A long-term ephemeris generation method and device for intelligent terminal accelerated positioning

Abstract

The application discloses a long-acting ephemeris generation method and device for intelligent terminal accelerated positioning, and belongs to the technical field of satellite navigation and positioning. The method comprises the following steps: a server receives multi-source GNSS observation data, optimizes a reference station, and solves historical precise orbit and clock error; a dynamic model is used for long-acting ephemeris basic prediction, a historical prediction error vector is constructed per epoch, and a historical error input window is stacked; time domain trend characteristics and frequency domain periodicity characteristics of the historical error input window are extracted through a time-frequency domain attention mechanism; future error is fused and compensated to the basic prediction value to generate high-precision long-acting ephemeris parameters; the long-acting ephemeris parameters are compressed, encoded, integrity checked and adaptively broadcast; after the terminal receives and analyzes the long-acting ephemeris parameters, satellite position and satellite clock error are calculated, and accelerated positioning is completed. The application can improve the prediction accuracy of long-acting ephemeris, and improve the positioning availability in a weak network or no network scene.

Description

Technical Field

[0001] This invention belongs to the field of satellite navigation and positioning technology, and specifically relates to a long-term ephemeris generation method and device for accelerating positioning of smart terminals. Background Technology

[0002] Global Navigation Satellite Systems (GNSS) are the core means for smart terminals to achieve positioning, navigation, and timing functions. In standard point positioning mode, the terminal needs to directly demodulate the navigation message from the satellite signal to obtain the satellite ephemeris. Due to the low message broadcast rate, receiving a complete set of valid ephemeris usually takes a long time, especially when the terminal is in a cold start state, the initial positioning time is too long, which seriously affects the user experience.

[0003] To overcome this bottleneck, Assisted GNSS (A-GNSS) technology was proposed. This technology rapidly transmits auxiliary data to the terminal via cellular networks or wireless LANs, replacing the slow satellite message demodulation process, thus significantly shortening the time to first positioning. Currently, A-GNSS mainly includes online mode and offline mode. In online mode, the terminal acquires broadcast ephemeris with a short validity period in real time, requiring high stability of the network connection. The core of offline mode lies in ephemeris extension technology, which can be further divided into server-generated extended ephemeris (SGEE) and client-generated extended ephemeris (CGEE). The former generates long-term ephemeris from the cloud using reference station data and distributes it to the terminal, while the latter allows the terminal to autonomously generate short-term forecasts based on locally stored historical ephemeris using simplified algorithms.

[0004] However, existing technologies still have significant shortcomings in scenarios with no network or weak signal. First, online mode heavily relies on a continuous and stable network connection. Once users enter signal dead zones such as mountainous areas or wilderness, the terminal cannot acquire or update ephemeris data in a timely manner, and the initial positioning time will quickly degrade to a cold start state. Although server-generated ephemeris extensions alleviate some of the problems by pre-installing ephemeris data, their initial download and periodic updates still require a network connection. If users do not update the ephemeris data before traveling or spend more than the ephemeris's validity period outdoors, the positioning capability will also degrade. Second, the accuracy of ephemeris forecasts naturally decreases as the validity period extends. The accuracy of existing ephemeris extension technologies is limited by the quality of historical data and the accuracy of orbital models. Errors may accumulate or even diverge in long-term forecasts, rendering the ephemeris unusable in later stages of the forecast and making it difficult to support high-precision applications in long-term offline scenarios. Furthermore, smart terminals are limited in terms of power, storage, and computing power. Existing technologies often employ complex dynamic models to improve forecast accuracy, requiring a large amount of historical data and iterative calculations on the terminal side, resulting in significant power consumption and time consumption. Even with server-generated solutions, long-term ephemeris data spanning multiple days across the entire constellation still occupies considerable storage space. In addition, current ephemeris updates largely rely on manual operation or passive synchronization, lacking the ability to adaptively adjust based on network conditions or user scenarios. Users are prone to forgetting to update or being unable to update in offline environments, leading to expired ephemeris data and impacting the continuity of location services and user experience.

[0005] Furthermore, existing long-term satellite orbit and clock bias generation schemes typically include steps such as reference station selection, fused data processing, orbit prediction, clock bias prediction, correction broadcasting, and terminal adaptation. Their focus is on reducing the amount of broadcast data through orbital dynamics smoothing, clock bias prediction models, and correction fitting relative to the broadcast ephemeris. At the terminal side, a simplified orbit and clock bias predictor is used to generate short-term backup ephemeris. These schemes still primarily rely on mechanical models, clock bias statistical models, and terminal-side adaptive processing. They lack an attention error compensation mechanism that jointly models the temporal trend characteristics and frequency domain periodic characteristics of the historical error input window formed by the chronological stacking of historical prediction error vectors. They also lack a unified encoding, packet transmission, and missing packet retransmission mechanism for long-term ephemeris parameters for weak network communication scenarios of smart terminals.

[0006] Therefore, there is an urgent need for a long-term ephemeris generation and broadcasting mechanism that can extend the validity period of ephemeris while ensuring forecast accuracy, adapting to terminal resource constraints, reducing network dependence, and achieving intelligent updates. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a long-term ephemeris generation method and apparatus for accelerating positioning of smart terminals. It uses a time-frequency domain attention neural network to compensate for dynamic forecast errors to generate high-precision long-term ephemeris. Furthermore, it performs lightweight compression encoding on the generated long-term ephemeris parameters and adaptively selects a broadcast strategy based on network conditions to adapt to weak or no-network scenarios.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A long-term ephemeris generation method for accelerating positioning on smart terminals, the method comprising:

[0010] Step 1: The server receives multi-source GNSS data via the Internet, selects the best combination of high-quality reference stations, and calculates the historical precise orbit and clock error as reference values.

[0011] Step 2: Based on the historical precise orbit and clock error, perform basic predictions of long-term ephemeris using a dynamic model, and construct historical prediction error vectors for each epoch, stacking them in chronological order to form a historical error input window; input the historical error input window and satellite auxiliary state features into a time-frequency domain attention neural network, extract the time-domain trend features and frequency-domain periodic features of the historical error input window through the time-frequency domain attention mechanism, fuse them to predict future prediction errors, and compensate the future prediction errors into the results of the basic prediction to generate a corrected long-term ephemeris;

[0012] Step 3: Compress and encode the corrected long-term ephemeris parameters, and adaptively select a broadcast strategy based on the network status and local cache status of the terminal.

[0013] Step 4: After receiving and parsing the long-term ephemeris parameters, the terminal calculates the satellite position and satellite clock error, and uses the satellite position and satellite clock error as the terminal positioning calculation input to complete accelerated positioning.

[0014] Furthermore, in step 2, constructing the historical forecast error vector for a single historical epoch includes:

[0015] Using the historical precise orbit and clock error as reference values, the position error vector between the satellite reference position in the reference values ​​and the satellite forecast position obtained from the basic forecast is calculated in the geocentric Earth-fixed coordinate system. The position error vector is then projected onto the radial, trace, and normal directions to obtain the orbital error component. Based on the difference between the satellite reference clock error in the reference values ​​and the satellite forecast clock error obtained from the basic forecast, the clock error equivalent distance error component is obtained. The orbital error component and the clock error equivalent distance error component together constitute the historical forecast error vector for a single historical epoch.

[0016] Furthermore, in step 2, the time-frequency domain attention neural network includes a time-domain feature branch, a frequency-domain feature branch, an auxiliary feature branch, a feature fusion layer, and a prediction output layer;

[0017] The temporal feature branch uses a long short-term memory network to process the historical error input window, and extracts the temporal trend features and short-term correlation features of the historical error input window through input gate, forget gate, output gate and cell state;

[0018] The frequency domain feature branch performs a Fourier transform on the historical error input window and converts the complex frequency domain components obtained after the transform into real-valued frequency domain features.

[0019] The auxiliary feature branch encodes the satellite auxiliary state features and maps them through a fully connected layer to obtain auxiliary features;

[0020] The feature fusion layer fuses the time trend feature, the short-term correlation feature, the real-valued frequency domain feature, and the auxiliary feature to obtain the fused feature;

[0021] The prediction output layer outputs orbital error estimates and clock error equivalent distance estimates for multiple future forecast epochs based on the fusion features and the preset future forecast age.

[0022] Furthermore, in step 2, compensating the future forecast error into the result of the basic forecast includes:

[0023] After converting the orbital error estimate output by the prediction output layer back to the geocentric Earth-fixed coordinate system from the radial, track, and normal directions, it is added back to the satellite prediction position obtained from the basic prediction to obtain the corrected satellite position.

[0024] The estimated equivalent distance error of the clock error output by the prediction output layer is divided by the speed of light and then added back to the satellite prediction clock error obtained from the basic prediction to obtain the corrected satellite clock error.

[0025] The corrected long-term ephemeris parameters are composed of the corrected satellite position and the corrected satellite clock error.

[0026] Furthermore, the corrected long-term ephemeris parameters include common head parameters, coordinate reference frame and time system parameters, satellite orbit parameters, satellite clock error parameters, mass and integrity parameters, and optional error compensation parameters;

[0027] The common header parameters include protocol version, product batch number identifier, generation time, validity period start and end, and check code;

[0028] The coordinate reference frame and time system parameters include coordinate reference frame identifier, time system identifier, week number, week seconds, and reference epoch;

[0029] The satellite orbital parameters are expressed using piecewise Kepler fitting parameters, piecewise polynomial coefficients, or Chebyshev coefficients.

[0030] The satellite clock bias parameters are expressed using piecewise fitting coefficients or periodic term coefficients.

[0031] The quality and integrity parameters include satellite health status and accuracy level;

[0032] The optional error compensation parameters include orbital residuals and clock error residuals, which are used to compensate for the remaining systematic errors over the long forecast period.

[0033] Furthermore, in step 3, the adaptive selection of the broadcast strategy based on the network status and local cache status of the terminal includes:

[0034] When the terminal network status is broadband connection, the server transmits the complete long-term ephemeris parameters in one go.

[0035] When the terminal network is in a narrowband or weak network connection state, the server splits the long-term ephemeris parameters into multiple data packets according to satellites and effective time periods, and transmits them using small packet fragmentation, low-frequency transmission, breakpoint resumption, and missing packet retransmission mechanisms; the terminal performs out-of-order reassembly, missing packet detection, and integrity verification based on the sequence number and cyclic redundancy check code in each data packet, and requests the missing data packets from the server only when the network recovers;

[0036] When the terminal has no network connection, the server suspends the delivery, and the terminal uses the long-term ephemeris parameters that have been cached locally, are still valid, and have passed the integrity check.

[0037] Furthermore, the method also includes:

[0038] Before publishing the corrected long-term ephemeris parameters, the server divides the training set, validation set, and test set according to the satellite or forecast period, and evaluates the time-frequency domain attention neural network using orbital radial error, orbital three-dimensional error, clock error equivalent distance error, or spatial signal ranging error as evaluation indicators.

[0039] When the evaluation results meet the preset accuracy threshold and integrity threshold, the corrected long-term ephemeris parameters are published.

[0040] When the evaluation results do not meet the preset accuracy threshold or integrity threshold, strategies such as reducing the forecast validity period, reverting to the basic forecast results of the dynamic model, or using the previous version of the valid product are adopted.

[0041] On the other hand, the present invention provides a long-lasting ephemeris generation device for accelerating positioning of smart terminals, comprising:

[0042] The calculation module is used to set up the server to receive multi-source GNSS data via the Internet, select the best combination of high-quality reference stations, and calculate the historical precise orbit and clock error as reference values.

[0043] The compensation module is used to perform basic predictions of long-term ephemeris based on the historical precise orbit and clock error using a dynamic model, and to construct historical prediction error vectors for each epoch, stacking them in chronological order to form a historical error input window; the historical error input window and satellite auxiliary state features are input into a time-frequency domain attention neural network, and the time-frequency domain attention mechanism is used to extract the time-domain trend features and frequency-domain periodic features of the historical error input window, fuse them to predict future prediction errors, and compensate the future prediction errors into the results of the basic prediction to generate a corrected long-term ephemeris;

[0044] The broadcast module is used to compress and encode the corrected long-term ephemeris parameters and adaptively select a broadcast strategy based on the network status and local cache status of the terminal.

[0045] The positioning module is used to set the terminal to receive and parse the long-term ephemeris parameters, calculate the satellite position and satellite clock error, and use the satellite position and satellite clock error as the terminal positioning calculation input to complete accelerated positioning.

[0046] Thirdly, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned long-term ephemeris generation method for accelerated positioning of smart terminals.

[0047] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned long-term ephemeris generation method for accelerated positioning of smart terminals.

[0048] The beneficial effects of this invention are as follows:

[0049] Long-term ephemeris offers extended validity and controllable forecast accuracy. This application, building upon server-side calculation of historical precise orbits and clock errors using multi-source GNSS data, further utilizes a time-frequency domain attention mechanism to model and compensate for errors in the dynamical basis forecast. This mechanism simultaneously extracts the temporal trend and frequency-domain periodicity characteristics of the error sequence, which helps mitigate the increasing trend of ephemeris errors with forecast duration. Under example verification conditions, for a 28-day forecast age, the three-dimensional orbital error of the dynamical basis forecast is typically in the tens to hundreds of meters range; after error compensation by this application, the three-dimensional orbital error can be controlled within the tens of meters range, and the clock error equivalent distance error can be controlled within the tens of meters range, thereby improving the availability of long-term ephemeris over long forecast ages. This is suitable for rapid acquisition and assisted positioning by smart terminals, as well as cached positioning scenarios under weak or no network conditions.

[0050] The positioning timeliness is good in weak network or no network environments. This application adopts a unified encoding of long-term ephemeris parameters and a network status adaptive broadcasting mechanism. Under weak network conditions where communication is possible, it uses small packet fragmentation, breakpoint resumption, and missing packet retransmission to enable the terminal to obtain the satellite orbit parameters and satellite clock bias parameters required for positioning calculation as soon as possible. Under completely no network conditions, the terminal uses cached long-term ephemeris parameters that are still valid to complete the positioning, which significantly enhances the positioning availability and response speed in offline scenarios.

[0051] This application minimizes terminal resource consumption and lowers deployment barriers. It deploys highly complex tasks such as precise orbit determination, orbit prediction, and error compensation on the server side, requiring only lightweight verification, parsing, and fusion processing on the terminal. This avoids the high power consumption and computational overhead associated with running complex algorithms on the terminal. Furthermore, by compressing and encoding long-term ephemeris parameters and transmitting them in packets according to satellite and effective time periods, the amount of data required for storage and transmission on the terminal side is significantly reduced, making it more suitable for deployment on resource-constrained devices such as smartphones and smartwatches.

[0052] It offers strong service continuity and good adaptability to various scenarios. This application features integrity verification, retransmission upon failure, and intelligent incremental updates. It can automatically replenish missing long-term ephemeris parameter data packets when the network is briefly restored, preventing positioning capability degradation due to ephemeris expiration. It is suitable for long-term weak or no-network scenarios such as outdoor sports, emergency rescue, and field operations, and has high engineering practical value. Attached Figure Description

[0053] Figure 1 This is a flowchart of a long-term ephemeris generation method for accelerating positioning of smart terminals according to the present invention.

[0054] Figure 2 This is a flowchart of the long-term ephemeris error compensation prediction method based on the time-frequency domain attention mechanism of the present invention;

[0055] Figure 3This is a schematic diagram of a time-frequency domain attention error compensation neural network structure according to the present invention;

[0056] Figure 4 This is a flowchart of the long-term ephemeris parameter encoding and adaptive broadcasting process of the present invention. Detailed Implementation

[0057] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0058] This invention provides a long-term ephemeris generation method for accelerated positioning of smart terminals. Through a cloud-edge collaborative architecture, it solves the problem of terminals struggling to obtain effective ephemeris in a timely manner under weak network or no network conditions. The core lies in the server-side generation of long-term ephemeris, combined with compression encoding, integrity verification, and adaptive broadcasting mechanisms, enabling the terminal to shorten its initial positioning time when it has cached effective long-term ephemeris and can track satellite signals. Specifically, as shown... Figure 1 As shown, the method includes:

[0059] Step 1: Server-side reference value construction. The server receives multi-source GNSS observation data via the Internet, selects the best combination of high-quality reference stations, and calculates the historical precise orbit and clock error as the reference value;

[0060] The server-side architecture includes a data acquisition and processing module and a reference station selection strategy module. The server aggregates raw observation data from a global GNSS reference station network via the internet, including pseudorange, carrier phase, and navigation messages from multiple systems (GPS / BDS / Galileo / GLONASS). After receiving the observation data, it undergoes preprocessing and integrity checks, specifically including format standardization, time system unification, coordinate reference frame unification, gross error detection, cycle slip detection, and data integrity checks. Subsequently, the reference station selection strategy module dynamically selects a set of reasonably distributed and highly stable reference stations based on service area selection principles, satellite geometry, and reference station health monitoring (such as clock slips and signal loss rate), outputting a high-quality dataset for subsequent processing. Selection principles include prioritizing data from continuously operating stations with a data integrity rate greater than 98% to improve the reliability of orbit determination.

[0061] The precise orbit and clock error determination module uses numerical integration and mechanical models (such as the Earth's gravitational field model and solar radiation pressure model) to perform precise orbit determination and clock error estimation on the preprocessed observation data, and calculates a high-precision historical ephemeris (including orbit and clock error). This process requires the fusion of multi-day observation data to calibrate key parameters (such as the radiation pressure coefficient), and the generated precise product serves as a reference value for long-term ephemeris forecasts and error sequence construction.

[0062] Step 2: Basic Forecast and Time-Frequency Domain Error Compensation. Based on the historical precise orbits and clock errors, a long-term ephemeris basic forecast is performed using a dynamic model extrapolation. Historical forecast error vectors are constructed epoch-by-epoch and stacked chronologically to form a historical error input window. The time-domain trend features and frequency-domain periodic features of the historical error input window are extracted using a time-frequency domain attention mechanism to predict future forecast errors. These future forecast errors are then compensated into the basic forecast results according to the error definition, generating a corrected long-term ephemeris. The specific implementation process is as follows: Figure 2 As shown:

[0063] The orbit and clock error prediction module first acquires historical precise ephemeris (historical precise orbit and clock error products) generated by the precise orbit and clock error determination module. Using this as a high-precision reference value, it extrapolates using a precise orbital dynamics model (considering perturbations such as Earth's non-spherical gravity, lunar and solar gravity, and solar radiation pressure) to generate a basic ephemeris prediction value for a predetermined long-term validity period. Subsequently, the basic prediction values ​​generated at different times are compared with the corresponding historical precise ephemeris values, and the historical prediction error vector (including orbital error and clock error) is calculated epoch-by-epoch. These vectors are then stacked according to a predetermined window length to form a historical error input window, thereby constructing a sample dataset for training the error compensation model.

[0064] Construct a deep learning network specifically for error sequence prediction. The network takes a window of historical error input as input and its processing flow comprises two parallel and fused paths:

[0065] In one alternative embodiment, the structure of the time-frequency domain attention error compensation neural network is as follows: Figure 3 As shown. Using historical precision orbits and clock difference products as reference values, the basic forecast values ​​for the corresponding historical epochs are obtained using a dynamic model. Historical forecast error vectors are constructed epoch by epoch according to formula (1), and the historical forecast error vectors of multiple historical epochs are stacked in chronological order to form a historical error input window, which serves as the main input to the neural network:

[0066] (1)

[0067] The historical error is constructed by subtracting the dynamic fundamental prediction value from the aforementioned reference value. The orbital error is first subtracted in the geocentric-geofixed coordinate system (ECEF) and then projected onto the RTN coordinate system consisting of radial (Radial, R), tangential (Tangential, T), and normal (Normal, N). The clock error is uniformly represented as an equivalent distance in meters.

[0068] in: Indicates the satellite number; Indicates the first A historical era, Here, n is the total number of epochs in the input window; Indicates the first a satellite in The historical prediction error vector for a single historical epoch at any given time; , , These represent the radial, track, and normal trajectory error components, respectively, in meters; This represents the clock error equivalent distance error, expressed in meters. Indicates the historical precision orbit reference position; Indicates the location of the basic dynamic prediction; This represents the position error vector in the ECEF coordinate system. This represents the direction cosine matrix projected from the ECEF coordinate system to the RTN coordinate system; This indicates a precision clock error reference value, in seconds. This indicates the basic forecast clock difference, in seconds; Represents the speed of light; This represents the historical error input window; Let n represent the space of real numbers in rows of 4 columns.

[0069] The neural network input specifically includes: the historical error input window. (The historical forecast error vector of a single historical epoch, defined by formula (1), is obtained by stacking the historical forecasts in chronological order across n historical epochs) and satellite auxiliary state features. The error vector of each epoch includes radial, track, and normal orbital error components, as well as clock error equivalent distance components; satellite auxiliary state features include constellation identifier, satellite number, forecast age, solar radiation pressure related parameters, shadow status, orbital surface parameters, and clock type. The input window can be adjusted according to the sampling interval and forecast validity period. For example, 96 15-minute epochs can be used as a 1-day historical window, or it can be expanded into a multi-day sliding window; the basic input dimension of each epoch is 4-dimensional, corresponding to radial, track, normal, and clock error equivalent distance errors, respectively.

[0070] The neural network comprises a temporal feature branch, a frequency domain feature branch, an auxiliary feature branch, a feature fusion layer, and a prediction output layer. The temporal feature branch extracts the trend term and short-term correlation of historical errors; the frequency domain feature branch extracts periodic error features related to daily period, orbital period, and perturbation; and the auxiliary feature branch introduces prior knowledge of the satellite state. The feature fusion layer performs weighted fusion of temporal, frequency, and auxiliary features, and the prediction output layer outputs the orbital error and clock error equivalent distance error for multiple future forecast ages.

[0071] In an exemplary network parameter configuration, the temporal branch may include a normalization layer, a one-dimensional convolutional layer, and an LSTM (Long Short-Term Memory) layer; the frequency branch only performs Fast Fourier Transform (FFT) or Short-Time Fourier Transform (STFT) on the error sequence within the historical window, without introducing future precision products or future error labels, and extracts the real-valued features of the effective frequency components; the auxiliary feature branch may use a fully connected mapping. The fusion prediction layer performs attention weighting or splicing fusion on the three types of features, and outputs the error estimates for multiple forecast epochs within a preset long validity period. The network example structure is shown in Table 1, and the core mathematical relationships are shown in formulas (2) to (5).

[0072] Table 1

[0073]

[0074] The structures and parameters listed in Table 1 are merely examples and do not limit the scope of protection. Those skilled in the art can adjust the sequence length, number of filters, number of LSTM layers, number of neurons, and number of output epochs according to the sampling interval, satellite system, historical data length, and the required validity period of the terminal. The core technology lies in using the historical error input window as input (each historical epoch element is a historical forecast error vector) and predicting future forecast errors through the fusion of time-domain gated memory, frequency-domain periodic features, and auxiliary state priors.

[0075] Regarding formula (2): Time-domain LSTM gating:

[0076] (2)

[0077] The time trend and memory information in the historical error input window are extracted using the input gate, forget gate, output gate, and candidate cell state of an LSTM. (Note: Subscripts are also included.) This indicates the epoch number within the historical error input window. ; Indicates the first Each historical epoch input typically consists of a standardized historical forecast error vector for a single historical epoch. constitute; Indicates the input gate; Indicates the Gate of Oblivion; Indicates the output gate; Indicates the state of candidate cells; Indicates the current cell state; Indicates the cell state of the previous epoch; Indicates the current hidden state; Indicates the hidden state of the previous epoch; This represents the weight matrix input to each gate unit; This represents the weight matrix from the hidden state to each gate unit; Indicates the bias term; Represents the Sigmoid function; Represents the hyperbolic tangent function; This represents Hadamard element-wise multiplication.

[0078] Regarding formula (3): Real-valued characteristics in the frequency domain:

[0079] (3)

[0080] Performing FFT or STFT on the historical error input window results in a complex number in the frequency domain. Before entering the neural network, it should be converted into real-valued features such as real / imaginary parts or amplitude / phase.

[0081] in: This represents the historical error input window defined by formula (1); : Perform FFT or STFT transformation operators on the historical error input window; Indicates the first One frequency component; Indicates the number of effective frequency components selected; Indicates the first Historical error input window for each satellite After FFT or STFT at frequency Complex frequency domain components; Represents the real-valued frequency domain characteristics of the frequency domain branch; Represents the real part of a complex number; Represent the imaginary part of a complex number; Indicates amplitude; Indicates phase.

[0082] Regarding formula (4): Attention fusion:

[0083] (4)

[0084] Temporal attention assigns weights to different historical moments, while frequency domain attention assigns weights to different frequency components. These are then fused with auxiliary features to obtain a unified feature. Among these: Indicates the first Temporal attention weights for each historical epoch; Represents the temporal attention scoring function; Indicates the length of the historical error input window; The expression obtained by formula (2) is the first One hidden state; Represents temporal attention features; Indicates the first Frequency domain attention weights for each frequency component; This represents the frequency domain attention scoring function; Indicates the number of effective frequency components; The first expression obtained by formula (3) A real-valued frequency domain feature; Represents frequency domain attention features; Indicates the first Auxiliary status characteristics of each satellite; The encoding mapping function representing the auxiliary feature branch; This represents the auxiliary features obtained from satellite auxiliary state features; This indicates the unified characteristics after fusion; Represents the weight matrix of the fusion layer; Indicates the fusion layer bias term; This represents the nonlinear mapping of the fusion layer; This indicates feature splicing.

[0085] Regarding formula (5): Output of future multi-step forecast error:

[0086] (5)

[0087] Output the error estimates for multiple future forecast ages.

[0088] in: Indicates the first Satellites in the future The prediction error estimation vector; Indicates the start time of the current forecast; subscript Indicates the future output epoch number; Indicates the first One future predicted age; Indicates the predicted output layer; This represents the fusion feature obtained by formula (4); Indicates the number of output epochs within the preset long validity period; : Estimated value of the equivalent distance error of future clock bias, in meters.

[0089] Regarding formula (6): Track and clock error compensation:

[0090] (6)

[0091] Since the error in formula (1) is defined as "reference value minus base forecast value", the prediction error should be added back to the base forecast result during compensation. After the orbital error is predicted in the RTN coordinate system, it needs to be converted back to the ECEF coordinate system before being compensated for to the satellite position; if the clock error is directly expressed in seconds, the clock error compensation term is no longer divided by the speed of light c.

[0092] in: Indicates the compensated satellite position; Indicates the location of the basic dynamic prediction; This represents the transpose direction cosine matrix used to transform from the RTN coordinate system back to the ECEF coordinate system. , , These represent the radial, trace, and normal trajectory errors predicted by the neural network, respectively, in meters. This indicates the compensated satellite clock bias, expressed in seconds. This represents the clock difference in the fundamental dynamics prediction, in seconds. This represents the clock error equivalent distance error predicted by the neural network, in meters. Represents the speed of light; Indicates the time of future compensation.

[0093] To ensure consistency in training, validation, and online inference, the server standardizes the historical forecast error vector of a single historical epoch in the historical error input window according to satellite, coordinate components, and clock error components, and saves the scale parameters corresponding to the product version. The training set, validation set, and test set are isolated by time period or satellite set to avoid adjacent samples of the same forecast arc appearing in the training set and test set at the same time, thereby reducing the risk of overfitting and data leakage.

[0094] The estimated value of the future forecast error is output according to formula (5), and the corrected long-term ephemeris parameters are obtained according to formula (6). Since the historical error is defined as "reference value minus basic forecast value" according to formula (1), the prediction error should be added back to the basic forecast result during compensation. During the training phase, weighted mean square error, orbital radial error, orbital three-dimensional error, clock error equivalent distance error, or signal-in-space range error (SISRE) can be used as targets or evaluation indicators to improve the stability of long-term forecast error prediction.

[0095] After the network outputs a future forecast error sequence, the server compensates for this error sequence by incorporating it into the dynamical basis forecast results, and then further fits or encodes it into long-term ephemeris parameters with a preset long validity period. The terminal receives the encoded long-term ephemeris parameters, not the neural network model itself or historical precision product files.

[0096] To avoid terminal positioning errors caused by abnormal neural network predictions, the server performs model training and quality evaluation before releasing long-term ephemeris parameters: training, validation, and test sets are divided according to satellites, constellations, or forecast periods; the model is evaluated using orbital radial error, orbital three-dimensional error, clock error equivalent distance error, SISRE, or a combination thereof; when the evaluation results meet the preset accuracy threshold, integrity threshold, and abnormal satellite removal conditions, the corresponding product is released; when the evaluation results do not meet the thresholds, the model or parameters are adjusted, including reducing the validity period, reverting to dynamical basis forecast results, or using the previous version of the valid product, and retraining and re-evaluating are performed.

[0097] This data-driven approach helps compensate for systematic errors that are difficult to accurately model using purely mechanical models. Under selected training data, validation data, and release thresholds, orbital errors over a pre-defined long-term validity period can be controlled within a pre-defined accuracy level, for example, reaching tens of meters in a 28-day forecast period. Specific accuracy depends on constellation type, observation data quality, forecast age, solar activity, and the coverage of model training samples.

[0098] Step 3: Lightweight Encoding and Adaptive Broadcasting of Long-Term Ephemeris Parameters. The corrected long-term ephemeris parameters are lightweight compressed and encoded, and a broadcasting strategy is adaptively selected based on the terminal's network status and local cache status. The long-term ephemeris parameters include common header parameters, coordinate reference frame and time system parameters, satellite orbit parameters, satellite clock bias parameters, quality and integrity parameters, and optional error compensation parameters. After receiving the parameters, the terminal can calculate the satellite position and satellite clock bias under a unified coordinate reference frame and time system.

[0099] The server efficiently distributes the corrected long-term ephemeris parameters to the terminals through data compression and broadcasting modules. For example... Figure 4 As shown, in order to adapt to the narrowband communication environment and limited storage space of smart terminals, this step includes four stages: long-term ephemeris parameter expression, unified encoding, packet transmission, and terminal integrity verification.

[0100] The long-term ephemeris parameter expression can be functionally divided into common header and version fields, coordinate reference frame and time system fields, satellite identifier and health quality fields, satellite orbit expression fields, satellite clock bias expression fields, signal deviation and inter-system time auxiliary fields, and optional fitting or residual correction fields. Specific field examples and usage constraints are shown in Table 2. It should be noted that the long-term ephemeris parameters in this invention are not the original navigation messages directly broadcast by the satellite, nor are they historical precision product files. Instead, they are long-term valid orbit and clock bias expression parameters generated by the server based on historical precision orbits and clock biases, and compensated for by time-frequency domain attention errors. The orbit expression parameters in Table 2 are not limited to traditional broadcast ephemeris Kepler elements. In long-term valid scenarios, piecewise Kepler fitting parameters, piecewise polynomial coefficients, Chebyshev coefficients, or equivalent residual correction parameters can also be used, as long as the terminal can recover the satellite position and satellite clock bias within the specified valid time period. In multi-constellation scenarios, the coordinate reference frame, time system, week number / second within week, and inter-navigation system time bias identifiers should be clearly defined to ensure the consistency of terminal positioning calculations.

[0101] Table 2

[0102]

[0103] During encoding, the server uses "satellite - valid time period" as the smallest independently parseable unit to uniformly encode and packetize the long-term ephemeris parameters. Each data packet contains necessary fields from the product common header or packet control field, orbital clock error parameter area, coordinate reference frame and time system field, and quality and integrity field. The terminal performs out-of-order reordering, missing packet detection, and availability judgment based on the product version number or product IOD, coordinate reference frame, time system, validity period, packet sequence number, and CRC check result; where the product IOD is the batch identifier of this long-term ephemeris product and is not equivalent to the IODE or IODC parameters in the original satellite navigation message.

[0104] During broadcasting, the server can transmit complete long-term ephemeris parameters in one go under broadband network; when communication is still possible under narrowband or weak network conditions, the server uses small packet fragmentation, low-frequency transmission, breakpoint resumption, and missing packet retransmission; when there is no network at all, the server cannot transmit data in real time, and the terminal only uses long-term ephemeris parameters that have been cached locally and are still within the validity period and have passed integrity verification.

[0105] The above processing enables the terminal to parse the satellite orbit parameters and satellite clock bias parameters required for positioning calculation according to a unified message format, and to perform availability judgment by combining the product IOD, time system, validity period and integrity verification results, thereby reducing the complexity of terminal implementation and improving data consistency in weak network scenarios.

[0106] Intelligent broadcasting and progressive transmission: The server dynamically adjusts its broadcasting strategy based on real-time monitoring of the terminal network status (such as bandwidth and latency). During data transmission, the server splits the long-term ephemeris parameters into several data packets according to satellites and valid time periods, and configures each data packet with a sequence number, total number of packets, message type, and CRC checksum. The terminal determines whether there are missing or out-of-order packets based on the sequence number; if there are missing packets, the missing sequence number is recorded, and only the missing data packets are resent when the network recovers or in the next request.

[0107] Step 4: After receiving and parsing the long-term ephemeris parameters, the terminal completes accelerated positioning.

[0108] After receiving data broadcast by the server, the terminal module first classifies and caches the data according to the message type, satellite identifier, time period identifier, and version number in the packet header, and then performs CRC verification and sequence reconstruction. When the orbital parameters, clock bias parameters, coordinate reference frame, and time system information required for positioning calculation are complete and valid, the terminal uses the long-term ephemeris parameters to calculate the satellite position and satellite clock bias, and performs positioning calculation. If the data packet is missing, verification fails, the version is inconsistent, the validity period has expired, or the satellite health status is abnormal, the terminal does not use the corresponding satellite data for positioning, and requests the missing or updated data packet from the server when the network is available; when there is no network, the terminal only uses the locally cached long-term ephemeris parameters that are still valid.

[0109] Example:

[0110] Taking the outdoor use of smart wearable devices as an example:

[0111] The server generates long-term ephemeris parameters with a preset long validity period daily or according to a preset update cycle, and maintains forecast accuracy through the error compensation method and quality control method in step 2; in one embodiment, the preset long validity period can be configured to 28 days. To illustrate the feasibility of this embodiment, a set of server forecast accuracy verification data and terminal communication positioning verification data are given below. The listed values ​​are example results obtained under specific datasets, device conditions, and parameter configurations, and are not intended to limit the scope of protection of this invention.

[0112] Test Example 1: Server-side forecast error compensation verification. A continuous 90-day historical precise orbit and clock error product was selected as the training sample. Historical forecast error vectors were established epoch-by-epoch with 15-minute sampling, and stacked according to a preset window length to form a historical error input window. The subsequent 14 days of data were used as the validation set, and then the next 14 days of data were used as the test set. The satellites participating in the test were GPS, BDS, and Galileo satellites with normal health status and data integrity rate not less than 95%. The precise orbit and clock error product of the test set was used as a verification reference, and the orbital 3D RMS and clock error equivalent distance RMS were used for evaluation, resulting in the example data shown in Table 3.

[0113] Table 3

[0114]

[0115] As shown in Table 3, under this example condition, time-frequency domain attention error compensation can reduce orbital error and clock error equivalent distance error over a long forecast period. When a satellite has abnormal errors, abnormal health status, or exceeds the preset SISRE threshold in the test set, the server marks the corresponding time period of the satellite as unavailable or reduces the validity period before releasing it.

[0116] Test Example 2: Terminal Weak Network Broadcasting and Cache Location Verification. The server organizes the 28-day long-term ephemeris parameters of 42 satellites (GPS, BDS, and Galileo) according to "satellite-effective time period," quantizes and compresses them, and adds CRC checksums to form data packets. Each packet has a payload of 1024 bytes, the complete product size is approximately 0.86MB, and the terminal occupies approximately 1.05MB of storage space after caching the index and integrity information. The communication and positioning process is verified using a smart wearable device prototype, and the example data shown in Table 4 is obtained.

[0117] Table 4

[0118]

[0119] The initial positioning time in Table 4 is calculated when the terminal can receive GNSS signals, has a rough time or can obtain the time from the received signal, and the number of satellites participating in the positioning meets the positioning calculation conditions. The download time in Table 4 is matched with the product size, link rate, and retransmission mechanism. TTFF reflects the acceleration effect when the cached long-term ephemeris participates in positioning. If the terminal is in a deeply obstructed indoor environment, the satellite signal is untrackable, or the cached product integrity verification fails, the failure parameter is not used, and the failure degradation strategy is followed. The absence of a valid cache reference only represents the empirical range of a normal cold start; the specific time is affected by navigation message demodulation, satellite signal strength, and receiver strategy.

[0120] Before traveling, smart wearable devices receive long-term ephemeris parameter data packets via Wi-Fi. If the locally cached long-term ephemeris parameters are still valid and pass the consistency checks of CRC, product IOD, coordinate reference frame, and time system, the terminal can directly use local parameters to calculate satellite positions and clock errors when it is able to track GNSS satellite signals and meet the observation conditions for the number of satellites required for positioning, thereby shortening the first positioning time.

[0121] If a user stays for more than the preset update cycle (e.g., 7 days), the smart wearable device will automatically request the missing or updated long-term ephemeris parameter data packet via Bluetooth when the user connects to the smartphone during exercise breaks or when the cellular network is restored. If the network is unavailable, the device will continue to use the cached data that is still valid. If the cached data expires, the version is inconsistent, the coordinate reference frame or time system is inconsistent, or the integrity check fails, the device will prompt that the positioning capability may be degraded and will not use the corresponding invalid parameters.

[0122] Through the above-described embodiments and examples, this invention provides a complete process for server-side error compensation, product quality control, data packet retransmission, terminal cache verification, and location degradation processing, which can solve the technical problems of limited terminal resources, difficulty in updating in weak networks, and insufficient continuity of location without network cache.

[0123] On the other hand, the present invention provides a long-lasting ephemeris generation device for accelerating positioning of smart terminals, the various modules of which can implement the various steps of the aforementioned method, specifically including:

[0124] The calculation module is used to set up the server to receive multi-source GNSS data via the Internet, select the best combination of high-quality reference stations, and calculate the historical precise orbit and clock error as reference values.

[0125] The compensation module is used to perform basic predictions of long-term ephemeris based on the historical precise orbit and clock error using a dynamic model, and to construct historical prediction error vectors for each epoch, stacking them in chronological order to form a historical error input window; the historical error input window and satellite auxiliary state features are input into a time-frequency domain attention neural network, and the time-frequency domain attention mechanism is used to extract the time-domain trend features and frequency-domain periodic features of the historical error input window, fuse them to predict future prediction errors, and compensate the future prediction errors into the results of the basic prediction to generate a corrected long-term ephemeris;

[0126] The broadcast module is used to compress and encode the corrected long-term ephemeris parameters and adaptively select a broadcast strategy based on the network status and local cache status of the terminal.

[0127] The positioning module is used to set the terminal to receive and parse the long-term ephemeris parameters, calculate the satellite position and satellite clock error, and use the satellite position and satellite clock error as the terminal positioning calculation input to complete accelerated positioning.

[0128] Thirdly, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned long-term ephemeris generation method for accelerated positioning of smart terminals.

[0129] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned long-term ephemeris generation method for accelerated positioning of smart terminals.

[0130] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A long-term ephemeris generation method for accelerating positioning on smart terminals, characterized in that, The method includes: Step 1: The server receives multi-source GNSS data in real time via the Internet, selects the best combination of high-quality reference stations, and calculates the historical precise orbit and clock error as reference values. Step 2: Based on the historical precise orbit and clock error, perform basic predictions of long-term ephemeris using a dynamic model. Construct historical prediction error vectors for each epoch and stack them in chronological order to form historical error input windows. Input the historical error input windows and satellite auxiliary state features into a time-frequency domain attention neural network. Extract the time-domain trend features and frequency-domain periodic features of the historical error input windows through the time-frequency domain attention mechanism. After fusion, predict future prediction errors. Compensate the future prediction errors into the results of the basic predictions to generate corrected long-term ephemeris. Step 3: Compress and encode the corrected long-term ephemeris parameters, and adaptively select a broadcast strategy based on the network status and local cache status of the terminal. Step 4: After receiving and parsing the long-term ephemeris parameters, the terminal calculates the satellite position and satellite clock error, and uses the satellite position and satellite clock error as the terminal positioning calculation input to complete accelerated positioning.

2. The long-term ephemeris generation method for accelerated positioning of smart terminals according to claim 1, characterized in that, In step 2, constructing the historical forecast error vector for a single historical epoch includes: Using the historical precise orbit and clock error as reference values, the position error vector between the satellite reference position in the reference values ​​and the satellite forecast position obtained from the basic forecast is calculated in the geocentric Earth-fixed coordinate system. The position error vector is then projected onto the radial, trace, and normal directions to obtain the orbital error component. Based on the difference between the satellite reference clock error in the reference values ​​and the satellite forecast clock error obtained from the basic forecast, the clock error equivalent distance error component is obtained. The orbital error component and the clock error equivalent distance error component together constitute the historical forecast error vector for a single historical epoch.

3. The long-term ephemeris generation method for accelerated positioning of smart terminals according to claim 1, characterized in that, In step 2, the time-frequency domain attention neural network includes a time-domain feature branch, a frequency-domain feature branch, an auxiliary feature branch, a feature fusion layer, and a prediction output layer; The temporal feature branch uses a long short-term memory network to process the historical error input window, and extracts the temporal trend features and short-term correlation features of the historical error input window through input gate, forget gate, output gate and cell state; The frequency domain feature branch performs a Fourier transform on the historical error input window and converts the complex frequency domain components obtained after the transform into real-valued frequency domain features. The auxiliary feature branch encodes the satellite auxiliary state features and maps them through a fully connected layer to obtain auxiliary features; The feature fusion layer fuses the time trend feature, the short-term correlation feature, the real-valued frequency domain feature, and the auxiliary feature to obtain the fused feature; The prediction output layer outputs orbital error estimates and clock error equivalent distance estimates for multiple future forecast epochs based on the fusion features and the preset future forecast age.

4. The long-term ephemeris generation method for accelerated positioning of smart terminals according to claim 3, characterized in that, Step 2, which involves compensating the future forecast error into the result of the basic forecast, includes: After converting the orbital error estimate output by the prediction output layer back to the geocentric Earth-fixed coordinate system from the radial, track, and normal directions, it is added back to the satellite prediction position obtained from the basic prediction to obtain the corrected satellite position. The estimated equivalent distance error of the clock error output by the prediction output layer is divided by the speed of light and then added back to the satellite prediction clock error obtained from the basic prediction to obtain the corrected satellite clock error. The corrected long-term ephemeris parameters are composed of the corrected satellite position and the corrected satellite clock error.

5. A long-term ephemeris generation method for accelerated positioning of smart terminals according to claim 4, characterized in that, The corrected long-term ephemeris parameters include common head parameters, coordinate reference frame and time system parameters, satellite orbit parameters, satellite clock error parameters, mass and integrity parameters, and optional error compensation parameters; The common header parameters include protocol version, product batch number identifier, generation time, validity period start and end, and check code; The coordinate reference frame and time system parameters include coordinate reference frame identifier, time system identifier, week number, week seconds, and reference epoch; The satellite orbital parameters are expressed using piecewise Kepler fitting parameters, piecewise polynomial coefficients, or Chebyshev coefficients. The satellite clock bias parameters are expressed using piecewise fitting coefficients or periodic term coefficients. The quality and integrity parameters include satellite health status and accuracy level; The optional error compensation parameters include orbital residuals and clock error residuals, which are used to compensate for the remaining systematic errors over the long forecast period.

6. The long-term ephemeris generation method for accelerated positioning of smart terminals according to claim 1, characterized in that, In step 3, the adaptive selection of the broadcast strategy based on the network status and local cache status of the terminal includes: When the terminal network status is broadband connection, the server transmits the complete long-term ephemeris parameters in one go. When the terminal network is in a narrowband or weak network connection state, the server splits the long-term ephemeris parameters into multiple data packets according to satellites and effective time periods, and transmits them using small packet fragmentation, low-frequency transmission, breakpoint resumption, and missing packet retransmission mechanisms; the terminal performs out-of-order reassembly, missing packet detection, and integrity verification based on the sequence number and cyclic redundancy check code in each data packet, and requests the missing data packets from the server only when the network recovers; When the terminal has no network connection, the server suspends the delivery, and the terminal uses the long-term ephemeris parameters that have been cached locally, are still valid, and have passed the integrity check.

7. The long-term ephemeris generation method for accelerated positioning of smart terminals according to claim 1, characterized in that, The method further includes: Before publishing the corrected long-term ephemeris parameters, the server divides the training set, validation set, and test set according to the satellite or forecast period, and evaluates the time-frequency domain attention neural network using orbital radial error, orbital three-dimensional error, clock error equivalent distance error, or spatial signal ranging error as evaluation indicators. When the evaluation results meet the preset accuracy threshold and integrity threshold, the corrected long-term ephemeris parameters are published. When the evaluation results do not meet the preset accuracy threshold or integrity threshold, strategies such as reducing the forecast validity period, reverting to the basic forecast results of the dynamic model, or using the previous version of the valid product are adopted.

8. A long-lasting ephemeris generation device for accelerated positioning of smart terminals, characterized in that, include: The calculation module is used to set up the server to receive multi-source GNSS data via the Internet, select the best combination of high-quality reference stations, and calculate the historical precise orbit and clock error as reference values. The compensation module is used to make a basic prediction of long-term ephemeris based on the historical precise orbit and clock error using a dynamic model, and to construct a historical prediction error vector for each epoch, and stack them in chronological order to form a historical error input window. The historical error input window and satellite auxiliary state features are input into the time-frequency domain attention neural network. The time-frequency domain attention mechanism is used to extract the time-domain trend features and frequency-domain periodic features of the historical error input window. After fusion, the future forecast error is predicted. The future forecast error is compensated into the result of the basic forecast to generate the corrected long-term ephemeris. The broadcast module is used to compress and encode the corrected long-term ephemeris parameters and adaptively select a broadcast strategy based on the network status and local cache status of the terminal. The positioning module is used to set the terminal to receive and parse the long-term ephemeris parameters, calculate the satellite position and satellite clock error, and use the satellite position and satellite clock error as the terminal positioning calculation input to complete accelerated positioning.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the long-term ephemeris generation method for accelerated positioning of smart terminals as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the long-term ephemeris generation method for accelerated positioning of smart terminals as described in any one of claims 1-7.

Images