A time synchronization method based on GNSS tamed constant temperature crystal oscillator

Abstract

The application relates to the technical field of navigation, and particularly discloses a time synchronization method based on a GNSS tamed constant-temperature crystal oscillator, which comprises the following steps: in a locked state, a taming controller acquires a first time from a constant-temperature crystal oscillator and a second time from a GNSS receiver, and calculates a time deviation according to the first time and the second time; a time keeping controller determines a first frequency deviation according to the time deviation and a taming algorithm, generates a first voltage control signal of the constant-temperature crystal oscillator according to the first frequency deviation, and applies the first voltage control signal to the constant-temperature crystal oscillator through a digital-to-analog converter; the time keeping controller establishes a frequency deviation model according to continuously received first frequency deviation and temperature data; in an unlocked state, the time keeping controller predicts a second frequency deviation by using the frequency deviation model, generates a second voltage control signal of the constant-temperature crystal oscillator according to the second frequency deviation, and applies the second voltage control signal to the constant-temperature crystal oscillator through the digital-to-analog converter.

Description

Technical Field

[0001] This invention relates to the field of navigation technology, and more specifically to a time synchronization method based on a GNSS-disciplined thermostatic crystal oscillator. Background Technology

[0002] Time synchronization is crucial in modern network systems, ensuring time consistency across all nodes.

[0003] In existing technologies, under isothermal conditions, UTC (NTSC) is primarily used as the reference signal. The time difference between the 10 MHz output signal from the crystal oscillator, after frequency division to generate a 1 PPS signal, and the reference signal is measured. An SR620 is used to measure and smooth the time difference. The data is then input into a computer to calculate state parameters using a Kalman filter. The crystal oscillator is controlled based on the estimated state parameters. This method does not consider the influence of temperature, is only suitable for isothermal environments, and has a limited applicability. Furthermore, it employs a traditional Kalman filter algorithm, which is not intelligent enough.

[0004] In 2000, Hwang et al. proposed a time-holding method for the thermostatic crystal oscillator (DC-PLL) system. This model uses a second-order polynomial to describe the temperature and aging effects of the DC-PLL oscillator. Model parameters are obtained from experimental data using the least squares method, and these parameters are then used for simulation experiments. This method involves direct least-squares inversion, resulting in a large computational burden.

[0005] In existing technologies, the temperature is raised from 0℃ to 60℃ and then lowered back to 0℃ to obtain temperature and crystal oscillator frequency data from the controlled system. Null values ​​in the acquired isothermal crystal oscillator frequency sequence are processed to filter out high-frequency noise and smooth the frequency aging curve. The data is then fed into a two-factor BP neural network model, which fits the crystal oscillator aging data using the BP neural network. Time and temperature are selected as independent variables, and frequency difference as the dependent variable. The original data is divided into training and testing sets. The training set is used to train the network model, and the testing data outputs predicted and expected values ​​to determine whether the fitting results of the testing network meet the requirements. This method, using a BP neural network model, is intelligent but computationally intensive and unsuitable for crystal oscillators with limited computing power. Summary of the Invention

[0006] To address the aforementioned problems, the purpose of this invention is to provide a time synchronization method based on GNSS-disciplined cryogenic crystal oscillators, reducing frequency drift caused by temperature and aging of the cryogenic crystal oscillators, and achieving time synchronization between the cryogenic crystal oscillators and GNSS.

[0007] This invention provides a time synchronization method based on a GNSS-disciplined cryogenic crystal oscillator. The output of the cryogenic crystal oscillator and the output of the GNSS receiver are respectively connected to the input of a discipline controller; the output of the discipline controller and the output of a temperature sensor are respectively connected to the input of a time-holding controller; the outputs of the discipline controller and the time-holding controller are respectively connected to the input of a digital-to-analog converter (DAC) via selector switches; the output of the DAC is connected to the input of the cryogenic crystal oscillator; the time synchronization method includes: The temperature sensor collects the temperature data of the thermostatic crystal oscillator in real time. In the locked state, the selector switch of the digital-to-analog converter is connected to the discipline controller; the discipline controller acquires a first time from the temperature-controlled crystal oscillator and a second time from the GNSS receiver, calculates the time deviation based on the first time and the second time, and sends it to the time-holding controller; The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm, generates a first voltage control signal for the thermostatic crystal oscillator based on the first frequency difference, and applies the first voltage control signal to the thermostatic crystal oscillator through a digital-to-analog converter to compensate for the frequency difference of the thermostatic crystal oscillator, thereby achieving time synchronization between the thermostatic crystal oscillator and the GNSS in the locked state. The time-holding controller establishes a frequency difference model based on the continuously received first frequency difference and the temperature data; In the unlocked state, the selector switch of the digital-to-analog converter is connected to the time-holding controller; the time-holding controller uses the frequency difference model to predict the second frequency difference, generates a second voltage control signal for the thermostatic crystal oscillator based on the second frequency difference, and applies the second voltage control signal to the thermostatic crystal oscillator through the digital-to-analog converter to compensate for the frequency difference of the thermostatic crystal oscillator, thereby achieving time synchronization between the thermostatic crystal oscillator and the GNSS in the unlocked state.

[0008] In one possible implementation, the time-holding controller establishes a frequency difference model based on the continuously received first frequency difference and the temperature data, including: Initialize the velocity and position of the particles; the particles include: temperature coefficient, aging coefficient, and initial frequency difference; The fitness of the particle swarm is calculated based on the recursive mean square error. A particle swarm optimization algorithm with perturbation terms is used to find the globally optimal particle; the globally optimal particle includes: temperature coefficient, aging coefficient, and initial frequency difference; The velocity and position of particles are updated using the Discrete Particle Swarm Optimization (DPSO) algorithm.

[0009] In one possible implementation, the time-holding controller, based on the continuously received first frequency difference and the temperature data, further includes establishing a frequency difference model, including: The recursive mean square error is expressed by the following formula: ; in, For particles No. Mean square error at time, For particles No. Mean square error at time, Let n be the instantaneous squared error at time n. This is a correction term introduced due to parameter updates. For the first Frequency difference of time, For the first Observational data at any time For particles No. coefficient vector at time step For particles No. The speed of time From time 1 to time 2 real-time observation data With frequency difference The cumulative sum of cross-correlation vectors, From time 1 to time 2 The cumulative sum of the cross-correlation vectors between the time-observed data and the frequency difference. From time 1 to time 2 real-time observation data The cumulative sum of the autocorrelation matrix, From time 1 to time 2 real-time observation data The cumulative sum of the autocorrelation matrix, This is a transpose.

[0010] In one possible implementation, the time-holding controller, based on the continuously received first frequency difference and the temperature data, further includes establishing a frequency difference model, including: The fitness of a particle swarm is expressed by the following formula; ; in, For particles No. Adaptability at any time For the first Frequency difference vector at time step, For the first Transpose of the observation vector at time step For particles No. The coefficient vector at time step.

[0011] In one possible implementation, the time-holding controller, based on the continuously received first frequency difference and the temperature data, further includes establishing a frequency difference model, including: The velocity of the particle swarm can be expressed by the following formula: ; in, For the first Sub-iteration particles No. Dimensional speed, For the first Sub-iteration particles No. Dimensional speed, For inertial weights, For the first Moment Particle The Dimensional components, and For learning factors, For disturbance factor, and A random number between [0, 1] A random number between [-1, 1] For its own best historical position, This is the optimal position for the population.

[0012] In one possible implementation, the time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm by: The first frequency difference is expressed by the following formula. : ; In the formula, For aging coefficient, For temperature coefficient, The initial frequency difference of the locked state, for Temperature data at any time For time.

[0013] In one possible implementation, the time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm by: The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm; The first frequency difference is filtered by a moving average filter to obtain the filtered first frequency difference. The control quantity of the digital-to-analog converter is determined based on the frequency difference proportionality coefficient of the isothermal crystal oscillator and the first frequency difference after filtering.

[0014] In one possible implementation, determining the control quantity of the digital-to-analog converter based on the frequency difference scaling factor of the thermostatic crystal oscillator and the filtered first frequency difference includes: Perform moving average filtering according to the following formula: ; in, After filtering Frequency difference data at time, for The original frequency difference data at any given time For sliding windows, j relative to the current time t The offset; The control quantity is expressed by the following formula. : ; in, for Control quantities of the digital-to-analog converter at any time For the frequency difference proportionality coefficient of the temperature-controlled crystal oscillator, After filtering Frequency difference data at any given time.

[0015] In one possible implementation, the time-holding controller predicts the second frequency difference using the frequency difference model, including: The second frequency difference of the unlocked state is determined using the following formula. : ; in, for Aging coefficient over time The start time of the unlocked state for Temperature coefficient at time For the first Temperature data at any time For the first Temperature data at any time This represents the initial frequency difference in the unlocked state.

[0016] In one possible implementation, the time-holding controller further includes using the frequency difference model to predict the second frequency difference, including: The initial frequency difference of the unlocked state is determined using the following formula. : ; in, This is the time when the locked state ends. For the average filter window, The start time of the unlocked state for Aging coefficient over time for Temperature coefficient at time For the first Temperature data at any time For the first Temperature data at any given time.

[0017] The present invention provides a time synchronization method based on a GNSS-disciplined cryogenic crystal oscillator. In the unlocked state, the controller switch is connected to the time-holding controller. A temperature sensor continuously collects the temperature information of the cryogenic crystal oscillator. The time-holding controller uses a model established in the locked state to predict and compensate for the frequency difference of the cryogenic crystal oscillator using a digital-to-analog converter, thereby reducing the frequency drift of the cryogenic crystal oscillator caused by temperature and aging, and achieving time synchronization between the cryogenic crystal oscillator and GNSS. Attached Figure Description

[0018] Figure 1 A schematic diagram of a time synchronization system provided for an embodiment of the present invention; Figure 2 A flowchart illustrating the time synchronization method provided in an embodiment of the present invention. Detailed Implementation

[0019] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. The following detailed description of the embodiments and the accompanying drawings are used to illustrate the principles of the present invention by way of example, but should not be used to limit the scope of the present invention. That is, the present invention is not limited to the described preferred embodiments, and the scope of the present invention is defined by the claims.

[0020] In the description of this invention, it should be noted that, unless otherwise stated, "a plurality of" means two or more; the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance; those skilled in the art can understand the specific meaning of the above terms in this invention as appropriate.

[0021] Figure 1 A schematic diagram of a time synchronization system provided for an embodiment of the present invention, such as... Figure 1 As shown, the output terminals of the thermostatic crystal oscillator and the GNSS receiver are connected to the input terminal of the discipline controller, respectively; the output terminals of the discipline controller and the temperature sensor are connected to the input terminal of the time-hold controller, respectively; the output terminals of the discipline controller and the time-hold controller are connected to the input terminal of the digital-to-analog converter via selector switches, respectively; and the output terminal of the digital-to-analog converter is connected to the input terminal of the thermostatic crystal oscillator.

[0022] Figure 2 A flowchart illustrating the time synchronization method provided for embodiments of the present invention is shown below. Figure 2 As shown, this invention provides a time synchronization method based on a GNSS-disciplined cryogenic crystal oscillator, comprising: Step S1: The temperature sensor collects the temperature data of the thermostatic crystal oscillator in real time; In step S2, under the locked state, the selector switch of the digital-to-analog converter is connected to the discipline controller; the discipline controller obtains the first time from the temperature-controlled crystal oscillator and the second time from the GNSS receiver, calculates the time deviation based on the first time and the second time, and sends it to the time-holding controller; Step S3: The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm, generates a first voltage control signal for the thermostatic crystal oscillator based on the first frequency difference, and applies the first voltage control signal to the thermostatic crystal oscillator through a digital-to-analog converter to compensate for the frequency difference of the thermostatic crystal oscillator and realize time synchronization between the thermostatic crystal oscillator and the GNSS in the locked state. In one possible implementation, the first frequency difference is expressed by the following formula. : ; In the formula, For aging coefficient, For temperature coefficient, The initial frequency difference of the locked state, for Temperature data at any time For time.

[0023] In one possible implementation, the time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm; the first frequency difference is filtered by a moving average to obtain the filtered first frequency difference; and the control quantity of the digital-to-analog converter is determined based on the frequency difference proportionality coefficient of the temperature-controlled crystal oscillator and the filtered first frequency difference.

[0024] Since the frequency difference of a temperature-controlled crystal oscillator can be considered linear over a short period, moving average filtering is a suitable method for denoising and smoothing linear time series. Therefore, this invention applies moving average filtering to the initial frequency difference data to reduce the impact of noise. In practice, other filtering methods such as ARMA can also be used.

[0025] The process of determining the control quantity of the digital-to-analog converter based on the frequency difference proportionality coefficient of the isothermal crystal oscillator and the first frequency difference after filtering includes: Perform moving average filtering according to the following formula: ; in, After filtering Frequency difference data at time, for The original frequency difference data at any given time For sliding windows,j relative to the current time t The offset; The GNSS data update frequency is 1Hz, which means the frequency difference sampling period of the thermostatic crystal oscillator is 1s. As mentioned earlier, the thermostatic crystal oscillator has good short-term stability. Therefore, this invention uses the smoothed and filtered frequency difference data with a lag of q seconds as the frequency correction amount, which is the control amount of the digital-to-analog converter at time t.

[0026] The control quantity is expressed by the following formula. : ; in, for Control quantities of the digital-to-analog converter at any time For the frequency difference proportionality coefficient of the temperature-controlled crystal oscillator, After filtering Frequency difference data at any given time.

[0027] Step S4: The time-hold controller establishes a frequency difference model based on the continuously received first frequency difference and temperature data; In one possible implementation, temperature and aging factor are estimated using the Particle Swarm Optimization (DPSO) algorithm with perturbation terms, based on the clock difference data from the locked-phase phase. The aging factor is then... Temperature coefficient Frequency difference at the initial moment The process of solving for the temperature coefficient by considering the position of particles in three-dimensional space can be viewed as a process of finding the dynamic optimal position. Since particles in the population tend to converge to the same position in the later stages of iteration, they lose activity and eventually move at almost "zero speed," meaning that particles are insensitive to fluctuations in temperature and aging coefficients in the later stages of iteration. Therefore, a perturbation factor is added to the velocity formula of the particle swarm optimization algorithm to improve the problem of low speed in the later stages. Alternatively, other methods such as backpropagation (BP) neural networks can be used.

[0028] Specifically, the velocity and position of the particles are initialized; the particles include: temperature coefficient, aging coefficient, and initial frequency difference; the fitness of the particle swarm is calculated based on the recursive mean square error; the global optimal particle is found using a particle swarm optimization algorithm with perturbation terms; the global optimal particle includes: temperature coefficient, aging coefficient, and initial frequency difference; and the velocity and position of the particles are updated using the Discrete Particle Swarm Optimization (DPSO) algorithm.

[0029] In one possible implementation, the fitness of the particle swarm is represented by the following formula; ; in, For particles No. Adaptability at any time For the first Frequency difference vector at time step, For the first Transpose of the observation vector at time step For particles No. The coefficient vector at time step.

[0030] It can be seen that fitness is directly proportional to n, meaning the computational cost is directly proportional to the model building time, which leads to a large computational cost in the later stages of modeling. To reduce the computational cost, this invention uses recursive mean squared error, but other fitting metrics can also be used in practice. The formula... Substituting into the fitness formula above, we get: ; in, .

[0031] Will Substitute into the formula ,have to: ; Further analysis revealed: ; in, , , .

[0032] Therefore, the recursive form of the mean squared error can be obtained.

[0033] The recursive mean square error is expressed by the following formula: ; in, For particles No. Mean square error at time, For particles No. Mean square error at time, Let n be the instantaneous squared error at time n. This is a correction term introduced due to parameter updates. For the first Frequency difference of time, For the first Observational data at any time For particles No. coefficient vector at time step For particles No. The speed of time From time 1 to time 2 real-time observation data With frequency difference The cumulative sum of cross-correlation vectors, From time 1 to time 2 The cumulative sum of the cross-correlation vectors between the time-observed data and the frequency difference. From time 1 to time 2 real-time observation data The cumulative sum of the autocorrelation matrix, From time 1 to time 2 real-time observation data The cumulative sum of the autocorrelation matrix, This is a transpose.

[0034] It can be seen that only historical data needs to be stored each time. , , , Observational data and the state data of particle i , This allows us to calculate the fitness function of particle i at time n, avoiding the extensive calculations required to solve for the mean square error of the fit.

[0035] In one possible implementation, the velocity of the particle swarm is represented by the following formula: ; in, For the first Sub-iteration particles No. Dimensional speed, For the first Sub-iteration particles No. Dimensional speed, For inertial weights, For the first Moment Particle The Dimensional components, and As learning factors, the maximum learning step size towards its own historical best and the population best is adjusted respectively. For disturbance factor, and A random number between [0, 1] A random number between [-1, 1] For its own best historical position, This represents the optimal position for the population. The optimal position is determined by the fitness function. Furthermore, the search space of the particles can be restricted. and speed The vector form of the coefficient iteration formula is as follows: ; ; in, Let be the coefficient vector of particle i at time n; This is the best in its own history; This is the optimal position for the population.

[0036] In step S5, under the unlocked state, the selector switch of the digital-to-analog converter is connected to the time-holding controller; the time-holding controller uses the frequency difference model to predict the second frequency difference, generates the second voltage control signal of the thermostatic crystal oscillator based on the second frequency difference, and applies the second voltage control signal to the thermostatic crystal oscillator through the digital-to-analog converter to compensate for the frequency difference of the thermostatic crystal oscillator, thereby realizing the time synchronization between the thermostatic crystal oscillator and the GNSS under the unlocked state.

[0037] In one possible implementation, the second frequency difference of the unlocked state is determined according to the following formula. : ; in, for Aging coefficient over time The start time of the unlocked state for Temperature coefficient at time For the first Temperature data at any time For the first Temperature data at any time This represents the initial frequency difference in the unlocked state.

[0038] The initial frequency difference of the unlocked state is determined using the following formula. : ; in, This is the time when the locked state ends. For the average filter window, The start time of the unlocked state for Aging coefficient over time for Temperature coefficient at time For the first Temperature data at any time For the first Temperature data at any given time. This invention uses DPSO to estimate temperature and aging coefficient in real time; however, least squares method, BP neural network, etc., can also be used in practice.

[0039] The present invention provides a time synchronization method based on a GNSS-disciplined cryogenic crystal oscillator. In the locked state, the digital-to-analog converter (DAC) selector switch is connected to the discipline controller. The discipline controller first acquires the time deviation between the cryogenic crystal oscillator and the GNSS, and then calculates a first voltage control signal for the cryogenic crystal oscillator based on the discipline algorithm. This first voltage control signal is applied to the cryogenic crystal oscillator via the DAC, changing the output frequency of the cryogenic crystal oscillator. This loop enables time synchronization between the cryogenic crystal oscillator and the GNSS in the locked state. At this time, the time-holding controller continuously receives the frequency difference data output by the discipline controller and the temperature information measured by the temperature sensor, serving to build a model using the time-holding algorithm. In the unlocked state, the controller switch is connected to the time-holding controller. The temperature sensor needs to continuously collect the temperature information of the cryogenic crystal oscillator. The time-holding controller uses the model built in the locked state to predict and uses the DAC to compensate for the frequency difference of the cryogenic crystal oscillator, reducing the frequency drift of the cryogenic crystal oscillator caused by temperature and aging, and achieving time synchronization between the cryogenic crystal oscillator and the GNSS. Thus, real-time time synchronization is achieved with a small computational load in both the unlocked and locked GNSS states.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A time synchronization method based on GNSS tamed oven crystal oscillator, characterized in that, The output terminals of the thermostatic crystal oscillator and the GNSS receiver are respectively connected to the input terminal of the discipline controller; the output terminal of the discipline controller and the output terminal of the temperature sensor are respectively connected to the input terminal of the time-hold controller; the output terminals of the discipline controller and the time-hold controller are respectively connected to the input terminal of the digital-to-analog converter via selector switches. The output terminal of the digital-to-analog converter is connected to the input terminal of the thermostatic crystal oscillator; the time synchronization method includes: The temperature sensor collects the temperature data of the thermostatic crystal oscillator in real time. In the locked state, the selector switch of the digital-to-analog converter is connected to the discipline controller; the discipline controller acquires a first time from the temperature-controlled crystal oscillator and a second time from the GNSS receiver, calculates the time deviation based on the first time and the second time, and sends it to the time-holding controller; The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm, generates a first voltage control signal for the thermostatic crystal oscillator based on the first frequency difference, and applies the first voltage control signal to the thermostatic crystal oscillator through a digital-to-analog converter to compensate for the frequency difference of the thermostatic crystal oscillator, thereby achieving time synchronization between the thermostatic crystal oscillator and the GNSS in the locked state. The time-holding controller establishes a frequency difference model based on the continuously received first frequency difference and the temperature data; In the unlocked state, the selector switch of the digital-to-analog converter is connected to the time-holding controller; the time-holding controller uses the frequency difference model to predict the second frequency difference, generates a second voltage control signal for the thermostatic crystal oscillator based on the second frequency difference, and applies the second voltage control signal to the thermostatic crystal oscillator through the digital-to-analog converter to compensate for the frequency difference of the thermostatic crystal oscillator, thereby achieving time synchronization between the thermostatic crystal oscillator and the GNSS in the unlocked state.

2. The time synchronization method of claim 1, wherein, The time-holding controller establishes a frequency difference model based on the continuously received first frequency difference and the temperature data, including: Initialize the velocity and position of the particles; the particles include: temperature coefficient, aging coefficient, and initial frequency difference; The fitness of the particle swarm is calculated based on the recursive mean square error. A particle swarm optimization algorithm with perturbation terms is used to find the globally optimal particle; the globally optimal particle includes: temperature coefficient, aging coefficient, and initial frequency difference; The velocity and position of particles are updated using the Discrete Particle Swarm Optimization (DPSO) algorithm.

3. The time synchronization method of claim 2, wherein, The time-holding controller, based on the continuously received first frequency difference and the temperature data, further includes the following: The recursive mean square error is expressed by the following formula: ； in, For particles No. Mean square error at time, For particles No. Mean square error at time, Let n be the instantaneous squared error at time n. This is a correction term introduced due to parameter updates. For the first Frequency difference of time, For the first Observational data at any time For particles No. coefficient vector at time step For particles No. The speed of time From time 1 to time 2 real-time observation data With frequency difference The cumulative sum of cross-correlation vectors, From time 1 to time 2 The cumulative sum of the cross-correlation vectors between the time-series observation data and the frequency difference. From time 1 to time 2 real-time observation data The cumulative sum of the autocorrelation matrix, From time 1 to time 2 real-time observation data The cumulative sum of the autocorrelation matrix, This is a transpose.

4. The time synchronization method according to claim 2, characterized in that, The time-holding controller, based on the continuously received first frequency difference and the temperature data, further includes the following: The fitness of a particle swarm is expressed by the following formula; ； in, For particles No. Adaptability at any time For the first Frequency difference vector at time step, For the first Transpose of the observation vector at time step For particles No. The coefficient vector at time step.

5. The time synchronization method of claim 2, wherein, The time-holding controller, based on the continuously received first frequency difference and the temperature data, further includes the following: The velocity of the particle swarm can be expressed by the following formula: ； in, For the first Sub-iteration particles No. Dimensional speed, For the first Sub-iteration particles No. Dimensional speed, For inertial weights, For the first Moment Particle The Dimensional components, and For learning factors, For disturbance factor, and A random number between [0, 1] A random number between [-1, 1] For its own best historical position, This is the optimal position for the population.

6. The time synchronization method according to claim 1, characterized in that, The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm, including: The first frequency difference is expressed according to the following equation : ； wherein is an aging coefficient, is a temperature coefficient, is an initial frequency difference in the locked state, is temperature data at the moment, is time.

7. The time synchronization method of claim 1, wherein, The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm, including: The time-holding controller determines the first frequency difference based on the time deviation and the discipline algorithm; The first frequency difference is filtered by a moving average filter to obtain the filtered first frequency difference. The control quantity of the digital-to-analog converter is determined based on the frequency difference proportionality coefficient of the isothermal crystal oscillator and the first frequency difference after filtering.

8. The time synchronization method of claim 7, wherein, The process of determining the control quantity of the digital-to-analog converter based on the frequency difference proportionality coefficient of the isothermal crystal oscillator and the first frequency difference after filtering includes: Perform moving average filtering according to the following formula: ； in, After filtering Frequency difference data at time, for The original frequency difference data at any given time For sliding windows, j relative to the current time t The offset; The control quantity is expressed according to the following formula : ； wherein, is the control quantity of the time-to-digital converter, is the frequency difference proportional coefficient of the oven-controlled crystal oscillator, is the filtered frequency difference data of the time.

9. The time synchronization method of claim 1, wherein, The time-holding controller uses the frequency difference model to predict the second frequency difference, including: The second frequency difference in the lost lock state is determined according to the following equation : ； wherein is the aging coefficient at the time instant, is the start time instant of the loss of lock, is the temperature coefficient at the time instant, is the temperature data at the time instant, is the temperature data at the time instant, is the temperature data at the time instant, is the temperature data at the time instant, is the initial frequency difference of the loss of lock.

10. The time synchronization method of claim 9, wherein, The time-holding controller further includes using the frequency difference model to predict the second frequency difference, including: The initial frequency difference of the loss of lock state is determined according to the following formula : ； in, This is the time when the locked state ends. For the average filter window, The start time of the unlocked state. for Aging coefficient over time for Temperature coefficient at time For the first Temperature data at any time For the first Temperature data at any given time.

Images