A base station self-calibration method combined with direct non-line-of-sight error mitigation

Abstract

The present invention provides a base station self-calibration method combined with direct non-line-of-sight error mitigation. The non-line-of-sight error is directly mitigated through a parallel deep regression network, and the base station position is iteratively updated by combining global optimization and local optimization strategies, effectively solving the problems that the existing base station self-calibration methods are limited by non-line-of-sight occlusion and multipath effects, have a slow convergence speed, and have low self-calibration accuracy, and providing a more efficient and accurate solution for the self-calibration of ultra-wideband base station positions in complex environments.

Description

Technical Field

[0001] The invention relates to the field of radio signal processing, and particularly to a base station self-calibration method combined with direct non-line-of-sight error mitigation. Background Art

[0002] Ultra-wideband (UWB) positioning technology has become the most commonly used indoor positioning technology due to its high precision and anti-multipath interference ability. UWB technology locates through high-frequency pulse signals, which can provide centimeter-level accuracy in complex indoor environments, especially suitable for application scenarios with high precision requirements, such as intelligent manufacturing, robot navigation, and Internet of Things positioning. Its main advantage lies in being able to effectively solve the problem of performance degradation of traditional positioning technologies in non-line-of-sight (NLOS) environments, and having strong anti-interference ability and low power consumption.

[0003] Base station self-calibration is a crucial part of the UWB positioning system, and usually, it is necessary to accurately determine the position of the base station to ensure the accuracy of the positioning system. However, due to environmental factors, signal multipath effects, and NLOS errors, base station calibration is a cumbersome and challenging task. Traditional base station calibration methods usually rely on manual measurement or assume known environmental information, lacking automation and adaptability. Currently, common base station self-calibration technologies include methods based on least squares optimization, multidimensional scaling transformation, and Kalman filtering. For example, the "UWB base station self-calibration method based on multidimensional scaling transformation" (publication number: CN 117528413) applied by Dalian Haoru Technology Co., Ltd. constructs an inner product matrix through ranging between base stations, reduces the dimension to calculate the coordinates of the target base station, simplifies the calibration process, and improves the deployment efficiency and configuration flexibility; the "Base station self-calibration method for UWB positioning system based on Kalman filtering" (publication number: CN 107708204) applied by Changzhou Institute of Technology processes the distance and height data between base stations through Kalman filtering, recursively calculates the coordinates of each base station, realizes high-precision self-calibration, simplifies the system layout, and improves the mobility. However, these methods are easily affected by errors in a dynamic environment, especially the interference of the NLOS effect, resulting in the difficulty of traditional base station self-calibration methods to converge or converge to incorrect values;

[0004] Therefore, there is an urgent need to improve the existing technology. Summary of the Invention

[0005] For the problem that the base station positions arranged in the UWB indoor positioning system need to be manually measured and calibrated, traditional methods usually rely on manual measurement and precise calculation, which are time-consuming and vulnerable to human errors. At the same time, existing base station self-calibration methods are limited by NLOS occlusion and multipath effects, making it difficult to perform effective calibration or converge to the correct position, and there is a challenge of low self-calibration efficiency. To address these problems, the present invention proposes a base station self-calibration method combined with direct non-line-of-sight error mitigation. This method directly identifies and corrects NLOS errors through parallel depth regression prediction, provides a smaller initial iteration value to accelerate the base station self-calibration calculation, and realizes efficient and accurate automatic calibration of the base station positions by combining global and local optimization strategies, effectively improving the accuracy and convergence speed of self-calibration.

[0006] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0007] A base station self-calibration method combined with direct non-line-of-sight error mitigation, comprising the following steps:

[0008] S1: First, obtain the ranging values d between each base station through the bilateral two-way time-of-flight ranging method ij , for any two base stations i and j, their coordinates are p i =(x i , y i , z i ) and p j =(x j , y j , z j ), and the ranging value satisfies:

[0009]

[0010] S2: Mitigate the direct non-line-of-sight error for the ranging values between all base stations to obtain the ranging matrix D m ={dm ij}, provide a smaller initial input value for optimization iteration, and accelerate the calculation of the self-calibrated position of the base station;

[0011] S3: Construct the global error function:

[0012]

[0013] S4: Local optimization, divide all base stations into several subsets, and use local graph optimization to solve the positions within each subset, that is, three base stations form a triangle and the intersection of two spheres is used for solution, and four base stations form a tetrahedron and the intersection of three spheres is used for solution;

[0014] S5: Local position solution, select N≥3 base stations and initialize the positions using the following rules:

[0015] The coordinates of the fixed base station 1 are (0, 0, 0), so:

[0016]

[0017] Secondly, determine the position of base station 2. Assume it is located on the x-axis, i.e., (d 12 , 0, 0);

[0018] Next, determine the position of base station 3. According to the method of the intersection line of two spherical surfaces, limit it to the xy-plane:

[0019]

[0020] If N = 3, three base stations form a triangle, and the solution stops here.

[0021] S6: Determine the position of base station 4. According to the ranging values d 14 , d 24 , d 34 between base station 4 and base stations 1, 2, and 3 respectively, solve its position by the intersection of three spherical surfaces:

[0022] First, establish the equations for the intersection of three spherical surfaces:

[0023]

[0024] After expanding each equation and eliminating variables, we get:

[0025]

[0026] Simplify these two linear equations to:

[0027]

[0028] Iteratively solve this linear equation by the Levenberg-Marquardt method to obtain the optimal coordinates (x4, y4, z4) of base station 4;

[0029] S7: Global optimization. Use the global error E as the objective function, adjust the relative positions of each local geometric figure so that their overlapping regions meet the consistency constraints, and use the genetic algorithm to perform nonlinear optimization on the coordinates of all base stations;

[0030] S8: Stop when the global error E converges to the set threshold ε or reaches the maximum number of iterations;

[0031] S9: Output the relative coordinates p1, p2,..., p N ;

[0032] Furthermore, the bilateral two-way time-of-flight ranging method includes the following steps:

[0033] S11: First, the base station i sends a communication polling frame to the base station j, where the base stations i and j respectively record their own hardware timestamps: t0 and t1;

[0034] S12: After a fixed time delay D round1 the base station j replies with a response frame, and the corresponding timestamps are t2 and t3;

[0035] S13: After another fixed time delay D round2 the base station i sends the last frame to the base station j, and records the timestamps as t4 and t5;

[0036] S14: Then use these timestamps to calculate the two-way round-trip time of flight:

[0037]

[0038] Based on the two-way round-trip time of flight, calculate the ranging value between the base stations i and j:

[0039] d ij = Timediff × c (9)

[0040] where c is the propagation speed of the ultra-wideband radio wave in the current environment;

[0041] Furthermore, the direct non-line-of-sight error mitigation is realized by parallel deep regression prediction, and the parallel deep regression prediction specifically includes the following steps:

[0042] S21: First, the ultra-wideband base station receives the signals sent by other base stations, and extracts the channel impulse response data and seven artificial features therefrom;

[0043] S22: Input the channel impulse response data into the first branch to extract deep features, and input the seven artificial features into the second branch to extract further features;

[0044] S23: In the first branch of the channel impulse response data, first process it through a multi-layer deep convolutional feature extraction network. Secondly, further extract key features from the processed feature map through the squeeze-and-excitation attention mechanism. Finally, flatten the obtained key features and input them into the long short-term memory network layer to extract the temporal features, and input the finally processed feature map into the feature fusion layer;

[0045] S24: Input the seven artificial features into the second branch. First, process them through three fully connected layers to capture the non-linear relationships in the input data. Then, activate them through the sigmoid function to normalize the feature values to the range [0,1]. Finally, input them into the feature fusion layer;

[0046] S25: The deep features of the channel impulse response data extracted by the first and second branches and the further features of seven artificial features are fused through the feature fusion layer;

[0047] S26: The fused features are dimensionally reduced through two fully connected layers and finally input into the regression layer to obtain the predicted value Δd of the ranging error ij , and the ranging error is mitigated:

[0048] dm ij = d ij -Δd ij (10)

[0049] Furthermore, there is the following ranging constraint relationship between the base stations:

[0050] The distance d of each base station from base station 1 1i provides a spherical constraint:

[0051]

[0052] The ranging d between two base stations ij provides an ellipsoidal surface constraint;

[0053] The definition of the ellipsoidal surface constraint is as follows. Assume that the coordinates of the base station k to be solved are p k =(x k , y k , z k ), the distance between base station k and base station i is d ik , and the distance between base station k and base station j is d jk , to obtain the following two spherical constraints:

[0054]

[0055]

[0056] These two constraints mean that base station k must be located on the spheres centered at base stations i and j, with radii d ik and d jk ;

[0057] After splitting and merging, the ellipsoidal surface equation is obtained:

[0058]

[0059] Furthermore, the process of the Levenberg-Marquardt iteration for solving linear equations is as follows:

[0060] S61: Set the damping factor λ > 0 and the stopping condition;

[0061] S62: Construct the error function, and define the error function for each pair of base stations (i, j):

[0062] e ij =||p i -p j ||-d ij (15)

[0063] Combine all error terms into an error vector:

[0064] e=[e 12 e 13 ... e N-1,N (16)

[0065] S63: Define the objective function as the sum of squares of the error vector:

[0066] L=e T e (17)

[0067] S64: Construct the Jacobian matrix J, where each term is the partial derivative of the error function with respect to the k-th coordinate:

[0068]

[0069] For e ij , its partial derivative form is:

[0070]

[0071] S65: The Levenberg-Marquardt method is updated as follows:

[0072] Δp=-(J T J+λI) -1 J T e (20)

[0073] where I is the identity matrix;

[0074] The updated coordinate is:

[0075] p←p+Δp(21)

[0076] S66: Iteratively solve for the optimal value. First, calculate the new objective function value L. If the error decreases, then decrease λ and accept the update; otherwise, increase λ and reject the update;

[0077] Repeat steps S64 and S65 until convergence to obtain the optimal coordinates of base station 4;

[0078] Furthermore, in the parallel depth regression prediction, the feature extraction of the first branch and the second branch is carried out simultaneously. When there is no channel impulse response data input to the first branch, if there are seven artificial features input to the second branch, its feature extraction will still be carried out, and the output of the feature fusion layer is only the feature map after the extraction of the seven artificial features. When there are no seven artificial features input to the second branch, if there is channel impulse response data input to the first branch, it will still be carried out, and the output of the feature fusion layer is only the feature map after the extraction of the channel impulse response data.

[0079] In the multi-layer deep convolutional feature extraction network, a batch normalization layer and a max pooling layer are connected in sequence after each convolutional layer.

[0080] In the second branch, a batch normalization layer, a ReLU activation function, and a dropout layer are added in sequence between the first fully connected layer and the second fully connected layer.

[0081] The seven artificial features include: ranging value between base stations, received signal strength difference of the first path, signal received strength indication, distance difference of the first path, kurtosis factor, average excess delay, and root mean square delay spread.

[0082] The beneficial effects of the present invention are mainly manifested in:

[0083] (1) By combining global optimization and local optimization, the coordination problem of global error and local error in the multi-base station self-calibration process is solved. Global optimization improves the overall positioning accuracy, and local optimization specifically corrects the error between each pair of base stations, thereby further refining the accurate position of each base station while ensuring system consistency.

[0084] (2) By using a parallel depth regression network to simultaneously process the ranging errors between multiple base stations, directly identify and mitigate NLOS errors, provide a smaller initial iterative error, and solve the problems of high computational complexity and low optimization efficiency of traditional methods. This parallel processing significantly accelerates the process of solving the base station position and improves the iterative efficiency of self-calibration.

[0085] (3) By combining the fast convergence of the Levenberg-Marquardt algorithm with the feature extraction ability of the deep network, the problems of slow convergence speed and inaccurate error processing in the base station position optimization are effectively solved, and the self-calibration accuracy and optimization efficiency are significantly improved. At the same time, through the squeeze-and-excitation attention mechanism and parallel processing procedures in the deep regression network, the accuracy of regression prediction is further enhanced. Brief Description of the Drawings

[0086] Figure 1 is the operation flowchart of the base station self-calibration method for mitigating direct non-line-of-sight errors in combination with the present invention.

[0087] Figure 2It is a schematic diagram of the ultra-wideband two-way time-of-flight ranging method;

[0088] Figure 3 It is a flowchart of the operation for direct non-line-of-sight error mitigation using parallel depth regression prediction;

[0089] Figure 4 It is a flowchart of the operation for solving the linear equation of the base station position by Levenberg-Marquardt iteration. Specific implementation manner

[0090] To facilitate the understanding of the present invention, the present invention will be described more comprehensively and in detail below in conjunction with the accompanying drawings of the specification and preferred embodiments, but the protection scope of the present invention is not limited to the following specific embodiments.

[0091] As Figure 1 shown, a base station self-calibration method combining direct non-line-of-sight error mitigation in this embodiment includes the following steps:

[0092] S1: First, obtain the ranging values d between each base station through the two-way time-of-flight ranging method. For any two base stations i and j, their coordinates are p ij =(x i ,y i ,z i ) and p i =(x j ,y j ,z j ) respectively, and the ranging value satisfies:

[0093]

[0094] S2: Perform direct non-line-of-sight error mitigation on the ranging values between all base stations to obtain the ranging matrix D m ={dm ij}, providing a smaller initial value for the optimization iteration input and accelerating the calculation of the self-calibrated position of the base station;

[0095] S3: Construct a global error function:

[0096]

[0097] S4: Local optimization. Divide all base stations into several subsets, and use local graph optimization to solve the position within each subset, that is, three base stations form a triangle and solve using the intersection of two spheres, and four base stations form a tetrahedron and solve using the intersection of three spheres;

[0098] S5: Local position solution. Select N≥3 base stations and initialize the position using the following rules:

[0099] ​The coordinates of the fixed base station 1 are (0, 0, 0), so:

[0100]

[0101] Secondly, determine the position of base station 2. Assume it is located on the x-axis, i.e., (d 12 , 0, 0);

[0102] Next, determine the position of base station 3. According to the method of the intersection line of two spherical surfaces, confine it to the xy-plane:

[0103]

[0104] If N = 3, three base stations form a triangle, and the solution stops here.

[0105] S6: Determine the position of base station 4. According to the ranging values d 14 , d 24 , d 34 between base station 4 and base stations 1, 2, and 3 respectively, solve its position by the intersection of three spherical surfaces:

[0106] First, establish the equation of the intersection of three spherical surfaces:

[0107]

[0108] After expanding each equation and eliminating variables, we get:

[0109]

[0110] Simplify these two linear equations to:

[0111]

[0112] Solve this linear equation iteratively by the Levenberg-Marquardt method to obtain the optimal coordinates (x4, y4, z4) of base station 4;

[0113] S7: Global optimization. Use the global error E as the objective function to adjust the relative positions of each local geometric figure so that their overlapping regions meet the consistency constraints, and use the genetic algorithm to perform nonlinear optimization on the coordinates of all base stations;

[0114] S8: Stop when the global error E converges to the set threshold ε or reaches the maximum number of iterations;

[0115] S9: Output the relative coordinates p1, p2,..., p N ;

[0116] As Figure 2 shown, the bilateral two-way time-of-flight ranging method specifically includes the following steps:

[0117] S11: First, the base station i sends a communication polling frame to the base station j. The base stations i and j respectively record their own hardware timestamps: t0 and t1.

[0118] S12: After a fixed time delay D round1 the base station j replies with a response frame, and the corresponding timestamps are t2 and t3.

[0119] S13: After another fixed time delay D round2 the base station i sends the last frame to the base station j, and records the timestamps as t4 and t5.

[0120] S14: Then use these timestamps to calculate the two-way round-trip time of flight:

[0121]

[0122] Calculate the ranging value between the base stations i and j based on the two-way round-trip time of flight:

[0123] d ij = Timediff × c (9)

[0124] where c is the propagation speed of the ultra-wideband radio wave in the current environment;

[0125] As Figure 3 shown, the direct non-line-of-sight error mitigation is achieved through parallel deep regression prediction. The parallel deep regression prediction specifically includes the following steps:

[0126] S21: First, the ultra-wideband base station receives signals sent from other base stations, and extracts channel impulse response data and seven artificial features from them.

[0127] S22: Input the channel impulse response data into the first branch to extract deep features, and input the seven artificial features into the second branch to extract further features.

[0128] S23: In the first branch of the channel impulse response data, first process it through a multi-layer deep convolutional feature extraction network. Secondly, further extract key features from the processed feature map through the squeeze-and-excitation attention mechanism. Finally, flatten the obtained key features and input them into the long short-term memory network layer to extract temporal features, and input the finally processed feature map into the feature fusion layer.

[0129] S24: Input the seven artificial features into the second branch. First, process them through three fully connected layers to capture the non-linear relationships in the input data. Then, activate them through the sigmoid function to normalize the feature values to the range [0, 1]. Finally, input them into the feature fusion layer.

[0130] S25: Further fuse the deep features of the channel impulse response data and seven artificial features extracted by the first branch and the second branch through the feature fusion layer;

[0131] S26: The fused features are dimensionally reduced through two fully connected layers and finally input into the regression layer to obtain the predicted value Δd of the ranging error ij , and mitigate the ranging error:

[0132] dm ij = d ij -Δd ij (10)

[0133] There is the following ranging constraint relationship between the base stations: The distance d between each base station and base station 1 1i provides a spherical constraint:

[0134]

[0135] The ranging d between two base stations ij provides an ellipsoidal surface constraint;

[0136] The definition of the ellipsoidal surface constraint is as follows. Assume that the coordinates of the base station k to be solved are p k =(x k ,y k ,z k ), the distance between base station k and base station i is d ik , the distance between base station k and base station j is d jk , and the following two spherical constraints are obtained:

[0137]

[0138]

[0139] These two constraints mean that base station k must be located on the spheres centered at base stations i and j, with radii d ik and d jk ;

[0140] After splitting and merging, the ellipsoidal surface equation is obtained:

[0141]

[0142] As Figure 4 shown, the process of the Levenberg-Marquardt iteration for solving linear equations is as follows:

[0143] S61: Set the damping factor λ > 0 and the stopping condition;

[0144] S62: Construct the error function and define the error function for each pair of base stations (i,j):

[0145] e ij = ||p i - p j || - d ij (15)

[0146] Combine all error terms into an error vector:

[0147] e = [e 12 e 13 ... e N-1,N (16)

[0148] S63: Define the objective function as the sum of squares of the error vector:

[0149] L = e T e (17)

[0150] S64: Construct the Jacobian matrix J, where each term is the partial derivative of the error function with respect to the k-th coordinate:

[0151]

[0152] For e ij , its partial derivative form is:

[0153]

[0154] S65: The Levenberg - Marquardt method is updated as follows:

[0155] Δp = -(J T J + λI) -1 J T e (20)

[0156] where I is the identity matrix;

[0157] The updated coordinate is:

[0158] p ← p + Δp (21)

[0159] S66: Iteratively solve for the optimal value. First, calculate the new objective function value L. If the error decreases, then decrease λ and accept the update; otherwise, increase λ and reject the update;

[0160] Repeat steps S64 and S65 until convergence to obtain the optimal coordinates of base station 4;

[0161] In the parallel depth regression prediction, the feature extraction of the first branch and the second branch is carried out simultaneously. When there is no channel impulse response data input to the first branch, if there are seven artificial features input to the second branch, its feature extraction will still be carried out, and the output of the feature fusion layer is only the feature map after the extraction of the seven artificial features; when there are no seven artificial features input to the second branch, if there is channel impulse response data input to the first branch, it will still be carried out, and the output of the feature fusion layer is only the feature map after the extraction of the channel impulse response data;

[0162] In the multi-layer deep convolutional feature extraction network, a batch normalization layer and a max pooling layer are sequentially connected after each convolutional layer;

[0163] In the second branch, a batch normalization layer, a ReLU activation function and a dropout layer are sequentially added between the first fully connected layer and the second fully connected layer;

[0164] The seven artificial features include: ranging value between base stations, received signal strength difference of the first path, signal received strength indication, distance difference of the first path, kurtosis factor, average excess delay and root mean square delay spread;

[0165] With the help of the teachings present in the foregoing specification and the related drawings, those skilled in the art to which the present invention pertains will envision many modifications and other embodiments of the present invention. Accordingly, it is to be understood that the present invention is not limited to the specific embodiments disclosed, and that modifications and other embodiments are considered to be included within the scope of the appended claims. Although specific terms are employed herein, they are used in a general and descriptive sense only and not for purposes of limitation.

Claims

1. A base station self-calibration method combined with direct non-line-of-sight error mitigation, characterized in that: It includes the following steps: S1: First, obtain the ranging values d between each base station through the bilateral two-way time-of-flight ranging method ij , for any two base stations i and j, their coordinates are p i =(x i , y i , z i ) and p j =(x j , y j , z j ), and the ranging value satisfies: S2: Perform direct non-line-of-sight error mitigation on the ranging values between all base stations to obtain the ranging matrix D after error mitigation m ={dm ij}, provide a smaller initial value for the optimized iteration input, and accelerate the calculation of the self-calibrated positions of the base stations; S3: Construct a global error function: S4: Local optimization. Divide all base stations into several subsets. Use local graph optimization to solve the position within each subset. That is, when three base stations form a triangle, use the intersection of two spheres to solve; when four base stations form a tetrahedron, use the intersection of three spheres to solve; S5: Local position solution. Select N≥3 base stations and initialize the position using the following rules: Fix the coordinates of base station 1 as (0,0,0). Therefore: Next, determine the position of base station 2, assuming it is located on the x-axis, i.e., (d 12 , 0, 0); Determine the position of base station 3 again. According to the method of the intersection line of two spheres, limit it to the xy-plane: If N = 3, that is, there are three base stations forming a triangle, the solution stops here. S6: Determine the position of base station 4. According to the ranging values d 14 , d 24 , d 34 between base station 4 and base stations 1, 2, and 3 respectively, solve its position by the intersection of three spheres: First, establish the equation of the intersection of three spheres: After expanding each equation and eliminating variables, we get: Simplify these two linear equations to: Iteratively solve this linear equation using the Levenberg-Marquardt method to obtain the optimal coordinates (x4, y4, z4) of base station 4; S7: Global optimization. Use the global error E as the objective function, adjust the relative positions of each local geometric figure so that their overlapping regions satisfy the consistency constraint, and use the genetic algorithm to perform nonlinear optimization on the coordinates of all base stations; S8: Stop when the global error E converges to the set threshold ε or reaches the maximum number of iterations; S9: Output the relative coordinates p1, p2, …, p of N base stations N .

2. The base station self-calibration method combined with direct non-line-of-sight error mitigation according to claim 1, wherein: The two-way two-way time-of-flight ranging method includes the following steps: S11: First, base station i sends a communication polling frame to base station j, where base station i and base station j respectively record their respective hardware timestamps: t0 and t1; S12: After a fixed time delay D round1 the base station j replies with a response frame, with corresponding timestamps t2 and t3; S13: At another fixed time delay D round2 After that, base station i sends the last frame to base station j, recording the timestamps as t4 and t5; S14: Then use these timestamps to calculate the two-way two-way time of flight: Calculate the ranging value between base station i and base station j based on the two-way two-way time of flight: d ij = Timediff × c (9) where c is the propagation speed of ultra-wideband radio waves in the current environment.

3. The base station self-calibration method combined with direct non-line-of-sight error mitigation according to claim 1, characterized in that: The direct non-line-of-sight error mitigation is achieved through parallel deep regression prediction. The parallel deep regression prediction specifically includes the following steps: S21: First, the ultra-wideband base station receives signals sent from other base stations and extracts channel impulse response data and seven artificial features from them; S22: Input the channel impulse response data into the first branch to extract deep features, and input the seven artificial features into the second branch to extract further features; S23: In the first branch of the channel impulse response data, first process it through a multi-layer deep convolutional feature extraction network. Secondly, use the squeeze-and-excitation attention mechanism to further extract key features from the processed feature map. Finally, flatten the obtained key features and input them into the long short-term memory network layer to extract temporal features, and input the finally processed feature map into the feature fusion layer; S24: Input the seven artificial features into the second branch. First, process them through three fully connected layers to capture the non-linear relationships in the input data. Then, activate them through the sigmoid function, normalize the feature values to the range [0,1], and finally input them into the feature fusion layer; S25: The feature fusion layer fuses the deep features of the channel impulse response data extracted from the first branch and the further features of the seven artificial features; S26: The fused features are dimensionally reduced through two fully connected layers and finally input into the regression layer to obtain the predicted value Δd of the ranging error ij , and the ranging error is mitigated: dm ij = d ij - Δd ij (10).

4. The base station self-calibration method combining direct non-line-of-sight error mitigation according to claim 1, characterized in that: There is the following ranging constraint relationship between the base stations: The distance d of each base station from base station 1 1i A spherical constraint is provided: The ranging d between two base stations ij provides an elliptical surface constraint; The definition of the elliptical surface constraint is as follows. Assume that the coordinates of the base station k to be solved are p k =(x k ,y k ,z k ), the distance between the base station k and the base station i is d ik , the distance between the base station k and the base station j is d jk , and the following two spherical constraints are obtained: These two constraints mean that base station k must lie on the spheres centered at base stations i and j, with radii d ik and d jk ; After splitting and combining, the ellipsoid equation is obtained:

5. The base station self-calibration method combined with direct non-line-of-sight error mitigation according to claim 1, characterized in that: The process of iteratively solving the linear equation using the Levenberg-Marquardt method is as follows: S61: Set the damping factor λ>0 and the stop condition; S62: Construct an error function and define the error function for each pair of base stations (i, j): e ij = ||p i -p j || -d ij (15) Combine all error terms into an error vector: e = [e 12 e 13 ...e N-1,N (16) S63: Define the objective function as the sum of squares of the error vector: L = e T e (17) S64: Construct the Jacobian matrix J, where each term is the partial derivative of the error function with respect to the k-th coordinate: For e ij , its partial derivative form is: S65: The Levenberg-Marquardt method is updated as follows: Δp = -(J T J + λI) -1 J T e (20) where I is the identity matrix; The updated coordinate is: p←p + Δp(21) S66: Iteratively solve for the optimal value. First, calculate the new objective function value L. If the error decreases, then decrease λ and accept the update; otherwise, increase λ and reject the update; Repeat steps S64 and S65 until convergence to obtain the optimal coordinates of base station 4.

6. The base station self-calibration method combined with direct non-line-of-sight error mitigation according to claim 3, characterized in that: In the parallel deep regression prediction, the feature extraction of the first branch and the second branch is carried out simultaneously. When there is no channel impulse response data input to the first branch, if there are seven artificial features input to the second branch, its feature extraction will still be carried out, and the output of the feature fusion layer is only the feature map after the extraction of the seven artificial features; when there are no seven artificial features input to the second branch, if there is channel impulse response data input to the first branch, it will still be carried out, and the output of the feature fusion layer is only the feature map after the extraction of the channel impulse response data; In the multi-layer deep convolutional feature extraction network, a batch normalization layer and a max pooling layer are sequentially connected after each convolutional layer; In the second branch, a batch normalization layer, a ReLU activation function, and a dropout layer are sequentially added between the first fully connected layer and the second fully connected layer; The seven artificial features include: ranging values between base stations, first path received signal strength difference, signal received strength indication, first path distance difference, kurtosis factor, average excess delay, and root mean square delay spread.

Images