Clock module precision improving method based on GRU model prediction

Abstract

The invention relates to the technical field of clock module precision control, in particular to a clock module precision improving method based on GRU model prediction. Comprising the steps of obtaining frequency drift data and temperature data of a plurality of groups of crystal oscillators of the same model; performing wavelet change filtering on the obtained frequency drift data; constructing a GRU neural network model by using the filtered frequency drift curve and the temperature change curve, and training the GRU neural network model; the prediction model effect is verified through the test set; real-time frequency and temperature data of the crystal oscillator are obtained, and a frequency drift predicted value is obtained through calculation; and compensating the crystal oscillator voltage according to the frequency predicted value. According to the technical scheme, the problems that an existing high-precision clock module is high in power consumption and high in cost can be solved, and meanwhile high-precision time reference in various environments is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of clock module precision control, and in particular to a method for improving the precision of a clock module based on GRU model prediction. Background Art

[0002] Clock modules are an indispensable component of modern electronic products, used to provide accurate time reference. Traditional clock modules use crystal oscillators as signal sources, but the frequency of crystal oscillators changes due to the influence of ambient temperature, and the accuracy is limited. Oven-controlled crystal oscillator (OCXO) is a crystal oscillator that uses constant temperature control technology and can maintain high-precision frequency output at a constant temperature. Due to its extremely high stability and accuracy, OCXO is widely used in fields that require high-precision clock references, such as aerospace, military, and communications. However, the high cost and high power consumption of OCXO limit its application range. Temperature-compensated crystal oscillator (TCXO) is a crystal oscillator that uses temperature compensation technology and can maintain high-precision frequency output at a constant temperature. Unlike ordinary crystal oscillators, temperature-compensated crystal oscillators take the influence of ambient temperature into consideration in their design, and use built-in temperature sensors and temperature compensation circuits to achieve temperature compensation, overcoming the problem of large frequency deviation of ordinary crystal oscillators due to temperature changes, while achieving higher accuracy and stability. Compared with traditional OCXOs, TCXOs have lower costs and lower power consumption, and are suitable for large-scale integration and applications, but their accuracy is slightly worse than OCXO. Therefore, a low-cost, low-power, high-precision clock module solution is needed that can maintain a high-precision time reference in various environments and is suitable for large-scale applications. Summary of the invention

[0003] The purpose of the present invention is to propose a method for improving the accuracy of a clock module based on GRU model prediction. The technical solution can solve the problems of high power consumption and high cost of existing high-precision clock modules, and at the same time achieve high-precision time reference in various environments.

[0004] To achieve the above-mentioned purpose, the embodiment of the present disclosure provides a method for improving the accuracy of a clock module based on GRU model prediction, including: obtaining frequency drift data and temperature data of several groups of crystal oscillators of the same model; performing wavelet transform filtering on the obtained frequency drift data; using the filtered frequency drift curve and temperature change curve to construct a GRU neural network model and train it; verifying the effect of the prediction model through a test set; obtaining the real-time frequency and temperature data of the crystal oscillator and calculating the frequency drift prediction value; and compensating the crystal oscillator voltage according to the frequency prediction value.

[0005] Beneficial effects of the basic scheme: The present invention uses a temperature-compensated crystal as a signal source and adopts a prediction model to compensate for the time drift and temperature drift of the temperature-compensated crystal in real time, thereby improving frequency accuracy while avoiding excessive power consumption. The temperature-compensated crystal oscillator is dynamically controlled by a microprocessor preloaded with prediction model parameters to achieve high-precision output requirements for the temperature-compensated crystal oscillator. Since a large amount of temperature-compensated crystal frequency drift and environmental data are used as training to adjust the prediction model, there is no need to collect long-term data for model training when applying it.

[0006] The original frequency drift data is preprocessed through wavelet transform filtering technology to effectively remove noise and unnecessary fluctuations, making the data used to train the GRU model purer and more regular, improving the model's ability to fit the complex relationship between frequency drift and temperature, and thus improving the accuracy of the prediction.

[0007] Compared with ordinary convolutional neural networks (CNNs), which can only process inputs one by one and cannot reflect the changing relationship between the previous and next inputs, the present invention uses a gated recurrent network (GRU), one of the variants of the recurrent neural network (RNN), to explore the commonality between the frequency drifts of the same type of temperature-compensated crystals, which can reflect both the internal relationship of the inputs and the changes between the previous and next inputs, and solve the problem that the gradient of the hidden layer variables in the RNN network may disappear or explode.

[0008] When building and training the GRU neural network model, the correspondence between the filtered frequency drift curve and the temperature change curve is used to enable the model to learn the deep-level influence of temperature on frequency drift. This learning method enhances the generalization ability of the model under different temperature conditions, making the prediction results more reliable.

[0009] By acquiring the frequency and temperature data of the crystal oscillator in real time and calculating the predicted value of the frequency drift based on the trained GRU model, real-time compensation of the crystal oscillator frequency can be achieved, responding quickly to changes in ambient temperature, and effectively reducing the frequency deviation caused by temperature changes, thereby maintaining the high-precision output of the clock module.

[0010] As a core component of many electronic systems, the accuracy and stability of the clock module directly affect the performance of the entire system. Through this technical solution, the clock module can better resist the influence of ambient temperature fluctuations and maintain a stable frequency output, thereby improving the stability and reliability of the entire system. And compared with the traditional hardware adjustment method, this technical solution has higher flexibility and lower cost. In addition, since the prediction model can automatically adapt to environmental changes, it reduces the need for manual intervention and further reduces the maintenance cost of the system. It can be widely used in various fields that require high-precision clocks, such as electronic equipment, computers, communication equipment, etc.

[0011] 2. A method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: obtaining frequency drift data and temperature data of several groups of crystal oscillators of the same model, including the following contents: Prepare temperature control equipment and a frequency meter; install the temperature compensation crystal on the frequency meter and control the temperature change through the temperature control equipment; set different temperatures and time periods to obtain frequency drift data and temperature change data at multiple temperatures; the frequency drift data includes a frequency drift curve, and the temperature change data includes a temperature change curve.

[0012] As an implementable preferred solution, wavelet transform filtering is performed on the obtained frequency drift data, including the following contents: Select a wavelet basis function; use the selected wavelet basis function to perform wavelet decomposition on the frequency drift data to obtain wavelet coefficients and approximate coefficients of different scales to decompose the frequency drift curve into multiple layers, each layer corresponding to a different frequency component; by setting a threshold, set the smaller coefficients in the wavelet coefficients to zero and retain the larger coefficients; perform a reverse transformation on the processed wavelet coefficients and approximate coefficients to obtain the filtered frequency drift curve.

[0013] As an implementable preferred solution, the GRU neural network model is constructed and trained using the filtered frequency drift curve and temperature change curve, including the following contents: The filtered frequency drift curve and temperature change curve are used as feature quantities to remove outliers and noise in the data; the neural network structure is defined; the parameters of the neural network are initialized; and the model is trained using the training set.

[0014] As an implementable preferred solution, the formula of the GRU model is as follows: Defining the input sequence and the hidden state sequence ,in Represents a time step:

[0015]

[0016] in, represents the frequency drift curve after filtering, Represents the temperature change curve. Represents the basic unit of the GRU neural network; The calculation formula of the GRU unit is as follows:

[0017]

[0018]

[0019]

[0020] in, represents the sigmoid function, represents element-wise multiplication, Represents the reset gate, represents the update gate, represents the new candidate hidden state, represents the hidden state sequence, , , They represent the weight matrices, , and Represents the bias vector: Connect the output layer to the GRU layer and use the Softmax function to normalize the result to get the probability distribution of each category. The formula is as follows:

[0021] in, represents the weight matrix, They represent the bias vectors respectively.

[0022] As an implementable preferred solution, the prediction model effect is verified through a test set, including the following: For each crystal, input its filtered frequency drift curve and temperature change curve to obtain its corresponding predicted frequency drift curve; The root mean square error (RMSE) between the predicted frequency drift curve and the actual frequency drift curve is calculated to evaluate the accuracy of the prediction model. The specific formula is:

[0023] Where n represents the number of samples, Indicates the actual frequency drift value, Indicates the predicted frequency drift value; It is determined whether the RMSE value is lower than the preset threshold; if so, the GRU model parameters are input into the MCU processor, otherwise the parameter values ​​of the GRU model are readjusted and training is performed.

[0024] As an implementable preferred solution, obtaining the real-time frequency and temperature data of the crystal oscillator and calculating the frequency drift prediction value includes the following: During the operation of the clock module, the ambient temperature is monitored in real time through the temperature sensor, and the temperature data is input into the GRU model. The frequency data of the temperature-compensated crystal oscillator is regularly obtained through the MCU's timer and reference frequency source. The temperature data and frequency data are input into the GRU model as feature quantities for real-time frequency prediction.

[0025] As an implementable preferred solution, the crystal oscillator voltage is compensated in real time by the frequency prediction value to control the frequency output of the crystal oscillator. According to the prediction result, the voltage of the temperature compensated crystal oscillator is adjusted by a DAC digital-to-analog converter. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of the clock module structure.

[0027] Figure 2 The figure is a logical schematic diagram of a method for improving the accuracy of clock modules based on GRU model prediction.

[0028] Figure 3 Schematic diagram of the GRU neural network unit structure.

[0029] Figure 4 Schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the technical solution and advantages of the present application clearer, the technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings. It will be understood that the specific embodiments described herein are only partial embodiments of the present invention, which are only used to explain the present application, rather than to limit the present application. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated, and they can be combined with each other to achieve better technical effects. The same reference numerals appearing in the drawings of the following embodiments represent the same features or components, which can be applied to different embodiments.

[0031] In addition, unless otherwise defined, the technical terms or scientific terms used in the description of the present invention should have the common meanings understood by those skilled in the art in the art to which the present invention belongs.

[0032] The present invention is further described in detail below in conjunction with the accompanying drawings: Reference Figure 1 ,The clock module includes microprocessor (MCU), temperature sensor, digital-to-analog converter (DAC), temperature-compensated crystal oscillator (TCXO) and other components.

[0033] The microprocessor, as the core control unit of the system, is used to perform the calculation of the prediction model, data acquisition and processing, and real-time control of the digital-to-analog converter output to adjust the crystal frequency. The microprocessor is also equipped with a timer to periodically obtain the frequency data of the temperature-compensated crystal oscillator in combination with the reference frequency source. The microprocessor is pre-installed with the trained GRU (Gated Recurrent Unit) model parameters to quickly predict the temperature-compensated crystal oscillator during operation.

[0034] The temperature sensor is used to monitor the ambient temperature in real time and transmit the temperature data to the MCU. Temperature data is one of the important input features of the prediction model and is crucial to improving the prediction accuracy.

[0035] The digital-to-analog converter adjusts the voltage output to the temperature-compensated crystal oscillator according to the instructions of the microprocessor, thereby achieving fine-tuning of the crystal oscillator frequency. The accuracy and response speed of the digital-to-analog converter directly affect the adjustment effect of the crystal oscillator frequency.

[0036] Temperature Compensated Crystal Oscillator (TCXO), as the signal source of the clock module, realizes temperature compensation through built-in temperature sensor and temperature compensation circuit to overcome the problem of large frequency deviation of ordinary crystal oscillators due to temperature changes.

[0037] Reference Figure 2 , a method for improving the accuracy of a clock module based on GRU model prediction, specifically including the following contents: Step S100, obtain multiple sets of frequency drift and temperature data of the same type of temperature compensated crystals, and divide them into training sets and test sets. In this embodiment, taking 600 sets of data as an example, 500 sets can be divided into training sets and 100 sets into test sets. Specifically including: Step S101, prepare the temperature control equipment and frequency meter in the laboratory. The temperature control equipment can control the temperature to change within the required range and monitor the temperature change process. The frequency meter can measure the signal frequency output by the crystal oscillator and record the change over time. Adjust the accuracy of the temperature control equipment and frequency meter according to the experimental requirements, and calibrate them regularly to ensure the accuracy and stability of the collected data. At the same time, during the experiment, interference and noise such as electromagnetic interference or mechanical vibration should be avoided to affect the experimental results.

[0038] Step S102, installing a temperature compensation crystal on a frequency meter, and controlling the temperature change by a temperature control device.

[0039] Step S103, setting different temperatures and time periods to obtain frequency drift curves and temperature variation curves at multiple temperatures.

[0040] Step S104, storing the collected data on a reliable data storage medium for subsequent data processing and analysis.

[0041] Step S200, performing wavelet transform filtering on the obtained frequency drift curve, specifically includes: Step S201, select a wavelet basis function. A wavelet basis function is a basic function in wavelet transform. Different wavelet basis functions have different characteristics when decomposing signals. A wavelet basis function such as Haar, Daubechies or Symlets is selected according to needs.

[0042] Step S202, performing wavelet decomposition on the frequency drift curve. Perform wavelet decomposition on the frequency drift curve using the selected wavelet basis function to obtain wavelet coefficients and approximate coefficients of different scales to decompose the frequency drift curve into multiple layers, each layer corresponding to a different frequency component.

[0043] Step S203, by setting a threshold, smaller coefficients in the wavelet coefficients are set to zero, larger coefficients are retained, high-frequency noise is filtered out, and main low-frequency information is retained, so as to achieve a filtering effect.

[0044] Step S204: The processed wavelet coefficients and approximation coefficients are inversely transformed to obtain a filtered frequency drift curve.

[0045] Step S300, using the filtered frequency drift curve and temperature change curve to form the GRU neural network input feature quantity , refer to Figure 3 , establish a GRU neural network model and train it, including: Step S301, using the filtered frequency drift curve and temperature change curve as feature quantities, removing outliers and noise in the data, and standardizing it into a form acceptable to the neural network.

[0046] Step S302, determine the number of layers, number of neurons, activation function and other parameters of the neural network. The network structure includes an input layer, multiple GRU layers, and an output layer.

[0047] Step S303, initializing the parameters (weights and biases) of the neural network. In this embodiment, random numbers are used for initialization.

[0048] Step S304, train the model using the training set, calculate the output of the model through forward propagation, and update the parameters of the neural network through a back propagation algorithm to minimize the prediction error of the model.

[0049] The formula of the GRU model is as follows: Defining the input sequence and the hidden state sequence ,in Represents a time step:

[0050]

[0051] in, represents the frequency drift curve after filtering, Represents the temperature change curve. Represents the basic unit of the GRU neural network.

[0052] The calculation formula of the GRU unit is as follows:

[0053]

[0054]

[0055]

[0056] in, represents the sigmoid function, represents element-wise multiplication, Represents the reset gate, represents the update gate, represents the new candidate hidden state, represents the hidden state sequence, , , They represent the weight matrices, , and Represents the bias vector.

[0057] Finally, the output layer is connected to the GRU layer, and the result is normalized using the Softmax function to obtain the probability distribution of each category. The formula is as follows:

[0058] in, represents the weight matrix, They represent the bias vectors respectively.

[0059] Step S400, performing test set verification on the trained GRU model, specifically including: Step S401: for each crystal, input its filtered frequency drift curve and temperature change curve to obtain its corresponding predicted frequency drift curve.

[0060] Step S402, calculate the root mean square error (RMSE) between the predicted frequency drift curve and the actual frequency drift curve to evaluate the accuracy of the prediction model. The specific formula is:

[0061] Where n represents the number of samples, Indicates the actual frequency drift value, Indicates the predicted frequency drift value.

[0062] Step S403, determine whether the RMSE value is lower than a preset threshold. If yes, the GRU model parameters are input into the MCU processor, otherwise the GRU model parameter values ​​are readjusted and trained until a satisfactory result is achieved.

[0063] Step S500, obtaining real-time frequency drift and temperature data during the operation of the clock module, and calculating a predicted value of crystal oscillator frequency drift, specifically includes: Step S501, during the operation of the clock module, the ambient temperature is monitored in real time through a temperature sensor, and the temperature data is input into the GRU model.

[0064] Step S502: periodically obtain frequency data of the temperature compensated crystal oscillator through the timer of the MCU and the reference frequency source.

[0065] Step S503: Input the temperature data and frequency data as feature quantities into the GRU model to perform real-time frequency prediction.

[0066] Step S600: The voltage of the temperature compensated crystal oscillator is compensated in real time by the frequency prediction value so as to control the frequency output of the crystal oscillator. According to the prediction result, the voltage of the temperature compensated crystal oscillator is adjusted by a DAC digital-to-analog converter.

[0067] The disclosed embodiment also provides a system for improving the accuracy of a clock module based on GRU model prediction [1], which uses the above-mentioned method for improving the accuracy of a clock module based on GRU model prediction.

[0068] The embodiment of the present disclosure also provides a storage medium, in which a computer program is stored. When the computer program is executed by a processor, all steps of the method for improving the accuracy of the clock module based on GRU model prediction can be implemented.

[0069] Those of ordinary skill in the art can understand that all or part of the processes in the method for improving the accuracy of the clock module based on GRU model prediction can be completed by instructing related hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of various embodiments of the method for improving the accuracy of the clock module based on GRU model prediction. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0070] The embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method for improving the accuracy of the clock module based on GRU model prediction when executing the program. In the embodiment of the present application, the processor is the control center of the computer system, which can be a processor of a physical machine or a processor of a virtual machine.

[0071] Reference Figure 4 , the electronic device 500 includes: at least one processor 501, at least one communication interface 502, at least one memory 503 and at least one bus 504. Among them, the bus 504 is used to realize the connection and communication between these components, the communication interface 502 is used to communicate signaling or data with other node devices, and the memory 503 stores machine-readable instructions executable by the processor 501. When the electronic device 500 is running, the processor 501 communicates with the memory 503 through the bus 504, and when the machine-readable instructions are called by the processor 501, the steps of the method for improving the accuracy of the clock module based on GRU model prediction are executed.

[0072] The above contents are only embodiments of the present invention. The common sense such as the known specific structures and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field to which the invention belongs before the application date or the priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for the ordinary technicians in the relevant field to implement this application. It should be pointed out that for the technicians in this field, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to explain the content of the claims.

Claims

1. A method for improving the accuracy of a clock module based on GRU model prediction, characterized in that: include: Obtain frequency drift data and temperature data of several groups of crystal oscillators of the same model; perform wavelet transform filtering on the obtained frequency drift data; use the filtered frequency drift curve and temperature change curve to construct and train the GRU neural network model; verify the prediction model effect through the test set; obtain the real-time frequency and temperature data of the crystal oscillator and calculate the frequency drift prediction value; compensate the crystal oscillator voltage according to the frequency prediction value.

2. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: Obtain frequency drift and temperature data of several groups of crystal oscillators of the same model, including the following: Prepare temperature control equipment and a frequency meter; install the temperature compensation crystal on the frequency meter and control the temperature change through the temperature control equipment; set different temperatures and time periods to obtain frequency drift data and temperature change data at multiple temperatures; the frequency drift data includes a frequency drift curve, and the temperature change data includes a temperature change curve.

3. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 2, characterized in that: The obtained frequency drift data is subjected to wavelet transform filtering, including the following: Select a wavelet basis function; use the selected wavelet basis function to perform wavelet decomposition on the frequency drift data to obtain wavelet coefficients and approximate coefficients of different scales to decompose the frequency drift curve into multiple layers, each layer corresponding to a different frequency component; by setting a threshold, set the smaller coefficients in the wavelet coefficients to zero and retain the larger coefficients; perform a reverse transformation on the processed wavelet coefficients and approximate coefficients to obtain the filtered frequency drift curve.

4. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: Using the filtered frequency drift curve and temperature change curve, a GRU neural network model is constructed and trained, including the following: The filtered frequency drift curve and temperature change curve are used as feature quantities to remove outliers and noise in the data; the neural network structure is defined; the parameters of the neural network are initialized; and the model is trained using the training set.

5. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: The formula of the GRU model is as follows: Defining the input sequence and the hidden state sequence ,in Represents a time step: in, represents the frequency drift curve after filtering, Represents the temperature change curve. Represents the basic unit of the GRU neural network; The calculation formula of the GRU unit is as follows: in, represents the sigmoid function, represents element-wise multiplication, Represents the reset gate, represents the update gate, represents the new candidate hidden state, represents the hidden state sequence, , , They represent the weight matrices, , and Represents the bias vector: Connect the output layer to the GRU layer and use the Softmax function to normalize the result to get the probability distribution of each category. The formula is as follows: in, represents the weight matrix, They represent the bias vectors respectively.

6. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: The prediction model effect is verified through the test set, including the following: For each crystal, input its filtered frequency drift curve and temperature change curve to obtain its corresponding predicted frequency drift curve; The root mean square error (RMSE) between the predicted frequency drift curve and the actual frequency drift curve is calculated to evaluate the accuracy of the prediction model. The specific formula is: Where n represents the number of samples, Indicates the actual frequency drift value, Indicates the predicted frequency drift value; It is determined whether the RMSE value is lower than the preset threshold; if so, the GRU model parameters are input into the MCU processor, otherwise the parameter values ​​of the GRU model are readjusted and training is performed.

7. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: Obtain the real-time frequency and temperature data of the crystal oscillator and calculate the predicted value of frequency drift, including the following: During the operation of the clock module, the ambient temperature is monitored in real time through the temperature sensor, and the temperature data is input into the GRU model; The frequency data of the temperature-compensated crystal oscillator is regularly obtained through the MCU's timer and reference frequency source; the temperature data and frequency data are input into the GRU model as feature quantities for real-time frequency prediction.

8. The method for improving the accuracy of a clock module based on GRU model prediction according to claim 1, characterized in that: The crystal oscillator voltage is compensated in real time by the frequency prediction value to control the frequency output of the crystal oscillator. According to the prediction result, the voltage of the temperature compensated crystal oscillator is adjusted through the DAC digital-to-analog converter.

Images