Data optimization method, data optimization system and computing device cluster

Abstract

The invention discloses a data optimization method and system and a computing device cluster. After the data optimization system receives the physical attribute parameters of the passive device, feature extraction is carried out on the physical attribute parameters of the passive device, and poles, residuals and constant terms forming a scattering parameter frequency response curve of the passive device are obtained. The data optimization system calculates a scattering parameter frequency response curve of the passive device according to different frequencies, poles, residuals and constant terms in a set frequency range, performs passivity correction on the scattering parameter frequency response curve of the passive device, filters abnormal values in the scattering parameter frequency response curve of the passive device, or corrects the abnormal values, and finally performs passive correction on the scattering parameter frequency response curve of the passive device. And the curve is smoother. The scattering parameter of the scattering parameter frequency response curve of the passive device output by the data optimization system meets the physical property constraint. A designer selects the most suitable passive device for the radio frequency circuit based on the performance parameters of the passive device meeting the physical property constraint, so that the performance of the designed and developed radio frequency circuit is optimal.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a data optimization method, a data optimization system, and a computing device cluster. Background Art

[0002] Radio frequency circuits refer to circuits used to process radio frequency signals, and usually operate in a frequency range of tens of kHz to several thousand GHz. Radio frequency circuits generally include a large number of passive components, such as inductors, transmission lines, filters, transformers, etc. In the process of designing radio frequency circuits, designers need to perform electromagnetic simulations on each passive component in the circuit to predict the performance parameters of each passive component so that designers can select appropriate passive components for the radio frequency circuit to improve the performance of the entire radio frequency circuit. In related technologies, the overhead time for a single frequency sweep electromagnetic simulation of a single passive component is approximately 10-30 minutes, so the electromagnetic simulation of radio frequency circuits takes a relatively long time, resulting in relatively low efficiency in the design and development of radio frequency circuits. Summary of the Invention

[0003] To address the above-mentioned issues, an embodiment of the present application provides a data optimization method that takes into account the physical property constraints of the performance parameters of passive components, so that the output performance parameters meet the physical property constraints. Designers can select the most appropriate passive components for the RF circuit based on the performance parameters of the passive components that meet the physical property constraints, so that the performance of the designed and developed RF circuit is optimized. In addition, the present application also provides a data optimization system and computing device cluster corresponding to the data optimization method.

[0004] To this end, the following technical solutions are adopted in the embodiments of the present application:

[0005] In a first aspect, an embodiment of the present application provides a data optimization method, comprising: performing feature extraction on physical property parameters of a first passive device to obtain characteristic parameters of the first passive device; the first passive device is a passive device in a radio frequency circuit, or passive devices of the same type; the characteristic parameters include poles, residues, and constant terms; calculating a scattering parameter frequency response curve of the first passive device based on different frequencies within a set frequency range, the poles, the residues, and the constant terms; the scattering parameter frequency response curve represents the relationship between different frequencies and scattering parameters; and performing passivity correction on the scattering parameter frequency response curve of the first passive device to obtain a corrected scattering parameter frequency response curve of the first passive device.

[0006] In this embodiment, the data optimization system calculates the passive device's scattering parameter frequency response curve based on the passive device's physical properties and frequency. It then performs passivity correction on the passive device's scattering parameter frequency response curve, filtering out abnormal values ​​in the passive device's scattering parameter frequency response curve to smooth the overall curve. The scattering parameters of the passive device's scattering parameter frequency response curve output by the data optimization system ensure passivity while maintaining reciprocity and causality.

[0007] In one embodiment, the method further includes: using a parameter conversion formula to convert the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component into performance parameters other than the scattering parameters in the first passive component; the performance parameters include one or more of the scattering parameters, admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters and quality factor parameters.

[0008] In this embodiment, the data optimization system can calculate the performance parameters of the passive device at different frequencies, such as the admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters, and quality factor parameters, based on the scattering parameter frequency response curve of the passive device, so that the performance parameters of the passive device strictly maintain parameter consistency.

[0009] In one embodiment, calculating the scattering parameter frequency response curve of the first passive component based on different frequencies within a set frequency range, the poles, the residues, and the constant term specifically includes: using a differentiable transfer function to calculate the scattering parameters of multiple first passive components using the poles, the residues, and the constant term; and using the differentiable transfer function to calculate the scattering parameter frequency response curve of the first passive component using different frequencies within the set frequency range and the scattering parameters of the multiple first passive components.

[0010] In this embodiment, the data optimization system can use a differentiable transfer function to calculate the passive device's scattering parameters from its characteristic parameters, ensuring that the passive device's scattering parameters satisfy the reciprocity and causality constraints of physical properties. The data optimization system can then use the differentiable transfer function to calculate the passive device's scattering parameter frequency response curve based on different frequencies within a frequency range and the passive device's scattering parameters, facilitating subsequent passivity processing of the passive device's scattering parameters.

[0011] In one embodiment, the feature extraction of the physical property parameters of the first passive device to obtain the characteristic parameters of the first passive device specifically includes: sampling each parameter of the physical property parameters of the first passive device to obtain multiple first samples; each first sample includes a numerical value of each parameter; randomly sampling the multiple first samples to obtain a sample data set of the first passive device; the sample data set includes a portion of the multiple first samples; and feature extraction of the sample data set to obtain the characteristic parameters of the first passive device.

[0012] In this embodiment, the data optimization system samples various parameters of the physical property parameters of the passive device to obtain multiple samples, and obtains some samples from the multiple samples as samples for feature extraction, which can reduce the feature extraction overhead of the data optimization system.

[0013] In one embodiment, before extracting features from the physical property parameters of the first passive device to obtain the feature parameters of the first passive device, the method further includes: uniformly sampling each parameter of the physical property parameters of the first passive device to obtain a plurality of second samples; each second sample includes a numerical value of each parameter; randomly sampling the plurality of second samples to obtain a training data set for the first passive device; the training data set includes a portion of the plurality of second samples; inputting the training data set into a neural network for training to obtain training parameters for the first passive device; the training parameters are used to extract features from the physical property parameters of the first passive device through the neural network.

[0014] In this embodiment, the data optimization system uniformly samples various physical property parameters of the passive component to obtain multiple samples, and then extracts a portion of the samples from the multiple samples as a training data set. The data optimization system inputs the training data set into a neural network to obtain training parameters for the neural network, allowing the data optimization system to extract features from the physical property parameters of the passive component through the neural network.

[0015] In one embodiment, the method further includes: inputting second samples of the plurality of second samples other than the training data set and training parameters of the first passive component into the neural network to obtain test scattering parameters of the first passive component; and comparing the test scattering parameters of the first passive component with reference scattering parameters to detect performance of the training parameters of the first passive component.

[0016] In this embodiment, the data optimization system uses a second sample from the plurality of samples, other than the training dataset, as a test sample and inputs the training parameters of the passive device in the test sample into the neural network to obtain test scattering parameters of the passive device. The data optimization system can compare the test scattering parameters with actual reference scattering parameters obtained through numerical simulation or experiment, and evaluate the performance of the trained neural network based on the comparison results.

[0017] In one embodiment, the method further includes: obtaining a loss function value based on the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component and reference scattering parameters; and optimizing the training parameters of the first passive component using the loss function value using an optimization algorithm to obtain optimized training parameters of the first passive component.

[0018] In this embodiment, the data optimization system can compare the scattering parameters corresponding to the corrected passive device scattering parameter frequency response curve with actual reference scattering parameters obtained through numerical simulation or experimentation to obtain a loss function value. The data optimization system can adjust and optimize the training parameters of the neural network based on optimization algorithms such as stochastic gradient descent to improve the accuracy of the passive device scattering parameter frequency response curve.

[0019] In one embodiment, the method further includes: receiving physical property parameters of a second passive device; the second passive device and the first passive device are passive devices of the same type and have different manufacturing processes; sampling the training data set of the first passive device to obtain multiple third samples; replacing the physical property parameters in the multiple third samples with the physical property parameters of the second passive device to obtain a training data set of the second passive device; inputting the training data set of the second passive device into a neural network for training to obtain training parameters of the second passive device; the training parameters of the second passive device are used to perform feature extraction on the physical property parameters of the second passive device through the neural network.

[0020] In this embodiment, the data optimization system can obtain a training data set of passive components from an old manufacturing process and can sample the training data set of passive components from the old manufacturing process to obtain important samples. The data optimization system can use numerical simulation or experimental measurement to obtain the physical property parameters of passive components from the new manufacturing process corresponding to the physical property parameters in the important samples, thereby obtaining a training data set for passive components from the new manufacturing process. Based on the training data set of passive components from the new manufacturing process, the data optimization system can train training parameters for a neural network that can extract features of the physical property parameters of passive components from the new manufacturing process, thereby enabling the data optimization system to adapt to passive components from the new manufacturing process.

[0021] In a second aspect, an embodiment of the present application provides a data optimization system, including: an extraction module, used to perform feature extraction on physical property parameters of a first passive device to obtain characteristic parameters of the first passive device; the first passive device is a passive device in a radio frequency circuit, or passive devices of the same type; the characteristic parameters include poles, residues and constant terms; a transfer module, used to calculate a scattering parameter frequency response curve of the first passive device based on different frequencies within a set frequency range, the poles, the residues and the constant terms; the scattering parameter frequency response curve represents the relationship between different frequencies and scattering parameters; a correction module, used to perform passivity correction on the scattering parameter frequency response curve of the first passive device to obtain the corrected scattering parameter frequency response curve of the first passive device.

[0022] In one embodiment, the method further includes: a conversion module for converting the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component into performance parameters other than the scattering parameters in the first passive component using a parameter conversion formula; the performance parameters include one or more of the scattering parameters, admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters, and quality factor parameters.

[0023] In one embodiment, the transfer module is specifically configured to use a differentiable transfer function to calculate the scattering parameters of the plurality of first passive components using the poles, the residues, and the constant terms; and use the differentiable transfer function to calculate the scattering parameter frequency response curve of the first passive component using different frequencies within the set frequency range and the scattering parameters of the plurality of first passive components.

[0024] In one embodiment, the extraction module is specifically used to sample each parameter of the physical property parameters of the first passive component to obtain multiple first samples; each first sample includes a numerical value of each parameter; the multiple first samples are randomly sampled to obtain a sample data set of the first passive component; the sample data set includes a portion of the multiple first samples; and feature extraction is performed on the sample data set to obtain characteristic parameters of the first passive component.

[0025] In one embodiment, it also includes: a training module, which is used to uniformly sample each parameter of the physical property parameters of the first passive component to obtain multiple second samples; each second sample includes a numerical value of each parameter; randomly sample the multiple second samples to obtain a training data set of the first passive component; the training data set includes part of the multiple second samples; input the training data set into a neural network for training to obtain training parameters of the first passive component; the training parameters are used to extract features of the physical property parameters of the first passive component through the neural network.

[0026] In one embodiment, the training module is further configured to input second samples other than the training data set among the plurality of second samples and training parameters of the first passive component into the neural network to obtain test scattering parameters of the first passive component; the test scattering parameters of the first passive component are used to compare with reference scattering parameters to detect performance of the training parameters of the first passive component.

[0027] In one embodiment, the training module is further used to obtain a loss function value based on the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component and the reference scattering parameters; and use the optimization algorithm to optimize the training parameters of the first passive component using the loss function value to obtain the optimized training parameters of the first passive component.

[0028] In one embodiment, it also includes: an update module, which is used to receive physical property parameters of a second passive device; the second passive device and the first passive device are passive devices of the same type, and have different manufacturing processes; sampling the training data set of the first passive device to obtain multiple third samples; replacing the physical property parameters in the multiple third samples with the physical property parameters of the second passive device to obtain the training data set of the second passive device; inputting the training data set of the second passive device into the neural network for training to obtain the training parameters of the second passive device; the training parameters of the second passive device are used to perform feature extraction on the physical property parameters of the second passive device through the neural network.

[0029] In a third aspect, an embodiment of the present application provides a computing device comprising: a processor and a memory; the processor of the computing device is used to execute instructions stored in the memory of the computing device, so that the computing device executes various possible implementations of the first aspect.

[0030] In a fourth aspect, an embodiment of the present application provides a computing device cluster, characterized in that it includes at least one computing device, each computing device includes a processor and a memory; the processor of the at least one computing device is used to execute instructions stored in the memory of the at least one computing device, so that the computing device cluster executes the various possible implementation embodiments of the first aspect.

[0031] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, comprising computer program instructions. When the computer program instructions are executed by a computing device, the computing device executes the various possible implementations of the first aspect.

[0032] In a sixth aspect, an embodiment of the present application provides a computer program product comprising instructions, characterized in that the computer program product stores instructions that, when executed by a computing device, enable the computing device to implement various possible implementation embodiments of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The following is a brief introduction to the drawings required for describing the embodiments or prior art.

[0034] Figure 1 A schematic diagram of the structure of a data optimization system provided in an embodiment of the present application;

[0035] Figure 2 A flow chart of a data optimization method provided in an embodiment of the present application;

[0036] Figure 3 This is a schematic diagram of the distribution of the results output by the data optimization system provided in the embodiment of the present application that meet causality;

[0037] Figure 4 This is a schematic diagram showing how the performance parameter prediction error of the data optimization system provided in the embodiment of the present application decreases as the number of samples in the training data set of the new manufacturing process increases;

[0038] Figure 5 Schematic diagram of the relative error distribution of inductance, capacitance, resistance, and quality factor output by the data optimization system provided in an embodiment of the present application;

[0039] Figure 6 This is a structural diagram of a data optimization device provided in an embodiment of the present application;

[0040] Figure 7 A schematic diagram of the structure of a computing device provided in an embodiment of the present application;

[0041] Figure 8 A schematic diagram of the architecture of a computing device cluster provided in an embodiment of the present application;

[0042] Figure 9 This is a schematic diagram of the architecture of another computing device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0044] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship, for example, A / B means either A or B.

[0045] The terms "first" and "second" in this specification and claims are used to distinguish different objects rather than to describe a specific order of objects. For example, "first response message" and "second response message" are used to distinguish different response messages rather than to describe a specific order of response messages.

[0046] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0047] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, for example, multiple processing units means two or more processing units, etc.; multiple elements means two or more elements, etc.

[0048] Before introducing the technical solution protected by this application, several professional terms involved in the technical solution protected by this application are explained in advance, namely:

[0049] The performance parameters of passive components in RF circuits refer to scattering (S) parameters, admittance (Y) parameters, impedance (Z) parameters, capacitance (C) parameters, inductance (L) parameters, resistance (R) parameters and quality factor (Q) parameters.

[0050] Scattering parameters describe the characteristics of single-port or multi-port passive devices under RF working conditions, as well as the response of passive devices to the transmission, attenuation, coupling, etc. of RF signals.

[0051] Admittance parameters describe the conductivity and transmission characteristics of single-port or multi-port passive devices in RF circuits. They are generally complex numbers (single-port devices) or complex matrices (multi-port devices). The real part is the conductance and the imaginary part is the susceptance.

[0052] The impedance parameter is the reciprocal (single-port device) or inverse matrix (multi-port device) of the admittance parameter. It describes the degree of obstruction of the passive device to the RF signal. It is generally a complex number or a complex matrix. The real part is resistance and the imaginary part is reactance.

[0053] Capacitance parameters describe the capacitance characteristics of passive devices. The capacitance parameters of multi-port devices usually refer to their equivalent capacitance in the equivalent circuit model.

[0054] Inductance parameters describe the inductance characteristics of passive components. The inductance parameters of multi-port components usually refer to their equivalent inductance in the equivalent circuit model.

[0055] Resistance parameters describe the resistance characteristics of passive components. The resistance parameters of multi-port components usually refer to their equivalent resistance values ​​in the equivalent circuit model.

[0056] The quality factor parameter describes the effective performance of a passive component. The quality factor parameter depends on the passive component's loss, frequency response, bandwidth characteristics, etc.

[0057] The physical property constraints of passive devices refer to the passivity, reciprocity, causality, smoothness and parameter consistency of the performance parameters of passive devices.

[0058] Passivity means that passive devices do not actively generate energy at any frequency, but only transmit, attenuate, or couple input signals. In other words, the total output energy of a passive device is less than or equal to the total output energy.

[0059] Reciprocity refers to the fact that a passive device's response to an input signal is independent of the signal's direction of propagation. The passive device's transmission and coupling to the input signal remain unchanged by changes in the input signal's direction of propagation.

[0060] Causality means that the response of a passive device depends only on past and current input signals and will not be affected by future input signals.

[0061] Smoothness means that the response of a passive device changes continuously with the input signal without sudden or discontinuous changes.

[0062] Parameter consistency means that the property parameters of passive devices under different conditions remain consistent and can be converted into each other through formulas.

[0063] In order to improve the efficiency of designing and developing RF circuits, a model that can predict the performance parameters of RF passive components has been designed in related technologies, which is called a "proxy model". The workflow of the proxy model can be divided into a training phase and an inference phase. In the training phase, designers use a set of physical property parameters to describe a class of passive components, generate training samples and verification samples through sampling, and use numerical simulation or experimental measurement to obtain the labels corresponding to the samples, and train the proxy model. In the inference phase, designers can input the physical property parameters of one or a group of passive components of the same type as the training samples into the proxy model, and let the proxy model directly output the performance parameters of the passive components. The proxy model is a model composed of a network that can predict the performance parameters of passive components at extremely low computational cost and time overhead, thereby improving the design and development efficiency of RF circuits.

[0064] However, the proxy models used in related technologies can only perform purely data-driven modeling, resulting in poor physical constraints on the performance parameters of the predicted passive components. Designers, based on the performance parameters of the passive components output by the proxy models, only achieve limited performance improvements in RF circuits constructed from the passive components they select.

[0065] In order to solve the defects existing in the related art, the embodiment of the present application designs a data optimization system. In the training stage, after receiving the physical property parameters of the same type of passive devices, the data optimization system takes into account the physical property constraints of the performance parameters of the passive devices in the process of adjusting and optimizing the training parameters of the neural network within the data optimization system, so that the output performance parameters meet the physical property constraints. In the reasoning stage, after receiving one or a group of physical property parameters of the same type of passive devices as in the training stage, the data optimization system can directly output performance parameters that meet the physical property constraints. Designers can select the most suitable passive devices for the RF circuit based on the performance parameters of the passive devices that meet the physical property constraints, so that the performance of the designed and developed RF circuit is optimal.

[0066] Figure 1 This is a structural diagram of a data optimization system provided in an embodiment of the present application. Figure 1 As shown, the data optimization system 100 can be divided into an extraction module 110, a transfer module 120, a correction module 130, a conversion module 140, a training module 150 and an update module 160 according to execution functions.

[0067] Data optimization system 100 can be an application, software code, or the like. Data optimization system 100 can be deployed on a local device used by a designer, such as a computer, server, laptop, tablet, or smartphone. Data optimization system 100 can also be deployed on a cloud server. If data optimization system 100 is deployed on a cloud server, designers can use their local devices to access the cloud server and complete tasks using data optimization system 100.

[0068] The extraction module 110 is used to extract the physical property parameters of a passive device or passive devices of the same type after receiving them, and obtain the characteristic parameters of the passive device. The physical property parameters of the passive device refer to various attribute data of the passive device, such as material, length, width, height, area, volume, number of coil turns, spacing, and other parameters. Taking an inductor as an example, the physical property parameters of the inductor can be the radius of the inductor (rad), the aspect ratio of the inductor (x_y_ratio), the number of turns of the inductor (turn), the coil width of the inductor (space), the winding spacing of the inductor (width), etc.

[0069] After receiving the physical property parameters, the extraction module 110 can sample each parameter in the physical property parameters to construct multiple samples. The values ​​of the physical property parameters corresponding to each sample are not exactly the same. A portion of the multiple samples is sampled as a sample data set for the passive device.

[0070] Before the extraction module 110 performs feature extraction on the physical property parameters, it can detect whether there are training parameters of a neural network that extracts the characteristics of the passive device corresponding to the physical property parameters locally. If the extraction module 110 determines that the training parameters of the neural network that extracts the characteristics of the physical property parameters of the passive device do not exist locally, it indicates that the extraction module 110 may not be able to perform feature extraction on the physical property parameters of the passive device. In an embodiment of the present application, the training module 150 can train the neural network based on the physical property parameters to obtain training parameters of the neural network that can characterize the physical property parameters of the passive device, and input the obtained training parameters into the extraction module 110, so that the neural network in the extraction module 110 can perform feature extraction on the physical property parameters of the passive device based on the training parameters.

[0071] In one embodiment, taking an inductor as an example, the physical property parameters of the inductor are the radius of the inductor, the aspect ratio of the inductor, the number of windings of the inductor, the coil width of the inductor, and the winding spacing of the inductor. Among them, the radius of the inductor ranges from 30 microns to 120 microns, the aspect ratio of the inductor is 0.5, 0.75 or 1.0, the number of windings of the inductor ranges from 1 to 7.5, the coil width of the inductor ranges from 2.0 microns to 10.0 microns, and the winding spacing of the inductor ranges from 2 microns to 6 microns. The extraction module 110 uniformly samples the five parameters of the inductor to obtain 9318 samples. The extraction module 110 can randomly sample the 9318 samples and sample 9021 training samples as the training data set of the inductor, which is used to adjust the neural network in the training extraction module 110 to obtain the training parameters of the neural network. The remaining 297 samples are used as test samples to evaluate the performance of the trained neural network.

[0072] In an embodiment of the present application, the extraction module 110 can use a neural network model such as a fully connected neural network (FCNN) and a convolutional neural network (CNN) to perform feature extraction on a sample data set of the same type of passive devices input to output characteristic parameters such as poles (p), residues (r) and constant terms (c) corresponding to the scattering parameter frequency response curve of the passive device. Among them, poles, residues and constant terms are concepts that describe circuit response and transfer function in radio frequency circuits. Poles are points that cause the transfer function to diverge. When the frequency approaches the poles, it will cause the circuit to have more significant attenuation or gain near the frequency. The residues are the coefficients of the amplitude and phase at the poles of the transfer function, which are used to calculate the frequency response of the transfer function. The constant term is the low-frequency gain of the transfer function, that is, the system response when the input frequency approaches zero.

[0073] In one possible embodiment, the FCNN can perform feature extraction on a sample dataset and use poles, residues, and constant terms as network outputs. During training, the FCNN can use a pole-residue transfer function to extract features from the sample dataset to obtain the poles, residues, and constant terms of the passive device, and output the real part of the pole, the imaginary part of the pole, the real part of the residue, the imaginary part of the residue, and the constant term.

[0074] The transfer module 120 is configured to calculate the scattering parameters of the passive device based on the characteristic parameters of the passive device after receiving the characteristic parameters of the passive device. In the embodiment of the present application, after receiving the characteristic parameters of the passive device, the transfer module 120 can use a transfer function to calculate the scattering parameters of the passive device based on the characteristic parameters of the passive device, so that the scattering parameters of the passive device can meet the reciprocity and causality constraints of the physical properties.

[0075] Optionally, the transfer function of the transfer module 120 is a differentiable transfer function. A differentiable transfer function refers to a transfer function with a continuous derivative, which is usually used to describe the response characteristics of a system. A differentiable transfer function can help understand and optimize the frequency response of a circuit. The differentiable transfer function can be a function of formula (1) or formula (2), and formula (1) is specifically:

[0076]

[0077] Where ω = 2πf represents the angular frequency, f represents the frequency, and k represents the index of the poles and residues. is the imaginary unit, p represents the pole, r represents the residue, and c represents the constant term.

[0078] Formula (2) is specifically:

[0079]

[0080] Here, z represents the zero point.

[0081] In an embodiment of the present application, the transmission module 120 can obtain in advance the frequency range of the RF circuit under normal operating conditions. The transmission module 120 can use a differentiable transfer function to calculate the scattering parameter frequency response curve of the passive device based on different frequencies within the frequency range and the scattering parameters of the passive device, and transmit the scattering parameter frequency response curve of the passive device to other modules so that other modules can calculate other performance parameters based on the scattering parameter frequency response curve of the passive device, optimize the neural network, etc. The scattering parameter frequency response curve describes the response characteristics of the RF circuit at different frequencies and shows the relationship between the scattering parameters of the passive device and frequency.

[0082] The transfer module 120 can use poles / zeros and poles / residues to represent the scattering parameter frequency response curve. In the pole / zero form, the frequency response characteristics of the differentiable transfer function are described by poles and zeros. Poles are important factors that cause changes in system response and can determine the stability of the system and the shape of the frequency response. Zeros affect the gain and phase characteristics of the system. In the pole / residue form, the frequency response characteristics of the differentiable transfer function are described by poles and residues. Poles indicate the stability of the system and the shape of the frequency response. Residues indicate the response contribution of each pole in the system.

[0083] The correction module 130 is used to perform passivity correction on the scattering parameter frequency response curve of the passive device after receiving the scattering parameter frequency response curve of the passive device, so that the scattering parameter frequency response curve of the passive device is smoother, and the scattering parameters of the passive device meet passivity, reciprocity and causality. In the embodiment of the present application, the correction module 130 can use a causal filter to smooth the scattering parameter frequency response curve of the passive device, filter out abnormal values ​​in the scattering parameter frequency response curve of the passive device, or correct the abnormal values ​​in the scattering parameter frequency response curve of the passive device, so that the entire curve is smoother. The correction module 130 smoothes the scattering parameter frequency response curve of the passive device, so that the scattering parameters in the scattering parameter frequency response curve of the passive device are guaranteed to be passivated while maintaining reciprocity and causality. Among them, the causal filter can be constructed by formulas (3)-(5), specifically:

[0084]

[0085]

[0086]

[0087] in, represents the phase of the filter, σ max () represents the maximum singular value of the matrix, S(ω) represents the scattering parameter matrix at a certain frequency, and Hilbert() represents the Hilbert transform.

[0088] In an embodiment of the present application, the correction module 130 may input the corrected scattering parameter frequency response curve of the passive device into the training module 150. After obtaining the corrected scattering parameter frequency response curve of the passive device, the training module 150 may compare the scattering parameters corresponding to the corrected scattering parameter frequency response curve of the passive device with actual reference scattering parameters obtained through numerical simulation or experiment to obtain a loss function value. The training module 150 may adjust and optimize the training parameters of the neural network in the training module 150 based on an optimization algorithm such as stochastic gradient descent, thereby improving the accuracy of the scattering parameter frequency response curve of the passive device calculated by the transfer module 120 based on the characteristic parameters of the passive device output by the extraction module 110.

[0089] The conversion module 140 is used to calculate the performance parameters such as the admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters, and quality factor parameters of the passive device at different frequencies according to the scattering parameter frequency response curve of the passive device after receiving the corrected scattering parameter frequency response curve of the passive device, so that the performance parameters of the passive device strictly maintain parameter consistency. In the embodiment of the present application, in the process of calculating other performance parameters of the passive device, the conversion module 140 can use the parameter conversion formula within the automatic differentiation framework to perform conversion calculations between different parameters to ensure parameter consistency between the various parameters. The automatic differentiation framework can be a framework such as pytorch and tensorflow.

[0090] In one embodiment, if the passive device is a two-port passive device, the physical formula used by the conversion module 140 to calculate the admittance parameter is:

[0091] Y 11 =Y0((1-S 11 )(1+S 22 )+S 12 S 21 ) / ((1+s 11 )(1+s 22 )-s 12 s 21 );

[0092] Y 12 =Y0(-2s 12 ) / ((1+s 11 )(1+s 22 )-S 12 S 21 );

[0093] Y 21 =Y0(-2S 21 ) / ((1+S 11 )(1+S 22 )-S 12 S21 );

[0094] Y 12 =Y0((1+S 11 )(1-S 22 )+S 12 S 21 ) / ((1+S 11 )(1+S 22 )-S 12 S 21 ).

[0095] In one embodiment, if the passive device is a two-port passive device, the physical formula used by the conversion module 140 to calculate the impedance parameter is:

[0096] Z 11 =Y 22 / (Y 11 Y 22 -Y 12 Y 21 );

[0097] Z 12 =-Y 12 / (Y 11 Y 22 -Y 12 Y 21 );

[0098] Z 21 =-Y 21 / (Y 11 Y 22 -Y 12 Y 21 );

[0099] Z 22 =Y 11 / (Y 11 Y 22 -Y 12 Y 21 ).

[0100] In one embodiment, if the passive device is a two-port single-ended inductor, the physical formula used by the conversion module 140 to calculate the capacitance parameter is:

[0101] C=Imag(Y 11 +Y 21 ) / ω;

[0102] Where ω is the angular frequency and Imag() represents the imaginary part of the complex number.

[0103] In one embodiment, if the passive device is a two-port single-ended inductor, the physical formula used by the conversion module 140 to calculate the inductor parameters is:

[0104] L=Imag(1 / Y 11 ) / ω.

[0105] In one embodiment, if the passive device is a two-port single-ended inductor, the physical formula used by the conversion module 140 to calculate the resistance parameter is:

[0106] R=Real(1 / Y 11 );

[0107] Among them, Real() means taking the real part of the complex number.

[0108] In one embodiment, if the passive device is a two-port single-ended inductor, the physical formula used by the conversion module 140 to calculate the quality factor parameter is:

[0109] Q=-Imag(Y 11 ) / Real(Y 11 ).

[0110] In an embodiment of the present application, during electromagnetic simulation of a passive device by the data optimization system 100, the physical property parameters of the passive device can be input into the FCNN to obtain the characteristic parameters of the passive device. After calculating the scattering parameters at different frequencies based on the characteristic parameters of the passive device, the data optimization system 100 performs passivity correction on the scattering parameters and calculates other performance parameters of the passive device. This ensures that the performance parameters of the passive device output by the data optimization system 100 meet passivity, reciprocity, causality, smoothness of the frequency response curve, and parameter consistency. This also ensures that the performance of the RF circuit designed and developed by the designer based on the performance parameters of the passive device output by the data optimization system 100 is optimized.

[0111] Typically, the manufacturing process of passive components also affects their performance parameters. Therefore, when a network model for passive components using an older manufacturing process extracts features from the physical property parameters of passive components using a newer manufacturing process, the calculated performance parameters may deviate from their true values. In an embodiment of the present application, the update module 160 can obtain a training dataset for passive components using an older manufacturing process. The update module 160 can sample the training dataset for passive components using an older manufacturing process to obtain significant samples and record the physical property parameters in the significant samples. The update module 160 can utilize numerical simulation or experimental measurement to obtain the physical property parameters of passive components using a newer manufacturing process corresponding to the physical property parameters in the significant samples, thereby obtaining a training dataset for passive components using a newer manufacturing process. The update module 160 can input the training dataset for passive components using a newer manufacturing process into the training module 150, allowing the training module 150 to train training parameters for a neural network capable of extracting features from the physical property parameters of passive components using a newer manufacturing process, thereby enabling the data optimization system 100 to adapt to passive components using a newer manufacturing process.

[0112] In the embodiment of the present application, the update module 160 uses active learning and transfer learning technology, which can significantly reduce the number of training samples of new process devices while ensuring the performance parameter prediction accuracy of the data optimization system 100 corresponding to the new process.

[0113] The following is an introduction through the process Figure 1 The process of implementing the plan.

[0114] Figure 2 Schematic diagram of a data optimization method provided in the embodiment of this application. Figure 2 As shown, the data optimization method can be Figure 1 The data optimization system 100 in the embodiment is executed, and the specific implementation process is as follows:

[0115] Step S201 : extracting features of physical property parameters of a first passive component to obtain feature parameters of the first passive component.

[0116] The physical property parameters of a passive device refer to various attribute data of the passive device, such as parameters such as material, length, width, height, area, volume, number of coil turns, spacing, etc. In the embodiment of the present application, before the data optimization system 100 performs feature extraction on the physical property parameters, it can detect whether there are training parameters of a neural network for extracting features of the physical property parameters of the passive device locally. When the data optimization system 100 determines that there are no training parameters of a neural network for extracting features of the physical property parameters of the passive device locally, it indicates that the data optimization system 100 may not be able to perform feature extraction on the physical property parameters of the passive device.

[0117] At this time, the data optimization system 100 uniformly samples each parameter in the physical property parameters to obtain multiple samples. The data optimization system 100 can randomly sample the multiple samples to sample multiple training samples as a training data set for the passive device, which is used to adjust the training neural network to obtain the training parameters of the neural network. The remaining samples are used as test samples to evaluate the performance of the trained neural network. The data optimization system 100 can train the neural network based on the physical property parameters to obtain training parameters of the neural network that can characterize the physical property parameters of the passive device, so that the neural network can extract features of the physical property parameters of the passive device based on the training parameters.

[0118] After receiving the physical property parameters, the data optimization system 100 can sample each of the physical property parameters to construct multiple samples. The values ​​of the physical property parameters corresponding to each sample are not exactly the same. A portion of the multiple samples is sampled as a sample dataset for the passive device. The data optimization system 100 can utilize a neural network model to perform feature extraction on the input sample dataset to output characteristic parameters such as poles, residues, and constant terms corresponding to the scattering parameter frequency response curve of the passive device.

[0119] Step S202 : calculating a scattering parameter frequency response curve of the first passive component according to different frequencies, poles, residues, and constant terms within a set frequency range.

[0120] After receiving the characteristic parameters of the passive device, the data optimization system 100 can use the differentiable transfer function to calculate the scattering parameters of the passive device from the characteristic parameters of the passive device, so that the scattering parameters of the passive device can meet the reciprocity and causality of the physical property constraints.

[0121] Data optimization system 100 can obtain the frequency range of the RF circuit under normal operating conditions in advance. Using a differentiable transfer function, data optimization system 100 can calculate the passive component's scattering parameter frequency response curve based on the different frequencies within the frequency range and the passive component's scattering parameters. The scattering parameter frequency response curve describes the RF circuit's response characteristics at different frequencies and shows the relationship between the passive component's scattering parameters and frequency.

[0122] Exemplarily, the data optimization system 100 can input the poles, residues and constant terms in the characteristic parameters, as well as the numerical values ​​of the frequency range under normal working conditions of the radio frequency circuit, into the differentiable transfer function represented by formula (1) or formula (2), and output the scattering parameters at different frequencies. The data optimization system 100 can construct a scattering parameter frequency response curve based on the scattering parameters at different frequencies. The scattering parameter frequency response curve obtained by the data optimization system 100 is obtained based on the characteristic parameters trained based on the physical property parameters of the passive device, which ensures that the scattering parameters in the scattering parameter frequency response curve of the passive device can meet the causality of the physical property constraints. The scattering parameter frequency response curve obtained by the data optimization system 100 is calculated based on the differentiable transfer function. The differentiable transfer function is a transfer function with continuous derivatives, which can better understand and optimize the frequency response of the circuit, so that the scattering parameters in the scattering parameter frequency response curve of the passive device can meet the reciprocity of the physical property constraints.

[0123] Step S203 , performing passivity correction on the frequency response curve of the first passive component to obtain a corrected scattering parameter frequency response curve of the first passive component.

[0124] Data optimization system 100 can use a causal filter to smooth the scattering parameter frequency response curve of the passive device, filtering out abnormal values ​​in the scattering parameter frequency response curve of the passive device, or correcting abnormal values ​​in the scattering parameter frequency response curve of the passive device, thereby making the entire curve smoother. By smoothing the scattering parameter frequency response curve of the passive device, data optimization system 100 ensures the passivity of the scattering parameters in the scattering parameter frequency response curve of the passive device while maintaining reciprocity and causality.

[0125] In an embodiment of the present application, after obtaining the corrected scattering parameter frequency response curve of the passive device, data optimization system 100 can compare the scattering parameters corresponding to the corrected scattering parameter frequency response curve of the passive device with actual reference scattering parameters obtained through numerical simulation or experimentation to obtain a loss function value. Data optimization system 100 can adjust and optimize the training parameters of the neural network based on an optimization algorithm such as stochastic gradient descent to improve the accuracy of the calculated scattering parameter frequency response curve of the passive device.

[0126] In an embodiment of the present application, after receiving the corrected scattering parameter frequency response curve of the passive device, the data optimization system 100 can calculate the passive device's performance parameters, such as admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters, and quality factor parameters, at different frequencies based on the scattering parameter frequency response curve of the passive device, thereby ensuring that the performance parameters of the passive device strictly maintain parameter consistency. In the process of calculating other performance parameters of the passive device, the data optimization system 100 can use parameter conversion formulas within the automatic differentiation framework to perform conversion calculations between different parameters to ensure parameter consistency between the various parameters.

[0127] Generally, the manufacturing process of passive devices will also affect the performance parameters of passive devices. Therefore, when the network model of passive devices suitable for old manufacturing processes extracts features of the physical property parameters of passive devices of new manufacturing processes, it may cause the calculated performance parameters to deviate from the true values. In an embodiment of the present application, the data optimization system 100 can obtain a training data set of passive devices of old manufacturing processes, sample the training data set of passive devices of old manufacturing processes to obtain important samples, and record the physical property parameters in the important samples. The data optimization system 100 can use numerical simulation or experimental measurement to obtain the physical property parameters of passive devices of new manufacturing processes corresponding to the physical property parameters in the important samples, thereby obtaining a training data set of passive devices of new manufacturing processes. The data optimization system 100 can input the training data set of passive devices of new manufacturing processes into a neural network, train the training parameters of the neural network that can extract the features of the physical property parameters of passive devices of new manufacturing processes, thereby realizing that the data optimization system 100 can be adapted to passive devices of new manufacturing processes. The data optimization system 100 adopts active learning and transfer learning technology, which can significantly reduce the number of training samples of new process devices while ensuring the performance parameter prediction accuracy of the data optimization system 100 corresponding to the new process.

[0128] The following simulation experiment results are used to demonstrate the technical effect of the technical solution protected by this application.

[0129] Designers conducted simulation experiments on five passive components: a single-ended inductor (IND_SINGLE), a differential inductor (IND_DIFF), a single-ended transmission line (CPW_SINGLE), a differential transmission line (CPW_DIFF), a 1:2 transformer (XFMR_OVERLAP_1v2), and a 1:3 transformer (XFMR_OVERLAP_1v3). In verifying passivity, one simulation experiment used predictions based on the data optimization system 100 protected by the present application, and another simulation experiment used predictions based on a proxy model in the related art.

[0130] As shown in Table 1, the data optimization system 100 protected by this application has greatly improved the proportion of physical performance parameters output by six passive components, namely single-ended inductors, differential inductors, single-ended transmission lines, differential transmission lines, 1:2 transformers and 1:3 transformers, that meet the passivity constraints, and all are 100%.

[0131] Table 1 The proportion of the output results of six passive devices with and without passivity constraints that meet the passivity constraints

[0132]

[0133] In the process of predicting the scattering parameters of a passive device by the data optimization system 100 protected by the present application, the distribution of the generated poles on the complex plane is as follows: Figure 3 Since all the poles are stably distributed on the left half plane of the complex plane (ie, the real parts are all less than zero), the scattering parameters output by the data optimization system 100 all meet the causality constraint.

[0134] Assume that the data optimization system 100 is a system 100 obtained by the designer through training of a single-ended inductor of an old manufacturing process. If the designer needs a data optimization system for a single-ended inductor suitable for a new manufacturing process, he can use Figure 1-2 And the training scheme described in the related description, with fewer samples of passive components of the new manufacturing process, a data optimization system corresponding to the new manufacturing process with accuracy that meets the requirements is obtained. Specifically, the update module 160 selects important samples from the training samples of the old manufacturing process and records the physical property parameters in the important samples. Then, the update module 160 uses methods such as numerical simulation or experimental measurement to obtain the physical property parameters of the passive components of the new manufacturing process corresponding to the physical property parameters in the important samples, and forms a training data set for the passive components of the new manufacturing process. Finally, the training module 150 can be trained with the training data set of the passive components of the new manufacturing process to adjust and optimize the training parameters of the neural network in the data optimization system corresponding to the old manufacturing process, and obtain the data optimization system corresponding to the new manufacturing process. Taking a single-ended inductor as an example, the training data set of the data optimization system corresponding to the old manufacturing process contains 9021 training samples. Using the scheme proposed in this application, a data optimization system corresponding to the new process with accuracy that meets the requirements can be obtained with only 902 new process training samples.

[0135] Figure 4 This is a schematic diagram showing how the performance parameter prediction error of the data optimization system provided in the embodiment of the present application decreases as the number of samples in the training data set of the new manufacturing process increases. Figure 4As shown, (in the single-ended inductor, assuming that the number of samples in the training data set of the old manufacturing process is 9021, it is recorded as 100%. The dotted line corresponds to the solution (labeled as Full Dataset, w / Model Transfer), which is to retrain the data optimization system on the 9021 training samples of the new manufacturing process. In addition, the prediction error generated by 297 test samples is used as a benchmark. The dotted line labeled Important Samples, w / Model Transfer is the performance parameter prediction error of the new data optimization system obtained by training using the solution proposed in this application. Figure 4 As shown, the solution proposed in this application only requires 902 training samples of the new manufacturing process (equivalent to 10% of the number of training samples of the old process) to obtain a data optimization system for the new manufacturing process with a prediction accuracy reaching the baseline, and the overhead of generating training samples for the new manufacturing process can be significantly reduced.

[0136] Figure 5 Schematic diagram of the relative error distribution of inductance, capacitance, resistance, and quality factor output by the data optimization system provided in the embodiment of this application. The data optimization system 100 was trained using 902 training samples of the new manufacturing process and tested using 297 test samples, making the data optimization system 100 applicable to the system of the new manufacturing process. Figure 5 As shown in the figure, the relative error of inductance is concentrated in the range of less than 0.25%, and the maximum error is less than 1%. The relative errors of capacitance, resistance and quality factor are concentrated in the range of less than 1%, and the maximum error is less than 5%.

[0137] Figure 6 Schematic diagram of the structure of an image processing device provided in an embodiment of the present application. Figure 6 As shown, the image processing device 600 can be divided into a first processing unit 610, a second processing unit 620 and a third processing unit 630 according to the execution function. The specific implementation process of the image processing device 600 is as follows:

[0138] The first processing unit 610 is used to extract features of the physical property parameters of the first passive device to obtain characteristic parameters of the first passive device. The first passive device is a passive device in the radio frequency circuit, or passive devices of the same type. The characteristic parameters include poles, residues, and constant terms. The second processing unit 620 is used to calculate the scattering parameter frequency response curve of the first passive device based on different frequencies, poles, residues, and constant terms within a set frequency range. The scattering parameter frequency response curve represents the relationship between different frequencies and scattering parameters. The third processing unit 630 is used to perform passivity correction on the scattering parameter frequency response curve of the first passive device to obtain the corrected scattering parameter frequency response curve of the first passive device.

[0139] In one embodiment, the third processing unit 630 is further configured to convert the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component into performance parameters other than the scattering parameters of the first passive component using a parameter conversion formula. The performance parameters include one or more of a scattering parameter, an admittance parameter, an impedance parameter, a capacitance parameter, an inductance parameter, a resistance parameter, and a quality factor parameter.

[0140] In one embodiment, the second processing unit 620 is specifically configured to use a differentiable transfer function to calculate scattering parameters of the plurality of first passive components from poles, residues, and constant terms. The second processing unit 620 is specifically configured to use a differentiable transfer function to calculate a scattering parameter frequency response curve of the first passive components from different frequencies within a set frequency range and the scattering parameters of the plurality of first passive components.

[0141] In one embodiment, the first processing unit 610 is specifically configured to sample various parameters of the physical property parameters of the first passive component to obtain a plurality of first samples. Each first sample includes a numerical value for each parameter. The first processing unit 610 is specifically configured to randomly sample the plurality of first samples to obtain a sample dataset of the first passive component. The sample dataset includes a portion of the plurality of first samples. The first processing unit 610 is specifically configured to perform feature extraction on the sample dataset to obtain characteristic parameters of the first passive component.

[0142] In one embodiment, the first processing unit 610 is further configured to uniformly sample each parameter of the physical property parameters of the first passive component to obtain a plurality of second samples. Each second sample includes a numerical value for each parameter. The first processing unit 610 is further configured to randomly sample the plurality of second samples to obtain a training dataset for the first passive component. The training dataset includes a portion of the plurality of second samples. The first processing unit 610 is further configured to input the training dataset into a neural network for training to obtain training parameters for the first passive component. The training parameters are used to extract features from the physical property parameters of the first passive component via the neural network.

[0143] In one embodiment, the first processing unit 610 is further configured to input second samples other than the training dataset from the plurality of second samples and the training parameters of the first passive component into the neural network to obtain test scattering parameters of the first passive component. The test scattering parameters of the first passive component are used to compare with reference scattering parameters to detect the performance of the training parameters of the first passive component.

[0144] In one embodiment, the third processing unit 630 is further configured to obtain a loss function value based on the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component and the reference scattering parameters. The third processing unit 630 is further configured to optimize the training parameters of the first passive component using the loss function value using an optimization algorithm to obtain optimized training parameters of the first passive component.

[0145] In one embodiment, the third processing unit 630 is further configured to receive physical property parameters of a second passive device. The second passive device and the first passive device are of the same type and have different manufacturing processes. The third processing unit 630 is further configured to sample the training data set of the first passive device to obtain a plurality of third samples. The third processing unit 630 is further configured to replace the physical property parameters in the plurality of third samples with the physical property parameters of the second passive device to obtain a training data set of the second passive device. The third processing unit 630 is further configured to input the training data set of the second passive device into a neural network for training to obtain training parameters of the second passive device. The training parameters of the second passive device are used to perform feature extraction on the physical property parameters of the second passive device through the neural network.

[0146] Figure 7 This is a schematic diagram of the structure of a computing device provided in an embodiment of the present application. Figure 7 As shown, computing device 700 includes a bus 710, a processor 720, a memory 730, and a communication interface 740. Processor 720, memory 730, and communication interface 740 communicate with each other via bus 710. Computing device 700 may be a server, a computer, a portable notebook, a cabinet, etc. It should be understood that this application does not limit the number of processors and memories in computing device 700.

[0147] The bus 710 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The bus 710 may include a path for transmitting information between various components of the computing device 700 (eg, the processor 720, the memory 730, and the communication interface 740).

[0148] The processor 720 may be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0149] The memory 730 may include a volatile memory, such as a random access memory (RAM). The memory 730 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0150] Memory 730 stores executable program code. Processor 720 executes this executable program code to implement the functions of the aforementioned modules, such as extraction module 110, transfer module 120, correction module 130, conversion module 140, training module 150, and update module 160, thereby implementing the data optimization method. In other words, memory 730 stores instructions for executing the data optimization method.

[0151] Alternatively, the memory 730 stores executable codes, and the processor 720 executes the executable codes to respectively implement the functions of the aforementioned modules, thereby implementing the data optimization method. In other words, the memory 730 stores instructions for executing the data optimization method.

[0152] The communication interface 740 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 700 and other devices or a communication network.

[0153] Embodiments of the present application also provide a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smartphone.

[0154] like Figure 8 As shown, the computing device cluster includes at least one computing device 700. The memory 730 in one or more computing devices 700 in the computing device cluster may store the same instructions for executing the data optimization method.

[0155] In some possible implementations, the memory 730 of one or more computing devices 700 in the computing device cluster may also store some instructions for executing the data optimization method. In other words, the combination of one or more computing devices 100 can jointly execute the instructions for executing the data optimization method.

[0156] It should be noted that the memory 730 in different computing devices 700 in the computing device cluster may store different instructions, each for executing part of the functions of the aforementioned extraction module 110, transfer module 120, correction module 130, conversion module 140, training module 150, and update module 160. In other words, the instructions stored in the memory 730 in different computing devices 700 may implement the functions of one or more of the aforementioned extraction module 110, transfer module 120, correction module 130, conversion module 140, training module 150, and update module 160.

[0157] In some possible implementations, one or more computing devices in a computing device cluster may be connected via a network, which may be a wide area network or a local area network. Figure 9 A possible implementation is shown. Figure 9 As shown, two computing devices, computing device 700A and computing device 700B, are connected via a network. Specifically, the connection to the network is achieved through a communication interface in each computing device. In this type of possible implementation, the memory 730 in computing device 700A stores instructions for executing the functions of some modules among the above-mentioned extraction module 110, transfer module 120, correction module 130, conversion module 140, training module 150, and update module 160. At the same time, the memory 730 in computing device 700B stores instructions for executing the functions of another part of the above-mentioned extraction module 110, transfer module 120, correction module 130, conversion module 140, training module 150, and update module 160.

[0158] Figure 9 The connection method between the computing device clusters shown can be based on the fact that the data optimization method provided in this application requires a large amount of data storage, so it is considered to entrust the functions implemented by another part of the modules in the above-mentioned extraction module 110, transmission module 120, correction module 130, conversion module 140, training module 150 and update module 160 to be executed by the computing device 700B.

[0159] It should be understood that Figure 9 The functionality of the computing device 700A shown in FIG. 7 may also be implemented by multiple computing devices 700. Similarly, the functionality of the computing device 700B may also be implemented by multiple computing devices 700.

[0160] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similarly referred to as Figure 7 and Figure 8 The connection mode of the computing device cluster is different in that the memory 730 of one or more computing devices 700 in the computing device cluster may store the same instructions for executing the data optimization method.

[0161] In some possible implementations, the memory 730 of one or more computing devices 700 in the computing device cluster may also store partial instructions for executing the data optimization method. In other words, the combination of one or more computing devices 700 can jointly execute the instructions for executing the data optimization method.

[0162] It should be noted that the memory 730 in different computing devices 700 in the computing device cluster may store different instructions for executing partial functions of the computing device 700. That is, the instructions stored in the memory 730 in different computing devices 700 may implement the functions of one or more of the extraction module 110, transfer module 120, correction module 130, conversion module 140, training module 150, and update module 160 described above.

[0163] The present application also provides a computer program product comprising instructions. The computer program product may be software or a program product comprising instructions that can be executed on a computing device or stored in any available medium. When the computer program product is executed on at least one computing device, the at least one computing device executes the data optimization method.

[0164] The present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the data optimization method.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the protection scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A data optimization method, characterized in that: include: Extracting features of physical property parameters of a first passive component to obtain characteristic parameters of the first passive component; the first passive component is a passive component in a radio frequency circuit, or passive components of the same type; the characteristic parameters include poles, residues, and constant terms; Calculating a scattering parameter frequency response curve of the first passive component based on different frequencies within a set frequency range, the poles, the residues, and the constant term; the scattering parameter frequency response curve represents a relationship between different frequencies and scattering parameters; Passivity correction is performed on the scattering parameter frequency response curve of the first passive component to obtain a corrected scattering parameter frequency response curve of the first passive component.

2. The method according to claim 1, characterized in that The method further comprises: Using a parameter conversion formula, the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component are converted into performance parameters other than the scattering parameters in the first passive component; the performance parameters include one or more of the scattering parameters, admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters and quality factor parameters.

3. The method according to claim 1 or 2, characterized in that The calculating of the scattering parameter frequency response curve of the first passive component according to different frequencies within the set frequency range, the poles, the residues, and the constant term specifically includes: Calculating scattering parameters of a plurality of the first passive components by using the poles, the residues, and the constant terms using a differentiable transfer function; The differentiable transfer function is used to calculate a scattering parameter frequency response curve of the first passive component by combining different frequencies within the set frequency range with the scattering parameters of the plurality of first passive components.

4. The method according to any one of claims 1 to 3, characterized in that The extracting features of the physical property parameters of the first passive component to obtain the feature parameters of the first passive component specifically includes: Sampling each parameter of the physical property parameter of the first passive component to obtain a plurality of first samples; each first sample includes a value of each parameter; Randomly sampling the plurality of first samples to obtain a sample data set of the first passive component; the sample data set includes a portion of the plurality of first samples; Feature extraction is performed on the sample data set to obtain characteristic parameters of the first passive component.

5. The method according to any one of claims 1 to 4, characterized in that Before extracting the characteristics of the physical property parameters of the first passive component to obtain the characteristic parameters of the first passive component, the method further includes: uniformly sampling each parameter of the physical property parameter of the first passive component to obtain a plurality of second samples; each second sample includes a value of each parameter; Randomly sampling the plurality of second samples to obtain a training data set of the first passive component; the training data set includes a portion of the plurality of second samples; The training data set is input into a neural network for training to obtain training parameters of the first passive component; the training parameters are used to perform feature extraction on the physical property parameters of the first passive component through the neural network.

6. The method according to claim 5, characterized in that The method further comprises: Second samples other than the training data set among the plurality of second samples and the training parameters of the first passive component are input into the neural network to obtain test scattering parameters of the first passive component; the test scattering parameters of the first passive component are used to compare with reference scattering parameters to detect performance of the training parameters of the first passive component.

7. The method according to claim 5 or 6, characterized in that The method further comprises: Obtaining a loss function value according to the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component and a reference scattering parameter; The loss function value is used to optimize the training parameters of the first passive component by using an optimization algorithm to obtain optimized training parameters of the first passive component.

8. The method according to any one of claims 1 to 7, characterized in that The method further comprises: receiving physical property parameters of a second passive component; the second passive component and the first passive component are of the same type but manufactured using different processes; Sampling the training data set of the first passive component to obtain a plurality of third samples; Replacing the physical property parameters in the plurality of third samples with the physical property parameters of the second passive component to obtain a training data set of the second passive component; The training data set of the second passive component is input into the neural network for training to obtain the training parameters of the second passive component; the training parameters of the second passive component are used to extract features of the physical property parameters of the second passive component through the neural network.

9. A data optimization system, characterized in that: include: an extraction module, configured to perform feature extraction on physical property parameters of a first passive component to obtain characteristic parameters of the first passive component; the first passive component is a passive component in a radio frequency circuit, or passive components of the same type; the characteristic parameters include poles, residues, and constant terms; a transfer module, configured to calculate a scattering parameter frequency response curve of the first passive component based on different frequencies within a set frequency range, the poles, the residues, and the constant term; the scattering parameter frequency response curve represents a relationship between different frequencies and scattering parameters; The correction module is used to perform passivity correction on the scattering parameter frequency response curve of the first passive component to obtain a corrected scattering parameter frequency response curve of the first passive component.

10. The device according to claim 9, characterized in that Also includes: a conversion module, configured to convert, by using a parameter conversion formula, the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component into performance parameters other than the scattering parameters of the first passive component; the performance parameters comprising one or more of the scattering parameters, admittance parameters, impedance parameters, capacitance parameters, inductance parameters, resistance parameters, and quality factor parameters.

11. The device according to claim 9 or 10, characterized in that The transfer module is specifically configured to calculate the scattering parameters of the plurality of first passive components by using the poles, the residues and the constant terms using a differentiable transfer function; The differentiable transfer function is used to calculate a scattering parameter frequency response curve of the first passive component by combining different frequencies within the set frequency range with the scattering parameters of the plurality of first passive components.

12. The device according to any one of claims 9 to 11, characterized in that: The extraction module is specifically configured to sample each parameter of the physical property parameter of the first passive component to obtain a plurality of first samples; each first sample includes a value of each parameter; Randomly sampling the plurality of first samples to obtain a sample data set of the first passive component; the sample data set includes a portion of the plurality of first samples; Feature extraction is performed on the sample data set to obtain characteristic parameters of the first passive component.

13. The device according to any one of claims 9 to 12, characterized in that: Also includes: A training module, configured to uniformly sample each parameter of the physical property parameters of the first passive component to obtain a plurality of second samples; each second sample includes a value of each parameter; Randomly sampling the plurality of second samples to obtain a training data set of the first passive component; the training data set includes a portion of the plurality of second samples; Inputting the training data set into a neural network for training to obtain training parameters of the first passive component; The training parameters are used to perform feature extraction on the physical property parameters of the first passive component through the neural network.

14. The device according to claim 13, characterized in that The training module is further configured to input second samples other than the training data set among the plurality of second samples and the training parameters of the first passive component into the neural network to obtain test scattering parameters of the first passive component; the test scattering parameters of the first passive component are used to compare with reference scattering parameters to detect performance of the training parameters of the first passive component.

15. The device according to claim 13 or 14, characterized in that The training module is further configured to obtain a loss function value based on the scattering parameters in the corrected scattering parameter frequency response curve of the first passive component and a reference scattering parameter; The loss function value is used to optimize the training parameters of the first passive component by using an optimization algorithm to obtain optimized training parameters of the first passive component.

16. The device according to any one of claims 9 to 15, characterized in that Also includes: An updating module, configured to receive physical property parameters of a second passive component; the second passive component and the first passive component are of the same type but have different manufacturing processes; Sampling the training data set of the first passive component to obtain a plurality of third samples; Replacing the physical property parameters in the plurality of third samples with the physical property parameters of the second passive component to obtain a training data set of the second passive component; The training data set of the second passive component is input into the neural network for training to obtain the training parameters of the second passive component; the training parameters of the second passive component are used to extract features of the physical property parameters of the second passive component through the neural network.

17. A computing device cluster, characterized in that: comprising at least one computing device, each computing device including a processor and a memory; The processor of the at least one computing device is configured to execute instructions stored in a memory of the at least one computing device, so that the computing device cluster executes the method according to any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that The method comprises computer program instructions, and when the computer program instructions are executed by a computing device, the computing device performs the method according to any one of claims 1 to 8.

19. A computer program product comprising instructions, characterized in that The computer program product stores instructions, which, when executed by a computing device, enable the computing device to implement the method according to any one of claims 1 to 8.

Images