Method and device for determining scattering parameters based on deep learning, medium and product

Abstract

The invention provides a method and device for determining scattering parameters based on deep learning, a medium and a product, and the method comprises the following steps: obtaining to-be-processed scattering parameters obtained through the overall measurement of a to-be-tested device and a test fixture; the scattering parameters to be processed are input into a pre-trained parameter optimization model, a physical constraint framework is introduced when the parameter optimization model is trained to ensure that a prediction result conforms to physical laws such as causality, stability and passivity, and a physical constraint loss function and a frequency adaptive weighting strategy are adopted; obtaining an equivalent model which is output by the parameter optimization model and corresponds to the scattering parameter to be processed; and obtaining a target scattering parameter corresponding to the to-be-processed scattering parameter according to the equivalent model. According to the method, a physical constraint framework is introduced, so that the problem that a traditional 2xthru de-embedding method is in a high-frequency region (especially S11gt; s21), and physical rationality and stability of the result are ensured at the same time.

Description

Technical Field

[0001] The present invention relates to the technical fields of microwave engineering and signal integrity analysis, and particularly to a method, device, medium, and product for determining scattering parameters based on deep learning. Background Art

[0002] S parameters (Scattering Parameters) are key parameters used to describe signal transmission and reflection characteristics in high-frequency circuits, and are widely used especially in radio frequency (RF), microwave, and signal integrity analysis. For a two-port network, S 11 is the reflection coefficient of port 1, S 21 is the forward transmission coefficient from port 1 to port 2, S 12 is the reverse transmission coefficient from port 2 to port 1, and S 22 is the reflection coefficient of port 2.

[0003] The 2xthru de-embedding method in the IEEE P370 standard, that is, the double-length de-embedding method, is a method recommended in the IEEE P370 standard for high-frequency signal integrity measurement and calibration. Through mathematical analysis and processing of the measured S parameters by the 2xthru de-embedding method, its basic process is as follows: measure the S parameters of the complete system including the fixture; use mathematical methods to process the measured S parameters to create an equivalent model of the test fixture; eliminate the fixture influence through de-embedding technology to extract the intrinsic characteristics of the device; use the extracted characteristics for subsequent analysis and design.

[0004] Eliminating the fixture influence through the 2xthru de-embedding method is widely used in high-speed interconnect systems, PCB design, and high-frequency measurement fields. With the continuous increase in data transmission rate, accurate characterization in the high-frequency region has become increasingly important, but the existing 2xthru de-embedding method still has the following disadvantages: (1) The inherent instability of the mathematical model: The key formulas used in the 2xthru method, namely e 111 =(S 22r -e 002 ) / S 12r and e 112 =(S 11r -e 001 ) / S 21r have serious numerical instability. Among them, S 11r , S 12r , S 22r , S 21r are the measured S parameters, e 001 and e 002 represent the directivity error, and e 111 and e 112respectively represent the port matching error of the error box. When the transmission coefficient S 12r or S 21r is very small (when the insertion loss of the fixture is large), the division operation will produce an extremely large value. The insertion loss of the fixture increases with the increase of frequency, resulting in an extremely small transmission coefficient, which is particularly prominent in the high-frequency region. (2) High-frequency region accuracy problem: In the high-frequency region, especially when the reflection coefficient S 11 is greater than the transmission coefficient S 21 , the accuracy of the traditional method drops significantly. Due to the division operation in the error term calculation, when the reflection coefficient is less than the transmission coefficient (S 11 <S 21 ), the error will be further amplified. This leads to serious distortion of the extracted device characteristics in the high-frequency region. (3) Lack of physical constraints: The 2xthru de-embedding method lacks the constraints that a physical system must follow, such as causality, stability, and passivity. Under certain conditions (especially when S 11 <S 21 ), the calculated error terms may cause the system to violate the passivity principle, which will produce physically unrealizable results and affect subsequent system design and analysis. Summary of the Invention

[0005] The first object of the present invention is to provide a method for determining scattering parameters based on deep learning, which solves the problems of instability, distortion, and even non-compliance with physical laws of the existing 2xthru de-embedding method at high frequencies.

[0006] The second object of the present invention is to provide a computer device for implementing the above method for determining scattering parameters based on deep learning.

[0007] The third object of the present invention is to provide a computer-readable storage medium for implementing the above method for determining scattering parameters based on deep learning.

[0008] The fourth object of the present invention is to provide a computer program product for implementing the above method for determining scattering parameters based on deep learning.

[0009] To achieve the above first object, the present invention provides a method, which includes the following steps: obtaining the scattering parameters to be processed, where the scattering parameters to be processed are obtained by measuring the overall test device and the test fixture connecting the test device; inputting the scattering parameters to be processed into a pre-trained parameter optimization model; where the parameter optimization model is trained by deep learning in a physical constraint framework based on the measured scattering parameters and the de-embedded scattering parameters obtained by de-embedding and screening the measured scattering parameters; obtaining an equivalent model corresponding to the scattering parameters to be processed output by the parameter optimization model; and obtaining the first target scattering parameters corresponding to the scattering parameters to be processed according to the equivalent model.

[0010] As can be seen from the above solution, in the present invention, the de-embedding process of the scattering parameters to be processed is carried out through a pre-trained parameter optimization model. Since the parameter optimization model is trained by setting a physical constraint framework and combining deep learning methods, it can ensure that the extracted device characteristics conform to the basic laws of the physical system. In addition, the parameter optimization model is trained based on deep learning, and has a good de-embedding effect at high frequencies, is more accurate and reliable than the existing de-embedding methods, has a fast processing speed, and has strong generalization ability.

[0011] A further solution is that the physical constraint framework is a vector fitting model, and the vector fitting model is expressed as: H(s) = D + Σ(Ri / (s - pi)) + C, where pi is the pole, Ri is the corresponding residue, C is the proportionality coefficient, D is the constant term, and s is the complex variable in the complex frequency domain.

[0012] Thus, it can be seen that the vector fitting model can be used as the physical constraint framework, so that the prediction results always satisfy the physical laws.

[0013] A further solution is that the parameters in the vector fitting model are represented as one-dimensional vectors, and the one-dimensional vectors include the following types of physical parameters: real poles, real parts of complex poles, imaginary parts of complex poles, real pole residues, real parts of complex pole residues, imaginary parts of complex pole residues, C matrix parameters, D matrix parameters.

[0014] Thus, it can be seen that representing the parameters in the vector fitting model as one-dimensional vectors facilitates subsequent prediction.

[0015] A further solution is that in the process of training the parameter optimization model, it includes setting a loss function according to the vector fitting model, where the loss function is expressed as: the total loss is equal to the sum of the parameter difference term, the real pole penalty term in the pole stability penalty, the real part penalty term of the complex pole in the pole stability penalty, and the pole imaginary part constraint term.

[0016] Thus, it can be seen that in the design of the loss function, physical constraints and frequency response characteristics are combined to ensure that the prediction results always satisfy the physical laws.

[0017] A further solution is that the calculation process of the parameter difference term includes: determining the corresponding initial physical parameters in the vector fitting model according to the measured scattering parameters, and determining the weighted error based on the difference between the frequency responses corresponding to the initial physical parameters and the predicted physical parameters. Among them, for the region greater than the preset frequency and S 11 >S 21 a higher weight is assigned to this region compared to other regions.

[0018] Thus, it can be seen that for the preset frequency and S 11 >S 21The region is given a higher weight relative to other regions, which can specifically solve the precision problem of the 2xthru method in the high-frequency region, especially for S 11 >S 21 when.

[0019] A further solution is that the physical constraint framework is an equivalent circuit model in the form of Foster or Cauer.

[0020] Thus, it can be seen that other circuit models can also be selected as the physical constraint framework.

[0021] A further solution is that after the scattering parameter to be processed is input into the pre-trained parameter optimization model, it is judged whether the scattering parameter to be processed is a scattering parameter in a frequency region greater than a preset frequency. If so, the scattering parameter to be processed is input into the parameter optimization model, and the first target scattering parameter corresponding to the scattering parameter to be processed is obtained according to the equivalent model; if not, the second target scattering parameter corresponding to the scattering parameter to be processed is obtained through the 2xthru de-embedding method.

[0022] Thus, it can be seen that the de-embedding method can be determined according to whether the scattering parameter to be processed is in the high-frequency region. For the scattering parameter in a frequency region less than the preset frequency, de-embedding is performed through the 2xthru de-embedding method. For the scattering parameter in a frequency region greater than or equal to the preset frequency, de-embedding is performed through the parameter optimization model. The present invention expands the frequency range of de-embedding of the measured scattering parameter. For high-frequency data (especially S 11 >S 21 in the case), a deep learning optimization method based on vector fitting constraint is adopted; for low-frequency data, the traditional 2xthru de-embedding method is adopted, ensuring the applicability of the method in the full frequency band.

[0023] To achieve the above second object, a computer device provided by the present invention includes a processor and a memory, wherein: a computer program is stored on the memory, and when the computer program is executed by the processor, the method for determining scattering parameters based on deep learning as described above is implemented.

[0024] To achieve the above third object, a computer-readable storage medium provided by the present invention has a computer program stored thereon, wherein: when the computer program is executed by the processor, the method for determining scattering parameters based on deep learning as described above is implemented.

[0025] To achieve the above fourth object, a computer program product provided by the present invention includes computer instructions, wherein when the computer instructions are executed by the processor, the method for determining scattering parameters based on deep learning as described above is implemented. Description of the Drawings

[0026] Figure 1It is a flowchart for determining the scattering parameters of a device in an embodiment of the method for determining scattering parameters based on deep learning according to the present invention, showing the complete processing flow of frequency judgment, method selection, and physical constraint framework.

[0027] Figure 2 It is a flowchart for training a parameter optimization model in an embodiment of the method for determining scattering parameters based on deep learning according to the present invention, showing key steps such as data loading, physical constraint framework setting, loss function design, and neural network architecture construction.

[0028] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. Detailed implementation manners

[0029] The method for determining scattering parameters based on deep learning according to the present invention proposes a deep learning optimization method constrained by a physical constraint framework. By setting up the physical constraint framework and combining deep learning techniques, the scattering parameters of a device are determined, and it is ensured that the extracted device characteristics conform to the basic laws of the physical system. The present invention also provides a computer device, a computer-readable storage medium, and a computer program product for implementing the above method for determining scattering parameters based on deep learning.

[0030] Embodiment of the method for determining scattering parameters based on deep learning: In this embodiment, a Vector Fitting model is used as the physical constraint framework, and a parameter optimization model is trained by combining deep learning methods, so as to determine the scattering parameters of the device under test according to the trained parameter optimization model.

[0031] This embodiment is implemented by executing a computer program. The following will be combined with Figure 1 Describe the specific steps.

[0032] First, step S11 is executed to obtain the scattering parameters to be processed.

[0033] The scattering parameters to be processed refer to the scattering parameters obtained by measuring the device under test and the test fixture connecting the device under test as a whole. Since the scattering parameters to be processed are affected by the test fixture, the method of this embodiment needs to achieve de-embedding to extract the intrinsic characteristics of the device under test.

[0034] Continue to execute step S12 to determine whether the scattering parameters to be processed are scattering parameters in a frequency region greater than a preset frequency.

[0035] The scattering parameter in the frequency region greater than the preset frequency means that the frequency of each frequency point in the scattering parameter is greater than the preset frequency. In this embodiment, the preset frequency is taken as 50 GHz. When it is determined that the scattering parameter to be processed is the scattering parameter in the frequency region not greater than 50 GHz, jump to step S16, perform de-embedding by the 2xthru de-embedding method, and then continue to execute step S17 to obtain the second target scattering parameter corresponding to the scattering parameter to be processed.

[0036] When the judgment result of step S12 is "yes", continue to execute step S13, and input the scattering parameter to be processed into the pre-trained parameter optimization model. The parameter optimization model is pre-trained by deep learning technology, and the specific training process will be introduced in detail below and will not be elaborated here.

[0037] Then continue to execute step S14: Obtain the equivalent model corresponding to the scattering parameter to be processed output by the parameter optimization model.

[0038] The specific form of the equivalent model corresponds to the physical constraint framework. In this embodiment, the physical constraint framework is a vector fitting model, so the equivalent model refers to the target vector fitting model corresponding to the scattering parameter to be processed.

[0039] Finally, execute step S15: Obtain the first target scattering parameter corresponding to the scattering parameter to be processed according to the equivalent model.

[0040] The first target scattering parameter corresponding to the scattering parameter to be processed can be obtained according to the response of the equivalent model at different frequencies. Specifically, first determine the scattering parameter corresponding to the required target frequency, and then substitute the target frequency in complex form into the target vector fitting model corresponding to the scattering parameter to be processed obtained in step S14, and the scattering parameter matrix corresponding to the target frequency can be obtained, that is, the first target scattering parameter after de-embedding processing of the scattering parameter to be processed at the target frequency. For example, let the target frequency be f, then the angular frequency ω = 2πf corresponding to the target frequency can be obtained, and then the angular frequency is converted into complex form, that is, the complex variable in the complex frequency domain corresponding to the target frequency is obtained, that is, s = jω = j⋅2πf, and substituting the complex variable in the complex frequency domain corresponding to the target frequency into the target vector fitting model, the scattering parameter matrix corresponding to the target frequency can be obtained.

[0041] It should be noted that the above expressions of "first" and "second" are used to distinguish similar practices and do not have the meaning of sequence.

[0042] The parameter optimization model is trained by deep learning, and the following will be combined with Figure 2 Introduce the training process of the parameter optimization model step by step.

[0043] First, execute step S21 to load the data required for training.

[0044] In step S21, it includes step S211, step S122, and step S213.

[0045] First, step S211 is executed to import the measured scattering parameters. The measured scattering parameters are imported through s2p or s4p files. The measured scattering parameters refer to the scattering parameters obtained by measuring the device and the test fixture connecting the device together, and the measured scattering parameters are used for model training.

[0046] Then, step S212 is executed to de-embed by the 2xthru method. For the above imported measured scattering parameters, the 2xthru de-embedding method is used for de-embedding.

[0047] Then, step S213 is executed to screen the de-embedded results. The screening specifically refers to selecting from the results of the measured scattering parameters after de-embedding, and taking the available frequency bands (for example, taking the frequency bands where the de-embedded results satisfy passivity, or the frequency bands where the measured scattering parameter IL-RL > 5 dB), so that the obtained de-embedded scattering parameters are credible.

[0048] Thus, the data required for training is obtained. The data required for training includes the measured scattering parameters and the de-embedded scattering parameters obtained by de-embedding and screening the measured scattering parameters.

[0049] Then, step S22 is executed to set up the physical constraint framework.

[0050] In step S22, it includes step S221 and step S222.

[0051] First, step S211 is executed to construct a vector fitting model.

[0052] In this embodiment, the vector fitting model is used as the physical constraint framework.

[0053] In other embodiments, the physical constraint framework can also adopt an equivalent circuit model in the form of Foster or Cauer.

[0054] In other embodiments, more professional field knowledge constraints can also be introduced under the physical constraint framework, such as electromagnetic field distribution constraints.

[0055] The vector fitting model provides a rational function representation that conforms to the characteristics of the physical system: H(s) = D + Σ(Ri / (s - pi)) + C, where pi is the pole, Ri is the corresponding residue, C is the proportionality coefficient, D is the constant term, and s is the complex variable in the complex frequency domain.

[0056] By setting up a physical constraint framework, it can be ensured that the prediction results of all neural network models meet the causality and stability requirements of the physical system. As a physical constraint framework, the vector fitting model can meet the causality and passivity requirements of the physical system by using the form of residues and poles. Causality refers to the property that the output of a physical system cannot occur before the input. Passivity refers to the property that a physical system cannot generate energy. Correspondingly, in the vector fitting model, the real part of the system poles must be negative.

[0057] Then step S212 is executed, and the physical parameters of the vector fitting model are flattened and represented as a one-dimensional vector.

[0058] The physical parameters in the vector fitting model are flattened and represented as a one-dimensional vector, which includes the following types of physical parameters: real poles, real parts of complex poles, imaginary parts of complex poles, real pole residues, real parts of complex pole residues, imaginary parts of complex pole residues, C matrix parameters, and D matrix parameters. Each type of physical parameter in this one-dimensional vector has a clear physical meaning and is subject to corresponding physical constraints. Among them, the real poles, real parts of complex poles, and imaginary parts of complex poles correspond to the poles in the rational function representation of the vector fitting model; the real pole residues, real parts of complex pole residues, and imaginary parts of complex pole residues correspond to the corresponding residues in the rational function representation of the vector fitting model, the C matrix parameters correspond to the proportionality coefficients in the rational function representation of the vector fitting model, and the D matrix parameters correspond to the constant terms in the rational function representation of the vector fitting model.

[0059] Then step S23 is executed to set up the loss function. The setting of the loss function combines physical constraints and frequency response characteristics to ensure that physical laws are always satisfied during the optimization process.

[0060] In step S22, it includes step S231, step S232, and step S233.

[0061] First, step S231 is executed to set up the parameter difference term.

[0062] The parameter difference term measures the difference between the predicted physical parameters and the initial physical parameters using the mean squared error (MSE), encouraging the optimization process to maintain parameter stability. Specifically, the initial physical parameters corresponding to the selected de-embedded scattering parameters are obtained through the Vector Fitting method, and the vector fitting model corresponding to the initial physical parameters is called the initial vector fitting model; the predicted physical parameters corresponding to the measured scattering parameters are predicted through the parameter optimization model, and the vector fitting model corresponding to the predicted physical parameters is called the predicted vector fitting model. Frequency response calculations are performed on the initial vector fitting model and the predicted vector fitting model respectively to obtain the corresponding scattering parameters at different frequency points. The error is determined based on the difference between the scattering parameters obtained from the initial vector fitting model and the scattering parameters obtained from the predicted vector fitting model at the same frequency point. Specifically, it can be the sum of the differences of the same terms in the scattering parameters multiplied by the corresponding weight coefficients to obtain the error corresponding to this frequency point. By analogy, the errors corresponding to all frequency points are obtained, and the average of all errors is calculated to obtain the weighted error, which measures the difference in the frequency response between the initial physical parameters and the predicted physical parameters. It should be noted that for the frequency region greater than the preset frequency and S 11 >S 21 a higher weight is assigned to the frequency region relative to other frequency regions. For example, in this embodiment, the preset frequency is 50 GHz. For frequencies greater than 50 GHz and S 11 >S 21 the weight coefficient of the frequency points is 3 (3 is an empirical value obtained through experiments, and setting it to this value results in a better prediction effect for the parameter optimization model compared to other values), while the coefficient of the frequency points in other frequency regions is 1.

[0063] Then, step S232 is executed to set the physical constraint penalty term. The physical constraint penalty term includes a pole stability penalty term and a pole imaginary part constraint term.

[0064] The pole stability penalty term includes a real pole penalty term and a real part penalty term of the complex pole.

[0065] The real pole penalty term is used to ensure that the predicted real pole is less than zero. By setting the penalty weight, the influence of the loss term is balanced. When the predicted real pole is greater than 0, the penalty term is triggered, and the ReLU function layer outputs a positive value, which is then multiplied by the penalty coefficient (the penalty coefficient is taken as 10) to obtain the score of the real pole penalty term.

[0066] The real part penalty term of the complex pole is used to ensure that both the real part of the predicted complex pole and the real part of the complex pole residue are less than zero. By setting the penalty weight, the influence of the loss term is balanced. When the real part of the predicted complex pole is greater than 0, the penalty term is triggered, and the ReLU function layer outputs a positive value, which is then multiplied by the penalty coefficient (the penalty coefficient is taken as 10) to obtain the score of the pole stability penalty term.

[0067] The imaginary part constraint term of the pole is used to ensure that the imaginary part of the predicted complex pole and the imaginary part of the residue of the complex pole must be positive. When the imaginary part of the pole is less than 0, the penalty term is triggered, and the ReLU function layer outputs a positive value, which is then multiplied by the penalty coefficient (the penalty coefficient is taken as 10) to obtain the score of the imaginary part constraint term of the pole.

[0068] Then step S233 is executed to calculate the total loss.

[0069] The total loss of this embodiment is denoted as total_loss.

[0070] The loss function of this embodiment is expressed as: total_loss = param_diff + poles_real_penalty + poles_cmplx_real_penalty + cmplx_imag_penalty. Among them, param_diff represents the parameter difference term, poles_real_penalty represents the real pole penalty term in the pole stability penalty term, poles_cmplx_real_penalty represents the complex pole real part penalty term in the pole stability penalty, and cmplx_imag_penalty represents the imaginary part constraint term of the pole.

[0071] Then step S24 is executed to set up a neural network architecture integrated with the physical constraint framework.

[0072] In step S24, it includes step S241 and step S242.

[0073] First, step S241 is executed to set up the input layer, hidden layer, and output layer.

[0074] The neural network architecture integrated with the physical constraint framework includes an input layer, a hidden layer, and an output layer.

[0075] Among them, the input layer is used to receive the pre-measured scattering parameters. There are multiple hidden layers, and each hidden layer performs linear transformation, ReLU function activation, and batch normalization in sequence. The output layer is a linear layer used to map the output of the last hidden layer to the predicted physical parameters.

[0076] Then step S242 is executed for forward propagation and reverse ship training.

[0077] After the neural network architecture is defined, the input data is passed layer by layer to the output layer through the forward propagation process, and the output layer outputs the predicted physical parameters, and the model parameters are updated by combining the loss function and the backpropagation algorithm. Finally, step S25 is executed to obtain the parameter optimization model that has completed training.

[0078] After training is completed, a neural network model integrated with the physical constraint framework is obtained, that is, the above-mentioned parameter optimization model is obtained.

[0079] Since the parameter optimization model is specifically used to predict the physical parameters corresponding to the vector fitting model, and by integrating physical constraints into the loss function to ensure its compliance with the characteristics of the actual system, it takes into account both the expressive ability and training stability, and is suitable for dealing with complex non-linear relationships of high-frequency signals.

[0080] In other embodiments, the neural network architecture can also use a convolutional neural network (CNN) to process the frequency domain response, combined with a vector fitting physical constraint output layer.

[0081] In other embodiments, the neural network architecture can also adopt a graph neural network based on physical prior knowledge to capture the physical correlation between parameters.

[0082] In summary, through the deep learning optimization method based on the physical framework constraint, the present invention effectively extends the 2xthru de-embedding method to higher frequencies, and has the following significant advantages: 1. Greatly improved high-frequency accuracy: By introducing the VF model as the physical constraint framework and the frequency adaptive weighting strategy, the present invention improves the accuracy in the high-frequency region of S 11 >S 21 by more than 50%, and particularly optimizes the key frequency bands where the traditional methods fail; 2. Guarantee of physical consistency: All optimization results are restricted by the vector fitting physical constraint framework to ensure compliance with physical characteristics such as causality, stability, and passivity, and avoid generating physically unrealizable results; 3. Significantly improved numerical stability: Through the design of the physical constraint loss function, the problem of numerical instability caused by the division operation in the traditional 2xthru method is effectively solved; 4. Efficient exploration of the physical parameter space: The parameter optimization model efficiently explores the parameter space under the physical constraint framework and discovers the optimal physical parameter combinations that are difficult to find by traditional methods; 5. Balancing computational efficiency and physical rationality: Compared with traditional methods, the present invention improves the computational efficiency by 40% while ensuring physical rationality, and the results are more stable; 6. Strong generalization ability: The deep learning framework based on physical constraints has stronger generalization ability, can be applied to different test environments and device characteristics, and realizes the successful extension of the de-embedding method to higher frequencies.

[0083] The training of the above-mentioned parameter optimization model is specifically implemented through the Python language. The following code frameworks for loading the training data, setting the physical constraint framework, setting the loss function, and setting the neural network architecture integrated with the physical constraint framework are given in sequence: Code framework for loading data required for training and setting up the physical constraint framework: def objective_function_vf_object( params_flat , n_poles_real , n_poles_ cmplx_pairs , n_ports , has_c , has_d , freqs , target_s , z0 , vf_template ): # Unpack parameters # ... # Use the VF model to ensure the feasibility of physical implementation tmp_vf = vf(rf.Network( frequency =rf.Frequency.from_f(freqs, unit ='Hz'), s =np.zeros((len(freqs), n_ports, n_ports), dtype =complex), z0 =z0)) tmp_vf.poles = current_poles tmp_vf.residues = residues_reshaped.T # ... def pack_params_structured( poles_real , poles_cmplx_pairs , res_real , res_ cmplx , c_real , d_real ): """Pack the structured VF parameters into a flat real vector.""" params = np.array([], dtype =float) # Pack real poles params = np.concatenate((params, poles_real.real)) # Pack the real and imaginary parts of complex pole pairs if poles_cmplx_pairs is not None and len(poles_cmplx_pairs)>0: cmplx_pole_params = np.ravel(poles_cmplx_pairs).real params = np.concatenate((params, cmplx_pole_params)) # Pack residues (continue packing other parameters...) # ... Code framework for setting the loss function: def custom_loss( pred_params , input_params , target_score ): # Ensure that the predicted parameters are within the physically valid range clamped_params = torch.zeros_like(pred_params) for i, bounds in enumerate(param_bounds): low, high = bounds clamped_params[:, i] = torch.clamp(pred_params[:, i], low, high) # Calculate the parameter difference - encourage staying close to the initial parameters, use weighted MSE param_diff = torch.mean((clamped_params - input_params)**2) # Extract the parameters of the pole part poles_real_indices = list(range(n_poles_real)) poles_cmplx_real_indices = list(range(n_poles_real, n_poles_real + n_poles_cmplx_pairs)) # Pole stability penalty - ensure the real part is negative poles_real_penalty = torch.mean(torch.relu(clamped_params[:, poles_real_indices])) * 10.0 poles_cmplx_real_penalty = torch.mean(torch.relu(clamped_params[:,poles_cmplx_real_indices])) * 10.0 # Physical constraint that the imaginary part must be positive cmplx_imag_indices = list(range(n_poles_real + n_poles_cmplx_pairs, n_poles_real + 2 * n_poles_cmplx_pairs)) cmplx_imag_penalty = torch.mean(torch.relu(-clamped_params[:, cmplx_imag_indices])) * 10.0 # Combine losses total_loss = param_diff + poles_real_penalty + poles_cmplx_real_penalty + cmplx_imag_penalty return total_loss, loss_components_dict class VFParamNet(nn.Module): Code framework for setting up the neural network architecture integrated with the physical constraint framework """Deep neural network model for predicting VF parameters""" def __init__( self , input_size , output_size , hidden_layers =[512, 256,128]): super(VFParamNet, self).__init__() # Build network layers layers = [] prev_size = input_size for hidden_size in hidden_layers: layers.append(nn.Linear(prev_size, hidden_size)) layers.append(nn.ReLU()) layers.append(nn.BatchNorm1d(hidden_size)) prev_size = hidden_size # Output layer - generate parameters that meet physical constraints layers.append(nn.Linear(prev_size, output_size)) self.model = nn.Sequential(*layers) def forward( self , x ): # Forward propagation - the output will apply physical constraints in the loss function return self.model(x) Example of computer device: The computer device of this embodiment includes a processor and a memory. The memory stores a computer program. When the processor executes the computer program, it implements the above-mentioned embodiment of the permission configuration management method applied to BI analysis software.

[0084] The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the computer device may include more or fewer components, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0085] For example, the processor can be a Central Processing Unit (CPU), or it can also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microcontroller or any conventional processor, etc. The processor is the control center of the computer device, connecting all parts of the entire computer device through various interfaces and circuits.

[0086] The memory can be used to store computer programs and / or modules. The controller realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. For example, the memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text data, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, memory, plug-in hard disks, Smart Media Cards (SMCs), Secure Digital (SD) cards, Flash Cards, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices.

[0087] Examples of computer-readable storage media: If the modules integrated in the computer device of the above embodiments 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 this understanding, all or part of the process of implementing the method embodiments for determining scattering parameters based on deep learning can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a controller, the steps of the above method embodiments for determining scattering parameters based on deep learning can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The storage medium can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0088] Embodiment of computer program product: The computer program product of this embodiment includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes each step of the above method embodiments for determining scattering parameters based on deep learning.

[0089] Finally, it should be emphasized that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining scattering parameters based on deep learning, characterized in that It includes the following steps: Obtain the scattering parameters to be processed, which are obtained by measuring the overall device under test and the test fixture connecting the device under test; Input the scattering parameters to be processed into a pre-trained parameter optimization model; wherein, the parameter optimization model is trained by a deep learning method under a physical constraint framework based on the measured scattering parameters and the de-embedded scattering parameters obtained by de-embedding and screening the measured scattering parameters; Obtain the equivalent model corresponding to the scattering parameters to be processed output by the parameter optimization model; Obtain the first target scattering parameter corresponding to the scattering parameters to be processed according to the equivalent model.

2. The method for determining scattering parameters based on deep learning according to claim 1, wherein: The physical constraint framework is a vector fitting model, and the vector fitting model is expressed as: H(s) = D + Σ(Ri / (s - pi)) + C, where pi is the pole, Ri is the corresponding residue, C is the proportionality coefficient, D is the constant term, and s is the complex variable in the complex frequency domain.

3. The method for determining scattering parameters based on deep learning according to claim 2, wherein: The parameters in the vector fitting model are expressed as one-dimensional vectors, and the one-dimensional vectors include the following types of physical parameters: real poles, real parts of complex poles, imaginary parts of complex poles, real pole residues, real parts of complex pole residues, imaginary parts of complex pole residues, C matrix parameters, D matrix parameters.

4. The method for determining scattering parameters based on deep learning according to claim 3, wherein: In the process of training the parameter optimization model, it includes setting a loss function according to the vector fitting model, wherein the loss function is expressed as: the total loss is equal to the sum of the parameter difference term, the real pole penalty term in the pole stability penalty term, the real part penalty term of the complex pole in the pole stability penalty, and the pole imaginary part constraint term.

5. The method for determining scattering parameters based on deep learning according to claim 4, wherein: The calculation process of the parameter difference term includes: determining the corresponding initial physical parameters in the vector fitting model according to the measured scattering parameters, and determining the weighted error based on the difference between the frequency responses corresponding to the initial physical parameters and the predicted physical parameters, wherein for a region with a frequency greater than a preset frequency and S 11 >S 21 a higher weight is assigned to the region compared to other regions.

6. The method for determining scattering parameters based on deep learning according to claim 1, wherein: The physical constraint framework is an equivalent circuit model in Foster or Cauer form.

7. The method for determining scattering parameters based on deep learning according to any one of claims 1 to 6, wherein: After obtaining the scattering parameters to be processed, judge whether the scattering parameters to be processed are scattering parameters in a frequency region greater than a preset frequency. If so, input the scattering parameters to be processed into the parameter optimization model to obtain the equivalent model, and obtain the first target scattering parameter corresponding to the scattering parameters to be processed according to the equivalent model; If not, obtain the second target scattering parameter corresponding to the scattering parameters to be processed through the 2xthru de-embedding method.

8. A computer device, comprising a processor and a memory, wherein: A computer program is stored on the memory, and when the computer program is executed by the processor, the method for determining scattering parameters based on deep learning according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the method for determining scattering parameters based on deep learning according to any one of claims 1 to 7 above.

10. A computer program product, comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, they implement the method for determining scattering parameters based on deep learning according to any one of claims 1 to 7 above.

Images