagram of the screen which inputs the division | segmentation number of the selected object in Embodiment 1 of this invention (A) and (B) are conceptual diagrams of CAD data after object division in Embodiment 1 of the present invention. Conceptual diagram of creating a two-dimensional array according to the number of divisions in Embodiment 1 of the present invention Conceptual diagram of association between two-dimensional array and divided CAD data in Embodiment 1 of the present invention Conceptual diagram of generation of individuals of the initial generation in the genetic algorithm according to Embodiment 1 of the present invention Conceptual diagram of association between array values and corresponding object materials in Embodiment 1 of the present invention Explanatory drawing of the crossover method in Embodiment 1 of this invention Explanatory diagram when determining the crossover position and crossover range with random numbers (A) and (B) are explanatory diagrams of the mutation method used in the present invention. Process drawing explaining the optimal manufacturing method of the antenna of arbitrary three-dimensional shapes in Embodiment 2 of this invention Conceptual diagram of a CAD diagram including an antenna element to be optimized in the second embodiment of the present invention Conceptual diagram for selecting an object to be subjected to shape optimization by applying the genetic algorithm according to the second embodiment of the present invention The conceptual diagram of the screen which inputs the division | segmentation number of the selected object in Embodiment 2 of this invention (A) and (B) are conceptual diagrams of CAD data after object division according to Embodiment 2 of the present invention. (A)-(C) are the conceptual diagrams of the correlation of the CAD data divided | segmented and the set of the two-dimensional array in Embodiment 2 of this invention Conceptual diagram of association between array values and corresponding object material in Embodiment 2 of the present invention Explanatory drawing of the crossover method in Embodiment 2 of this invention Comparison diagram of one-dimensional array and two-dimensional array in genetic algorithm

Explanation of symbols

1 Bar type substrate 2 Dielectric 3 Antenna element (flat plate)
4 Feed line 5 Pointer 6 Window

Claims (9)

Simulation for creating CAD data of a two-dimensional antenna corresponding to a two-dimensional array, performing electromagnetic field simulation based on the CAD data, obtaining characteristic data of the antenna, and obtaining fitness of the antenna from the characteristic data Process,
A determination step of determining whether a convergence condition is satisfied based on the fitness after the simulation step;
If the convergence condition is not satisfied after this determination step, based on the fitness as an operation of a genetic algorithm, a part of the CAD data is replaced with a part of other CAD data corresponding to the part. An operation step of newly generating CAD data by performing crossover to be exchanged in the two-dimensional array state, or performing a mutation to convert a part of the CAD data in the two-dimensional array state,
An antenna manufacturing method in which the simulation process is repeated for CAD data newly generated in this operation process. 2. The method of manufacturing an antenna according to claim 1, wherein in the operation step, the mutation method is a method of inverting a part of randomly selected bits in the two-dimensional array. The antenna manufacturing method according to claim 1, wherein a two-dimensional array other than the two-dimensional array converted by the crossover or the mutation in one generation is generated by random numbers. The antenna manufacturing method according to claim 1, wherein the crossover is performed by exchanging randomly selected one row or one column. Simulation that creates CAD data of a three-dimensional antenna corresponding to a three-dimensional array, performs electromagnetic field simulation based on the CAD data, obtains characteristic data of the antenna, and obtains fitness of the antenna from the characteristic data Process,
A determination step of determining whether a convergence condition is satisfied based on the fitness after the simulation step;
If the convergence condition is not satisfied after this determination step, based on the fitness as an operation of a genetic algorithm, a part of the CAD data is replaced with a part of other CAD data corresponding to the part. An operation step of newly generating CAD data by performing crossover to be exchanged in the three-dimensional array state or performing mutation to convert a part of the CAD data in the three-dimensional array state,
An antenna manufacturing method in which the simulation process is repeated for CAD data newly generated in this operation process. 6. The method of manufacturing an antenna according to claim 5, wherein in the operation step, the mutation method is a method of inverting a part of randomly selected bits in the three-dimensional array. 6. The method for manufacturing an antenna according to claim 5, wherein a three-dimensional array other than the three-dimensional array converted by the crossover or the mutation in one generation is generated by a random number. The method for manufacturing an antenna according to claim 1 or 5, wherein the number of individuals to be mutated in one generation is dynamically changed in accordance with the total number of calculations or the update status of an individual having the highest fitness. 6. The method for manufacturing an antenna according to claim 5, wherein the crossover is performed by exchanging bits at the same position in a two-dimensional array of the same array, which is selected at random or in the same vertical or horizontal row.

Images