High-precision positioning method based on 5G

Abstract

The invention relates to the technical field of communication, in particular to a high-precision positioning method based on 5G. Comprising the following steps: collecting signal data through a 5G base station, building a hierarchical positioning model by using a graph neural network modeling environment and signal features, and optimizing the weight in real time; a mixed near-far field algorithm is used for cooperatively processing signals, and the positioning precision is improved. An error correction model is constructed based on transfer learning, and adaptive optimization in a complex environment is realized. A user position is dynamically tracked through sparse representation learning, prediction is performed in combination with Kalman filtering, a positioning result is updated in real time by using a 5G network, feedback optimization is performed, and high-precision positioning is ensured. According to the method, the positioning precision is improved, the environmental adaptability is enhanced, an error correction mechanism is perfected, a complete high-precision positioning system is formed, and the positioning method has higher reliability and efficiency in a complex environment.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and particularly to a high-precision positioning method based on 5G. Background Art

[0002] With the rapid development of modern communication technologies, communication towers, as an important part of base stations, play an increasingly important role in wireless communication networks. Especially under the promotion of 5G technology, communication towers not only undertake the traditional communication coverage function but also gradually develop into a new type of infrastructure supporting high-precision positioning. As a kind of tower facility, communication towers have natural advantages such as high height, wide coverage, and open vision, and are suitable as carriers of 5G base stations to provide efficient and accurate positioning services for users within the range. However, there are still the following problems: existing positioning methods are difficult to obtain high-precision positioning results in complex physical environments (such as urban multipath effects, signal occlusion, etc.), especially in areas with dense high-rise buildings, where signal interference and reflection cause a significant decline in positioning accuracy; traditional positioning methods cannot adjust the model in real time to adapt to changes in different environments, and positioning results are prone to significant deviations in dynamic scenarios; existing technologies mostly adopt a single positioning algorithm, which is difficult to balance the processing of near-field signals and far-field signals, resulting in a limited positioning range, and at the same time, the coordination between the near field and the far field is poor, restricting the performance of the overall positioning system. Summary of the Invention

[0003] To solve the above problems, the present invention provides a high-precision positioning method based on 5G, which solves the problem of how to improve the high-precision positioning ability of communication towers based on 5G in complex environments, especially realizes the adaptive adjustment of the model in dynamic scenarios, and effectively balances the coordination of near-field and far-field signal processing, thereby improving the positioning accuracy and making the positioning method have higher reliability and efficiency in complex environments.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A high-precision positioning method based on 5G, comprising the following steps:

[0006] S1: Collect signal data sent by user equipment through a 5G base station deployed on an iron tower;

[0007] S2: Based on the signal data, use a graph neural network algorithm to perform spatial topology modeling on the physical environment data around the iron tower, establish a hierarchical positioning model in combination with signal characteristics, and adapt to different complex environments by adjusting the model weights in real time;

[0008] S3: Based on the hierarchical positioning model, adopt a hybrid near-far field positioning algorithm to decompose and coordinate the processing of near-field positioning signals and far-field positioning signals;

[0009] S4: Using the transfer learning algorithm, construct an error correction model through historical positioning data and real-time collected data, and adaptively correct the positioning error in a complex environment through the knowledge transfer mechanism of transfer learning;

[0010] S5: Dynamically track the user's location through sparse representation learning, use the 5G network for real-time location updates, output the final positioning result to the application terminal, and at the same time feedback correction information for adaptive optimization.

[0011] Further, the signal data includes time of arrival, angle of arrival, relative time delay, and received signal strength.

[0012] Further, the specific steps of step S2 include the following steps:

[0013] Collect the physical environment data around the iron tower, including terrain, building distribution, and vegetation coverage rate, and generate an environmental feature dataset;

[0014] Based on the environmental feature dataset, construct an environmental feature map, and based on the signal data, form a node feature matrix;

[0015] Use the graph neural network algorithm to perform spatial topology modeling on the node feature matrix, and generate a high-dimensional topological structure containing environmental features and signal features;

[0016] Based on the high-dimensional topological structure, design a hierarchical positioning model, divide the area around the iron tower into multiple levels, optimize the dynamic weights of the hierarchical positioning model, and adjust the model weights in real time.

[0017] Even further, the formula of the hierarchical positioning model is as follows:

[0018]

[0019] Among them, P represents the final positioning result of the user equipment; w represents the weight parameter of the graph neural network; h (K) represents the node feature after K-layer iteration of the graph neural network; d near (h (K) ) represents the positioning error function of the near-field signal; d far (h (K) ) represents the positioning error function of the far-field signal; α and β respectively represent the weights of the near-field and far-field signal errors; H hist represents the historical positioning data feature matrix; γ represents the weight of the error correction term.

[0020] Further, the specific steps of step S3 include the following steps:

[0021] Perform classification processing on the collected signal data for near-field signals and far-field signals, and use a clustering algorithm for signal grouping;

[0022] For near-field signals, based on the short-path propagation model, combined with the characteristics of multipath effects, the time-reversal algorithm is used to analyze the signal path and generate the near-field signal positioning result;

[0023] For far-field signals, based on the long-distance propagation model, combined with the line-of-sight propagation characteristics, the time difference of arrival algorithm and the angle of arrival algorithm are used for positioning processing to generate the far-field signal positioning result;

[0024] The near-field signal positioning result and the far-field signal positioning result are synergistically fused, and the near-field and far-field positioning weights are dynamically adjusted based on the hybrid weight optimization model;

[0025] The position information of the user equipment is corrected through the fusion result to generate the precise positioning result after collaborative processing, and the result is input into the subsequent steps for error correction.

[0026] Furthermore, the formula of the short-path propagation model is as follows:

[0027]

[0028] Where, d represents the model output result, that is, the estimated distance between the user equipment and the base station; C represents the speed of light; N p represents the total number of multiple paths for the signal to reach the receiving end; T i represents the time difference of arrival of the signal on the i-th path; W i represents the signal quality weight of the i-th path; A i represents the energy loss of the signal during propagation; S i represents the received signal strength of the i-th path; B represents the loss exponent of the signal propagation path; F i represents the multipath interference factor of the i-th path; G i represents the angle correction factor of the i-th path; L i represents the angle of arrival of the i-th path; E represents the comprehensive error correction term introduced by multipath effects, environmental noise, and equipment deviation.

[0029] Furthermore, the long-distance propagation model specifically combines the line-of-sight propagation characteristics and multipath effects, adopts the joint positioning method of time difference of arrival and angle of arrival, constructs a path loss model including comprehensive corrections of propagation delay, signal attenuation, and noise interference, and uses the parameter optimization algorithm to dynamically adjust the weights of the path attenuation factor and the noise correction term in the model to achieve high-precision positioning of far-field signals.

[0030] Further, the transfer learning algorithm specifically adopts a hierarchical transfer mechanism, which divides the error correction task into two parts: global feature transfer and local dynamic adjustment. Global feature transfer is used for cross-scenario error compensation, generating global error correction parameters based on the signal propagation mode and environmental factors. Local dynamic adjustment is used to combine the real-time environmental characteristics and real-time signal fluctuations, and use the incremental learning algorithm to quickly update the model weights.

[0031] Further, the sparse representation learning specifically performs sparse coding processing on the multi-source signals of the user device and hierarchically screens the signal features. Among them, high-confidence signals are extracted through the dictionary matching algorithm for positioning modeling, and low-confidence signals are dynamically filtered through the noise elimination mechanism, and further combined with the Kalman filtering algorithm to predict and correct the time series trajectory of the user's position.

[0032] The beneficial effects of the present invention are as follows:

[0033] Based on the signal data collected by 5G base stations, the present invention combines the graph neural network algorithm to perform spatial topology modeling, fully considering the complexity of the physical environment around the iron tower, and can establish a more accurate hierarchical positioning model, thus significantly improving the positioning accuracy, especially adapting to complex urban multipath environments. By adjusting the model weights in real time, it can adapt to different complex environmental changes. Whether it is building occlusion, reflection interference or changes in user dynamic behavior, the positioning method can provide stable results, with strong environmental adaptability and robustness. The application of the hybrid near-field and far-field positioning algorithm effectively decomposes and collaboratively processes near-field and far-field signals, not only making full use of the high bandwidth and low latency characteristics of 5G signals, but also improving the positioning coverage and accuracy under the coordination of multi-source signals, avoiding the limitations of a single algorithm. Through the transfer learning algorithm, combining historical positioning data with real-time collected data to construct an error correction model can adaptively correct the positioning errors in complex environments, thus significantly improving the reliability and accuracy of positioning. With the help of sparse representation learning technology and the real-time transmission ability of 5G networks, it can dynamically track the user's position and achieve real-time update. At the same time, the model is continuously optimized through the feedback correction mechanism, further improving the positioning accuracy and real-time performance. Brief Description of the Drawings

[0034] Figure 1 is a schematic flowchart of a 5G-based high-precision positioning method of the present invention.

[0035] Figure 2 is a schematic flowchart of step S2 provided by an embodiment of the present invention.

[0036] Figure 3 is a schematic flowchart of step S3 provided by an embodiment of the present invention. Detailed Embodiments

[0037] Please refer to Figures 1-3 as shown, the present invention relates to a high-precision positioning method based on 5G.

[0038] Embodiment

[0039] A high-precision positioning method based on 5G includes the following steps:

[0040] S1: Collect signal data sent by user equipment through 5G base stations deployed on iron towers; the signal data includes time of arrival, angle of arrival, relative time delay, and received signal strength.

[0041] It should be noted that multi-antenna systems (such as MIMO array antennas) are installed at the top and different heights of the iron tower, and the coverage angles of the antennas are precisely designed to ensure full coverage of the target area. The base stations deployed on the iron tower support high-bandwidth data transmission and ultra-low latency communication, and are combined with Doppler frequency shift compensation functions to adapt to the dynamic movement of user equipment. The base station location needs to be optimized in combination with the user distribution density and the environmental characteristics of the target area (such as urban building complexes or rural open areas) to ensure signal quality and coverage.

[0042] The radio frequency front-end module in the base station filters and amplifies the received user equipment signals to ensure that the signal strength is suitable for subsequent processing. Through a high-precision clock synchronization system (such as a synchronization module based on GPS or PPS signals), nanosecond-level time synchronization is achieved between multiple base stations. An array antenna is used to collect different spatial dimension characteristics of multi-path signals, including time of arrival, angle of arrival, relative time delay, and received signal strength.

[0043] Time of arrival (ToA) measurement: The base station calculates the signal propagation time difference by capturing the time of arrival of the first path of the user equipment signal and combining an accurate timestamp. Using the high time resolution characteristics of ultra-wideband (UWB) or high-bandwidth 5G signals, the measurement accuracy of ToA reaches the nanosecond level, thus achieving sub-meter-level distance estimation.

[0044] Angle of arrival (AoA) measurement: The base station uses an array antenna to calculate the angle of arrival of the signal through a beamforming algorithm based on the phase difference of the signal arriving at different antenna elements. High-resolution algorithms such as MUSIC or ESPRIT are used for AoA measurement, which can effectively distinguish the angle characteristics of multi-path signals and avoid angle deviation caused by occlusion and reflection.

[0045] Relative time delay measurement: In the base station processing unit, the relative time delay characteristics of multi-path signals are extracted through cross-correlation operations on the signals. By comparing the propagation times of signals on different paths, the propagation mode of the signal reflection path can be analyzed to assist subsequent environmental modeling.

[0046] Received Signal Strength (RSS) Measurement: The power detection module within the base station collects the strength data of the received signal in real-time, which is used to estimate the loss characteristics of the signal propagation path. The RSS data is smoothed and denoised to reduce the intensity fluctuations caused by environmental interference (such as signal blockage or multipath fading).

[0047] The collected signal feature data (ToA, AoA, relative time delay, and RSS) is transmitted to the central processing server via a high-speed optical fiber link. In the central server, preprocessing algorithms are used to denoise, normalize, and standardize the signal data to improve the accuracy of subsequent modeling and calculations. The signal feature data, together with metadata such as the acquisition time, base station location, and antenna parameters, is stored in a database as the input for subsequent positioning algorithms.

[0048] S2: Based on the signal data, use the graph neural network algorithm to perform spatial topology modeling on the physical environment data around the iron tower, establish a hierarchical positioning model in combination with signal features, and adapt to different complex environments by adjusting the model weights in real-time;

[0049] Among them, the step S2 specifically includes the following steps:

[0050] Collect the physical environment data around the iron tower, including terrain, building distribution, and vegetation coverage rate, to generate an environmental feature dataset;

[0051] Based on the environmental feature dataset, construct an environmental feature map, and based on the signal data, form a node feature matrix;

[0052] It should be noted that the construction of the environmental feature map is specifically as follows:

[0053] Node creation: Divide the area around the iron tower into a grid or small blocks, and each area is used as a node; the physical features of the node include terrain height, building density, vegetation coverage rate, etc.

[0054] Edge relationship establishment: Adjacent nodes are connected according to the geographical distance, reflecting the spatial correlation. If a signal propagates between two nodes, an edge connection is added, and the weight of the edge can represent the signal strength or attenuation degree.

[0055] Generation of the node feature matrix, the features of each node consist of two parts:

[0056] Environmental features: Include physical attributes such as the terrain, buildings, and vegetation of the node.

[0057] Signal features: Include real-time signal data such as time of arrival, angle of arrival, and received signal strength.

[0058] Convert this information into a standardized data matrix for subsequent graph neural network processing.

[0059] Use the graph neural network algorithm to perform spatial topology modeling on the node feature matrix, generating a high-dimensional topological structure that includes environmental features and signal features;

[0060] Specifically, take the environmental feature map and the node feature matrix as inputs and perform modeling through a graph neural network (GNN). The graph neural network fuses the features of each node with the features of its neighboring nodes and gradually transmits information. The aggregated result not only contains the characteristics of the node itself but also reflects the correlation between the node and the surrounding environment, such as how the signal is affected by surrounding buildings and vegetation. The model generates a high-dimensional topological structure that describes the spatial relationship and signal propagation pattern in the area around the iron tower. This structure contains both physical environmental characteristics and incorporates signal propagation features.

[0061] Based on the high-dimensional topological structure, design a hierarchical positioning model, divide the area around the iron tower into multiple levels, and perform dynamic weight optimization on the hierarchical positioning model by adjusting the model weights in real time.

[0062] Specifically, divide the area into multiple levels according to the characteristics of the environment around the iron tower:

[0063] Core area: The area close to the iron tower, where the signal mainly follows a direct path.

[0064] Multipath reflection area: Contains signal reflection paths caused by buildings and vegetation.

[0065] Far-field area: The area far from the iron tower, where the signal may experience significant attenuation or complex multipath propagation.

[0066] For the design of the hierarchical positioning model, design sub-models for different levels: The core area model focuses on the time of arrival and angle of the signal to quickly calculate the user's position. The multipath reflection area model combines topological modeling to analyze the signal reflection path and identify reliable reflected signals. The far-field area model uses the signal strength attenuation pattern to predict the position of users at a long distance. Finally, integrate the outputs of the sub-models at each level through a weighted fusion strategy to form a global positioning result. Monitor signal data (such as signal-to-noise ratio and signal strength) in real time and dynamically adjust the weights of the hierarchical model according to the current environment. When significant changes occur in the environment (such as new obstacles or signal interference), update the model weights through incremental learning. Regularly compare the positioning results with the actual positions to generate error data. Feed the error data back to the hierarchical positioning model for optimizing parameters and improving subsequent positioning accuracy.

[0067] Furthermore, the formula of the hierarchical positioning model is as follows:

[0068]

[0069] Among them, P represents the final positioning result of the user device; w represents the weight parameters of the graph neural network; h (K) represents the node features after K-layer iteration of the graph neural network; d near (h (K) ) represents the positioning error function of the near-field signal; d far (h (K) ) represents the positioning error function of the far-field signal; α and β respectively represent the weights of the near-field and far-field signal errors; H hist represents the historical positioning data feature matrix; γ represents the weight of the error correction term.

[0070] S3: Based on the hierarchical positioning model, adopt a hybrid near-far field positioning algorithm to decompose and cooperate with the near-field positioning signal and the far-field positioning signal;

[0071] Among them, the step S3 specifically includes the following steps:

[0072] Perform classification processing on the collected signal data for near-field signals and far-field signals, and use a clustering algorithm to group the signals;

[0073] Specifically, use a collection device (such as a Wi-Fi module, UWB module, or Bluetooth sensor) to collect signal data in real time. Perform preliminary sorting on the collected signals, including filtering abnormal signals and smoothing signal fluctuations (such as removing environmental interference). Normalize the signal data, and adjust the signal features (such as intensity, time, frequency, etc.) to a unified dimension range to ensure the accuracy of subsequent processing.

[0074] Extract the key features of the signal, such as signal intensity, time of arrival, propagation path, etc., and construct a signal feature data set. Select a clustering algorithm (such as density clustering or grouping clustering) to classify the signals: the features of near-field signals are usually higher intensity and lower path loss; the features of far-field signals are usually lower intensity and higher path loss.

[0075] Check whether the classification result is accurate, for example, verify the classification result by comparing it with known positioning points in the collection environment. If a large classification error is found, the parameters of the clustering algorithm (such as the clustering radius or the number of groups) can be adjusted to optimize the classification effect. Output the grouped near-field signal and far-field signal data to provide a basis for subsequent positioning processing.

[0076] For near-field signals, based on the short-path propagation model, combined with the characteristics of multipath effects, use the time-reversal algorithm to analyze the signal path and generate the near-field signal positioning result;

[0077] Specifically, analyze the propagation characteristics of the near-field signal, and record the main path and secondary paths when the signal propagates over a short distance. For the multipath characteristics of the signal, mark the direct path (the shortest path) and the reflected path (the path reflected by obstacles). Play back the received signal in chronological order, identify the direct-path signal, and eliminate the interference of the secondary-path signal. Correct the multipath signal to enhance the characteristics of the direct-path signal. Combine the processed signal to determine the relative position between the device and the signal source. Conduct in-depth analysis of the signal characteristics of the direct path, such as identifying the impact of obstacles on the path on the signal. Generate the positioning data of the near-field signal, including the preliminary position coordinates of the device and the signal coverage range.

[0078] For far-field signals, based on the long-distance propagation model, combined with the line-of-sight propagation characteristics, use the time difference of arrival algorithm and the angle of arrival algorithm for positioning processing to generate the far-field signal positioning result;

[0079] Specifically, analyze the propagation path of the far-field signal, which is usually mainly line-of-sight propagation, but may be interfered by some multipath effects. Record the arrival time of the signal, the signal direction, and the environmental characteristics on the path (such as building occlusion). Use multiple base stations to collect signals, compare the time difference of arrival of the signals from the device to each base station, and infer the distance difference between the device and the base stations. Use a multi-antenna device to measure the angle of arrival of the signal, and calculate the position of the device based on the relationship between the signal direction and the positions of the base stations. Correct the signal positioning result, considering the influence of interference factors in the environment (such as signal reflection, occlusion, etc.) on the far-field signal positioning. Select the most reliable result from the multi-base station positioning results and exclude the positioning outliers. Output the positioning data of the far-field signal, including the preliminary position and coverage area of the device.

[0080] Perform collaborative fusion on the near-field signal positioning result and the far-field signal positioning result, and dynamically adjust the near-field and far-field positioning weights based on the hybrid weight optimization model;

[0081] Specifically, design a dynamic weight allocation mechanism to perform weighted fusion on the near-field positioning result and the far-field positioning result according to the credibility of the signal. The credibility of the signal is determined by the signal strength, the noise level, and the degree of environmental interference: when the credibility of the near-field signal is high, a higher weight is assigned; when the far-field signal has high credibility in a low-noise environment, a higher weight is assigned. Combine the positioning results of the near-field and far-field signals, conduct comprehensive analysis of the position of the user equipment, and generate the fused position information. During the fusion process, use weighted average or other fusion algorithms to dynamically adjust the contribution ratio of the near-field and far-field signals. Optimize the fused result, such as correcting it using historical positioning data and environmental information, and excluding possible error points. Generate accurate device position information for subsequent error correction and high-precision applications.

[0082] The location information of the user device is corrected based on the fusion result to generate an accurate positioning result after collaborative processing, and the result is input into the subsequent steps for error correction.

[0083] Specifically, the long-distance propagation model combines the line-of-sight propagation characteristics and the multipath effect, adopts the joint positioning method of time difference of arrival and angle of arrival, constructs a path loss model that comprehensively corrects the propagation delay, signal attenuation, and noise interference, and uses a parameter optimization algorithm to dynamically adjust the weights of the path attenuation factor and the noise correction term in the model to achieve high-precision positioning of far-field signals.

[0084] Furthermore, the formula of the short-path propagation model is as follows:

[0085]

[0086] Among them, d represents the output result of the model, that is, the estimated distance between the user device and the base station; C represents the speed of light; N p represents the total number of multiple paths of the signal reaching the receiving end; T i represents the time difference of arrival of the signal on the i-th path; W i represents the signal quality weight of the i-th path, usually assigned based on the signal-to-noise ratio (SNR). The larger the weight, the higher the contribution of this path to the positioning result; A i represents the energy loss of the signal during propagation; S i represents the received signal strength of the i-th path; B represents the loss exponent of the signal propagation path, reflecting the attenuation characteristics of the signal in different environments; F i represents the multipath interference factor of the i-th path, used to describe the complexity of path interference, calculated by statistically analyzing the power distribution of the path; G i represents the angle correction factor of the i-th path; L i represents the angle of arrival of the i-th path; E represents the comprehensive error correction term introduced by multipath effects, environmental noise, and device deviations.

[0087] S4: Using the transfer learning algorithm, an error correction model is constructed through historical positioning data and real-time collected data, and the positioning error in a complex environment is adaptively corrected through the knowledge transfer mechanism of transfer learning; the transfer learning algorithm specifically adopts a hierarchical transfer mechanism, which divides the error correction task into two parts: global feature transfer and local dynamic adjustment. Global feature transfer is used for cross-scenario error compensation, generating global error correction parameters based on the signal propagation mode and environmental factors, and local dynamic adjustment is used to combine the real-time environmental characteristics and real-time signal fluctuations, and use the incremental learning algorithm to quickly update the model weights.

[0088] Specifically, a two-stage transfer learning model is designed to handle global feature transfer and local dynamic adjustment respectively.

[0089] Global feature migration: Analyze and summarize signal propagation patterns across scenarios, such as the multipath signal effect in high-rise dense areas or the direct signal characteristics in open areas. Extract general error correction features from different environments and establish a basic global error correction model.

[0090] Local dynamic adjustment: Dynamically analyze the signal features collected in the real-time environment, such as a sudden attenuation of signal strength or an abnormal change in the arrival angle. Design a real-time adaptive adjustment mechanism to finely adjust the error correction model according to these changes in signal features.

[0091] Implementation of global feature migration: Use historical data from different environments (such as urban areas, suburbs, highways) to train a basic model for predicting error characteristics in common scenarios.

[0092] Implementation of local dynamic adjustment: Based on the signal characteristics collected in real-time (such as instantaneous signal fluctuations or occlusion effects), trigger the dynamic adjustment mechanism. Quickly adjust the key weight parameters in the model to adapt to the changes in the real-time environment.

[0093] Cross-scenario error compensation: The user equipment moves from one scenario (such as an open field) to another scenario (such as a high-rise dense urban area). The global feature migration model can automatically load the error correction features applicable to the new environment.

[0094] Real-time correction in a dynamic environment: When the user equipment enters a dynamic environment (such as a traffic-congested street), the local dynamic adjustment mechanism optimizes the correction result in real-time.

[0095] Adaptive optimization and feedback: All corrected error data will be recorded and stored in the system's historical database in real-time for subsequent optimization of the global feature migration ability of the model.

[0096] Output and optimization of error correction: Combine the corrected positioning result with the final positioning model and feedback it to the user application terminal in real-time to ensure the output of high-precision positioning results. Throughout the process, the system continuously optimizes the error correction model through the feedback mechanism to further enhance its adaptability in cross-scenario and dynamic environments.

[0097] S5: Dynamically track the user's location through sparse representation learning, use the 5G network for real-time location updates, output the final positioning result to the application terminal, and at the same time feedback correction information for adaptive optimization. The sparse representation learning specifically performs sparse coding processing on the multi-source signals of the user equipment and conducts hierarchical screening on the signal features; among them, high-confidence signals are extracted through the dictionary matching algorithm for positioning modeling, low-confidence signals are dynamically filtered through the noise rejection mechanism, and further combined with the Kalman filtering algorithm to predict and correct the time series trajectory of the user's location.

[0098] Specifically, for the sparse representation of multi-source signals, the signal matrix is converted into a sparse vector to retain important features and compress redundant information. The specific steps are as follows: Using the lasso regression method, high-value signal features are extracted through regularization constraints, and low-contribution signal components are eliminated. A sparse representation model is constructed to generate a sparse coding matrix, where each column corresponds to different feature dimensions of the signal (such as time, angle, intensity). Reliable signal features are matched in the sparse coding matrix through a dictionary matching algorithm, including: signals with a relatively high consistency in time series; signal paths verified through modeling in spatial topology.

[0099] Combining the historical trajectory data of the user device and the current signal features, the Kalman filter is used to predict the time series trajectory of the user's position. The kinematic model (such as a uniform linear motion or accelerated motion model) is used to predict the position at the next moment. The predicted position is updated by combining the real-time collected signal data to reduce the error caused by environmental dynamic changes. For the positioning deviation caused by occlusion or signal interference, the update equation of the Kalman filter is used for adaptive correction, especially significantly improving the positioning accuracy in fast-moving scenarios (such as vehicle movement or pedestrian dynamic walking).

[0100] The position result of the user device is sent to the application terminal (such as a navigation system or a dispatching center) with a millisecond-level delay through the 5G network. The real-time positioning result is compared with the actual trajectory data of the user to calculate the error and generate feedback data; the feedback data is sent into the sparse representation learning model to dynamically adjust the model weights and the dictionary matching algorithm. The sparse coding dictionary is updated regularly to ensure that the model adapts to real-time environmental changes. For new environments (such as new obstacles, vehicle parking), the signal processing model is updated in real time without the need for complete retraining.

[0101] In summary, the present invention utilizes the high bandwidth and low latency characteristics of the 5G network, and combines the high time resolution of ultra-wideband signals, and multi-dimensional signal features such as ToA, AoA, RSS, etc., to achieve nanosecond-level time synchronization and sub-meter-level positioning accuracy, meeting the high-precision positioning requirements in complex scenarios. The deployment and optimization design of multi-antenna systems at the top of the iron tower and at different heights ensure the wide-area coverage and targeted adaptability of the signals, adapting to different positioning requirements in urban high-density building or open rural environments. Through the spatial dimension analysis of multi-path signals, the positioning error caused by building occlusion or reflection is effectively reduced.

[0102] The neural network modeling in the present invention combines signal characteristics with physical environment characteristics to accurately model the complex spatial topological structure around the tower, thereby improving the robustness of the positioning algorithm in complex environments (such as cities with severe multipath effects or open areas dominated by line-of-sight propagation). Through the layered positioning model, the core area, multipath reflection area and far-field area are processed separately, and combined with the dynamic weight optimization strategy, accurate positioning at different environmental levels is achieved. The model can adjust the weights and optimize the positioning output in real time, enhancing the dynamic adaptability of the system.

[0103] The present invention uses a hybrid near-field and far-field positioning algorithm to decompose and coordinate near-field and far-field signals. The near field uses a time reversal algorithm to enhance the direct path signal, and the far field uses the arrival time difference and arrival angle algorithm for positioning. Combined with dynamic weight fusion, the positioning accuracy is significantly improved. An error correction model is constructed based on a transfer learning algorithm, and a method combining global feature migration and local dynamic adjustment is used to adaptively correct positioning errors in complex environments. Through the incremental learning mechanism, the model can quickly respond to environmental changes and achieve high reliability across scenarios and dynamic environments.

[0104] In the present invention, sparse representation learning performs sparse coding and hierarchical screening on multi-source signals, predicts user location trajectories through Kalman filtering, dynamically updates positioning results, and realizes real-time location tracking with millisecond delay. Dynamic signal filtering and model weight adjustment improve the reliability and real-time performance of the system. The sparse representation of signal features and dictionary matching algorithm are used to effectively compress data redundancy, extract high-value signal features, and reduce computational complexity. At the same time, the organic combination of signal features and environmental modeling optimizes resource utilization efficiency.

[0105] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A high-precision positioning method based on 5G, characterized in that: The following steps are involved: S1: Collect signal data sent by user equipment through 5G base stations deployed on towers; S2: Based on the signal data, a graph neural network algorithm is used to perform spatial topological modeling on the physical environment data around the tower, a hierarchical positioning model is established in combination with signal characteristics, and the model weights are adjusted in real time to adapt to different complex environments; S3: Based on the hierarchical positioning model, a hybrid near-field and far-field positioning algorithm is used to decompose and collaboratively process near-field positioning signals and far-field positioning signals; S4: Using the transfer learning algorithm, an error correction model is constructed based on historical positioning data and real-time collected data, and the positioning error in a complex environment is adaptively corrected through the knowledge transfer mechanism of transfer learning; S5: Dynamically track the user's location through sparse representation learning, use the 5G network to perform real-time location updates, output the final positioning result to the application terminal, and feedback the correction information for adaptive optimization.

2. A high-precision positioning method based on 5G according to claim 1, characterized in that: The signal data includes arrival time, arrival angle, relative delay and received signal strength.

3. A high-precision positioning method based on 5G according to claim 1, characterized in that: The step S2 specifically includes the following steps: Collect physical environment data around the tower, including topography, building distribution and vegetation coverage, to generate an environmental feature data set; Based on the environmental feature data set, an environmental feature map is constructed, and based on the signal data, a node feature matrix is ​​formed; Using a graph neural network algorithm to perform spatial topological modeling on the node feature matrix, generating a high-dimensional topological structure including environmental features and signal features; Based on the high-dimensional topological structure, a hierarchical positioning model is designed, and the area around the tower is divided into multiple levels. The hierarchical positioning model is dynamically optimized by adjusting the model weights in real time.

4. A 5G-based high-precision positioning method according to claim 3, characterized in that: The formula of the hierarchical positioning model is as follows: Where P represents the final user device positioning result; W represents the weight parameter of the graph neural network; h (K) represents the node features after K layers of graph neural network iteration; d near (h (K) ) represents the positioning error function of the near-field signal; d far (h (K) ) represents the positioning error function of the far-field signal; α and β represent the weights of the near-field and far-field signal errors, respectively; H hist represents the feature matrix of historical positioning data; γ represents the weight of the error correction term.

5. The high-precision positioning method based on 5G according to claim 1, characterized in that: The step S3 specifically comprises the following steps: The collected signal data is classified into near-field signals and far-field signals, and the signals are grouped using a clustering algorithm; For near-field signals, based on the short-path propagation model and combined with the multipath effect characteristics, a time reversal algorithm is used to analyze the signal path and generate near-field signal positioning results; For far-field signals, based on the long-distance propagation model and combined with the line-of-sight propagation characteristics, the arrival time difference algorithm and arrival angle algorithm are used for positioning processing to generate far-field signal positioning results; The near-field signal positioning results are synergistically integrated with the far-field signal positioning results, and the near-field and far-field positioning weights are dynamically adjusted based on the hybrid weight optimization model; The location information of the user equipment is corrected through the fusion results to generate an accurate positioning result after collaborative processing, and the result is input into the subsequent steps for error correction.

6. A 5G-based high-precision positioning method according to claim 5, characterized in that: The formula of the short path propagation model is as follows: Where d is the model output, i.e., the estimated distance between the user equipment and the base station; C is the speed of light; N p Indicates the total number of paths for the signal to reach the receiving end; T i represents the signal arrival time difference of the i-th path; W i A represents the signal quality weight of the i-th path; i Indicates the energy loss of the signal during propagation; S i represents the received signal strength of the i-th path; B represents the loss index of the signal propagation path; F i represents the multipath interference factor of the i-th path; G i represents the angle correction factor of the i-th path; L i represents the arrival angle of the i-th path; E represents the comprehensive error correction term introduced by multipath effect, environmental noise and equipment deviation.

7. The high-precision positioning method based on 5G according to claim 5, characterized in that: The long-distance propagation model specifically combines the line-of-sight propagation characteristics and the multipath effect, adopts the arrival time difference and arrival angle joint positioning method, constructs a path loss model that includes comprehensive corrections for propagation delay, signal attenuation and noise interference, and uses a parameter optimization algorithm to dynamically adjust the weights of the path attenuation factor and the noise correction term in the model to achieve high-precision positioning of far-field signals.

8. The high-precision positioning method based on 5G according to claim 1, characterized in that: The transfer learning algorithm specifically adopts a hierarchical migration mechanism, which divides the error correction task into two parts: global feature migration and local dynamic adjustment. The global feature migration is used for cross-scenario error compensation, and global error correction parameters are generated based on signal propagation patterns and environmental factors. The local dynamic adjustment is used to combine real-time environmental characteristics and real-time signal fluctuations, and use an incremental learning algorithm to quickly update the model weights.

9. The high-precision positioning method based on 5G according to claim 1, characterized in that: The sparse representation learning specifically performs sparse coding processing on the multi-source signals of the user device and performs hierarchical screening on the signal features; among them, high-confidence signals are extracted through a dictionary matching algorithm for positioning modeling, and low-confidence signals are dynamically filtered through a noise removal mechanism, and combined with the Kalman filtering algorithm to predict and correct the time series trajectory of the user's location.

Images