Stripline coupler optimization method, system and coupler based on neural network

Abstract

The present invention proposes a stripline coupler optimization method, system, and coupler based on a neural network, belonging to the field of artificial intelligence technology. The method comprises: 1) obtaining a design coupling degree C of the stripline coupler; 2) inputting the design coupling degree C into a neural network, and outputting structural parameters of the stripline coupler through learning of the neural network; 3) determining a stripline coupler model based on the structural parameters of the stripline coupler, performing an EM simulation on the stripline coupler model, and obtaining simulation results of the stripline coupler; 4) judging whether the stripline coupler meets design requirements based on the simulation results of the stripline coupler; if not, using an Adam optimizer to reversely update the neural network according to a loss function, repeating steps 2) and 3) until the stripline coupler meets the design requirements, and terminating the optimization process; if it meets the requirements, then terminating the optimization process.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a stripline coupler optimization method, system and coupler based on a neural network. Background Art

[0002] Ground-penetrating radar (GPR) is a nondestructive underground target detection technology capable of accurately detecting buried objects. It is widely used in pipeline detection, roadbed inspection, resource exploration, and other fields. Based on the principles of a vector network analyzer (VNA), swept-frequency GPR offers the advantages of wide bandwidth, signal reconfigurability, high sensitivity, and the ability to balance shallow-layer resolution with deep-layer detection depth. Couplers are key components of swept-frequency GPR, and wideband coupler designs can enhance GPR performance in practical applications.

[0003] During the structural design process of the coupler, it is necessary to design the model structure and perform simulation verification in EM software. The simulation verification process often requires a long design time. When the model structure is complex, a very long design cycle is required. Summary of the Invention

[0004] In order to solve the problem that the model structure design of the coupler mentioned above requires a long design time, the present invention proposes a stripline coupler optimization method, system and coupler based on a neural network.

[0005] The present invention introduces an optimization algorithm into the coupler, shortening the coupler design time while designing a more diverse and complex model structure that meets the design requirements. The specific technical solution is as follows:

[0006] The stripline coupler optimization method based on neural network includes the following steps:

[0007] 1) Obtain the design coupling degree C of the stripline coupler;

[0008] 2) Input the designed coupling degree C into the neural network, and output the structural parameters of the stripline coupler through the learning of the neural network;

[0009] 3) determining a stripline coupler model according to the structural parameters of the stripline coupler, performing EM simulation on the stripline coupler model, and obtaining simulation results of the stripline coupler;

[0010] 4) Determine whether the stripline coupler meets the design requirements based on the simulation results of the stripline coupler;

[0011] If not, the Adam optimizer is used to reversely update the neural network according to the loss function L, and steps 2) and 3) are repeated until the stripline coupler meets the design requirements, and the optimization process ends;

[0012] If satisfied, the optimization process ends.

[0013] It is further defined that the structural parameter of the stripline coupler is the vertical coordinate value of the stripline.

[0014] It is further defined that the neural network includes a plurality of base layers connected in sequence, and an output layer arranged at the output end of the base layer, and the base layer includes a fully connected layer and an activation layer connected in sequence.

[0015] It is further defined that the learning calculation method of the neural network is:

[0016]

[0017] represents the jth neuron in the n+1th base layer, f represents the Sigmoid activation function, N represents the number of neurons in the nth base layer, represents the i-th neuron in the n-th base layer, j represents the number of neurons in the n+1-th base layer, i represents the number of neurons in the n-th base layer, n represents the number of base layers, w represents the neuron weight of the n-th base layer, and b represents the neuron bias term of the n-th activation layer.

[0018] It is further defined that the basic layer has three layers, the number of neurons in the first basic layer is 81, the number of neurons in the second basic layer is 128, the number of neurons in the third basic layer is 64, and the number of neurons in the output layer is 8.

[0019] Further defined, the calculation formula of the loss function L is:

[0020] L=MSE(CEM,C)+MSE(DEM,D)

[0021] C represents the design coupling degree, D represents the design directivity coefficient, CEM represents the coupling degree obtained by EM simulation, DEM represents the directivity coefficient obtained by EM simulation, MSE(CEM,C) represents the mean square error between the design coupling degree and the design directivity coefficient, and MSE(DEM,D) represents the mean square error between the simulated coupling degree and the simulated directivity coefficient.

[0022] The stripline coupler optimization system based on neural network includes coupling degree acquisition module, neural network optimization module, EM simulation module and result judgment module.

[0023] The coupling degree acquisition module is used to obtain the design coupling degree C of the stripline coupler;

[0024] The neural network optimization module is used to input the design coupling degree C into the neural network, and output the structural parameters of the stripline coupler through the learning of the neural network;

[0025] The EM simulation module is used to determine a stripline coupler model according to the structural parameters of the stripline coupler, perform EM simulation on the stripline coupler model, and obtain a simulation result of the stripline coupler;

[0026] The result judgment module is used to judge whether the stripline coupler meets the design requirements according to the simulation results of the stripline coupler;

[0027] If not, the Adam optimizer is used to update the neural network in reverse according to the loss function L until the stripline coupler meets the design requirements, and the optimization process ends;

[0028] If satisfied, the optimization process ends.

[0029] A stripline coupler is formed using the above-mentioned neural network-based stripline coupler optimization method. The stripline coupler includes a dielectric plate and multiple striplines. The multiple striplines are arranged in parallel on a surface of the dielectric plate. Both ends of each stripline are waveguide ports, and the waveguide ports extend to the edge of the dielectric plate.

[0030] It is further defined that the waveguide port of each stripline is extended to the edge of the dielectric plate through a connecting line.

[0031] A computer-readable storage medium stores a program file, wherein the program file is executed to implement the above-mentioned stripline coupler optimization method based on neural network.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention discloses a neural network-based stripline coupler optimization method. The method uses a neural network to learn the design coupling degree C of the stripline coupler. Simultaneously, the structural parameters of the stripline coupler are simulated using an EM simulation process, thereby facilitating judgment of the simulation results of the stripline coupler until the stripline coupler meets the requirements. The present invention introduces the optimization algorithm of the neural network and EM simulation into the structural design of the coupler. While shortening the coupler design time, the present invention designs a more diversified and complex stripline coupler model structure that meets the design requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of the process of the stripline coupler optimization method based on a neural network in this application;

[0035] Figure 2 A schematic diagram of a stripline coupler optimization system based on a neural network according to the present application;

[0036] Figure 3This is a schematic structural diagram of the stripline coupler of the present application;

[0037] Figure 4 Schematic diagram of the structure of the first strip line;

[0038] Figure 5 Schematic diagram of the simulation results of the return loss, coupling and isolation of the stripline coupler;

[0039] Figure 6 Schematic diagram of simulation results of the directivity coefficient of the stripline coupler;

[0040] Among them, 101 is a first stripline, 102 is a second stripline, 201 is a first waveguide port, 202 is a second waveguide port, 203 is a third waveguide port, 204 is a fourth waveguide port, 301 is a first connecting line, 302 is a second connecting line, 303 is a third connecting line, 304 is a fourth connecting line, and 4 is a dielectric board. DETAILED DESCRIPTION

[0041] The technical solution of the invention is further explained below with reference to the accompanying drawings and embodiments, but the present invention is not limited to the embodiments described below.

[0042] Example 1

[0043] See also Figure 1 The stripline coupler optimization method based on a neural network in this embodiment includes the following steps:

[0044] 1) Obtaining a design coupling degree C of the stripline coupler; specifically, the design coupling degree C of the stripline coupler here is obtained according to design requirements;

[0045] 2) Input the designed coupling degree C into the neural network, and output the structural parameters of the stripline coupler through learning of the neural network; Specifically, the horizontal coordinate value of the stripline structure in this embodiment is fixed and does not need to be optimized. In this step, only the vertical coordinate value of the stripline structure needs to be optimized and learned through the designed coupling degree C, that is, the structural parameters of the stripline coupler in this step are the vertical coordinate values ​​of the stripline; see Figure 4 Taking a stripline as an example, the length L0 of the first stripline 101 is 122.7 mm and the thickness is 0.1 mm. The geometric structure on the XOY plane is composed of a broken line connecting eight points: A1(x1, y1), A2(x2, y2), A3(x3, y3), A4(x4, y4), A5(x5, y5), A6(x6, y6), A7(x7, y7), and A8(x8, y8). x1 = x8 = 0, x2 = x7 = L0 / 3, x3 = x6 = 2*L0 / 3, x4 = x5 = L0, and y1, y2, ..., y8 are obtained using a neural network-based optimization method.

[0046] 3) determining a stripline coupler model according to the structural parameters of the stripline coupler, performing EM simulation on the stripline coupler model, and obtaining simulation results of the stripline coupler;

[0047] 4) Determine whether the stripline coupler meets the design requirements based on the simulation results of the stripline coupler;

[0048] If not, the Adam optimizer is used to reversely update the neural network according to the loss function L, and steps 2) and 3) are repeated until the stripline coupler meets the design requirements, and the optimization process ends;

[0049] If satisfied, the optimization process ends.

[0050] The calculation formula of the loss function L is:

[0051] L=MSE(CEM,C)+MSE(DEM,D)

[0052] C represents the design coupling degree, which is the same as the coupling degree C in step 2), D represents the design directivity coefficient, CEM represents the coupling degree obtained by EM simulation, DEM represents the directivity coefficient obtained by EM simulation, MSE(CEM,C) represents the mean square error of the design coupling degree and the design directivity coefficient, and MSE(DEM,D) represents the mean square error of the simulated coupling degree and the simulated directivity coefficient.

[0053] The coupler design requirements for this embodiment are: an operating frequency band of 0.05 to 4 GHz, a coupling degree of -20 dB ± 0.5 dB, a return loss of less than 20 dB, and a directivity coefficient of more than 17 dB. Using the aforementioned optimization method, this embodiment optimizes the coupler structure based on a neural network, completing the optimization within 50 iterations and obtaining the coupler structural parameters that meet the design requirements: y1 = 1.5 mm, y2 = 2.1 mm, y3 = 2.7 mm, y4 = 3.1 mm, y5 = 1.4 mm, y6 = 1.2 mm, y7 = 0.7 mm, and y8 = 0.2 mm. The design of the stripline coupler is complete.

[0054] The neural network includes multiple base layers connected in sequence, and an output layer set at the output end of the base layer. The base layer includes a fully connected layer and an activation layer connected in sequence. The activation layers of two adjacent base layers are connected to the fully connected layer, that is, the output value of the activation layer corresponding to the previous base layer is the input value of the fully connected layer corresponding to the next base layer.

[0055] The learning calculation method of the neural network is:

[0056]

[0057] represents the jth neuron in the n+1th base layer, f represents the Sigmoid activation function, N represents the number of neurons in the nth base layer, represents the i-th neuron in the n-th base layer, j represents the number of neurons in the n+1-th base layer, i represents the number of neurons in the n-th base layer, n represents the number of base layers, w represents the neuron weight of the n-th base layer, and b represents the neuron bias term of the n-th activation layer.

[0058] Preferably, the basic layer has three layers, the number of neurons in the first basic layer is 81, the number of neurons in the second basic layer is 128, the number of neurons in the third basic layer is 64, and the number of neurons in the output layer is 8.

[0059] Example 2

[0060] See also Figure 2 The stripline coupler optimization system based on a neural network in this embodiment is formed on the basis of the stripline coupler optimization method based on a neural network in Example 1, and includes a coupling degree acquisition module, a neural network optimization module, an EM simulation module, and a result judgment module.

[0061] Coupling degree acquisition module: used to obtain the design coupling degree C of the stripline coupler; specifically, the design coupling degree C of the stripline coupler here is obtained according to the design requirements;

[0062] Neural network optimization module: used to input the designed coupling degree C into the neural network, and output the structural parameters of the stripline coupler through learning of the neural network; specifically, the abscissa value of the stripline structure in this embodiment is fixed and does not require optimization design. In this step, only the ordinate value of the stripline structure needs to be optimized based on the designed coupling degree C, that is, the stripline structural parameter in this step is the ordinate value of the stripline;

[0063] EM simulation module: used to determine the stripline coupler model according to the structural parameters of the stripline coupler, perform EM simulation on the stripline coupler model, and obtain the simulation results of the stripline coupler;

[0064] Result judgment module: used to judge whether the stripline coupler meets the design requirements according to the simulation results of the stripline coupler;

[0065] If not, the Adam optimizer is used to reversely update the neural network according to the loss function L, and steps 2) and 3) are repeated until the stripline coupler meets the design requirements, and the optimization process ends;

[0066] If satisfied, the optimization process ends.

[0067] The calculation formula of the loss function L is:

[0068] L=MSE(CEM,C)+MSE(DEM,D)

[0069] C represents the design coupling degree, which is the same as the coupling degree C in step 2), D represents the design directivity coefficient, CEM represents the coupling degree obtained by EM simulation, DEM represents the directivity coefficient obtained by EM simulation, MSE(CEM,C) represents the mean square error of the design coupling degree and the design directivity coefficient, and MSE(DEM,D) represents the mean square error of the simulated coupling degree and the simulated directivity coefficient.

[0070] Example 3

[0071] This embodiment provides a stripline coupler formed using the neural network-based stripline coupler optimization method of Example 1. The stripline coupler includes a dielectric plate and multiple striplines. The multiple striplines are arranged in parallel on a surface of the dielectric plate. Both ends of each stripline are waveguide ports, and the waveguide ports extend to the edge of the dielectric plate.

[0072] The waveguide port of each stripline is extended to the edge of the dielectric plate through a connecting line.

[0073] See also Figure 3 This embodiment is described using two strip lines as an example. It should be noted that in addition to two strip lines, the present application can also use three, four, five, etc. strip lines, depending on the design requirements of the strip line coupler.

[0074] The two strip lines are a first strip line 101 and a second strip line 102, wherein the first strip line 101 and the second strip line 102 are arranged side by side on the upper surface of the dielectric plate 4 along the length direction of the dielectric plate 4, a first waveguide port 201 is provided at one end of the first strip line 101, a second waveguide port 202 is provided at the other end of the first strip line 101, a third waveguide port 203 is provided at one end of the second strip line 102, and a fourth waveguide port 204 is provided at the other end of the second strip line 102, a first connecting line 301 is provided between one end of the first strip line 101 and the first waveguide port 201, the first waveguide port 201 extends to the side edge of the dielectric plate 4 through the first connecting line 301, and the second connecting line 302 is provided between the other end of the first strip line 101 and the second waveguide port 203. 02, the second waveguide port 202 extends to the side edge of the dielectric plate 4 through the second connecting line 302, wherein the first waveguide port 201 and the second waveguide port 202 are respectively arranged on opposite sides of the dielectric plate 4; the third connecting line 303 is arranged between one end of the second stripline 102 and the third waveguide port 203, and the third waveguide port 203 extends to the side edge of the dielectric plate 4 through the third connecting line 303; the fourth connecting line 304 is arranged between the other end of the second stripline 102 and the fourth waveguide port 204, and the fourth waveguide port 204 extends to the side edge of the dielectric plate 4 through the fourth connecting line 304, wherein the third waveguide port 203 and the fourth waveguide port 204 are arranged on the same side of the dielectric plate 4, and are both adjacent to the first waveguide port 201 and the second waveguide port 202.

[0075] Preferably, the second stripline 102 of this embodiment is symmetrical with the first stripline 101 along the Y axis, the first waveguide port 201 and the second waveguide port 202 are respectively perpendicular to the first stripline 101, the third waveguide port 203 and the fourth waveguide port 204 are respectively perpendicular to the second stripline 102, the width of the first waveguide port 201, the width of the second waveguide port 202, the width of the third waveguide port 203 and the width of the fourth waveguide port 204 are all 7 mm, the first connecting line 301, the second connecting line 302, the third connecting line 303 and the fourth connecting line 304 are all 50-ohm strip lines, the width of the first connecting line 301, the width of the second connecting line 302, the width of the third connecting line 303 and the width of the fourth connecting line 304 are all 1.68 mm, the length of the first connecting line 301 and the length of the second connecting line 302 are both 7 mm, the length of the third connecting line 303 and the length of the fourth connecting line 304 are both 8 mm; the dielectric constant of the dielectric plate 4 is 2.2, the length of the dielectric plate 4 is 136.7 mm, the width of the dielectric plate 4 is 22 mm, and the height of the dielectric plate 4 is 2.032 mm.

[0076] Figure 5This is a schematic diagram of the simulation results of the return loss, coupling degree and isolation of the stripline coupler in this embodiment. It can be seen from the figure that the stripline coupler in this embodiment has a return loss of less than 20dB, a coupling degree of about -20dB, an in-band ripple of less than 0.5dB, and a directivity coefficient greater than 17dB within the operating frequency band of 0.05 to 4GHz, which meets the design requirements of the swept-frequency ground penetrating radar system.

[0077] Figure 6 Schematic diagram of the simulation results of the directivity coefficient of the stripline coupler in this embodiment. It can be seen from the figure that the directivity coefficient of the stripline coupler in this embodiment is greater than 17dB within the working frequency band of 0.05 to 4GHz, which meets the design requirements of the swept-frequency ground penetrating radar system.

[0078] Example 4

[0079] This embodiment provides a computer-readable storage medium storing a program file. The program file is executed to implement the stripline coupler optimization method based on a neural network according to the first embodiment.

Claims

1. A stripline coupler optimization method based on neural network, characterized in that: The following steps are involved: 1) Obtain the design coupling degree C of the stripline coupler; 2) Inputting the designed coupling degree C into the neural network, and outputting the structural parameters of the stripline coupler through learning of the neural network, wherein the structural parameters of the stripline coupler are the ordinate values ​​of the stripline; 3) determining a stripline coupler model according to the structural parameters of the stripline coupler, performing EM simulation on the stripline coupler model, and obtaining simulation results of the stripline coupler; 4) Determine whether the stripline coupler meets the design requirements based on the simulation results of the stripline coupler; If not, the Adam optimizer is used to reversely update the neural network according to the loss function L, and steps 2) and 3) are repeated until the stripline coupler meets the design requirements, and the optimization process is terminated. The calculation formula of the loss function L is: L=MSE(CEM,C)+MSE(DEM,D) C represents the design coupling degree, D represents the design directivity coefficient, CEM represents the coupling degree obtained by EM simulation, DEM represents the directivity coefficient obtained by EM simulation, MSE(CEM,C) represents the mean square error between the design coupling degree and the design directivity coefficient, and MSE(DEM,D) represents the mean square error between the simulated coupling degree and the simulated directivity coefficient; If satisfied, the optimization process ends.

2. The stripline coupler optimization method based on neural network according to claim 1, characterized in that: The neural network includes a plurality of base layers connected in sequence, and an output layer arranged at an output end of the base layer, wherein the base layer includes a fully connected layer and an activation layer connected in sequence.

3. The stripline coupler optimization method based on neural network according to claim 2, characterized in that: The learning calculation method of the neural network is: represents the jth neuron in the n+1th base layer, f represents the Sigmoid activation function, N represents the number of neurons in the nth base layer, represents the i-th neuron in the n-th base layer, j represents the number of neurons in the n+1-th base layer, i represents the number of neurons in the n-th base layer, n represents the number of base layers, w represents the neuron weight of the n-th base layer, and b represents the neuron bias term of the n-th activation layer.

4. The stripline coupler optimization method based on neural network according to claim 2, wherein: The basic layer has three layers, the number of neurons in the first basic layer is 81, the number of neurons in the second basic layer is 128, the number of neurons in the third basic layer is 64, and the number of neurons in the output layer is 8.

5. A stripline coupler optimization system based on neural network, characterized in that: Including coupling degree acquisition module, neural network optimization module, EM simulation module and result judgment module, The coupling degree acquisition module is used to obtain the design coupling degree C of the stripline coupler; The neural network optimization module is used to input the design coupling degree C into the neural network, and output the structural parameters of the stripline coupler through learning of the neural network, where the structural parameters of the stripline coupler are the vertical coordinate values ​​of the stripline; The EM simulation module is used to determine a stripline coupler model according to the structural parameters of the stripline coupler, perform EM simulation on the stripline coupler model, and obtain a simulation result of the stripline coupler; The result judgment module is used to judge whether the stripline coupler meets the design requirements according to the simulation results of the stripline coupler; If not, the Adam optimizer is used to reversely update the neural network according to the loss function L until the stripline coupler meets the design requirements, and the optimization process ends. The calculation formula of the loss function L is: L=MSE(CEM,C)+MSE(DEM,D) C represents the design coupling degree, D represents the design directivity coefficient, CEM represents the coupling degree obtained by EM simulation, DEM represents the directivity coefficient obtained by EM simulation, MSE(CEM,C) represents the mean square error between the design coupling degree and the design directivity coefficient, and MSE(DEM,D) represents the mean square error between the simulated coupling degree and the simulated directivity coefficient; If satisfied, the optimization process ends.

6. A stripline coupler, characterized in that: The stripline coupler is formed using the neural network-based stripline coupler optimization method described in any one of claims 1 to 4. The stripline coupler includes a dielectric plate and multiple striplines, wherein the multiple striplines are arranged in parallel on a surface of the dielectric plate, and both ends of each stripline are waveguide ports, and the waveguide ports extend to the edge of the dielectric plate.

7. The stripline coupler according to claim 6, wherein The waveguide port of each strip line is extended to the edge of the dielectric plate through a connecting line.

8. A computer-readable storage medium, characterized in that A program file is stored, and the program file is executed to implement the stripline coupler optimization method based on neural network according to any one of claims 1 to 4.

Images