CN114826337A

CN114826337A - Pre-compensation method and system for Doppler frequency offset of satellite communication signal

Info

Publication number: CN114826337A
Application number: CN202210494244.4A
Authority: CN
Inventors: 李红光; 刘垚圻; 卓蕊潋; 石晶林; 许凯波; 陈丽; 盛秋明; 蒋佳佳
Original assignee: Institute of Computing Technology of CAS
Current assignee: Institute of Computing Technology of CAS
Priority date: 2022-03-23
Filing date: 2022-05-07
Publication date: 2022-07-29

Abstract

The embodiment of the invention provides a precompensation method for Doppler frequency offset of satellite communication signals, which comprises the following steps: predicting a plurality of orbit parameters in the satellite orbit data at the current moment by using the trained neural network model, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, an argument of a near point and an argument of a near point; predicting rising point right ascension parameters in satellite orbit data at the current moment by using a trained linear function module; determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the perigee argument, the mean perigee angle and the ascension parameter of the ascending point obtained by prediction; and according to the relative position and the relative speed of the satellite relative to the mobile terminal, performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite.

Description

Pre-compensation method and system for Doppler frequency offset of satellite communication signal

Technical Field

The present invention relates to the field of satellite communication technologies, and in particular, to a method and a system for precompensating doppler frequency offset of a satellite communication signal.

Background

The high dynamic state is one of the biggest characteristics of a low-orbit satellite communication system, an LEO satellite moves at a speed of about 7km/s, and has a very high relative movement speed for a ground static target, and the LEO satellite moves circularly around the earth center, and the radial speed of the LEO satellite and a communication target carrier is time-varying, so that the LEO satellite has a certain relative acceleration, the high dynamic state of carrier Doppler causes the low-orbit satellite communication system to adopt a link synchronization technology which is more complex than that of the high-orbit satellite communication system or the ground communication system, and the synchronization problem of the satellite and a terminal is mainly divided into four parts, namely Doppler frequency offset pre-compensation, time delay compensation, signal frequency offset compensation and phase compensation. The pre-synchronization technology of satellite communication mainly relates to Doppler frequency offset pre-compensation and signal frequency offset compensation, and in order to ensure reliable communication, the Doppler frequency offset with high ground accuracy of a satellite in a visual range is acquired.

In the prior art, signal frequency offset estimation and compensation are generally performed in a traditional angle based on signal recovery, but a low-earth orbit satellite moves at a high speed and generates a great relative motion speed with a ground communication target, so that the doppler frequency offset caused by the movement increases the signal recovery difficulty of a user terminal, and therefore, for a satellite system with high dynamics, large frequency offset and large frequency offset change rate, rapid signal frequency offset estimation and compensation are difficult to perform from the angle of signal recovery. The Doppler frequency offset estimation and compensation are the precondition for accessing the mobile terminal user. The inventor finds that the orbit data of the satellite can be predicted in the research process, so that a precompensation algorithm based on ephemeris data is provided, but simultaneously finds that the orbit parameters of the satellite orbit data in the ephemeris data are different in change rule, the prediction is not easy, a large error exists in the predicted result, the accuracy of the orbit data of the satellite cannot be guaranteed, the calculated Doppler frequency offset precompensation value is inaccurate, the compensation effect is not good enough, and the accuracy of subsequent signal decoding is greatly reduced.

Therefore, a method for improving the accuracy of the doppler frequency offset precompensation value to perform accurate frequency offset precompensation is needed.

Disclosure of Invention

Accordingly, the present invention is directed to overcoming the above-mentioned shortcomings of the prior art and providing a method and system for pre-compensating for doppler frequency offset of satellite communication signals.

The purpose of the invention is realized by the following technical scheme:

according to a first aspect of the present invention, there is provided a method for pre-compensation of doppler frequency offset of satellite communication signals, comprising: predicting a plurality of orbit parameters in the satellite orbit data at the current moment by using the trained neural network model, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, an argument of a near point and an argument of a near point; predicting rising point right ascension parameters in satellite orbit data at the current moment by using a trained linear function module; determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the perigee argument, the mean perigee angle and the ascension parameter of the ascending point obtained by prediction; and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal.

In some embodiments of the present invention, the step of performing doppler frequency offset pre-compensation on a down-conversion signal corresponding to a received signal obtained by a mobile terminal from a satellite according to a relative position and a relative speed of the satellite relative to the mobile terminal includes: acquiring position information and speed information of the mobile terminal and relative position and relative speed of the satellite relative to the mobile terminal when the mobile terminal obtains a received signal from the satellite so as to determine a pre-compensation value of Doppler frequency offset; and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to the received signal obtained by the mobile terminal based on the pre-compensation value of the Doppler frequency offset.

In some embodiments of the invention, the trained neural network model includes a plurality of LSTM neural network elements, wherein each of the plurality of orbit parameters corresponds to one LSTM neural network element.

In some embodiments of the present invention, the LSTM neural network element corresponding to each of the plurality of trajectory parameters is trained in the following manner: acquiring time sequence information corresponding to corresponding orbit parameters in satellite orbit data to form a first training set of the orbit parameters, wherein the time sequence information corresponding to the orbit parameters comprises a plurality of time points and the value of the orbit parameters at each time point; training the LSTM neural network unit corresponding to the track parameter by using the first training set, and determining the estimated value of the track parameter at the next time point of the corresponding time point according to the value of the track parameter at the corresponding time point; and calculating a loss value according to the estimated value of the track parameter at the next time point of the corresponding time point and the real value of the track parameter at the next time point of the time point recorded in the time sequence information corresponding to the track parameter, and updating the parameter of the LSTM neural network unit corresponding to the track parameter based on the loss value.

In some embodiments of the present invention, the first training set for each of the plurality of orbit parameters is constructed by: acquiring historical ephemeris data at preset time intervals to obtain satellite orbit data at different time points; selecting the values of the orbit parameters in the satellite orbit data at different time points aiming at the corresponding orbit parameters of the orbit parameters to obtain an initial data set of the orbit parameters; based on the threshold range set for the track parameter, eliminating data which is not in the threshold range in the initial data set of the track parameter to obtain a first training set of the track parameter, wherein each track parameter of a plurality of track parameters corresponds to one set threshold range.

In some embodiments of the present invention, the updating the parameters of the corresponding LSTM neural network element based on the total loss includes: and calculating gradient values according to the total loss, and updating parameters of corresponding LSTM neural network units in a back propagation mode.

In some embodiments of the invention, the trained linear function module is trained in the following manner: acquiring time sequence information corresponding to the ascension parameter of the ascending node in the satellite orbit data to form a second training set of the ascension parameter of the ascending node, wherein the time sequence information corresponding to the ascension parameter of the ascending node comprises a plurality of corresponding time points and values of the ascension parameter of the ascending node corresponding to the time points; training a linear function module by using a second training set to determine an estimated value of the ascension parameter of the next time point of the corresponding time point according to the value of the ascension parameter of the corresponding time point; and calculating an error based on the estimated value of the ascension point parameter of the next time point of the corresponding time point and the real value of the ascension point parameter of the next time point of the corresponding time point recorded in the time sequence information corresponding to the ascension point parameter, and updating the parameter of the linear function module based on the error.

In some embodiments of the present invention, the method for constructing the second training set comprises: acquiring historical ephemeris data at preset time intervals to obtain satellite orbit data at different time points; selecting elevation intersection right ascension parameters in satellite orbit data at different time points to obtain an elevation intersection right ascension data set; and based on the threshold range set for the ascending crossing point right ascension parameters, eliminating the ascending crossing point right ascension parameters which are not in the threshold range in the ascending crossing point right ascension data set to obtain a second training set.

In some embodiments of the present invention, the linear function module predicts the ascension parameter of the ascending intersection by using a least square method.

In some embodiments of the present invention, the obtaining the historical ephemeris data at the predetermined time interval to obtain satellite orbit data at different time points includes: acquiring satellite orbit data of a corresponding time point in historical ephemeris data, a real relative position and a real relative speed of a satellite relative to a mobile terminal at a preset time interval; determining the relative position and the relative speed of the satellite at the corresponding time point by adopting an orbit extrapolation algorithm based on the satellite orbit data at the corresponding time point, and obtaining a total difference value based on the position difference between the determined relative position and the real relative position and the speed difference between the determined relative speed and the real relative speed; and based on the total difference value corresponding to each time point, satellite orbit data of which the total difference value is greater than the time point corresponding to the preset threshold value are removed, and satellite orbit data corresponding to a time point in preset time before the time point are obtained, so that satellite orbit data of different time points are obtained.

In some embodiments of the present invention, the calculating the relative position and the relative velocity of the satellite with respect to the mobile terminal comprises determining based on the predicted semi-major axis, eccentricity, orbit inclination, perigee argument, mean perigee argument, and ascension parameter of the ascent point using an orbit extrapolation algorithm, wherein the orbit extrapolation algorithm comprises: a two-body model based orbit extrapolation algorithm, an SGP4 orbit extrapolation algorithm, a J2 orbit extrapolation algorithm, and a J4 orbit extrapolation algorithm.

According to a second aspect of the present invention, there is provided a system for pre-compensation of doppler frequency offset of satellite communication signals, comprising: the trained neural network model is used for predicting a plurality of orbit parameters in the satellite orbit data of the current time point, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, a near point argument and a near point argument; the trained linear function module is used for predicting rising point right ascension parameters in the satellite orbit data at the current time point; the orbit extrapolation module is used for determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the argument of the near place, the argument of the mean near point and the ascension parameter of the ascending intersection point obtained by prediction; and the pre-compensation module is used for pre-compensating Doppler frequency offset of a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal.

According to a third aspect of the present invention, there is provided an electronic apparatus comprising: one or more processors; and a memory, wherein the memory is to store executable instructions; the one or more processors are configured to implement the steps of the method of the first aspect of the invention via execution of the executable instructions.

Compared with the prior art, the invention has the advantages that:

1. the invention considers the different properties of the change rule of different orbit parameters, predicts the nonlinear orbit parameters (semimajor axis, eccentricity, orbit dip angle, perigee argument and perigee angle) of the change rule by utilizing the trained neural network model, and the linear function module is used for predicting the orbit parameters (rising point right ascension parameters) with linear change rules, the obtained satellite orbit data (namely the semimajor axis, the eccentricity, the orbit inclination angle, the perigee argument, the mean and perigee argument and the rising point right ascension parameters) are more accurate, a more accurate satellite orbit data estimation method is provided for the Doppler frequency offset problem, in addition, the relative position and the relative speed of the satellite relative to the mobile terminal are determined based on the predicted semi-major axis, eccentricity, orbit inclination, perigee argument, mean perigee angle and ascension parameter of the ascending point; according to the relative position and the relative speed of the satellite relative to the mobile terminal, performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite; the final decoding accuracy can be improved, in addition, Doppler frequency offset pre-compensation is carried out on the corresponding down-conversion signals, and the problems that when signal recovery is carried out on the signals obtained after the down-conversion signals are analyzed, signal frequency offset estimation is difficult, signal frequency offset compensation cannot be carried out quickly and the like are solved.

2. According to the method, on one hand, after part of the acquired satellite orbit data is removed, the accuracy of the acquired satellite orbit data is ensured, and meanwhile, the satellite orbit data corresponding to a time point in a preset time before the time point is obtained again to be used as supplement for removing the corresponding satellite orbit data, so that the richness of a data set is ensured. On the other hand, based on the threshold range set for the orbit parameter, the data which is not in the threshold range in the initial data set of the orbit parameter is removed, wherein each orbit parameter of a plurality of orbit parameters is set with a threshold range correspondingly, so that the first training set and the second training set of the corresponding orbit parameters are obtained, and the prediction accuracy of the neural network model and the linear function module can be greatly improved.

Drawings

Embodiments of the invention are further described below with reference to the accompanying drawings, in which:

FIG. 1 is a schematic diagram of a neural network model according to one embodiment of the present invention;

FIG. 2 is a schematic diagram of the input and output of the LSTM neural network unit corresponding to the corresponding orbit parameter in the neural network model at the time point t according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of the input and output of the LSTM neural network unit corresponding to the corresponding orbit parameter in the neural network model at the time point t-1 immediately preceding the time point t according to one embodiment of the present invention;

FIG. 4 is a schematic diagram of the structure and internal data processing of an LSTM neural network element according to one embodiment of the present invention;

FIG. 5 is a diagram illustrating calculation of a frequency offset pre-compensation value based on a neural network model and a linear function model according to an embodiment of the present invention;

FIG. 6 is a flow chart of a method for pre-compensating for Doppler frequency offset of a satellite communication signal according to one embodiment of the invention;

FIG. 7 is a schematic illustration of a satellite orbit according to one embodiment of the invention;

FIG. 8 is a schematic diagram illustrating the fitting effect of hyper-parametric training of the ascension crossing point according to an embodiment of the present invention;

FIG. 9a is a schematic diagram illustrating a comparison between predicted values and actual observed values of the semi-major axis according to an embodiment of the present invention;

FIG. 9b is a diagram illustrating a prediction error of a semi-major axis according to an embodiment of the present invention;

FIG. 10a is a schematic diagram illustrating a comparison between a predicted value and a true observed value of eccentricity according to an embodiment of the present invention;

FIG. 10b is a graphical illustration of the prediction error of eccentricity according to one embodiment of the present invention;

FIG. 11a is a schematic diagram illustrating a comparison between a predicted value and a true observed value of a track inclination according to an embodiment of the present invention;

FIG. 11b is a schematic illustration of the prediction error of the track inclination according to one embodiment of the present invention;

FIG. 12a is a schematic diagram illustrating a comparison between a predicted value and a true observed value of a argument of a near location according to an embodiment of the present invention;

FIG. 12b is a schematic diagram of the prediction error of the perigee argument according to one embodiment of the present invention;

FIG. 13a is a schematic diagram illustrating a comparison between a predicted value and a true observed value of a mean anomaly according to an embodiment of the present invention;

FIG. 13b is a schematic illustration of prediction error of mean anomaly according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail by embodiments with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As mentioned in the background section, the synchronization problem between the satellite and the mobile terminal is mainly divided into four parts, doppler frequency offset pre-compensation, delay compensation, doppler frequency offset compensation and phase compensation. In the prior art, the conventional angle based on signal recovery is generally used for Doppler frequency offset estimation and compensation, but for a satellite system with high dynamic, large frequency offset and frequency offset change rate, rapid estimation and compensation are difficult to perform. The Doppler frequency offset estimation and compensation are the precondition for accessing the mobile terminal user. The inventor finds that the orbit data of the satellite can be predicted in the research process. However, the applicant finds that if all orbit parameters are directly predicted by using a pre-trained neural network, the prediction accuracy is poor, so that the compensation effect is not good enough. After numerous analyses and studies, the applicant found that the change law of some orbital parameters (i.e. semimajor axis, eccentricity, orbital inclination, perigee argument and perigee angle) is non-linear, while the change law of other orbital parameters (i.e. ascension parameter at ascending intersection) is (or close to) linear. Therefore, the trajectory parameters of different transformation laws need to be predicted in different ways on a targeted basis.

Based on the above problem, the present invention provides a method for pre-compensating doppler frequency offset of satellite communication signals, comprising: predicting a plurality of orbit parameters in the satellite orbit data at the current moment by using a trained neural network model (equivalent to a nonlinear model trained in advance), wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity, an orbit inclination angle, a perigee argument and a perigee angle; predicting rising point right ascension parameters in satellite orbit data at the current moment by using a trained linear function module; determining the relative position and relative speed of the satellite relative to the mobile terminal based on the predicted semi-major axis, eccentricity, orbit inclination, perigee argument, mean perigee angle and ascension parameter of the ascending intersection point; and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal. The method considers the different properties of the change rules of different orbit parameters, predicts the nonlinear orbit parameters (semimajor axis, eccentricity, orbit dip angle, perigee argument and perigee angle) of the change rules by utilizing a trained neural network model, and the linear function module is used for predicting the orbit parameters (rising point right ascension parameters) with linear change rules, the obtained satellite orbit data (namely the semimajor axis, the eccentricity, the orbit inclination angle, the perigee argument, the mean and perigee argument and the rising point right ascension parameters) are more accurate, a more accurate satellite orbit data estimation method is provided for the Doppler frequency offset problem, in addition, the relative position and the relative speed of the satellite relative to the mobile terminal are determined based on the predicted semi-major axis, eccentricity, orbit inclination, perigee argument, mean perigee angle and ascension parameter of the ascending point; according to the relative position and the relative speed of the satellite relative to the mobile terminal, performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite; the method and the device can improve the accuracy of final decoding, and the Doppler frequency offset pre-compensation is carried out at the position of the down-conversion signal corresponding to the received signal instead of carrying out compensation after the down-conversion signal is analyzed, so that the problems that when the signal obtained after the down-conversion signal is analyzed is recovered, the signal frequency offset estimation is difficult, the signal frequency offset compensation cannot be carried out quickly and the like are solved.

According to an embodiment of the present invention, the plurality of orbit parameters in the satellite orbit data include five orbit parameters, namely a semi-major axis, an eccentricity ratio, an orbit inclination angle, a perigee argument and a perigee angle, and the five orbit parameters and the ascension parameter of the rising intersection in the satellite orbit data constitute six satellite orbits, that is, the five orbit parameters and the ascension parameter of the rising intersection are satellite orbit parameters. The semi-major axis is denoted by symbol a, the eccentricity is denoted by symbol e, the track inclination is denoted by symbol i, the perigee argument is denoted by symbol ω, the equatorial argument is denoted by symbol M, and the ascension parameter at the rising intersection is denoted by symbol Ω.

When performing doppler frequency offset precompensation, a trained neural Network model needs to be obtained, and according to an embodiment of the present invention, the obtained trained neural Network model includes a plurality of LSTM (Long Short Term Memory Network) neural Network units. Referring to fig. 1, the present invention needs to use a neural network model to predict five orbit parameters, i.e., semi-major axis, eccentricity, orbit inclination, perigee argument and perigee argument, wherein the five orbit parameters are represented by symbols

Model prediction of representation neural networkSemi-major axis, symbol

Representing eccentricity, sign predicted by neural network model

Representing predicted orbit inclination, sign of neural network model

Representing argument and sign of near place predicted by neural network model

And (3) representing the mean anomaly predicted by the neural network model, and correspondingly predicting an orbit parameter by an LSTM neural network unit. The trained neural network model thus obtained includes five LSTM neural network elements for predicting respective ones of a plurality of orbital parameters.

Based on the above neural network model structure, according to an embodiment of the present invention, first, an LSTM neural network unit should be trained independently for each of the plurality of orbit parameters, that is, a semi-major axis, an eccentricity, an orbit inclination angle, an argument of a perigee, and an argument of a perigee are respectively trained correspondingly. And correspondingly training an LSTM neural network unit by one parameter to obtain the LSTM neural network unit suitable for predicting the corresponding parameter, thereby ensuring the prediction precision of each track parameter.

It should be understood that in the above embodiments, it is only one preferred embodiment that each of the parameters of the semi-major axis, eccentricity, track inclination, argument of perigee and argument of perigee corresponds to one LSTM neural network element, and this way more accurate prediction values can be obtained. However, in practice, an LSTM model may be trained jointly with the semi-major axis, eccentricity, orbit tilt, perigee argument, and perigee angle, with five inputs (semi-major axis, eccentricity, orbit tilt, perigee argument, and perigee angle at the previous time) and five outputs (predicted semi-major axis, eccentricity, orbit tilt, perigee argument, and perigee angle).

Before independently training an LSTM neural network unit for predicting a corresponding track parameter in a plurality of track parameters, a first training set for training the corresponding track parameter of the LSTM neural network unit corresponding to the corresponding track parameter needs to be constructed. According to one embodiment of the present invention, the first training set for each of the plurality of track parameters is constructed in a manner comprising steps a1, a2, and A3:

step A1: and acquiring historical ephemeris data at preset time intervals to obtain satellite orbit data at different time points.

Because the mobile terminal may not be powered on for a long time, ephemeris may be a certain past time, and the acquired historical ephemeris data itself may have errors, the acquired ephemeris data with errors needs to be removed. According to an embodiment of the present invention, step a1 specifically includes:

step A101: satellite orbit data, a true relative position and a true relative velocity of the satellite with respect to the mobile terminal at respective time points in the historical ephemeris data are acquired at predetermined time intervals.

According to an embodiment of the present invention, assuming that there are M days of historical ephemeris data, each of which performs data acquisition at an interval of H hours, for example, ephemeris data in the previous 100 days is acquired at an interval of 6 hours, the ephemeris data is calculated at 24 hours a day, and 4 pieces of ephemeris data are acquired each day, 400 pieces of ephemeris data are selected, and each piece of ephemeris data corresponds to satellite orbit data including a corresponding time point, a true relative position of a satellite relative to the mobile terminal, and a true relative velocity. The satellite orbit data comprises semi-major axis, eccentricity, orbit inclination angle, perigee argument, mean perigee angle and ascension parameter of ascending point.

Step A102: and determining the relative position and the relative speed of the satellite at the corresponding time point by adopting an orbit extrapolation algorithm based on the satellite orbit data at the corresponding time point, and obtaining a total difference value based on the position difference between the determined relative position and the real relative position and the speed difference between the determined relative speed and the real relative speed. The calculation of the relative position and the relative speed of the satellite relative to the mobile terminal comprises the following steps of determining by using an orbit extrapolation algorithm based on parameters of a semi-major axis, an eccentricity, an orbit inclination angle, a perigee argument, a mean perigee angle and a rising intersection right ascension included in satellite orbit data, wherein the orbit extrapolation algorithm can be adopted as follows: an SGP4 orbit extrapolation algorithm, a J2 orbit extrapolation algorithm, and a J4 orbit extrapolation algorithm.

Step A103: and based on the total difference value corresponding to each time point, satellite orbit data of which the total difference value is greater than the time point corresponding to the preset threshold value are removed, and satellite orbit data corresponding to a time point in preset time before the time point are obtained, so that satellite orbit data of different time points are obtained. For example, satellite orbit data at 6 o ' clock of a certain day is rejected, satellite orbit data corresponding to a time point within 1 hour before the 6 o ' clock of the day is acquired, and the satellite orbit data at 5 o ' clock of the day replaces the rejected satellite orbit data. According to the method, the satellite orbit data of the corresponding time point with the difference value larger than the preset threshold value are removed, so that the accuracy of the obtained satellite orbit data is guaranteed, and meanwhile, the satellite orbit data corresponding to a time point in the preset time before the time point is obtained again to supplement the removed corresponding satellite orbit data, and the richness of the data set is guaranteed. The prediction accuracy of the neural network model is further improved.

Step A2: and selecting the values of the orbit parameters in the satellite orbit data at different time points aiming at the corresponding orbit parameters of the orbit parameters to obtain an initial data set of the orbit parameters.

According to an embodiment of the present invention, the plurality of orbit parameters include five orbit parameters, i.e., a semi-major axis, an eccentricity ratio, an orbit inclination angle, a perigee argument, and a perigee angle, and values of the orbit parameters in the satellite orbit data at different time points are selected for corresponding orbit data of the five orbit parameters, so as to obtain an initial data set of the semi-major axis parameters, an initial data set of the eccentricity ratio parameters, an initial data set of the orbit inclination angle parameters, an initial data set of the perigee argument parameters, and an initial data set of the perigee argument parameters.

Step A3: based on the threshold range set for the track parameter, eliminating data which is not in the threshold range in the initial data set of the track parameter to obtain a first training set of the track parameter, wherein each track parameter of a plurality of track parameters is set with a threshold range correspondingly.

According to an embodiment of the invention, five orbit parameters of long-time historical ephemeris data are respectively subjected to numerical range statistics, and are sorted from large to small, the most-valued distribution in the initial data set of the corresponding orbit parameters is observed, meanwhile, threshold ranges which are set correspondingly by the orbit parameters and accord with respective conditions are removed from the first and last 3% (it should be understood that the above steps are only schematic, and can be 2% or 4% and the like, and can be set according to the needs of specific conditions, and the invention does not make any limitation on the above steps) of the initial data set of the corresponding orbit parameters, and the corresponding orbit parameters which accord with the actual conditions reasonably are reserved. Therefore, a first training set of semimajor axis parameters, a first training set of eccentricity parameters, a first training set of track inclination angle parameters, a first training set of perigee argument parameters and a first training set of mean perigee angle parameters with more accurate data are obtained. And improving the prediction precision of the corresponding LSTM neural network unit.

And finally, training by using the first training set of the corresponding track parameters to obtain the LSTM neural network unit corresponding to the trained corresponding track parameters. According to an embodiment of the present invention, the LSTM neural network unit corresponding to each of the plurality of track parameters performs a plurality of rounds of training according to the following steps, each round of training comprising steps B1, B2, and B3:

step B1: the method comprises the steps of obtaining time sequence information corresponding to corresponding orbit parameters in satellite orbit data to form a first training set of the orbit parameters, wherein the time sequence information corresponding to the orbit parameters comprises a plurality of time points and the value of the orbit parameters at each time point.

According to an embodiment of the present invention, a timing information corresponding to a track parameter is expressed in terms of (time, value of the track parameter corresponding to a time point), for example, the timing information corresponding to a semi-major axis track parameter may include { (t-n, a) _t-n )…(t-1，a _t-1 )，(t，a _t )，(t+1，a _t+1 )…}，a _t Representing the value of the semi-major axis orbit parameter at a point in time t, a _t+1 The value of the semi-major axis orbit parameter at a time point next to the time point t. Likewise, e _t 、i _t 、ω _t And M _t Respectively representing the value of the eccentricity orbit parameter, the value of the orbit inclination angle orbit parameter, the value of the perigee argument orbit parameter and the value of the perigee angle orbit parameter at the time point t.

Step B2: and training the LSTM neural network unit corresponding to the track parameter by using the first training set, and determining the estimated value of the track parameter at the next time point of the corresponding time point according to the value of the track parameter at the corresponding time point.

According to one embodiment of the present invention, referring to FIG. 2, at time t, the input of the corresponding LSTM neural network element has x _t 、h _t-1 And c _t-1 Output h _t And c _t . Referring to FIG. 3, at time t-1 (i.e., the time immediately preceding time t), the input to the LSTM neural network element has x _t-1 、h _t-2 And c _t-2 Output h _t-1 And c _t-1 . Wherein x is _t For the value of the true corresponding orbit parameter for the current time point t, (x) _t ∈{a _t ，e _t ，i _t ，ω _t ，M _t })，h _t-1 An estimated value c of the corresponding orbit parameter for the next time point t of the time point t-1 output by the LSTM neural network unit _t-1 The long-term unit state c at the last time point t-1 output by the LSTM neural network unit _t-1 ，h _t An estimate of the corresponding orbit parameter at a time point t +1 next to the time point t and a long-term cell state c at the corresponding time point t _t ，x _t-1 The value of the true corresponding orbit parameter corresponding to the last time point t-1 of t, (x) _t-1 ∈{a _t-1 ，e _t-1 ，i _t-1 ，ω _t-1 ，M _t-1 })，h _t-2 An estimate of the corresponding orbit parameter at a time point t-1 next to the time point t-2, output by the LSTM neural network unit, c _t-2 Is the long-term cell state at time point t-2, where x _t-1 、x _t 、h _t-2 、h _t-1 、h _t 、c _t-2 、c _t-1 、c _t Are all vectors.

Further, the input data is processed as follows in combination with the specific structure of the LSTM neural network unit.

According to one embodiment of the present invention, referring to fig. 4, the LSTM neural network unit uses three control switches internally to solve the core problem of controlling the long-term state c, and the switch 1 is responsible for controlling to continue to save the long-term state c; the switch 2 is responsible for controlling the input of the instant state of the current time point into the long-term unit state c; the switch 3 is responsible for controlling whether the long-term state c is taken as the nonlinear training output of the current time point. W _c A weight matrix of a long-term state c, tanh is a hyperbolic cosine function and is used for determining the cell state of the current time point t, the control switch is realized by the concept of gates in the algorithm, sigma 1, sigma 2 and sigma 3 are a forgetting gate, an input gate and an output gate respectively, W _f 、W _i And W _o The weight matrixes of corresponding gates are respectively, each gate is a layer of full-connection layer, the input of each gate is a vector, the working mechanism is that the output vector of the gate is multiplied by a vector needing to be controlled according to elements, and the output is a real number vector between 0 and 1. Wherein if the output of the gate is 0, the vector of control cannot pass; conversely, if the gate input is 1, then all the vectors of the control can pass through without loss; if the output of the gate is between 0 and 1, the vector of control will pass with different weights. The forgetting gate is used for controlling how much the long-term unit state of the last time point is reserved to the current time point; the input gate is used to control how much the input of the network at the current time point is saved to the long-term unit state at the current time point. The output gate (output gate) is used to control how much the long-term cell state at the current time point is output to the current output value.

The expression for a forgetting gate is as follows:

f _t ＝σ1(W _f ·[h _t-1 ,x _t ]+b _f )

wherein f is _t Is the output of a forgetting gate, σ 1 is the sigmoid function, W _f Is the weight matrix of the forgetting gate, [ h ] _t-1 ,x _t ]Indicating handleOutput value h at time point t-1 _t-1 And input value x of corresponding track parameter at t time point _t Spliced into a vector, b _f Is a biased term for a forgetting gate. When inputting x _t Is d in the dimension of _x ，h _t-1 Is d in the dimension of _h Cell state C _t-1 Is d in the dimension of _c (in general d) _c ＝d _h ) Then forget the weight matrix W of the gate _f Dimension is d _c ×(d _h +d _x ). Weight matrix W of forgetting gate _f By inputting h left behind _t-1 Weight matrix W of _fh Inputting x _t Weight matrix W of _fx Are composed of _f And inputting h of forgetting gate _t-1 And x _t The respective weight matrix correspondence is as follows:

＝W _fh h _t-1 +W _fx x _t

the expression for the input gate is as follows:

I _t ＝σ2(W _i ·[h _t-1 ,x _t ]+b _i )

wherein, I _t Is the output of the input gate, σ 1 is the sigmoid function, W _i Is a weight matrix of the input gate, b _i Is the offset term of the input gate.

H output at last time point t-1 _t-1 And x input at the current time point t _t Obtaining the unit state c 'of the current time point' _t ：

c′ _t ＝tanh(W _c ·[h _t-1 ,x _t ]+b _c )

Wherein, W _c As a weight matrix, b _c Is the bias term.

Long term cell state c at the current point in time _t Can be represented by the following formula:

wherein, the symbol

Denotes multiplication by element, c _t-1 Representing the long-term cell state at the last time point t-1, f _t Is a forgetting gate, c' _t Is the cell state at the current point in time, I _t Is an input gate. The elements are multiplied by the corresponding elements of the tensor, which is the calculation of element multiplication, and the result of multiplication at each position is used as a return value. The control of the forgetting gate allows history information to be stored for a long time, and the control of the input gate prevents unimportant contents from being memorized. The output gate controls the influence of long-term memory on the current output, and the expression of the output gate is as follows:

o _t ＝σ3(W _o ·[h _t-1 ,x _t ]+b _o )

wherein o is _t To output the output of the gate, W _o As a weight matrix of output gates, b _o Is the bias term of the output gate.

Output o from output gate _t And a long-term cell state c of the current time point t _t The final output of the joint determination is as follows:

wherein h is _t The long-term unit state of the last time point t-1 output by the LSTM neural network unit and the corresponding orbit parameter estimated value h of the current time point t output by the LSTM neural network unit are based on the corresponding orbit parameter of the input t time point for the corresponding LSTM neural network unit _t-1 And outputting a corresponding track parameter estimated value h of the next time point t +1 _t 。

Step B3: and calculating a loss value according to the estimated value of the track parameter at the next time point of the corresponding time point and the real value of the track parameter at the next time point of the time point recorded in the time sequence information corresponding to the track parameter, and updating the parameter of the LSTM neural network unit corresponding to the track parameter based on the loss value. And when the loss value is within a preset threshold value or reaches a preset training turn, stopping updating the parameters of the LSTM neural network unit corresponding to the track parameters, and finishing the training.

According to one embodiment of the invention, updating parameters of the respective LSTM neural network element includes: and calculating gradient values according to the total loss, and updating parameters of corresponding LSTM neural network units in a back propagation mode. Specifically, updating the parameters in a back propagation manner includes: calculating the output value of each neuron in forward direction, i.e. calculating f _t 、I _t 、c _t 、o _t 、h _t The values of the five vectors. And calculating the loss value of each neuron reversely to obtain the total loss. A gradient value is calculated for each weight based on the corresponding loss value. Wherein the backward propagation includes backward propagation in time or propagation to an upper layer.

In order to adapt to the regularity of different orbit parameters, according to an embodiment of the present invention, when performing doppler frequency offset precompensation, a trained linear function module is further required to be obtained, and the obtained linear function module predicts the right ascension parameter of the intersection point by using a least square method. Better ascension point right ascension parameter can be obtained, and the prediction precision of the ascension point right ascension parameter is improved.

First, to obtain a trained linear function module, a second training set for training the ascension parameters of the ascending intersection of the linear function module is constructed. According to an embodiment of the present invention, the construction method of the second training set includes steps C1, C2, and C3:

step C1: and acquiring historical ephemeris data at preset time intervals to obtain satellite orbit data at different time points. Wherein the step C1 is the same as the step a1 of constructing the first training set.

Step C2: and selecting elevation intersection right ascension parameters in the satellite orbit data at different time points to obtain an elevation intersection right ascension data set.

Step C3: and based on the threshold range set for the ascending crossing point right ascension parameters, eliminating the ascending crossing point right ascension parameters which are not in the threshold range in the ascending crossing point right ascension data set to obtain a second training set.

According to an embodiment of the invention, the numerical range statistics is performed on the ascension point right ascension parameter of the long-time historical ephemeris data, the ascending intersection point right ascension parameters are sorted from large to small, the most-valued distribution in the ascending intersection point right ascension data set is observed, then a threshold range is set, the unreasonable ascending intersection point right ascension parameters of the top and the last 3% (it should be understood that the steps are only illustrated here, and can be 2%, 4% and the like, and the steps can be set according to the needs of specific situations, and the invention does not make any limitation on the steps), and the ascending intersection point right ascension parameters which are reasonable according with the actual situations are reserved. Thus, a second training set of ascending crossing right ascension parameters is obtained.

And finally, training by using the second training set for obtaining the ascension parameter of the ascending intersection point to obtain a trained linear function module. According to an embodiment of the present invention, the trained linear function module is obtained by performing multiple rounds of training in the following manner, wherein each round of training comprises steps D1, D2 and D3:

step D1: and acquiring time sequence information corresponding to the ascension parameter of the ascending intersection in the satellite orbit data to form a second training set of the ascension parameter of the ascending intersection, wherein the time sequence information corresponding to the ascension parameter of the ascending intersection comprises a plurality of corresponding time points and values of the ascension parameter of the ascending intersection corresponding to the time points.

Step D2: and training the linear function module by using the second training set to determine the estimated value of the rising-point right ascension parameter at the next time point of the corresponding time point according to the value of the rising-point right ascension parameter at the corresponding time point.

According to an embodiment of the present invention, n samples are selected from the second training set, where the n samples include n time points and a value of the ascension parameter at each time point, and the linear function module estimates the ascension parameter at the ascension point by using a least square method, and the determination method is as follows:

wherein the content of the first and second substances,

is an estimated value of the ascension parameter of the ascending intersection point,

as a function of the number of the coefficients,

t _i for the point in time of the ith sample of the n samples,

is the average of the rising point right ascension parameter, Ω, of n samples _i Is the true ascension point right ascension parameter of the ith sample,

is the average time of the time points of n samples, mod is the remainder operation, t _next Next point in time, t, being the corresponding point in time ₀ To select a reference time point, T _Ω Is the acquisition period of the ascension parameter of the ascending intersection point,

in order to obtain the intercept of the signal,

step D3: and calculating an error based on the estimated value of the ascension point parameter of the next time point of the corresponding time point and the real value of the ascension point parameter of the next time point of the corresponding time point recorded in the time sequence information corresponding to the ascension point parameter, and updating the parameter of the linear function module based on the error.

According to one embodiment of the invention, five LSTM neural network units in the invention are trained in parallel with the linear function model to improve the training speed.

After the neural network model and the linear function module are trained, referring to fig. 5, according to an embodiment of the present invention, the trained neural network model and linear function module are used to predict the satellite at the current time pointTrack data to obtain more accurate semimajor axis

Eccentricity ratio

Inclination angle of track

Argument of near place

Flat near point angle

Right ascension parameter of intersection

The relative position and the relative speed of the satellite relative to the mobile terminal are obtained by adopting an orbit extrapolation algorithm according to accurate satellite orbit data, a Doppler frequency offset pre-compensation value with small error is further obtained according to the relative position and the relative speed, and the frequency offset pre-compensation is carried out on the signal received by the mobile terminal based on the pre-compensation value obtained in the mode, so that the signal accuracy can be effectively improved.

According to an embodiment of the present invention, a method for pre-compensating for doppler frequency offset of satellite communication signals is provided, and referring to fig. 6, the method includes steps S1, S2, S3 and S4, and for better understanding of the present invention, each step is described in detail below with reference to specific embodiments.

Step S1: and acquiring the trained neural network model, and predicting a plurality of orbit parameters in the satellite orbit data at the current moment by using the neural network model, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, an argument of a near point and an argument of a near point.

According to one embodiment of the invention, the trained neural network model includes five LSTM neural network elements for predicting the semi-major axis, eccentricity, orbital inclination, perigee argument, and perigee angle in the satellite orbit data at the current time point, respectively.

Step S2: and acquiring the trained linear function module, and predicting the ascent point and ascent point parameters in the satellite orbit data at the current moment by using the linear function module.

Step S3: and determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the perigee argument, the mean perigee angle and the ascension parameter of the ascending point obtained by prediction.

According to one embodiment of the present invention, the calculating the relative position and the relative velocity of the satellite with respect to the mobile terminal includes determining based on the predicted semi-major axis, eccentricity, orbit inclination, perigee argument, mean perigee argument, and ascension parameter of the ascension point using an orbit extrapolation algorithm, wherein the orbit extrapolation algorithm includes: a two-body model based orbit extrapolation algorithm, an SGP4 orbit extrapolation algorithm, a J2 orbit extrapolation algorithm, and a J4 orbit extrapolation algorithm. The relative position and relative velocity are calculated below by taking a two-body model-based orbit extrapolation algorithm as an example.

Firstly, calculating a position vector and a velocity vector of a satellite at a current time point in an ECI (earth centered inertial coordinate system) coordinate system, and specifically comprising the following steps:

knowing the satellite orbit data of 6 o 'clock at the current time point (i.e. six orbits), the mean apogee angle M between the integer points (e.g. between 6 o' clock and 7 o 'clock) is calculated, e.g. the satellite orbit data of 6 o' clock is predicted in the neural network model and the linear function module, then based on the satellite orbit data of 6 o 'clock, the mean apogee angle M in the satellite orbit data of 6.002 time point is extrapolated by the two-body model (TwoBody), and the other five orbit parameters are taken as the ideal case, i.e. the five orbit parameters of the semi-major axis a, eccentricity e, orbit inclination i and apogee argument ω at 6.002 time point are the same as those predicted by 6 o' clock. The manner of solving the mean-anomaly angle by the two-body model (TwoBody) extrapolation is as follows:

M(t)＝ω _s (t-t _p )，

wherein M (t) is the mean-nearest-point angle of the current time point t (e.g. 6.002 time point),

GM＝3.986005×10 ¹⁴ m ³ /s ² is the constant of the earth's gravity, t _p Is the point in time of the satellite at the near site.

Further calculating the angle of approach point E in the following way:

E-esinE＝M(t)，

wherein e is the eccentricity.

The sine and cosine of the true anomaly angle f are calculated, the sine being expressed as:

the cosine is expressed as:

an exemplary satellite orbit is shown in fig. 7, wherein it is assumed that the unit vector of the azimuth of the satellite at the perigee is P, the vector Q is the unit vector perpendicular to P on the orbital plane, r is the distance of the satellite from the geocenter, a is the semimajor axis, ω is the perigee argument, and f is the true perigee angle. The unit vector P is expressed as follows:

vector Q is represented as follows:

based on sine and cosine of the true anomaly angle f and unit vectors P and Q, the position vector of the satellite in the ECI coordinate system can be obtained, and the specific calculation mode is as follows:

r _ECI ＝rcosfP+rsinfQ，

where, r ═ a (1-ecosE), E is the eccentricity, and E is the approximate point angle.

And (3) obtaining the sine and cosine of the true anomaly angle f to obtain a position vector as follows:

since the velocity vector is the derivative of the position vector with respect to time, i.e. with respect to r _ECI With respect to the time t derivative, the velocity vector of the satellite in the ECI coordinate system is:

secondly, the position vector and the velocity vector in the ECI coordinate system are converted into an ECEF (earth-centered, earth-fixed, earth-center-earth-fixed) coordinate system. Because in the process of calculating the motion parameters of the satellite, the coordinates of the satellite in the ECI coordinate system are generally required to be converted into the coordinates in the ECEF coordinate system. The coordinates of the satellite in the ECI coordinate system rotate the Greenwich mean time angle theta around the Z axis _g The positions of the available satellites in the ECEF coordinate system are:

r _ECEF ＝Rz(θ _g )r _ECI ，

wherein the content of the first and second substances,

θ _g ＝θ _g0 +w _e (t-t ₀ )，θ _g0 greenwich mean time angle at time t0, t being the current time, w _e Is the rotational angular velocity of the earth, about 7.2692X 10 ^-5 rad/s。

To r _ECEF With respect to the t derivative, the velocity of the satellite in the ECEF coordinate system is:

acquiring the elevation, longitude, latitude and longitude of the geographic position of the mobile terminal under the ECEF coordinate systemVelocity v _r . Obtaining the position r of the mobile terminal based on the elevation, longitude and latitude of the geographical position by using the following formula _r ：

Wherein the content of the first and second substances,

a _G is the major semi-axis of a general reference ellipsoid, e _G For the first eccentricity, L represents the longitude of the geographical position of the mobile terminal, B represents the latitude of the geographical position of the mobile terminal, and h represents the elevation of the geographical position of the mobile terminal.

Finally, the relative position r and the relative speed v of the satellite relative to the mobile terminal are obtained according to the following modes:

step S4: and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal.

According to an embodiment of the present invention, step S4 specifically includes the following steps:

step S401: and acquiring the position information and the speed information of the mobile terminal when the mobile terminal obtains the received signal from the satellite and the relative position and the relative speed of the satellite relative to the mobile terminal so as to determine a pre-compensation value of the Doppler frequency offset.

According to an embodiment of the present invention, the method for determining the pre-compensation value of the doppler frequency offset may be to calculate the doppler frequency offset according to a doppler frequency offset formula, and the specific calculation method is as follows:

wherein the content of the first and second substances,

is Doppler frequency offset, lambda is satellite signal carrier wavelength, f ₀ Is the carrier frequency of the satellite signal and,

is the angle between the relative position r and the relative velocity v, and c is the speed of light.

Step S402: and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to the received signal obtained by the mobile terminal based on the pre-compensation value of the Doppler frequency offset.

After the mobile terminal of the present invention is powered on and completes initialization, it may further determine whether a time interval between a latest ephemeris of a satellite in a current ephemeris database and a current time point exceeds a threshold, and if not, it indicates that there is no error or little error in the ephemeris data, so according to another embodiment of the present invention, another pre-compensation method for doppler frequency offset of a satellite communication signal is provided, and when the mobile terminal determines that there is no error or little error in the ephemeris data, the satellite orbit data of the current time is directly acquired, and step S3 and step S4 in the above embodiments are executed based on the acquired satellite orbit data, thereby completing the pre-compensation.

According to one embodiment of the present invention, there is provided a system for pre-compensation of doppler frequency offset of satellite communication signals, comprising: the trained neural network model is used for predicting a plurality of orbit parameters in the satellite orbit data of the current time point, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, an argument of a near point and an argument of a near point; the trained linear function module is used for predicting the ascension parameter of the rising point in the satellite orbit data at the current time point; the orbit extrapolation module is used for determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the argument of the near place, the argument of the mean near point and the ascension parameter of the ascending intersection point obtained by prediction; and the pre-compensation module is used for pre-compensating Doppler frequency offset of a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal.

To verify the effect of the present invention, the inventors tested a plurality of LSTM neural network elements of a trained linear function module and a trained neural network model, and the experimental effect is illustrated as follows:

the trained linear function module predicts the condition of the ascension at the ascending intersection: referring to fig. 8, the right ascension point sample according to 6 hours exhibited a saw-tooth wave with a period of 80 days, and the use of linear fitting had a good effect.

LSTM neural network units corresponding to the trained semi-major axis orbit parameters: hyper-parameters during training: the number of neurons is 300, the maximum number of iterations is 120, the learning rate of the first 80 is 0.005, and the learning rate of the following 40 is 0.2. Referring to fig. 9a, the abscissa of the coordinate system represents the corresponding time point, and the ordinate represents the true observed value and the predicted value of the semi-major axis, and referring to fig. 9b, the abscissa of the coordinate system represents the corresponding time point, and the ordinate represents the error value between the true observed value and the predicted value of the semi-major axis. Fig. 9a and 9b show that the semi-major axis is predicted well from 10 months and 11 days by using the LSTM neural network unit corresponding to the trained track parameters, the prediction effect of the semi-major axis is good, the maximum prediction error is about 750m, and the RMSE is 391.1401.

LSTM neural network units corresponding to the trained eccentricity orbit parameters: hyper-parametric situation during training: the number of neurons is 200, the maximum number of iterations is 120, the learning rate of the first 80 is 0.005, and the learning rate of the following 40 is 0.2. Referring to fig. 10a, the abscissa of the coordinate system represents a corresponding time point, and the ordinate represents the true observed value and the predicted value of the eccentricity, referring to fig. 10b, the abscissa of the coordinate system represents a corresponding time point, and the ordinate represents an error value between the true observed value and the predicted value of the eccentricity. FIGS. 10a and 10b show the results of using the LSTM neural network unit corresponding to the trained eccentricity orbit parameters to predict the eccentricity from 10 months and 11 days, which shows that the eccentricity prediction effect is better, and the maximum prediction error is 5 × 10 ^-5 Left and right, RMSE 2.2361 function10 ^-5 The eccentricity itself is small, so the maximum error and RMSE are not large.

LSTM neural network unit corresponding to the trained orbit inclination angle orbit parameter: hyper-parametric situation during training: the number of neurons is 200, the maximum number of iterations is 120, the learning rate of the first 80 is 0.005, and the learning rate of the following 40 is 0.2. Referring to fig. 11a, the abscissa of the coordinate system represents the corresponding time point, and the ordinate represents the true observed and predicted values of the track inclination, referring to fig. 11b, the abscissa of the coordinate system represents the corresponding time point, and the ordinate represents the error value between the true observed and predicted values of the track inclination. Fig. 11a and 11b show the results of predicting the orbit tilt angle from 10 months and 11 days by using the LSTM neural network unit corresponding to the trained orbit tilt angle orbit parameters, and the effect is more ideal.

LSTM neural network units corresponding to the trained perigee argument orbit parameters: hyper-parametric situation during training: the number of neurons is 200, the maximum number of iterations is 120, the learning rate of the first 80 is 0.005, and the learning rate of the following 40 is 0.2. Referring to fig. 12a, the abscissa of the coordinate system represents the corresponding time point, and the ordinate represents the true observed value and the predicted value of the argument of the near point, referring to fig. 12b, the abscissa of the coordinate system represents the corresponding time point, and the ordinate represents the error value between the true observed value and the predicted value of the argument of the near point. Fig. 12a and 12b show the results of predicting the argument of the perigee from 10 months and 11 days by using the LSTM neural network unit corresponding to the trained argument orbit parameters, and it can be seen that the maximum prediction error in the overall prediction effect of the argument of the perigee reaches about 0.4.

LSTM neural network units corresponding to the trained mean-anomaly orbital parameters: hyper-parametric situation during training: the number of neurons is 200, the maximum number of iterations is 120, the learning rate of the first 80 is 0.005, and the learning rate of the following 40 is 0.2. Referring to fig. 13a, the abscissa of the coordinate system thereof indicates the corresponding time point, and the ordinate indicates the true observed value and the predicted value of the mean-near-point angle, referring to fig. 13b, the abscissa of the coordinate system thereof indicates the corresponding time point, and the ordinate indicates the error value between the true observed value and the predicted value of the mean-near-point angle. Fig. 13a and 13b show the results of predicting the flat proximal angle from 10 months and 11 days by using the LSTM neural network unit corresponding to the trained flat proximal angle orbit parameter, which shows that the overall prediction effect of the flat proximal angle is better, and the maximum prediction error reaches about 0.6.

It should be noted that, although the steps are described in a specific order, the steps are not necessarily performed in the specific order, and in fact, some of the steps may be performed concurrently or even in a changed order as long as the required functions are achieved.

The present invention may be a system, method and/or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied therewith for causing a processor to implement various aspects of the present invention.

The computer readable storage medium may be a tangible device that retains and stores instructions for use by an instruction execution device. The computer readable storage medium may include, for example, but is not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing.

Having described embodiments of the present invention, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for precompensating for doppler frequency offset of satellite communication signals, comprising:

predicting a plurality of orbit parameters in the satellite orbit data at the current moment by using the trained neural network model, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, an argument of a near point and an argument of a near point;

predicting rising point right ascension parameters in satellite orbit data at the current moment by using a trained linear function module;

determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the perigee argument, the mean perigee angle and the ascension parameter of the ascending point obtained by prediction;

and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal.

2. The method of claim 1, wherein the step of pre-compensating for the doppler frequency offset of the down-converted signal corresponding to the received signal obtained by the mobile terminal from the satellite according to the relative position and relative velocity of the satellite with respect to the mobile terminal comprises:

acquiring position information and speed information of the mobile terminal and relative position and relative speed of the satellite relative to the mobile terminal when the mobile terminal obtains a received signal from the satellite so as to determine a pre-compensation value of Doppler frequency offset;

and performing Doppler frequency offset pre-compensation on a down-conversion signal corresponding to the received signal obtained by the mobile terminal based on the pre-compensation value of the Doppler frequency offset.

3. The method of claim 1, wherein the trained neural network model comprises a plurality of LSTM neural network elements, wherein each of the plurality of trajectory parameters corresponds to one LSTM neural network element.

4. The method of claim 3, wherein the LSTM neural network element for each of the plurality of trajectory parameters is trained by:

acquiring time sequence information corresponding to corresponding orbit parameters in satellite orbit data to form a first training set of the orbit parameters, wherein the time sequence information corresponding to the orbit parameters comprises a plurality of time points and the value of the orbit parameters at each time point;

training the LSTM neural network unit corresponding to the track parameter by using the first training set, and determining the estimated value of the track parameter at the next time point of the corresponding time point according to the value of the track parameter at the corresponding time point;

and calculating a loss value according to the estimated value of the track parameter at the next time point of the corresponding time point and the real value of the track parameter at the next time point of the time point recorded in the time sequence information corresponding to the track parameter, and updating the parameter of the LSTM neural network unit corresponding to the track parameter based on the loss value.

5. The method of claim 3, wherein the first training set for each of the plurality of orbit parameters is constructed by:

acquiring historical ephemeris data at preset time intervals to obtain satellite orbit data at different time points;

selecting the values of the orbit parameters in the satellite orbit data at different time points aiming at the corresponding orbit parameters of the orbit parameters to obtain an initial data set of the orbit parameters;

based on the threshold range set for the track parameter, eliminating data which is not in the threshold range in the initial data set of the track parameter to obtain a first training set of the track parameter, wherein each track parameter of a plurality of track parameters corresponds to one set threshold range.

6. The method of claim 3, wherein updating the parameters of the respective LSTM neural network element based on the total loss comprises: and calculating gradient values according to the total loss, and updating parameters of corresponding LSTM neural network units in a back propagation mode.

7. The method of claim 1, wherein the trained linear function module is trained in the following manner:

acquiring time sequence information corresponding to the ascension parameter of the ascending node in the satellite orbit data to form a second training set of the ascension parameter of the ascending node, wherein the time sequence information corresponding to the ascension parameter of the ascending node comprises a plurality of corresponding time points and values of the ascension parameter of the ascending node corresponding to the time points;

training a linear function module by using a second training set to determine an estimated value of the ascension parameter of the next time point of the corresponding time point according to the value of the ascension parameter of the corresponding time point;

and calculating an error based on the estimated value of the ascension point parameter of the next time point of the corresponding time point and the real value of the ascension point parameter of the next time point of the corresponding time point recorded in the time sequence information corresponding to the ascension point parameter, and updating the parameter of the linear function module based on the error.

8. The method of claim 7, wherein the second training set is constructed by:

selecting elevation intersection right ascension parameters in satellite orbit data at different time points to obtain an elevation intersection right ascension data set;

and based on the threshold range set for the ascending crossing point right ascension parameters, eliminating the ascending crossing point right ascension parameters which are not in the threshold range in the ascending crossing point right ascension data set to obtain a second training set.

9. The method of claim 7, wherein the linear function module predicts the ascension parameter at the intersection point using a least squares method.

10. The method according to claim 5 or 8, wherein the obtaining of the historical ephemeris data at predetermined time intervals to obtain satellite orbit data at different time points comprises:

acquiring satellite orbit data of a corresponding time point in historical ephemeris data, a real relative position and a real relative speed of a satellite relative to a mobile terminal at a preset time interval;

determining the relative position and the relative speed of the satellite at the corresponding time point by adopting an orbit extrapolation algorithm based on the satellite orbit data at the corresponding time point, and obtaining a total difference value based on the position difference between the determined relative position and the real relative position and the speed difference between the determined relative speed and the real relative speed;

and based on the total difference value corresponding to each time point, satellite orbit data of which the total difference value is greater than the time point corresponding to the preset threshold value are removed, and satellite orbit data corresponding to a time point in preset time before the time point are obtained, so that satellite orbit data of different time points are obtained.

11. The method according to one of claims 1 to 9, wherein calculating the relative position and relative velocity of the satellite with respect to the mobile terminal comprises determining based on the predicted semi-major axis, eccentricity, orbital inclination, perigee argument, mean perigee angle and ascension at ascending intersection parameters using an orbital extrapolation algorithm comprising: a two-body model based orbit extrapolation algorithm, an SGP4 orbit extrapolation algorithm, a J2 orbit extrapolation algorithm, and a J4 orbit extrapolation algorithm.

12. A system for precompensating for doppler frequency offset of satellite communication signals, comprising:

the trained neural network model is used for predicting a plurality of orbit parameters in the satellite orbit data of the current time point, wherein the plurality of orbit parameters comprise a semi-major axis, an eccentricity ratio, an orbit inclination angle, a near point argument and a near point argument;

the trained linear function module is used for predicting rising point right ascension parameters in the satellite orbit data at the current time point;

the orbit extrapolation module is used for determining the relative position and the relative speed of the satellite relative to the mobile terminal based on the semi-major axis, the eccentricity, the orbit inclination angle, the argument of the near place, the argument of the mean near point and the ascension parameter of the ascending intersection point obtained by prediction;

and the pre-compensation module is used for pre-compensating Doppler frequency offset of a down-conversion signal corresponding to a receiving signal obtained by the mobile terminal from the satellite according to the relative position and the relative speed of the satellite relative to the mobile terminal.

13. A computer-readable storage medium, on which a computer program is stored which is executable by a processor for carrying out the steps of the method according to any one of claims 1 to 11.

14. An electronic device, comprising:

one or more processors; and a memory, wherein the memory is to store executable instructions;

the one or more processors are configured to implement the steps of the method of any of claims 1-11 via execution of the executable instructions.