Robust analog filter auxiliary design method and system based on prediction uncertainty

Abstract

The present invention discloses a robust analog filter auxiliary design method and system based on prediction uncertainty, including the collection of training data and the training of uncertainty prediction model, and the use of uncertainty model for auxiliary design; firstly, a simulation method is used to obtain data with added noise and the uncertainty prediction model is trained using these data, and then the robustness of the filter is evaluated during the filter design process using the uncertainty prediction model. The method of the present invention can estimate the robustness of specific filter parameters after actual implementation without conducting physical experiments, can effectively reduce experimental costs, and also allows the use of lower precision devices, thereby reducing the production cost of analog filters.

Description

Technical Field

[0001] The invention discloses a robust analog filter auxiliary design method and system based on prediction uncertainty, belonging to the technical field of communication and signal processing. Background Art

[0002] Analog filters are widely used in the fields of communications and signal processing. In order to meet the filtering performance required in specific scenarios, the parameters of the filter usually need to be redesigned. With the increasing demand for filter accuracy, the physical implementation of analog filters often places high demands on the performance of components, which greatly increases the production cost; on the other hand, the impact of the performance of each component on the performance of the analog filter is very complex and difficult to calculate through the accuracy of a single component, so it can only be evaluated by experimental methods. If high-precision components are simply used, the production cost will be high, but ensuring the robustness of the filter when using lower-precision components requires a lot of experimental evaluation and expert experience, which significantly increases the design cost.

[0003] There are some studies based on machine learning methods for auxiliary design of filters, mainly from two directions, but none of them considers the robustness of filters. One direction is to use machine learning methods to learn a prediction model for the filter curve, so that the performance of a specific filter can be predicted without experiments. However, if robustness is to be considered, the performance of the filter has a certain random volatility, so its performance cannot be accurately predicted using the model. Another direction is to use a machine learning model to directly predict the filter parameters according to the required response curve. This method also requires that one response curve corresponds to one filter parameter, but one-to-one correspondence does not exist when considering random fluctuations. Summary of the invention

[0004] Purpose of the invention: In view of the problems and shortcomings in the prior art, the present invention provides a robust analog filter auxiliary design method and system based on prediction uncertainty.

[0005] Technical solution: A robust analog filter auxiliary design method based on prediction uncertainty, including data generation and uncertainty model training steps, and uncertainty model auxiliary design analog filter step; in the data generation and uncertainty model training, batch filter parameters are randomly generated, and set random noise is added to the parameters to obtain noisy parameters X, and then a simulator is used to predict the response curve Y corresponding to these parameters X, and the parameters X and the response curve Y are used as training data to train the uncertainty prediction model M; in the uncertainty model auxiliary design analog filter step, in the filter design process of a given target response curve, the trained uncertainty prediction model M is used to evaluate the accuracy and robustness of the parameters, so that analog filter parameters with both accuracy and robustness can be designed without conducting experiments.

[0006] The data generation and uncertainty model training steps are specifically as follows:

[0007] Step 100, initializing the random generator and setting the structure of the analog filter and other information;

[0008] Step 101, using a random generator to generate a preset number of analog filter parameters;

[0009] Step 102, adding corresponding random noise to the parameters generated in step 101, and the filter parameters with added noise are recorded as X;

[0010] Step 103, using a simulator to calculate a response curve Y corresponding to the parameter X;

[0011] Step 104: Use the data X and Y obtained above to train the uncertainty prediction model M.

[0012] The uncertainty model aided design of analog filter step comprises:

[0013] Step 200, an initialization step, loading the uncertainty prediction model M obtained in step 104;

[0014] Step 201, input the design target response curve y;

[0015] Step 202, designing a set of filter parameters to be evaluated;

[0016] Step 203, using the uncertainty prediction model M to predict the accuracy and robustness of the set of parameters;

[0017] Step 204, determine whether the accuracy and robustness meet the requirements, if yes, proceed to step 205, otherwise repeat steps 202 to 204;

[0018] Step 205: Output filter parameters that meet the accuracy and robustness requirements, and end the design process.

[0019] In the data generation and uncertainty model training steps: first, the expert sets the structure and other information of the analog filter, and uses a random generator to generate a sufficient number of analog filter parameters, such as resistance, capacitance, power amplifier ratio and other parameters. Then add a certain amount of random noise to the parameters. The intensity of the noise is determined by the performance of the components to be used in the final physical production. For example, if the error of the resistor is 0.001, then Gaussian noise with a standard deviation of 0.001 is added, and the filter parameter with the added noise is recorded as X. Use the simulator to calculate the response curve Y corresponding to the parameter X to obtain the training data set X, Y. Then use the aforementioned training data set to train the uncertainty estimation model M. Here, a feedforward neural network with a Dropout layer is used as the uncertainty estimation model M. Its input is the filter parameter X, and its output is the predicted response curve Y. The training process uses the stochastic gradient descent method to optimize the mean square loss function (MSE) of the target Y.

[0020] The steps of the uncertainty model-assisted analog filter design are: first load the trained uncertainty prediction model M, then input the target response curve y to be designed, and the designer designs a set of filter parameters to be evaluated. Use the uncertainty prediction model M to evaluate the accuracy and robustness of this set of parameters. Specifically, since Dropout can bring certain uncertainties in model prediction, we can turn on the Dropout function during prediction. At this time, the prediction results of multiple groups of the same parameter are not the same. When the prediction results are close to the target response curve (measured by MSE), the group of parameters is considered to be accurate and robust; if it deviates from the target y, it is considered that the accuracy is poor, and if the prediction fluctuation is large, it is considered that the robustness is poor. The evaluation criteria are related to the specific design goals. Determine whether the accuracy and robustness meet the requirements. If not, repeat the above design and evaluation steps. If it meets the requirements, output the corresponding analog filter parameters and end the design process.

[0021] A robust analog filter auxiliary design system based on prediction uncertainty, including a data generation and uncertainty model training module, and an uncertainty model auxiliary design analog filter module;

[0022] The data generation and uncertainty model training module randomly generates batch filter parameters, adds set random noise to the parameters to obtain noisy parameters X, then uses a simulator to predict response curves Y corresponding to these parameters X, and uses the parameters X and response curves Y as training data to train the uncertainty prediction model M;

[0023] The uncertainty model assists in the design of the analog filter module. In the filter design process of a given target response curve, the trained uncertainty prediction model M is used to evaluate the accuracy and robustness of the parameters, thereby designing the analog filter parameters with both accuracy and robustness without conducting experiments.

[0024] The implementation process of the robust analog filter auxiliary design system based on prediction uncertainty is the same as the implementation process of the method.

[0025] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the robust analog filter auxiliary design method based on prediction uncertainty as described above is implemented.

[0026] A computer-readable storage medium stores a computer program for executing the above-mentioned method for aiding robust analog filter design based on prediction uncertainty.

[0027] Beneficial effects: Compared with the prior art, the robust analog filter auxiliary design method based on prediction uncertainty provided by the present invention uses a computer to run a simulator to simulate the performance fluctuations of physical components, produces noisy training data, and then uses a neural network model with a Dropout layer to train the uncertainty prediction model. The uncertainty prediction model can be used to evaluate the parameters of the analog filter, and can simultaneously evaluate the accuracy and robustness without conducting physical production experiments, which can effectively improve the design efficiency of the analog filter; at the same time, components with lower precision can also be used to ensure that the performance of the final product meets the requirements from a design perspective, thereby reducing production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flowchart of data generation and uncertainty model training according to an embodiment of the present invention;

[0029] Figure 2 This is a flow chart of the uncertainty model-assisted design of an analog filter according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The present invention is further explained below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0031] like Figure 1As shown, in order to obtain the uncertainty prediction model M, data generation and uncertainty model training processes are required. Specifically, a random data generation method is first specified, mainly to set some filter parameters of the random generator, such as the size of the resistor and capacitor (step 100); then a sufficient number of simulated filter parameters, such as 10,000, are generated using the random generator (step 101); the noise distribution is set according to the accuracy of the physical components, for example, if the accuracy of the components is 0.001, a normal distribution with a standard deviation of 0.001 can be used as the distribution of the noise. A random noise sampled from the noise distribution is superimposed on each parameter data, and the parameter after superimposing the noise is recorded as X (step 102); the response curve Y corresponding to each parameter under ideal conditions is calculated using the existing simulator software (step 103); and the uncertainty prediction model M is trained using data X and Y (step 104). Here, a feedforward neural network with a Dropout layer is used as the uncertainty prediction model M, whose input is the filter parameter X and output is the predicted response curve Y. The training process uses the stochastic gradient descent method to optimize the mean square loss function (MSE) of the target Y.

[0032] like Figure 2 As shown, in the design-experimental evaluation process, the experimental evaluation process is replaced by an uncertainty prediction model, that is, the uncertainty model assists the design of analog filters. Specifically, first load the uncertainty prediction model M (step 200); input the design target response curve Y (step 201); the designer designs a set of analog filter parameters X (step 202); use the uncertainty prediction model M to evaluate the accuracy and robustness of the parameter X (step 203); if the accuracy and robustness meet the requirements, proceed to step 205, otherwise repeat steps 202 to 204 (step 204); output the filter parameter Y that meets the accuracy and robustness requirements, and the design process ends (step 205).

[0033] Since Dropout can bring a certain degree of uncertainty in model prediction, we can turn on the Dropout function during prediction. At this time, the prediction results for multiple groups of the same parameter are not the same. When the prediction results are close to the target response curve (measured by MSE), the group of parameters is considered to have accuracy and robustness; if it deviates from the target Y, it is considered that the accuracy is poor, and if the prediction fluctuates greatly, it is considered that the robustness is poor. The evaluation criteria are related to the specific design goals. Determine whether the accuracy and robustness meet the requirements. If not, repeat the above design and evaluation steps. If they meet the requirements, output the corresponding analog filter parameters and end the design process.

[0034] A robust analog filter auxiliary design system based on prediction uncertainty, including a data generation and uncertainty model training module, and an uncertainty model auxiliary design analog filter module;

[0035] The data generation and uncertainty model training module randomly generates batch filter parameters, adds the set random noise to the parameters to obtain the noisy parameters X, and then uses the simulator to predict the response curve Y corresponding to these parameters X. The parameters X and the response curve Y are used as training data to train the uncertainty prediction model M.

[0036] Uncertainty model-assisted design of analog filter module, in the filter design process of a given target response curve, the trained uncertainty prediction model M is used to evaluate the accuracy and robustness of the parameters, so that the analog filter parameters with both accuracy and robustness can be designed without conducting experiments.

[0037] Obviously, those skilled in the art should understand that the above-mentioned steps of the robust analog filter auxiliary design method based on prediction uncertainty or the modules of the robust analog filter auxiliary design system based on prediction uncertainty of the embodiment of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiment of the present invention is not limited to any specific combination of hardware and software.

Claims

1. A robust analog filter aided design method based on prediction uncertainty, It is characterized in that The method comprises the steps of data generation and uncertainty model training and uncertainty model assisted simulation filter design; in the data generation and uncertainty model training, batch filter parameters are randomly generated, and set random noise is added to the parameters to obtain noisy parameters X, and then a simulator is used to predict response curves Y corresponding to these parameters X, and the parameters X and the response curve Y are used as training data to train an uncertainty prediction model M; in the uncertainty model assisted simulation filter design step, in the filter design process of a given target response curve, the trained uncertainty prediction model M is used to evaluate the accuracy and robustness of the parameters, so as to design simulation filter parameters with both accuracy and robustness without conducting experiments; The data generation and uncertainty model training steps are specifically as follows: Step 100, initializing a random generator and setting structural information of an analog filter; Step 101, using a random generator to generate a preset number of analog filter parameters; Step 102, adding corresponding random noise to the parameters generated in step 101, and the filter parameters with added noise are recorded as X; Step 103, using a simulator to calculate a response curve Y corresponding to the parameter X; Step 104, using the obtained data X, Y to train the uncertainty prediction model M; In the data generation and uncertainty model training steps: firstly, the structural information of the analog filter is set, a set number of analog filter parameters are generated by a random generator, and then a certain amount of random noise is added to the parameters, wherein the intensity of the noise is determined by the performance of the components to be used in the final physical production; a response curve Y corresponding to the parameter X is calculated using a simulator to obtain a training data set X, Y; Then, the uncertainty prediction model M is trained using the aforementioned training data set. A feedforward neural network with a Dropout layer is used as the uncertainty prediction model M, whose input is the filter parameter X and output is the predicted response curve Y. The training process uses the stochastic gradient descent method to optimize the average square loss function of the target Y.

2. The robust analog filter aided design method based on prediction uncertainty according to claim 1, It is characterized in that The uncertainty model aided design of analog filter step comprises: Step 200, an initialization step, loading the uncertainty prediction model M obtained in the data generation and uncertainty model training steps; Step 201, input the design target response curve y; Step 202, designing a set of filter parameters to be evaluated; Step 203, using the uncertainty prediction model M to predict the accuracy and robustness of the set of parameters; Step 204, determine whether the accuracy and robustness meet the requirements, if yes, proceed to step 205, otherwise repeat steps 202 to 204; Step 205: Output filter parameters that meet the accuracy and robustness requirements, and end the design process.

3. The robust analog filter aided design method based on prediction uncertainty according to claim 1, It is characterized in that The steps of the uncertainty model-assisted simulation filter design are: firstly, loading the trained uncertainty estimation model M, then inputting the target response curve Y to be designed, and designing a set of filter parameters to be evaluated; The uncertainty prediction model M is used to estimate the accuracy and robustness of this group of parameters. For multiple groups of prediction results of the same parameter, they are not the same. When the prediction results are close to the target response curve, the group of parameters is considered to be accurate and robust; if it deviates from the target y, it is considered that the accuracy is poor, and if the prediction fluctuation is large, it is considered that the robustness is poor. The evaluation criteria for whether the prediction fluctuation is large are related to the specific design goals; judge whether the accuracy and robustness meet the requirements. If not, repeat the above design and evaluation steps. If they meet the requirements, output the corresponding analog filter parameters and end the design process; The MSE distance is used to determine whether the prediction result is close to the target response curve.

4. A system for implementing the robust analog filter auxiliary design method based on prediction uncertainty as claimed in claim 1, It is characterized in that It includes data generation and uncertainty model training module and uncertainty model assisted design simulation filter module; The data generation and uncertainty model training module randomly generates batch filter parameters, adds set random noise to the parameters to obtain noisy parameters X, then uses a simulator to predict response curves Y corresponding to these parameters X, and uses the parameters X and response curves Y as training data to train the uncertainty prediction model M; The uncertainty model assists in the design of the analog filter module. In the filter design process of a given target response curve, the trained uncertainty prediction model M is used to evaluate the accuracy and robustness of the parameters, thereby designing the analog filter parameters with both accuracy and robustness without conducting experiments.

5. A computer device, Features: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the robust analog filter auxiliary design method based on prediction uncertainty as described in any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, Features: The computer-readable storage medium stores a computer program for executing the prediction uncertainty-based robust analog filter auxiliary design method according to any one of claims 1 to 3.

Images