CN112394376B - A method for parallel processing of large-scale GNSS network observation data with non-differential whole network - Google Patents

Abstract

The invention relates to a non-difference network-wide parallel processing method for large-scale GNSS network observation data, specifically based on the underlying multi-core parallel computing technology, to realize non-difference network-wide parallel computing of large-scale GNSS network observation data. First, select a certain number of measuring stations from the large-scale GNSS network as the core stations, estimate the satellite orbit and satellite clock error in parallel, perform parallel non-difference floating-point solution calculations between stations, and estimate the wide-lane and narrow-lane fractional phase deviations at the satellite end in parallel ;Secondly, carry out the parallel calculation of non-difference floating-point solutions between stations for other stations in the GNSS network, use the wide-lane and narrow-lane decimal phase deviations estimated by the core station to fix the non-difference ambiguities of all stations in parallel, and generate the GNSS network in parallel The carrier distance of all stations in the network; finally, use the pseudo-range observation value and carrier distance to carry out the unified calculation of the whole network in the non-difference mode. The parameters to be estimated include the coordinates of the station, satellite orbit, receiver and satellite clock difference, double difference Ambiguity, tropospheric delay, fractional phase bias and Earth rotation parameters, etc.

Description

A method for parallel processing of large-scale GNSS network observation data with non-differential whole network

technical field

The invention relates to a large-scale GNSS network observation data non-difference whole network parallel processing method, and belongs to the technical field of large-scale GNSS network observation data processing.

Background technique

As an important ground infrastructure of the country, many countries and regions have established a GNSS (Global Navigation Satellite System) reference station network consisting of hundreds or even thousands of continuously operating reference stations. Large-scale global GNSS network data provides abundant computing resources, but it also brings great challenges in data processing. The time-consuming data processing and calculation increases geometrically with the expansion of the scale of the GNSS network. The data processing of the global GNSS large network faces the problem of low computational efficiency or even inability to solve the problem. There is an urgent need for corresponding high-efficiency data processing models and methods as support. , to give full play to the advantages of the large network and improve the accuracy, reliability and real-time performance of large network services. Seeking efficient and fast processing of large-scale GNSS network data has become one of the current hot topics and has received more and more attention and attention.

Research institutions and researchers at home and abroad have paid extensive attention and research on improving models and algorithms, and applying parallel computing technology. "Parallel resolution of large-scale GNSS network un-difference ambiguity", "Anew data processing strategy for huge GNSS global networks" of "Journal of Geodesy" and "An enhanced strategy for GNSS data processing of massive networks", domestic "Journal of Surveying and Mapping" "Research on Parallel Processing of GNSS Data in a Multi-core Environment", "Parallel Fast Calculation Method of GNSS Large-scale Double-difference Model" and "Fast and Precise Calculation of PPP Network Solving UPD Ambiguity Fixed Large-scale CORS Stations without Base Station Difference", " "A Fast and Efficient Processing Strategy for GNSS Large Network Data" in Geodesy and Geodynamics.

In the large-scale GNSS network non-difference whole network calculation method, the existing solution adopts the serial calculation method to synchronously process the data of all reference stations in the large-scale GNSS network, and serially estimate satellite orbits and clock errors sequentially. Wide-lane and narrow-lane fractional phase deviation, serial fixation of non-difference ambiguity, serial generation of carrier distance and serial solution of the whole network. In the parallel calculation of large-scale GNSS network observation data, existing research has focused on baseline vector network adjustment, double-difference baseline calculation, single-station non-difference floating-point solution and fixed solution calculation, etc.

Contents of the invention

The purpose of the present invention is to provide a large-scale GNSS network observation data non-differential network parallel processing method to solve the problem that the current GNSS network scale continues to expand and the time-consuming data processing and solution increases exponentially.

To achieve the above object, the solutions of the present invention include:

A kind of large-scale GNSS network observation data non-difference whole network parallel processing method of the present invention comprises the following steps:

1) Select a set number of measuring stations as core stations, and calculate satellite orbit, satellite clock error and earth rotation parameters in parallel according to the observation data of said core stations;

2) Estimate the wide-lane fractional phase deviation and the narrow-lane fractional phase deviation at the satellite end by using the satellite orbit, satellite clock error and earth rotation parameters;

3) Construct the non-difference ambiguity for each core station, and use the rounded wide-lane fractional phase deviation and narrow-lane fractional phase deviation to fix the ionospheric-free combination ambiguity;

For stations other than the core station, use the satellite orbit, satellite clock error, earth rotation parameters, and the wide-lane fractional phase deviation and narrow-lane fractional phase deviation at the satellite end to perform inter-station parallel non-difference fixed solution calculations, for each other The station constructs the undifferenced ambiguity, and fixes the combined ambiguity without the ionosphere;

4) According to the carrier phase observation value of each station and the fixed ionosphere-free combination ambiguity, generate the carrier distance observation value of the corresponding station in parallel;

5) Combining the pseudo-range observations and carrier distance observations of each station, constructing the normal equation in parallel, and inverting the normal equation in parallel, and performing parallel calculations for the entire network based on the non-difference mode.

The present invention provides a new large-scale GNSS network observation data non-difference whole network multi-core parallel processing method, which realizes the parallel execution of many links in the large-scale GNSS network observation data non-difference whole network calculation, and improves the computational efficiency of the multi-core platform The utilization rate improves the efficiency of the whole network solution.

Further, it is characterized in that the core stations are not less than 100 measuring stations evenly distributed around the world.

Further, in step 1), the satellite orbit and earth rotation parameters are estimated by Kalman filtering in the non-difference mode, and the non-difference observation equation is constructed with a single core station as the parallel granularity, and the double difference ambiguity is constructed and the ambiguity Fixed, constraining the undifferenced observation equation.

Further, in step 2), the wide-lane fractional phase deviation and the narrow-lane fractional phase deviation at the satellite end are estimated in parallel by performing inter-station parallel non-difference floating-point solution calculations on the core station.

Further, in step 4), the carrier phase observation value is corrected to obtain the carrier distance observation value by using the fixed ionosphere-free combination ambiguity and the antenna phase winding effect.

Further, in step 5), the pseudo-range hardware delay of the satellite terminal is deducted from the pseudo-range observation value.

Further, in step 5), the method for parallel calculation of the entire network based on the non-difference mode is: use the pseudorange observation value and the carrier distance observation value as observations; Equation; use the parameter elimination algorithm based on recursive least squares to solve the normal equation; use matrix block inversion and Cholesky parallel decomposition method to invert the normal equation.

At present, the unified calculation of large-scale GNSS network observation data is more and more demanding on real-time and calculation efficiency. For the timeliness of large-scale GNSS network observation data non-difference network calculation, the multi-core parallel computing technology - parallel task library (taskparallel library, TPL) is used to design the large-scale GNSS network observation data non-difference network calculation method under the multi-core platform to realize the core Parallel calculation of satellite orbit, satellite clock error and earth rotation parameters by stations; parallel estimation of wide-lane fractional phase deviation and narrow-lane fractional phase deviation by core stations; parallel fixation of non-difference ambiguity of the whole network, parallel generation of carrier distance of the whole network and non-difference mode Under the multi-core parallel execution of the entire network solution.

The method of the present invention is simple and easy to operate, improves the utilization rate of the multi-core platform, shortens the time for non-difference whole network calculation of large-scale GNSS network observation data, and improves the efficiency of non-difference whole network calculation of large-scale GNSS network observation data .

Furthermore, the parameters to be estimated obtained by the parallel calculation of the entire network include: station coordinates, satellite orbits, double-difference ambiguities, tropospheric delays, receiver and satellite clock errors, fractional phase deviations, and earth rotation parameters.

Further, for the unfixed ionosphere-free combination ambiguity of each station in step 3), construct the ionosphere-free combination ambiguity of station-satellite double difference, and fix the ionosphere-free combination ambiguity.

For the ambiguities that cannot be fixed by the non-difference mode at each station, further try to use the double-difference mode to fix the ambiguity, so as to improve the success rate of ambiguity fixation, so that more observation data can be used in the calculation of the whole network, and then the estimated Accuracy and reliability of parameter solution.

Description of drawings

Fig. 1 is a flow chart of the non-differential network-wide parallel calculation of large-scale GNSS network observation data under the multi-core platform of the present invention.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

As shown in Figure 1, to the observation data of large-scale GNSS network reference station (measuring station), the method of the present invention discloses the flow process of large-scale GNSS network observation data non-difference whole network parallel processing under the multi-core platform, is divided into the following steps:

Step 1: Select a certain number of globally uniformly distributed measuring stations in the large-scale GNSS network as core stations, and calculate satellite orbits, satellite clock errors and earth rotation parameters in parallel based on task parallelism and data parallelism;

Step 2: Using the satellite orbit, satellite clock error and earth rotation parameters, perform parallel non-difference floating-point solution calculations for the core station, and estimate the wide-lane fractional phase deviation and narrow-lane fractional phase deviation at the satellite side in parallel;

Step 3: Perform parallel calculation of non-differenced floating-point solutions for other stations in the large-scale GNSS network, and use the wide-lane and narrow-lane fractional phase deviations estimated by the core station to perform non-differenced ambiguity calculations for each station fixed, to obtain the fixed non-difference ionosphere-free combined ambiguity;

Step 4: For the carrier phase observations of all stations in the large-scale GNSS network, based on the inter-station and inter-satellite two-level parallel mechanism, the corresponding fixed non-difference ionospheric-free combination ambiguities are deducted, and converted into ambiguity-free The carrier distance observation value, for the pseudo-range observation value of all stations, deducts the satellite-side pseudo-range hardware delay;

Step 5: Combine the pseudorange and carrier distance obtained in step 4, construct the normal equation in parallel, and invert the normal equation in parallel, and perform parallel calculation of the entire network based on the non-difference mode.

In step 1, the number of core stations evenly distributed globally is not less than 100, and the strategy for estimating satellite orbit, satellite clock error, and earth rotation parameters is the Kalman filter in the non-difference mode. The difference observation equation, by constructing the double-difference ambiguity and fixing it, constrains the non-difference equation.

In step 3), the method of parallel calculation of non-difference fixed solution is:

For the core station, there is no need to calculate the non-difference floating-point solution, and directly use the wide-lane fractional phase deviation and narrow-lane fractional phase deviation obtained in step 2), and obtain the wide-lane ambiguity and narrow-lane ambiguity after rounding, and then use The fixed wide-lane ambiguity and narrow-lane ambiguity are calculated to obtain a fixed non-difference ionosphere-free combined ambiguity;

For other stations in the GNSS network, use satellite orbit, satellite clock error, earth rotation parameters, wide-lane fractional phase deviation and narrow-lane fractional phase deviation at the satellite end to perform inter-station parallel non-difference fixed solution calculations.

In step 4), the carrier distance is generated using the fixed non-difference ionosphere-free combination ambiguity in step 3), and the antenna phase winding effect calculated by the model, and the ionosphere-free combination carrier phase observation is corrected to obtain Carrier distance

为载波距离，L为原始无电离层组合载波相位观测值，λ为无电离层组合波长，N为固定了的非差无电离层组合模糊度，ξ为天线相位缠绕效应。In the formula,

is the carrier distance, L is the original ionospheric combination carrier phase observation value, λ is the ionospheric combination wavelength, N is the fixed non-difference ionospheric combination ambiguity, ξ is the antenna phase winding effect.

For the ionosphere-free combination ambiguity that has not been fixed by non-difference fixed at each station in step 3), further construct the ionosphere-free combination ambiguity of station-satellite double-difference, try to fix the double-difference ambiguity, and effectively improve the ambiguity of the fixed ambiguity. The success rate increases the effective carrier distance observation value that can be used for the parallel calculation of the whole network in step 5), thereby improving the accuracy and reliability of the calculation of the parameters to be estimated.

In step 5), the parallel calculation method for the entire network in the non-difference mode is:

First, the obtained pseudorange and carrier distance are used as observations. When the normal equation is constructed, the epoch is used as the granularity, and the normal equation of each epoch is constructed in parallel based on data parallelism. The parameter elimination algorithm based on recursive least squares Equations are solved; when inverting normal equations, matrix block inversion and Cholesky parallel decomposition methods can be used.

The parameters to be estimated for the whole network solution in the non-difference mode include station coordinates, satellite orbits, double-difference ambiguities, tropospheric delays, receiver and satellite clock errors, fractional phase deviations, and earth rotation parameters, etc.

The observation data of the present invention include GPS, Galileo, BDS and GLONASS systems, multiple frequency pseudorange and carrier phase observation values.

The above is only a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited thereto. Any person familiar with the technical field within the technical scope disclosed in the present invention can obviously obtain the simplicity of the technical solution. Changes or equivalent replacements all fall within the protection scope of the present invention.

The essence of the method for parallel processing of large-scale GNSS network observation data non-difference whole network of the present invention is to realize the multi-core parallel execution of many links of large-scale GNSS network observation data non-difference whole network calculation. The experiment uses the observation data of 500 GNSS reference stations distributed around the world, and adopts dual-core, quad-core and six-core parallel schemes respectively. Through testing, the large-scale GNSS network observation data non-differential network multi-core parallel processing method proposed by the present invention greatly shortens the calculation time and improves the calculation efficiency compared with the traditional single-core serial method. Dual-core, quad-core, and six-core parallel processing The speedup ratios reached 1.60, 3.05, and 4.50 times respectively. The actual application effect is closely related to the performance of the hardware system, the size of the GNSS network and the quality of the observation data.

Therefore, compared with the prior art, the present invention has the following outstanding beneficial technical effects:

(1) Improve the timeliness of non-difference whole network calculation of large-scale GNSS network observation data.

The invention proposes a large-scale GNSS network observation data non-difference whole network parallel calculation method, and realizes a certain number of uniformly distributed core stations in the GNSS network to solve satellite orbits, satellite clock errors and earth rotation parameters in parallel under the multi-core platform ; the core station parallelly estimates the wide-lane and narrow-lane fractional phase deviation; and the parallel calculation of the whole network non-difference ambiguity, the parallel generation of the whole network carrier distance and the parallel calculation of the whole network solution in the non-difference mode shortens the large-scale GNSS network observation The calculation time of the whole network without data difference improves the calculation efficiency.

(2) Easy to expand.

The method proposed by the invention has wide applicability and strong expansibility, is suitable for processing multi-frequency observation values of various satellite navigation systems, and performs calculation of the whole network based on the non-difference mode. The observation data of systems such as GPS, Galileo, BDS and GLONASS can all be incorporated into the non-difference whole network parallel computing method of the present invention, and the method proposed by the present invention can be used to carry out non-differentiated multi-frequency observation data of the selected satellite navigation system. Parallel computing of the entire network under differential mode, and not limited to the above 2 frequencies, but still applicable to 3, 4 and 5 frequencies, the present invention is effectively applied to the "Geodesy and Surveying In the field of "engineering" technology, the large-scale GNSS network observation data is realized, and the parallel calculation of the whole network is not different, and the economic and social benefits are huge.

Claims (8)

1. A large-scale GNSS network observation data non-differential whole network parallel processing method is characterized by comprising the following steps:

1) Selecting a set number of measuring stations as core stations, and parallelly resolving satellite orbit, satellite clock error and earth rotation parameters according to the observation data of the core stations; estimating satellite orbit and earth rotation parameters through Kalman filtering in a non-differential mode, constructing a non-differential observation equation by taking a single core station as parallel granularity, and restraining the non-differential observation equation by constructing double-differential ambiguity and fixing the ambiguity;

2) Estimating the wide lane decimal phase deviation and the narrow lane decimal phase deviation of a satellite end by utilizing satellite orbit, satellite clock error and earth rotation parameters;

3) Constructing non-differential ambiguity for each core station, and fixing ionosphere-free combined ambiguity by using the rounded wide lane decimal phase deviation and narrow lane decimal phase deviation;

for other stations except the core station, carrying out inter-station parallel non-difference fixed solution calculation by utilizing satellite orbit, satellite clock error, earth rotation parameters, wide lane decimal phase deviation and narrow lane decimal phase deviation of a satellite end, constructing non-difference ambiguity for each other station, and fixing ionosphere-free combined ambiguity;

4) Generating carrier distance observation values of the corresponding stations in parallel according to the carrier phase observation values of each station and the fixed ionosphere-free combined ambiguity;

5) And combining the pseudo-range observation value and the carrier distance observation value of each measuring station, constructing a normal equation in parallel, carrying out parallel inversion on the normal equation, and carrying out whole-network parallel calculation based on a non-difference mode.

2. The method for parallel processing of non-differential whole network of large-scale GNSS network observations according to claim 1, wherein the core stations are not less than 100 stations distributed uniformly around the world.

3. The method for non-differential whole network parallel processing of large-scale GNSS network observation data according to claim 1, wherein in the step 2), the satellite-side wide lane decimal phase deviation and the satellite-side narrow lane decimal phase deviation are estimated in parallel by performing inter-station parallel non-differential solution calculation on a core station.

4. The method according to claim 1, wherein in step 4), the carrier distance observations are obtained by correcting the carrier phase observations using a fixed ionosphere-free combined ambiguity and antenna phase wrapping effect.

5. The method according to claim 1, wherein in step 5), the pseudorange observations are subtracted by a satellite-side pseudorange hardware delay.

6. The method for performing non-differential whole network parallel processing on large-scale GNSS network observation data according to claim 1, wherein in step 5), the method for performing whole network parallel calculation based on the non-differential mode is as follows: taking the pseudo-range observation value and the carrier distance observation value as observed quantities; constructing a normal equation of each epoch in parallel by taking the epoch as granularity; solving a normal equation by adopting a parameter elimination algorithm based on recursive least square; the normal equation is inverted by matrix block inversion and Cholesky parallel decomposition.

7. The method for parallel processing of non-differential whole network of large-scale GNSS network observation data according to claim 1, wherein the parameters to be estimated obtained by parallel calculation of whole network include: station coordinates, satellite orbit, double-difference ambiguity, tropospheric delay, receiver and satellite clock bias, fractional phase bias and earth rotation parameters.

8. The method for parallel processing of non-differential whole network of large-scale GNSS network observation data according to claim 1, wherein for the non-ionospheric combined ambiguities of each station unfixed in the step 3), the non-ionospheric combined ambiguities of station star double differences are constructed, and the fixation of the non-ionospheric combined ambiguities is performed.

Images