Circuit board impedance matching and signal integrity optimization method and system

Abstract

The invention provides a circuit board impedance matching and signal integrity optimization method and system, and relates to the technical field of circuit board design, and the method comprises the steps: dividing a signal integrity analysis region, and setting a signal test point; signal characteristic parameters are collected and classified and analyzed; identifying a signal quality abnormal region and establishing a multi-dimensional evaluation index system; constructing a signal compensation model and a recursive filter; constructing a deep reinforcement learning model, and dynamically adjusting a compensation coefficient and a filtering coefficient; carrying out processing and real-time evaluation on the signal quality abnormal region; and when the score continuously reaches a preset threshold value, determining that optimization is completed.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit board design, and in particular to a circuit board impedance matching and signal integrity optimization method and system. Background Art

[0002] As integrated circuits and electronic devices develop towards higher speed, miniaturization and higher integration, signal integrity issues are increasingly becoming a key factor restricting system performance. During high-speed signal transmission, factors such as impedance mismatch, frequency dispersion effects and temperature changes on circuit boards will lead to a decrease in signal quality. To ensure reliable operation of the system, impedance matching and signal integrity optimization of circuit boards are required, which has become an important part of high-speed circuit design. Traditional signal integrity optimization methods mainly improve signal quality by adjusting compensation parameters and filtering parameters, and perform manual optimization and debugging based on test data.

[0003] In the existing technology, the single-point test method is difficult to fully reflect the signal characteristic differences in various areas of the circuit board, and it is easy to miss local signal quality abnormalities; the signal processing strategy with fixed parameters fails to fully consider the influence of frequency dispersion effects, and it is difficult to effectively deal with the loss and distortion of different frequency components; the optimization method that relies on empirical parameter settings and manual debugging lacks adaptive capabilities and cannot respond to the impact of changes in operating temperature and signal rate in a timely manner; the evaluation method of a single indicator cannot accurately reflect the overall status of the signal quality, and the static optimization method that lacks real-time feedback and dynamic adjustment mechanism has low optimization efficiency.

[0004] In summary, the systematic analysis and adaptive optimization of circuit board signal characteristics are achieved, a comprehensive multi-dimensional evaluation system is established, and on this basis, an optimization method with real-time feedback and dynamic adjustment capabilities is constructed to improve the accuracy and efficiency of signal integrity optimization. The present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiment of the present invention provides a circuit board impedance matching and signal integrity optimization method and system, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention, A circuit board impedance matching and signal integrity optimization method is provided, comprising: Divide the circuit board into multiple signal integrity analysis areas, and set multiple signal test points in each signal integrity analysis area, collect signal characteristic parameters of the signal test points, including characteristic impedance value, signal transmission rate, signal rise time, and signal response under different operating temperatures, classify and analyze the signal characteristic parameters based on an adaptive weighted clustering algorithm, identify signal quality abnormal areas, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing, and power consumption, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system; For the signal quality abnormal area, a signal compensation model considering the frequency dispersion effect is constructed, and the initial value of the compensation coefficient is calculated based on the signal characteristic parameters; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameters; based on the initial value of the compensation coefficient and the initial value of the filter coefficient, the signal in the signal quality abnormal area is preliminarily processed, and the processing result is evaluated by the multi-dimensional evaluation index system to obtain the score change relative to the initial score, and determine the coefficient adjustment amount of the compensation coefficient and the filter coefficient; A deep reinforcement learning model is constructed, and the characteristic parameters of the signal test points are used as state inputs, coefficient adjustment amounts are used as action outputs, and score changes are used as reward functions to construct the deep reinforcement learning model; using the deep reinforcement learning model, based on the state inputs and the reward function, the compensation coefficients and the filter coefficients are dynamically adjusted to obtain optimized compensation coefficients and optimized filter coefficients, signal processing is performed on areas with abnormal signal quality, and real-time evaluation is performed, and the evaluation results are fed back to the deep reinforcement learning model for iterative optimization; when the scores of the multi-dimensional evaluation index system continuously reach a preset threshold, it is determined that the signal integrity optimization of the circuit board is completed.

[0007] In an optional embodiment, The circuit board is divided into multiple signal integrity analysis areas, and multiple signal test points are set in each signal integrity analysis area. The signal characteristic parameters of the signal test points are collected, including characteristic impedance value, signal transmission rate, signal rise time and signal response under different operating temperatures. Based on the adaptive weighted clustering algorithm, the signal characteristic parameters are classified and analyzed to identify the signal quality abnormal area, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power consumption loss, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system, including: Based on the signal test points, the nodes of the graph are determined, and based on the signal transmission path, the edges of the graph are determined, and the wiring topology of the circuit board is converted into a graph structure; the local features and global features of the signal test points are extracted, and the local features include characteristic impedance values, signal transmission rates, and signal rise times, and the global features include signal responses at different operating temperatures; the association strength between the signal test points is calculated by using a graph attention mechanism, and a region segmentation criterion is constructed based on the association strength to divide the circuit board into multiple signal integrity analysis regions; A multimodal sensor array is arranged in the signal integrity analysis area to collect signal characteristic parameters of the signal test point; a sampling strategy is determined according to the signal characteristic parameters, including a sampling rate and a sampling duration; a wavelet transform is used to perform time-frequency analysis on the signal characteristic parameters to extract time-varying characteristics; a three-dimensional distribution model of the signal characteristic parameters is constructed based on the spatial distribution of the multimodal sensor array; segmentation information of the signal integrity analysis area is combined with the three-dimensional distribution model to generate regional characteristic data; Construct an adaptive weighted clustering model and input regional feature data into the adaptive weighted clustering model; dynamically adjust the weight coefficient of the adaptive weighted clustering model based on the time-varying characteristics of signal feature parameters, classify and analyze signal feature parameters, and identify areas with abnormal signal quality; A multi-dimensional evaluation index system is established, including a signal integrity index, a crosstalk suppression index, a signal timing index, and a power consumption loss index; an index weight calculation model is constructed based on the time-varying characteristics of the signal characteristic parameters; the correlation strength of the signal test point is used as the basis for calculating the coupling degree between the indicators in the multi-dimensional evaluation index system, and the weight of each indicator in the multi-dimensional evaluation index system is dynamically adjusted according to the signal characteristic parameters; combined with the identification results of the abnormal signal quality area, the initial score of the abnormal signal quality area based on the multi-dimensional evaluation index system is calculated.

[0008] In an optional embodiment, The region segmentation criteria constructed based on the association strength include: Acquire the initial correlation strength between the signal test points, map the initial correlation strength to an interval from zero to one according to the maximum and minimum value range, and obtain a first correlation strength; calculate the physical distance between the signal test points according to the spatial coordinates of the signal test points, and based on the physical distance, generate a distance attenuation coefficient that decreases as the physical distance increases, and multiply the distance attenuation coefficient by the first correlation strength to obtain a second correlation strength; extract the signal waveform features of the signal test points, calculate the waveform similarity between different signal test points, and obtain a waveform similarity coefficient, and multiply the waveform similarity coefficient by the second correlation strength to obtain a third correlation strength; Calculating the mean value of the third correlation strength as the initial correlation threshold; establishing a region division rule, and dividing the adjacent signal test points whose third correlation strength is greater than the initial correlation threshold into the same region according to the region division rule, to obtain an initial region division result; Calculate the maximum physical distance between signal test points in the same area. If the maximum physical distance exceeds a preset distance threshold, separate the signal test point with the farthest distance to form a new area to obtain the first optimized area division result; count the number of signal test points in each area in the first optimized area division result, divide the area where the number of signal test points exceeds the preset upper limit, and merge the area where the number of signal test points is lower than the preset lower limit to obtain the second optimized area division result; identify the change trend of the third correlation strength at the area boundary in the second optimized area division result, determine that the third correlation strength decreases from the inside of the area to the boundary, and if the decreasing trend is not met, adjust the area boundary to obtain the third optimized area division result; Calculating the average third correlation strength of the signal test points between adjacent areas, extracting the signal waveform features of the adjacent areas, and merging the adjacent areas if the similarity between the average third correlation strength and the signal waveform features exceeds a set threshold; The spatial distribution density of the signal test points in each area after merging and the third correlation strength distribution are statistically calculated to construct a regional uniformity evaluation parameter; the initial correlation threshold is adjusted according to the regional uniformity evaluation parameter, and the adjusted correlation threshold is reused for regional division, and the optimization process is repeatedly executed until the regional division result is stable.

[0009] In an optional embodiment, Based on the spatial distribution of the multimodal sensor array, constructing a three-dimensional distribution model of the signal characteristic parameters; combining the segmentation information of the signal integrity analysis area with the three-dimensional distribution model to generate regional characteristic data includes: Taking the plane of the circuit board as a reference, a rectangular coordinate system is established, the physical coordinate information of each sensor in the multimodal sensor array is obtained, and the physical coordinate information is mapped into the rectangular coordinate system to obtain the sensor mapping coordinates; Collect the original signal features of the multimodal sensor array, perform maximum and minimum value normalization processing on the original signal features to obtain a first signal feature; calculate the distance between adjacent sensor mapping coordinates, determine the grid division interval based on the distance, and perform grid division in the plane of the rectangular coordinate system to obtain a two-dimensional grid; count the value distribution range of the first signal feature, divide the value distribution range into multiple value layers, and superimpose the value layers on the basis of the two-dimensional grid to obtain a three-dimensional grid; calculate the distance from each grid point in the three-dimensional grid to the surrounding sensor mapping coordinates, calculate the spatial influence factor of the grid point based on the distance, and multiply the spatial influence factor by the first signal feature of the corresponding sensor to obtain a second signal feature; according to a preset sparsity threshold, identify the sparse distribution area of ​​the sensor mapping coordinates, and perform local weighted smoothing processing on the second signal features of the grid points in the sparse area to obtain a third signal feature; Acquire boundary information of a signal integrity analysis area, map the boundary information to a three-dimensional grid containing the third signal feature, and extract a set of grid points within each analysis area; Calculate the mean and variance of the third signal feature in the grid point set to obtain regional statistical parameters, calculate the spatial gradient of the third signal feature to obtain regional spatial parameters; combine the regional statistical parameters with the regional spatial parameters to form a regional feature description, calculate the Euclidean distance between the regional feature descriptions of adjacent analysis areas to obtain regional similarity; extract the position of the local extreme point of the third signal feature in the analysis area as the feature point coordinates, and combine the regional feature description, the regional similarity and the feature point coordinates to generate regional feature data.

[0010] In an optional embodiment, For the signal quality abnormal area, a signal compensation model considering the frequency dispersion effect is constructed, and an initial value of the compensation coefficient is calculated based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter, including: Perform fast Fourier transform on the signal in the signal quality abnormality area to obtain the frequency domain amplitude spectrum and phase spectrum, calculate the transmission constant based on the dielectric constant, loss tangent, line width, line thickness and interlayer spacing of the transmission line, construct a frequency response function based on the transmission constant, and determine the amplitude attenuation and phase delay in the signal transmission process based on the frequency response function; Acquire signal characteristic parameters, map the rise time in the signal characteristic parameters to high-frequency compensation strength, map the overshoot to medium-frequency compensation strength, and map the jitter value to low-frequency compensation strength; construct a piecewise linear frequency compensation function, and determine the initial value of the compensation coefficient based on the high-frequency compensation strength, the medium-frequency compensation strength, and the low-frequency compensation strength, combined with amplitude attenuation and phase delay, wherein the compensation coefficient includes a phase compensation coefficient and an amplitude compensation coefficient; A complementary filter with a three-branch recursive structure is constructed, wherein the complementary filter comprises a high-pass filter branch, a medium-pass filter branch and a low-pass filter branch; the cutoff frequency of the complementary filter is determined based on the signal bandwidth, the quality factor of the complementary filter is determined based on the signal-to-noise ratio, and an initial value of the filter coefficient is obtained.

[0011] In an optional embodiment, Based on the initial value of the compensation coefficient and the initial value of the filter coefficient, the signal quality abnormal area is initially processed, and the processing result is evaluated by the multi-dimensional evaluation index system to obtain the score change relative to the initial score. The coefficient adjustment amount of the compensation coefficient and the filter coefficient is determined, including: The signal in the signal quality abnormality area is processed through the frequency response function to obtain the transmitted signal, and the transmitted signal is processed through the frequency compensation function to obtain the compensated signal; the compensated signal is processed through the high-pass filter branch, the middle-pass filter branch and the low-pass filter branch respectively to obtain the high-pass output signal, the middle-pass output signal and the low-pass output signal; Performing weighted combination on the high-pass output signal, the mid-pass output signal and the low-pass output signal to obtain a preliminary processed signal, calculating the eye opening, signal-to-noise ratio and timing jitter of the preliminary processed signal to obtain a processed score; calculating the difference between the processed score and the initial score, determining the score change, and using the gradient descent method to calculate the adjustment direction of the compensation coefficient and the filter coefficient; The segmentation points of the frequency compensation function, the adjustment amount of the compensation strength, and the cutoff frequency and the adjustment amount of the quality factor of the complementary filter are calculated respectively, and applied to the current compensation coefficient and the filter coefficient to obtain a new parameter combination, and the signal processing is re-executed using the new parameter combination to obtain a new score change; when the new score change is less than a preset threshold, the parameter optimization is stopped to obtain the optimized compensation coefficient and the filter coefficient.

[0012] In an optional embodiment, A deep reinforcement learning model is constructed, and the characteristic parameters of the signal test point are used as state input, the coefficient adjustment amount is used as action output, and the score change is used as a reward function to construct a deep reinforcement learning model; using the deep reinforcement learning model, based on the state input and the reward function, the compensation coefficient and the filter coefficient are dynamically adjusted to obtain the optimized compensation coefficient and the optimized filter coefficient, and the signal quality abnormal area is processed and evaluated in real time, and the evaluation result is fed back to the deep reinforcement learning model for iterative optimization; when the score of the multi-dimensional evaluation index system continuously reaches a preset threshold, determining that the signal integrity optimization of the circuit board is completed includes: Standardize the characteristic parameters of the signal test points, establish a symbol description mapping table based on the standardized characteristic parameters, determine the mapping relationship, and build an initial parameter adjustment rule base, wherein the initial parameter adjustment rule base includes compensation coefficient adjustment strategies and filter coefficient adjustment strategies under different characteristic combinations; The compensation coefficient, the filter coefficient and the standardized feature parameter are used as nodes of a causal dependency graph, and the compensation coefficient adjustment strategy and the filter coefficient adjustment strategy are used as edges of the causal dependency graph to construct a causal dependency graph; a graph attention network is constructed, wherein the input layer of the graph attention network receives the causal dependency graph, and the weight of the edge is calculated by using the attention mechanism through the message passing layer, the information transmission between the nodes is determined, and the compensation coefficient adjustment amount and the filter coefficient adjustment amount are obtained; Establishing a semantic mapping relationship between the initial parameter adjustment rule base and the compensation coefficient adjustment amount and the filter coefficient adjustment amount, and verifying the consistency of the rule reasoning result and the network output result through a bidirectional constraint mechanism; based on the consistency verification result, updating the initial parameter adjustment rule base to obtain an optimized parameter adjustment rule base; Processing the signal according to the compensation coefficient adjustment amount and the filter coefficient adjustment amount to obtain characteristic parameters of the processed signal; calculating a score change value of the characteristic parameters of the processed signal; Using the score change value as a reward value, updating the weights of the edges in the causal dependency graph and the parameters of the graph attention network; performing a counterfactual analysis on the compensation coefficient adjustment amount and the filter coefficient adjustment amount based on the score change value to obtain a potential adjustment result; updating the adjustment strategy in the optimization parameter adjustment rule base according to the potential adjustment result, and dynamically adjusting the structure of the causal dependency graph; and re-optimizing the parameters based on the updated causal dependency graph and the optimization parameter adjustment rule base; When the score reaches the preset threshold value K times in a row, the current compensation coefficient is output as the optimized compensation coefficient, and the current filtering coefficient is output as the optimized filtering coefficient, thereby completing the signal integrity optimization method.

[0013] According to a second aspect of the embodiments of the present invention, A circuit board impedance matching and signal integrity optimization system is provided, comprising: The first unit is used to divide the circuit board into multiple signal integrity analysis areas, and set multiple signal test points in each signal integrity analysis area, collect signal characteristic parameters of the signal test points, including characteristic impedance value, signal transmission rate, signal rise time and signal response under different working temperatures, classify and analyze the signal characteristic parameters based on an adaptive weighted clustering algorithm, identify signal quality abnormal areas, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power consumption loss, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system; The second unit is used to construct a signal compensation model that takes into account the frequency dispersion effect for the signal quality abnormal area, and calculate the initial value of the compensation coefficient based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter; based on the initial value of the compensation coefficient and the initial value of the filter coefficient, the signal quality abnormal area is preliminarily processed, and the processing result is evaluated by the multi-dimensional evaluation index system to obtain the score change relative to the initial score, and determine the coefficient adjustment amount of the compensation coefficient and the filter coefficient; The third unit is used to construct a deep reinforcement learning model, which takes the characteristic parameters of the signal test point as the state input, the coefficient adjustment amount as the action output, and the score change as the reward function to construct the deep reinforcement learning model; using the deep reinforcement learning model, based on the state input and the reward function, dynamically adjust the compensation coefficient and the filter coefficient to obtain the optimized compensation coefficient and the optimized filter coefficient, perform signal processing on the signal quality abnormal area, and perform real-time evaluation, and feed back the evaluation result to the deep reinforcement learning model for iterative optimization; when the score of the multi-dimensional evaluation index system continuously reaches the preset threshold, it is determined that the signal integrity optimization of the circuit board is completed.

[0014] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0016] In an embodiment of the present invention, a circuit board is divided into multiple signal integrity analysis areas, and multiple signal test points are set in each area to collect multiple signal characteristic parameters. An adaptive weighted clustering algorithm is used to classify and analyze the parameters to identify abnormal signal quality areas. At the same time, a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power loss is established, so that the analysis of circuit board signal quality is more comprehensive and accurate; a signal compensation model considering frequency dispersion effect is constructed, and a complementary recursive filter is used to process the signal. By combining the compensation coefficient and the filter coefficient, the signal quality abnormal area is optimized in a targeted manner, which effectively improves the signal quality; a deep reinforcement learning model is introduced, and the characteristic parameters of the signal test point are used as the state input, the coefficient adjustment amount is used as the action output, and the score change is used as the reward function. In this way, the system can dynamically adjust the compensation coefficient and the filter coefficient to achieve adaptive optimization of signal processing. At the same time, through real-time evaluation and feedback mechanism, the system can continuously iterate and optimize until the preset performance threshold is reached, thereby ensuring the continuous improvement of circuit board signal integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a circuit board impedance matching and signal integrity optimization method according to an embodiment of the present invention; Figure 2 It is a schematic diagram of the structure of a circuit board impedance matching and signal integrity optimization system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0019] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0020] Figure 1 FIG. 1 is a flow chart of a method for optimizing circuit board impedance matching and signal integrity according to an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Divide the circuit board into multiple signal integrity analysis areas, and set multiple signal test points in each signal integrity analysis area, collect signal characteristic parameters of the signal test points, classify and analyze the signal characteristic parameters based on an adaptive weighted clustering algorithm, identify signal quality abnormal areas, establish a multi-dimensional evaluation index system, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system; S102. For the signal quality abnormal area, a signal compensation model considering the frequency dispersion effect is constructed, and the initial value of the compensation coefficient is calculated based on the signal characteristic parameters; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameters; based on the initial value of the compensation coefficient and the initial value of the filter coefficient, the signal in the signal quality abnormal area is preliminarily processed, and the processing result is evaluated by a multi-dimensional evaluation index system to obtain the score change relative to the initial score, and determine the coefficient adjustment amount of the compensation coefficient and the filter coefficient; S103. Construct a deep reinforcement learning model, take the characteristic parameters of the signal test point as the state input, the coefficient adjustment amount as the action output, and the score change as the reward function to construct a deep reinforcement learning model; use the deep reinforcement learning model to dynamically adjust the compensation coefficient and the filter coefficient based on the state input and the reward function to obtain the optimized compensation coefficient and the optimized filter coefficient, perform signal processing on the signal quality abnormal area, and conduct real-time evaluation, and feed back the evaluation results to the deep reinforcement learning model for iterative optimization; when the score of the multi-dimensional evaluation index system continuously reaches the preset threshold, it is determined that the signal integrity optimization of the circuit board is completed.

[0021] First, divide the circuit board into multiple signal integrity analysis areas, for example, it can be divided into 10 areas according to features such as functional modules or wiring density. Set multiple signal test points in each area, for example, set 20 test points in each area. Collect the signal characteristic parameters of these test points, including characteristic impedance value, signal transmission rate, signal rise time, and signal response curves at different operating temperatures such as 25℃, 45℃, and 65℃.

[0022] Next, an adaptive weighted clustering algorithm is used to classify and analyze the collected signal feature parameters. Specifically, each parameter is first normalized, and then the weighted Euclidean distance between the test points is calculated, and the weight is dynamically adjusted according to the importance of the parameter. The K-means algorithm is used for clustering, and the number of clusters is gradually increased from 2 until the clustering results are stable. By analyzing the clustering results, areas with abnormal signal quality are identified.

[0023] Then, a multi-dimensional evaluation index system is established, including signal integrity indicators (such as overshoot and undershoot percentage), crosstalk suppression indicators (such as near-end crosstalk and far-end crosstalk amplitude), signal timing indicators (such as transmission delay and jitter), and power loss indicators (such as voltage drop). Each indicator is given a weight, such as 0.4 for signal integrity, 0.3 for crosstalk suppression, 0.2 for signal timing, and 0.1 for power loss. The initial score of the abnormal signal quality area under this index system is calculated.

[0024] For the identified signal quality abnormal areas, a signal compensation model considering the frequency dispersion effect is constructed. This model is based on the transmission line theory and takes into account the influence of factors such as conductor loss and dielectric loss on the signal. According to the collected signal characteristic parameters, the initial value of the compensation coefficient is calculated, including the amplitude compensation coefficient and the phase compensation coefficient.

[0025] At the same time, a complementary recursive filter is designed to process the signal in the abnormal area. The filter is composed of a cascade of a forward filter and a backward filter, which can effectively suppress high-frequency noise while maintaining the fast response characteristics of the signal. The initial value of the filter coefficient is set according to the signal characteristic parameters, including the forward filter coefficient and the backward filter coefficient.

[0026] The initial compensation coefficient and filter coefficient are used to perform preliminary signal processing on the signal quality abnormal area. After processing, the signal characteristic parameters are collected again and the processing results are evaluated through a multi-dimensional evaluation index system to calculate the score change relative to the initial score. The adjustment amount of the compensation coefficient and filter coefficient is determined based on the score change.

[0027] Next, a deep reinforcement learning model is constructed, using a deep Q network structure. The characteristic parameters of the signal test point are used as state input, the coefficient adjustment amount is used as the action output, and the score change is used as the reward function. Specifically, the state space includes parameters such as characteristic impedance value, signal transmission rate, and signal rise time; the action space is the adjustment amount of the compensation coefficient and the filter coefficient, which is discrete into 11 levels (-0.5, -0.4, ..., 0.4, 0.5); the reward function is the difference between the score after processing and the score before processing.

[0028] Using this deep reinforcement learning model, the compensation coefficient and filter coefficient are dynamically adjusted based on the current state input and reward function. Each time an adjustment is made, the signal quality abnormal area is processed and evaluated in real time through a multi-dimensional evaluation index system. The evaluation results are fed back to the model as new states and rewards, and the optimization is continuously iterated. When the scores reach 95 points or above for 10 consecutive times, the signal integrity optimization of the circuit board is determined to be complete.

[0029] In this embodiment, an adaptive weighted clustering algorithm is used to classify and analyze the signal characteristic parameters, which can accurately identify the abnormal signal quality area and provide a precise target for subsequent optimization. At the same time, a multi-dimensional evaluation index system is established to comprehensively evaluate the signal quality and avoid the one-sided optimization caused by a single indicator. A signal compensation model considering the frequency dispersion effect is constructed, and a complementary recursive filter is designed, which can effectively compensate for signal loss and suppress noise, thereby improving signal quality. The deep reinforcement learning model is used to dynamically adjust the parameters, overcoming the problem of difficult parameter optimization in traditional methods. The deep reinforcement learning model is used to realize the adaptive optimization of signal processing parameters, and real-time evaluation and iterative optimization are performed, making the optimization process more intelligent and efficient. Ultimately, the precise control of circuit board impedance matching and the significant improvement of signal integrity are achieved.

[0030] In an optional implementation, the circuit board is divided into a plurality of signal integrity analysis areas, and a plurality of signal test points are set in each signal integrity analysis area, and signal characteristic parameters of the signal test points are collected, including characteristic impedance value, signal transmission rate, signal rise time, and signal response at different operating temperatures. Based on an adaptive weighted clustering algorithm, the signal characteristic parameters are classified and analyzed, and signal quality abnormal areas are identified. A multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing, and power consumption is established, and an initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system is calculated, including: Based on the signal test points, the nodes of the graph are determined, and based on the signal transmission path, the edges of the graph are determined, and the wiring topology of the circuit board is converted into a graph structure; the local features and global features of the signal test points are extracted, and the local features include characteristic impedance values, signal transmission rates, and signal rise times, and the global features include signal responses at different operating temperatures; the association strength between the signal test points is calculated by using a graph attention mechanism, and a region segmentation criterion is constructed based on the association strength to divide the circuit board into multiple signal integrity analysis regions; A multimodal sensor array is arranged in the signal integrity analysis area to collect signal characteristic parameters of the signal test point; a sampling strategy is determined according to the signal characteristic parameters, including a sampling rate and a sampling duration; a wavelet transform is used to perform time-frequency analysis on the signal characteristic parameters to extract time-varying characteristics; a three-dimensional distribution model of the signal characteristic parameters is constructed based on the spatial distribution of the multimodal sensor array; segmentation information of the signal integrity analysis area is combined with the three-dimensional distribution model to generate regional characteristic data; Construct an adaptive weighted clustering model and input regional feature data into the adaptive weighted clustering model; dynamically adjust the weight coefficient of the adaptive weighted clustering model based on the time-varying characteristics of signal feature parameters, classify and analyze signal feature parameters, and identify areas with abnormal signal quality; A multi-dimensional evaluation index system is established, including a signal integrity index, a crosstalk suppression index, a signal timing index, and a power consumption loss index; an index weight calculation model is constructed based on the time-varying characteristics of the signal characteristic parameters; the correlation strength of the signal test point is used as the basis for calculating the coupling degree between the indicators in the multi-dimensional evaluation index system, and the weight of each indicator in the multi-dimensional evaluation index system is dynamically adjusted according to the signal characteristic parameters; combined with the identification results of the abnormal signal quality area, the initial score of the abnormal signal quality area based on the multi-dimensional evaluation index system is calculated.

[0031] Specifically, the circuit board is first divided into multiple signal integrity analysis areas, the nodes of the graph are determined based on the signal test points, the edges of the graph are determined based on the signal transmission path, and the wiring topology of the circuit board is converted into a graph structure. Specifically, each signal test point can be regarded as a node of the graph, and the signal line connecting two signal test points can be regarded as an edge of the graph. For example, for a circuit board with 100 signal test points, a graph structure containing 100 nodes can be constructed.

[0032] Next, extract the local and global features of the signal test point. Local features include characteristic impedance value, signal transmission rate, and signal rise time. For example, the characteristic impedance value (such as 50 ohms), signal transmission rate (such as 1Gbps), and signal rise time (such as 100ps) of each signal test point can be measured. Global features include signal response at different operating temperatures. The signal response can be measured at different temperatures such as -40°C, 25°C, and 85°C to obtain the amplitude, phase, and other characteristics of the signal at different temperatures.

[0033] Then, the graph attention mechanism is used to calculate the correlation strength between signal test points. Specifically, the similarity matrix between nodes can be calculated and the attention coefficient can be obtained by normalizing the similarity matrix, and then the correlation strength between signal test points can be calculated. Based on the correlation strength, a regional segmentation criterion is constructed to divide the circuit board into multiple signal integrity analysis areas. For example, signal test points with a correlation strength greater than 0.8 can be divided into the same area.

[0034] A multimodal sensor array is set up in the divided signal integrity analysis area to collect the signal characteristic parameters of the signal test points. The multimodal sensor array can include voltage probes, current probes, temperature sensors, etc. to comprehensively collect the characteristic parameters of the signal such as voltage, current, and temperature. The sampling strategy is determined according to the signal characteristic parameters, including the sampling rate and sampling duration. For example, for a 1Gbps high-speed signal, a sampling rate of 10GS / s can be used, and the sampling duration is 1ms.

[0035] Wavelet transform is used to perform time-frequency analysis on the characteristic parameters of the signal and extract the time-varying characteristics. Daubechies wavelet can be used to decompose the signal into 5 layers to obtain the characteristics of the energy distribution of the signal in different frequency bands over time. Based on the spatial distribution of the multimodal sensor array, a three-dimensional distribution model of the signal characteristic parameters is constructed. For example, the x and y coordinates can represent the position of the sensor on the circuit board, and the z coordinate can represent the value of the signal characteristic parameter to construct a three-dimensional distribution of the parameters.

[0036] Combine the segmentation information of the signal integrity analysis area with the three-dimensional distribution model to generate regional feature data. Statistics such as the mean and variance of the feature parameters can be calculated for each region as the feature data of the region. Construct an adaptive weighted clustering model and input the regional feature data into the model. The model can use an improved K-means algorithm to dynamically adjust the feature weights according to the importance of the feature.

[0037] Dynamically adjust the weight coefficient of the adaptive weighted clustering model based on the time-varying characteristics of the signal feature parameters. For example, the coefficient of variation of each feature parameter can be calculated based on the time-varying characteristics obtained by wavelet transform, and the feature with a larger coefficient of variation is given a higher weight. Perform classification analysis on the signal feature parameters to identify areas with abnormal signal quality. Categories with a long distance from the center point of the clustering result can be identified as abnormal areas.

[0038] Establish a multi-dimensional evaluation index system, including signal integrity index, crosstalk suppression index, signal timing index and power consumption index. Signal integrity index can include signal amplitude, rise time, etc.; crosstalk suppression index can include near-end crosstalk, far-end crosstalk, etc.; signal timing index can include signal delay, jitter, etc.; power consumption index can include static power consumption, dynamic power consumption, etc.

[0039] Based on the time-varying characteristics of signal characteristic parameters, an indicator weight calculation model is constructed. The entropy weight method can be used to calculate the weight according to the information entropy of each indicator. The smaller the information entropy, the more effective information the indicator provides, and the higher the weight should be given. The correlation strength of the signal test point is used as the basis for calculating the coupling degree between the indicators in the multi-dimensional evaluation indicator system. The higher the correlation strength, the stronger the coupling between the indicators, and this coupling relationship should be considered when calculating the comprehensive score.

[0040] Dynamically adjust the weight of each indicator in the multi-dimensional evaluation index system according to the signal characteristic parameters. A sliding time window can be set to calculate the coefficient of variation of each indicator within the window. The indicator with a larger coefficient of variation is given a higher weight. Combined with the identification results of the abnormal signal quality area, calculate the initial score of the abnormal signal quality area based on the multi-dimensional evaluation index system. The comprehensive score can be calculated by weighted summation, and the weight is determined by the previous steps.

[0041] In this embodiment, by dividing the circuit board into multiple signal integrity analysis areas and setting multiple signal test points in each area, the signal quality of the circuit board can be analyzed more finely and comprehensively. This regionalized analysis method can accurately locate the signal quality abnormal area, which is conducive to targeted optimization and improvement. The signal characteristic parameters are classified and analyzed using an adaptive weighted clustering algorithm, and the signal quality abnormal area can be automatically identified. The algorithm can dynamically adjust the weight coefficient according to the time-varying characteristics of the signal characteristic parameters, thereby improving the accuracy and adaptability of anomaly detection. A multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power consumption is established, and the coupling relationship between the indicators is considered, so that the evaluation of signal quality is more comprehensive and accurate. By dynamically adjusting the indicator weights, this method can adapt to the signal quality evaluation needs under different working conditions.

[0042] In an optional implementation, constructing a region segmentation criterion based on the association strength includes: Acquire the initial correlation strength between the signal test points, map the initial correlation strength to an interval from zero to one according to the maximum and minimum value range, and obtain a first correlation strength; calculate the physical distance between the signal test points according to the spatial coordinates of the signal test points, and based on the physical distance, generate a distance attenuation coefficient that decreases as the physical distance increases, and multiply the distance attenuation coefficient by the first correlation strength to obtain a second correlation strength; extract the signal waveform features of the signal test points, calculate the waveform similarity between different signal test points, and obtain a waveform similarity coefficient, and multiply the waveform similarity coefficient by the second correlation strength to obtain a third correlation strength; Calculating the mean value of the third correlation strength as the initial correlation threshold; establishing a region division rule, and dividing the adjacent signal test points whose third correlation strength is greater than the initial correlation threshold into the same region according to the region division rule, to obtain an initial region division result; Calculate the maximum physical distance between signal test points in the same area. If the maximum physical distance exceeds a preset distance threshold, separate the signal test point with the farthest distance to form a new area to obtain the first optimized area division result; count the number of signal test points in each area in the first optimized area division result, divide the area where the number of signal test points exceeds the preset upper limit, and merge the area where the number of signal test points is lower than the preset lower limit to obtain the second optimized area division result; identify the change trend of the third correlation strength at the area boundary in the second optimized area division result, determine that the third correlation strength decreases from the inside of the area to the boundary, and if the decreasing trend is not met, adjust the area boundary to obtain the third optimized area division result; Calculating the average third correlation strength of the signal test points between adjacent areas, extracting the signal waveform features of the adjacent areas, and merging the adjacent areas if the similarity between the average third correlation strength and the signal waveform features exceeds a set threshold; The spatial distribution density of the signal test points in each area after merging and the third correlation strength distribution are statistically calculated to construct a regional uniformity evaluation parameter; the initial correlation threshold is adjusted according to the regional uniformity evaluation parameter, and the adjusted correlation threshold is reused for regional division, and the optimization process is repeatedly executed until the regional division result is stable.

[0043] Specifically, it is obtained by calculating the correlation coefficient or mutual information of signals at different test points. For example, for the signal sequence of two test points A and B, the Pearson correlation coefficient of the two test points can be calculated as the initial correlation strength. Then the initial correlation strength is mapped to the interval of 0 to 1 according to the maximum and minimum value range to obtain the first correlation strength. Specifically, a linear mapping method can be used, assuming that the maximum value of the initial correlation strength is max and the minimum value is min, then the first correlation strength can be expressed as: (initial correlation strength-min) / (max-min).

[0044] Next, the physical distance between the test points is calculated based on the spatial coordinates of the signal test points. The Euclidean distance calculation method can be used. Assuming that the three-dimensional coordinates of the two test points are (x1, y1, z1) and (x2, y2, z2), the physical distance between the two test points is: √((x1-x2) 2 +(y1-y2) 2 +(z1-z2) 2 ). Based on the calculated physical distance, a distance decay coefficient is generated that decreases as the distance increases. An exponential decay function can be used, such as e (-d / d0) , where d is the physical distance and d0 is the characteristic distance. The distance attenuation coefficient is multiplied by the first association strength to obtain the second association strength.

[0045] Then, the signal waveform features of the signal test point are extracted, which may include the amplitude, frequency, phase and other features of the signal. The similarity of the waveform features between different test points is calculated to obtain the waveform similarity coefficient. The similarity between the waveform feature vectors can be calculated using methods such as cosine similarity. The waveform similarity coefficient is multiplied by the second correlation strength to obtain the third correlation strength.

[0046] The mean value of the third correlation strength is calculated as the initial correlation threshold, and a region division rule is established to divide adjacent test points whose third correlation strength is greater than the initial correlation threshold into the same region to obtain an initial region division result.

[0047] Optimize the initial area division results. First, calculate the maximum physical distance between test points in the same area. If it exceeds the preset distance threshold (for example, 10 meters), separate the test points with the farthest distance to form a new area to obtain the first optimization result. Then count the number of test points in each area, split the areas with a number exceeding the preset upper limit (such as 100), and merge the areas with a number below the preset lower limit (such as 10) to obtain the second optimization result. Then identify the change trend of the third correlation strength at the boundary of the area to ensure that it decreases from the inside of the area to the boundary. If it does not meet the requirements, adjust the boundary of the area to obtain the third optimization result.

[0048] The average third correlation strength of the test points between adjacent regions is calculated, and the signal waveform features of the adjacent regions are extracted. If both the average third correlation strength and the waveform feature similarity exceed a set threshold (such as 0.8), the adjacent regions are merged.

[0049] The spatial distribution density and the third correlation strength distribution of the test points in each region after merging are statistically analyzed to construct regional uniformity evaluation parameters. The variance of the test point spatial coordinates and the third correlation strength can be calculated as a uniformity evaluation index. The initial correlation threshold is adjusted according to the uniformity evaluation parameter, for example, when the uniformity is poor, the threshold is appropriately lowered. The adjusted threshold is reused for regional division, and the optimization process is cyclically executed until the regional division result is stable, that is, the difference between two consecutive division results is less than the set threshold.

[0050] In this embodiment, by calculating the multi-level correlation strength, the signal correlation, spatial distance and waveform characteristics are comprehensively considered, making the regional division more reasonable and comprehensive. Adopting the iterative optimization method, the adaptive optimization of regional division is achieved through multiple adjustments, which improves the accuracy and stability of the division results. The uniformity evaluation parameter is introduced to ensure the uniformity and consistency of the division results, avoiding the problem of over-segmentation or merging.

[0051] In an optional implementation, based on the spatial distribution of the multimodal sensor array, constructing a three-dimensional distribution model of the signal characteristic parameters; combining the segmentation information of the signal integrity analysis area with the three-dimensional distribution model to generate regional characteristic data includes: Taking the plane of the circuit board as a reference, a rectangular coordinate system is established, the physical coordinate information of each sensor in the multimodal sensor array is obtained, and the physical coordinate information is mapped into the rectangular coordinate system to obtain the sensor mapping coordinates; Collect the original signal features of the multimodal sensor array, perform maximum and minimum value normalization processing on the original signal features to obtain a first signal feature; calculate the distance between adjacent sensor mapping coordinates, determine the grid division interval based on the distance, and perform grid division in the plane of the rectangular coordinate system to obtain a two-dimensional grid; count the value distribution range of the first signal feature, divide the value distribution range into multiple value layers, and superimpose the value layers on the basis of the two-dimensional grid to obtain a three-dimensional grid; calculate the distance from each grid point in the three-dimensional grid to the surrounding sensor mapping coordinates, calculate the spatial influence factor of the grid point based on the distance, and multiply the spatial influence factor by the first signal feature of the corresponding sensor to obtain a second signal feature; according to a preset sparsity threshold, identify the sparse distribution area of ​​the sensor mapping coordinates, and perform local weighted smoothing processing on the second signal features of the grid points in the sparse area to obtain a third signal feature; Acquire boundary information of a signal integrity analysis area, map the boundary information to a three-dimensional grid containing the third signal feature, and extract a set of grid points within each analysis area; Calculate the mean and variance of the third signal feature in the grid point set to obtain regional statistical parameters, calculate the spatial gradient of the third signal feature to obtain regional spatial parameters; combine the regional statistical parameters with the regional spatial parameters to form a regional feature description, calculate the Euclidean distance between the regional feature descriptions of adjacent analysis areas to obtain regional similarity; extract the position of the local extreme point of the third signal feature in the analysis area as the feature point coordinates, and combine the regional feature description, the regional similarity and the feature point coordinates to generate regional feature data.

[0052] The spatial impact factor specifically refers to the weight coefficient used to characterize the spatial correlation between the grid point and the surrounding sensors in the three-dimensional grid space. This factor is calculated based on the distance from the grid point to the sensor, reflecting the degree of influence of the distance of the spatial position on the signal characteristics. The closer the sensor is, the greater the influence on the grid point is, and the farther the sensor is, the smaller the influence on the grid point is. The weighted calculation of this factor can achieve a reasonable distribution of signal characteristics in space.

[0053] Based on the spatial distribution of the multimodal sensor array, a three-dimensional distribution model of signal characteristic parameters is constructed, and the segmentation information of the signal integrity analysis area is combined with the three-dimensional distribution model to generate a specific implementation method of regional characteristic data: First, a rectangular coordinate system is established based on the plane of the circuit board. The physical coordinate information of each sensor in the multimodal sensor array is obtained, such as the coordinates of sensor 1 are (10mm, 20mm), the coordinates of sensor 2 are (30mm, 40mm), etc. These physical coordinate information are mapped to the established rectangular coordinate system to obtain the mapping coordinates of the sensor. For example, the mapping coordinates of sensor 1 are (1, 2), and the mapping coordinates of sensor 2 are (3, 4).

[0054] Next, collect the original signal features of the multimodal sensor array. Assume that the original signal feature values ​​collected are 0.5V, 1.2V, 0.8V, etc. These original signal features are normalized to the maximum and minimum values ​​to obtain the first signal feature. The specific method is to subtract the minimum value from each original signal feature value, and then divide it by the difference between the maximum and minimum values. For example, the normalized first signal feature value may be 0.3, 0.9, 0.5, etc.

[0055] Then, the distance between the mapping coordinates of adjacent sensors is calculated. Taking the Euclidean distance as an example, the distance between sensor 1 and sensor 2 is calculated to be 2.83. The grid division interval is determined based on the calculated distance, such as taking 0.5 as the interval. Grid division is performed in the plane of the rectangular coordinate system to obtain a two-dimensional grid.

[0056] The numerical distribution range of the first signal feature is counted, such as 0 to 1. The numerical distribution range is divided into multiple numerical layers, for example, 10 layers, each layer having an interval of 0.1. These numerical layers are superimposed on the two-dimensional grid to obtain a three-dimensional grid.

[0057] For each grid point in the three-dimensional grid, calculate its distance to the mapped coordinates of the surrounding sensors. Based on these distances, calculate the spatial influence factor of the grid point. The inverse distance weighting method can be used, and the closer the sensor, the greater the influence. Multiply the spatial influence factor with the first signal feature of the corresponding sensor to obtain the second signal feature.

[0058] According to the preset sparsity threshold, if the distance between adjacent sensors is greater than 5, identify the sparse distribution area of ​​sensor mapping coordinates. Perform local weighted smoothing on the second signal features of the grid points in these sparse areas. A Gaussian smoothing algorithm can be used to perform weighted averaging on the grid points within a certain range around. After processing, the third signal feature is obtained.

[0059] Obtain the boundary information of the signal integrity analysis area, such as the boundary coordinates of a certain analysis area are (1, 1), (1, 3), (3, 3), (3, 1). Map these boundary information to a three-dimensional grid containing the third signal feature, and extract the grid point set in each analysis area.

[0060] Calculate the mean and variance of the third signal feature in the grid point set to obtain regional statistical parameters. For example, the mean of a certain analysis area is 0.6 and the variance is 0.04. Calculate the spatial gradient of the third signal feature to obtain regional spatial parameters. The central difference method can be used to calculate the gradient to obtain the gradient values ​​in the x and y directions.

[0061] Combine regional statistical parameters with regional spatial parameters to form regional feature descriptions. Calculate the Euclidean distance between regional feature descriptions of adjacent analysis regions to obtain regional similarity. Extract the local extreme point positions of the third signal feature in the analysis region as feature point coordinates. Finally, combine regional feature descriptions, regional similarity, and feature point coordinates to generate regional feature data.

[0062] In this embodiment, by constructing a three-dimensional distribution model of signal characteristic parameters, spatial visualization of data collected by a multimodal sensor array is achieved, which improves the intuitiveness and accuracy of data analysis. The use of grid division and spatial interpolation technology overcomes the data sparsity problem caused by uneven sensor distribution and enhances the robustness and applicability of the model. By calculating regional feature descriptions and similarities and extracting feature point coordinates, accurate quantification and comparison of signal features in different analysis areas are achieved, providing a reliable data basis for subsequent signal integrity analysis.

[0063] In an optional implementation, for the signal quality abnormal area, a signal compensation model considering frequency dispersion effect is constructed, and an initial value of a compensation coefficient is calculated based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter, including: Perform fast Fourier transform on the signal in the signal quality abnormality area to obtain the frequency domain amplitude spectrum and phase spectrum, calculate the transmission constant based on the dielectric constant, loss tangent, line width, line thickness and interlayer spacing of the transmission line, construct a frequency response function based on the transmission constant, and determine the amplitude attenuation and phase delay in the signal transmission process based on the frequency response function; Acquire signal characteristic parameters, map the rise time in the signal characteristic parameters to high-frequency compensation strength, map the overshoot to medium-frequency compensation strength, and map the jitter value to low-frequency compensation strength; construct a piecewise linear frequency compensation function, and determine the initial value of the compensation coefficient based on the high-frequency compensation strength, the medium-frequency compensation strength, and the low-frequency compensation strength, combined with amplitude attenuation and phase delay, wherein the compensation coefficient includes a phase compensation coefficient and an amplitude compensation coefficient; A complementary filter with a three-branch recursive structure is constructed, wherein the complementary filter comprises a high-pass filter branch, a medium-pass filter branch and a low-pass filter branch; the cutoff frequency of the complementary filter is determined based on the signal bandwidth, the quality factor of the complementary filter is determined based on the signal-to-noise ratio, and an initial value of the filter coefficient is obtained.

[0064] This embodiment provides a compensation processing method for signal quality abnormal areas. The method first constructs a signal compensation model that takes frequency dispersion effect into consideration, and calculates the initial value of the compensation coefficient based on the signal characteristic parameter. Then, based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter.

[0065] Specifically, a fast Fourier transform is performed on the signal in the signal quality abnormal area to obtain the frequency domain amplitude spectrum and phase spectrum. The transmission constant is calculated based on the parameters such as the dielectric constant, loss tangent, line width, line thickness and layer spacing of the transmission line. For example, for a microstrip line with a characteristic impedance of 50Ω and a dielectric constant of 4.4, its transmission constant can be expressed as α+jβ, where α is the attenuation constant and β is the phase constant. Based on the obtained transmission constant, a frequency response function H(f) is constructed. The frequency response function describes the amplitude attenuation and phase delay characteristics of the signal during transmission.

[0066] Next, obtain the signal characteristic parameters, including rise time, overshoot, and jitter value. Map the rise time to high-frequency compensation strength, for example, a rise time of 100ps corresponds to a high-frequency compensation strength of 0.8. Map the overshoot to the medium-frequency compensation strength, for example, an overshoot of 10% corresponds to a medium-frequency compensation strength of 0.6. Map the jitter value to the low-frequency compensation strength, for example, a jitter value of 20ps corresponds to a low-frequency compensation strength of 0.4. Construct a piecewise linear frequency compensation function G(f), which corresponds to different compensation strengths in the low-frequency, medium-frequency, and high-frequency ranges. Combine the amplitude attenuation and phase delay information in the frequency response function H(f) to determine the initial value of the compensation coefficient. The compensation coefficient includes a phase compensation coefficient and an amplitude compensation coefficient, which are used to correct the phase and amplitude characteristics of the signal, respectively.

[0067] Then, a complementary filter with a three-branch recursive structure is constructed, including three filtering branches: high-pass, mid-pass, and low-pass. The cutoff frequency of the complementary filter is determined based on the signal bandwidth. For example, for a 5Gbps signal, the cutoff frequency of the high-pass filter can be set to 3GHz, and the cutoff frequency of the low-pass filter can be set to 1GHz. The quality factor of the complementary filter is determined based on the signal-to-noise ratio. For example, for a signal-to-noise ratio of 20dB, the quality factor can be set to 0.707. In this way, the initial value of the filter coefficient is obtained.

[0068] In practical applications, the signal in the abnormal signal quality area is first sent to the complementary filter for processing. The high-pass filter branch is mainly used to compensate for the loss of high-frequency components, the medium-pass filter branch is used to adjust the intermediate frequency characteristics, and the low-pass filter branch is used to suppress low-frequency noise. The filtered signal is then multiplied by the compensation coefficient to achieve amplitude and phase compensation of the signal. Finally, the compensated signal is compared with the original signal, and the compensation coefficient and filter coefficient are dynamically adjusted according to the compensation effect to achieve the best signal quality improvement effect.

[0069] In this embodiment, by considering the frequency dispersion effect, a more accurate signal compensation model is constructed, which can effectively improve the transmission quality of high-speed signals. A complementary recursive filter structure is adopted to achieve separate processing of different frequency bands of the signal, thereby improving the flexibility and accuracy of compensation. The introduction of signal characteristic parameters to guide the setting of compensation coefficients and filter coefficients makes the compensation process more intelligent and adaptive, and can perform personalized processing according to different signal characteristics.

[0070] In an optional implementation, preliminary signal processing is performed on the signal quality abnormal area based on the initial value of the compensation coefficient and the initial value of the filter coefficient, and the processing result is evaluated by the multi-dimensional evaluation index system to obtain a score change relative to the initial score, and the coefficient adjustment amount of the compensation coefficient and the filter coefficient is determined, including: The signal in the signal quality abnormality area is processed through the frequency response function to obtain the transmitted signal, and the transmitted signal is processed through the frequency compensation function to obtain the compensated signal; the compensated signal is processed through the high-pass filter branch, the middle-pass filter branch and the low-pass filter branch respectively to obtain the high-pass output signal, the middle-pass output signal and the low-pass output signal; Performing weighted combination on the high-pass output signal, the mid-pass output signal and the low-pass output signal to obtain a preliminary processed signal, calculating the eye opening, signal-to-noise ratio and timing jitter of the preliminary processed signal to obtain a processed score; calculating the difference between the processed score and the initial score, determining the score change, and using the gradient descent method to calculate the adjustment direction of the compensation coefficient and the filter coefficient; The segmentation points of the frequency compensation function, the adjustment amount of the compensation strength, and the cutoff frequency and the adjustment amount of the quality factor of the complementary filter are calculated respectively, and applied to the current compensation coefficient and the filter coefficient to obtain a new parameter combination, and the signal processing is re-executed using the new parameter combination to obtain a new score change; when the new score change is less than a preset threshold, the parameter optimization is stopped to obtain the optimized compensation coefficient and the filter coefficient.

[0071] The eye opening refers to the degree of opening in the vertical direction of the signal eye diagram, which indicates the fault tolerance of the signal at the sampling time. The larger the opening, the better the signal discrimination and the stronger the anti-interference ability; the smaller the opening, the worse the signal quality and the more likely it is to produce bit errors.

[0072] The cutoff frequency specifically refers to the frequency corresponding to when the signal amplitude attenuates to 70.7% of the maximum value of the passband (i.e. -3dB point) on the filter frequency response characteristic curve. Above and below the corresponding frequency point, the filter's signal passing characteristics change significantly, which can be used to distinguish between the passband and the stopband.

[0073] The quality factor specifically refers to the selectivity index of the filter, indicating the steepness of the filter's attenuation characteristics near the cutoff frequency. The higher the quality factor, the better the frequency selectivity of the filter, but at the same time it may introduce greater phase distortion; the lower the quality factor, the smoother the filtering characteristics and the smaller the phase distortion.

[0074] The specific implementation method of performing preliminary signal processing on the signal quality abnormal area based on the initial value of the compensation coefficient and the initial value of the filter coefficient is as follows: First, the signal in the abnormal signal quality area is processed through the frequency response function to obtain the transmitted signal. The frequency response function can simulate the attenuation and distortion of the signal during the transmission process. For example, for a high-speed signal with a bandwidth of 10GHz, the frequency response function can be set to a low-pass characteristic, maintaining an amplitude of 1 in the frequency range below 10GHz and gradually attenuating the frequency above 10GHz.

[0075] Next, the transmitted signal is processed through the frequency compensation function to obtain a compensated signal. The frequency compensation function is used to compensate for the attenuation and distortion caused by transmission. Taking the above 10 GHz bandwidth signal as an example, the frequency compensation function can be set to perform gain compensation in the frequency range above 10 GHz to make the overall frequency response of the signal flat.

[0076] Then, the compensated signal is processed through a high-pass filter branch, a medium-pass filter branch, and a low-pass filter branch, respectively, to obtain a high-pass output signal, a medium-pass output signal, and a low-pass output signal. The cutoff frequency of the high-pass filter branch can be set to 7GHz, the passband of the medium-pass filter branch is 3-7GHz, and the cutoff frequency of the low-pass filter branch is 3GHz. In this way, the signal can be divided into three parts: high frequency, medium frequency, and low frequency for independent processing.

[0077] The high-pass output signal, the mid-pass output signal and the low-pass output signal are weighted and combined to obtain a preliminary processed signal. The weighting coefficient can be adjusted according to actual conditions, for example, the weight ratio of high-pass: mid-pass: low-pass can be set to 1:2:1.

[0078] Calculate the eye opening, signal-to-noise ratio and timing jitter of the preliminary processed signal to get the score after processing. The eye opening can be obtained by calculating the vertical and horizontal opening degree of the eye diagram; the signal-to-noise ratio can be calculated by the ratio of signal power to noise power; the timing jitter can be obtained by the time deviation statistics of the signal zero transition. Combine these three indicators with a certain weight to get a comprehensive score.

[0079] The difference between the processed score and the initial score is calculated to determine the score change. The gradient descent method is used to calculate the adjustment direction of the compensation coefficient and the filter coefficient. For example, if the score increases, the adjustment continues in the current adjustment direction; if the score decreases, the parameters are adjusted in the opposite direction.

[0080] Calculate the segmentation points of the frequency compensation function, the adjustment amount of the compensation strength, and the cutoff frequency and quality factor of the complementary filter. The size of the adjustment amount can be proportional to the change in the score. For example, if the score increases by 1%, each parameter is adjusted by 0.1%. Apply the calculated adjustment amount to the current compensation coefficient and filter coefficient to obtain a new parameter combination.

[0081] The signal processing is performed again using the new parameter combination to obtain a new score change. The above process is repeated until the new score change is less than a preset threshold (such as 0.1%), and the parameter optimization is stopped to obtain the optimized compensation coefficient and filter coefficient.

[0082] In this embodiment, by combining frequency compensation and multi-branch filtering, the signal characteristics in the signal quality abnormal area can be effectively improved, the eye opening degree and signal-to-noise ratio of the signal can be improved, and the timing jitter can be reduced, thereby improving the overall quality of the signal. By adopting multi-dimensional evaluation indicators and gradient descent optimization methods, the compensation and filtering parameters can be adaptively adjusted so that the processing results gradually tend to the optimal, avoiding the blindness and inefficiency of manual debugging. This method has good versatility and scalability, and can flexibly adjust the evaluation indicators and optimization strategies according to the actual application scenarios, and is suitable for various types of signal quality improvement needs.

[0083] In an optional implementation, a deep reinforcement learning model is constructed, and the characteristic parameters of the signal test point are used as state input, the coefficient adjustment amount is used as action output, and the score change is used as a reward function to construct a deep reinforcement learning model; using the deep reinforcement learning model, based on the state input and the reward function, the compensation coefficient and the filter coefficient are dynamically adjusted to obtain an optimized compensation coefficient and an optimized filter coefficient, and signal processing is performed on the abnormal signal quality area, and real-time evaluation is performed, and the evaluation results are fed back to the deep reinforcement learning model for iterative optimization; when the score of the multi-dimensional evaluation index system continuously reaches a preset threshold, determining that the signal integrity optimization of the circuit board is completed includes: Standardize the characteristic parameters of the signal test points, establish a symbol description mapping table based on the standardized characteristic parameters, determine the mapping relationship, and build an initial parameter adjustment rule base, wherein the initial parameter adjustment rule base includes compensation coefficient adjustment strategies and filter coefficient adjustment strategies under different characteristic combinations; The compensation coefficient, the filter coefficient and the standardized feature parameter are used as nodes of a causal dependency graph, and the compensation coefficient adjustment strategy and the filter coefficient adjustment strategy are used as edges of the causal dependency graph to construct a causal dependency graph; a graph attention network is constructed, wherein the input layer of the graph attention network receives the causal dependency graph, and the weight of the edge is calculated by using the attention mechanism through the message passing layer, the information transmission between the nodes is determined, and the compensation coefficient adjustment amount and the filter coefficient adjustment amount are obtained; Establishing a semantic mapping relationship between the initial parameter adjustment rule base and the compensation coefficient adjustment amount and the filter coefficient adjustment amount, and verifying the consistency of the rule reasoning result and the network output result through a bidirectional constraint mechanism; based on the consistency verification result, updating the initial parameter adjustment rule base to obtain an optimized parameter adjustment rule base; Processing the signal according to the compensation coefficient adjustment amount and the filter coefficient adjustment amount to obtain characteristic parameters of the processed signal; calculating a score change value of the characteristic parameters of the processed signal; Using the score change value as a reward value, updating the weights of the edges in the causal dependency graph and the parameters of the graph attention network; performing a counterfactual analysis on the compensation coefficient adjustment amount and the filter coefficient adjustment amount based on the score change value to obtain a potential adjustment result; updating the adjustment strategy in the optimization parameter adjustment rule base according to the potential adjustment result, and dynamically adjusting the structure of the causal dependency graph; and re-optimizing the parameters based on the updated causal dependency graph and the optimization parameter adjustment rule base; When the score reaches the preset threshold value K times in a row, the current compensation coefficient is output as the optimized compensation coefficient, and the current filtering coefficient is output as the optimized filtering coefficient, thereby completing the signal integrity optimization method.

[0084] First, the characteristic parameters of the signal test point are standardized. Specifically, the maximum and minimum normalization method can be used to map each characteristic parameter to the interval [0, 1]. For example, for the rise time feature, the minimum value can be set to 0ns and the maximum value to 10ns, then the actual measured 5ns rise time can be standardized to 0.5.

[0085] Next, a symbol description mapping table is established based on the standardized feature parameters. The continuous numerical features are discretized into several levels, such as "low", "medium", "high", etc., and the corresponding relationship between the numerical interval and the symbol description is established. For example, [0, 0.3] can be mapped to "low", [0.3, 0.7] can be mapped to "medium", and [0.7, 1] can be mapped to "high".

[0086] Then, an initial parameter adjustment rule base is constructed, which includes compensation coefficient and filter coefficient adjustment strategies under different feature combinations. The rule can be expressed as: "If the rise time is low and the overshoot is high, increase the compensation coefficient by 0.1 and decrease the filter coefficient by 0.05". The initial rule base can be set according to expert experience.

[0087] The compensation coefficient, filter coefficient and standardized characteristic parameter are used as nodes of the causal dependency graph, and the adjustment strategy is used as the edge to construct the causal dependency graph. For example, "rise time", "overshoot" and "compensation coefficient" can be used as nodes, and "increase compensation coefficient" can be used as an edge to connect the corresponding nodes.

[0088] Construct a graph attention network, where the input layer receives the causal dependency graph. Through the message passing layer, the attention mechanism is used to calculate the edge weights. The softmax function can be used to normalize the edge weights to ensure that the sum of the weights is 1. According to the calculated weights, the information transfer between nodes is determined, and finally the adjustment amount of the compensation coefficient and the filter coefficient is obtained.

[0089] Establish a semantic mapping relationship between the initial parameter adjustment rule base and the network output results. For example, the compensation coefficient adjustment amount of the network output is 0.15, which can be mapped to "increase the compensation coefficient by 0.1" in the rule base. Through the two-way constraint mechanism, verify the consistency of the rule reasoning results and the network output results. If there is an inconsistency, update the rule base according to the network output results to obtain the optimized parameter adjustment rule base.

[0090] The signal is processed according to the adjustment amount of the compensation coefficient and the filter coefficient to obtain the characteristic parameters of the processed signal. The score change value of the characteristic parameters before and after processing is calculated and fed back to the reinforcement learning model as a reward value. For example, if the rise time score is improved from 80 points to 90 points after processing, the reward value is 10.

[0091] Based on the reward value, the weights of the edges in the causal dependency graph and the parameters of the graph attention network are updated. The gradient descent method can be used to optimize the network parameters so that the network output results are more consistent with the actual effect. At the same time, a counterfactual analysis is performed on the adjustment amount of the compensation coefficient and the filter coefficient, assuming the possible results of using different adjustment amounts, and obtaining potential adjustment results.

[0092] Update the adjustment strategy in the optimization parameter adjustment rule base according to the potential adjustment results. For example, if the analysis finds that increasing the compensation coefficient by 0.2 is more effective, the original rule "increase the compensation coefficient by 0.1" is modified to "increase the compensation coefficient by 0.2". At the same time, the structure of the causal dependency graph is dynamically adjusted, which may add new edges or delete invalid edges.

[0093] Based on the updated causal dependency graph and the optimization parameter adjustment rule base, the parameters are re-optimized. Repeat the above steps to continuously iterate and optimize the compensation coefficient and filter coefficient. When the score reaches the preset threshold (such as 90 points) for K consecutive times (such as K=5), the current compensation coefficient and filter coefficient are output as the final optimization result, completing the signal integrity optimization process.

[0094] In this embodiment, by constructing a deep reinforcement learning model, the signal integrity optimization process is intelligent and automated, greatly improving the optimization efficiency. The introduction of causal dependency graphs and graph attention networks effectively captures the complex relationship between parameters and improves the accuracy and interpretability of the optimization results. The use of a two-way constraint mechanism and counterfactual analysis achieves the coordinated optimization of the rule base and the deep learning model, enhancing the robustness and generalization ability of the model.

[0095] Figure 2 FIG. 1 is a schematic diagram of the structure of the circuit board impedance matching and signal integrity optimization system according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to divide the circuit board into multiple signal integrity analysis areas, and set multiple signal test points in each signal integrity analysis area, collect signal characteristic parameters of the signal test points, including characteristic impedance value, signal transmission rate, signal rise time and signal response under different working temperatures, classify and analyze the signal characteristic parameters based on an adaptive weighted clustering algorithm, identify signal quality abnormal areas, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power consumption loss, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system; The second unit is used to construct a signal compensation model that takes into account the frequency dispersion effect for the signal quality abnormal area, and calculate the initial value of the compensation coefficient based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter; based on the initial value of the compensation coefficient and the initial value of the filter coefficient, the signal quality abnormal area is preliminarily processed, and the processing result is evaluated by the multi-dimensional evaluation index system to obtain the score change relative to the initial score, and determine the coefficient adjustment amount of the compensation coefficient and the filter coefficient; The third unit is used to construct a deep reinforcement learning model, which takes the characteristic parameters of the signal test point as the state input, the coefficient adjustment amount as the action output, and the score change as the reward function to construct the deep reinforcement learning model; using the deep reinforcement learning model, based on the state input and the reward function, dynamically adjust the compensation coefficient and the filter coefficient to obtain the optimized compensation coefficient and the optimized filter coefficient, perform signal processing on the signal quality abnormal area, and perform real-time evaluation, and feed back the evaluation result to the deep reinforcement learning model for iterative optimization; when the score of the multi-dimensional evaluation index system continuously reaches the preset threshold, it is determined that the signal integrity optimization of the circuit board is completed.

[0096] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0097] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0098] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A circuit board impedance matching and signal integrity optimization method, characterized in that: include: Divide the circuit board into multiple signal integrity analysis areas, and set multiple signal test points in each signal integrity analysis area, collect signal characteristic parameters of the signal test points, including characteristic impedance value, signal transmission rate, signal rise time, and signal response under different operating temperatures, classify and analyze the signal characteristic parameters based on an adaptive weighted clustering algorithm, identify signal quality abnormal areas, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing, and power consumption, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system; For the signal quality abnormal area, a signal compensation model considering the frequency dispersion effect is constructed, and the initial value of the compensation coefficient is calculated based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter; Performing preliminary signal processing on the signal quality abnormal area based on the initial value of the compensation coefficient and the initial value of the filter coefficient, and evaluating the processing result through the multi-dimensional evaluation index system to obtain the score change relative to the initial score, and determining the coefficient adjustment amount of the compensation coefficient and the filter coefficient; A deep reinforcement learning model is constructed, and the characteristic parameters of the signal test points are used as state inputs, coefficient adjustment amounts are used as action outputs, and score changes are used as reward functions to construct the deep reinforcement learning model; using the deep reinforcement learning model, based on the state inputs and the reward function, the compensation coefficients and the filter coefficients are dynamically adjusted to obtain optimized compensation coefficients and optimized filter coefficients, signal processing is performed on areas with abnormal signal quality, and real-time evaluation is performed, and the evaluation results are fed back to the deep reinforcement learning model for iterative optimization; when the scores of the multi-dimensional evaluation index system continuously reach a preset threshold, it is determined that the signal integrity optimization of the circuit board is completed.

2. The method according to claim 1, characterized in that The circuit board is divided into multiple signal integrity analysis areas, and multiple signal test points are set in each signal integrity analysis area. The signal characteristic parameters of the signal test points are collected, including characteristic impedance value, signal transmission rate, signal rise time and signal response under different operating temperatures. Based on the adaptive weighted clustering algorithm, the signal characteristic parameters are classified and analyzed to identify the signal quality abnormal area, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power consumption loss, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system, including: Based on the signal test points, the nodes of the graph are determined, and based on the signal transmission path, the edges of the graph are determined, and the wiring topology of the circuit board is converted into a graph structure; the local features and global features of the signal test points are extracted, and the local features include characteristic impedance values, signal transmission rates, and signal rise times, and the global features include signal responses at different operating temperatures; the association strength between the signal test points is calculated by using a graph attention mechanism, and a region segmentation criterion is constructed based on the association strength to divide the circuit board into multiple signal integrity analysis regions; A multimodal sensor array is arranged in the signal integrity analysis area to collect signal characteristic parameters of the signal test point; a sampling strategy is determined according to the signal characteristic parameters, including a sampling rate and a sampling duration; a wavelet transform is used to perform time-frequency analysis on the signal characteristic parameters to extract time-varying characteristics; a three-dimensional distribution model of the signal characteristic parameters is constructed based on the spatial distribution of the multimodal sensor array; segmentation information of the signal integrity analysis area is combined with the three-dimensional distribution model to generate regional characteristic data; Construct an adaptive weighted clustering model and input regional feature data into the adaptive weighted clustering model; dynamically adjust the weight coefficient of the adaptive weighted clustering model based on the time-varying characteristics of signal feature parameters, classify and analyze signal feature parameters, and identify areas with abnormal signal quality; A multi-dimensional evaluation index system is established, including a signal integrity index, a crosstalk suppression index, a signal timing index, and a power consumption loss index; an index weight calculation model is constructed based on the time-varying characteristics of the signal characteristic parameters; the correlation strength of the signal test point is used as the basis for calculating the coupling degree between the indicators in the multi-dimensional evaluation index system, and the weight of each indicator in the multi-dimensional evaluation index system is dynamically adjusted according to the signal characteristic parameters; combined with the identification results of the abnormal signal quality area, the initial score of the abnormal signal quality area based on the multi-dimensional evaluation index system is calculated.

3. The method according to claim 2, characterized in that The region segmentation criteria constructed based on the association strength include: Acquire the initial correlation strength between the signal test points, map the initial correlation strength to an interval from zero to one according to the maximum and minimum value range, and obtain a first correlation strength; calculate the physical distance between the signal test points according to the spatial coordinates of the signal test points, and based on the physical distance, generate a distance attenuation coefficient that decreases as the physical distance increases, and multiply the distance attenuation coefficient by the first correlation strength to obtain a second correlation strength; extract the signal waveform features of the signal test points, calculate the waveform similarity between different signal test points, and obtain a waveform similarity coefficient, and multiply the waveform similarity coefficient by the second correlation strength to obtain a third correlation strength; Calculating the mean value of the third correlation strength as the initial correlation threshold; establishing a region division rule, and dividing the adjacent signal test points whose third correlation strength is greater than the initial correlation threshold into the same region according to the region division rule, to obtain an initial region division result; Calculate the maximum physical distance between signal test points in the same area. If the maximum physical distance exceeds a preset distance threshold, separate the signal test point with the farthest distance to form a new area to obtain the first optimized area division result; count the number of signal test points in each area in the first optimized area division result, divide the area where the number of signal test points exceeds the preset upper limit, and merge the area where the number of signal test points is lower than the preset lower limit to obtain the second optimized area division result; identify the change trend of the third correlation strength at the area boundary in the second optimized area division result, determine that the third correlation strength decreases from the inside of the area to the boundary, and if the decreasing trend is not met, adjust the area boundary to obtain the third optimized area division result; Calculating the average third correlation strength of the signal test points between adjacent areas, extracting the signal waveform features of the adjacent areas, and merging the adjacent areas if the similarity between the average third correlation strength and the signal waveform features exceeds a set threshold; The spatial distribution density of the signal test points in each area after merging and the third correlation strength distribution are statistically calculated to construct a regional uniformity evaluation parameter; the initial correlation threshold is adjusted according to the regional uniformity evaluation parameter, and the adjusted correlation threshold is reused for regional division, and the optimization process is repeatedly executed until the regional division result is stable.

4. The method according to claim 2, characterized in that: Based on the spatial distribution of the multimodal sensor array, constructing a three-dimensional distribution model of the signal characteristic parameters; combining the segmentation information of the signal integrity analysis area with the three-dimensional distribution model to generate regional characteristic data includes: Taking the plane of the circuit board as a reference, a rectangular coordinate system is established, the physical coordinate information of each sensor in the multimodal sensor array is obtained, and the physical coordinate information is mapped into the rectangular coordinate system to obtain the sensor mapping coordinates; Collect the original signal features of the multimodal sensor array, perform maximum and minimum value normalization processing on the original signal features to obtain a first signal feature; calculate the distance between adjacent sensor mapping coordinates, determine the grid division interval based on the distance, and perform grid division in the plane of the rectangular coordinate system to obtain a two-dimensional grid; count the value distribution range of the first signal feature, divide the value distribution range into multiple value layers, and superimpose the value layers on the basis of the two-dimensional grid to obtain a three-dimensional grid; calculate the distance from each grid point in the three-dimensional grid to the surrounding sensor mapping coordinates, calculate the spatial influence factor of the grid point based on the distance, and multiply the spatial influence factor by the first signal feature of the corresponding sensor to obtain a second signal feature; according to a preset sparsity threshold, identify the sparse distribution area of ​​the sensor mapping coordinates, and perform local weighted smoothing processing on the second signal features of the grid points in the sparse area to obtain a third signal feature; Acquire boundary information of a signal integrity analysis area, map the boundary information to a three-dimensional grid containing the third signal feature, and extract a set of grid points within each analysis area; Calculate the mean and variance of the third signal feature in the grid point set to obtain regional statistical parameters, calculate the spatial gradient of the third signal feature to obtain regional spatial parameters; combine the regional statistical parameters with the regional spatial parameters to form a regional feature description, calculate the Euclidean distance between the regional feature descriptions of adjacent analysis areas to obtain regional similarity; extract the position of the local extreme point of the third signal feature in the analysis area as the feature point coordinates, and combine the regional feature description, the regional similarity and the feature point coordinates to generate regional feature data.

5. The method according to claim 1, characterized in that: For the signal quality abnormal area, a signal compensation model considering the frequency dispersion effect is constructed, and an initial value of the compensation coefficient is calculated based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter, including: Perform fast Fourier transform on the signal in the signal quality abnormality area to obtain the frequency domain amplitude spectrum and phase spectrum, calculate the transmission constant based on the dielectric constant, loss tangent, line width, line thickness and interlayer spacing of the transmission line, construct a frequency response function based on the transmission constant, and determine the amplitude attenuation and phase delay in the signal transmission process based on the frequency response function; Acquire signal characteristic parameters, map the rise time in the signal characteristic parameters to high-frequency compensation strength, map the overshoot to medium-frequency compensation strength, and map the jitter value to low-frequency compensation strength; construct a piecewise linear frequency compensation function, and determine the initial value of the compensation coefficient based on the high-frequency compensation strength, the medium-frequency compensation strength, and the low-frequency compensation strength, combined with amplitude attenuation and phase delay, wherein the compensation coefficient includes a phase compensation coefficient and an amplitude compensation coefficient; A complementary filter with a three-branch recursive structure is constructed, wherein the complementary filter comprises a high-pass filter branch, a medium-pass filter branch and a low-pass filter branch; the cutoff frequency of the complementary filter is determined based on the signal bandwidth, the quality factor of the complementary filter is determined based on the signal-to-noise ratio, and an initial value of the filter coefficient is obtained.

6. The method according to claim 5, characterized in that Based on the initial value of the compensation coefficient and the initial value of the filter coefficient, the signal quality abnormal area is initially processed, and the processing result is evaluated by the multi-dimensional evaluation index system to obtain the score change relative to the initial score. The coefficient adjustment amount of the compensation coefficient and the filter coefficient is determined, including: The signal in the signal quality abnormality area is processed through the frequency response function to obtain the transmitted signal, and the transmitted signal is processed through the frequency compensation function to obtain the compensated signal; the compensated signal is processed through the high-pass filter branch, the middle-pass filter branch and the low-pass filter branch respectively to obtain the high-pass output signal, the middle-pass output signal and the low-pass output signal; Performing weighted combination on the high-pass output signal, the mid-pass output signal and the low-pass output signal to obtain a preliminary processed signal, calculating the eye opening, signal-to-noise ratio and timing jitter of the preliminary processed signal to obtain a processed score; calculating the difference between the processed score and the initial score, determining the score change, and using the gradient descent method to calculate the adjustment direction of the compensation coefficient and the filter coefficient; The segmentation points of the frequency compensation function, the adjustment amount of the compensation strength, and the cutoff frequency and the adjustment amount of the quality factor of the complementary filter are calculated respectively, and applied to the current compensation coefficient and the filter coefficient to obtain a new parameter combination, and the signal processing is re-executed using the new parameter combination to obtain a new score change; when the new score change is less than a preset threshold, the parameter optimization is stopped to obtain the optimized compensation coefficient and the filter coefficient.

7. The method according to claim 1, characterized in that A deep reinforcement learning model is constructed, and the characteristic parameters of the signal test point are used as state input, the coefficient adjustment amount is used as action output, and the score change is used as a reward function to construct the deep reinforcement learning model; using the deep reinforcement learning model, based on the state input and the reward function, the compensation coefficient and the filter coefficient are dynamically adjusted to obtain the optimized compensation coefficient and the optimized filter coefficient, and the signal quality abnormal area is processed and evaluated in real time, and the evaluation result is fed back to the deep reinforcement learning model for iterative optimization; When the score of the multi-dimensional evaluation index system continuously reaches a preset threshold, determining that the signal integrity optimization of the circuit board is completed includes: Standardize the characteristic parameters of the signal test points, establish a symbol description mapping table based on the standardized characteristic parameters, determine the mapping relationship, and build an initial parameter adjustment rule base, wherein the initial parameter adjustment rule base includes compensation coefficient adjustment strategies and filter coefficient adjustment strategies under different characteristic combinations; The compensation coefficient, the filter coefficient and the standardized feature parameter are used as nodes of a causal dependency graph, and the compensation coefficient adjustment strategy and the filter coefficient adjustment strategy are used as edges of the causal dependency graph to construct a causal dependency graph; a graph attention network is constructed, wherein the input layer of the graph attention network receives the causal dependency graph, and the weight of the edge is calculated by using the attention mechanism through the message passing layer, the information transmission between the nodes is determined, and the compensation coefficient adjustment amount and the filter coefficient adjustment amount are obtained; Establishing a semantic mapping relationship between the initial parameter adjustment rule base and the compensation coefficient adjustment amount and the filter coefficient adjustment amount, and verifying the consistency of the rule reasoning result and the network output result through a bidirectional constraint mechanism; based on the consistency verification result, updating the initial parameter adjustment rule base to obtain an optimized parameter adjustment rule base; Processing the signal according to the compensation coefficient adjustment amount and the filter coefficient adjustment amount to obtain characteristic parameters of the processed signal; calculating a score change value of the characteristic parameters of the processed signal; Using the score change value as a reward value, updating the weights of the edges in the causal dependency graph and the parameters of the graph attention network; performing a counterfactual analysis on the compensation coefficient adjustment amount and the filter coefficient adjustment amount based on the score change value to obtain a potential adjustment result; updating the adjustment strategy in the optimization parameter adjustment rule base according to the potential adjustment result, and dynamically adjusting the structure of the causal dependency graph; and re-optimizing the parameters based on the updated causal dependency graph and the optimization parameter adjustment rule base; When the score reaches the preset threshold value K times in a row, the current compensation coefficient is output as the optimized compensation coefficient, and the current filtering coefficient is output as the optimized filtering coefficient, thereby completing the signal integrity optimization method.

8. A circuit board impedance matching and signal integrity optimization system, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to divide the circuit board into multiple signal integrity analysis areas, and set multiple signal test points in each signal integrity analysis area, collect signal characteristic parameters of the signal test points, including characteristic impedance value, signal transmission rate, signal rise time and signal response under different working temperatures, classify and analyze the signal characteristic parameters based on an adaptive weighted clustering algorithm, identify signal quality abnormal areas, establish a multi-dimensional evaluation index system including signal integrity, crosstalk suppression, signal timing and power consumption loss, and calculate the initial score of the signal quality abnormal area based on the multi-dimensional evaluation index system; The second unit is used to construct a signal compensation model considering the frequency dispersion effect for the signal quality abnormal area, and calculate the initial value of the compensation coefficient based on the signal characteristic parameter; based on the signal compensation model, the signal in the signal quality abnormal area is processed by a complementary recursive filter, and the initial value of the filter coefficient is set according to the signal characteristic parameter; Performing preliminary signal processing on the signal quality abnormal area based on the initial value of the compensation coefficient and the initial value of the filter coefficient, and evaluating the processing result through the multi-dimensional evaluation index system to obtain the score change relative to the initial score, and determining the coefficient adjustment amount of the compensation coefficient and the filter coefficient; The third unit is used to construct a deep reinforcement learning model, which takes the characteristic parameters of the signal test point as the state input, the coefficient adjustment amount as the action output, and the score change as the reward function to construct the deep reinforcement learning model; using the deep reinforcement learning model, based on the state input and the reward function, dynamically adjust the compensation coefficient and the filter coefficient to obtain the optimized compensation coefficient and the optimized filter coefficient, perform signal processing on the signal quality abnormal area, and perform real-time evaluation, and feed back the evaluation result to the deep reinforcement learning model for iterative optimization; when the score of the multi-dimensional evaluation index system continuously reaches the preset threshold, it is determined that the signal integrity optimization of the circuit board is completed.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images