Program, method, trained model, and method of generating trained model

Abstract

To provide a program, a method, and a trained model for improving the accuracy of an S-parameter of a high frequency circuit.SOLUTION: An estimation apparatus 20 functions as: an acquisition unit 22 which acquires first information pertaining to a linear circuit having a plurality of ports for receiving or outputting a high frequency signal, and second information pertaining to a frequency of the high frequency signal; a first estimation unit 24 which estimates, when the second information indicates that the frequency is 0 Hz, an S-parameter for two ports of the plurality of ports when the frequency is 0 Hz, from the first information based on a first trained model; and a second estimation unit 26 which estimates, when the second information indicates that the frequency is other than 0 Hz, an S-parameter at the frequency from the first information and the second information, based on a second trained model.SELECTED DRAWING: Figure 8

Description

Technical Field

[0001] The present invention relates to a program, a method, a learned model, and a method for generating a learned model.

Background Art

[0002] In the circuit design of a high-frequency circuit, an equivalent circuit model using a lumped constant circuit with lumped constant elements is used (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in a high-frequency band with a high frequency, the wavelength of the high-frequency signal is not sufficiently large with respect to the lumped constant element. For this reason, when the frequency changes, the high-frequency characteristics of the high-frequency circuit may not match the equivalent circuit model using the lumped constant circuit. Also, in a model representing a high-frequency circuit by S-parameters, it is difficult to create a model every time the configuration of the high-frequency circuit is changed. Although a method of estimating the S-parameters of a high-frequency circuit using a machine-learned model can be considered, the accuracy of the S-parameters may be low.

[0005] The present disclosure has been made in view of the above problems, and an object thereof is to improve the accuracy of S-parameters.

Means for Solving the Problems

[0006] One embodiment of the present disclosure causes a computer to function as an acquisition unit that acquires first information regarding a linear circuit having a plurality of ports for inputting or outputting high-frequency signals and second information regarding the frequency of the high-frequency signals, a first estimation unit that estimates S-parameters for two ports out of the plurality of ports when the frequency is 0 Hz based on the first learned model from the first information when the second information indicates that the frequency is 0 Hz, and a second estimation unit that estimates the S-parameters at the frequency based on the first information and the second information from the second learned model when the second information indicates that the frequency is other than 0 Hz. The first learned model is generated by machine learning a plurality of first teacher data that defines the relationship between the plurality of first information of the linear circuit and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz. The second learned model is generated by machine learning a plurality of second teacher data that defines the relationship between the plurality of first information of the linear circuit, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies. It is a program.

[0007] One embodiment of the present disclosure is a learned model for estimating S-parameters for two ports out of the plurality of ports based on first information regarding a linear circuit having a plurality of ports for inputting or outputting high-frequency signals and second information regarding the frequency of the high-frequency signals. The learned model includes a first learned model generated by machine learning a plurality of first teacher data that defines the relationship between the plurality of first information and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz, and a second learned model generated by machine learning a plurality of second teacher data that defines the relationship between the plurality of first information, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies.

[0008] One embodiment of the present disclosure relates to a method for generating a learned model for estimating S-parameters for two ports among a plurality of ports based on first information regarding a linear circuit having a plurality of ports for inputting or outputting high-frequency signals and second information regarding frequency, the method including: generating a first learned model by machine learning a plurality of first teacher data defining a relationship between a plurality of the first information and a plurality of the S-parameters calculated for the plurality of the first information with the frequency set to 0 Hz; and generating a second learned model by machine learning a plurality of second teacher data defining a relationship between the plurality of the first information and a plurality of frequencies including frequencies other than 0 Hz and a plurality of the S-parameters calculated for each of the plurality of the first information and the plurality of frequencies.

[0009] The present disclosure can be realized not only as such a characteristic estimation program and estimation method, but also as an estimation device that processes such characteristic steps. Further, it can be realized as a semiconductor integrated circuit that realizes part or all of the estimation device, or as an estimation system including the estimation device.

Advantages of the Invention

[0010] According to the present disclosure, the accuracy of the S-parameters can be improved.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Mode for Carrying Out the Invention

[0012] [Description of Embodiments of the Present Disclosure] First, the contents of the embodiments of the present disclosure will be listed and described. (1) One embodiment of the present disclosure causes a computer to function as an acquisition unit that acquires first information regarding a linear circuit having a plurality of ports for inputting or outputting high-frequency signals and second information regarding the frequency of the high-frequency signals, a first estimation unit that, when the second information indicates that the frequency is 0 Hz, estimates S-parameters for two of the plurality of ports when the frequency is 0 Hz based on the first learned model from the first information, and a second estimation unit that, when the second information indicates that the frequency is other than 0 Hz, estimates the S-parameters at the frequency based on the first information, the second information, and the second learned model. The first learned model is generated by machine learning a plurality of first teacher data that defines the relationship between the plurality of first information of the linear circuit and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz. The second learned model is generated by machine learning a plurality of second teacher data that defines the relationship between the plurality of first information of the linear circuit, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies. This can improve the accuracy of estimating S-parameters. (2) In the above (1), the S-parameters in the first teacher data may be normalized by a first maximum value and a first minimum value, and the S-parameters in the second teacher data may be normalized by a second maximum value and a second minimum value. This can further improve the accuracy of estimating S-parameters. (3) In the above (2), the difference between the first maximum value and the first minimum value may be smaller than the difference between the second maximum value and the second minimum value. This can further improve the accuracy of estimating S-parameters. (4) In the above (3), the first estimation unit may estimate the S parameter by decoding a value generated based on the first learned model based on the first maximum value and the first minimum value, and the second estimation unit may estimate the S parameter by decoding a value generated based on the second learned model based on the second maximum value and the second minimum value. Thereby, the accuracy of estimating the S parameter can be further improved. (5) In any one of the above (1) to (4), the two ports may be open or shorted between each other at 0 Hz. Thereby, the accuracy of estimating the S parameter can be further improved. (6) In the above (5), when the two ports are port 1 and port 2, the S parameter may include S21. Thereby, the accuracy of estimating the S parameter can be further improved. (7) In any one of the above (1) to (6), the second learned model may be generated by machine learning a plurality of second teacher data that define a relationship between the plurality of first information of the linear circuit, the plurality of frequencies including 0 Hz and other than 0 Hz, and the plurality of S parameters calculated for each of the plurality of first information and the plurality of frequencies. Thereby, the S parameter can be accurately estimated. (8) One embodiment of the present disclosure includes steps of obtaining first information regarding a linear circuit having a plurality of ports to which a high-frequency signal is input or output and second information regarding the frequency of the high-frequency signal; when the second information indicates that the frequency is 0 Hz, estimating S-parameters for two of the plurality of ports when the frequency is 0 Hz based on the first information and a first learned model; and when the second information indicates that the frequency is other than 0 Hz, estimating the S-parameters at the frequency based on the first information, the second information, and a second learned model. The first learned model is generated by machine learning a plurality of first teacher data defining a relationship between the plurality of first information of the linear circuit and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz. The second learned model is generated by machine learning a plurality of second teacher data defining a relationship between the plurality of first information of the linear circuit, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies. Thereby, the accuracy of estimating the S-parameters can be improved. (9) One embodiment of the present disclosure is a learned model for estimating S-parameters for two of the plurality of ports based on first information regarding a linear circuit having a plurality of ports to which a high-frequency signal is input or output and second information regarding the frequency of the high-frequency signal, the learned model including a first learned model generated by machine learning a plurality of first teacher data defining a relationship between the plurality of first information and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz, and a second learned model generated by machine learning a plurality of second teacher data defining a relationship between the plurality of first information, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies. Thereby, the accuracy of estimating the S-parameters can be improved. (10) One embodiment of the present disclosure relates to a method for generating a learned model for estimating S-parameters for two ports among the plurality of ports based on first information regarding a linear circuit having a plurality of ports for inputting or outputting high-frequency signals and second information regarding frequency. The method includes generating a first learned model by machine learning a plurality of first teacher data that defines a relationship between a plurality of the first information and a plurality of the S-parameters calculated for the plurality of first information with the frequency set to 0 Hz; and generating a second learned model by machine learning a plurality of second teacher data that defines a relationship between the plurality of the first information and a plurality of frequencies including frequencies other than 0 Hz and a plurality of the S-parameters calculated for each of the plurality of the first information and the plurality of frequencies. This can improve the accuracy of estimating S-parameters. (11) One embodiment of the present disclosure includes an acquisition unit that acquires first information regarding a linear circuit having a plurality of ports for inputting or outputting high-frequency signals and second information regarding the frequency of the high-frequency signal; a first estimation unit that estimates S-parameters for two ports among the plurality of ports when the frequency is 0 Hz based on the first learned model from the first information when the second information indicates that the frequency is 0 Hz; and a second estimation unit that estimates the S-parameters at the frequency based on the first information, the second information, and the second learned model when the second information indicates that the frequency is other than 0 Hz. The first learned model is generated by machine learning a plurality of first teacher data that defines a relationship between a plurality of the first information of the linear circuit and a plurality of the S-parameters calculated for the plurality of first information with the frequency set to 0 Hz, and the second learned model is generated by machine learning a plurality of second teacher data that defines a relationship between a plurality of the first information of the linear circuit and a plurality of frequencies including frequencies other than 0 Hz and a plurality of the S-parameters calculated for each of the plurality of the first information and the plurality of frequencies. This can improve the accuracy of estimating S-parameters. (12) One embodiment of the present disclosure relates to a processor that acquires first information regarding a memory and a linear circuit having a plurality of ports through which a high-frequency signal is input or output, and second information regarding the frequency of the high-frequency signal, and when the second information indicates that the frequency is 0 Hz, based on a first learned model from the first information, estimates S-parameters for two ports out of the plurality of ports when the frequency is 0 Hz, and when the second information indicates that the frequency is other than 0 Hz, estimates the S-parameters at the frequency based on a second learned model from the first information and the second information. The first learned model is generated by machine learning a plurality of first teacher data that defines the relationship between the plurality of the first information of the linear circuit and the plurality of the S-parameters calculated for the plurality of the first information with the frequency being 0 Hz. The second learned model is generated by machine learning a plurality of second teacher data that defines the relationship between the plurality of the first information of the linear circuit and a plurality of the frequencies including other than 0 Hz, and the plurality of the S-parameters calculated for each of the plurality of the first information and the plurality of the frequencies. This can improve the accuracy of estimating S-parameters.

[0013] [Details of Embodiments of the Present Disclosure] Specific examples of a program, a method, a learned model, and a method for generating a learned model according to an embodiment of the present disclosure will be described below with reference to the drawings. Note that the present disclosure is not limited to these examples, and is intended to cover all modifications within the meaning and scope equivalent to the claims, as indicated by the claims.

[0014] At least a part of the embodiments described below may be arbitrarily combined. The estimation device is configured to include a computer, and each function of the estimation device is exhibited by a computer program stored in a storage device of the computer being executed by a CPU (Central Processing Unit) of the computer. The computer program can be stored in a storage medium such as a CD-ROM (Compact Disc Read Only Memory) or a DVD (Digital Versatile Disc).

[0015] [Example 1] [Explanation of Linear Circuit] FIG. 1 is a block diagram of a linear circuit for estimating S parameters in Example 1. As shown in FIG. 1, the linear circuit 10 includes a plurality of ports P1, P2,..., Pi to Pn−1, Pn through which a high-frequency signal is input or output. Here, n is an integer corresponding to the number of ports P1 to Pn of the linear circuit 10, and is 2 or more.

[0016] [Specific Example 1 of Linear Circuit] FIG. 2 is a plan view of Specific Example 1 of the linear circuit in Example 1. As shown in FIG. 2, in the linear circuit 10a of Specific Example 1, a metal pattern 12 is provided on a dielectric substrate 11. The metal pattern 12 forms spiral inductors L1 and L2. Both ends of the inductor L1 are used as ports P1 and P2, and both ends of the inductor L2 are used as ports P3 and P4. Let the line width of the inductor L1 be W1, the line length be D1, and the number of turns be N1. Let the line width of the inductor L2 be W2, the line length be D2, and the number of turns be N2. Let the distance between the inductors L1 and L2 be D3.

[0017] [Specific Example 2 of Linear Circuit] FIG. 3 is a plan view and a cross-sectional view of a specific example 2 of the linear circuit in Example 1. As shown in FIG. 3, in the linear circuit 10b of the specific example 2, metal patterns 12a and 12b are provided on the dielectric substrate 11. A dielectric layer 13 is provided between the metal patterns 12a and 12b. The dielectric layer 13, and the metal patterns 12a and 12b sandwiching the dielectric layer 13, form a capacitor C. Let the thickness of the dielectric layer 13 be t, and the area of the region where the metal patterns 12a and 12b overlap with the dielectric layer 13 in between be Ar. The ends of the metal patterns 12a and 12b are ports P1 and P2, respectively.

[0018] The information for estimating the S-parameters of the linear circuit 10 is at least one of, for example, dimensional information regarding dimensions and physical property information regarding the physical properties of materials. In the specific example 1, the dimensional information is the line widths W1, W2, the line lengths D1, D2, the number of turns N1, N2, and the distance D3. The physical property information is the permittivity, the dielectric loss tangent of the dielectric substrate 11, and the conductivity of the metal pattern 12, etc. In the specific example 2, the dimensional information is the thickness t and the area Ar. The physical property information is the permittivity, the dielectric loss tangent of the dielectric layer 13, and the conductivity of the metal patterns 12a and 12b, etc.

[0019] For the design of the linear circuit 10, an equivalent circuit represented by a lumped-constant circuit using linear elements such as resistors, inductors, capacitors, and transmission lines is used. Or, the scattering matrix (S-parameters) for each frequency of the high-frequency signal is used. In particular, for circuit designs in, for example, microwaves (from 300 MHz to 30 GHz) or millimeter waves (from 30 GHz to 300 GHz), S-parameters are often used.

[0020] [Explanation of S-parameters] The S-parameters (scattering matrix) of the linear circuit 10 will be explained. Equation 1 is the matrix of the S-parameters of the linear circuit 10. [Equation]

[0021] Let the S-parameter from port Pk to port Pl be the element Skl. The element Skl of the S-parameter is a complex number. In a linear circuit, Skl = Slk. In Specific Example 1 of Figure 2, n = 4, and in Specific Example 2 of Figure 3, n = 2.

[0022] [Linear Circuit Model Using a Trained Model] An equivalent circuit represented by lumped constants has difficulty expressing high-frequency characteristics with high accuracy over a wide bandwidth. Since the S-parameters represent the high-frequency characteristics for each frequency, they can express high-frequency characteristics with high accuracy over a wide bandwidth. However, by performing electromagnetic field analysis each time the parameters of the linear circuit are different, the S-parameters are calculated. Electromagnetic field analysis is time-consuming and labor-intensive. Therefore, it is difficult to design a circuit by changing the dimensional information of the linear circuit.

[0023] Therefore, a circuit model based on a neural network is proposed, with the parameters of the linear circuit as explanatory variables and the S-parameters as objective variables.

[0024] Hereinafter, an estimation method for the S-parameters of the linear circuit in Example 1 will be described.

[0025] [Block Diagram of a Computer] Figure 4 is a block diagram of the computer in Example 1. The computer 30 functions as an estimation device that cooperates with software to estimate the S-parameters of the linear circuit 10. The computer 30 executes an estimation program and performs an estimation method.

[0026] Computer 30 includes a processor 32, a memory 34, an input / output device 36, and an internal bus 38. The processor 32 is, for example, a CPU (Central Processing Unit) and executes programs and methods. The memory 34 is, for example, a volatile memory or a non-volatile memory and stores data and the like used when the processor 32 executes programs and methods. The memory 34 may store the programs executed by the processor 32. The input / output device 36 inputs data acquired by the processor 32 from an external device and outputs data output by the processor 32 to the external device. The external device is another computer or another program in the same computer. The internal bus 38 connects the processor 32, the memory 34, and the input / output device 36 and transmits data and the like. The program is stored in a storage medium 35. The storage medium 35 is, for example, a non-transitory tangible medium such as a CD-ROM or a DVD.

[0027] [Method for Generating Teacher Data] FIG. 5 is a flowchart showing the method for generating teacher data in the first embodiment. Each step in FIG. 5 may be executed by the computer shown in FIG. 4 or by a human. FIG. 6 is a diagram showing an example of teacher data in the first embodiment.

[0028] As shown in FIG. 5, set i = 1 and j = 0 (step S30). i is an integer from 1 to N, and j is an integer from 0 to M.

[0029] Next, set information A to A(i) and set frequency f to f(j) (step S32). A(i) is, for example, dimension information and physical property information. When the material used for the linear circuit 10 is fixed, the physical property information may not be used. In Specific Example 1, the dimension information is D1(i), D2(i), W1(i), W2(i), N1(i), N2(i), and D3(i). i corresponds from 1 to N, and the values of the dimension information are made different. The frequency f(j) includes 0 Hz and other than 0 Hz. When j = 0, for example, f(0) = 0 Hz.

[0030] Perform electromagnetic field analysis at information A(i) and frequency f(j) to calculate the S-parameter S(i,j) (step S34). Here, the matrix of S-parameters is an n-by-n matrix such as Equation 1. In Specific Example 1, it is a 4-by-4 matrix, and in Specific Example 2, it is a 2-by-2 matrix. The S-parameter S(i,j) is at least one element among the elements of the matrix of Equation 1 for which estimation is performed.

[0031] Determine whether j = 0 (step S36). If No, proceed to step S42. If Yes, normalize the calculated S-parameter S(i,j) (step S38). When S(i,j) is a plurality of elements among the matrix of Equation 1, normalize each element. Let the maximum value for normalization be MAX1 and the minimum value be MIN1. When S(i,j) is a plurality of elements, the maximum value MAX1 and the minimum value MIN1 can be set for each element. The normalized S-parameter NS1(i,j) = (S(i,j) - MIN1) / (MAX1 + MIN1).

[0032] Generate teacher data T1(i,0) (step S40). For example, when i = 1 and j = 0, as shown in FIG. 6, teacher data T1(1,0) defining the relationship between information A(1), frequency f(0) = 0 Hz, and the normalized S-parameter NS1(1,0) is generated.

[0033] Next, normalize the calculated S(i,j) (step S42). Let the maximum value for normalization be MAX2 and the minimum value be MIN2. MAX2 is different from MAX1, and MIN2 is different from MIN1. MAX2 - MIN2 is larger than MAX1 - MIN1. When S(i,j) is a plurality of elements, the maximum value MAX2 and the minimum value MIN2 can be set for each element. The normalized S-parameter NS2(i,j) = (S(i,j) - MIN2) / (MAX2 + MIN2).

[0034] Generate teacher data T2(i, j) (step S44). For example, when i = 1 and j = 0, as shown in FIG. 6, teacher data T2(1, 0) that defines the relationship between information A(1) and the normalization S parameter NS2(1, 0) with a frequency f(0) = 0 Hz is generated.

[0035] Next, determine whether i = N (step S46). If No, return to step S32. If Yes, determine whether j = M (step S48). If No, return to step S32.

[0036] As described above, when the frequency f(0) = 0 Hz, N pieces of teacher data T1(i, 0) that define the relationship between information A(i) and the S parameter S(i, 0) are generated.

[0037] Also, N × (M + 1) pieces of teacher data T2(i, j) that define the relationship between information A(i), the frequency f(j), and the S parameter S(i, 0) are generated.

[0038] In step S48, if Yes, output teacher data T1 and T2 to the memory 34 or an external device (step S50). Then end.

[0039] [Method for Generating a Trained Model] FIG. 7 is a flowchart showing a method for generating a trained model in the first embodiment. As shown in FIG. 7, the processor 32 acquires N pieces of teacher data T1 from the memory 34 or an external device (step S60). The processor 32 performs machine learning based on the acquired N pieces of teacher data T1 to generate a first trained model M1 (step S61). The processor 32 or a human verifies the first trained model M1 (step S62). For example, the S parameter S is estimated from known information A using the first trained model M1 for the relationship between information A and the S parameter. If the estimated S parameter substantially matches the known S parameter, it is determined as Yes in step S62, and if not, it is determined as No. If No, return to step S61.

[0040] When the answer is Yes in step S62, the processor 32 acquires N×(M + 1) pieces of teacher data T2 from the memory 34 or an external device (step S63). Based on the acquired N×(M + 1) pieces of teacher data T2, the processor 32 performs machine learning to generate a second learned model M2 (step S64). The processor 32 or a human verifies the second learned model M2 (step S65). The determination method is the same as in step S62. When the answer is No, the process returns to step S64. When the answer is Yes, the first learned model M1 and the second learned model M2 are output to the memory 34 or an external device (step S66). Then the process ends. The execution order of steps S60 to S62 and steps S63 and S65 is arbitrary. As described above, the first learned model M1 and the second learned model M2 are generated.

[0041] [Estimation method of S parameter] [Functional block diagram] FIG. 8 is a functional block diagram of the estimation device according to the first embodiment. As shown in FIG. 8, the estimation device 20 includes an acquisition unit 22, a first estimation unit 24, a second estimation unit 26, and an output unit 28. The processor 32 cooperates with software and functions as the acquisition unit 22, the first estimation unit 24, the second estimation unit 26, and the output unit 28. The acquisition unit 22 acquires information A and frequency f of the linear circuit 10 from an external device via the input / output device 36. When the frequency f is 0 Hz, the first estimation unit 24 estimates the S parameter S based on the information A and the frequency f from the first learned model M1. When the frequency f is other than 0 Hz, the second estimation unit 26 estimates the S parameter S based on the information A and the frequency f from the second learned model M2. The output unit 28 outputs the S parameter S estimated by the first estimation unit 24 or the second estimation unit 26 to an external device via the input / output device 36.

[0042] [Flowchart] FIG. 9 is a flowchart showing a method for estimating the S parameter according to the first embodiment. As shown in FIG. 9, the acquisition unit 22 acquires information A (first information) regarding the linear circuit 10 and information regarding the frequency f of the high-frequency signal (second information) (step S10).

[0043] The processor 32 determines whether the frequency f = 0 Hz (step S12). When Yes, the first estimation unit 24 acquires the first learned model M1 (step S14A). The first estimation unit 24 may acquire the maximum value MAX1 and the minimum value MIN1 for normalization together with the first learned model M1. The first estimation unit 24 applies the first learned model M1 to the information A and estimates the normalization S parameter NS (step S16A).

[0044] The first estimation unit 24 decodes the normalization S parameter NS (step S18A). The decoded S parameter S can be calculated by S = NS × (MAX1 + MIN1) + MIN1 using the normalization S parameter NS. When the normalization S parameter NS is a plurality of elements of a matrix, the S parameter may be calculated for each element. Then, the output unit 28 outputs the S parameter S to the memory 34 or an external device (step S20). Then it ends.

[0045] In step S12, when No, the second estimation unit 26 acquires the second learned model M2 (step S14B). The second estimation unit 26 may acquire the maximum value MAX2 and the minimum value MIN2 for normalization together with the second learned model M2. The second estimation unit 26 applies the second learned model M2 to the information A and the frequency f and estimates the normalization S parameter NS (step S16B).

[0046] The second estimation unit 26 decodes the normalization S parameter NS (step S18B). The decoded S parameter S can be calculated by S = NS × (MAX2 + MIN2) + MIN2. When the normalization S parameter NS is a plurality of elements of a matrix, the S parameter may be calculated for each element. Then, the output unit 28 outputs the S parameter S to the memory 34 or an external device. Then it ends.

[0047] [Comparative Example 1] FIG. 10 is a flowchart showing a method for estimating the S parameter in Comparative Example 1. As shown in FIG. 10, in Comparative Example 1, step S12 in FIG. 9 of Example 1 is not provided. After step S10, regardless of the frequency f, the processor 32 obtains a model M corresponding to the second learned model M2 as a learned model (step S14). The learned model M is applied to the information A and the frequency f to estimate a normalized S parameter NS (step S16). The processor 32 decodes the normalized S parameter NS using MAX2 and MIN2 (step S18). Thereby, the S parameter S is generated. Thereafter, the processor 32 outputs the S parameter S (step S20).

[0048] The problem of Comparative Example 1 will be described with a linear circuit as a resistor R. FIG. 11 is an equivalent circuit of the linear circuit. As shown in FIG. 11, in the linear circuit 10c, a resistor R is electrically connected between the ports P1 and P2. The ports P1 and P2 are terminated by a reference impedance Z0 (for example, 50 Ω).

[0049] When it is direct current (that is, when the frequency is 0 Hz), consider S21 in the matrix of the S parameters of the linear circuit 10c. Since the resistor R has almost no reactance component, S21 is almost a real number. Therefore, S21 will be described as the real part of S21. Let the resistance value of the resistor R be R and the impedance (resistance value because it is a real number) of the reference impedance be Z0, then S21 = 2×Z0 / (R + 2×Z0).

[0050] Here, let Z0 be 50 Ω. FIG. 12 is a diagram showing S21 with respect to the resistance value R of the resistor R. As shown in FIG. 12, when the resistance value R is 100 Ω, S21 is 0.5. When the resistance value R is 1 Ω or less, S21 is almost 1. When the resistance value R is 10000 Ω or more, S21 is almost 0. When the resistance value R is small, it corresponds to a short between the ports P1 and P2, and when the resistance value R is large, it corresponds to an open between the ports P1 and P2.

[0051] FIG. 13 is a diagram showing S21 when there is a short between ports P1 and P2 in direct current. In FIG. 13, the horizontal direction shows S21. The lower part is an enlarged view of the S21 in the upper part near +1. In the case of a short in direct current, the range RDC that S21 can take is close to +1, for example, from +0.99945 to +1. This corresponds to a resistance value R from 0.055 Ω to 0 Ω. On the other hand, in alternating current, S21 can take a range RAC from -0.04 to +1. For example, between ports P1 and P2 in Specific Example 1, there is a short in direct current. However, in alternating current with a frequency f, this is because the impedance of inductor L1 becomes j(2πf)L1 (j is the imaginary unit).

[0052] FIG. 14 is a diagram showing S21 when there is an open between ports P1 and P2 in direct current. In FIG. 14, the horizontal direction shows S21. The lower part is an enlarged view of the S21 in the upper part near 0. In the case of an open in direct current, the range RDC that S21 can take is close to 0, for example, from 0 to +0.0000021. This corresponds to an infinite Ω to 47.6 MΩ in terms of resistance value. On the other hand, in alternating current, S21 can take a range RAC from -0.04 to +0.16. For example, between ports P1 and P2 in Specific Example 2, there is an open in direct current. However, in alternating current with a frequency f, this is because the impedance of capacitor C becomes -1 / j(2πf)C (j is the imaginary unit).

[0053] In Comparative Example 1, the same learned model is used regardless of direct current or alternating current. For example, when there is an open in direct current, normalization is performed using the range RAC. In this case, the maximum value MAX2 and the minimum value MIN2 are used.

[0054] Assume that the resistance value R between ports P1 and P2 in direct current is 100 MΩ. At this time, the real part of S21 is 1×10 -6 . When normalizing S21 with MAX2 = 0.16 and MIN2 = -0.04, NS21 = (1×10 -6 - (-0.04)) / (0.16 + (-0.04)) = 0.200005. A learned model is generated based on this NS21.

[0055] In step S18 of FIG. 10, when estimating NS21 using this learned model, if NS21 included an error of 0.001 with respect to 0.200005, in the decoding step in step S18, S21 = 0.201005×(0.16 - (-0.04)) + (-0.04) = 0.000201. This corresponds to 500 kΩ, and due to an error of 0.0001, the resistance value R becomes 1 / 200.

[0056] As shown in FIG. 12, when there is a short or open in direct current, even if the resistance value R changes, S21 hardly changes. Therefore, when estimating S21 using normalization and decoding in the cases of alternating current and direct current, the error of the resistance value R in the case of direct current becomes large.

[0057] [Description of Example 1] Therefore, in Example 1, when there is an open in direct current, for the normalization and decoding of data with a frequency of 0 Hz, the range RDC is used. In this case, they are the maximum value MAX1 and the minimum value MIN1.

[0058] The resistance value in direct current between port P1 and P2 is 100 MΩ, and the real part of S21 is 1×10 -6 is. When normalizing S21 with MAX1 = +0.0000021 and MIN2 = 0, NS21 = (1×10 -6 -0) / (0.0000021 + 0) = 0.47619.

[0059] If NS21 included an error of 0.001 with respect to 0.47619, in the decoding step in step S18, S21 = 0.47719×(0.0000021 + 0) - 0 = 1.0021×10 -6 corresponds to, and the resistance value R is approximately 100 MΩ. Thus, even with an error, the resistance value R does not change significantly.

[0060] According to Example 1, as shown in FIGS. 8 and 9, when the frequency f is 0 Hz, the first estimation unit 24 estimates the S parameter S based on the information A and the first learned model M1. When the second estimation unit 26 indicates that the frequency f is other than 0 Hz, the S parameter S at the frequency f is estimated based on the information A and the frequency f and the second learned model M2. As shown in FIGS. 5 to 6, the first learned model M1 is obtained by machine learning a plurality of pieces of teacher data T1(i,0) (first teacher data) that define the relationship between a plurality of pieces of information A(i) of the linear circuit 10 and a plurality of S parameters S(i,0) calculated for the plurality of pieces of information A(i) with the frequency f being 0 Hz. The second learned model M2 is obtained by machine learning a plurality of pieces of teacher data T2(i,j) (second teacher data) that define the relationship between a plurality of pieces of information A(i) and a plurality of frequencies f(j) including other than 0 Hz and a plurality of S parameters S(i,j) calculated for each of the plurality of pieces of information A(i) and the plurality of frequencies f(j).

[0061] In this way, the S parameter S of direct current with the frequency f being 0 Hz is estimated using the first learned model M1 generated using the teacher data T1(i,0) when the frequency f is 0 Hz. The S parameter S of alternating current with the frequency f being other than 0 Hz is estimated using the second learned model M2 generated using the teacher data T2(i,0) including when the frequency f is other than 0 Hz. Thereby, the accuracy of the estimation of the S parameter of direct current and the estimation of the S parameter of alternating current can be improved. In this way, the accuracy of the model can be improved. The S parameter to be estimated may be the S parameter for two ports out of the plurality of ports P1 to Pn in FIG. 1.

[0062] For all elements in the S parameter matrix, the S parameter may be estimated. For some elements in the S parameter matrix, the S parameter may not be estimated. For some elements of the S parameter, the S parameter may be estimated using the first learned model M1 and the second learned model M2, and for other elements, the S parameter may be estimated using the second learned model M2 regardless of whether the frequency f is 0 Hz or other than 0 Hz.

[0063] The S parameter in the teacher data T1(i,0) is normalized by the maximum value MAX1 (the first maximum value) and the minimum value MIN1 (the first minimum value), and the S parameter in the teacher data T2(i,j) is normalized by the maximum value MAX2 (the second maximum value) and the minimum value MIN2 (the second minimum value). In this way, the teacher data T1(i,0) and T2(i,j) are normalized using different maximum and minimum values. Thereby, even when the maximum and minimum values of the DC S parameter and the maximum and minimum values of the AC S parameter are significantly different, the S parameter can be accurately estimated.

[0064] The difference between the maximum value MAX1 and the minimum value MIN1 is smaller than the difference between the maximum value MAX2 and the minimum value MIN2. Thereby, even when the range RDC of the DC S parameter is smaller than the range RAC of the AC S parameter, the S parameter can be accurately estimated. The range RDC is, for example, 1 / 10 or less of the range RAC and 1 / 100 times or less.

[0065] The first estimation unit 24 estimates the S parameter by decoding the value generated based on the first learned model M1 based on the maximum value MAX1 and the minimum value MIN1. The second estimation unit 26 estimates the S parameter by decoding the value generated based on the second learned model M2 based on the maximum value MAX2 and the minimum value MIN2. Thereby, even when the maximum and minimum values of the DC S parameter and the maximum and minimum values of the AC S parameter are significantly different, the S parameter can be accurately estimated.

[0066] The S parameter to be estimated has the ports open or shorted at a frequency f of 0 Hz. In this case, when the method of Comparative Example 1 is used, the accuracy of estimating the S parameter decreases. Therefore, by estimating the S parameter using the method of Example 1, the accuracy of estimating the S parameter can be improved.

[0067] When the S parameter is S21, as described in FIGS. 13 and 14, if the method of Comparative Example 1 is used, the accuracy of estimating the S parameter decreases. Therefore, by estimating the S parameter using the method of Example 1, the accuracy of estimating the S parameter can be improved. Note that in the linear circuit 10, S21 and S12 are equivalent.

[0068] Consider the case of estimating the S parameter at a frequency f between the frequency f(0) = 0 Hz and the lowest frequency f(1) other than 0 Hz among the teacher data when the teacher data T2(i,j) does not include data at the frequency f = 0 Hz. In this case, it is necessary to estimate the S parameter at a frequency f outside the frequency range of the teacher data T2, and the S parameter cannot be accurately estimated. Therefore, the second trained model M2 is generated by machine learning the teacher data T2(i,j) that includes both 0 Hz and frequencies other than 0 Hz. Thereby, even when estimating the S parameter at a frequency f between f(0) and f(1), the S parameter can be accurately estimated.

[0069] Each process (each function) of the above-described embodiment is realized by a processing circuit including one or more processors. The processing circuit may be configured by, in addition to the one or more processors, an integrated circuit in which one or more memories, various analog circuits, and various digital circuits are combined. The one or more memories store programs (instructions) for causing the one or more processors to execute the respective processes. The one or more processors may execute the respective processes according to the programs read from the one or more memories, or may execute the respective processes according to a logic circuit designed in advance to execute the respective processes.

[0070] The above-mentioned processor may be various processors suitable for controlling a computer, such as a CPU, a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an FPGA (Field Programmable Gate Array), an ASIC (Application Specication Integrated Circuit), etc. Note that the above-mentioned plurality of physically separated processors may cooperate with each other to execute the above-mentioned respective processes. For example, the processors installed in each of a plurality of physically separated computers may cooperate with each other via a network such as a LAN (Local Area Network), a WAN (Wide Area Network), or the Internet to execute the above-mentioned respective processes.

[0071] The above-mentioned program may be installed in the above-mentioned memory via the above-mentioned network from an external server device or the like, or may be distributed in a state stored in a recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory, and may be installed from the above-mentioned recording medium into the above-mentioned memory.

[0072] It should be considered that the embodiments disclosed this time are illustrative in all respects and not restrictive. The scope of the present disclosure is shown not by the above meaning, but by the scope of the claims, and it is intended that all modifications within the meaning and scope equivalent to the scope of the claims are included.

Explanation of Reference Numerals

[0073] 10, 10a, 10b, 10c Linear Circuit 11 Dielectric Substrate 12, 12a, 12b Metal Pattern 13 Dielectric Layer 20 Estimation Device 22 Acquisition Unit 24 First Estimation Unit 26 Second Estimation Unit 28 Output Unit 30 Computer 32 Processor 34 Memory 36 Input / Output Device 38 Internal Bus M1 First Learned Model M2 Second Learned Model MAX1 (First Maximum Value), MAX2 (Second Maximum Value) Maximum Value MIN1 (First Minimum Value), MIN1 (Second Minimum Value) Minimum Value T1 (First Teacher Data), T2 (Second Teacher Data) Teacher Data

Claims

1. A computer, an acquisition unit that acquires first information regarding a linear circuit having a plurality of ports to which a high-frequency signal is input or output and second information regarding the frequency of the high-frequency signal; a first estimation unit that estimates S-parameters for two ports among the plurality of ports when the frequency is 0 Hz based on the first learned model from the first information when the second information indicates that the frequency is 0 Hz; a second estimation unit that estimates the S-parameters at the frequency based on the second learned model from the first information and the second information when the second information indicates that the frequency is other than 0 Hz; causes to function as, the first learned model is generated by machine learning a plurality of first teacher data defining a relationship between the plurality of first information of the linear circuit and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz; the second learned model is generated by machine learning a plurality of second teacher data defining a relationship between the plurality of first information of the linear circuit and a plurality of frequencies including other than 0 Hz and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies, a program.

2. the S-parameters in the first teacher data are normalized by a first maximum value and a first minimum value, the program according to claim 1, wherein the S-parameters in the second teacher data are normalized by a second maximum value and a second minimum value.

3. the program according to claim 2, wherein a difference between the first maximum value and the first minimum value is smaller than a difference between the second maximum value and the second minimum value.

4. the first estimation unit estimates the S-parameters by decoding a value generated based on the first learned model based on the first maximum value and the first minimum value; The program according to claim 3, wherein the second estimation unit estimates the S parameter by decoding a value generated based on the second learned model based on the second maximum value and the second minimum value.

5. The program according to any one of claims 1 to 4, wherein the two ports are open or shorted at a frequency of 0 Hz.

6. The program according to claim 5, wherein the S parameter includes S21 when the two ports are port 1 and port 2.

7. The program according to any one of claims 1 to 4, wherein the second learned model is generated by machine learning a plurality of second teacher data defining a relationship between the plurality of first information of the linear circuit and the plurality of frequencies including 0 Hz and other than 0 Hz, and the plurality of S parameters calculated for each of the plurality of first information and the plurality of frequencies.

8. A step of obtaining first information regarding a linear circuit having a plurality of ports for inputting or outputting a high-frequency signal and second information regarding the frequency of the high-frequency signal; When the second information indicates that the frequency is 0 Hz, a step of estimating S parameters for two of the plurality of ports when the frequency is 0 Hz based on the first information and a first learned model; When the second information indicates that the frequency is other than 0 Hz, a step of estimating the S parameter at the frequency based on the first information, the second information, and a second learned model; including The first learned model is generated by machine learning a plurality of first teacher data defining a relationship between the plurality of first information of the linear circuit and the plurality of S parameters calculated with the frequency being 0 Hz for the plurality of first information, The method, wherein the second trained model is generated by machine learning a plurality of second teacher data defining a relationship between the plurality of first information of the linear circuit, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies. Claim 9 A trained model for estimating S-parameters for two ports among the plurality of ports based on first information regarding a linear circuit having a plurality of ports to which a high-frequency signal is input or output and second information regarding the frequency of the high-frequency signal, a first trained model generated by machine learning a plurality of first teacher data defining a relationship between the plurality of first information and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz; a second trained model generated by machine learning a plurality of second teacher data defining a relationship between the plurality of first information, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies; and a trained model including the same. Claim 10 A method for generating a trained model for estimating S-parameters for two ports among the plurality of ports based on first information regarding a linear circuit having a plurality of ports to which a high-frequency signal is input or output and second information regarding the frequency, generating a first trained model by machine learning a plurality of first teacher data defining a relationship between the plurality of first information and the plurality of S-parameters calculated for the plurality of first information with the frequency being 0 Hz; generating a second trained model by machine learning a plurality of second teacher data defining a relationship between the plurality of first information, the plurality of frequencies including other than 0 Hz, and the plurality of S-parameters calculated for each of the plurality of first information and the plurality of frequencies; and a method for generating a trained model including the same.

Images