Electromagnetic shielding compensation method and system for FPC high performance computing chip interface

Abstract

The invention relates to the technical field of electromagnetic shielding compensation, and discloses an electromagnetic shielding compensation method and system for an FPC high-performance computing chip interface, wherein the method comprises: acquiring impedance signals of the FPC high-performance computing chip interface to obtain time domain signals and second-order rate of change curves, wherein the time domain signals are acquired by applying an AC excitation signal between a signal layer and a ground layer; extracting features of the time domain signals and the second-order rate of change curves to obtain feature vectors; inputting the feature vectors into a support vector machine model to classify and identify electromagnetic interference to obtain electromagnetic interference mode types; performing weighted fusion analysis on the time domain signals to obtain electromagnetic interference state parameters; constructing a hierarchical compensation controller, and adjusting the impedance parameters of the shielding layer through an adaptive backstepping control algorithm to obtain compensation control parameters. The method realizes precise adjustment of the impedance parameters of the shielding layer and ensures the stability and effectiveness of electromagnetic shielding compensation.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic shielding compensation, and in particular to an electromagnetic shielding compensation method and system for an FPC high-performance computing chip interface. Background Art

[0002] During high-speed data transmission, the FPC interface is susceptible to various electromagnetic interferences such as power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference, resulting in a decrease in signal transmission quality and affecting the stability and reliability of the autonomous driving system.

[0003] The existing FPC interface electromagnetic shielding compensation method mainly adopts a distributed electronic and electrical architecture to achieve electromagnetic interference suppression by setting a multi-layer shielding structure on the FPC. This method not only increases the thickness and cost of the PCB, but also lacks the ability to accurately identify interference characteristics and dynamically compensate for them, making it difficult to deal with multi-source interference problems in complex electromagnetic environments, and the shielding effect is difficult to meet the strict requirements of L5 autonomous driving. Summary of the invention

[0004] The present invention provides an electromagnetic shielding compensation method and system for an FPC high-performance computing chip interface, which realizes accurate adjustment of shielding layer impedance parameters and ensures the stability and effectiveness of electromagnetic shielding compensation.

[0005] In a first aspect, the present invention provides an electromagnetic shielding compensation method for an FPC high-performance computing chip interface, the electromagnetic shielding compensation method for an FPC high-performance computing chip interface comprising:

[0006] The impedance signal is collected on the FPC high-performance computing chip interface to obtain a time domain signal and a second-order rate of change curve, wherein the time domain signal is collected by applying an AC excitation signal between the signal layer and the ground layer;

[0007] Performing feature extraction on the time domain signal and the second-order rate of change curve to obtain time domain feature parameters and frequency domain feature parameters, and combining the time domain feature parameters and the frequency domain feature parameters into a feature vector;

[0008] Inputting the feature vector into a support vector machine model to classify and identify electromagnetic interference to obtain electromagnetic interference pattern types, wherein the electromagnetic interference pattern types include power frequency interference, switching interference, capacitive coupling interference, and inductive coupling interference;

[0009] A dual-frequency extended state observer is constructed according to the time domain signal and the electromagnetic interference mode type, and a weighted fusion analysis is performed on the time domain signal through a high-frequency observer and a low-frequency observer to obtain an electromagnetic interference state parameter;

[0010] A layered compensation controller is constructed based on the electromagnetic interference mode type and the electromagnetic interference state parameter, and the shielding layer impedance parameter is adjusted through an adaptive backstepping control algorithm to obtain a compensation control parameter.

[0011] In a second aspect, the present invention provides an electromagnetic shielding compensation system for an FPC high-performance computing chip interface, the electromagnetic shielding compensation system for an FPC high-performance computing chip interface comprising:

[0012] An acquisition module is used to acquire impedance signals from the FPC high-performance computing chip interface to obtain a time domain signal and a second-order rate of change curve, wherein the time domain signal is acquired by applying an AC excitation signal between the signal layer and the ground layer;

[0013] An extraction module, used for performing feature extraction on the time domain signal and the second-order rate of change curve to obtain time domain feature parameters and frequency domain feature parameters, and combining the time domain feature parameters and the frequency domain feature parameters into a feature vector;

[0014] An identification module, used for inputting the feature vector into a support vector machine model to classify and identify electromagnetic interference, and obtaining electromagnetic interference pattern types, wherein the electromagnetic interference pattern types include power frequency interference, switching interference, capacitive coupling interference, and inductive coupling interference;

[0015] An analysis module, configured to construct a dual-frequency extended state observer according to the time domain signal and the electromagnetic interference mode type, and perform weighted fusion analysis on the time domain signal through a high frequency observer and a low frequency observer to obtain electromagnetic interference state parameters;

[0016] The regulating module is used to construct a layered compensation controller based on the electromagnetic interference mode type and the electromagnetic interference state parameter, and to adjust the shielding layer impedance parameter through an adaptive backstepping control algorithm to obtain a compensation control parameter.

[0017] The third aspect of the present invention provides an electromagnetic shielding compensation device for an FPC high-performance computing chip interface, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the electromagnetic shielding compensation device for the FPC high-performance computing chip interface to execute the above-mentioned electromagnetic shielding compensation method for the FPC high-performance computing chip interface.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, which, when executed on a computer, enable the computer to execute the electromagnetic shielding compensation method for the FPC high-performance computing chip interface.

[0019] In the technical solution provided by the present invention, an AC excitation signal is applied between the signal layer and the ground layer to collect impedance signals, and the feature extraction of the time domain signal and the second-order rate of change curve is combined to achieve accurate characterization of the electromagnetic characteristics of the FPC interface, and a support vector machine model is used to classify and identify electromagnetic interference on the feature vector, accurately distinguishing various interference types such as power frequency interference, switch interference, capacitive coupling interference and inductive coupling interference, thereby improving the accuracy of electromagnetic interference identification. Based on the design of a dual-frequency extended state observer, through the weighted fusion analysis of high-frequency observers and low-frequency observers, real-time and accurate observation of the electromagnetic interference state is achieved, effectively avoiding the problem of observation noise amplification. A hierarchical compensation controller structure is adopted, combined with an adaptive backstepping control algorithm, to achieve accurate adjustment of the impedance parameters of the shielding layer, ensuring the stability and effectiveness of electromagnetic shielding compensation. A dual network structure of an evaluation network and an action network is introduced for reinforcement learning optimization, and through online learning and parameter optimization, continuous improvement of the shielding compensation effect is achieved, so that the system has adaptive optimization capabilities.

[0020] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of an embodiment of an electromagnetic shielding compensation method for an FPC high-performance computing chip interface in an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of an embodiment of an electromagnetic shielding compensation system of an FPC high-performance computing chip interface in an embodiment of the present invention;

[0024] Figure 3 It is a schematic diagram of an embodiment of an electromagnetic shielding compensation device of an FPC high-performance computing chip interface in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0027] To facilitate understanding of this embodiment, firstly, a method for electromagnetic shielding compensation of an FPC high-performance computing chip interface disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps:

[0028] 101. Collect impedance signals from the FPC high-performance computing chip interface to obtain time domain signals and second-order rate of change curves. The time domain signals are collected by applying an AC excitation signal between the signal layer and the ground layer.

[0029] It is understandable that the execution subject of the present invention may be an electromagnetic shielding compensation system of an FPC high-performance computing chip interface, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as an execution subject.

[0030] Specifically, a 1V AC excitation signal is applied between the signal layer and the ground layer of the FPC high-performance computing chip interface to stimulate the current response signal of the FPC interface, and the response signal is collected by an impedance analyzer to obtain the current response signal. The sampling frequency of 1MHz is used for collection, and the sampling time is set to 100ms to fully capture the detailed features in the current response signal and reduce the information loss caused by insufficient sampling. The collected original impedance data contains the current response of the FPC interface under electromagnetic excitation. Based on the ratio of the AC excitation signal to the current response signal, an impedance value sequence is obtained to reflect the impedance characteristics of the FPC interface under specific electromagnetic excitation. The impedance value sequence is subjected to high-pass filtering. The Butterworth high-pass filter is used to filter out interference components below 100Hz. The Butterworth filter has good frequency response smoothing characteristics during the filtering process, does not cause excessive distortion to the high-frequency part, and can better retain the useful components in the impedance signal. Through the processing of the high-pass filter, the low-frequency interference is effectively filtered out to obtain the filtered impedance signal. The filtered impedance signal is subjected to 50Hz power frequency interference elimination processing. 50Hz power frequency interference mainly comes from the power supply network. It is a common environmental interference source and will have adverse effects on signal analysis and electromagnetic characteristics extraction. The 50Hz power frequency interference can be effectively eliminated through the interference suppression algorithm, improving the quality of the impedance signal and obtaining a cleaner time domain signal. The second-order differential operation is performed on the time domain signal to obtain the second-order rate of change curve, revealing the acceleration characteristics of the signal on the time axis.

[0031] 102. Perform feature extraction on the time domain signal and the second-order rate of change curve to obtain time domain feature parameters and frequency domain feature parameters, and form a feature vector with the time domain feature parameters and the frequency domain feature parameters;

[0032] Specifically, the statistical parameters of the time domain signal are calculated to obtain the time domain characteristic parameters, including mean, variance, skewness, kurtosis, etc., which describe the overall trend and change characteristics of the time domain signal and reflect the response characteristics of the signal in the electromagnetic interference environment. The second-order rate of change curve is decomposed by wavelet packet to decompose the original signal into multiple frequency band sub-signals. Wavelet packet decomposition has the ability of time-frequency localization and can analyze signals at different frequency resolutions, which is suitable for processing complex electromagnetic interference signals. Through wavelet packet decomposition, the original second-order rate of change curve is decomposed into several frequency band sub-signals, which cover the interference characteristics in different frequency bands. Energy calculation is performed on multiple frequency band sub-signals to obtain the energy value of each frequency band. The frequency band energy value is the power characteristic of the signal in each frequency band, reflecting the contribution of different frequency components to the overall signal. The frequency band energy value is normalized to map all the frequency band energy values ​​to the same numerical range to obtain the frequency domain characteristic parameters. The time domain characteristic parameters and the frequency domain characteristic parameters are spliced ​​and combined to obtain the original feature vector. The original feature vector is subjected to dimensionality reduction processing. The principal component analysis model is used to decompose the eigenvalues, calculate the eigencontribution rate, and obtain the principal component mapping matrix. The principal component mapping matrix can project the original high-dimensional eigenvector into a low-dimensional space while retaining as much original data information as possible. The principal component mapping matrix is ​​used to reduce the dimension of the original eigenvector to obtain the reduced-dimensional feature data. The reduced-dimensional feature data is standardized to adjust the feature data to the same scale range to obtain the eigenvector.

[0033] 103. Input the feature vector into a support vector machine model to classify and identify electromagnetic interference, and obtain electromagnetic interference pattern types, which include power frequency interference, switch interference, capacitive coupling interference, and inductive coupling interference;

[0034] Specifically, the feature vector is input into the support vector machine model, and the inner product mapping of the feature space is calculated by the radial basis kernel function of the support vector machine model, so that the original feature vector is mapped to the high-dimensional feature space to obtain the high-dimensional feature mapping data. The radial basis kernel function can transform the nonlinear problem into a linearly separable problem, so that the model can more accurately classify and identify the electromagnetic interference in the high-dimensional space. The high-dimensional feature mapping data is input into the first hidden layer of the support vector machine model for power frequency interference feature extraction. The support vector machine determines the possibility that the input data belongs to power frequency interference by calculating the distance between the support vector and the input data, and obtains the probability value of the power frequency interference. The high-dimensional feature mapping data is input into the second hidden layer of the support vector machine model to extract the switch interference feature. In a similar way, the SVM model calculates the distance between the support vector and the input data to obtain the probability value of the switch interference. Since the characteristics of the switch interference usually have the characteristics of transient mutation, the feature extraction of the second hidden layer can effectively identify whether the data contains the components of the switch interference, and further estimate its probability size. The high-dimensional feature mapping data is input into the third hidden layer of the SVM model to extract the capacitive coupling interference feature. Capacitive coupling interference is manifested as mutual influence caused by parasitic capacitance on the signal line. By calculating the distance between the support vector in the third hidden layer and the input data, the characteristics of capacitive coupling interference are effectively captured, and the corresponding interference probability value is obtained. The high-dimensional feature mapping data is input into the fourth hidden layer of the SVM model to extract the characteristics of inductive coupling interference. Inductive coupling interference is interference caused by the interaction of magnetic fields between adjacent conductors. Through the fourth hidden layer, the SVM model calculates the distance between the support vector and the input data to obtain the probability value of inductive coupling interference. The maximum values ​​of the power frequency interference probability value, the switch interference probability value, the capacitive coupling interference probability value and the inductive coupling interference probability value are compared, and the interference type corresponding to the maximum probability value is determined as the electromagnetic interference mode type of the current input data.

[0035] 104. Construct a dual-frequency extended state observer according to the time domain signal and the electromagnetic interference mode type, perform weighted fusion analysis on the time domain signal through the high frequency observer and the low frequency observer, and obtain the electromagnetic interference state parameter;

[0036] Specifically, the observer bandwidth is calculated for the electromagnetic interference mode type. By analyzing the frequency characteristics of power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference, the bandwidth allocation scheme of the observer is obtained. According to the frequency characteristics of these interferences, the bandwidth of the observer is reasonably allocated so that the observer can accurately capture and process the corresponding interference signal. The bandwidth parameters of the high-frequency observer and the low-frequency observer are set according to the observer bandwidth allocation scheme. The bandwidth of the high-frequency observer is set to 1kHz to ensure that it can effectively capture the characteristics of the high-frequency interference signal, and the bandwidth of the low-frequency observer is set to 100Hz so that it can focus on capturing the power frequency and low-frequency interference characteristics. The bandwidth setting ensures the separation and independent observation of high-frequency and low-frequency interference characteristics, thereby improving the accuracy of interference analysis and the response performance of the observer. According to the mode type of electromagnetic interference, the time domain signal is allocated to the high-frequency observer and the low-frequency observer for processing. The input time domain signal is processed in parallel by the high-frequency observer and the low-frequency observer to obtain dual-frequency observation data, thereby effectively separating and capturing the characteristics of different interference components. The output data of the high-frequency observer is weighted, and the weighted high-frequency data is obtained by multiplying the high-frequency observation data with the first weight coefficient. The output data of the low-frequency observer is weighted, and the weighted low-frequency data is obtained by multiplying the low-frequency observation data with the second weight coefficient. The influence of the high-frequency and low-frequency components is adjusted according to the actual interference situation to obtain a more accurate comprehensive interference feature. The high-frequency weighted data and the low-frequency weighted data are time-series fused to obtain the fused interference data. The high-frequency and low-frequency weighted data are combined on the time axis to comprehensively reflect the interaction and cumulative effect between different frequency components. The state parameters of the fused interference data are analyzed to obtain the interference feature quantity. The interference feature quantity is a representation of the characteristics of electromagnetic interference in the entire time domain and frequency domain, reflecting the relationship between different frequency components and their influence on the FPC interface signal. The state vector is constructed based on the interference feature quantity. The state vector is a unified description of the interference characteristics in time and frequency. By analyzing the state vector, the change rate and cumulative effect of the interference quantity are calculated to obtain the state parameters of the electromagnetic interference. The rate of change of the interference reflects the dynamic characteristics of electromagnetic interference, especially the quantification of the severity of interference at different times; while the cumulative effect reflects the overall impact of interference on the system over a period of time.

[0037] 105. A hierarchical compensation controller is constructed based on the electromagnetic interference mode type and the electromagnetic interference state parameters, and the shielding layer impedance parameters are adjusted through an adaptive backstepping control algorithm to obtain compensation control parameters.

[0038] Specifically, a suitable basic shielding structure is selected according to the type of electromagnetic interference mode. Different interference mode types correspond to different shielding requirements. Therefore, it is necessary to flexibly set the thickness of the shielding layer, the type of material, and the interlayer connection method to obtain the first layer compensation parameters suitable for the current electromagnetic environment. For example, when facing power frequency interference, materials with high conductivity and good shielding performance are selected to reduce the impact of low-frequency interference; for high-frequency interference, the thickness of the shielding layer and the interlayer connection method need to be optimized to ensure that high-frequency signals can be effectively absorbed and reflected. The Lyapunov function is constructed based on the state parameters of electromagnetic interference to evaluate the stability of the system. The Lyapunov function is a mathematical tool for analyzing the stability of dynamic systems. By constructing a quadratic function of the state variables, the stability criterion of the electromagnetic shielding system is obtained to ensure that the design of the compensation control will not destroy the stability of the system. The electromagnetic interference state parameters are the main factors affecting the stability of the system. Through the construction of the Lyapunov function, these parameters are associated with the state variables of the system to ensure that the design of the controller is stable and effective. On this basis, the first layer compensation parameters are backstepped and calculated to recursively solve the control law step by step and decompose the control gain matrix. Backstepping control is a hierarchical control algorithm. By decomposing complex control objectives layer by layer, the high-dimensional system control problem is decomposed into multiple low-dimensional sub-problems, so as to gradually approach the final control objective. In the calculation process of backstepping control, the control law is recursively solved to obtain the control gain matrix, which is used to describe the relationship between different state variables and the control effect. The state parameters of electromagnetic interference are input into the control gain matrix, and the initial value of the compensation amount is obtained through matrix multiplication. According to the initial value of the compensation amount, the impedance matching calculation is performed to make the impedance of the shielding layer match the impedance of the electromagnetic interference source as much as possible, thereby minimizing the generation of reflected interference and energy leakage. By adjusting the matching degree between the shielding layer impedance and the interference source impedance, the impedance adjustment coefficient is obtained, which is used to guide the real-time adjustment of the shielding layer to ensure effective shielding under different interference environments. The impedance adjustment coefficient is optimized online, and the shielding effect is improved by adjusting the gain parameters of the backstepping controller to obtain the second-layer compensation parameters. The online optimization process can respond to changes in external interference in real time, ensure that the compensation controller can be flexibly adjusted, and improve the real-time and accuracy of the compensation effect. The first-layer compensation parameters and the second-layer compensation parameters are hierarchically combined, and a complete compensation strategy sequence is obtained by constructing a hierarchical compensation controller. Hierarchical compensation control can achieve comprehensive suppression of electromagnetic interference through different levels of control measures, and each layer of control measures is independent and cooperative, thereby improving the overall compensation control effect. After constructing the compensation strategy sequence, the execution order is sorted to ensure that each layer of control measures can be executed in the correct order to avoid poor interference suppression effect caused by improper execution order.The compensation control parameters are obtained by reasonably sorting the execution order of the compensation strategy sequence.

[0039] The compensation control parameters are input into the 16-node input layer of the evaluation network, and each compensation parameter is nonlinearly transformed and mapped through the ReLU activation function to obtain a 16-dimensional feature vector. The ReLU activation function is used to ensure that the feature data can retain the positive characteristics of the parameters and suppress negative values ​​after mapping, thereby improving the representation ability of the network and ensuring that the feature extraction of the compensation control parameters can fully reflect the actual effect of electromagnetic shielding. The 16-dimensional feature vector is input into the 32-node hidden layer of the evaluation network, and the weighted sum and bias of each node are calculated through the hyperbolic tangent (tanh) function to obtain the 32-dimensional hidden layer features. The advantage of the tanh function is that its output range is -1 to 1, which can represent the relative size of the negative and positive values ​​of the input features, ensuring that the hidden layer has a balanced expression ability when extracting features. The 32 nodes of the hidden layer enable the network to perform more complex combinations and abstractions on the input features and extract higher-level feature information. The 32-dimensional hidden layer features are input into the 1-node output layer of the evaluation network, and the features are compressed to the range of 0 to 1 through the Sigmoid activation function to obtain the final compensation effect score. The compression characteristics of the Sigmoid function allow the compensation effect to be quantified as a probability value, which is convenient for evaluating the effectiveness of the compensation strategy. A score close to 1 indicates a good compensation effect, while a score close to 0 indicates a poor compensation effect. At the same time, the compensation control parameters are input into the Actor layer of the action network, and the compensation strategy features are extracted through a three-layer fully connected network structure to obtain an action strategy vector. The three-layer fully connected structure ensures that the network can extract the feature information of the compensation parameters at different abstract levels to form the final action strategy vector, which is used to guide the subsequent compensation strategy adjustment to ensure that the shielding compensation can be adaptively optimized in the electromagnetic interference environment. The compensation effect score and the action strategy vector are input into the Critic layer of the action network, and the state-action value function is calculated through the temporal difference algorithm to obtain the compensation adjustment strategy. The temporal difference algorithm is a strategy evaluation method in reinforcement learning. It effectively judges the quality of the compensation strategy by evaluating the value function relationship between the current compensation state and the action strategy taken. This method can continuously adjust the compensation strategy according to the feedback information of the compensation effect to gradually approach the optimal shielding effect. The parameters of the compensation adjustment strategy are updated. The adjustment direction and step size of the compensation parameters are calculated by the policy gradient method to ensure that the adjustment of the parameters can be carried out in the direction of improving the compensation effect. The policy gradient method calculates the parameter update amount by evaluating the impact of each parameter on the final compensation effect. The calculated compensation parameter update amount is superimposed on the compensation control parameter, and the optimized compensation parameter is obtained through the cumulative update of the parameter. This process is a gradual optimization process. By continuously adjusting the parameters in small increments, the compensation effect can gradually approach the optimal state. The performance of the optimized compensation parameters is evaluated. By setting the performance indicator threshold of 0.95, the convergence of the compensation parameters is judged.If the compensation effect score of the compensation parameter can stably exceed the threshold of 0.95 after multiple updates, it is considered that the parameter has converged and the compensation effect has reached the expected standard, thus determining that the compensation parameter is the optimal compensation parameter; if this threshold is not reached, it is necessary to continue updating and optimizing the parameter until the compensation effect meets the requirements.

[0040] In the embodiment of the present invention, an AC excitation signal is applied between the signal layer and the ground layer to collect impedance signals, and the feature extraction of the time domain signal and the second-order rate of change curve is combined to achieve accurate characterization of the electromagnetic characteristics of the FPC interface, and a support vector machine model is used to classify and identify electromagnetic interference on the feature vector, accurately distinguishing various interference types such as power frequency interference, switch interference, capacitive coupling interference and inductive coupling interference, thereby improving the accuracy of electromagnetic interference identification. Based on the design of a dual-frequency extended state observer, through the weighted fusion analysis of high-frequency observers and low-frequency observers, real-time and accurate observation of the electromagnetic interference state is achieved, effectively avoiding the problem of observation noise amplification. A hierarchical compensation controller structure is adopted, combined with an adaptive backstepping control algorithm, to achieve accurate adjustment of the impedance parameters of the shielding layer, ensuring the stability and effectiveness of electromagnetic shielding compensation. A dual network structure of an evaluation network and an action network is introduced for reinforcement learning optimization, and through online learning and parameter optimization, continuous improvement of the shielding compensation effect is achieved, so that the system has adaptive optimization capabilities.

[0041] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0042] A 1V AC excitation signal is applied between the signal layer and the ground layer of the FPC high-performance computing chip interface, and an impedance analyzer is used to collect the current response signal;

[0043] The current response signal is sampled at a frequency of 1 MHz, and the sampling time is set to 100 ms to obtain the original impedance data;

[0044] An impedance value sequence is obtained by performing calculation based on the ratio of the AC excitation signal to the current response signal;

[0045] The impedance value sequence is input into a Butterworth high-pass filter to filter out interference components below 100 Hz, thereby obtaining a filtered impedance signal;

[0046] The filtered impedance signal is processed to eliminate the 50 Hz power frequency interference to obtain a time domain signal, and a second-order differential operation is performed on the time domain signal to obtain a second-order rate of change curve.

[0047] Specifically, a 1V AC excitation signal is applied between the signal layer and the ground layer of the FPC interface, and the system is measured by an impedance analyzer to obtain a current response signal. The AC excitation signal effectively excites the impedance characteristics in the interface, so that the current response signal can reflect the characteristics and dynamic response of the interface under different frequency interference. The excitation signal voltage is set to 1V to ensure that the signal amplitude is moderate, which can effectively drive the parasitic parameters in the interface without introducing too many nonlinear effects. The current response signal is collected by an impedance analyzer. The use of the impedance analyzer can capture the response characteristics of the current with high precision and ensure the accuracy and precision of the data. The current response signal is collected at a sampling frequency of 1MHz, and the sampling time is set to 100ms to obtain the original impedance data. The sampling frequency of 1MHz is selected to ensure that the high-frequency components in the signal will not be lost, which meets the requirement of "the sampling frequency is greater than twice the highest frequency of the signal" in the sampling theorem. At the same time, the sampling time is set to 100ms to obtain sufficient data volume, so that frequency analysis and feature extraction can be better performed. The original impedance data contains the dynamic current response of the FPC interface under the AC excitation signal, which is an important basis for subsequent calculation of impedance value and removal of interference. The impedance value sequence is obtained by calculating the ratio of the AC excitation signal to the current response signal. ) is expressed as the excitation voltage ( ) and current response ( ) ratio:

[0048] ;

[0049] in, represents the applied AC excitation signal voltage, represents the current response signal, Represents time. The impedance value sequence is calculated by the above formula, which reflects the impedance change of the FPC interface under AC excitation at different time points. Impedance is a key parameter to characterize the electrical characteristics of the circuit, and its change directly reflects the change of the electrical characteristics of the system under the influence of interference. In order to remove the low-frequency interference components in the signal, especially the interference caused by environmental noise or other low-frequency interference sources, a Butterworth high-pass filter is used to filter out the interference components below 100Hz. The Butterworth filter is characterized by a smooth frequency response and no obvious ripples in the passband, so that it can effectively filter out unnecessary low-frequency signals while retaining useful signal components. Assume that the transfer function of the filter is:

[0050] ;

[0051] in, represents the transfer function of the filter, is a complex frequency variable, is the cut-off frequency, set rad / s, is the order of the filter. By selecting a suitable cutoff frequency (i.e. 100Hz), low-frequency interference can be effectively filtered out, and the high-frequency characteristic components in the signal can be retained to obtain the filtered impedance signal. The filtered impedance signal is processed to eliminate the 50Hz power frequency interference. Power frequency interference is a common interference source in power systems, usually caused by the power supply system of the power grid, and frequently appears in various signal processing at a frequency of 50Hz. In order to eliminate power frequency interference, a notch filter is used, and its transfer function is:

[0052] ;

[0053] in, is the angular frequency of the power frequency, is the damping ratio, which is used to control the width of the notch. The notch filter effectively removes the 50Hz interference component to obtain a purer time domain signal, reflecting the impedance characteristics of the FPC interface in other frequency ranges. Perform a second-order differential operation on the time domain signal to obtain a second-order rate of change curve. The second-order differential reflects the acceleration change of the signal and is a measure of the rate of change of the signal. The second-order rate of change ( ) by analyzing the time domain signal ( ) and perform a second derivative operation to obtain:

[0054] ;

[0055] The second-order rate of change curve can reveal the trend and intensity of the signal change on the time axis, which is helpful for analyzing the response characteristics of the system under different interference conditions. For example, in the case of electromagnetic interference, the second-order rate of change of the signal may have a spike, reflecting that the system is subject to strong transient interference. For a normal stable state, the second-order rate of change will maintain a small value, indicating that the impedance of the system changes relatively slowly.

[0056] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0057] Calculate the statistical parameters of the time domain signal to obtain the time domain characteristic parameters, and perform wavelet packet decomposition on the second-order rate of change curve to obtain multiple frequency band sub-signals;

[0058] Performing energy calculation on multiple frequency band sub-signals to obtain frequency band energy values, and normalizing the frequency band energy values ​​to obtain frequency domain feature parameters;

[0059] The time domain feature parameters and the frequency domain feature parameters are combined to obtain the original feature vector;

[0060] The original eigenvector is input into the principal component analysis model, and the characteristic contribution rate is calculated by eigenvalue decomposition to obtain the principal component mapping matrix;

[0061] The original eigenvector is transformed into a dimension-reduced one through the principal component mapping matrix to obtain the eigendata after dimensionality reduction, and the eigendata after dimensionality reduction is standardized to obtain the eigenvector.

[0062] Specifically, the statistical characteristics of the time domain signal are extracted to obtain the time domain characteristic parameters. The time domain signal reflects the response information of the system in the time domain, and its characteristic parameters are extracted through a variety of statistical methods, including mean, variance, skewness, kurtosis and other statistics. The mean is calculated and expressed as:

[0063] ;

[0064] in, represents the mean value of the time domain signal, is the total number of signal samples, Indicates The mean reflects the overall deviation of the signal. The variance is used to measure the discreteness of the signal, and the calculation formula is:

[0065] ;

[0066] in, is the variance, which indicates the dispersion of the signal relative to the mean. Skewness and kurtosis are used to reflect the symmetry and peak characteristics of the signal. The skewness ( ) and kurtosis ( ) are:

[0067] ;

[0068] By calculating these statistical features, the time domain characteristic parameters are obtained, which can effectively describe the changing characteristics of the time domain signal under different electromagnetic interference conditions. Wavelet packet decomposition is performed on the second-order rate of change curve to obtain multiple frequency band sub-signals. Wavelet packet decomposition is a signal decomposition method that can perform multi-level frequency band division on the signal and decompose the signal into multiple sub-signals with different frequency characteristics. Assume that the second-order rate of change curve is , then it is decomposed into multiple frequency band sub-signals by wavelet packet decomposition, and each sub-signal is Indicates that Indicates the level of frequency bands. Perform energy calculation on the frequency band sub-signals to obtain the energy value of each frequency band. Assume that the frequency band sub-signal is , then its energy value is expressed by the formula:

[0069] ;

[0070] in, Indicates The energy of each frequency band sub-signal is calculated to evaluate the contribution of different frequency components to the entire signal. In order to avoid the influence of large energy differences between different frequency bands on subsequent feature extraction, the frequency band energy values ​​are normalized to obtain frequency domain feature parameters. The normalization process is implemented by the following formula:

[0071] ;

[0072] in, is the normalized frequency band energy value, is the total number of sub-signals in the frequency band. The time domain feature parameters are combined with the frequency domain feature parameters to obtain the original feature vector. The original feature vector is input into the principal component analysis model for dimensionality reduction. Principal component analysis is a commonly used dimensionality reduction method. By performing eigenvalue decomposition on the covariance matrix of the feature vector, the main direction of data change is found, thereby retaining the main information of the data and reducing the dimension of the data. Assume that the covariance matrix of the original feature vector is , then its eigenvalue decomposition is expressed as:

[0073] ;

[0074] in, The covariance matrix is The matrix composed of the eigenvectors of is the corresponding eigenvalue diagonal matrix. The principal component mapping matrix is ​​constructed by selecting the eigenvector with the largest eigenvalue , perform dimension reduction transformation on the original feature vector. The feature vector after dimension reduction It is expressed by the following formula:

[0075] ;

[0076] in, is the principal component mapping matrix, is the feature data after dimensionality reduction. In this way, the redundant information in the original high-dimensional feature vector is effectively removed, and the features that contribute most to classification in the data are retained, thereby reducing computational complexity and improving processing efficiency. The feature data after dimensionality reduction is standardized and adjusted to a standard normal distribution with a mean of 0 and a standard deviation of 1, so that the subsequent classification or regression model can better process the feature data and avoid training difficulties caused by different feature scales, and obtain the feature vector.

[0077] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0078] The feature vector is input into the support vector machine model, and the inner product mapping of the feature space is calculated by the radial basis kernel function of the support vector machine model to obtain high-dimensional feature mapping data;

[0079] The high-dimensional feature mapping data is input into the first hidden layer of the support vector machine model to extract the power frequency interference features, and the power frequency interference probability value is obtained by calculating the support vector distance;

[0080] The high-dimensional feature mapping data is input into the second hidden layer of the support vector machine model to extract the switch interference features, and the switch interference probability value is obtained by calculating the support vector distance;

[0081] The high-dimensional feature mapping data is input into the third hidden layer of the support vector machine model to extract the capacitive coupling interference features, and the capacitive coupling interference probability value is obtained by calculating the support vector distance;

[0082] The high-dimensional feature mapping data is input into the fourth hidden layer of the support vector machine model to extract the inductive coupling interference features, and the inductive coupling interference probability value is obtained by calculating the support vector distance;

[0083] The maximum values ​​of the power frequency interference probability value, the switch interference probability value, the capacitive coupling interference probability value and the inductive coupling interference probability value are compared, and the interference type corresponding to the maximum probability is determined as the electromagnetic interference mode type.

[0084] Specifically, the feature vector is input into the support vector machine (SVM) model, and the radial basis kernel function is used to perform inner product mapping of the high-dimensional feature space. The radial basis kernel function is a commonly used kernel function that maps low-dimensional nonlinear features to high-dimensional space, making the features that were originally linearly inseparable in the high-dimensional space become linearly separable. The radial basis kernel function is defined as:

[0085] ;

[0086] in, and are the input feature vector, represents the Euclidean distance, is the parameter of the kernel function, which is used to control the width of the Gaussian distribution. Through the mapping of the radial basis kernel function, high-dimensional feature mapping data is obtained to better describe the nonlinear relationship between features. The high-dimensional feature mapping data is input into different hidden layers of the support vector machine model to extract the features of a specific type of electromagnetic interference. The high-dimensional feature mapping data is input into the first hidden layer of the SVM model for feature extraction of power frequency interference. Power frequency interference is usually generated by power supply from the power system and has a relatively stable low-frequency characteristic. In the first hidden layer, the SVM determines the degree of similarity to the power frequency interference by calculating the distance between the high-dimensional feature data and the support vector. The support vector is the feature point selected by the SVM classifier during the training process, and these points play a key role in determining the classification hyperplane. The formula for calculating the support vector distance is:

[0087] ;

[0088] in, Represents the distance between high-dimensional feature data and the power frequency support vector, is the number of support vectors for the power frequency interference category, is the weight coefficient of the support vector, is the support vector in the power frequency interference category. Through the above formula, the probability value of power frequency interference is obtained, reflecting the possibility that the input feature belongs to the power frequency interference category. The same high-dimensional feature mapping data is input into the second hidden layer of the SVM model for feature extraction of switching interference. Switching interference is usually a transient interference signal generated by the device during the opening and closing process, which has high-frequency characteristics and obvious mutations. In the second hidden layer, the probability value of switching interference is obtained through similar support vector distance calculation:

[0089] ;

[0090] in, Represents the distance between the high-dimensional feature data and the switch interference support vector, is the number of support vectors of the switching interference category. By calculating this distance, it is effective to identify whether the input feature contains characteristic components related to switching interference. The high-dimensional feature mapping data is input into the third hidden layer of the SVM model to extract the characteristics of capacitive coupling interference. Capacitive coupling interference is caused by the influence of parasitic capacitance on the signal line and has a certain frequency distribution characteristic. In the third hidden layer, the probability value of capacitive coupling interference is also obtained by calculating the distance between the high-dimensional feature data and the support vector of the capacitive coupling interference category:

[0091] ;

[0092] in, Represents the distance between the high-dimensional feature data and the capacitive coupling interference support vector, is the number of support vectors for the capacitive coupling interference category. By calculation, the degree of influence of capacitive coupling on the system is identified. The high-dimensional feature map data is input into the fourth hidden layer of the SVM model to extract the characteristics of inductive coupling interference. Inductive coupling interference is caused by the magnetic field induction between adjacent conductors and affects the system. In the fourth hidden layer, SVM obtains the probability value of inductive coupling interference through support vector distance calculation:

[0093] ;

[0094] in, represents the distance between the high-dimensional feature data and the inductive coupling interference support vector, is the number of support vectors of the inductive coupling interference category. Thus, the inductive coupling interference in the input features is effectively evaluated. The maximum values ​​of the power frequency interference probability value, the switch interference probability value, the capacitive coupling interference probability value and the inductive coupling interference probability value are compared, and the interference type corresponding to the maximum probability is determined as the mode type of electromagnetic interference.

[0095] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0096] The observer bandwidth is calculated for the electromagnetic interference mode type, and the observer bandwidth allocation scheme is obtained by analyzing the frequency characteristics of power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference;

[0097] The bandwidth parameters of the high-frequency observer and the low-frequency observer are set according to the observer bandwidth allocation scheme. By configuring the high-frequency observer bandwidth to 1kHz and the low-frequency observer bandwidth to 100Hz, a dual-frequency extended state observer is obtained.

[0098] According to the electromagnetic interference mode type, the time domain signal is distributed into the high-frequency observer and the low-frequency observer, and the dual-frequency observation data is obtained by parallel processing of the electromagnetic interference signal;

[0099] The data output by the high-frequency observer is weighted, and the high-frequency weighted data is obtained by multiplying the first weight coefficient by the observed data, and the data output by the low-frequency observer is weighted, and the low-frequency weighted data is obtained by multiplying the second weight coefficient by the observed data;

[0100] The high-frequency weighted data and the low-frequency weighted data are time-series fused to obtain fused interference data, and the fused interference data is analyzed for state parameters to obtain interference feature quantities;

[0101] The state vector is constructed according to the interference characteristic quantity, and the change rate and cumulative effect of the interference quantity are calculated to obtain the electromagnetic interference state parameters.

[0102] Specifically, the frequency characteristics of power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference are analyzed to reasonably allocate the bandwidth of the observer. The characteristics of power frequency interference are concentrated at 50Hz, which belongs to the low frequency range, while switching interference is manifested as high frequency components, with frequencies ranging from hundreds of Hz to several kHz; capacitive coupling interference and inductive coupling interference also have different frequency characteristics. Capacitive coupling interference contains high frequency components, while the frequency range of inductive coupling interference is relatively wide. Based on the analysis of these interference characteristics, a reasonable bandwidth allocation scheme for the observer is obtained to effectively capture the characteristics of different types of interference. The bandwidth parameters of the high-frequency observer and the low-frequency observer are set according to the observer bandwidth allocation scheme. The bandwidth of the high-frequency observer is set to 1kHz, and the bandwidth of the low-frequency observer is set to 100Hz to obtain a dual-frequency extended state observer. The high-frequency observer captures electromagnetic interference with high-frequency characteristics, such as switching interference and part of capacitive coupling interference, while the low-frequency observer is used to capture power frequency interference and some low-frequency inductive coupling interference characteristics. The bandwidth of the high-frequency observer is selected to be 1kHz, covering switching interference and other high-frequency components; while the bandwidth of the low-frequency observer is 100Hz, which can accurately analyze the power frequency interference components in the low-frequency band. Through the setting, a dual-frequency extended state observer is formed, so that the entire observer system can cover both low-frequency interference and high-frequency interference characteristics, and has good spectrum range adaptability. According to the type of electromagnetic interference mode, the time domain signal is allocated to the high-frequency observer and the low-frequency observer for processing. The power frequency interference signal is mainly allocated to the low-frequency observer, while the switching interference and high-frequency capacitive coupling and inductive coupling interference are allocated to the high-frequency observer, ensuring that the observer can independently observe and process the interference characteristics in different frequency ranges. By processing these time domain signals in parallel, dual-frequency observation data is obtained. The dual-frequency observation data is weighted to better describe the contribution of each frequency band to the overall interference. The data output by the high-frequency observer is weighted by multiplying the first weight coefficient by the high-frequency observation data. The formula is:

[0103] ;

[0104] in, is the data output by the high-frequency observer. The setting of the weight coefficient depends on the characteristics of the interference signal and the characteristics of the self-standard interference. For the case where the high-frequency component in the signal is dominant, a higher high-frequency weight coefficient is set. Similarly, the data output by the low-frequency observer is weighted, and the formula is:

[0105] ;

[0106] in, is the weighted low-frequency data, is the second weight coefficient, The data output by the low-frequency observer. Through weighted operation, the influence of high-frequency and low-frequency components on the final observation results can be flexibly adjusted. The weighted high-frequency data and low-frequency data are time-series fused to obtain fused interference data. By integrating information from different frequency bands, the final fused data can more comprehensively reflect the characteristics of the interference signal. Time series fusion is achieved through weighted summation, and the formula is as follows:

[0107] ;

[0108] in, Indicates at time The fused interference data on the point, and are the high-frequency and low-frequency weighted data in time The value of the point. Through the fusion operation, a comprehensive interference signal containing high-frequency and low-frequency components is obtained. The state parameter analysis of the fused interference data is performed to extract the characteristic quantity of the interference. The interference characteristic quantity is a quantitative indicator that reflects different aspects of the electromagnetic interference signal, such as the change in amplitude, the characteristics of the frequency distribution, etc. The state parameter analysis is achieved by analyzing the change trend of the data. For example, by performing a derivative operation on the fused data, the change rate of the interference is obtained:

[0109] ;

[0110] in, It represents the instantaneous rate of change of the fused interference data, reflecting the intensity of the change of the interference signal over time. For stronger interference, the rate of change is usually larger, while for weaker interference, the rate of change is relatively smaller. The state vector is constructed based on the interference feature quantity. The state vector is a high-dimensional description used to comprehensively characterize the state of electromagnetic interference at each moment. Let the state vector be , which is composed of fused interference data, change rate and other feature quantities, and is expressed as:

[0111] ;

[0112] The state vector contains the instantaneous value of the interference signal and the dynamic change information of the interference, which can describe the overall characteristics of the interference. The interference characteristic quantity in the state vector is calculated to obtain the change rate and cumulative effect of the interference quantity. The change rate of the interference quantity is represented by the first-order derivative of the state vector, and the cumulative effect is obtained by integrating and summing the state vector:

[0113] ;

[0114] in, Indicates at time The cumulative effect of interference at a certain point reflects the overall impact of interference on the system over a period of time. The cumulative effect is used to evaluate the impact of long-term interference on the FPC high-performance computing chip interface. If the cumulative effect is too large, it will have a significant impact on the stability and performance of the chip.

[0115] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0116] Select the basic shielding structure according to the type of electromagnetic interference mode, and obtain the first layer compensation parameters by setting the shielding layer thickness, material type and interlayer connection method;

[0117] Based on the electromagnetic interference state parameters, the Lyapunov function is constructed, and the stability criterion is obtained by calculating the quadratic function of the system state variables.

[0118] The first-layer compensation parameters are backstepped and calculated. The control gain matrix is ​​obtained by recursively solving the step-by-step decomposition of the control law. The electromagnetic interference state parameters are input into the control gain matrix. The initial value of the compensation amount is obtained by matrix multiplication.

[0119] The impedance matching calculation is performed according to the initial value of the compensation amount. The impedance adjustment coefficient is obtained by adjusting the matching degree between the impedance of the shielding layer and the impedance of the interference source. The impedance adjustment coefficient is optimized online. The second-layer compensation parameter is obtained by adjusting the gain parameter of the backstepping controller.

[0120] The first layer compensation parameters and the second layer compensation parameters are hierarchically combined, and a hierarchical compensation controller is constructed to obtain a compensation strategy sequence, and the compensation strategy sequence is sorted in execution order to obtain compensation control parameters.

[0121] Specifically, the electromagnetic interference is analyzed and the appropriate shielding layer configuration is selected. This involves the design of the shielding layer thickness, material type and interlayer connection method so that it can effectively counteract specific types of electromagnetic interference. The modes of electromagnetic interference include power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference. The characteristics of each type of interference are different, and the selection requirements of shielding materials are also different. For example, for low-frequency interference (such as power frequency interference), it is necessary to select high-permeability materials to enhance the magnetic field shielding effect; and for high-frequency interference, select materials with high conductivity to reduce the coupling of the electric field. Assume that the thickness of the shielding layer is , material type is conductivity and magnetic permeability By setting these parameters reasonably, the first layer of compensation parameters are obtained for basic shielding design. Based on the state parameters of electromagnetic interference, in order to ensure the stability of the system, the Lyapunov function is constructed. The Lyapunov function is a mathematical tool for analyzing the stability of the system. By constructing a suitable function, it is determined whether the dynamic characteristics of the system tend to be stable. The Lyapunov function uses a quadratic function to represent the state variables of the system. Suppose the state variables of the system are , then the Lyapunov function is expressed as:

[0122] ;

[0123] in, is the state variable vector, is a symmetric positive definite matrix used to determine the weighted relationship of the system state. The derivative of the Lyapunov function is used to determine the stability of the system. If , then the system is asymptotically stable. By constructing the Lyapunov function and calculating its derivative, the stability criterion of the system is obtained, thereby evaluating the effectiveness of the shielding design and control strategy. The first-layer compensation parameters are backstepped and calculated. Backstepping control is a method of recursively solving the control law step by step. By decomposing the complex system step by step and designing the controller layer by layer, the final stability is guaranteed. The control objective of the system is decomposed into several sub-objectives, and the control law is designed layer by layer. Let the first-layer compensation parameter be , and the nonlinear control model of the system is expressed as:

[0124] ;

[0125] in, and are the nonlinear function and control gain matrix of the system respectively, is the compensation parameter of the first layer. By decomposing the control law step by step and recursively solving the control gain of each layer, we can get the control gain matrix Then the state parameters of electromagnetic interference are input into the control gain matrix, and the initial value of compensation is obtained by matrix multiplication:

[0126] ;

[0127] in, is the initial value of the compensation amount, which indicates the initial control amount obtained by backstepping control calculation under the current interference state. After obtaining the initial value of the compensation amount, impedance matching calculation is performed to improve the shielding layer's suppression effect on the interference source. The goal of impedance matching is to make the shielding layer's impedance match the interference source's impedance as much as possible to reduce reflected energy and improve absorption effect. Assume that the shielding layer impedance is , the interference source impedance is , the impedance matching is calculated by the following formula:

[0128] ;

[0129] in, The smaller the value, the better the matching effect. By adjusting the physical parameters of the shielding layer (such as thickness , Material conductivity etc.), so that As small as possible, thereby improving the shielding effect. After obtaining the impedance matching, the impedance adjustment coefficient is calculated to optimize the shielding parameters, and the adjustment coefficient is continuously adjusted using the online optimization method. By adjusting the gain parameter of the backstepping controller, the second layer compensation parameter is obtained. , to ensure the dynamic adaptive ability of shielding compensation in the actual operating environment. The first layer compensation parameters and the second layer compensation parameters are hierarchically combined to construct a hierarchical compensation controller. Assume that the first layer compensation parameters are , the second layer compensation parameter is , then the compensation control amount of the layered combination is expressed as:

[0130] ;

[0131] in, is the final compensation control quantity, combining the static compensation of the basic shielding design and the dynamic compensation of the adaptive backstepping control. Through the hierarchical combination, the comprehensive suppression of different types of electromagnetic interference is achieved. In order to ensure the reasonable execution order of each compensation strategy, the compensation strategy sequence is sorted in the execution order to obtain the final compensation control parameter. Let the compensation strategy sequence be ,By analyzing the priority of each compensation strategy and the degree of influence on the system, the execution order is determined, thereby sorting the compensation strategy sequence.,The final compensation control parameters are used to guide the configuration of the shielding structure and the adjustment of the controller to achieve the best shielding effect.

[0132] In a specific embodiment, the electromagnetic shielding compensation method for executing the FPC high performance computing chip interface further includes:

[0133] The compensation control parameters are input into the 16-node input layer of the judgment network, and each compensation parameter is nonlinearly transformed and mapped through the ReLU activation function to obtain a 16-dimensional feature vector;

[0134] The 16-dimensional feature vector is input into the 32-node hidden layer of the judgment network, and the weighted sum and bias of each node are calculated by the hyperbolic tangent function to obtain the 32-dimensional hidden layer features;

[0135] The 32-dimensional hidden layer features are input into the 1-node output layer of the evaluation network, and the features are compressed to the range of 0 to 1 through the Sigmoid function to obtain the compensation effect score;

[0136] The compensation control parameters are input into the Actor layer of the action network, and the compensation strategy features are extracted through a three-layer fully connected network structure to obtain the action strategy vector;

[0137] The compensation effect score and action strategy vector are input into the Critic layer of the action network, and the state-action value function is calculated by the temporal difference algorithm to obtain the compensation adjustment strategy;

[0138] Update the parameters of the compensation adjustment strategy, calculate the adjustment direction and step size of the compensation parameters through the policy gradient method, obtain the compensation parameter update amount, and superimpose the compensation parameter update amount with the compensation control parameter, and obtain the optimized compensation parameter through parameter cumulative update;

[0139] The performance of the optimized compensation parameters was evaluated, and the convergence was determined by setting a performance index threshold of 0.95 to obtain the optimal compensation parameters.

[0140] Specifically, a neural network structure is constructed to evaluate the effect of the compensation control parameters. The input layer of the evaluation network contains 16 nodes, each node corresponds to a compensation control parameter, the compensation control parameters are input into the input layer of the network, and each compensation parameter is nonlinearly transformed through the ReLU activation function. The form of the ReLU activation function is:

[0141] ;

[0142] in, As the input value, the role of ReLU is to set negative values ​​to zero and retain positive values, thereby introducing nonlinearity to ensure that the neural network can capture more complex feature patterns. Through the nonlinear mapping of the ReLU activation function, a 16-dimensional feature vector is obtained to represent the performance of the compensation control parameters under different conditions. The 16-dimensional feature vector is input to the 32-node hidden layer of the judgment network. The hidden layer calculates the weighted sum and bias of each node through the hyperbolic tangent function (tanh) to obtain a 32-dimensional hidden layer feature. The form of the hyperbolic tangent function is:

[0143] ;

[0144] in, is the input value. The output range of the tanh function is between -1 and 1, which makes the distribution of features more balanced and provides positive and negative expression capabilities for the input features. In the hidden layer, the feature calculation process is:

[0145] ;

[0146] in, For the The output of hidden layer nodes, Is the connection input layer node With hidden layer nodes The weight of is the hidden layer node By calculating the weighted sum and bias of each node and applying the tanh function, 32-dimensional hidden layer features are obtained. These features represent the comprehensive features of the input compensation parameters in the hidden layer. The 32-dimensional hidden layer features are input to the 1-node output layer of the evaluation network. The output layer compresses the features to the range of 0 to 1 through the Sigmoid function to obtain the compensation effect score. The form of the Sigmoid function is:

[0147] ;

[0148] in, The input value is the Sigmoid function, which maps the input value to the interval (0,1) and interprets the output as a probability or score. Through this process, the compensation effect score is obtained. The closer the score is to 1, the better the current compensation effect is. The closer the score is to 0, the poorer the compensation effect is. The compensation control parameters are input into the Actor layer of the action network. The Actor layer extracts the compensation strategy features through a three-layer fully connected network structure to generate an action strategy vector. The role of the three-layer fully connected network is to gradually abstract the input features and extract the effective features of the compensation strategy. Assume that the input is , the output of the three-layer fully connected network is:

[0149] ;

[0150] in, To compensate the control parameters, are the weight matrices of each layer, is the bias vector, is the activation function of each layer. Through this process, the action strategy vector Describe the control strategy that should be taken under different interference conditions. The compensation effect score and action strategy vector are input into the Critic layer of the action network. The Critic layer calculates the state-action value function through the temporal difference algorithm (TD algorithm) to evaluate the quality of the current compensation strategy. The temporal difference algorithm is a strategy evaluation method based on reinforcement learning. It calculates the state-action value function by estimating the future cumulative benefits of the current strategy. Let the current state be s and the action be , the state-action value function is expressed as:

[0151] ;

[0152] in, is the state-action value function, For the current reward, is the discount factor used to measure the weight of future rewards, is the next state after executing action a. The quality of the current compensation strategy is calculated through the temporal difference algorithm to obtain the compensation adjustment strategy. The compensation parameters of the compensation adjustment strategy are updated. The policy gradient method is used to calculate the adjustment direction and step size of the compensation parameters so that the compensation effect can be improved. The update process of the policy gradient method is:

[0153] ;

[0154] in, is the update amount of the compensation parameter, is the learning rate, is the gradient of the state-action value function to the compensation parameter. By calculating the gradient, the adjustment direction of the compensation parameter is obtained, and the adjustment step size is determined according to the learning rate. The compensation parameter update amount is superimposed on the original compensation control parameter, and the optimized compensation parameter is obtained through parameter cumulative update:

[0155] ;

[0156] in, is the updated compensation parameter. The optimized compensation parameter is evaluated to determine whether it has converged to the optimal state. The performance index threshold is set to 0.95. If the current compensation effect score is higher than 0.95, it is considered that the current compensation parameter has reached the optimal state and the update is stopped; otherwise, the parameter adjustment needs to be continued until the score reaches or exceeds the threshold.

[0157] The electromagnetic shielding compensation method of the FPC high-performance computing chip interface in the embodiment of the present invention is described above. The electromagnetic shielding compensation system of the FPC high-performance computing chip interface in the embodiment of the present invention is described below. Figure 2 In one embodiment of the present invention, an electromagnetic shielding compensation system of an FPC high-performance computing chip interface includes:

[0158] The acquisition module 201 is used to acquire impedance signals from the FPC high-performance computing chip interface to obtain time domain signals and second-order rate of change curves. The time domain signals are acquired by applying an AC excitation signal between the signal layer and the ground layer.

[0159] An extraction module 202 is used to extract features from the time domain signal and the second-order rate of change curve to obtain time domain feature parameters and frequency domain feature parameters, and to form a feature vector with the time domain feature parameters and the frequency domain feature parameters;

[0160] The identification module 203 is used to input the feature vector into the support vector machine model to classify and identify the electromagnetic interference, and obtain the electromagnetic interference mode type, which includes power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference;

[0161] The analysis module 204 is used to construct a dual-frequency extended state observer according to the time domain signal and the electromagnetic interference mode type, and perform weighted fusion analysis on the time domain signal through the high frequency observer and the low frequency observer to obtain the electromagnetic interference state parameter;

[0162] The adjustment module 205 is used to construct a layered compensation controller based on the electromagnetic interference mode type and the electromagnetic interference state parameters, and adjust the shielding layer impedance parameters through an adaptive backstepping control algorithm to obtain compensation control parameters.

[0163] Through the cooperation of the above components, the impedance signal is collected by applying AC excitation signal between the signal layer and the ground layer, and the feature extraction of time domain signal and second-order rate of change curve is combined to achieve accurate characterization of the electromagnetic characteristics of the FPC interface. The support vector machine model is used to classify and identify the electromagnetic interference of the feature vector, accurately distinguishing various interference types such as power frequency interference, switching interference, capacitive coupling interference and inductive coupling interference, thereby improving the accuracy of electromagnetic interference identification. Based on the design of dual-frequency extended state observer, the real-time and accurate observation of the electromagnetic interference state is achieved through weighted fusion analysis of high-frequency observer and low-frequency observer, effectively avoiding the problem of observation noise amplification. The hierarchical compensation controller structure is adopted, combined with the adaptive backstepping control algorithm, to achieve accurate adjustment of the shielding layer impedance parameters, ensuring the stability and effectiveness of electromagnetic shielding compensation. The dual network structure of the judgment network and the action network is introduced for reinforcement learning optimization. Through online learning and parameter optimization, the shielding compensation effect is continuously improved, so that the system has adaptive optimization capabilities.

[0164] above Figure 2 The electromagnetic shielding compensation system of the FPC high-performance computing chip interface in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The electromagnetic shielding compensation device of the FPC high-performance computing chip interface in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0165] Figure 3It is a structural schematic diagram of an electromagnetic shielding compensation device of an FPC high-performance computing chip interface provided by an embodiment of the present invention. The electromagnetic shielding compensation device 300 of the FPC high-performance computing chip interface may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the electromagnetic shielding compensation device 300 of the FPC high-performance computing chip interface. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the electromagnetic shielding compensation device 300 of the FPC high-performance computing chip interface to implement the steps of the electromagnetic shielding compensation method of the FPC high-performance computing chip interface.

[0166] The electromagnetic shielding compensation device 300 of the FPC high-performance computing chip interface may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The electromagnetic shielding compensation device structure of the FPC high-performance computing chip interface shown does not constitute a limitation on the electromagnetic shielding compensation device of the FPC high-performance computing chip interface provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0167] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the electromagnetic shielding compensation method of the FPC high-performance computing chip interface.

[0168] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0169] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0170] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An electromagnetic shielding compensation method for an FPC high-performance computing chip interface, characterized in that: The method comprises: The impedance signal is collected on the FPC high-performance computing chip interface to obtain a time domain signal and a second-order rate of change curve, wherein the time domain signal is collected by applying an AC excitation signal between the signal layer and the ground layer; Performing feature extraction on the time domain signal and the second-order rate of change curve to obtain time domain feature parameters and frequency domain feature parameters, and combining the time domain feature parameters and the frequency domain feature parameters into a feature vector; Inputting the feature vector into a support vector machine model to classify and identify electromagnetic interference to obtain electromagnetic interference pattern types, wherein the electromagnetic interference pattern types include power frequency interference, switching interference, capacitive coupling interference, and inductive coupling interference; A dual-frequency extended state observer is constructed according to the time domain signal and the electromagnetic interference mode type, and a weighted fusion analysis is performed on the time domain signal through a high-frequency observer and a low-frequency observer to obtain an electromagnetic interference state parameter; A layered compensation controller is constructed based on the electromagnetic interference mode type and the electromagnetic interference state parameter, and the shielding layer impedance parameter is adjusted through an adaptive backstepping control algorithm to obtain a compensation control parameter.

2. The electromagnetic shielding compensation method of the FPC high-performance computing chip interface according to claim 1 is characterized in that: The impedance signal is collected from the FPC high-performance computing chip interface to obtain a time domain signal and a second-order rate of change curve, wherein the time domain signal is collected by applying an AC excitation signal between the signal layer and the ground layer, including: A 1V AC excitation signal is applied between the signal layer and the ground layer of the FPC high-performance computing chip interface, and an impedance analyzer is used to collect the current response signal; The current response signal is sampled at a sampling frequency of 1 MHz, and the sampling time is set to 100 ms to obtain the original impedance data; Performing calculation based on the ratio of the AC excitation signal to the current response signal to obtain an impedance value sequence; Inputting the impedance value sequence into a Butterworth high-pass filter to filter out interference components below 100 Hz to obtain a filtered impedance signal; The filtered impedance signal is subjected to 50 Hz power frequency interference elimination processing to obtain a time domain signal, and a second-order differential operation is performed on the time domain signal to obtain a second-order rate of change curve.

3. The electromagnetic shielding compensation method of the FPC high-performance computing chip interface according to claim 2 is characterized in that: The feature extraction of the time domain signal and the second-order rate of change curve to obtain a time domain feature parameter and a frequency domain feature parameter, and combining the time domain feature parameter and the frequency domain feature parameter into a feature vector includes: Performing statistical parameter calculation on the time domain signal to obtain time domain characteristic parameters, and performing wavelet packet decomposition on the second-order rate of change curve to obtain multiple frequency band sub-signals; Performing energy calculation on the multiple frequency band sub-signals to obtain frequency band energy values, and normalizing the frequency band energy values ​​to obtain frequency domain feature parameters; The time domain feature parameters and the frequency domain feature parameters are combined to obtain an original feature vector; The original eigenvector is input into the principal component analysis model, and the characteristic contribution rate is calculated by eigenvalue decomposition to obtain the principal component mapping matrix; The original eigenvector is transformed into a dimension-reduced one by the principal component mapping matrix to obtain the eigendata after dimensionality reduction, and the eigendata after dimensionality reduction is standardized to obtain the eigenvector.

4. The electromagnetic shielding compensation method of the FPC high-performance computing chip interface according to claim 3 is characterized in that: The inputting the feature vector into the support vector machine model to classify and identify electromagnetic interference to obtain electromagnetic interference pattern types, wherein the electromagnetic interference pattern types include power frequency interference, switching interference, capacitive coupling interference, and inductive coupling interference, including: The feature vector is input into a support vector machine model, and an inner product mapping of the feature space is calculated by a radial basis kernel function of the support vector machine model to obtain high-dimensional feature mapping data; Inputting the high-dimensional feature mapping data into the first hidden layer of the support vector machine model to extract the power frequency interference features, and obtaining the power frequency interference probability value by calculating the support vector distance; Inputting the high-dimensional feature mapping data into the second hidden layer of the support vector machine model to extract switch interference features, and obtaining a switch interference probability value by calculating the support vector distance; Inputting the high-dimensional feature mapping data into the third hidden layer of the support vector machine model to extract the capacitive coupling interference features, and obtaining the capacitive coupling interference probability value by calculating the support vector distance; Inputting the high-dimensional feature mapping data into the fourth hidden layer of the support vector machine model to extract the inductive coupling interference features, and obtaining the inductive coupling interference probability value by calculating the support vector distance; A maximum comparison is performed on the power frequency interference probability value, the switch interference probability value, the capacitive coupling interference probability value, and the inductive coupling interference probability value, and an interference type corresponding to the maximum probability is determined as an electromagnetic interference mode type.

5. The electromagnetic shielding compensation method of the FPC high-performance computing chip interface according to claim 4 is characterized in that: The dual-frequency extended state observer is constructed according to the time domain signal and the electromagnetic interference mode type, and the time domain signal is subjected to weighted fusion analysis by a high-frequency observer and a low-frequency observer to obtain electromagnetic interference state parameters, including: Calculating the observer bandwidth for the electromagnetic interference mode type, and obtaining an observer bandwidth allocation scheme by analyzing the frequency characteristics of power frequency interference, switching interference, capacitive coupling interference, and inductive coupling interference; The bandwidth parameters of the high-frequency observer and the low-frequency observer are set according to the observer bandwidth allocation scheme, and a dual-frequency extended state observer is obtained by configuring the high-frequency observer bandwidth to be 1 kHz and the low-frequency observer bandwidth to be 100 Hz; Distributing the time domain signal into the high-frequency observer and the low-frequency observer according to the electromagnetic interference mode type, and obtaining dual-frequency observation data by processing the electromagnetic interference signal in parallel; Weighting the data output by the high-frequency observer, obtaining high-frequency weighted data by multiplying the first weight coefficient by the observed data, and weighting the data output by the low-frequency observer, obtaining low-frequency weighted data by multiplying the second weight coefficient by the observed data; Performing time series fusion on the high-frequency weighted data and the low-frequency weighted data to obtain fused interference data, and performing state parameter analysis on the fused interference data to obtain interference feature quantities; A state vector is constructed according to the interference characteristic quantity, and the change rate and cumulative effect of the interference quantity are calculated to obtain the electromagnetic interference state parameter.

6. The electromagnetic shielding compensation method of the FPC high-performance computing chip interface according to claim 5, characterized in that: The layered compensation controller is constructed based on the electromagnetic interference mode type and the electromagnetic interference state parameter, and the shielding layer impedance parameter is adjusted by an adaptive backstepping control algorithm to obtain compensation control parameters, including: Selecting a basic shielding structure according to the electromagnetic interference mode type, and obtaining first layer compensation parameters by setting the shielding layer thickness, material type and interlayer connection method; Constructing a Lyapunov function based on the electromagnetic interference state parameters, and obtaining a stability criterion by calculating a quadratic function of the system state variables; Performing backstepping control calculation on the first layer compensation parameters, obtaining a control gain matrix by recursively solving the step-by-step decomposition of the control law, and inputting the electromagnetic interference state parameters into the control gain matrix, and obtaining an initial value of the compensation amount by matrix multiplication operation; An impedance matching calculation is performed according to the initial value of the compensation amount, an impedance adjustment coefficient is obtained by adjusting the matching degree between the impedance of the shielding layer and the impedance of the interference source, and the impedance adjustment coefficient is optimized online, and a second-layer compensation parameter is obtained by adjusting the gain parameter of the backstepping controller; The first layer compensation parameters and the second layer compensation parameters are hierarchically combined, a hierarchical compensation controller is constructed to obtain a compensation strategy sequence, and the compensation strategy sequence is sorted in execution order to obtain compensation control parameters.

7. The electromagnetic shielding compensation method of the FPC high-performance computing chip interface according to claim 6 is characterized in that: The electromagnetic shielding compensation method of the FPC high-performance computing chip interface also includes: Input the compensation control parameters into the 16-node input layer of the judgment network, and perform nonlinear transformation mapping on each compensation parameter through the ReLU activation function to obtain a 16-dimensional feature vector; Input the 16-dimensional feature vector into the 32-node hidden layer of the evaluation network, calculate the weighted sum and bias of each node by the hyperbolic tangent function, and obtain the 32-dimensional hidden layer feature; The 32-dimensional hidden layer features are input into the 1-node output layer of the evaluation network, and the features are compressed to the range of 0 to 1 by the Sigmoid function to obtain the compensation effect score; The compensation control parameters are input into the Actor layer of the action network, and the compensation strategy features are extracted through a three-layer fully connected network structure to obtain an action strategy vector; Input the compensation effect score and the action strategy vector into the Critic layer of the action network, calculate the state-action value function through the temporal difference algorithm, and obtain the compensation adjustment strategy; The compensation adjustment strategy is updated, and the adjustment direction and step size of the compensation parameter are calculated by the policy gradient method to obtain the compensation parameter update amount, and the compensation parameter update amount is superimposed on the compensation control parameter to obtain the optimized compensation parameter by cumulative parameter update; The optimized compensation parameters are evaluated for performance, and a performance indicator threshold of 0.95 is set to determine whether convergence has occurred, thereby obtaining the optimal compensation parameters.

8. An electromagnetic shielding compensation system for an FPC high-performance computing chip interface, characterized in that: A method for performing electromagnetic shielding compensation of an FPC high-performance computing chip interface according to any one of claims 1 to 7, the system comprising: An acquisition module is used to acquire impedance signals from the FPC high-performance computing chip interface to obtain a time domain signal and a second-order rate of change curve, wherein the time domain signal is acquired by applying an AC excitation signal between the signal layer and the ground layer; An extraction module, used for performing feature extraction on the time domain signal and the second-order rate of change curve to obtain time domain feature parameters and frequency domain feature parameters, and combining the time domain feature parameters and the frequency domain feature parameters into a feature vector; An identification module, used for inputting the feature vector into a support vector machine model to classify and identify electromagnetic interference, and obtaining electromagnetic interference pattern types, wherein the electromagnetic interference pattern types include power frequency interference, switching interference, capacitive coupling interference, and inductive coupling interference; An analysis module, configured to construct a dual-frequency extended state observer according to the time domain signal and the electromagnetic interference mode type, and perform weighted fusion analysis on the time domain signal through a high frequency observer and a low frequency observer to obtain electromagnetic interference state parameters; The regulating module is used to construct a layered compensation controller based on the electromagnetic interference mode type and the electromagnetic interference state parameter, and to adjust the shielding layer impedance parameter through an adaptive backstepping control algorithm to obtain a compensation control parameter.

9. An electromagnetic shielding compensation device for an FPC high-performance computing chip interface, characterized in that: The electromagnetic shielding compensation device of the FPC high-performance computing chip interface includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electromagnetic shielding compensation device of the FPC high-performance computing chip interface executes the electromagnetic shielding compensation method of the FPC high-performance computing chip interface as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the electromagnetic shielding compensation method of the FPC high-performance computing chip interface as described in any one of claims 1-7 is implemented.

Images