International Atomic Time Deviation Extraction Method Based on Pulsar Timing Noise Frequency Division Suppression
By employing a two-stage method that combines mid-frequency noise suppression and low-frequency common TAI bias estimation, the problem of decoupling spectral overlap noise from TAI bias is solved, improving the extraction accuracy and robustness of TAI bias, adapting to different pulsar observation conditions, and achieving more stable TAI bias estimation.
Patent Information
- Application Number
- CN202611334148.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-31
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to effectively decouple low-frequency noise from spectral overlap from long-term TAI bias, resulting in insufficient accuracy and poor robustness in extracting weak long-term TAI bias. Furthermore, differences in the observation quality of different pulsars affect the stability of public TAI estimation.
A two-stage method is adopted, which combines mid-frequency noise suppression and low-frequency common TAI bias estimation. By using Fourier design matrix and B-spline smoothing constraints, mid-frequency noise and low-frequency common TAI bias are processed respectively. Combined with the error term of pulsar rotation model and the shared common TAI bias term, the signal-to-noise ratio is improved and the contribution weight is adjusted.
It significantly improves the retention and extraction sensitivity of long-period, low-amplitude TAI bias, enhances the estimation accuracy and stability of long-term TAI bias sequences, adapts to differences in observations of different pulsars, and improves the reliability and repeatability of extraction results.
Smart Images

Figure CN123113274A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of time and frequency metrology and astrometry technology, specifically involving pulsar timing data processing and atomic time performance evaluation technology. It further provides a method for extracting International Atomic Time (TAI) deviations based on pulsar timing noise frequency division suppression, which can be used for long-term stability evaluation of TAI and construction of comprehensive pulsar time. Background Technology
[0002] International Atomic Time (TAI) is generated by the International Bureau of Weights and Measures through weighted calculations of observational data from multiple atomic clocks in 85 time service laboratories worldwide. It is an indispensable time standard for fields such as satellite navigation and positioning, long-distance communication time synchronization, cross-regional time and frequency transfer, and high-precision measurements in fundamental physics. Currently, the generation, calibration, and long-term stability assessment of TAI heavily rely on time comparisons and frequency calibrations within the atomic clock cluster. The entire assessment system is homogeneous, lacking an external reference standard with a completely independent physical mechanism separate from the atomic clock system, making it difficult to independently verify the long-term stability of TAI.
[0003] Millisecond pulsars possess rotation periods with high stability on the order of billions of years, allowing radio telescopes to continuously record the arrival times of their pulse signals over long periods. By converting the observed times to a unified time reference frame and calculating the difference between this timeframe and the theoretically predicted arrival times from the pulsar timing model, the pulsar timing residual sequence can be obtained. When there is a time deviation in the ground-based reference time standard, this deviation component will be synchronously superimposed on the timing residuals of all observed pulsars, manifesting as a low-frequency common component shared by multiple stars. Based on this physical characteristic, the millisecond pulsar timescale is recognized as an effective technical approach, outside of the atomic time system, for the long-term independent monitoring and verification of TAI stability.
[0004] The timing residuals of pulsars obtained from actual observations are complex, containing not only the target common timescale (TAI) deviation component but also measurement white noise, pulsar rotation red noise, random fluctuations in interstellar dispersion (DM), instrument observation noise, and timing model residuals, among other interference terms. Rotation red noise and interstellar DM variations are predominantly low-frequency components, with their spectral distributions highly overlapping with the low-frequency TAI deviation. This spectral aliasing between low-frequency signals makes it difficult to decouple the common TAI deviation from the low-frequency noise specific to individual pulsars. During noise fitting, if the constraint range of the low-frequency noise model is set too large, the noise fitting term will absorb some of the true TAI low-frequency deviation, resulting in target signal loss. If the noise suppression constraint is too weak, the residual low-frequency noise of individual pulsars will be mixed into the common timescale estimation results, directly reducing the estimation accuracy of the TAI deviation.
[0005] Current mainstream pulsar timescale construction schemes generally adopt a multi-pulsar global joint fitting strategy, simultaneously solving for the timing model parameters of each pulsar and the common timescale components shared across pulsars. The extraction of TAI common bias relies on the constraint of "multi-star shared" characteristics. This approach can achieve basic calculations of common timescale components and shows good noise separation in the mid-to-high frequency range. However, when pulsar rotation red noise, DM low-frequency fluctuations, and TAI bias highly overlap in the low-frequency spectrum, relying solely on the single constraint of cross-star commonality cannot effectively separate the two types of low-frequency components. The extraction effect on TAI low-frequency bias, which has weak amplitude and slow long-term variation, is limited. Furthermore, due to limitations in radio telescope observation plans, the observation time span, sampling interval, and single measurement noise level vary significantly among different pulsars. During the global fitting process, the weight of each pulsar is naturally determined by the observation data and cannot be actively adjusted. Pulsars with poor observation quality or incomplete data will introduce additional noise disturbances, resulting in insufficient stability of the common TAI estimation results.
[0006] Therefore, there is an urgent need for a TAI bias extraction method that can suppress low-frequency noise of individual pulsars while fully preserving low-frequency TAI bias information, and can adaptively adjust the contribution weight of different pulsars in joint estimation, so as to improve the estimation accuracy and robustness of weak long-term TAI bias. Summary of the Invention
[0007] The purpose of this invention is to address the shortcomings of existing technologies by proposing a TAI deviation extraction method based on pulsar timing noise frequency division suppression. This method aims to solve the technical problems of existing technologies, such as difficulty in decoupling low-frequency noise with spectral overlap from long-term TAI deviations, insufficient extraction accuracy of weak long-term TAI deviations, and poor robustness. The scheme comprises two stages: mid-frequency noise suppression and joint estimation of low-frequency common TAI deviations. In the first stage, based on DM variations and the degree of separability between red noise and TAI deviations in the frequency domain, relatively separable mid-frequency noise is estimated and suppressed, improving the effective signal-to-noise ratio in the timing residuals while preserving low-frequency common TAI deviation information. In the second stage, using the noise-suppressed timing residuals as input, a joint model is constructed, consisting of error terms from each pulsar rotation model and a shared common TAI deviation term. The low-frequency TAI deviation is recovered using cross-pulsar commonality. Furthermore, the stability of common TAI deviation extraction is improved through weak component protection, single-star contribution adjustment, and B-spline smoothing constraints, forming a technical flow of "mid-frequency noise suppression—low-frequency cross-star common joint estimation."
[0008] The steps of this invention to achieve the above objectives are as follows:
[0009] (1) Obtain long-term multi-frequency timing observation data of multiple millisecond pulsars, generate timing residuals with reference to the Earth time scale TT (TAI) obtained based on the conversion of International Atomic Time (TAI), and record the observation span, pulse arrival time TOA epoch, observation frequency and measurement uncertainty of each pulsar.
[0010] (2) For the interstellar dispersion DM variation component and the pulsar rotation red noise component, the corresponding cutoff frequency and harmonic number are set respectively. The starting harmonic order of the Fourier expansion is determined according to the observation span of each pulsar. A Fourier design matrix containing continuous harmonics and subject to Tukey window function constraints is constructed starting from the starting harmonic. For the DM variation component, the Fourier design matrix is frequency scaled according to the physical property that the dispersion delay is inversely proportional to the square of the observation frequency.
[0011] (3) Based on the measurement uncertainty corresponding to TOA, construct fitting weights and use the noise design matrix obtained in step (2) to perform mid-frequency noise component estimation; including: firstly, jointly fitting the pulsar rotation model error term and the DM change component to solve the DM change reconstruction component; subtracting the DM change reconstruction component from the timing residual sequence, and then jointly fitting the pulsar rotation model error term and the rotation red noise component to solve the rotation red noise reconstruction component;
[0012] (4) Perform weak component protection constraints on the DM change reconstruction component and the spin red noise reconstruction component obtained in step (3); for each millisecond pulsar, use the product of the median of its TOA measurement uncertainty and the preset weak component protection threshold as the judgment threshold. Only when the standard deviation of the corresponding reconstruction component is not less than the judgment threshold, the reconstruction component is deducted from the original timing residual to obtain the timing residual sequence after intermediate frequency noise suppression.
[0013] (5) The low-frequency common international atomic time (TAI) bias process to be estimated is characterized by cubic B-spline basis functions. The basis functions are sampled according to the TOA epoch of each millisecond pulsar to construct a common TAI design matrix corresponding to the observation epoch of each pulsar.
[0014] (6) Perform reduced QR decomposition on the spin model error term design matrix of each millisecond pulsar to obtain the orthogonal complement space of the column space of the spin model error term design matrix; project the common TAI design matrix corresponding to the current pulsar onto the orthogonal complement space to obtain the common TAI design matrix after applying polynomial orthogonal constraints, so as to eliminate the linear correlation between the low-frequency common TAI bias estimation and the pulsar spin model parameters;
[0015] (7) Based on the common TAI design matrix after applying polynomial orthogonal constraints obtained in step (6), and the fitting weight constructed from the TOA measurement uncertainty, calculate the initial relative contribution of each millisecond pulsar to the joint estimation of common TAI bias; when the initial relative contribution exceeds the preset contribution adjustment threshold, apply the same reduction factor to the common TAI design matrix after applying polynomial orthogonal constraints and the timing residual sequence after intermediate frequency noise suppression corresponding to the pulsar, so as to adjust the relative contribution of the pulsar in the joint estimation of common TAI bias;
[0016] (8) Construct a multi-pulse star joint model that includes the error terms of each millisecond pulse star's rotation model and the common TAI deviation term shared by all millisecond pulse stars; stack the intermediate frequency noise suppression timing residuals after the contribution weight adjustment across pulse stars, and apply first-order and second-order difference smoothing constraints to the B-spline coefficients that characterize the common TAI deviation; based on the weighted least squares solution, jointly estimate the coefficients of each millisecond pulse star's rotation model error term and the B-spline coefficients corresponding to the common TAI deviation.
[0017] (9) Using the common TAI deviation B-spline coefficients, construct the orthogonalized B-spline design matrix under the same low-order polynomial orthogonal constraint as in step (6) on a unified time grid; reconstruct the common TAI deviation sequence based on the matrix and B-spline coefficients, and perform sign conversion to output the deviation sequence relative to TT(TAI) when synthesizing pulsars.
[0018] (10) Align the deviation sequence output in step (9) with the BIPM reference deviation sequence TT(BIPM)-TT(TAI) to the same evaluation epoch and conduct a post-hoc comparison. The extraction results are evaluated using three evaluation indicators: Pearson correlation coefficient, amplitude ratio, and root mean square error.
[0019] Compared with the prior art, the present invention has the following advantages:
[0020] First, this invention combines the frequency distribution characteristics of timing noise with the cross-pulsar common inherent properties of TAI deviation, and makes differentiated use of the two types of constraint information in stages. This includes suppressing mid-frequency noise by relying on frequency domain separability in the mid-frequency stage, and calculating TAI deviation by relying on cross-pulsar common constraints in the low-frequency stage. The constraint effects of the two types of information are brought into play in different stages, realizing deep decoupling and separation of noise components and target TAI deviation components. This effectively avoids the problem of low-frequency noise fitting terms absorbing and masking weak long-term TAI deviations in the traditional global fitting process, and significantly improves the ability to retain and extract long-period, low-amplitude TAI deviation signals, so that weak long-term TAI deviations can be more effectively retained and extracted.
[0021] Secondly, this invention utilizes multiple independently observed millisecond pulsars to apply joint constraints to the same common TAI bias, comprehensively utilizing the common time-varying information dispersed in the timing residuals of different pulsars. By complementing the multi-source observation information, the disturbances caused by single-star random noise are offset, reducing the interference of accidental observation errors on the estimated sequence, effectively improving the estimation accuracy of the long-term TAI bias sequence, and enhancing the characterization and resolution capabilities of the slow drift and long-term evolution trend of TAI.
[0022] Third, this scheme does not require all pulsars to have consistent observation duration, sampling interval and observation accuracy. It can adapt to the actual engineering observation scenarios in which the observation span, measurement error, red noise intensity and interstellar dispersion noise level of each star in the pulsar timing array are uneven. It comprehensively adapts to the differences of different pulsars and maintains the stability of the array joint estimation through corresponding noise protection, contribution adjustment and smoothing processing, thereby improving the reliability and repeatability of the extraction results. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating the overall implementation of the method of the present invention;
[0024] Figure 2 This is a simulation result of pulsar intermediate frequency noise fitting in this invention;
[0025] Figure 3 This is a simulation result diagram of the TAI deviation extracted in this invention;
[0026] Figure 4 This is a comparison chart of simulation results for deviation extraction using the method of this invention and existing methods. Detailed Implementation
[0027] The present invention will now be further described with reference to the accompanying drawings.
[0028] Example 2: Refer to Figure 1 This invention proposes a method for extracting international atomic time deviation based on pulsar timing noise frequency division suppression, specifically including the following steps:
[0029] Step 1) Acquire long-term, multi-frequency timing observation data of multiple millisecond pulsars. After unified time conversion and combined with a pulsar timing model, generate timing residuals using the Earth Time Time Scale (TT(TAI)) calculated based on International Atomic Time (TAI). Record the observation span, TOA epoch of pulse arrival, observation frequency, and measurement uncertainty for each pulsar. In this embodiment, the first... The timing residuals of a pulsar satisfy the following model:
[0030] ,
[0031] in, For the first The timing residual vector of a pulsar; This represents the error term of the rotation model; This represents the common TAI bias term shared across pulsars; and These represent the red noise and DM variation components, respectively. This represents white noise and other unmodeled residuals.
[0032] Step 2) For the interstellar dispersion DM variation component and the pulsar rotation red noise component, set corresponding cutoff frequencies and harmonic quantities respectively. Determine the starting harmonic order of the Fourier expansion based on the observation span of each pulsar, and construct a Fourier design matrix containing continuous harmonics and subject to Tukey window function constraints starting from the starting harmonics. For the DM variation component, perform frequency scaling processing on the Fourier design matrix based on the physical property that the dispersion delay is inversely proportional to the square of the observation frequency.
[0033] This embodiment constructs a Fourier design matrix containing continuous harmonics and subject to Tukey window function constraints, starting from the initial harmonic. The specific implementation steps include:
[0034] (2.1) Let the noise type be , define the first The initial harmonics of the pulsar are:
[0035] ,
[0036] in, Noise type In the The initial harmonic order on a pulsar; Noise type The corresponding cutoff frequency; For the first The observation span of pulsars; Indicates rounding up; Indicates changes in DM. Indicates red noise;
[0037] (2.2) Obtain the first according to the following formula The first pulsar Normalized time of each TOA :
[0038] ;
[0039] In the formula, For the first The first pulsar The observation epochs of each TOA; For the first The earliest TOA epoch of this pulsar; This represents the observation span of the pulsar;
[0040] (2.3) Retaining from the initial harmonic The Fourier basis of a series of consecutive harmonics before frequency scaling is:
[0041] ,
[0042] in, superscript This indicates that DM frequency dependence has not yet been applied; For the first A pulsar in the epoch The Tukey window function is used to reduce spectral leakage at the boundaries of a finite observation interval; Pi is the mathematical constant of a circle.
[0043] The aforementioned red noise is directly designed using a Fourier design matrix. The DM variation is frequency-scaled according to the negative square power of the observed frequency, and its design matrix is as follows:
[0044] ;
[0045] in, For the first Design matrix of DM variation for a pulsar; Design the Fourier transform matrix for DM before frequency scaling; Scaling the diagonal matrix for DM frequency; For the first The first pulsar The observation frequency of each TOA; For a fixed reference frequency; For the first The number of TOA (Time of Arrival) of each pulsar.
[0046] Step 3) Construct fitting weights based on the measurement uncertainty corresponding to TOA, and use the noise design matrix constructed in step (2) to perform mid-frequency noise component estimation; including: firstly, jointly fitting the pulsar rotation model error term and the DM change component to solve the DM change reconstruction component; subtracting the DM change reconstruction component from the timing residual sequence, and then jointly fitting the pulsar rotation model error term and the rotation red noise component to solve the rotation red noise reconstruction component;
[0047] Step 4) Apply weak component protection constraints to the DM change reconstruction component and the rotation red noise reconstruction component obtained in Step 3). For each millisecond pulsar, the product of the median of its TOA measurement uncertainty and the preset weak component protection threshold is used as the judgment threshold. Only when the standard deviation of the corresponding reconstruction component is not less than the judgment threshold, the reconstruction component is subtracted from the original timing residual to obtain the timing residual sequence after intermediate frequency noise suppression. In this embodiment, the weak component protection constraints are applied to the obtained DM change reconstruction component and the rotation red noise reconstruction component. Specifically, the reconstruction component is subtracted from the original timing residual when it satisfies the following formula:
[0048] ,
[0049] in, For weak component protection threshold, For the first The TOA uncertainty vector of a pulsar; std represents the standard deviation, and median represents the median;
[0050] The timing residual sequence after intermediate frequency noise suppression is determined according to the following formula:
[0051] ,
[0052] in, and These are the DM variation and red noise reconstruction components actually used for subtraction after weak component protection judgment.
[0053] Step 5) The low-frequency common international atomic time (TAI) bias process to be estimated is characterized using cubic B-spline basis functions. The basis functions are sampled based on the TOA epochs of each millisecond pulsar to construct a common TAI design matrix corresponding to each pulsar observation epoch. This is achieved according to the following formula:
[0054] ,
[0055] in, node vector The first A cubic B-spline basis function, For the corresponding common TAI deviation coefficient, The number of B-spline coefficients.
[0056] Step 6) Perform reduced QR decomposition on the spin model error term design matrix for each millisecond pulsar to obtain the orthogonal complement space of the column space of the spin model error term design matrix; project the common TAI design matrix corresponding to the current pulsar onto the orthogonal complement space to obtain the common TAI design matrix after applying polynomial orthogonal constraints, so as to eliminate the linear correlation between the low-frequency common TAI bias estimation and the pulsar spin model parameters. This is achieved in the following way in this embodiment:
[0057] The first Projecting the common TAI bias design matrix of the pulsars onto Orthogonal complement of column space:
[0058] ,
[0059] in, for An identity matrix of order 1. For cubic B-spline basis in the th case The value matrix at the TOA epoch of a pulsar. Design a matrix for the common TAI deviation after projection.
[0060] Step 7) Based on the common TAI design matrix obtained in Step 6) after applying polynomial orthogonal constraints, and the fitting weights constructed from the TOA measurement uncertainty, calculate the initial relative contribution of each millisecond pulsar to the joint estimation of the common TAI bias. When the initial relative contribution exceeds a preset contribution adjustment threshold, apply the same reduction factor to both the common TAI design matrix after applying polynomial orthogonal constraints and the timing residual sequence after suppressing intermediate frequency noise corresponding to that pulsar, to adjust the relative contribution of that pulsar in the joint estimation of the common TAI bias. In this embodiment, the initial relative contribution and contribution adjustment factor of each millisecond pulsar to the joint estimation of the common TAI bias are calculated, and scaling is applied to the projected common TAI design matrix and the timing residual after suppressing intermediate frequency noise based on the contribution adjustment factor. Specifically, this includes:
[0061] by normalized value Characterizing the first The initial relative contributions of the pulsars, of which The trace of the matrix is represented by the contribution adjustment factor, which is defined as:
[0062] ,
[0063] in, To adjust the threshold for the preset contribution, For the first The TOA weight matrix corresponding to a millisecond pulsar;
[0064] After contribution adjustment, the intermediate frequency noise suppression timing residual and the projected common TAI design matrix satisfy the following:
[0065] ,
[0066] ;
[0067] In the formula, and These are the timing residuals of mid-frequency noise suppression before and after contribution adjustment; The common TAI design matrix after projection and before contribution adjustment in step (6); The common TAI design matrix after projection and adjustment for contribution.
[0068] Step 8) Construct a multi-pulsar joint model that includes the error terms of each millisecond pulsar's rotation model and the common TAI bias term shared by all millisecond pulsars; stack the intermediate frequency noise suppression timing residuals after contribution weight adjustment across pulsars, and apply first-order and second-order difference smoothing constraints to the B-spline coefficients characterizing the common TAI bias; based on weighted least squares solution, jointly estimate the coefficients of each millisecond pulsar's rotation model error term and the B-spline coefficients corresponding to the common TAI bias.
[0069] In this embodiment, the multi-pulsar joint model described above has a matrix representation including a multi-pulsar joint parameter vector and a joint design matrix; the multi-pulsar joint parameter vector is as follows:
[0070] ,
[0071] The corresponding joint design matrix is:
[0072] ,
[0073] in, For the first The coefficient vector of the rotation model error term corresponding to each pulsar. This is the common TAI bias B-spline coefficient vector shared by all pulsars. The number of pulsars participating in the joint estimation.
[0074] The above-mentioned first-order and second-order difference smoothing constraints are applied to the B-spline coefficients characterizing the common TAI bias, and the overall smoothing weight matrix is constructed as follows:
[0075] ,
[0076] in, and These are first-order and second-order difference operators, respectively. For overall smooth strength, and These are the relative weights corresponding to the first-order and second-order difference constraints, respectively;
[0077] The smoothness constraint is embedded in the joint regularization matrix. Afterwards, the joint parameters satisfy:
[0078] ,
[0079] in, The TOA weight matrix is the joint pulse arrival time. The stacking vector of adjusted timing residuals contributed to each pulsar; obtained from the solution Extract It is used to reconstruct the low-frequency common TAI deviation process.
[0080] Step 9) Using the common TAI deviation B-spline coefficients, construct the orthogonalized B-spline design matrix under the same low-order polynomial orthogonal constraints as in Step 6) on a unified time grid; reconstruct the common TAI deviation sequence based on this matrix and B-spline coefficients, perform sign transformation, and output the deviation sequence relative to TT(TAI) when synthesizing pulsars.
[0081] Step 10) Align the deviation sequence output in Step 9) with the BIPM reference deviation sequence TT(BIPM)-TT(TAI) to the same evaluation epoch and conduct a post-hoc comparison. Use three evaluation indicators, namely Pearson correlation coefficient, amplitude ratio, and root mean square error, to evaluate the extraction results.
[0082] Example 2: Refer to Figure 1-4 This embodiment is based on long-term multi-frequency timing observation data from the Parkes Pulsar Timing Array Third Data Release (PPTADR3). Common TAI bias is extracted from the multi-pulsar timing residuals to obtain a long-term bias sequence of the comprehensive pulsar time relative to the Earth time scale TT (TAI) calculated based on International Atomic Time (TAI). For example... Figure 1 As shown, the TAI deviation extraction process in this embodiment includes two stages: mid-frequency noise suppression and joint estimation of low-frequency common TAI deviation. The first stage estimates and suppresses mid-frequency DM variations and red noise that are relatively separable from the TAI deviation, while retaining low-frequency common TAI deviation information. The second stage uses the noise-suppressed timing residuals as input to jointly estimate the error terms of each pulsar rotation model and the shared common TAI deviation term. Specifically, it includes the following steps:
[0083] Step 1: Acquire long-term, multi-frequency timing observation data from multiple millisecond pulsars, perform a unified time conversion on the arrival times of each pulsar, and combine this with a pulsar timing model to form a pulsar timing residual with TT (TAI) as the ground time reference:
[0084] ,
[0085] In the formula, For the first The timing residual vector of a pulsar; This is the design matrix for the pulsar rotation model parameter errors in the timing residuals, including constant terms, first-order terms, and second-order terms. For the corresponding coefficient vector; Design a matrix for common TAI deviation. Its corresponding coefficient vector; and These are the red noise Fourier design matrix and its coefficient vector, respectively; and These are the DM variation design matrix and its coefficient vector, respectively; This represents white noise and other unmodeled residuals.
[0086] For the The first pulsar Let each TOA (Time of Observation) have an observation epoch, observation frequency, and measurement uncertainty denoted as _____. , and , No. The observation span of a pulsar is denoted as .
[0087] Step 2: Construct the Fourier design matrix for DM variation and red noise, for the noise type According to the first step obtained in step 1 Pulsar observation span and the cutoff frequency of the corresponding noise component Determine the initial harmonic order of the Fourier expansion:
[0088] ,
[0089] In the formula, Noise type In the The initial harmonic order on a pulsar; Noise type The corresponding cutoff frequency; For the first The observation span of pulsars; Indicates rounding up; Indicates changes in DM. This indicates red noise.
[0090] Construct a Fourier design matrix over the observation range of each pulsar. The remaining chord and sine sequences are respectively
[0091]
[0092] in, superscript This indicates that DM frequency dependence has not yet been applied; For the first A pulsar in the epoch The Tukey window function is used to reduce spectral leakage at the boundaries of a finite observation interval; Pi is the mathematical constant of a circle. For the first Normalized TOA epochs of pulsars:
[0093] ,
[0094] In the formula, For the first The first pulsar The observation epochs of each TOA; For the first The earliest TOA epoch of this pulsar; This represents the observation span of the pulsar.
[0095] Red noise is a non-dispersive time-correlated process, therefore no observation frequency scaling is applied, and the above Fourier design matrix is directly used:
[0096] ,
[0097] In the formula, For the first Red noise Fourier design matrix for pulsars. This is the Fourier design matrix constructed according to the above method.
[0098] The dispersion delay caused by the DM variation is inversely proportional to the square of the observation frequency; therefore, frequency scaling is applied to the fundamental Fourier design matrix of the DM variation.
[0099] ,
[0100] In the formula, For the first Design matrix of DM variation for a pulsar; Design the Fourier transform matrix for DM before frequency scaling; Scaling the diagonal matrix for DM frequency; For the first The first pulsar The observation frequency of each TOA; For a fixed reference frequency; For the first The number of TOA (Time of Arrival) of each pulsar; This indicates that the elements within the parentheses are placed sequentially on the main diagonal of the matrix. Factor Used to characterize the dispersion delay of DM Frequency dependent.
[0101] Step 3: Estimate DM variation and red noise sequentially. Using the noise design matrix constructed in Step 2 and the weighted least squares method, first estimate the DM variation, then estimate the red noise on the timing residual after subtracting the DM reconstruction component; the fitting results are as follows. Figure 2 As shown.
[0102] For noise type The design matrix of the error term of the rotation model defined in step 1 is then used. With the corresponding noise design matrix Combining and constructing a joint fitting design matrix :
[0103] ,
[0104] when hour, Design matrix for DM changes ;when hour, Design a matrix for red noise Noise type The corresponding fitted input residual is denoted as The joint parameters are estimated using the weighted least squares method:
[0105] ,
[0106] In the formula, Noise type The corresponding joint fitting parameter estimation vector includes the coefficients corresponding to the error term of the rotation model and the noise Fourier coefficients; the estimated noise Fourier coefficients are denoted as... The superscript T indicates matrix transpose. For the first TOA weight matrix of pulsars:
[0107] ,
[0108] In the formula, For the first The first pulsar Measurement uncertainty of one TOA; For the first The number of TOA (Time of Arrival) of each pulsar; This represents the operation of constructing a diagonal matrix; This represents the inverse variance weight corresponding to TOA. The corresponding noise components are reconstructed based on the obtained noise Fourier coefficients. :
[0109] ,
[0110] Through the above processing, step 3 obtains the first... The DM variation and red noise reconstruction component of the pulsar; whether to subtract them is determined by the weak component protection criterion in step 4.
[0111] Step 4: Implement weak component protection and obtain noise suppression residuals. First, calculate the... Median of TOA measurement uncertainty for pulsars:
[0112] ,
[0113] In the formula, For the first The median of the TOA measurement uncertainty for a single pulsar; For the first The first pulsar Measurement uncertainty of one TOA; This represents the number of TOA (Time of Arrival) for this pulsar; This indicates taking the median. For noise types... The weak component protection method is as follows:
[0114] ,
[0115] In the formula, This refers to the noise component actually used for subtraction after weak component protection judgment. The noise type obtained in step 3 The fitted noise component; This indicates the calculation of the standard deviation; This is the threshold parameter for weak component protection.
[0116] After completing the weak component protection judgment, the two types of fitted noise components that are actually retained are subtracted from the original timing residual to obtain the noise-suppressed timing residual:
[0117] ,
[0118] In the formula, For the first Timing residuals of pulsars after intermediate frequency noise suppression; This is the original timing residual; and These are the DM variation component and the red noise component actually used for subtraction after weak component protection.
[0119] Step 5: Represent the low-frequency common TAI bias process to be jointly estimated after noise suppression in Step 4 using cubic B-splines:
[0120] ,
[0121] In the formula, For a moment The low-frequency common TAI deviation component at the location; The number of B-spline basis functions and their corresponding coefficients; Index of B-spline basis functions; For the first One common TAI B spline coefficient; For node sequence Upper A cubic B-spline basis function; It is a sequence of B-spline nodes.
[0122] Based on the first obtained in step 1 Each pulsar's TOA epoch Calculate the corresponding B-spline basis function values and construct the first... Common TAI design matrix for pulsars Its matrix elements are represented as:
[0123] ,
[0124] In the formula, Design matrix for public TAI No. line, number Column elements; For the first TOA index of a pulsar; Index of B-spline basis functions; For the first The first pulsar The observation epochs of each TOA.
[0125] Therefore, the first The common TAI deviation component at each TOA epoch of the pulsar can be obtained from It means that among them This is the B-spline coefficient vector shared by all pulsars.
[0126] Step 6: Apply polynomial constraints to the common TAI design matrix. The first polynomial constraint defined in Step 1... Design matrix of error terms for pulsar rotation model This is used to represent the constant, linear, and quadratic terms in its timing residuals that correspond to the error of the rotation model. Perform reduced QR decomposition:
[0127] ,
[0128] and satisfy In the formula, for A set of orthogonal normal bases for column spaces; The upper triangular matrix is obtained by reducing the QR decomposition; the superscript T indicates matrix transpose; is the identity matrix of the corresponding dimension.
[0129] use The public TAI design matrix constructed in step 5 Project to Orthogonal complement of column space:
[0130]
[0131] In the formula, The common TAI design matrix after applying polynomial constraints; for An identity matrix of order 1; for Orthogonal projection matrix on column space; It is the projection matrix on its orthogonal complement space.
[0132] Through the above processing, the public TAI design matrix is projected onto the observation interval of each pulsar. Orthogonal complement of column space.
[0133] Step 7: Calculate and adjust the initial relative contributions of each pulsar, and use the projected common TAI design matrix obtained in Step 6. and the TOA weight matrix constructed in step 3 Calculate the first The initial relative contributions of each pulsar to the joint estimate of the common TAI:
[0134] ,
[0135] In the formula, For the first The initial relative contribution of each pulsar; Represents the trace of a matrix; Sum index for pulsars; The number of pulsars participating in the joint estimation; and The first The common TAI design matrix and TOA weight matrix after projection of pulsars.
[0136] Based on the initial relative contribution Define contribution adjustment factor:
[0137] ,
[0138] In the formula, For the first The contribution adjustment factor of each pulsar; Adjust the threshold based on contribution; This indicates taking the smaller of two numbers. When hour, The pulsar does not perform contribution reduction; when hour, This reduces the impact of the pulsar on subsequent joint estimates of the public TAI. Used to determine the triggering conditions for contribution adjustments; when an adjustment is triggered, the reduction factor is adjusted by... Determined together with the initial relative contribution.
[0139] Using contribution adjustment factor The projected common TAI design matrix and the noise suppression timing residual obtained in step 4 are uniformly scaled:
[0140] ,
[0141] ,
[0142] In the formula, The adjusted public TAI design matrix is contributed; To contribute the adjusted noise suppression timing residual; superscript This indicates the amount after contribution adjustment.
[0143] Step 8: Jointly estimate the common TAI bias coefficient across pulsars. Stack the adjusted timing residuals of each pulsar contribution obtained in Step 7 to form a joint residual vector. The coefficients of the rotation model error terms of each pulsar to be estimated are combined with the B-spline coefficients of the common TAI to form a joint parameter vector:
[0144] ,
[0145] In the formula, For joint parameter vectors; For the first The coefficient vector of the rotation model error term of the pulsar to be estimated in this stage; This is a common TAIB spline coefficient vector shared by all pulsars.
[0146] Design matrices for the rotation model error terms of each pulsar. The contribution-adjusted common TAI design matrix obtained in step 7 Direct combination to construct a multi-pulsar joint design matrix:
[0147] ,
[0148] In the formula, Matrix for joint design of multiple pulsars; Each is a zero matrix of the corresponding dimension; These correspond to the error term coefficients of the independent rotation model for each pulsar, and are in the last column. B-spline coefficients that correspond to the same set of common TAIs .
[0149] First-order and second-order difference constraints are applied to the B-spline coefficients of the common TAI to control the smoothness of the common TAI deviation curve:
[0150] ,
[0151] In the formula, The smoothing constraint matrix for the common TAIB spline coefficients; and These are the first-order difference matrix and the second-order difference matrix, respectively; For smoothing weights; coefficients These are the weights of the second-order difference terms. Placed in joint parameters The corresponding positions are set to zero, and the rest are set to zero to form a joint regularization matrix. .
[0152] The TOA weight matrices of each pulsar are combined into a joint weight matrix:
[0153] ,
[0154] In the formula, This is the joint TOA weight matrix; This means that the input matrices are arranged sequentially on the main diagonal block of the joint matrix, and the remaining positions are set to zero.
[0155] The joint parameters are estimated using a weighted least squares method with smoothing constraints:
[0156] ,
[0157] In the formula, This is the joint parameter estimation vector; superscript Indicates matrix transpose; superscript This represents finding the inverse of a matrix. By extracting the corresponding part of the common TAI parameters, the estimated vector of common TAI spline coefficients is obtained. This is used in step 9 to reconstruct the common TAI bias sequence.
[0158] Step 9: Reconstruct and output the common TAI bias sequence, using the common TAI B spline coefficient estimation vector obtained in Step 8. The common TAI bias sequence is reconstructed on a unified output time grid. Assume the output time grid contains... Each epoch is denoted as . Based on these epochs, construct a polynomial design matrix consisting of constant terms, linear terms, and quadratic terms. .
[0159] right Perform the same reduced QR decomposition as in step 6 to obtain the orthogonal normal basis of its column space. The B-spline design matrix is calculated on the output time grid. Following the same method as step 6, Project to Orthogonal complement of column space:
[0160] ,
[0161] In the formula, The common TAI design matrix is output after applying polynomial constraints on the time grid; for An identity matrix of order 1; for The orthogonal projection matrix in column space. Through this process, the common TAI design matrix on the output time grid adopts the same polynomial constraints as in step 6, ensuring that the final reconstruction is consistent with the joint estimation parameterization in step 8.
[0162] The final bias sequence is reconstructed using the common TAIB spline coefficient estimation vector obtained in step 8:
[0163]
[0164] In the formula, When the final output is a composite pulsar, relative to deviation sequence; This is the common TAIB spline coefficient estimation vector obtained in step 8; the negative sign is used to unify the definition direction of the extraction result and the reference deviation; the extracted deviation sequence is as follows: Figure 3 As shown.
[0165] Step 10: Conduct a post-event evaluation.
[0166] The result obtained in step 9 Deviation sequence and independent reference sequence Compare the waveform consistency, amplitude consistency, and absolute difference of the two at the same evaluation epoch.
[0167] Assume there is a total The first evaluation period, the first The extraction bias and reference bias values for each evaluation epoch are denoted as follows: and The arithmetic mean of the two sequences is denoted as follows: and .
[0168] The Pearson correlation coefficient is used to evaluate the waveform consistency between the two sequences.
[0169] ,
[0170] In the formula, To extract the Pearson correlation coefficient between the sequence and the reference sequence; To evaluate the number of epochs; To evaluate the epoch index; and The first Extracted and reference values for each evaluation epoch; and These are the arithmetic means of the two sequences, respectively.
[0171] The amplitude ratio is used to evaluate the amplitude consistency between two sequences:
[0172] ,
[0173] In the formula, The magnitude ratio of the extracted sequence to the reference sequence is given by the numerator, which is the root mean square magnitude of the extracted sequence after removing the mean, and the denominator is the root mean square magnitude of the reference sequence after removing the mean.
[0174] The root mean square error is used to evaluate the absolute difference between two sequences:
[0175] ,
[0176] In the formula, This is the root mean square error; For the first The difference between the extracted value and the reference value of each evaluation epoch.
[0177] The TT(BIPM2021)-TT(TAI) reference sequence was used for post-evaluation in step 10. The comparison results are as follows: Figure 4 As shown, the extraction results of this invention have higher consistency with the reference sequence and lower RMSE, achieving an accuracy improvement of one order of magnitude compared to traditional methods. The above examples and results analysis demonstrate the correctness and effectiveness of the method proposed in this invention.
[0178] The parts of this invention not described in detail are common knowledge to those skilled in the art.
[0179] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Obviously, those skilled in the art, after understanding the content and principle of the present invention, may make various modifications and changes in form and detail without departing from the principle and structure of the present invention. However, these modifications and changes based on the concept of the present invention are still within the scope of protection of the claims of the present invention.
Claims
1. A method for extracting international atomic time deviation based on pulsar timing noise frequency division suppression, characterized in that, Includes the following steps: (1) Obtain long-term multi-frequency timing observation data of multiple millisecond pulsars, generate timing residuals with reference to the Earth time scale TT (TAI) based on the conversion of International Atomic Time (TAI), and record the observation span, pulse arrival time TOA epoch, observation frequency and measurement uncertainty of each pulsar. (2) For the interstellar dispersion DM variation component and the pulsar rotation red noise component, the corresponding cutoff frequency and harmonic number are set respectively. The starting harmonic order of the Fourier expansion is determined according to the observation span of each pulsar. A Fourier design matrix containing continuous harmonics and subject to Tukey window function constraints is constructed starting from the starting harmonic. For the DM variation component, the Fourier design matrix is frequency scaled according to the physical property that the dispersion delay is inversely proportional to the square of the observation frequency. (3) Based on the measurement uncertainty corresponding to TOA, construct fitting weights and use the noise design matrix obtained in step (2) to perform mid-frequency noise component estimation; including: firstly, jointly fitting the pulsar rotation model error term and the DM change component to solve the DM change reconstruction component; subtracting the DM change reconstruction component from the timing residual sequence, and then jointly fitting the pulsar rotation model error term and the rotation red noise component to solve the rotation red noise reconstruction component; (4) Perform weak component protection constraints on the DM change reconstruction component and the spin red noise reconstruction component obtained in step (3); for each millisecond pulsar, use the product of the median of its TOA measurement uncertainty and the preset weak component protection threshold as the judgment threshold. Only when the standard deviation of the corresponding reconstruction component is not less than the judgment threshold, the reconstruction component is deducted from the original timing residual to obtain the timing residual sequence after intermediate frequency noise suppression. (5) The low-frequency common international atomic time (TAI) bias process to be estimated is characterized by cubic B-spline basis functions. The basis functions are sampled according to the TOA epoch of each millisecond pulsar to construct a common TAI design matrix corresponding to the observation epoch of each pulsar. (6) Perform reduced QR decomposition on the spin model error term design matrix of each millisecond pulsar to obtain the orthogonal complement space of the column space of the spin model error term design matrix; project the common TAI design matrix corresponding to the current pulsar onto the orthogonal complement space to obtain the common TAI design matrix after applying polynomial orthogonal constraints, so as to eliminate the linear correlation between the low-frequency common TAI bias estimation and the pulsar spin model parameters; (7) Based on the common TAI design matrix after applying polynomial orthogonal constraints obtained in step (6), and the fitting weight constructed from the TOA measurement uncertainty, calculate the initial relative contribution of each millisecond pulsar to the joint estimation of common TAI bias; when the initial relative contribution exceeds the preset contribution adjustment threshold, apply the same reduction factor to the common TAI design matrix after applying polynomial orthogonal constraints and the timing residual sequence after intermediate frequency noise suppression corresponding to the pulsar, so as to adjust the relative contribution of the pulsar in the joint estimation of common TAI bias; (8) Construct a multi-pulse star joint model that includes the error terms of each millisecond pulse star's rotation model and the common TAI deviation term shared by all millisecond pulse stars; stack the intermediate frequency noise suppression timing residuals after the contribution weight adjustment across pulse stars, and apply first-order and second-order difference smoothing constraints to the B-spline coefficients that characterize the common TAI deviation; based on the weighted least squares solution, jointly estimate the coefficients of each millisecond pulse star's rotation model error term and the B-spline coefficients corresponding to the common TAI deviation. (9) Using the common TAI deviation B-spline coefficients, construct the orthogonalized B-spline design matrix under the same low-order polynomial orthogonal constraint as in step (6) on a unified time grid; reconstruct the common TAI deviation sequence based on the matrix and B-spline coefficients, and perform sign conversion to output the deviation sequence relative to TT(TAI) when synthesizing pulsars. (10) Align the deviation sequence output in step (9) with the BIPM reference deviation sequence TT(BIPM)-TT(TAI) to the same evaluation epoch and conduct a post-hoc comparison. The extraction results are evaluated using three evaluation indicators: Pearson correlation coefficient, amplitude ratio, and root mean square error.
2. The method according to claim 1, characterized in that: Step (1) involves acquiring long-term, multi-frequency timing observation data from multiple millisecond pulsars, performing a unified time conversion, and combining this data with a pulsar timing model to generate a timing residual referenced to the Earth time scale TT (TAI) calculated based on International Atomic Time (TAI). The first... The timing residuals of a pulsar satisfy the following model: , in, For the first The timing residual vector of a pulsar; This represents the error term of the rotation model; This represents the common TAI bias term shared across pulsars; and These represent the red noise and DM variation components, respectively. This represents white noise and other unmodeled residuals.
3. The method according to claim 1, characterized in that: Step (2) involves constructing a Fourier design matrix that includes continuous harmonics and applies Tukey window function constraints, starting from the initial harmonic. The steps include: (2.1) Let the noise type be , define the first The initial harmonics of the pulsar are: , in, Noise type In the The initial harmonic order on a pulsar; Noise type The corresponding cutoff frequency; For the first The observation span of pulsars; Indicates rounding up; Indicates changes in DM. Indicates red noise; (2.2) Obtain the first according to the following formula The first pulsar Normalized time of each TOA : ; In the formula, For the first The first pulsar The observation epochs of each TOA; For the first The earliest TOA epoch of this pulsar; This represents the observation span of the pulsar; (2.3) Retaining from the initial harmonic The Fourier basis of a series of consecutive harmonics before frequency scaling is: , in, superscript This indicates that DM frequency dependence has not yet been applied; For the first A pulsar in the epoch The Tukey window function is used to reduce spectral leakage at the boundaries of a finite observation interval; Pi is the mathematical constant of a circle.
4. The method according to claim 3, characterized in that: The red noise is directly designed using a Fourier design matrix, and the DM variation is frequency-scaled according to the negative square power of the observed frequency. The design matrix is as follows: ; in, For the first Design matrix of DM variation for a pulsar; Design the Fourier transform matrix for DM before frequency scaling; Scaling the diagonal matrix for DM frequency; For the first The first pulsar The observation frequency of each TOA; For a fixed reference frequency; For the first The number of TOA (Time of Arrival) of each pulsar.
5. The method according to claim 1, characterized in that: Step (4) involves applying weak component protection constraints to the DM change reconstruction component and the self-rotation red noise reconstruction component obtained in step (3). Specifically, this is achieved by subtracting the reconstruction component from the original timing residual when the reconstruction component satisfies the following formula: , in, For weak component protection threshold, For the first The TOA uncertainty vector of a pulsar; std represents the standard deviation, and median represents the median; The timing residual sequence after intermediate frequency noise suppression is determined according to the following formula: , in, and These are the DM variation and red noise reconstruction components actually used for subtraction after weak component protection judgment.
6. The method according to claim 1, characterized in that: Step (5) describes the use of cubic B-spline basis functions to characterize the low-frequency common international atomic time (TAI) bias process to be estimated. The low-frequency common TAI bias process is achieved according to the following formula: , in, node vector The first A cubic B-spline basis function, For the corresponding common TAI deviation coefficient, The number of B-spline coefficients.
7. The method according to claim 1, characterized in that: Step (6) involves projecting the common TAI design matrix corresponding to the current pulsar onto the orthogonal complement space to obtain the common TAI design matrix after applying polynomial orthogonal constraints. This is achieved in the following way: The first Projecting the common TAI bias design matrix of the pulsars onto Orthogonal complement of column space: , in, for An identity matrix of order 1. For cubic B-spline basis in the th case The value matrix at the TOA epoch of a pulsar. Design a matrix for the common TAI deviation after projection.
8. The method according to claim 7, characterized in that: Step (7) calculates the initial relative contribution and contribution adjustment factor of each millisecond pulsar to the joint estimation of the common TAI bias, and applies scaling to the projected common TAI design matrix and the timing residual after intermediate frequency noise suppression based on the contribution adjustment factor. Specifically, this includes: by normalized value Characterizing the first The initial relative contributions of the pulsars, of which The trace of the matrix is represented by the contribution adjustment factor, which is defined as: , in, To adjust the threshold for the preset contribution, For the first The TOA weight matrix corresponding to a millisecond pulsar; After contribution adjustment, the intermediate frequency noise suppression timing residual and the projected common TAI design matrix satisfy the following: , ; In the formula, and These are the timing residuals of mid-frequency noise suppression before and after contribution adjustment; The common TAI design matrix after projection and before contribution adjustment in step (6); The common TAI design matrix after projection and contribution adjustment.
9. The method according to claim 8, characterized in that: The multi-pulsar joint model described in step (8) has a matrix representation including a multi-pulsar joint parameter vector and a joint design matrix; the multi-pulsar joint parameter vector is as follows: , The corresponding joint design matrix is: , in, For the first The coefficient vector of the rotation model error term corresponding to each pulsar. This refers to the common TAI bias B-spline coefficient vector shared by all pulsars. The number of pulsars participating in the joint estimation.
10. The method according to claim 9, characterized in that: In step (8), first-order and second-order difference smoothing constraints are applied to the B-spline coefficients characterizing the common TAI bias to construct the overall smoothing weight matrix: , in, and These are first-order and second-order difference operators, respectively. For overall smoothness strength, and These are the relative weights corresponding to the first-order and second-order difference constraints, respectively. The smoothness constraint is embedded in the joint regularization matrix. Afterwards, the joint parameters satisfy: , in, The TOA weight matrix is the joint pulse arrival time. The stacking vector of adjusted timing residuals contributed to each pulsar; obtained from the solution Extract It is used to reconstruct the low-frequency common TAI deviation process.
