W-band narrow-band filter design method based on SPP

Abstract

The invention discloses an SPP-based W-waveband narrow-band filter design method. The method comprises the following steps: S1, SPP structure selection; a specific structure is designed according to characteristics of a W wave band, propagation characteristics of the W wave band on metal and medium interfaces of different materials, a dispersion relation and an interaction mechanism with electromagnetic waves so as to realize excitation and control of the SPP. S2, loss compensation; s2, performing loss compensation on the SPP structure obtained through design in S1; s3, system integration; the SPP structure is integrated with other components in the system, so that the compatibility and performance of the SPP structure in the whole system are ensured. Through accurate structural design, loss compensation and system integration, not only is the performance of the filter improved, but also the adaptability and reliability of the filter in practical application are enhanced. The application of the method is expected to promote the development of related technologies, and provides a more efficient and stable filtering solution for a future communication system.

Description

Technical Field

[0001] The present invention relates to the technical field of communication and microwave engineering, and in particular to a W-band narrowband filter design method based on SPP. Background Art

[0002] In the field of communication and microwave engineering, with the rapid development of wireless communication technology, the demand for high-performance narrowband filters is growing. As part of the millimeter wave frequency band, the W band has a wide range of applications in radar, remote sensing, satellite communications, etc. due to its unique propagation characteristics and high frequency characteristics. However, the design of W-band filters faces the challenges of material loss, structural complexity, and precise control of electromagnetic wave propagation. Summary of the invention

[0003] The present invention aims to provide a W-band narrowband filter design method based on SPP to solve the above problems.

[0004] The present invention is achieved through the following technical solutions: A W-band narrowband filter design method based on SPP includes the following steps: S1: SPP structure selection; According to the characteristics of the W band, its propagation characteristics at the metal-dielectric interface of different materials, dispersion relations, and interaction mechanism with electromagnetic waves, a specific structure is designed to achieve the excitation and manipulation of SPP.

[0005] S2: loss compensation; Perform loss compensation on the SPP structure designed in S1; S3: System Integration; Integrate the SPP structure with other components in the system to ensure its compatibility and performance in the overall system.

[0006] The present invention not only improves the performance of the filter, but also enhances its adaptability and reliability in practical applications through precise structural design, loss compensation and system integration. The application of this method is expected to promote the development of related technologies and provide more efficient and stable filtering solutions for future communication systems.

[0007] Preferably, in S1, it specifically includes the following steps: S1.1: Data collection and preprocessing, collecting electromagnetic wave propagation data of different SPP structures in the W band, including resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor; S1.2: Model training: Use the data collected in S1.1 to train the deep learning model and predict the SPP behavior under different structural parameters through the deep learning model; S1.3: Structural parameter prediction, using the trained model to predict the optimal structural parameters for a given application target.

[0008] The present invention provides rich training data for the deep learning model by collecting electromagnetic wave propagation data of different SPP structures in the W band. This helps the model to learn and predict SPP behavior more accurately, thereby improving the accuracy and reliability of filter design. By continuously scrutinizing and refining data, the essence is formed, ensuring the hard core of the design process, that is, the content can withstand scrutiny, laying a solid foundation for subsequent steps; Preferably, in S1, the deep learning model is specifically a convolutional neural network model, wherein the different SPP structures in S1.1 specifically include a metal grating structure, an artificial surface plasmon structure, and a surface plasmon resonance mode hybrid structure.

[0009] The present invention uses a convolutional neural network to effectively learn the relationship between the structural parameters of the SPP and the electromagnetic wave propagation characteristics from a large amount of data. The innovation of this method lies in the ability to automatically extract features and make predictions, reducing the process of human trial and error and improving design efficiency and accuracy.

[0010] Preferably, the structural data of the metal grating structure includes: grating period, grating duty cycle, metal layer thickness, dielectric layer thickness, magnesium fluoride thickness, grating number, lattice constant, substrate height, width, height and length of metal nanorods, and groove width and depth; The structural data of the artificial surface plasmon structure include: the period of the metal grating, the width and depth of the metal strip, and the thickness of the metal layer; The structural data of the surface plasmon resonance mode hybrid structure include: the period of the nano-grating, the central wavelength of the photonic crystal, the width of the gold film, and the thickness of the gold film and the photonic crystal.

[0011] In the present invention, by collecting detailed simulation data of the structure data of metal grating structure, artificial surface plasmon structure, and surface plasmon resonance mode hybrid structure, rich feature information is provided for the model. The accuracy of these data directly affects the prediction effect of the model, thereby affecting the performance of the final filter. Preferably, in S1.1, the resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor of the metal grating structure, artificial surface plasmon structure and surface plasmon resonance mode hybrid structure are obtained through simulation and experiments to obtain their corresponding structural data.

[0012] The present invention obtains structural data through simulation and experiment, and combines theory with practice to more accurately verify and adjust the results predicted by the model. This method helps to improve the prediction accuracy of the model and ensure that the filter finally designed can meet the needs of practical applications.

[0013] Preferably, in the convolutional neural network model, the convolution layer extracts structural data corresponding to different SPP structures in the W band, the pooling layer identifies the propagation characteristics of the SPP at different scales, and identifies the changes in the resonant frequency and propagation constant of the SPP at different magnifications. The fully connected layer maps the input features to the final output, which specifically includes the input of the resonant wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor, and the output is the predicted SPP structure and its structural data.

[0014] In the design of the convolutional neural network model of the present invention, the SPP structure data can be effectively processed and analyzed through the collaborative work of the convolution layer, the pooling layer and the fully connected layer. This method can identify the SPP propagation characteristics at different scales and predict the optimal structural parameters, thereby improving the performance and efficiency of the filter design.

[0015] Preferably, in the convolutional neural network model, the loss function is specifically: in: is the total loss function; is the number of samples, which specifically refers to the number of samples of the metal grating structure, the artificial surface plasmon structure, and the surface plasmon resonance mode hybrid structure whose corresponding structural data are obtained through simulation and experiment; It is The true labels of samples are the structural data obtained through simulation and experiments; The model is The predicted output of each sample is the predicted SPP structure and its structural data; Represents the model's predicted probability for each category; is the weight coefficient, determined by cross-validation.

[0016] Preferably, It is calculated by the softmax function, and its function expression is: in , is the input vector, specifically the original score of each category output by the output layer of the convolutional neural network model, that is, the original output value of the predicted SPP structure and its structural data; K is the total number of categories, which is specifically the total number of structural data categories of the SPP structure; is the model parameter, which is optimized during the training process so that the predicted probability resolves the true label as much as possible; is the final output of the Softmax function, indicating that the model predicts that the input i-th sample belongs to the The probability of each category; The indexed scores of all categories are summed up. This summation operation is to convert the score of each category into a probability value so that the sum of the probabilities of all categories is 1. It is the index of the denominator, and its value ranges from 1 to K. For each element in the input vector, the Softmax function calculates its exponential, where f For the This step converts the raw scores into non-negative numbers in preparation for the next step of normalization.

[0017] The loss function and softmax function of the present invention provide an effective optimization target and prediction probability calculation method for the model. By adjusting the weight coefficient and optimizing the model parameters, the prediction accuracy of the model can be improved, thereby ensuring the accuracy of the filter design.

[0018] Preferably, the convolutional neural network model adjusts and optimizes the model according to the effect feedback of the model's prediction results until the model's prediction accuracy meets the requirements.

[0019] Preferably, in S2, the loss compensation is specifically to compensate for the precision loss caused by the structural error caused by manufacturing through experimental adjustment.

[0020] The present invention adjusts and optimizes the design through the actual manufacturing and testing process, and can compensate for the influence caused by the structural error. This method helps to improve the actual performance of the filter and ensure its stability and reliability in a real environment.

[0021] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. The present invention not only improves the performance of the filter, but also enhances its adaptability and reliability in practical applications through precise structural design, loss compensation and system integration. The application of this method is expected to promote the development of related technologies and provide more efficient and stable filtering solutions for future communication systems.

[0022] 2. The present invention provides rich training data for the deep learning model by collecting electromagnetic wave propagation data of different SPP structures in the W band. This helps the model to learn and predict SPP behavior more accurately, thereby improving the accuracy and reliability of filter design. By constantly scrutinizing and refining data, the essence is formed, ensuring the hard core of the design process, that is, the content can withstand scrutiny, laying a solid foundation for subsequent steps; 3. The present invention uses a convolutional neural network to effectively learn the relationship between the structural parameters of the SPP and the electromagnetic wave propagation characteristics from a large amount of data. The innovation of this method lies in the ability to automatically extract features and make predictions, reducing the process of human trial and error and improving design efficiency and accuracy.

[0023] 4. In the present invention, by collecting detailed simulation data of the structure data of metal grating structure, artificial surface plasmon structure, and surface plasmon resonance mode hybrid structure, rich feature information is provided for the model. The accuracy of these data directly affects the prediction effect of the model, thereby affecting the performance of the final filter. 5. The present invention uses a method of obtaining structural data through simulation and experiment, combining theory with practice, to more accurately verify and adjust the results of model prediction. This method helps to improve the prediction accuracy of the model and ensure that the final designed filter can meet the needs of practical applications.

[0024] 6. In the design of the convolutional neural network model of the present invention, the SPP structure data can be effectively processed and analyzed through the collaborative work of the convolution layer, the pooling layer and the fully connected layer. This method can identify the SPP propagation characteristics at different scales and predict the optimal structural parameters, thereby improving the performance and efficiency of the filter design.

[0025] 7. The present invention adjusts and optimizes the design through the actual manufacturing and testing process, which can compensate for the impact of structural errors. This method helps to improve the actual performance of the filter and ensure its stability and reliability in the real environment; DETAILED DESCRIPTION

[0026] The detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of the present invention. It should be noted that relational terms such as the terms "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0027] Moreover, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a list of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus. In the absence of more restrictions, an element defined by the phrase "comprises a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0028] Example This embodiment includes: a W-band narrowband filter design method based on SPP, comprising the following steps: S1: SPP structure selection; According to the characteristics of the W band, its propagation characteristics at the metal-dielectric interface of different materials, dispersion relations, and interaction mechanism with electromagnetic waves, a specific structure is designed to achieve the excitation and manipulation of SPP.

[0029] SPP, also known as surface plasmon polaritons, is an electromagnetic wave mode formed at the interface between metal and dielectric. The W band is part of the millimeter wave band, and its frequency range is approximately between 75 GHz and 110 GHz. The high frequency of the W band provides a higher data transmission rate and is suitable for high-speed communication systems.

[0030] S2: loss compensation; Perform loss compensation on the SPP structure designed in S1; S3: System Integration; Integrate the SPP structure with other components in the system to ensure its compatibility and performance in the overall system.

[0031] The present invention not only improves the performance of the filter through precise structural design, loss compensation and system integration, but also enhances its adaptability and reliability in practical applications. The application of this method is expected to promote the development of related technologies and provide more efficient and stable filtering solutions for future communication systems. Preferably, in S1, it specifically includes the following steps: S1.1: Data collection and preprocessing, collecting electromagnetic wave propagation data of different SPP structures in the W band, including resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor; S1.2: Model training: Use the data collected in S1.1 to train the deep learning model and predict the SPP behavior under different structural parameters through the deep learning model; S1.3: Structural parameter prediction, using the trained model to predict the optimal structural parameters for a given application target.

[0032] The present invention provides rich training data for the deep learning model by collecting electromagnetic wave propagation data of different SPP structures in the W band. This helps the model to learn and predict SPP behavior more accurately, thereby improving the accuracy and reliability of filter design. By continuously scrutinizing and refining data, the essence is formed, ensuring the hard core of the design process, that is, the content can withstand scrutiny, laying a solid foundation for subsequent steps; Preferably, in S1, the deep learning model is specifically a convolutional neural network model, wherein the different SPP structures in S1.1 specifically include a metal grating structure, an artificial surface plasmon structure, and a surface plasmon resonance mode hybrid structure.

[0033] The present invention uses a convolutional neural network to effectively learn the relationship between the structural parameters of the SPP and the electromagnetic wave propagation characteristics from a large amount of data. The innovation of this method lies in the ability to automatically extract features and make predictions, reducing the process of human trial and error and improving design efficiency and accuracy.

[0034] Preferably, the structural data of the metal grating structure includes: grating period, grating duty cycle, metal layer thickness, dielectric layer thickness, magnesium fluoride thickness, grating number, lattice constant, substrate height, width, height and length of metal nanorods, and groove width and depth; The structural data of the artificial surface plasmon structure include: the period of the metal grating, the width and depth of the metal strip, and the thickness of the metal layer; The structural data of the surface plasmon resonance mode hybrid structure include: the period of the nano-grating, the central wavelength of the photonic crystal, the width of the gold film, and the thickness of the gold film and the photonic crystal.

[0035] In the present invention, by collecting detailed simulation data of the structure data of metal grating structure, artificial surface plasmon structure, and surface plasmon resonance mode hybrid structure, rich feature information is provided for the model. The accuracy of these data directly affects the prediction effect of the model, thereby affecting the performance of the final filter. Preferably, in S1.1, the resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor of the metal grating structure, artificial surface plasmon structure and surface plasmon resonance mode hybrid structure are obtained through simulation and experiments to obtain their corresponding structural data.

[0036] The present invention obtains structural data through simulation and experiment, and combines theory with practice to more accurately verify and adjust the results predicted by the model. This method helps to improve the prediction accuracy of the model and ensure that the filter finally designed can meet the needs of practical applications.

[0037] Preferably, in the convolutional neural network model, the convolution layer extracts structural data corresponding to different SPP structures in the W band, the pooling layer identifies the propagation characteristics of the SPP at different scales, and identifies the changes in the resonant frequency and propagation constant of the SPP at different magnifications. The fully connected layer maps the input features to the final output, which specifically includes the input of the resonant wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor, and the output is the predicted SPP structure and its structural data.

[0038] In the design of the convolutional neural network model of the present invention, the SPP structure data can be effectively processed and analyzed through the collaborative work of the convolution layer, the pooling layer and the fully connected layer. This method can identify the SPP propagation characteristics at different scales and predict the optimal structural parameters, thereby improving the performance and efficiency of the filter design.

[0039] Preferably, in the convolutional neural network model, the loss function is specifically: in: is the total loss function; is the number of samples, which specifically refers to the number of samples of the metal grating structure, the artificial surface plasmon structure, and the surface plasmon resonance mode hybrid structure whose corresponding structural data are obtained through simulation and experiment; It is The true labels of samples are the structural data obtained through simulation and experiments; The model is The predicted output of each sample is the predicted SPP structure and its structural data; Represents the model's predicted probability for each category; is the weight coefficient, determined by cross-validation.

[0040] The loss function of the present invention can handle classification and regression tasks at the same time, and improves the generalization ability of the model, so that the model can learn to distinguish the decision boundaries of different categories, and can also reduce the overfitting phenomenon of the model.

[0041] Preferably, It is calculated by the softmax function, and its function expression is: in , is the input vector, specifically the original score of each category output by the output layer of the convolutional neural network model, that is, the original output value of the predicted SPP structure and its structural data; K is the total number of categories, which is specifically the total number of structural data categories of the SPP structure; is the model parameter, which is optimized during the training process to make the predicted probability as close to the true label as possible; is the final output of the Softmax function, indicating that the model predicts that the input i-th sample belongs to the The probability of each category; The indexed scores of all categories are summed up. This summation operation is to convert the score of each category into a probability value so that the sum of the probabilities of all categories is 1. It is the index of the denominator, and its value ranges from 1 to K. For each element in the input vector, the Softmax function calculates its exponential, where f For the This step converts the raw scores into non-negative numbers in preparation for the next step of normalization.

[0042] The loss function and softmax function of the present invention provide an effective optimization target and prediction probability calculation method for the model. By adjusting the weight coefficient and optimizing the model parameters, the prediction accuracy of the model can be improved, thereby ensuring the accuracy of the filter design.

[0043] Preferably, the convolutional neural network model adjusts and optimizes the model according to the effect feedback of the model's prediction results until the model's prediction accuracy meets the requirements.

[0044] Preferably, in S2, the loss compensation is adjusted through experiments to compensate for the precision loss caused by the structural error caused by manufacturing, and the specific steps include: S2.1: Performance test: The fabricated SPP structure is tested to obtain the performance parameters of resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor; S2.2: Data analysis; S2.2.1: Dimensional data analysis; The parameters that do not meet the requirements are analyzed, and the performance parameters of resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor measured by S2.1 are reversely input into the deep learning model through the deep learning model of S1, and the size parameters of the corresponding SPP structure are output; Compare the obtained dimensional parameters with the dimensional parameters of the manufactured SPP structure. If the dimensional parameters are the same, the losses are caused by the dimensional errors during manufacturing, and compensation is achieved by optimizing the production accuracy. If the dimensional parameters are different but within a reasonable range, other loss reasons are considered; S2.2.2: Material data analysis; When the loss caused entirely by size issues is excluded, material data analysis is performed, including experimental analysis of the material's ohmic loss, absorption loss, and material lattice defects or impurities; S2.3: Prepare a design adjustment report. According to the data analysis steps in S2.2, obtain a loss cause analysis report of the manufactured SPP structure, including dimensional data analysis and material data analysis.

[0045] S2.4: Iterative optimization; Repeat steps S2.1-S2.3 for the compensated SPP structure. If the loss compensation is not ideal, perform multiple adjustments and tests until satisfactory performance is achieved.

[0046] The present invention adjusts and optimizes the design through the actual manufacturing and testing process, and can compensate for the influence caused by the structural error. This method helps to improve the actual performance of the filter and ensure its stability and reliability in a real environment.

[0047] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A W-band narrowband filter design method based on SPP, characterized in that: The following steps are involved: S1: SPP structure selection; According to the characteristics of the W band, its propagation characteristics at the metal-dielectric interface of different materials, dispersion relations, and interaction mechanisms with electromagnetic waves, a specific structure is designed to achieve the excitation and manipulation of SPPs; S2: loss compensation; Perform loss compensation on the SPP structure designed in S1; S3: System Integration; Integrate the SPP structure with other components in the system.

2. The method for designing a W-band narrowband filter based on SPP according to claim 1, characterized in that: In S1, it specifically includes the following steps: S1.1: Data collection and preprocessing, collecting electromagnetic wave propagation data of different SPP structures in the W band, including resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor; S1.2: Model training: Use the data collected in S1.1 to train the deep learning model and predict the SPP behavior under different structural parameters through the deep learning model; S1.3: Structural parameter prediction, using the trained model to predict the optimal structural parameters for a given application target.

3. The method for designing a W-band narrowband filter based on SPP according to claim 2, characterized in that: In S1, the deep learning model is specifically a convolutional neural network model, wherein the different SPP structures in S1.1 specifically include a metal grating structure, an artificial surface plasmon structure, and a surface plasmon resonance mode hybrid structure.

4. The method for designing a W-band narrowband filter based on SPP according to claim 3, characterized in that: The structural data of the metal grating structure include: grating period, grating duty cycle, metal layer thickness, dielectric layer thickness, magnesium fluoride thickness, grating number, lattice constant, substrate height, width, height and length of metal nanorods, and groove width and depth; The structural data of the artificial surface plasmon structure include: the period of the metal grating, the width and depth of the metal strip, and the thickness of the metal layer; The structural data of the surface plasmon resonance mode hybrid structure include: the period of the nano-grating, the central wavelength of the photonic crystal, the width of the gold film, and the thickness of the gold film and the photonic crystal.

5. The method for designing a W-band narrowband filter based on SPP according to claim 4, characterized in that: In S1.1, the resonance wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor of the metal grating structure, artificial surface plasmon structure and surface plasmon resonance mode hybrid structure are all obtained through simulation and experiments.

6. The SPP-based W-band narrowband filter design method according to claim 5, characterized in that: In the convolutional neural network model, the convolution layer extracts the structural data corresponding to different SPP structures in the W band, the pooling layer identifies the propagation characteristics of SPP at different scales, and identifies the changes in the resonant frequency and propagation constant of SPP at different magnifications. The fully connected layer maps the input features to the final output, which specifically includes the input of resonant wavelength, quality factor, sensitivity, propagation length, field intensity enhancement, slow light effect, coupling efficiency, propagation constant and loss factor, and the output is the predicted SPP structure and its structural data.

7. The method for designing a W-band narrowband filter based on SPP according to claim 6, characterized in that: In the convolutional neural network model, the loss function is specifically: in: is the total loss function; is the number of samples, which specifically refers to the number of samples of the metal grating structure, the artificial surface plasmon structure, and the surface plasmon resonance mode hybrid structure whose corresponding structural data are obtained through simulation and experiment; It is The true labels of samples are the structural data obtained through simulation and experiments; The model is The predicted output of each sample is the predicted SPP structure and its structural data; Represents the model's predicted probability for each category; is the weight coefficient, determined by cross-validation.

8. The method for designing a W-band narrowband filter based on SPP according to claim 7, characterized in that: It is calculated by the softmax function, and its function expression is: in , is the input vector, specifically the original score of each category output by the output layer of the convolutional neural network model, that is, the original output value of the predicted SPP structure and its structural data; K is the total number of categories, which is specifically the total number of structural data categories of the SPP structure; are model parameters, which are optimized during the training process; is the final output of the Softmax function, indicating that the model predicts that the input i-th sample belongs to the The probability of each category; is the sum of the indexed raw scores of all categories; It is the index of the denominator, and its value ranges from 1 to K. For each element in the input vector, the Softmax function calculates its exponent, converting the original score into a non-negative number, where f For the The raw scores of the categories.

9. The method for designing a W-band narrowband filter based on SPP according to claim 8, characterized in that: The convolutional neural network model adjusts and optimizes the model according to the feedback of the model's prediction results until the model's prediction accuracy meets the requirements.

10. The SPP-based W-band narrowband filter design method according to claim 9, characterized in that: In S2, loss compensation is specifically to compensate for the precision loss caused by structural errors caused by manufacturing through experimental adjustment.

Images