JPH09312509A - Method for adjusting dielectric filter and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain high efficiency in work and to execute adjustment without depending on the knowhow of a worker by automatically analyzing and deciding the adjustment quantity of a dielectric filter. SOLUTION: The characteristic of an adjusting element in the dielectric filter 2 is varied from a set value within a fixed range so as to be assigned to a high frequency simulator 12, to be adopted as data of a frequency characteristic corresponding to the deviation of an adjusting element value and a neural network 14 is permitted to learn it. At the time of adjustment, the frequency characteristic measured from an actual object is inputted to the neural network 14 and the deviation quantity of the adjusting element is obtained. The difference of the frequency characteristic at the time of design from that as against the deviation quantity is the improvement expecting value of the caracteristic at the time of correcting deviation quantity. The difference is added to the measured characteristic, the adjusting element value is changed little by little unless it satisfy standard and the characteristic is inspected. After that, difference from the characteristic at the time of design is, moreover, added. When standard is finally satisfied, the total sum of deviation quantity till this time is adjusted as the adjustment quantity of the adjusting element.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for adjusting a dielectric filter, and more particularly to a method and an apparatus for adjusting a dielectric filter for automatically analyzing and determining an adjustment amount without depending on know-how of a skilled person.

[0002]

2. Description of the Related Art Hitherto, in order to obtain desired frequency characteristics of a dielectric filter, it has been necessary to adjust a part of components constituting the dielectric filter after assembling the components. For example,
There are various types of dielectric filters used in mobile phones, etc. that combine dielectric resonators, coils, and capacitors. In the process of manufacturing this type, the effects of variations in the electrical characteristics of the components used Thus, the desired frequency characteristics cannot be obtained immediately after assembling, and the entire frequency characteristics are adjusted by processing the dielectric resonator and changing the resonance frequency of the dielectric resonator itself.

[0003] Such adjustment work has conventionally been performed manually while looking at the display of the measuring instrument.

As a conventional example in which manual adjustment work is automated, a tuner automatic adjustment apparatus disclosed in Japanese Patent Application Laid-Open No. 4-28150 is a fuzzy rule based on the know-how of a skilled person as a knowledge base. There is. FIG. 7 is a block diagram showing this conventional example, in which a fuzzy inference unit 40, an oscilloscope 41, a feature recognition unit 42, a coil 43, and a tuner 4 are shown.
4, a hand 45, 46 and a high-frequency signal generator 47. First, the signal of the high frequency signal generator 47 is applied to the tuner 44, and the output spectrum thereof is displayed on the oscilloscope 41. At the same time, feature values such as a tuning peak value, a tuning characteristic gradient, and an area are extracted from the output spectrum by the feature recognition unit 42.

Next, a fuzzy inference unit 40 applies a fuzzy rule to the feature amount and calculates an adjustment amount for changing the coil interval. Based on the adjustment amount, the fuzzy inference unit 40 operates the hands 45 and 46 to change the interval between the coils 43 to perform the adjustment. In applying the fuzzy rule, a membership function determined based on the results of adjustment performed by a skilled person is used.

[0006]

In the above-described adjustment of the dielectric filter, since the adjustment is performed manually, there is a disadvantage that the adjustment takes a long time and is affected by the know-how of the operator. In addition, since the tuner automatic adjustment device uses the adjustment record of an expert to determine the membership function of the fuzzy rule used for the adjustment, there is a drawback that it takes a lot of time and effort to record the record, and it is also a characteristic Has a drawback that it is difficult to derive a unified fuzzy rule due to the strong. Further, there is a problem that fuzzy rules must be created by trial and error in the case of adjusting a new product for which there is no expert in adjustment.

The method and apparatus for adjusting a dielectric filter according to the present invention automatically analyze the adjustment amount regardless of the results of a skilled person.
The purpose of the determination is to make the work more efficient and to make adjustments that do not depend on the know-how of the worker.

[0008]

SUMMARY OF THE INVENTION A method of adjusting a dielectric filter according to the present invention is a method of adjusting a frequency characteristic of a dielectric filter. A first step of setting an element value other than the adjustment element to a design value, and the simulator calculates frequency characteristics of the dielectric filter by varying n adjustment element values r1 to rn;
A second step of storing data obtained by combining these adjustment element values r1 to rn and frequency characteristics as teacher data, and setting the adjustment element values performed in the second step to values different from the previous values. A third step to repeat until a predetermined combination is completed, a fourth step to measure the frequency characteristic of the dielectric filter to be adjusted, and the second step
Inputting the teacher data of the combination stored in the step and the third step and the frequency characteristics of the dielectric filter to be adjusted measured in the fourth step to a neural network, A fifth step of calculating a deviation of the adjustment element value with respect to the
A sixth step of calculating, with the simulator, the frequency characteristic with respect to the deviation of the adjustment element value calculated in the step, and a seventh step of setting the frequency of all the elements to the design values and calculating the frequency characteristic with the simulator. An eighth step of calculating a difference between the frequency characteristic calculated in the seventh step and the frequency characteristic calculated in the sixth step;
One of the adjustment elements is slightly changed from the design value, the other adjustment element is set to the design value, the frequency characteristics are calculated by the simulator, and then the frequency characteristics calculated in the seventh step are compared with the frequency characteristics calculated in the seventh step. The ninth step of calculating the difference and the frequency characteristics of the dielectric filter to be adjusted measured in the fourth step are described in the eighth step and the ninth step.
Adding a difference between the frequency characteristics calculated in the step (a), and calculating a margin, which is a difference between a standard value and a result of the addition processing, at a specific standard point of the frequency characteristics, and An eleventh step in which the ninth step and the tenth step are repeated until the margin in the step (c) exceeds a predetermined threshold, and the margin is predetermined in the eleventh step. For adjusting the dielectric filter to be adjusted by the sum of the shift amount of the adjustment element value in the ninth step when the threshold value is exceeded and the deviation of the adjustment element value calculated in the fifth step. And a twelfth step of adjusting the element value.

The dielectric filter adjusting method of the present invention is characterized in that the frequency characteristic of the second step is a difference from the frequency characteristic when the value of the adjusting element is set to a design value.

An apparatus for adjusting a dielectric filter according to the present invention comprises: an element value generating means for varying a plurality of adjustment element values of a dielectric filter in a plurality of stages, and setting element values other than the adjustment elements to design values; Teacher data generating means comprising simulation means for calculating the frequency characteristic A at the element value, and teacher data storing means for storing teacher data obtained by combining the element value and the frequency characteristic A. And waveform storage means for storing the frequency characteristic B of the dielectric filter to be adjusted; learning the combination of the teacher data and the frequency characteristic B stored in the teacher data storage means; −
A neural network learning means for calculating a deviation of the adjustment element value with respect to the data, a search element value generation means for setting all the elements to the design value or slightly changing one of the adjustment elements from the design value, The simulation for calculating the frequency characteristic C with respect to the deviation of the adjustment element value, the frequency characteristic D when all the elements are set to the design value, and the frequency characteristic E when one of the adjustment elements is slightly changed from the design value. Means for adding the frequency characteristic B, the difference between the frequency characteristic D and the frequency characteristic C, and the difference between the frequency characteristic E and the frequency characteristic D; Calculate the margin, which is the difference between the value and the addition result. If the margin exceeds a predetermined threshold, an instruction for adjustment is made.If the margin does not exceed the threshold, the adjustment is performed until the margin is exceeded. An adjustment control unit including a standard determination unit that instructs to reset the element value and recalculate the frequency characteristic E, and an adjustment instruction unit that performs adjustment based on an instruction from the standard determination unit and the search element value generation unit. , Is characterized by having.

[0011]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention, which comprises a dielectric filter adjusting device 1, a dielectric filter 2, a network analyzer 3, and an adjusting means 4. Further, the dielectric filter adjustment device 1 includes a teacher data generation unit 1a and an adjustment control unit 1b. In addition, the teacher data generation unit 1a includes the element value generation unit 1a.
0, the teacher data storage unit 11, the high-frequency simulator 12, the adjustment control unit 1b, the high-frequency simulator 12, the search element value generation unit 13, the neural network 14, the waveform processing unit 15, It comprises a storage unit 16, an adjustment amount instruction unit 17, and a standard determination unit 18.

FIG. 2 is a mounting diagram showing an example of the dielectric filter 2, in which a plurality of dielectric resonators 22, coils 23 and capacitors 24 are mounted on a mounting substrate 21.
On top of that, the case 20 is covered. Dielectric resonator 2
The electrode 2 is formed, for example, by baking electrodes on the outer periphery of a prismatic ceramic as shown in FIG. When this dielectric filter 2 is represented by an electric equivalent circuit, each dielectric resonator 2
2 can be represented as an element characterized by a resonance frequency. If the frequency characteristic of the dielectric filter 2 is out of the standard, the resonance frequency of the corresponding dielectric resonator 22 is changed by slightly shaving the electrode at an exposed portion of the electrode such as part A, for example. Satisfies the entire frequency characteristics standard. That is, the adjustment of the frequency characteristics of the dielectric filter 2 is performed by selecting one of the plurality of dielectric resonators 22 that can change the frequency characteristics in a good direction and shaving off an appropriate amount of electrodes. The frequency characteristic described here is a transfer characteristic of the dielectric filter 2 in a preset frequency region, and corresponds to an S21 element in the S parameter.

Next, an outline of the operation of the present invention will be described with reference to FIG.
And FIG.

First, a certain range of deviation from the design value is set for the resonance frequency of the dielectric resonator 22 shown in FIG. 2 as an adjusting element, and then the range is varied at equal intervals.
For example, when the range of deviation is ± 10% and the interval is equally divided, the values are set to -10%, -5%, 0%, 5%, and 10%. The above setting is performed for all the dielectric resonators 22, and then, any one value is assigned to each of the dielectric resonators 22 to make various combinations. For example, when there are three dielectric resonators 22, there are 125 combinations of 5 to the third power.

For all these combinations, the high frequency simulator 12 calculates the frequency characteristics of the dielectric filter 2. The amount of displacement of each of the dielectric resonators 22 and the frequency characteristic are used as a pair as teacher data. When data on all combinations are collected, the neural network is used.
To learn. This time, the frequency characteristic of the dielectric filter 2 to be adjusted is input to the neural network 14 to output the deviation amount of the dielectric resonator 22, and the dielectric resonator 22 is adjusted based on this value.

Next, let us turn to the detailed description of the present invention.

First, referring to FIG. 3 showing a flowchart for generating teacher data, "from generation of teacher data to learning by a neural network" in the present invention.
Will be described in detail.

First, in the element value generating section 10, as described above, a deviation of the resonance frequency of the dielectric resonator 22 from a design value within a certain range is set, and the range is dispersed at equal intervals. Then, any one of the values is assigned to each of the dielectric resonators 22 to make various combinations. Select one of all generated combinations. Coil 23 other than dielectric resonator 22, capacitor 2
All elements other than those set here, such as 4, are set to ideal values (design values), but their electrical characteristics are remarkably varied, and their influence on the frequency characteristics is affected by the dielectric resonance as an adjustment element. If the effect is equal to or greater than the influence of the element 22, the element is also set to a certain range of deviation from the design value, and is treated in the same manner as the dielectric resonator 22 described above (step 101).

A simulation is performed by the high-frequency simulator 12 based on each element value set in step 101, and a frequency characteristic is calculated. The point to be calculated is a transfer characteristic at a preselected frequency point within a specific frequency range, as shown in FIG. 6, for example. If it is determined that the transfer characteristic (S21) alone is not sufficient for the calculation of the element value deviation performed later, other parameters such as S11 and S22 may be included. Here, when S11 and S22 are included, all frequency characteristics include them in the following description (step 102).

The combination of the deviation amount of each of the dielectric resonators 22 set by the element value generation unit 10 in step 101 and the corresponding frequency characteristic calculated by the high frequency simulator 12 in step 102 is stored in the teacher data storage unit 11. To memorize. Although not particularly shown in the figure,
Instead of storing the frequency characteristics themselves, the teacher data storage unit 11 can also take the difference from the frequency characteristics when all the element values are set as design values and store them. Referring to FIG. 6 as an example, the difference referred to here is the difference between the design value and the above-calculated frequency characteristic at each of the measurement points O1 to On at the corresponding same frequency point. From O1 'to On', and all the differences represent the same processing.
Also, O1 'to On' obtained by the difference processing are referred to as a difference waveform. The reason why the difference processing is performed is to detect the difference in the input data with respect to the displacement of the dielectric resonator 22 so that the learning of the neural network 14 described later proceeds quickly (step 10).
3).

Next, the teacher data storage unit 11 confirms whether the processing of steps 101 to 103 has been executed for all combinations of deviations of the dielectric resonators 22. If all combinations have not been completed, the teacher data storage unit 1
1 instructs the element value generation unit 10 and repeats from the processing of step 101. If all combinations are completed, the teacher data storage unit 11 outputs teacher data of all combinations to the neural network 14. Advances to step 105 (step 104).

Upon receiving the teacher data of all the combinations stored in the teacher data storage unit 11, the neural network 14 performs the learning described below. The neural network 14 is configured, for example, as shown in FIG. In FIG. 5, the neural network 14 includes, from the teacher data described above, the measured values of the dielectric filter 2 from O1 to On.
Input layer 3 having the number of input elements corresponding to the number of data
0, an output of each of these input layers, an intermediate layer 31 having a predetermined number of intermediate elements, and an output of the intermediate layer 31 as an input, and a shift amount of the dielectric resonator 22 being output. And the same number of output elements 32 as the dielectric resonator 22. As the values of the input elements, O1 to On in FIG. 6 are sequentially assigned. The intermediate element calculates the sum of the outputs of the input elements after weighting the outputs of the input elements in advance, and outputs a value corresponding to the sum using a sigmoid function. As an example of the sigmoid function, a known Y = 1 / {1 + exp (-λX)} or the like is used.

Similarly, the output element 32 also applies a predetermined weight to the outputs of all the intermediate elements.
The sum is obtained, and a value corresponding to the sum is converted by a sigmoid function and output. The value output from the output element 32 has a real value between 0 and 1. This value corresponds to the amount of displacement of the dielectric resonator 22. Since the actual displacement of the dielectric resonator 22 is not always a number between 0 and 1, the maximum displacement of the dielectric resonator 22 in the teacher data is 1.0, The value is assigned to 0, and the intermediate value is scaled to a value between 0 and 1.0 by proportional calculation, and then used for learning the neural network 14.

When the measured value of the dielectric filter 2 to be adjusted is input to the neural network 14 and the amount of displacement of the dielectric resonator 22 is calculated, the above scaling is inversely converted to obtain the actual value. To obtain the adjustment amount.

The operation of causing the neural network 14 to learn the teacher data is, specifically, performed by setting the weights of the connections between the input layer 30 and the intermediate layer 31 and between the intermediate layer 31 and the output element 32 of the neural network 14 to correct values. The algorithm of the correction uses, for example, an error back propagation method. For the backpropagation method,
This is a known method and will not be described. When the output error of the output element becomes smaller than a predetermined threshold value for all the teacher data, the correction is terminated (step 105).

Next, with reference to FIG. 4 which is a flowchart for performing the adjustment, the operation of "from measurement to adjustment of the dielectric filter to be adjusted" in the present invention will be described in detail.

First, similarly to the step 102, the frequency characteristic at a predetermined frequency point is measured by the network analyzer 3 using the dielectric filter 2 to be adjusted. The frequency characteristics measured by the network analyzer 3 are stored in the waveform storage 16 (step 201).

The frequency characteristic measured in step 201 is
Input from the network analyzer 3 to the neural network 14,
The calculation described in step 105 is performed. The output of the neural network 14 is the displacement of the dielectric resonator 22. If the difference from the frequency characteristics at the design values of all the elements is stored instead of storing the frequency characteristics in step 103, the neural network is also stored in step 202.
The same process is performed on the frequency characteristics input to the loop 14, and the result is stored in the waveform storage unit 16 (step 202).

The neural network 14 outputs the displacement of the dielectric resonator 22 to the high frequency simulator 12, and the high frequency simulator 12 simulates frequency characteristics. The calculation result by the high frequency simulator 12 is output to the waveform processing unit 15 (step 203).

The search element value generator 13 inputs design values to all elements constituting the dielectric filter 2 according to an instruction from the high frequency simulator 12, and the high frequency simulator 12 calculates frequency characteristics. Instead of performing this process frequently, the process may be performed only once, stored in an external storage device or the like as design frequency characteristics, and used. In addition,
The calculation result is sent from the high-frequency simulator 12 to the waveform processing unit 15.
(Step 204).

In the waveform processing section 15, the difference processing of (design frequency characteristic calculated in step 204)-(frequency characteristic calculated in step 203) input from the high frequency simulator 12 is performed. The difference waveform obtained in this way becomes an expected value indicating how much the measured frequency characteristic moves when the adjustment is made by the amount of deviation of the dielectric resonator 22 (step 205).

Before proceeding to step 207, the search element value generation unit 13 sets all element values set in the high frequency simulator 12 to design values in accordance with an instruction from the high frequency simulator 12 (step 206).

In the search element value generator 13, one of the dielectric resonators 22 is selected, and this resonance frequency is changed by a physically adjustable minimum amount. The minimum amount that can be adjusted refers to the resolution of adjustment, and indicates the minimum amount that can be controlled when the electrode of the dielectric filter 2 to be adjusted is cut. The value of the resonance frequency of the dielectric resonator 22 after the change is held in the search element value generator 13. A search element value is output from the search element value generator 13 to the high frequency simulator 12,
The high frequency simulator 12 calculates the frequency characteristic in the element value for search. The calculation result is the high frequency simulator 1
2 to the waveform processing unit 15 (step 20).
7).

In the waveform processing unit 15, (Step 20
A difference corresponding to (frequency characteristic in the search element value calculated in step 7)-(frequency characteristic in the design value calculated in step 204) is calculated (step 208).

The waveform processing section 15 stores the frequency characteristic of step 201 stored in the waveform storage section 16 into the waveform storage section 16.
From (frequency characteristics measured in step 201) +
(The difference waveform calculated in step 205) + (step 2
Of the difference waveform calculated in step 08). The addition processing is performed for each frequency point similarly to the difference processing. The processing result is a frequency characteristic that can be expected to be obtained when the deviation amount of the dielectric resonator 22 obtained in step 202 and the shift amount set in step 207 are applied to the dielectric filter 2 to be adjusted. The data is output from the processing unit 15 to the standard determination unit 18.

Next, the standard determining unit 18 determines whether the waveform processing unit 15
Step 208 based on the addition processing result input from
At the specific standard point of the frequency characteristic calculated in step 1, how much the standard margin is. The standard point is a point in the frequency characteristics that must satisfy a preset condition. Taking FIG. 6 as an example, generally, standard values are not set for all points from O1 to On, but some standard points and standard values to be satisfied there are set. .

An example will be described in which two standard values are set. The standard point and the standard value are represented by, for example, Ok and the transfer coefficient S
21 is −2 dB or more, and the transfer coefficient S21 is −10 dB at Oj.
B or less. The margin of the standard is a difference between the standard value and the result of the addition processing. For example, if the result of the addition processing of Ok is -1 dB, the margin is 1 dB, and the result of the addition processing of Oj is -9 dB. If there is, the margin is -1 dB.
In actual adjustment, the margin is adjusted so as to be as large as possible, but it is necessary to balance the margin of a plurality of standard points. Therefore, for example, the sum of the margins of all the standard points is calculated to be the total margin. Other candidates such as a weighted average, a geometric mean, and the like are candidates for the overall margin, but are determined according to the characteristics of the dielectric filter 2 to be adjusted (step 209).

The standard determining unit 18 evaluates the total margin calculated in step 209, and when the margin is larger than a predetermined threshold value, notifies the adjustment amount instructing unit 17 of the completion of the adjustment from the standard determining unit 18 and notifies step 211 move on. When the margin is smaller than the predetermined threshold value, the standard determination unit 18 outputs a repeat instruction to the search element value generation unit 13 to return to step 207, and the dielectric resonance at step 207 performed last time is returned. It repeats again with a slight deviation from the set value of the child 22. In order to determine the direction in which the element value is shifted at this time, although not shown in the figure, the set value of the dielectric resonator 22 in the process 207 and the step 20 in that case are not shown.
The total margin calculated in step 9 is stored in the search element value generator 13, and the search element value generator 13 refers to this history and shifts the direction in which the margin is expected to increase. For this processing, a known search algorithm such as a hill-climbing method can be used (step 210).

In response to an instruction from the standard determining section 18, the adjustment amount instructing section 17 outputs from the search element value generating section 13 when the total margin finally becomes larger than a predetermined threshold value. The sum of the shift amount and the shift amount calculated in step 202 output from the neural network 14 is output to the adjusting means 4 as an adjustment amount. For output of the adjustment amount, for example, when instructing an operator, a monitor or the like is used as the adjusting means 4 and an instruction sentence is displayed on the screen. When a tool for shaving an electrode and a robot for positioning the tool are used, an electric signal suitable for the tool is used. The method of displaying the monitor screen and the method of using the robot are well-known techniques and will not be described in particular.

The above embodiment shows the case where the present invention is applied to the adjustment of a dielectric filter. In addition, the present invention is applied to a high frequency amplifier or the like whose characteristics can be simulated by a simulator. It is clear that the invention is applicable.

[0042]

As described above, the method and apparatus for adjusting a dielectric filter according to the present invention uses the neural network 14 for adjustment, and the teacher data of the neural network 14 is a high-frequency simulator. Since it is automatically generated by the step 12, it is not necessary to manually investigate the achievements of the skilled person, and there is an effect that the work efficiency is improved. Also,
Since it does not depend on the know-how of each worker, there is an effect that an adjustment system can be constructed without being affected by the personality of the worker and even when there is no skilled adjustment worker.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a mounting diagram illustrating an example of a dielectric filter.

FIG. 3 is a flowchart for generating teacher data.

FIG. 4 is a flowchart for performing adjustment.

FIG. 5 is a configuration diagram of a neural network.

FIG. 6 is an example of a frequency characteristic.

FIG. 7 is a configuration diagram showing a conventional example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Dielectric filter adjustment apparatus 1a Teacher data generation part 1b Adjustment control part 2 Dielectric filter 3 Network analyzer 4 Adjustment means 10 Element value generation part 11 Teacher data storage part 12 High frequency simulator 13 Element value generation part for search 14 Neural network Reference Signs List 15 Waveform processing unit 16 Waveform storage unit 17 Adjustment amount indicating unit 18 Standard determination unit 20 Case 21 Mounting substrate 22 Dielectric resonator 23 Coil 24 Capacitor 30 Input layer 31 Intermediate layer 32 Output element

Claims (3)

[Claims] 1. A method for adjusting a frequency characteristic of a dielectric filter, the method comprising: adjusting a frequency characteristic of the dielectric filter;
A first step of setting element values other than the adjustment elements to design values, and the simulator calculates frequency characteristics of the dielectric filter by varying n adjustment element values r1 to rn; The second is to store data obtained by combining the adjustment element values r1 to rn and frequency characteristics as teacher data.
And the third step of setting the adjustment element value set in the second step to a value different from the previous value and repeating until a predetermined combination is completed, and the frequency characteristic of the dielectric filter to be adjusted. Fourth to measure
Step, the combination of the teacher data stored in the second step and the third step, and the frequency characteristic of the dielectric filter to be adjusted measured in the fourth step in a neural network. A fifth step of inputting and calculating a deviation of the adjustment element value with respect to the teacher data, and a sixth step of calculating the frequency characteristic with respect to the deviation of the adjustment element value calculated in the fifth step by the simulator A step, a seventh step in which all the elements are set to design values and frequency characteristics are calculated by the simulator; a frequency characteristic calculated in the seventh step and a frequency characteristic calculated in the sixth step. In the eighth step of calculating the difference between the above-mentioned simulation, one of the adjusting elements is slightly changed from the designed value, and the other adjusting elements are set to the designed values. In the ninth step of calculating the frequency characteristics in, and then calculating the difference with the frequency characteristics calculated in the seventh step, and the frequency characteristics of the dielectric filter to be adjusted measured in the fourth step, Adding the differences in the frequency characteristics calculated in the eighth step and the ninth step,
Thereafter, at a specific standard point of the frequency characteristic, a tenth margin, which is a difference between the standard value and the result of the addition processing, is calculated.
And the ninth step and the first step until the margin in the tenth step exceeds a predetermined threshold.
An eleventh step of repeating step 0, the amount of shift of the adjustment element value in the ninth step when the margin exceeds a predetermined threshold value in the eleventh step, and the fifth step. A twelfth step of adjusting the adjustment element value of the dielectric filter to be adjusted by an amount corresponding to the sum of the deviations of the adjustment element values calculated in the step (b). . 2. The method for adjusting a dielectric filter according to claim 1, wherein the frequency characteristic in the second step is a difference from a frequency characteristic when a value of the adjustment element is a design value. . 3. An element value generating means for varying a plurality of adjustment element values of a dielectric filter in a plurality of steps, and setting element values other than the adjustment elements to design values, and a frequency characteristic A in the element value. Teacher data generating means comprising simulation means for calculating, teacher data storage means for storing teacher data obtained by combining the element values and the frequency characteristic A, and a dielectric filter to be adjusted The waveform storage means for storing the frequency characteristic B of the above, the combined teacher data and the frequency characteristic B stored in the teacher data storage means are learned, and the adjustment element values for the teacher data are learned. A neural network learning means for calculating a deviation; a search element value generating means for setting all elements to design values or slightly changing one of the adjustment elements from the design value; The simulation means for calculating a wave number characteristic C, a frequency characteristic D when all the elements are set to design values, and a frequency characteristic E when one of the adjusting elements is slightly changed from the design value; and the frequency characteristic B A waveform processing means for adding a difference between the frequency characteristic D and the frequency characteristic C, a difference between the frequency characteristic E and the frequency characteristic D, and a difference between a standard value and the addition result at a specific standard point of the frequency characteristic. Is calculated, and if the margin exceeds a predetermined threshold, adjustment is instructed.If the margin does not exceed the threshold, the adjustment element value is reset until the margin is exceeded, and the frequency characteristic E is changed. It is characterized by having standard specification means for instructing recalculation, and adjustment control means consisting of adjustment instruction means for performing adjustment in accordance with instructions from the standard judgment means and the search element value generation means. Dielectric filter adjustment device.