Training Method and Reflection Prediction Method for PCB Signal Reflection Prediction Model

Abstract

The present invention relates to the technical field of electronic circuits, and provides a method for training a reflection prediction model of PCB signals and a reflection prediction method. Among them, the method for training a reflection prediction model of PCB signals includes: constructing a training set, where the training set includes: a plurality of training samples and corresponding true values; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data samples are sample information of the driving end, receiving end, and transmission line of the connection sequence; inputting the training samples into a preset convolutional neural network to perform signal reflection prediction and obtain predicted values; based on the gap between the predicted values and the corresponding true values, performing iterative training on the convolutional neural network to obtain a trained reflection prediction model. This trained reflection prediction model can be used for predicting PCB signal reflections of PCB boards, and has strong universality and generality and low cost.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic circuits, and in particular, to a method for training a reflection prediction model of PCB signals and a reflection prediction method. Background Art

[0002] With the continuous improvement of the manufacturing process of PCB (Printed Circuit Board), the signal integrity problem of PCB boards has received increasing attention. Signal integrity refers to the signal starting from the transmitting end (driver end), passing through the transmission line structure, and reaching the receiving end in a lossless state and being correctly received. However, lossless transmission of signals is usually impossible. During the transmission process, the signal will inevitably be interfered with, resulting in a deterioration of the signal transmission quality. What affects the signal integrity of PCB signals are usually problems such as crosstalk and reflection.

[0003] Currently, for the prediction of PCB signal reflection, existing reflection simulation software is usually used to perform simulation modeling for different PCB boards, so as to predict the reflection of PCB signals. However, this method is inconvenient to implement, has a high cost, and has weak versatility. Summary of the Invention

[0004] The present invention provides a method for training a reflection prediction model of PCB signals and a reflection prediction method, which are used to solve the problems of inconvenient prediction, high cost, and weak versatility of PCB signal reflection in the prior art.

[0005] The present invention provides a method for training a reflection prediction model of PCB signals, including:

[0006] Constructing a training set, the training set includes: a plurality of training samples and corresponding true values; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data samples are sample information of the driver end, receiving end, and transmission line of the connection sequence;

[0007] Inputting the training samples into a preset convolutional neural network to perform signal reflection prediction to obtain predicted values;

[0008] Based on the gap between the predicted values and the corresponding true values, iteratively training the convolutional neural network to obtain the trained reflection prediction model.

[0009] Optionally, the step of obtaining the training samples includes:

[0010] Obtain all the signal reflection correlation data samples of the connection sequence, where the signal reflection correlation data samples include: the termination impedance value samples of the driving end, the rise time samples and / or fall time samples of the PCB signals at the driving end; the parallel termination resistance value samples and parallel termination capacitance value samples of the receiving end; the total time delay samples and characteristic impedance value samples of the transmission line;

[0011] Combine and normalize all the signal reflection correlation data samples to obtain the training samples, where the connection sequence corresponds to the training samples one by one;

[0012] When the number of the obtained training samples is greater than or equal to the preset sample number threshold, complete the acquisition of the training samples.

[0013] Optionally, the termination impedance value samples are obtained by averaging the output impedance value samples of the active devices and the resistance value samples of the passive devices in the connection sequence;

[0014] The parallel termination capacitance value samples are obtained by averaging the capacitance value samples between the input pins of the active devices connected in parallel at the receiving end and the ground, and the capacitance value samples of the passive devices;

[0015] The total time delay samples are obtained by summing the time delay samples of each section of the transmission line in the connection sequence.

[0016] Optionally, the steps of obtaining the true values corresponding to the training samples include:

[0017] Use a preset reflection simulation software to perform reflection simulation on multiple connection sequences to obtain signal reflection judgment results, and determine the signal reflection judgment results as the true values of the corresponding connection sequences.

[0018] Optionally, the convolutional neural network includes: an input layer, at least two feature extraction structures, a flattening layer, a fully connected layer, and an output layer connected in sequence;

[0019] The feature extraction structure includes: at least two convolutional layers for feature extraction and a max pooling layer connected in sequence.

[0020] The present invention also provides a PCB signal reflection prediction method, including:

[0021] Obtain the signal reflection correlation data of the connection sequence of the PCB board to be predicted, where the signal reflection correlation data is the information of the driving end, receiving end and transmission line of the PCB board to be predicted;

[0022] Combining and normalizing all the signal reflection correlation data of any of the said connection sequences to obtain the sequence data to be predicted;

[0023] Inputting the sequence data to be predicted into the reflection prediction model as described in any of the above items to perform signal reflection prediction and obtain a reflection prediction result.

[0024] The present invention also provides a training system for a reflection prediction model of PCB signals, including:

[0025] A training set construction module for constructing a training set, where the training set includes: a plurality of training samples and corresponding true values; the training samples are obtained by combining and normalizing all the signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data samples are the sample information of the driving end, receiving end, and transmission line of the connection sequence;

[0026] A prediction module for inputting the training samples into a preset convolutional neural network to perform signal reflection prediction and obtain predicted values;

[0027] A training module for iteratively training the convolutional neural network based on the gap between the predicted values and the corresponding true values to obtain the trained reflection prediction model.

[0028] The present invention also provides a PCB signal reflection prediction system, including:

[0029] An associated data acquisition module for acquiring the signal reflection correlation data of the connection sequence of the PCB board to be predicted, where the signal reflection correlation data is the information of the driving end, receiving end, and transmission line of the PCB board to be predicted;

[0030] A processing module for combining and normalizing all the signal reflection correlation data of any of the said connection sequences to obtain the sequence data to be predicted;

[0031] A reflection prediction module for inputting the sequence data to be predicted into the reflection prediction model as described in any of the above items to perform signal reflection prediction and obtain a reflection prediction result.

[0032] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it implements the reflection prediction model training method of the PCB signal as described in any of the above, or the PCB signal reflection prediction method as described above.

[0033] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for training a reflection prediction model of PCB signals as described in any one of the above, or the method for predicting PCB signal reflections as described above.

[0034] Advantages of the present invention: The present invention provides a method for training a reflection prediction model of PCB signals and a method for predicting reflections. Among them, the method for training a reflection prediction model of PCB signals constructs a training set, which includes: a plurality of training samples and corresponding true values; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data samples are sample information of the driving end, receiving end, and transmission line of the connection sequence; the training samples are input into a preset convolutional neural network for signal reflection prediction to obtain predicted values; based on the gap between the predicted values and the corresponding true values, the convolutional neural network is iteratively trained to obtain a trained reflection prediction model. This trained reflection prediction model can be used for predicting PCB signal reflections on the PCB board, with relatively high prediction accuracy, strong universality and generality, and low cost. Description of the Drawings

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0036] Figure 1 It is a flowchart showing the method for training a reflection prediction model of PCB signals provided by the present invention;

[0037] Figure 2 It is a flowchart showing the process of obtaining training samples in the method for training a reflection prediction model of PCB signals provided by the present invention;

[0038] Figure 3 It is a structural diagram of the convolutional neural network or the reflection prediction model in the method for training a reflection prediction model of PCB signals provided by the present invention;

[0039] Figure 4 It is a flowchart showing the method for predicting PCB signal reflections provided by the present invention;

[0040] Figure 5 It is a structural diagram of the system for training a reflection prediction model of PCB signals provided by the present invention;

[0041] Figure 6It is a schematic structural diagram of the PCB signal reflection prediction system provided by the present invention;

[0042] Figure 7 It is a schematic structural diagram of the electronic device provided by the present invention. Specific embodiments

[0043] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0044] Below, in the form of embodiments, in combination with Figures 1-7 Describe the reflection prediction model training method and reflection prediction method of the PCB signal provided by the present invention.

[0045] Please refer to Figure 1 , the reflection prediction model training method of the PCB signal provided by this embodiment includes:

[0046] S110: Construct a training set, the training set includes: a plurality of training samples and true values corresponding to the training samples; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data sample is sample information of the driving end, receiving end and transmission line of the connection sequence.

[0047] It should be noted that the PCB board usually includes one or more connection sequences, and the connection sequence refers to a sequence including a driving end, a transmission line and a receiving end, that is, a sequence capable of completing the transmission of PCB signals. The signal reflection correlation data sample refers to the data sample related to the signal reflection problem in the connection sequence. By constructing a training set in step S110, it is convenient to train the convolutional neural network based on the training set later.

[0048] S120: Input the training samples into a preset convolutional neural network for signal reflection prediction to obtain prediction values. By setting the convolutional neural network in this step, it helps to improve the accuracy of signal reflection prediction.

[0049] S130: Iteratively train the convolutional neural network based on the gap between the predicted value and the corresponding true value to obtain the trained reflection prediction model. That is, iteratively train the convolutional neural network based on the gap between the predicted value and the corresponding true value until convergence to obtain the final reflection prediction model. The PCB signal reflection prediction model training method in this embodiment can obtain a trained reflection prediction model. This reflection prediction model can be used for PCB signal reflection prediction of PCB boards, with relatively high prediction accuracy, strong universality and generality, and low cost.

[0050] It should be noted that signal reflection means that when a signal is transmitted, it encounters an instantaneous impedance of a transmission line at each moment. When the instantaneous impedance changes, a part of the signal will be reflected and another part will continue to be transmitted forward. Signal reflection will have a greater impact on the signal transmission quality. In the prior art, for signal reflection prediction, reflection simulation software is usually used to separately simulate and model each PCB board to complete the reflection prediction. This method brings great inconvenience to signal reflection prediction, and has high cost and poor generality. The PCB signal reflection prediction model training method in this embodiment, by adopting the above method, obtains a trained reflection prediction model, which can better overcome the above problems and improve the universality of reflection prediction.

[0051] Please refer to Figure 2 , in some embodiments, the step of obtaining the training samples includes:

[0052] S210: Obtain all the signal reflection associated data samples of the connection sequence. The signal reflection associated data samples include: the termination impedance value samples of the driver end, the rise time samples and / or fall time samples of the PCB signal at the driver end; the parallel termination resistance value samples and parallel termination capacitance value samples at the receiver end; the total time delay samples and characteristic impedance value samples of the transmission line. Specifically, the parallel termination resistance value samples at the receiver end refer to the resistance values of the resistors connected in parallel at the receiver end.

[0053] In some embodiments, the termination impedance value samples are obtained by averaging the output impedance value samples of the active devices and the resistance value samples of the passive devices in the connection sequence. Specifically, denote the output impedance value samples of the active devices as a x R x , x = 1, 2... n, and denote the resistance value samples of the passive devices as n y R y , y = 1, 2... k. Then the mathematical expression for obtaining the termination impedance value samples is:

[0054] R termial =(a1R1 + a2R2 + … + a nR n + n1R1 + n2R2 + … + n k R k ) / (n + k)

[0055] Among them, R termial is the termination impedance value sample. By adopting the above-mentioned averaging method to obtain the termination impedance value sample, it can better avoid the influence of the impedance value or resistance value of one or several devices being too large on the accuracy of the entire reflection prediction.

[0056] In some embodiments, the parallel termination capacitance value sample is obtained by averaging the capacitance value samples between the input pins of the active devices and the ground in the parallel termination at the receiving end, as well as the capacitance value samples of the passive devices. Specifically, denote the capacitance value sample between the input pin of the active device and the ground as a x C x , and the capacitance value sample of the passive device as n y C y . Then the mathematical expression for obtaining the parallel termination capacitance value sample is:

[0057] C receiving = (a1C1 + a2C2 + … + a n C n + n1C1 + n2C2 + … + n k C k ) / (n + k)

[0058] Among them, C receiving is the parallel termination capacitance value sample. By adopting the above-mentioned averaging method to obtain the parallel termination capacitance value sample, it can better avoid the influence of the capacitance value samples of one or several devices being too large on the accuracy of the entire reflection prediction.

[0059] In some embodiments, the total time delay sample is obtained by summing the time delay samples of each section of the transmission line in the connection sequence. Specifically, denote that there are h transmission lines in the connection sequence. Then the mathematical expression for obtaining the total time delay sample is:

[0060] d = d1 + d2 + … + d h

[0061] Among them, d refers to the total time delay sample, and d1, d2 … d h refer to the time delay samples of h transmission lines.

[0062] It should be noted that the acquisition method or calculation method of the characteristic impedance value sample of different types of transmission lines is different.

[0063] Specifically, in the case where the transmission line is a coaxial cable, the calculation formula for its characteristic impedance value sample is:

[0064] z = 138 * ln(D / f)

[0065] Wherein, z represents the characteristic impedance value sample, D is the outer diameter of the coaxial cable, and f is the inner diameter of the coaxial cable.

[0066] In the case where the transmission line is a microstrip line, the calculation formula for the characteristic impedance value sample is:

[0067]

[0068] Wherein, εr is the dielectric constant of the microstrip line, W is the width of the microstrip line, and U is the thickness of the microstrip line.

[0069] In the case where the transmission line is a twisted pair, the calculation formula for the characteristic impedance value sample is:

[0070]

[0071] Wherein, j is the wire diameter of the twisted pair, and b is the outer diameter of the insulation layer of the twisted pair.

[0072] S220: Combine and normalize all the signal reflection correlation data samples to obtain the training samples, and the connection sequence corresponds to the training samples one by one.

[0073] Specifically, the mathematical expression for combining all the signal reflection correlation data samples is:

[0074] data = [T, Rtermial, Rreceiving, Creceiving, d, z]

[0075] Wherein, data represents all the combined signal reflection correlation data samples, T represents the rise time sample and / or fall time sample of the driving end, and R receiving represents the parallel termination resistance value sample.

[0076] Furthermore, normalize this combination so that the range of each data is between 0 and 1, which is convenient for subsequent reflection prediction. Specifically, the normalization process is carried out in the following manner:

[0077] X = log(i, 10) / log(max, 10)

[0078] Among them, X represents the data obtained after normalization, log() represents the logarithmic function, i represents any value in the data of data, and max represents the maximum value in the data of data. The above method uses the log function with base 10 to obtain the log values corresponding to i and max respectively, and takes the ratio of these two log values as the normalized number, so that the range of the normalized value is kept between 0 and 1, which helps to perform reflection prediction subsequently. By using the non-linear log function to normalize the data of data, it can effectively take into account both the larger values and the smaller values in the data of data, without affecting the overall proportion problem, and to a certain extent helps to improve the accuracy of subsequent reflection prediction.

[0079] S230: When the number of the obtained training samples is greater than or equal to the preset sample number threshold, the acquisition of the training samples is completed. That is, repeat the steps of S210 - S220 above to obtain training samples of the connection sequences of multiple PCB boards. The size of the sample number threshold can be set according to the actual situation and will not be elaborated here. By using the above method to obtain training samples, it helps to improve the accuracy of subsequent reflection prediction.

[0080] In some embodiments, the step of obtaining the true value corresponding to the training sample includes:

[0081] Using a preset reflection simulation software to perform reflection simulation on multiple said connection sequences to obtain a signal reflection judgment result, and determining the signal reflection judgment result as the true value corresponding to the said connection sequence. Specifically, the reflection simulation software is such as HyperLynx (a simulation tool). By using the reflection simulation software to perform reflection simulation on the connection sequences on multiple PCB boards, the relationship data between the maximum amplitude and the original amplitude (the initial amplitude detected at the driving end) of the PCB signal can be obtained. Based on this relationship data, the signal reflection judgment result can be obtained. For example: if the maximum amplitude is greater than 10% of the original amplitude, it is determined that the signal reflection judgment result is that signal reflection has occurred or reflection has occurred.

[0082] In order to improve the reflection prediction accuracy of the convolutional neural network or the reflection prediction model, this embodiment proposes that the above convolutional neural network or reflection prediction model includes: an input layer (input), at least two feature extraction structures, a flattening layer (Flatten), a fully connected layer (Dense), and an output layer (output) connected in sequence.

[0083] The feature extraction structure includes: at least two consecutive convolution layers for feature extraction, and a max pooling layer. Specifically, the convolution layer is a one-dimensional convolution layer for feature extraction. By adding a max pooling layer after at least two convolution layers, the main features can be retained to reduce the computational amount. Each convolution layer can use ReLU (a non-linear activation function) as the activation function to improve the accuracy of the convolution layer.

[0084] The flattening layer serves as a transition between the max pooling layer and the fully connected layer, and is used to convert multi-dimensional data into one-dimensional data. The fully connected layer is used to map the output vector to a numerical value. The fully connected layer is a 1*1 fully connected layer. The output layer uses the sigmoid function to control the output of the model between 0 and 1.

[0085] Exemplarily, please refer to Figure 3 , Figure 3 where the convolutional neural network or reflection prediction model includes: an input layer, two feature extraction structures (the first feature extraction structure (the first convolution layer, the second convolution layer, and the first max pooling layer), the second feature extraction structure (the third convolution layer, the fourth convolution layer, and the second max pooling layer)), a flattening layer, a fully connected layer, and an output layer. By adopting the above network structure, the hidden features in the input data can be deeply extracted and finally mapped to a numerical output, which corresponds to or refers to the corresponding reflection prediction result. For example, when the output value is greater than 0.1, the reflection prediction result is that a reflection problem has occurred; when the output value is less than 0.1, the reflection prediction result is that no reflection problem has occurred. To a certain extent, the accuracy of the reflection prediction model is improved.

[0086] Furthermore, the convolutional neural network is trained using the following loss function:

[0087]

[0088] where E represents the number of training samples, y pred represents the predicted value, and y true represents the true value. By adopting the above loss function, it is helpful to improve the reflection prediction accuracy of the reflection prediction model.

[0089] In addition, the obtained training samples and the corresponding true values form a data set. In the specific implementation process, 80% of the data in this data set can be used as the training set, and 20% of the data can be used as the test set. The test set is used to test the trained reflection prediction model, thereby improving the accuracy of the reflection prediction model.

[0090] The reflection prediction model obtained in the above embodiments can be applied to the PCB signal reflection prediction of different PCB boards, with high prediction accuracy, strong universality, low cost, no need for simulation modeling, and reduced time cost.

[0091] Exemplarily, the above method for training the reflection prediction model of PCB signals will be described below with a specific embodiment.

[0092] First, a training set is constructed. Specifically, the training set includes: a plurality of training samples and the true values corresponding to the training samples; the training samples are obtained by combining and normalizing all the signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data samples are the sample information of the driver end, receiver end, and transmission line of the connection sequence; the signal reflection correlation data samples include: the termination impedance value sample of the driver end, the rise time sample and / or fall time sample of the PCB signal of the driver end; the parallel termination resistance value sample and parallel termination capacitance value sample of the receiver end; the total delay sample and characteristic impedance value sample of the transmission line.

[0093] Secondly, a convolutional neural network is built. Specifically, the convolutional neural network includes: an input layer, at least two feature extraction structures, a flattening layer, a fully connected layer, and an output layer connected in sequence; the feature extraction structure includes: at least two convolutional layers for feature extraction connected in sequence and a max pooling layer.

[0094] Then, the training samples are input into a preset convolutional neural network for signal reflection prediction to obtain predicted values.

[0095] Finally, based on the gap between the predicted values and the corresponding true values, the convolutional neural network is iteratively trained to obtain the trained reflection prediction model. By obtaining this trained reflection prediction model, when a new PCB board needs to be predicted for reflection problems, the trained reflection prediction model can be used for signal reflection prediction, without the need for simulation modeling for different PCB boards, which is more convenient to implement, has strong universality, and high accuracy.

[0096] Please refer to Figure 4 , this embodiment also provides a method for predicting PCB signal reflection, including:

[0097] S410: Obtain the signal reflection correlation data of the connection sequence of the PCB board to be predicted, where the signal reflection correlation data is the information of the driver end, receiver end, and transmission line of the PCB board to be predicted.

[0098] In some embodiments, the signal reflection correlation data includes: the termination impedance value of the PCB signal at the driving end, as well as the rise time and / or fall time; the parallel termination resistance value and parallel termination capacitance value at the receiving end; the total time delay and characteristic impedance value of the transmission line.

[0099] S420: Combine and normalize all the signal reflection correlation data of any one of the connection sequences to obtain the sequence data to be predicted. By combining and normalizing all the signal reflection correlation data of any one of the connection sequences in this step to obtain the sequence data to be predicted, it is convenient for subsequent signal reflection prediction.

[0100] The normalization process in this embodiment also uses a non-linear log function, which can more effectively take into account both the larger and smaller values of the data and will not affect the overall proportion problem.

[0101] S430: Input the sequence data to be predicted into the reflection prediction model as described in any one of the above, perform signal reflection prediction, and obtain the reflection prediction result. By inputting the sequence data to be predicted into the reflection prediction model as described in any one of the above in this embodiment for signal reflection prediction, a reflection prediction result with higher accuracy can be obtained.

[0102] Further, after the step of inputting the sequence data to be predicted into the reflection prediction model as described in any one of the above, performing signal reflection prediction, and obtaining the reflection prediction result, it further includes:

[0103] a. In the case where the reflection prediction result indicates reflection, label the connection sequence corresponding to the reflection prediction result as a problematic connection sequence.

[0104] b. Based on the value output by the reflection prediction model in the reflection prediction result and a preset prompt information matching strategy, match the prompt information corresponding to the value. It can be understood that the larger the value output by the reflection prediction model, the higher the severity of the reflection problem in the corresponding problematic connection sequence. In this embodiment, different severity levels of reflection problems correspond to different prompt information, which can help relevant personnel make corresponding corrections or improvements to the problematic connection sequences with different severity levels.

[0105] c. Based on the problematic connection sequence and the prompt information, conduct targeted warnings. Specifically, send the problematic connection sequence and the prompt information to an associated terminal, such as a mobile phone, computer, etc., to facilitate relevant personnel to make corresponding corrections to the problematic connection sequence.

[0106] In some embodiments, the prompt information is sequence correction guidance information, which is used to guide relevant personnel to make corresponding sequence corrections or improvements according to the severity of the reflection problem in the problematic connection sequence.

[0107] In some embodiments, the step of obtaining the termination impedance value includes:

[0108] Obtain the output impedance value of the active device and the resistance value of the passive device in the connection sequence of the PCB to be predicted; perform an averaging process on all the output impedance values and the resistance values to obtain the termination impedance value.

[0109] In some embodiments, the step of obtaining the parallel termination capacitance value of the receiving end includes:

[0110] Obtain the first capacitance value between the input pin of the active device and the ground and the second capacitance value of the passive device in the parallel termination of the receiving end;

[0111] Perform an averaging process on all the first capacitance values and the second capacitance values to obtain the parallel termination capacitance value.

[0112] In some embodiments, the step of obtaining the total time delay of the transmission line includes:

[0113] Sum the time delays of each section of the transmission line in the connection sequence; determine the obtained sum value as the total time delay.

[0114] The PCB signal reflection prediction model training system provided by the present invention will be described below. The PCB signal reflection prediction model training system described below can be correspondingly referred to the PCB signal reflection prediction model training method described above.

[0115] Please refer to Figure 5 , this embodiment also provides a PCB signal reflection prediction model training system, including:

[0116] A training set construction module 510, configured to construct a training set, where the training set includes: a plurality of training samples and true values corresponding to the training samples; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data samples are sample information of the driving end, receiving end, and transmission line of the connection sequence;

[0117] A prediction module 520, configured to input the training samples into a preset convolutional neural network for signal reflection prediction to obtain prediction values;

[0118] A training module 530 is configured to iteratively train the convolutional neural network based on the gap between the predicted value and the corresponding true value, so as to obtain the trained reflection prediction model. The training set construction module 510, the prediction module 520, and the training module 530 are connected. The PCB signal reflection prediction model training system in this embodiment can obtain a reflection prediction model with relatively high reflection prediction accuracy. The trained reflection prediction model can be used for PCB signal reflection prediction of a PCB board, with relatively high prediction accuracy, strong universality and generality, low cost, and strong feasibility.

[0119] In some embodiments, the training set construction module 510 includes: a training sample acquisition unit, configured to acquire all the signal reflection correlation data samples of the connection sequence. The signal reflection correlation data samples include: the termination impedance value samples of the driving end, the rise time samples and / or fall time samples of the PCB signal at the driving end; the parallel termination resistance value samples and parallel termination capacitance value samples of the receiving end; the total time delay samples and characteristic impedance value samples of the transmission line. Combine and normalize all the signal reflection correlation data samples to obtain the training samples, and the connection sequence corresponds to the training samples one by one. When the number of the obtained training samples is greater than or equal to a preset sample number threshold, the acquisition of the training samples is completed.

[0120] In some embodiments, the training set construction module 510 further includes: a true value acquisition unit, configured to use a preset reflection simulation software to perform reflection simulation on a plurality of the connection sequences, obtain a signal reflection judgment result, and determine the signal reflection judgment result as the true value of the corresponding connection sequence.

[0121] The PCB signal reflection prediction system provided by the present invention will be described below. The PCB signal reflection prediction system described below can be mutually corresponding and referred to with the PCB signal reflection prediction method described above.

[0122] Please refer to Figure 6 , this embodiment further provides a PCB signal reflection prediction system, including:

[0123] An association data acquisition module 610 is configured to acquire signal reflection association data of a connection sequence of a PCB board to be predicted. The signal reflection association data is information on the driving end, receiving end, and transmission line of the PCB board to be predicted.

[0124] A processing module 620 is configured to combine and normalize all the signal reflection association data of any one of the connection sequences to obtain data of a sequence to be predicted.

[0125] A reflection prediction module 630, configured to input the sequence data to be predicted into the reflection prediction model as described in any one of the above, perform signal reflection prediction, and obtain a reflection prediction result. The correlation data acquisition module 610, the processing module 620, and the reflection prediction module 630 are connected.

[0126] In some embodiments, the correlation data acquisition module 610 includes:

[0127] A termination impedance value acquisition unit, configured to acquire the termination impedance value of the PCB signal at the driving end;

[0128] A rise / fall time acquisition unit, configured to acquire the rise time and / or fall time of the PCB signal at the driving end;

[0129] A parallel termination resistor value acquisition unit, configured to acquire the parallel termination resistor value at the receiving end;

[0130] A parallel termination capacitor value acquisition unit, configured to acquire the parallel termination capacitor value at the receiving end;

[0131] An overall time delay acquisition unit, configured to acquire the overall time delay of the transmission line;

[0132] A characteristic impedance value acquisition unit, configured to acquire the characteristic impedance value of the transmission line.

[0133] In some embodiments, the termination impedance value acquisition unit is specifically configured to acquire the output impedance value of the active device and the resistance value of the passive device in the connection sequence of the PCB to be predicted; perform an averaging process on all the output impedance values and the resistance values to obtain the termination impedance value.

[0134] In some embodiments, the parallel termination capacitor value acquisition unit is specifically configured to acquire the first capacitor value between the input pin of the active device and the ground and the second capacitor value of the passive device that are connected in parallel at the receiving end; perform an averaging process on all the first capacitor values and the second capacitor values to obtain the parallel termination capacitor value.

[0135] In some embodiments, the overall time delay acquisition unit is specifically configured to sum the time delays of each segment of the transmission line in the connection sequence; determine the obtained sum value as the overall time delay.

[0136] In some embodiments, the PCB signal reflection prediction system further includes: a warning module, configured to, when the reflection prediction result indicates reflection occurs, label the connection sequence corresponding to the reflection prediction result as a problematic connection sequence; match, based on the value output by the reflection prediction model in the reflection prediction result and a preset hint information matching strategy, hint information corresponding to the value; and perform targeted warning based on the problematic connection sequence and the hint information.

[0137] Figure 7 An exemplary physical structure diagram of an electronic device is shown as Figure 7 shown. The electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communication bus 740. Among them, the processor 710, the communications interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call logical instructions in the memory 730 to execute the training method of the reflection prediction model for PCB signals or the PCB signal reflection prediction method. The training method of the reflection prediction model for PCB signals includes: constructing a training set, where the training set includes: a plurality of training samples and true values corresponding to the training samples; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in the PCB board, and the signal reflection correlation data sample is sample information of the driving end, receiving end, and transmission line of the connection sequence; inputting the training samples into a preset convolutional neural network to perform signal reflection prediction to obtain a predicted value; and performing iterative training on the convolutional neural network based on the gap between the predicted value and the corresponding true value to obtain a trained reflection prediction model. The PCB signal reflection prediction method includes: obtaining signal reflection correlation data of the connection sequence of the PCB board to be predicted, where the signal reflection correlation data is information on the driving end, receiving end, and transmission line of the PCB board to be predicted; combining and normalizing all signal reflection correlation data of any connection sequence to obtain data of the sequence to be predicted; and inputting the data of the sequence to be predicted into the reflection prediction model as described in any one of the above to perform signal reflection prediction to obtain a reflection prediction result.

[0138] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0139] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the PCB signal reflection prediction model training method or the PCB signal reflection prediction method provided by the above-mentioned various methods.

[0140] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0141] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, optical discs, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0142] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for training a reflection prediction model of PCB signals, characterized in that, Including: Constructing a training set, the training set including: a plurality of training samples and true values corresponding to the training samples; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in a PCB board, and the signal reflection correlation data samples are sample information of a driving end, a receiving end, and a transmission line of the connection sequence; The steps of obtaining the training samples include: Obtaining all the signal reflection correlation data samples of the connection sequence, the signal reflection correlation data samples including: a termination impedance value sample of the driving end, a rise time sample and / or a fall time sample of the PCB signal of the driving end; a parallel termination resistance value sample and a parallel termination capacitance value sample of the receiving end; a total time delay sample and a characteristic impedance value sample of the transmission line; Combining and normalizing all the signal reflection correlation data samples to obtain the training samples, and the connection sequence corresponds to the training sample one by one; When the number of the obtained training samples is greater than or equal to a preset sample number threshold, the obtaining of the training samples is completed; The termination impedance value sample is obtained by averaging an output impedance value sample of an active device and a resistance value sample of a passive device in the connection sequence; The parallel termination capacitance value sample is obtained by averaging a capacitance value sample between an input pin of the active device connected in parallel at the receiving end and the ground and a capacitance value sample of the passive device; The total time delay sample is obtained by summing up the time delay samples of each section of the transmission line in the connection sequence; Inputting the training samples into a preset convolutional neural network for signal reflection prediction to obtain a predicted value; Based on the gap between the predicted value and the corresponding true value, iteratively training the convolutional neural network to obtain the trained reflection prediction model.

2. The method for training a reflection prediction model of PCB signals according to claim 1, characterized in that, The steps of obtaining the true value corresponding to the training samples include: Using a preset reflection simulation software to perform reflection simulation on a plurality of the connection sequences to obtain a signal reflection judgment result, and determining the signal reflection judgment result as the true value of the corresponding connection sequence.

3. The method for training a reflection prediction model of PCB signals according to claim 1, characterized in that, The convolutional neural network includes: an input layer, at least two feature extraction structures, a flattening layer, a fully connected layer, and an output layer connected in sequence; The feature extraction structure includes: at least two convolutional layers for feature extraction and a max pooling layer connected in sequence.

4. A method for predicting PCB signal reflection, characterized in that, Including: Obtaining signal reflection correlation data of a connection sequence of a PCB board to be predicted, the signal reflection correlation data being information of a driving end, a receiving end, and a transmission line of the PCB board to be predicted; Combining and normalizing all the signal reflection correlation data of any one of the connection sequences to obtain data of a sequence to be predicted; Inputting the data of the sequence to be predicted into the reflection prediction model according to any one of claims 1 to 3 for signal reflection prediction to obtain a reflection prediction result.

5. A system for training a reflection prediction model of PCB signals, characterized in that, Including: A training set construction module for constructing a training set, where the training set includes: a plurality of training samples and corresponding true values; the training samples are obtained by combining and normalizing all signal reflection correlation data samples of any connection sequence in a PCB board, and the signal reflection correlation data samples are sample information of the driving end, receiving end, and transmission line of the connection sequence; the steps of obtaining the training samples include: obtaining all the signal reflection correlation data samples of the connection sequence, where the signal reflection correlation data samples include: the termination impedance value sample of the driving end, the rise time sample and / or fall time sample of the PCB signal of the driving end; the parallel termination resistance value sample and parallel termination capacitance value sample of the receiving end; the total time delay sample and characteristic impedance value sample of the transmission line; combining and normalizing all the signal reflection correlation data samples to obtain the training samples, and the connection sequence corresponds to the training sample one by one; when the number of the obtained training samples is greater than or equal to a preset sample number threshold, the obtaining of the training samples is completed; the termination impedance value sample is obtained by averaging the output impedance value sample of the active device and the resistance value sample of the passive device in the connection sequence; the parallel termination capacitance value sample is obtained by averaging the capacitance value sample between the input pin of the active device connected in parallel at the receiving end and the ground and the capacitance value sample of the passive device; the total time delay sample is obtained by summing the time delay samples of each section of the transmission line in the connection sequence; A prediction module for inputting the training samples into a preset convolutional neural network to perform signal reflection prediction and obtain prediction values; A training module for iteratively training the convolutional neural network based on the difference between the prediction values and the corresponding true values to obtain the trained reflection prediction model.

6. A system for predicting PCB signal reflection, characterized in that, It includes: An associated data acquisition module for acquiring signal reflection associated data of the connection sequence of the PCB board to be predicted, where the signal reflection associated data is information of the driving end, receiving end, and transmission line of the PCB board to be predicted; A processing module for combining and normalizing all the signal reflection associated data of any connection sequence to obtain data of the sequence to be predicted; A reflection prediction module for inputting the data of the sequence to be predicted into the reflection prediction model according to any one of claims 1 to 3 to perform signal reflection prediction and obtain a reflection prediction result.

7. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for training the reflection prediction model of the PCB signal according to any one of claims 1 to 3, or the method for predicting the reflection of the PCB signal according to claim 4.

8. A non-transitory computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by the processor, it implements the method for training the reflection prediction model of the PCB signal according to any one of claims 1 to 3, or the method for predicting the reflection of the PCB signal according to claim 4.

Images