Method and device for deducting frequency drift of hydrogen atomic clock

Abstract

The invention relates to a method and a device for deducting frequency drift of a hydrogen atomic clock. The method comprises the following steps: acquiring a historical frequency signal which is output by the hydrogen atomic clock and traces first time forward from a current moment; performing wavelet decomposition on the historical frequency signal to obtain a plurality of component signals; performing long-term stability analysis on each component signal, taking the component signal with the worst long-term stability as a long-term component signal, and taking the other component signals as general component signals; predicting a component signal corresponding to the long-term component signal in a future second time; predicting a component signal corresponding to each general component signal in a second time in the future; coupling the predicted component signals in the second time in the future to obtain predicted frequency signals in the second time in the future; determining a predicted frequency deviation sequence in a second time in the future; and in the second time in the future, performing frequency drift deduction on the frequency signal output by the hydrogen atomic clock based on the predicted frequency deviation sequence in the second time in the future.

Description

Technical Field

[0001] This invention relates to the field of hydrogen atomic clock technology, and more specifically to a method and apparatus for frequency drift reduction of a hydrogen atomic clock. Background Technology

[0002] While hydrogen atomic clocks possess excellent short-term frequency stability, their output frequency undergoes a slow and monotonous change over time, known as frequency drift. This drift is primarily caused by long-term variations in the internal physical systems of the atomic clock, such as the pressure of hydrogen gas within the storage bulb and the aging effect of the resonant cavity. These factors cause the output frequency of the hydrogen clock to deviate from its nominal value, exhibiting a long-term, non-linear trend.

[0003] The direct consequence of frequency drift is that it severely limits the long-term frequency stability (such as daily and monthly stability) of hydrogen atomic clocks. On the Allen variance curve, which evaluates frequency stability, the drift effect manifests as a "plateau" or rising edge over long sampling times (such as hours to days), rather than the ideal continuous downward trend. This significantly reduces the performance of hydrogen clocks in high-precision applications that require long-term continuous operation. Summary of the Invention

[0004] The purpose of this invention is to provide a frequency drift reduction method for hydrogen atomic clocks to improve their long-term stability.

[0005] To achieve the above objectives, the present invention provides a method for frequency drift reduction of a hydrogen atomic clock, comprising:

[0006] Obtain the historical frequency signal output by the hydrogen atomic clock, tracing back to the first time point from the current moment;

[0007] Wavelet decomposition is performed on historical frequency signals to obtain multiple component signals;

[0008] Long-term stability analysis is performed on each of the component signals to determine the component signal with the worst long-term stability; the component signal with the worst long-term stability is taken as the long-term component signal, and the remaining component signals are taken as general component signals.

[0009] The long-term component signal is used to fit a preset linear model, and the fitted linear model is used to predict the component signal in the second time period corresponding to the long-term component signal.

[0010] For each general component signal, the general component signal is input into a pre-trained signal prediction model so that the signal prediction model can predict the component signal corresponding to the general component signal in the second future time period.

[0011] The component signals in the second future time period corresponding to the predicted long-term component signals are coupled with the component signals in the second future time period corresponding to each general component signal to obtain the frequency signals in the predicted second future time period.

[0012] Subtracting the ideal reference frequency signal for the second time period from the predicted frequency signal for the second time period yields the predicted frequency deviation sequence for the second time period.

[0013] In the second future time period, frequency drift is subtracted from the frequency signal output by the hydrogen atomic clock based on the predicted frequency deviation sequence for the second future time period.

[0014] Optionally, the wavelet decomposition has J layers, and each component signal includes the low-frequency component signal of the Jth layer and the high-frequency component signals from the first to the Jth layers.

[0015] Optionally, long-term stability analysis is performed on each of the component signals to determine the component signal with the worst long-term stability, specifically including:

[0016] For each component signal, calculate the Allan bias of that component signal;

[0017] The component signal with the largest Allan deviation is obtained as the component signal with the worst long-term stability.

[0018] Optionally, a pre-defined linear model is fitted using long-term component signals, specifically including:

[0019] The long-term component signal is filtered once to obtain the processed long-term component signal.

[0020] The parameters of a pre-defined linear model are fitted using the processed long-term component signal.

[0021] Optionally, the training method for the signal prediction model includes:

[0022] Multiple training samples and their corresponding labels are obtained. Each training sample is the frequency signal of the hydrogen atomic clock in the first time period, and the label corresponding to the training sample is the frequency signal of the hydrogen atomic clock in the second time period. The first time period and the second time period are connected and do not overlap.

[0023] Each training sample is input into a preset signal prediction model so that the signal prediction model outputs the predicted value of the frequency signal in the second time period corresponding to the training sample.

[0024] The training sample is concatenated with the predicted value of the frequency signal in the second time period corresponding to the training sample to obtain the concatenated signal; the Allan bias of the concatenated signal is calculated as the first loss term.

[0025] The deviation between the predicted value of the frequency signal and the label in the second time period corresponding to the training sample is calculated and used as the second loss term;

[0026] The loss of the signal prediction model is obtained by adding the first loss term and the second loss term.

[0027] With the goal of minimizing the loss of the signal prediction model, the parameters of the signal prediction model are updated to obtain a trained signal prediction model.

[0028] Another aspect of the present invention provides a frequency drift reduction device for a hydrogen atomic clock, comprising:

[0029] The acquisition module is used to acquire the historical frequency signal output by the hydrogen atomic clock, which traces back to the first time from the current moment.

[0030] The decomposition module is used to perform wavelet decomposition on historical frequency signals to obtain multiple component signals;

[0031] The long-term stability analysis module is used to perform long-term stability analysis on each of the component signals to determine the component signal with the worst long-term stability; the component signal with the worst long-term stability is taken as the long-term component signal, and the remaining component signals are taken as general component signals.

[0032] The fitting module is used to fit a preset linear model with the long-term component signal and use the fitted linear model to predict the component signal in the second future time period corresponding to the long-term component signal.

[0033] The prediction module is used to input each general component signal into a pre-trained signal prediction model so that the signal prediction model can predict the component signal corresponding to the general component signal in the next time period.

[0034] The coupling module is used to couple the component signals in the second future time period corresponding to the predicted long-term component signal with the component signals in the second future time period corresponding to each general component signal, so as to obtain the frequency signal in the predicted second future time period.

[0035] The frequency offset calculation module is used to subtract the ideal reference frequency signal for the second time period from the predicted frequency signal for the second time period to obtain the predicted frequency offset sequence for the second time period.

[0036] The subtraction module is used to subtract frequency drift from the frequency signal output by the hydrogen atomic clock based on the predicted frequency deviation sequence for the second time period in the future.

[0037] Optionally, the wavelet decomposition has J layers, and each component signal includes the low-frequency component signal of the Jth layer and the high-frequency component signals from the first to the Jth layers.

[0038] Optionally, long-term stability analysis is performed on each of the component signals to determine the component signal with the worst long-term stability, specifically including:

[0039] For each component signal, calculate the Allan bias of that component signal;

[0040] The component signal with the largest Allan deviation is obtained as the component signal with the worst long-term stability.

[0041] Optionally, a pre-defined linear model is fitted using long-term component signals, specifically including:

[0042] The long-term component signal is filtered once to obtain the processed long-term component signal.

[0043] The parameters of a pre-defined linear model are fitted using the processed long-term component signal.

[0044] Optionally, the training method for the signal prediction model includes:

[0045] Multiple training samples and their corresponding labels are obtained. Each training sample is the frequency signal of the hydrogen atomic clock in the first time period, and the label corresponding to the training sample is the frequency signal of the hydrogen atomic clock in the second time period. The first time period and the second time period are connected and do not overlap.

[0046] Each training sample is input into a preset signal prediction model so that the signal prediction model outputs the predicted value of the frequency signal in the second time period corresponding to the training sample.

[0047] The training sample is concatenated with the predicted value of the frequency signal in the second time period corresponding to the training sample to obtain the concatenated signal; the Allan bias of the concatenated signal is calculated as the first loss term.

[0048] The deviation between the predicted value of the frequency signal and the label in the second time period corresponding to the training sample is calculated and used as the second loss term;

[0049] The loss of the signal prediction model is obtained by adding the first loss term and the second loss term.

[0050] With the goal of minimizing the loss of the signal prediction model, the parameters of the signal prediction model are updated to obtain a trained signal prediction model. Attached Figure Description

[0051] Figure 1 A flowchart of a hydrogen atomic clock frequency drift reduction method according to an embodiment of the present invention;

[0052] Figure 2 This is a structural block diagram of the frequency drift reduction device for a hydrogen atomic clock according to an embodiment of the present invention. Detailed Implementation

[0053] The preferred embodiments of the present invention are given below with reference to the accompanying drawings and described in detail.

[0054] like Figure 1 As shown, this embodiment of the invention provides a method for frequency drift reduction of a hydrogen atomic clock, which includes the following steps:

[0055] S100: Acquire the historical frequency signal output by the hydrogen atomic clock, which traces back to the first time from the current moment.

[0056] When a hydrogen atomic clock is operating, it outputs a continuous frequency signal. By sampling this continuous frequency signal at a preset sampling rate, discrete frequency signals can be obtained. The historical frequency signal is a discrete frequency signal, which includes multiple frequency values ​​traced back to the first time period from the current moment. The first time period can be set as needed, for example, 15 days.

[0057] In some embodiments, the hydrogen atomic clock outputs a phase signal rather than a frequency signal. Therefore, in order to obtain the frequency signal, the phase signal needs to be preprocessed. Specifically, a first-order differential circuit can be used to convert the phase signal into a frequency signal.

[0058] S200: Perform wavelet decomposition on the historical frequency signal to obtain multiple component signals.

[0059] Wavelet decomposition is a time-frequency analysis method that decomposes the original signal into low-frequency and high-frequency components by convolving it with wavelet basis functions. Discrete wavelet decomposition decomposes historical frequency signals at a series of different scales and resolutions, continuously separating high-frequency components. This decouples the mixed original signal into fundamental components with different time-frequency characteristics, revealing the signal's intrinsic structure and hierarchical features.

[0060] Assuming the historical frequency signal is Its length is N, and the number of wavelet decomposition levels is J. The wavelet decomposition process is as follows:

[0061] First-level decomposition: Filter and downsample the historical signal frequency to obtain the first-level low-frequency component (also known as low-frequency coefficient) and the first-level high-frequency component (also known as high-frequency coefficient).

[0062] Second-level decomposition: Filter and downsample the first-level low-frequency components to obtain the second-level low-frequency components and the second-level high-frequency components.

[0063] Similarly, each layer of decomposition filters and downsamples the low-frequency components of the previous layer to obtain the low-frequency and high-frequency components of that layer.

[0064] Low-frequency coefficients of layer j for:

[0065] (1)

[0066] in, , Here are the low-pass filter coefficients, and k is the index after downsampling, indicating the position of the nth sampling point (or filter window) in the next decomposition layer, with values ​​ranging from 1 to 10. ,in, for The length.

[0067] High-frequency coefficients of layer j from The layers are separated as follows:

[0068] (2)

[0069] in, These are the coefficients of the high-pass filter.

[0070] After J-level decomposition, one low-frequency coefficient and J high-frequency coefficients can be obtained:

[0071] (3)

[0072] Each coefficient represents a component of the signal at a certain scale. Therefore, the low-frequency coefficients of the Jth layer and the high-frequency coefficients from the first to the Jth layers form the component signals, meaning there are a total of J+1 component signals. The value of the decomposition layer J can be selected as needed.

[0073] S300: Perform long-term stability analysis on each of the component signals to determine the component signal with the worst long-term stability; take the component signal with the worst long-term stability as the long-term component signal, and take the remaining component signals as general component signals.

[0074] In some embodiments, overlapping Allan bias can be used for long-term stability analysis, performing long-term stability analysis on each of the component signals to determine the component signal with the worst long-term stability, specifically including:

[0075] For each component signal, calculate the Allan bias of that component signal;

[0076] The component signal with the largest Allan deviation is obtained as the component signal with the worst long-term stability.

[0077] The calculation method for Allen's bias is as follows:

[0078] (4)

[0079] in, Let M be the Allan deviation of the component signal y, and M be the length of the component signal. The time interval for calculating the Allen variance, The basic measurement interval (i.e., the sampling interval between any two adjacent signal values ​​of the component signal) is denoted by m, and the averaging factor is denoted by m over time. The number of signal values ​​contained within. This represents the (i+m)th signal value of the component signal y. Let s represent the i-th signal value of component signal y, where i and s are both positive integers.

[0080] By adjusting the value of m, Approaching 86,400 seconds, the Allan deviation calculated at this point can represent the long-term stability of the component signal.

[0081] The larger the Allan deviation, the worse the long-term stability of the component signal. Therefore, the component signal with the largest Allan deviation is the signal with the worst long-term stability.

[0082] Long-term signals retain the best components of long-term trends and are the components with the most obvious long-term characteristics. General component signals, on the other hand, retain more short-term characteristics and their long-term characteristics are not obvious.

[0083] S400: Fits a preset linear model using the long-term component signal, and uses the fitted linear model to predict the component signal in the second future time period corresponding to the long-term component signal.

[0084] In some embodiments, the preset linear model is:

[0085] (5)

[0086] Where y is the signal value (i.e., the frequency value), t is time, and a and b are the parameters to be fitted.

[0087] Fitting a pre-defined linear model using long-term component signals specifically includes:

[0088] The long-term component signal is filtered once to obtain the processed long-term component signal, thereby reducing noise and interference from short-term fluctuations.

[0089] The signal values ​​of the processed long-term component signals are substituted into the above linear model, and the parameters a and b of the linear model are obtained by fitting based on a preset fitting method (such as the least squares method).

[0090] Substitute multiple time points in the second future time period into the fitted linear model to obtain signal values ​​at different time points. The signal values ​​at different time points constitute the component signal in the second future time period corresponding to the long-term component signal.

[0091] The value of the second time period can be selected according to actual needs. The second time period can be the same as or different from the first time period. For example, both the first and second time periods are 15 days.

[0092] S500: For each general component signal, the general component signal is input into a pre-trained signal prediction model so that the signal prediction model can predict the component signal corresponding to the general component signal in the second future time period.

[0093] The signal prediction model can be any suitable machine learning model, such as an RNN (Recurrent Neural Network) model, an LSTM (Long Short-Term Memory Artificial Neural Network) model, or any other suitable machine learning model. For example, this invention uses an optimized PatchTST model. The PatchTST model normalizes the input univariate sequence to stabilize the data distribution and performs block processing, cutting the univariate sequence into adjacent local segments (patches) with a 50% overlap rate, thereby effectively capturing local dependencies and reducing sequence length. Each block is then transformed into a vector sequence that can be input into a Transformer through linear projection and position embedding. Specifically, the size of each patch after block division is... ,in This is the patch length, which is also the number of time steps; there are a total of... Each block can be viewed as a matrix. Through matrix multiplication Each length is The block is mapped to the new In 3D space, the result The Transformer's self-attention mechanism is permutation-invariant; it cannot perceive the order of the input sequence. Therefore, positional encoding is added to inject a unique "position signal" (i.e., a learnable matrix) into each position. This allows the model to understand the temporal relationships of the sequences. The final input to the Transformer is... The core component of the PatchTST model is the Transformer encoder, which mines global long-term dependencies between blocks through a multi-head self-attention mechanism and ensures training stability through residual connections and layer normalization. The encoder output is flattened and then mapped back to the predicted sequence by the linear head. For the specific principles of the PatchTST model, please refer to the literature Yuqi N, Nam HN, Phanwadee S, Jayant K, et al. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers[J], Computing Research Repository, 2023, which will not be elaborated here.

[0094] The optimized PatchTST model used in this invention differs from existing PatchTST models in that the optimized model incorporates Allan variance into the loss during training, while existing PatchTST models do not. The training method for the signal prediction model of this invention is as follows:

[0095] Multiple training samples and their corresponding labels are obtained. Each training sample is the frequency signal of the hydrogen atomic clock in the first time period, and the label corresponding to the training sample is the frequency signal of the hydrogen atomic clock in the second time period. The first time period and the second time period are connected and do not overlap.

[0096] Each training sample is input into a preset signal prediction model so that the signal prediction model outputs the predicted value of the frequency signal in the second time period corresponding to the training sample.

[0097] The training sample is concatenated with the predicted value of the frequency signal in the second time period corresponding to the training sample to obtain the concatenated signal; the Allan bias of the concatenated signal is calculated as the first loss term.

[0098] The deviation between the predicted value of the frequency signal and the label in the second time period corresponding to the training sample is calculated and used as the second loss term;

[0099] The loss of the signal prediction model is obtained by adding the first loss term and the second loss term together.

[0100] The parameters of the signal prediction model are updated with the goal of minimizing the loss of the signal prediction model in order to obtain a well-trained signal prediction model.

[0101] After obtaining the trained signal prediction model, each general component signal can be input into it, and the signal prediction model will output the predicted component signal corresponding to the second time period in the future for each general component signal.

[0102] S600: Couple the predicted long-term component signal corresponding to the component signal in the second future time period with the component signal in the second future time period corresponding to each general component signal to obtain the predicted frequency signal in the second future time period.

[0103] As described in step S200, since each component signal is obtained by wavelet decomposition of historical frequency signals, the component signal corresponding to each component signal in the future second time period is also a component signal of the frequency signal in the future second time period. In order to obtain the frequency signal in the future second time period, the component signals corresponding to each predicted component signal in the future second time period can be coupled. The coupling process is the reverse process of wavelet decomposition in step S200, that is, the original signal is coupled out through the low-frequency coefficients and high-frequency coefficients after wavelet decomposition. The specific coupling method is as follows: starting from the low-frequency coefficients and high-frequency coefficients of the bottom layer (i.e., the Jth layer), the low-frequency coefficients of the previous layer are restored, and then the low-frequency coefficients of the previous layer are restored using the low-frequency coefficients and high-frequency coefficients of the previous layer, and so on, layer by layer, until the original signal is finally obtained.

[0104] Let's take the example of reconstructing the low-frequency coefficients of layer j from the low-frequency and high-frequency coefficients of layer j+1 to illustrate:

[0105] First, upsampling is performed, inserting 0s into the low-frequency and high-frequency coefficients of the (j+1)th layer to restore the length of the low-frequency coefficients of the jth layer:

[0106] (6)

[0107] (7)

[0108] Then, convolve the two upsampled coefficients separately to obtain the convolved coefficients:

[0109] (8)

[0110] (9)

[0111] in, The coefficients are the result of convolving the low-frequency coefficients of the (j+1)th layer. Let m be the coefficients after convolving the high-frequency coefficients of the (j+1)th layer, and m be the index of the zero-plugging coefficients. and It is a reconstruction filter for low-frequency and high-frequency components.

[0112] Will and By adding them together, we can obtain the low-frequency coefficients of the j-th layer. .

[0113] S700: Subtract the ideal reference frequency signal for the second time period from the predicted frequency signal for the second time period to obtain the predicted frequency deviation sequence for the second time period.

[0114] The ideal reference frequency signal can be obtained in advance. The difference between the predicted frequency signal in the second time period and the ideal reference frequency signal in the second time period is the predicted frequency deviation sequence in the second time period, which includes frequency deviation values ​​at multiple different time points.

[0115] S800: In the second future time period, frequency drift is subtracted from the frequency signal output by the hydrogen atomic clock based on the predicted frequency deviation sequence in the second future time period.

[0116] After obtaining the frequency deviation sequence for the predicted future time period, when the hydrogen atomic clock outputs the frequency at any time in the second future time period, the frequency deviation value corresponding to that time in the predicted frequency deviation sequence for the second future time period can be subtracted from the output frequency of the hydrogen atomic clock at that time to subtract the frequency deviation value at that time in the predicted frequency deviation sequence for the second future time period, thereby subtracting the frequency offset to obtain the frequency after frequency drift subtraction; the frequency after frequency drift subtraction at different time points in the second future time period is the frequency signal for the second future time period.

[0117] In some embodiments, since a phase signal is required to be output in the end, the phase signal in the second time interval can be obtained by accumulating the frequency signal (i.e., the inverse process of differential).

[0118] The frequency drift reduction method for hydrogen atomic clocks in this invention decomposes historical frequency signals into multiple component signals. Based on the characteristics of each component signal, different prediction models are used to predict the component signals corresponding to each component signal in the future period. The predicted future component signals are then coupled to obtain the frequency signals in the future period. Frequency drift reduction is performed based on the frequency signals in the future period, thereby improving the long-term frequency stability of the hydrogen atomic clock.

[0119] like Figure 2 As shown, this embodiment of the invention also provides a frequency drift subtraction device for a hydrogen atomic clock, which includes: an acquisition module 10, a decomposition module 20, a long-term stability analysis module 30, a fitting module 40, a prediction module 50, a coupling module 60, a frequency offset calculation module 70, and a subtraction module 80.

[0120] The acquisition module 10 is used to acquire the historical frequency signal output by the hydrogen atomic clock, which traces back to the first time from the current moment.

[0121] The decomposition module 20 is used to perform wavelet decomposition on the historical frequency signal to obtain multiple component signals;

[0122] The long-term stability analysis module 30 is used to perform long-term stability analysis on each of the component signals to determine the component signal with the worst long-term stability; the component signal with the worst long-term stability is taken as the long-term component signal, and the remaining component signals are taken as general component signals.

[0123] The fitting module 40 is used to fit a preset linear model with the long-term component signal and use the fitted linear model to predict the component signal in the second time period corresponding to the long-term component signal.

[0124] For each general component signal, the prediction module 50 inputs the general component signal into a pre-trained signal prediction model so that the signal prediction model predicts the component signal corresponding to the general component signal in the next time period.

[0125] The coupling module 60 is used to couple the component signal in the second time period of the future corresponding to the predicted long-term component signal with the component signal in the second time period of the future corresponding to each general component signal, so as to obtain the frequency signal in the predicted second time period of the future.

[0126] The frequency offset calculation module 70 is used to subtract the ideal reference frequency signal in the second time period from the predicted frequency signal in the second time period to obtain the predicted frequency offset sequence in the second time period.

[0127] The subtraction module 80 is used to subtract frequency drift from the frequency signal output by the hydrogen atomic clock based on the predicted frequency deviation sequence for the second time period in the future.

[0128] The acquisition module 10, decomposition module 20, long-term stability analysis module 30, fitting module 40, prediction module 50, coupling module 60, frequency offset calculation module 70, and subtraction module 80 are functional modules that implement steps S100-S800 in the method embodiment. Their specific implementation methods are the same as those in the method embodiment, and can be referred to the description in the method embodiment. They will not be repeated here.

[0129] The frequency drift reduction device for a hydrogen atomic clock according to this invention decomposes a historical frequency signal into multiple component signals. Based on the characteristics of each component signal, different prediction models are used to predict the component signals corresponding to each component signal in the future period. The predicted future component signals are then coupled to obtain the frequency signals in the future period. Frequency drift reduction is performed based on the frequency signals in the future period, thereby improving the long-term frequency stability of the hydrogen atomic clock.

[0130] Another embodiment of the present invention provides a readable storage medium having a computer program stored thereon. When the computer program is executed in a computer, it causes the computer to perform the steps of the frequency drift subtraction method for the hydrogen atomic clock in the above embodiments of the present invention.

[0131] Another embodiment of the present invention provides an electronic device, which includes a memory and a processor. The memory stores executable code. When the processor executes the executable code, it performs the steps of the frequency drift subtraction method for the hydrogen atomic clock in the above embodiments of the present invention.

[0132] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0133] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] In a typical configuration, an electronic device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0139] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by electronic devices. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0142] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0144] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0145] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0146] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. That is, all simple and equivalent changes and modifications made based on the claims and description of this invention fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.

Claims

1. A method of frequency drift subtraction for a hydrogen atomic clock, characterized in that, The method comprises the following steps: obtaining a historical frequency signal of a hydrogen atomic clock output at a first time before a current time; wavelet-decomposing the historical frequency signal to obtain a plurality of component signals; performing long-term stability analysis on each of the component signals to determine a component signal with the worst long-term stability, taking the component signal with the worst long-term stability as a long-term component signal, and taking the remaining component signals as general component signals; fitting a preset linear model by using the long-term component signal, and predicting a component signal corresponding to the long-term component signal in a second time in the future by using the fitted linear model; for each general component signal, inputting the general component signal into a pre-trained signal prediction model to enable the signal prediction model to predict a component signal corresponding to the general component signal in the second time in the future; coupling the predicted component signal corresponding to the long-term component signal in the second time in the future and the component signals corresponding to the general component signals in the second time in the future to obtain a predicted frequency signal in the second time in the future; subtracting an ideal reference frequency signal in the second time in the future from the predicted frequency signal in the second time in the future to obtain a predicted frequency deviation sequence in the second time in the future; and based on the predicted frequency deviation sequence in the second time in the future, performing frequency drift deduction on the frequency signal of the hydrogen atomic clock output in the second time in the future.

2. The frequency drift subtraction method of a hydrogen maser as claimed in claim 1, characterized in that, The number of layers of wavelet decomposition is J Each of the component signals includes a low frequency component signal of the first J layer and high frequency component signals from the first to the J seventh layers.

3. The frequency drift subtraction method of a hydrogen maser as claimed in claim 1, characterized by, The method of performing long-term stability analysis on each of the component signals to determine a component signal with the worst long-term stability comprises the following steps: for each component signal, calculating an Allan deviation of the component signal; obtaining a component signal with the largest Allan deviation as the component signal with the worst long-term stability.

4. The frequency drift subtraction method of a hydrogen maser as claimed in claim 1, characterized by, The method of fitting a preset linear model by using the long-term component signal comprises the following steps: performing filtering processing on the long-term component signal to obtain a processed long-term component signal; fitting parameters of the preset linear model by using the processed long-term component signal.

5. The frequency drift subtraction method of a hydrogen maser as claimed in claim 1, characterized by, The training method of the signal prediction model comprises the following steps: obtaining a plurality of training samples and labels corresponding to the training samples, wherein each training sample is a frequency signal of a hydrogen atomic clock in a first time, and a label corresponding to each training sample is a frequency signal of the hydrogen atomic clock in a second time, the first time and the second time being consecutive and having no overlap; inputting each training sample into a preset signal prediction model to enable the signal prediction model to output a predicted value of the frequency signal of the training sample in the second time; splicing the training sample and the predicted value of the frequency signal of the training sample in the second time to obtain a spliced signal, and calculating an Allan deviation of the spliced signal as a first loss term; calculating a deviation between the predicted value of the frequency signal of the training sample in the second time and the label as a second loss term; adding the first loss term and the second loss term to obtain a loss of the signal prediction model; updating parameters of the signal prediction model to obtain a trained signal prediction model, with the aim of minimizing the loss of the signal prediction model.

6. A frequency drift subtraction device for a hydrogen maser, characterized by The method comprises the following steps: an obtaining module, configured to obtain a historical frequency signal of a hydrogen atomic clock output at a first time before a current time; a decomposing module, configured to wavelet-decompose the historical frequency signal to obtain a plurality of component signals; The long-term analysis module is configured to perform long-term analysis on each of the component signals to determine a component signal with the worst long-term stability, and determine the component signal with the worst long-term stability as a long-term component signal and the remaining component signals as general component signals. The fitting module is configured to fit a preset linear model with the long-term component signal, and predict the component signal corresponding to the long-term component signal in a future second time with the fitted linear model. The prediction module is configured to input each general component signal into a pre-trained signal prediction model to enable the signal prediction model to predict the component signal corresponding to the general component signal in the future second time. The coupling module is configured to couple the predicted component signal corresponding to the long-term component signal in the future second time and the predicted component signal corresponding to each general component signal in the future second time to obtain a predicted frequency signal in the future second time. The frequency offset calculation module is configured to subtract an ideal reference frequency signal in the future second time from the predicted frequency signal in the future second time to obtain a predicted frequency offset sequence in the future second time. The deduction module is configured to deduct frequency drift of a frequency signal output by the hydrogen atomic clock in the future second time based on the predicted frequency offset sequence in the future second time.

7. The apparatus for canceling the frequency drift of a hydrogen maser as defined in claim 6, wherein The number of layers of wavelet decomposition is J , and each component signal includes a low-frequency component signal of the first J layer and high-frequency component signals from the first to the J layer.

8. The apparatus for canceling the frequency drift of a hydrogen maser as defined in claim 6, wherein The long-term analysis on each of the component signals to determine a component signal with the worst long-term stability includes: For each component signal, calculating an Allan deviation of the component signal. Obtaining a component signal with the largest Allan deviation as the component signal with the worst long-term stability.

9. The apparatus for canceling the frequency drift of a hydrogen maser as defined in claim 6, wherein The fitting of the preset linear model with the long-term component signal includes: Performing a filtering process on the long-term component signal to obtain a processed long-term component signal. Fitting a parameter of the preset linear model with the processed long-term component signal.

10. The apparatus for canceling the frequency drift of a hydrogen maser as defined in claim 6, wherein The training method of the signal prediction model includes: Obtaining a plurality of training samples and labels corresponding to the training samples, wherein each training sample is a frequency signal of the hydrogen atomic clock in a first time, and the label corresponding to the training sample is a frequency signal of the hydrogen atomic clock in a second time, the first time and the second time being consecutive and non-overlapping. Inputting each training sample into a preset signal prediction model to enable the signal prediction model to output a predicted value of the frequency signal of the training sample in the second time. Splicing the training sample and the predicted value of the frequency signal of the training sample in the second time to obtain a spliced signal, and calculating an Allan deviation of the spliced signal as a first loss term. Calculating a deviation between the predicted value of the frequency signal of the training sample in the second time and the label as a second loss term. Adding the first loss term and the second loss term to obtain a loss of the signal prediction model. Updating parameters of the signal prediction model to obtain a trained signal prediction model, with the objective of minimizing the loss of the signal prediction model.

Images