A method for Beidou satellite single point positioning based on non-direct signal interference suppression

Abstract

The application provides a Beidou satellite single point positioning method based on non-direct signal interference suppression, comprising constructing a pre-training sample data set and a NLOS signal identification model based on a double self-attention network and pre-training; inputting GNSS measurement data and ephemeris data into the pre-trained NLOS signal identification model to predict a signal type; correcting a weight method based on an elevation angle by using a prediction result; and generating a correction weight matrix; and calculating a user's position by using a weighted least square single point positioning algorithm. The NLOS signal identification model uses a double self-attention network to establish two channels; a correction coefficient is introduced to enhance the weighting based on the elevation angle and reduce the interference of the NLOS signal on positioning; a suitable weight is assigned to the identification result of the NLOS signal, and finally the user's position is calculated by using a weighted least square single point positioning algorithm, so as to improve the Beidou satellite positioning precision in a complex urban environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of satellite positioning technology, and in particular to a Beidou satellite single point positioning method based on non-direct signal interference suppression. BACKGROUND

[0002] The Beidou global navigation satellite system can provide real-time absolute position information, and can realize high-precision positioning services for automatic driving, intelligent mobile devices, Internet of Things devices, etc. However, in urban areas, Beidou satellite signals are easily blocked by buildings, resulting in non-line-of-sight (NLOS) signals. Due to the additional propagation path caused by reflection, NLOS signals will introduce serious bias in Beidou global navigation satellite system measurements, thereby reducing positioning accuracy.

[0003] Traditional NLOS interference mitigation methods, such as ray tracing and shadow matching, require three-dimensional building models or three-dimensional maps to identify NLOS signals, and their three-dimensional modeling costs are high. There are also methods that use fisheye cameras to exclude NLOS signals or rotate satellite antennas to reduce NLOS multipath errors; however, the need for auxiliary equipment limits the scene applicability of these methods.

[0004] In recent years, machine learning-based methods have been widely used in global navigation satellite system (GNSS) positioning to mitigate NLOS signal interference. Such methods, such as support vector machines (SVM) and decision trees (DT), classify NLOS signals by using machine learning models, remove identified NLOS signals from the observation equation, and eliminate NLOS interference to improve positioning accuracy. Using deep learning models to identify NLOS signals is a direct and effective method to mitigate NLOS interference and improve GNSS positioning accuracy; when using deep learning models to identify NLOS signals across multiple locations or scenarios, changes in environmental features and incident signal states can severely affect identification performance. However, existing methods typically only consider single-level feature extraction, i.e., immediate observation features of the signal itself; existing methods directly exclude identified NLOS signals or reduce NLOS interference by directly multiplying the predicted probability of NLOS signals with a weight matrix to improve positioning accuracy. However, since NLOS interference is mainly affected by environmental and incident signal states, existing NLOS signal identification models fail to effectively model environmental features and signal timing features, resulting in inability to achieve high-precision NLOS identification results across multiple locations or scenarios.

[0005] Secondly, in the case of serious occlusion, directly removing NLOS signals may reduce the number of satellites available to the positioning algorithm, resulting in reduced positioning accuracy; if the prediction probability of the NLOS signal is directly multiplied by the weight matrix, since the value of the classification prediction is binary (0 or 1), the final allocation weight of the NLOS signal is much lower than the normal value, completely ignoring the role of the NLOS signal, also resulting in reduced positioning accuracy. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a Beidou satellite single point positioning method based on non-direct signal interference suppression, which simultaneously models environmental features and signal timing features to improve the identification performance of NLOS signals across multiple locations or scenarios; and the present application assigns appropriate weights to NLOS signals to ensure the actual number of positioning satellites while mitigating the interference of NLOS signals on satellite positioning accuracy.

[0007] The technical solution of the present application is: a Beidou satellite single point positioning method based on non-direct signal interference suppression, comprising the following steps:

[0008] S1), acquiring satellite GNSS measurement data of multiple locations or scenarios and a sky map of the current location; and constructing a pre-training sample data set;

[0009] S2), constructing an NLOS signal identification model based on a double self-attention network;

[0010] S3), pre-training the NLOS signal identification model based on the double self-attention network using the pre-training sample data set;

[0011] S4), inputting the acquired GNSS measurement data and ephemeris data into the pre-trained NLOS signal identification model based on the double self-attention network to predict the signal type;

[0012] S5), using the prediction result in step S4) to correct the elevation-based weight assignment method; and generating a corrected weight matrix;

[0013] S6), calculating the user's location using a weighted least squares single point positioning algorithm.

[0014] As a preferred, in step S1), satellite GNSS measurement data of multiple locations or scenarios is recorded by a satellite receiver, and a sky map of the current location is photographed using a camera, the position of each satellite projected onto the photographed sky map is calculated according to the elevation angle and azimuth angle of the satellite in the satellite GNSS measurement data, whether the satellite is an NLOS signal is determined by whether the satellite projection position overlaps with a building, the satellite GNSS measurement data is marked, and finally the pre-training sample data set is constructed.

[0015] As preferred, in step S1), the satellite GNSS measurement data includes a plurality of continuous ephemeris observation value feature sequences of all satellites, denoted as:

[0016]

[0017] As preferred, in step S1), the satellite GNSS measurement data includes target satellite observation value features, target satellite historical ephemeris feature sequences, and current ephemeris all satellite feature sequences.

[0018] As preferred, in step S1), the target satellite observation value features include satellite elevation angle, azimuth angle, carrier-to-noise ratio, pseudo-range residual.

[0019] As preferred, in step S1), the position of each satellite projected onto the photographed sky map is calculated according to the satellite elevation angle (elevation angle) and azimuth angle in the satellite GNSS measurement data, specifically including the following steps:

[0020] S11), the elevation angle of the satellite is obtained, and the distance r of the satellite projection point from the center of the photographed image is calculated according to the equal solid angle projection calculation formula of the fisheye camera, and the calculation formula is:

[0021]

[0022] Wherein, f is the focal length of the camera lens, and θ is the elevation angle of the satellite;

[0023] S12), the azimuth β of the satellite projection point on the photographed image is calculated according to the azimuth of the satellite and the camera lens;

[0024] S13), according to the value r of the satellite projection point from the center of the photographed image and its azimuth β on the image calculated in step S11), the specific position of the projection point on the photographed sky map is solved, that is:

[0025] (r*cosβ, r*sinβ);

[0026] Wherein, the origin of the coordinates is the center point of the photographed image.

[0027] As preferred, in step S2), the NLOS signal recognition model based on the double self-attention network includes a spatial channel attention module and a time channel attention module, as well as a feedforward network and a classifier, and the spatial environment features and the time features of the signal are simulated through the spatial channel attention module and the time channel attention module.

[0028] As preferred, in step S3), the pre-training process of the NLOS signal recognition model based on the double self-attention network is:

[0029] S31), input the observation feature of the target satellite, the satellite feature sequence of the current epoch, and the feature sequence of the historical epoch of the target satellite into the NLOS signal identification model of the double self-attention network, and the satellite feature sequence of the current epoch is input into the spatial channel attention module; the feature sequence of the historical epoch of the target satellite is fed into the time channel attention module;

[0030] S32), the dimensions of the satellite feature sequence and the feature sequence of the historical epoch of the target satellite are expanded through the embedding layer, and a corresponding representation vector X is obtained; that is:

[0031] X=W e S+b e ;

[0032] Wherein, S is the input of the embedding layer, W e is the weight matrix of the neural network embedding layer, and b e is the bias value of the neural network embedding layer;

[0033] S33), the spatial channel attention module generates a value vector V1 and a key vector K1 from the satellite feature sequence of the current epoch through two linear mapping layers;

[0034] The time channel attention module generates a value vector V2 and a key vector K2 from the corresponding feature sequence of the historical epoch of the target satellite through two linear mapping layers;

[0035] And another linear mapping layer is used to generate a query vector Q from the observation feature of the target satellite;

[0036] S34), the attention weight matrix between the satellites of the current epoch and the attention weight matrix between the historical epochs of the target satellite are calculated according to the obtained value vector, key vector and query vector;

[0037] S35), the attention weight matrix between the satellites of the current epoch and the attention weight matrix between the historical epochs of the target satellite are connected and transferred to the feedforward network, which has a higher dimensional intermediate layer representation, allowing the model to capture more complex patterns and relationships from the fused attention weights;

[0038] S36), the output result of the feedforward network is input into the classifier for predicting the label type of the target satellite.

[0039] As a preferred, in step S3), in the pre-training process of the NLOS signal identification model of the double self-attention network, the loss function is optimized by using the mini-batch gradient descent method, and the loss function adopts the cross-entropy loss function l, and the expression is:

[0040]

[0041] where x i is the input of the NLOS signal identification model, y i is the label corresponding to the input, and N is the number of samples.

[0042] As a preferred, in step S35), the feedforward network comprises two fully connected neural network layers; the attention weight matrix between the satellites of the current epoch and the attention weight matrix between the target satellite historical epochs are connected to the feedforward network; and the calculation formula is:

[0043] F out = Relu([Z S ; Z T ]W1+b1)W2+b2

[0044] where F out is the output of the feedforward network; Z S and Z T represent the attention weight matrix between the satellites of the current epoch and the attention weight matrix between the target satellite historical epochs, respectively; Relu is the ReLU activation function between the two fully connected neural network layers; W1 and W2 are the weight parameters of the two fully connected neural network layers, respectively; and b1 and b2 are the bias values of the two fully connected neural network layers, respectively.

[0045] As a preferred, in step S36), the classifier is composed of two fully connected layers, and the sigmoid function is used as the activation function of the last fully connected layer of the classifier.

[0046] As a preferred, in step S5), the prediction result in step S4) is corrected by using the elevation-based weight determination method; and the expression of the corrected weight matrix is:

[0047]

[0048] where ω is the corrected weight matrix; P NLPS is the probability of the model predicting NLOS; j is the correction coefficient; and σ e represents the elevation-based weight determination.

[0049] As a preferred, in step S5), the expression of ω is:

[0050]

[0051] where a and b are variable parameters; and θ is the satellite elevation angle, i.e., the elevation angle.

[0052] As preferred, in step S6), the position of the user is calculated according to the revised weight matrix ω, and the GNSS measurement data and ephemeris data, and by using a weighted least square single point positioning algorithm, and the calculation expression is:

[0053]

[0054] In the formula, represents the user position estimation; A represents the coefficient matrix of unknown parameters; W represents the weight matrix; and L represents the free term vector.

[0055] As preferred, in step S6), the position of the user is calculated according to the revised weight matrix ω, and the GNSS measurement data and ephemeris data, and by using a weighted least square single point positioning algorithm, and the calculation expression is:

[0056] S61), the initial position of the user is set as (x0, y0, z0); and the offset is (△x,△y,△z);

[0057] S62), according to the GNSS measurement data of the satellite; the estimated pseudo-range ρ is calculated i , and the calculation expression is:

[0058]

[0059] In the formula, P i is the Euclidean distance between satellite i and the user; and c represents the speed of light; and δt r are the clock offsets of satellite i and the GNSS receiver respectively; I i and T i represent the ionosphere and troposphere errors respectively; and ∈ represents the observation error;

[0060] S63), formula (1) is linearized by Taylor series expansion to obtain formula (2):

[0061]

[0062] In the formula, is the Euclidean distance between satellite i and the initial position of the user; is a unit LOS vector;

[0063] S64), the right side of formula (2) is represented by l i , that is,

[0064]

[0065] Therefore, the matrix form of the pseudo-range observation equation is obtained

[0066] ∈=AX-L

[0067] wherein L represents a free vector; X = [△x,△y,△z,c·δt r ] T ;

[0068] wherein n is the number of captured satellites;

[0069] S65), solving by using a least square method, wherein a loss function of the least square method is expressed as:

[0070] J(X) = ∈ T ∈ = (AX-L) T (AX-L)

[0071] In order to minimize the loss function, the gradient of J(X) should be:

[0072]

[0073] wherein, represents the user position estimation;

[0074] Meanwhile, the measurement bias between different satellites is considered, and a correction weight matrix ω is introduced; and the user position estimation

[0075]

[0076] S66), after the user position solution is calculated, the initial position of the user is replaced, and iterative calculation is performed until the result meets the change threshold of the solution requirement, and finally the iteration is stopped, and the final position result is output.

[0077] As preferred, in step S63), the expression of the Euclidean distance between the satellite i and the initial position of the user is:

[0078]

[0079] The unit LOS vector is expressed as:

[0080]

[0081] wherein (x i ,y i ,z i ) is the position of the satellite, which can be obtained from ephemeris data.

[0082] The beneficial effects of the present application are:

[0083] 1. The NLOS signal recognition model of this invention uses a dual self-attention network to establish two channels: one learns space environment information from the satellite sequence of the current epoch, and the other extracts signal timing information from historical epochs; and enhances the elevation angle-based weighting by introducing a correction coefficient to reduce the interference of NLOS signals on positioning; first, NLOS signals are identified, then the identified signals are assigned an appropriate weight, and finally the weighted least squares single-point positioning algorithm is used to calculate the user's position, thereby improving the positioning accuracy of BeiDou satellites in complex urban environments;

[0084] 2. This invention improves the accuracy of NLOS identification across multiple locations or scenarios by simultaneously extracting spatial environment features and signal timing features, thereby suppressing the interference of NLOS signals on the positioning accuracy of BeiDou satellites;

[0085] 3. This invention uses NLOS recognition results to correct the elevation angle-based scheme, and can assign more reasonable weights to NLOS signals according to the recognition results. Attached Figure Description

[0086] Figure 1 This is a schematic diagram of the framework of the NLOS signal recognition model based on dual self-attention networks of the present invention;

[0087] Figure 2 This is a flowchart illustrating the single-point positioning method of the present invention. Detailed Implementation

[0088] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0089] like Figure 1 and 2 As shown, this embodiment provides a BeiDou satellite point positioning method based on non-direct signal interference suppression, including the following steps:

[0090] S1) Acquire satellite GNSS measurement data from multiple locations or scenarios, as well as a sky map of the current location; and construct a pre-trained sample dataset;

[0091] In this embodiment, a satellite receiver is used to record satellite GNSS measurement data at multiple locations or scenes. The satellite GNSS measurement data includes the observation characteristics of the target satellite and the characteristic sequence of the target satellite's historical epochs. And the characteristic sequences of all satellites in the current epoch (S1, S2, ..., S...). n ), where n represents the number of satellites in the current epoch; t represents the current epoch; and k is the length of the historical epoch.

[0092] The observed characteristics of the target satellite include satellite elevation angle, azimuth angle, carrier-to-noise ratio, and pseudorange residual.

[0093] The embodiment calculates the position of each satellite projected onto the photographed sky map according to the elevation angle and the azimuth angle of the satellite in the satellite GNSS measurement data, judges whether the satellite is an NLOS signal by whether the satellite projection position overlaps with the building, marks the satellite GNSS measurement data, and finally constructs a pre-training sample data set.

[0094] The position of each satellite projected onto the photographed sky map according to the elevation angle and the azimuth angle of the satellite in the satellite GNSS measurement data specifically includes the following steps:

[0095] S11), the elevation angle of the satellite is obtained, and the distance r of the satellite projection point from the center of the photographed image is calculated according to the equal solid angle projection formula of the fisheye camera, and the calculation formula is:

[0096]

[0097] Wherein, f is the focal length of the camera lens, and θ is the elevation angle of the satellite.

[0098] S12), the azimuth β of the satellite projection point on the photographed image is calculated according to the azimuth of the satellite and the camera lens.

[0099] S13), the specific position of the projection point on the photographed sky map is solved according to the value r of the satellite projection point from the center of the image and its azimuth β on the image calculated in step S11), that is:

[0100] (r*cosβ, r*sinβ);

[0101] Wherein, the origin of the coordinates is the center point of the photographed image.

[0102] S2), constructing an NLOS signal recognition model based on a double self-attention network;

[0103] The framework diagram of the NLOS signal recognition model constructed in the embodiment is shown in Figure 1 The NLOS signal recognition model based on the double self-attention network in the embodiment includes a spatial channel attention module and a time channel attention module, as well as a feedforward network and a classifier. The spatial channel attention module and the time channel attention module are used to simulate the spatial environment characteristics and the time characteristics of the signal, so as to improve the recognition ability of the NLOS signal recognition model across multiple locations or scenes.

[0104] S3), pre-training the NLOS signal recognition model based on the double self-attention network by using the pre-training sample data set; in the pre-training process of the NLOS signal recognition model based on the double self-attention network, the loss function is optimized by using the small batch gradient descent method, and the loss function adopts the cross-entropy loss function l, and the expression is as follows:

[0105]

[0106] Wherein, x i is the input of the model, y i is the label corresponding to the input, and N is the number of samples.

[0107] The pre-training process of the NLOS signal recognition model based on the double self-attention network specifically includes the following steps:

[0108] S31), inputting the observation value feature of the target satellite, the satellite feature sequence (S1, S2,..., S n ) of the current epoch, the feature sequence of the historical epoch of the target satellite into the NLOS signal recognition model based on the double self-attention network, and the satellite feature sequence (S1, S2,..., S n ) of the current epoch is input into the spatial channel attention module; the feature sequence of the historical epoch of the target satellite is fed into the time channel attention module;

[0109] S32), the dimensions of the satellite feature sequence and the feature sequence of the historical epoch of the target satellite are expanded by the embedding layer, and the corresponding representation vector X is obtained; that is:

[0110] X=W e S+b e ;

[0111] Wherein, S is the input of the embedding layer, W e is the weight matrix of the neural network embedding layer, and b e is the bias value of the neural network embedding layer;

[0112] S33), the spatial channel attention module generates a value vector V1 and a key vector K1 from the satellite feature sequence of the current epoch by two linear mapping layers;

[0113] The time channel attention module generates a value vector V2 and a key vector K2 from the corresponding feature sequence of the historical epoch of the target satellite by two linear mapping layers;

[0114] And another linear mapping layer is used to generate a query vector Q from the observation value feature of the target satellite;

[0115] S34), calculating the attention weight matrix between satellites at the current epoch and the attention weight matrix between satellites at the historical epochs of the target satellite according to the obtained value vector, key vector and query vector; the calculation expression is:

[0116]

[0117] wherein, softmax is an activation function; is a scaling factor.

[0118] S35), connecting the attention weight matrix between satellites at the current epoch and the attention weight matrix between satellites at the historical epochs of the target satellite to transfer to a feedforward network, which has a higher dimensional intermediate layer representation, allowing the model to capture more complex patterns and relationships from the fused attention weights; wherein, the feedforward network comprises two fully connected neural network layers; by connecting the attention weight matrix between satellites at the current epoch and the attention weight matrix between satellites at the historical epochs of the target satellite to transfer to a feedforward network; the calculation expression is:

[0119] F out =Relu([Z S ;Z T ]W1+b1)W2+b2

[0120] wherein, F out is the output of the feedforward network; Z S and Z T represent the attention weight matrix between satellites at the current epoch and the attention weight matrix between satellites at the historical epochs of the target satellite, respectively; Relu is the ReLU activation function between the two fully connected neural network layers; W1 and W2 are the weight parameters of the two fully connected neural network layers, respectively; b1 and b2 are the bias values of the two fully connected neural network layers, respectively.

[0121] S36), inputting the output result of the feedforward network into a classifier for predicting the label type of the target satellite; the classifier is composed of two fully connected layers, and the sigmoid function is used as the activation function of the last fully connected layer of the classifier.

[0122] S4), inputting the obtained GNSS measurement data and ephemeris data into the pre-trained NLOS signal identification model of the dual self-attention network to predict the signal type;

[0123] S5), using the prediction result in step S4) to correct the elevation-based weighting method; and generating a corrected weight matrix, the expression of which is:

[0124]

[0125] wherein, ω is the corrected weight matrix; P NLOSThe model predicts the probability of NLOS occurring; j is the correction coefficient; σ e This indicates weighting based on elevation angle.

[0126] in, The expression is:

[0127]

[0128] In the formula, a and b are variable parameters; θ is the satellite elevation angle, i.e., the angle of elevation.

[0129] This embodiment uses an elevation angle-based weighting method to assign more accurate weights to NLOS satellites during the positioning process, thereby reducing the impact of NLOS interference on satellite positioning accuracy.

[0130] S6) The user's location is calculated using the weighted least squares single-point positioning algorithm. In this embodiment, the user's location is calculated based on the corrected weight matrix ω, GNSS measurement data, and ephemeris data, using the weighted least squares single-point positioning algorithm to improve the accuracy of BeiDou satellite positioning in complex urban environments. The specific steps include the following:

[0131] S61) Let the user's initial position be (x0, y0, z0); and the offset be (△x, △y, △z);

[0132] S62) Based on the satellite's GNSS measurement data, calculate the estimated pseudorange ρ. i Calculate the expression:

[0133]

[0134] Among them, P i It is the Euclidean distance between satellite i and the user, where c represents the speed of light; and δt r These are the clock offsets of satellite i and the GNSS receiver, respectively; I i and T i These represent errors in the ionosphere and troposphere, respectively; ∈ represents observation error.

[0135] S63), by linearizing equation (1) through Taylor series expansion, we obtain equation (2):

[0136]

[0137] In the formula; Let i be the Euclidean distance between satellite i and the user's initial position; It is a unit LOS vector;

[0138] Wherein, the Euclidean distance between satellite i and the user's initial position The expression is:

[0139]

[0140] Unit LOS vector is expressed as:

[0141]

[0142] where (x i ,y i ,z i ) is the position of the satellite, which can be obtained from ephemeris data;

[0143] S64), the right part of equation (2) is expressed as l i

[0144]

[0145] Therefore, the matrix form of the pseudo-range observation equation is obtained

[0146] ∈=AX-L

[0147] where L represents the free term vector; X = [△x,△y,△z,c·δt r ] T ;

[0148] where n is the number of captured satellites;

[0149] S65), the loss function of the least square method is expressed as:

[0150] J(X)=∈ T ∈=(AX-L) T (AX-L)

[0151] In order to minimize the loss function, the gradient of J(X) should be:

[0152]

[0153] where represents the user position estimate;

[0154] At the same time, the measurement bias between different satellites is considered, and a correction weight matrix ω is introduced; the user position estimate

[0155]

[0156] S66), after the user position solution is calculated, the user initial position is replaced, and iterative calculation is performed again until the result meets the change threshold of the solution requirement, and finally the iteration is stopped, and the final position result is output.​

[0157] The foregoing embodiments and descriptions are only illustrative of the principles and the best mode of the application, and various changes and modifications can be made thereto without departing from the spirit and scope of the application.

Claims

1. A method for BeiDou satellite single point positioning based on non-direct signal interference suppression, characterized in that, The method comprises the following steps: S1), acquiring satellite GNSS measurement data of multiple positions or scenes and a sky map of a current position; and constructing a pre-training sample data set; S2), constructing a NLOS signal recognition model based on a double self-attention network; The NLOS signal recognition model based on the double self-attention network comprises a space channel attention module and a time channel attention module, and a feedforward network and a classifier, and the space channel attention module and the time channel attention module are used to simulate the space environment characteristics and the time characteristics of the signal; S3), pre-training the NLOS signal recognition model based on the double self-attention network by using the pre-training sample data set; S4), inputting the acquired GNSS measurement data and ephemeris data into the pre-trained NLOS signal recognition model based on the double self-attention network to predict the signal type; S5), using the prediction result in step S4) to correct the elevation-based weight determination method; and generating a corrected weight matrix; the expression of the generated corrected weight matrix is: where ω is the correction weight matrix; P NLOS is the probability of NLOS occurrence predicted by the model; j is the correction coefficient; σ e denotes the elevation-based weight, where, The expression for is: In the formula, a and b are variable parameters; and θ is the elevation angle of the satellite; S6), calculating the position of the user by using a weighted least square single point positioning algorithm according to the corrected weight matrix, the GNSS measurement data and the ephemeris data.

2. The method according to claim 1, wherein the method is based on non-direct signal interference suppression for BDS-PPP. The satellite GNSS measurement data comprises the observation value characteristics of the target satellite, the feature sequence of the historical epochs of the target satellite, and the feature sequence of all satellites at the current epoch.

3. The method according to claim 2, wherein the method is based on non-direct signal interference suppression. In step S1), the satellite GNSS measurement data of multiple positions or scenes is recorded by using a satellite receiver, and a sky map of a current position is photographed by using a camera, the position of each satellite projected onto the photographed sky map is calculated according to the elevation angle and the azimuth angle of the satellite in the satellite GNSS measurement data, whether the satellite is an NLOS signal is judged by whether the satellite projection position overlaps with a building, the satellite GNSS measurement data is marked, and finally the pre-training sample data set is constructed.

4. The method according to claim 1, wherein the method is based on non-direct signal interference suppression for BDS-PPP. In step S3), the pre-training process of the NLOS signal recognition model based on the double self-attention network is as follows: S31), the observation value characteristics of the target satellite, the satellite feature sequence at the current epoch, and the feature sequence of the historical epochs of the target satellite are input into the NLOS signal recognition model based on the double self-attention network, and the satellite feature sequence at the current epoch is input into the space channel attention module; the feature sequence of the historical epochs of the target satellite is fed into the time channel attention module; S32), the dimensions of the satellite feature sequence and the feature sequence of the historical epochs of the target satellite are expanded by using an embedding layer, and a corresponding representation vector X is obtained; that is: X = W e S + b e ; wherein S is the input of the embedding layer, W e is the weight matrix of the neural network embedding layer, b e is the bias value of the neural network embedding layer; S33), the value vector V1 and the key vector K1 are generated from the satellite feature sequence at the current epoch by using two linear mapping layers in the space channel attention module; the value vector V2 and the key vector K2 are generated from the corresponding feature sequence of the historical epochs of the target satellite by using two linear mapping layers in the time channel attention module; and a query vector Q is generated from the observation value characteristics of the target satellite by using another linear mapping layer; S34), calculating the attention weight matrix between satellites at the current epoch and the attention weight matrix between the target satellite and historical epochs according to the obtained value vector, key vector and query vector; S35), connecting the attention weight matrix between satellites at the current epoch and the attention weight matrix between the target satellite and historical epochs to transfer to the feedforward network; S36), inputting the output result of the feedforward network into the classifier for predicting the label type of the target satellite.

5. The method according to claim 4, wherein the method is based on non-direct signal interference suppression for BDS-PPP. In step S35), the feedforward network comprises two fully connected neural network layers; the attention weight matrix between satellites at the current epoch and the attention weight matrix between the target satellite and historical epochs are connected to transfer to the feedforward network; and the calculation formula is: F out = Relu([Z S ; Z T ] W1 + b1) W2 + b2 where F out is the output of the feedforward network; Z S and Z T represent the attention weight matrix between satellites at the current epoch and the attention weight matrix between target satellites at the historical epochs, respectively; Relu is the ReLU activation function between the two fully connected neural network layers; W1 and W2 are the weight parameters of the two fully connected neural network layers, respectively; and b1 and b2 are the bias values of the two fully connected neural network layers, respectively.

6. The method according to claim 5, wherein the method is based on non-direct signal interference suppression. In step S6), the position of the user is calculated according to the corrected weight matrix ω and the GNSS measurement data and ephemeris data, and by using the weighted least squares single point positioning algorithm, and the calculation expression is: wherein represents the user position estimate; A represents a coefficient matrix of unknown parameters; W represents a weight matrix; and L represents a vector of free terms.

7. The method according to claim 6, wherein the method is based on non-direct signal interference suppression for BDS-PPP. In step S6), the position of the user is calculated according to the corrected weight matrix ω and the GNSS measurement data and ephemeris data, and by using the weighted least squares single point positioning algorithm, and the calculation expression is: S61), setting the initial position of the user as (x0, Y0, z0) and the offset as (△x, △y, △z); S62), from GNSS measurement data of the satellite; calculating an estimated pseudorange p i , calculating the expression: where P i is the Euclidean distance between satellite i and the user, c represents the speed of light; and δt r are the clock offsets of satellite i and GNSS receiver, respectively; i and T i represent the ionosphere and troposphere errors, respectively; ∈ represents the observation error; S63), linearizing formula (1) by Taylor series expansion to obtain formula (2): wherein is the Euclidean distance between satellite i and the user's initial position; is the unit LOS vector; S64), with l i to represent the right-hand side of equation (2), i.e.: Therefore, the matrix form of the pseudo-range observation equation is obtained ∈=AX-L wherein L represents a free vector; X = [△x,△y,△z, c·δt r ] T ; where n is the number of satellites captured; S65), solving by using the least squares method, and the loss function of the least squares method is represented as: J(X) = ∈ T ∈ = (AX - L) T (AX - L) In order to minimize the loss function, the gradient of J(X) should be: In the formula, represents the user position estimate; The measurement bias between different satellites is considered simultaneously, and a correction weight matrix ω is introduced; and a user position estimation is obtained S66), after obtaining the solution of the position of the user, replacing the initial position of the user and performing iterative calculation until the result meets the change threshold value of the solution requirement, and finally stopping iteration and outputting the final position result.

8. The method according to claim 7, wherein the method is based on non-direct signal interference suppression for BDS-PPP. In step S63) the Euclidean distance between the satellite i and the user initial position The expression for the Euclidean distance between the satellite i and the user initial position is: Unit LOS vector is represented as: where (x i ,y i ,z i ) is the position of the satellite, obtained from ephemeris data.

Images