Method for eliminating near-far effect of broadcast radio positioning system, storage medium and terminal device

Abstract

The present invention discloses a method for eliminating the near-far effect in a broadcast radio positioning system, a storage medium, and a terminal device, comprising the following steps: multi-source data acquisition and preprocessing; constructing a dynamic graph data structure; constructing a Transformer-GNN joint model; model training and optimization; and model evaluation and adaptive optimization. The present invention implements global signal correlation analysis through the self-attention mechanism of the Transformer. Based on the alternating stacked GNN-Transformer architecture, it introduces residual connections and a gating mechanism to dynamically fuse local-global features. It combines phased training with a dynamic weight adjustment strategy to optimize positioning accuracy and interference suppression capabilities. It supports multimodal data fusion, significantly improving anti-multipath performance and dynamic adaptability in complex scenarios, and solves the defects of traditional methods such as high computational complexity, strong hardware dependence, and insufficient environmental robustness. It is suitable for suppressing the near-far effect when satellites are unavailable, improving system positioning accuracy, and enabling the system to be better applied in high-precision positioning scenarios such as indoor positioning and drone navigation.

Description

Technical Field

[0001] The present invention relates to the field of wireless positioning technology, and in particular to a method for eliminating the near-far effect of a broadcast radio positioning system, a storage medium and a terminal device. Background Art

[0002] In broadcast radio positioning systems, because signal propagation loss is proportional to the square of the distance, the signal strength received by a radio positioning receiver terminal from a closer ground base station is significantly greater than that from a more distant one. This can cause the signal from a closer ground base station to overwhelm the signal from a more distant one. This phenomenon is known as the near-far effect in broadcast radio positioning systems. The near-far effect is a critical technical issue that urgently needs to be addressed in radio positioning. It affects navigation system signal reception, severely degrading positioning accuracy or even causing system failure, impacting the accuracy and reliability of the entire navigation system.

[0003] GNSS systems use satellites in medium- and high-orbit orbits to broadcast ranging signals, which are then received by satellite navigation receivers for real-time positioning and velocity measurement. However, due to the high orbits of satellite navigation systems, the signal power is extremely weak near the ground and can be easily blocked or interfered with by radar, communications, or malicious actors, causing malfunctions in the satellite navigation receivers. In high-security areas such as those related to national economy and people's livelihoods, or in military applications, emergency navigation and positioning measures are needed to meet navigation and positioning needs in the event of satellite signal interference or system failure. To this end, research institutions at home and abroad have been conducting research on technologies related to broadcast radio positioning systems.

[0004] A broadcast radio positioning system consists of two components: a radio ground base station and a radio positioning receiver. Multiple radio ground base stations are deployed on the ground, and the signal strength of a single station can effectively cover a distance of hundreds of kilometers. The radio positioning receiver terminal simultaneously receives ranging signals from multiple radio ground base stations to perform real-time positioning and velocity measurement. However, due to the strong transmission signals from the ground base stations, the radio positioning receiver terminal's ability to capture signals from distant base stations when the base station is close to the receiver terminal can be affected, resulting in reduced positioning accuracy or even inability to locate a position.

[0005] Currently, solutions to the near-far effect in broadcast radio positioning systems primarily rely on ground radio base stations. This effect is mitigated by improving the signal spectrum and signal strength at the base station's transmitter. Common methods include out-of-band emission, frequency offset, frequency hopping, time hopping, and power regulation. The first three methods are difficult to implement, requiring significant modifications to the radio frequency (RF) or correlators of the radio positioning receiver terminal, and are therefore not widely used. The time hopping method achieves good results within a small area, but due to the time slot division, the transmitted time-hopping signal has an overall lower power than the continuous signal, affecting the system's anti-interference performance. The power regulation method can mitigate the near-far effect to a certain extent, but requires signal strength analysis and power regulation based on the application scenario. This results in a low level of automation and is not suitable for flexible deployment.

[0006] The invention patent, "A Method for Suppressing the Near-Far Effect of Pseudosatellites Based on Multi-Constrained Beamforming," proposes an improved signal reception and processing method for a radio positioning receiver terminal. The signal received by the antenna array undergoes sampling, bandpass filtering, amplitude and phase error correction, calculation of the covariance matrix and its inverse, MUSIC direction finding, pseudosatellite and navigation satellite signal power estimation, calculation of the constrained response vector corresponding to the pseudosatellite signal direction, calculation of multi-constrained beamforming weights, digital beamforming, and navigation solution, ultimately obtaining local position information. This invention is more versatile and can operate in interference environments. It is convenient and flexible, requiring no modifications to the transmitting pseudosatellite or the receiving radio frequency. Summary of the Invention

[0007] The technical solution of the present invention to solve the above technical problems is to provide a method for eliminating the near-far effect of a broadcast radio positioning system, comprising the following steps:

[0008] S1: Multi-source data acquisition and preprocessing: Collect radio ground base station data, radio positioning receiver data, interference source data, and reflection source data; perform signal alignment, feature standardization, environmental perception fusion, and outlier repair on the data;

[0009] S2: Construct a dynamic graph data structure: Define ground base stations, receivers, interference sources, and fictitious multipath reflection nodes as graph nodes. Based on the signal propagation relationship, construct ground station-receiver connection edges, interference source-receiver connection edges, and spatiotemporal trajectory connection edges. Design a dynamic update mechanism to achieve real-time updates of node positions and edge weights.

[0010] S3: Build a Transformer-GNN joint model: Implement local-global joint modeling by alternating stacked GNN layers and Transformer layers, where:

[0011] The GNN layer defines a message function based on the node type and extracts local topological features;

[0012] The Transformer layer introduces edge attribute encoding and captures global signal correlation through sparse attention mechanism;

[0013] Dynamically fuse GNN and Transformer features using residual connections and gating mechanisms;

[0014] S4: Model training and optimization: A phased training strategy is adopted, first pre-training the GNN layer, and then jointly training the Transformer-GNN model; dynamic weight adjustment (AdaLoss) is used to balance the localization loss and interference suppression loss;

[0015] S5: Model evaluation and adaptive optimization: Evaluate positioning accuracy based on root mean square error (RMSE) and mean absolute error (MAE), adjust model structure parameters based on the evaluation results, and use data augmentation to improve generalization capabilities.

[0016] Furthermore, in step S2:

[0017] Node types include:

[0018] For radio ground base station nodes, the feature vector includes three-dimensional coordinates, transmission power, carrier frequency and pseudo-random code number;

[0019] Radio positioning receiver node, the feature vector includes real-time position, velocity, received signal strength vector and carrier-to-noise ratio;

[0020] Interference source node, the feature vector contains the interference location, power, bandwidth and type code;

[0021] The imaginary multipath reflection node, the feature vector includes the reflection surface coordinates, incident angle, reflection coefficient and path delay;

[0022] Edge weights are calculated as follows:

[0023] Ground station-receiver side: based on the free space path loss formula and obstacle attenuation model;

[0024] Interference-receiver edge: based on the correlation between spectrum overlap integral and spatial distance;

[0025] Spatiotemporal trajectory edges: based on time decay factors or prediction confidence scores.

[0026] Furthermore, the dynamic update mechanism of step S2 includes:

[0027] The update frequency is adjustable from 50ms to 5000ms, with the default being 3000ms;

[0028] The reconfiguration mechanism is triggered when a new interference source is detected or the signal is lost. The reconfiguration conditions include:

[0029] The spectrum energy mutation exceeds the threshold of 3 times the background noise;

[0030] No signal is detected for N consecutive epochs, where N is dynamically adjustable and the default value is 5.

[0031] Furthermore, the construction of the Transformer-GNN joint model in step S3 further includes:

[0032] The GNN layer uses a graph convolutional network (GCN) or a graph attention network (GAT) to reduce computational complexity through neighbor sampling or TopK aggregation;

[0033] The Transformer layer uses a sparse attention mechanism to encode edge attributes into position embeddings;

[0034] Alternately stack GNN layers and Transformer layers 3 to 5 times, in the order of GNN→Transformer loop structure.

[0035] Furthermore, the residual connection and gating mechanism are implemented as follows:

[0036] The residual connection directly adds the GNN output features to the Transformer output features;

[0037] The gating mechanism dynamically controls the feature fusion weights based on GRU, and selects retained features through update gates and reset gates.

[0038] Furthermore, the dynamic weight adjustment (AdaLoss) in step S4 satisfies:

[0039] The joint loss function is ,in, is the positioning loss, To suppress the loss of interference, the weight λ is dynamically adjusted during the training process;

[0040] The training data is divided into training set, validation set and test set according to the ratio of 6:3:1, and GPU is used to accelerate parallel training.

[0041] Furthermore, the adaptive optimization in step S5 includes:

[0042] Adjust the number of GNN layers, number of Transformer heads, and hidden layer dimensions based on RMSE and MAE evaluation results;

[0043] Enhance the training set by rotating and scaling data to improve the model's adaptability to dynamic environments.

[0044] In order to solve the above technical problems, the present invention further proposes a computer-readable storage medium storing a computer program, which implements the above-mentioned method for eliminating the near-far effect when executed by a processor.

[0045] In order to solve the above technical problems, the present invention also proposes a broadcast radio positioning system terminal device, including a memory, a processor and a computer program stored in the memory, and the processor executes the method for eliminating the near-far effect as described above when executing the program.

[0046] Compared with the prior art, the present invention has the following technical effects:

[0047] 1. Creatively constructing a Transformer-GNN dual-modal dynamic fusion network, breaking through the boundaries of traditional single-model technology. This approach leverages the Transformer's global attention mechanism to capture kilometer-level signal correlations, combined with GNN's graph-based learning to analyze centimeter-level signal propagation characteristics, achieving cross-scale feature fusion in the spatial dimension.

[0048] 2. The first attention-weighted signal balancing algorithm to eliminate intensity deviation through feature space mapping.

[0049] 3. Enhance robustness in a multi-physics field coupling environment, build an electromagnetic-spatial joint representation learning framework, and implement multipath propagation modeling through spatiotemporal graph convolution.

[0050] 4. Develop a dynamic parameter transfer learning mechanism.

[0051] 5. Innovatively introduce a meta-learning optimizer to achieve the co-evolution of model parameters and environmental characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0053] Figure 1 This is a flow chart of the steps of the method for eliminating the near-far effect of the broadcast radio positioning system described in the present invention. DETAILED DESCRIPTION

[0054] The present invention proposes a method for eliminating the near-far effect of a broadcast radio positioning system, a storage medium and a terminal device, aiming to design a method for eliminating the near-far effect of a broadcast radio positioning system based on a Transformer-GNN model.

[0055] The method for eliminating the near-far effect of the broadcast radio positioning system proposed by the present invention will be described in detail below in a specific embodiment:

[0056] Example 1: A method for eliminating the near-far effect of a broadcast radio positioning system, such as Figure 1 As shown, the following steps are included:

[0057] S1: Multi-source data acquisition and preprocessing: Collect radio ground base station data, radio positioning receiver data, interference source data, and reflection source data; perform signal alignment, feature standardization, environmental perception fusion, and outlier repair on the data;

[0058] S2: Construct a dynamic graph data structure: Define ground base stations, receivers, interference sources, and fictitious multipath reflection nodes as graph nodes. Based on the signal propagation relationship, construct ground station-receiver connection edges, interference source-receiver connection edges, and spatiotemporal trajectory connection edges. Design a dynamic update mechanism to achieve real-time updates of node positions and edge weights.

[0059] S3: Build a Transformer-GNN joint model: Implement local-global joint modeling by alternating stacked GNN layers and Transformer layers, where:

[0060] The GNN layer defines a message function based on the node type and extracts local topological features;

[0061] The Transformer layer introduces edge attribute encoding and captures global signal correlation through sparse attention mechanism;

[0062] Dynamically fuse GNN and Transformer features using residual connections and gating mechanisms;

[0063] S4: Model training and optimization: A phased training strategy is adopted, first pre-training the GNN layer, and then jointly training the Transformer-GNN model; dynamic weight adjustment (AdaLoss) is used to balance the localization loss and interference suppression loss;

[0064] S5: Model evaluation and adaptive optimization: Evaluate positioning accuracy based on root mean square error (RMSE) and mean absolute error (MAE), adjust model structure parameters based on the evaluation results, and use data augmentation to improve generalization capabilities.

[0065] Furthermore, in step S2:

[0066] Node types include:

[0067] For radio ground base station nodes, the feature vector includes three-dimensional coordinates, transmission power, carrier frequency and pseudo-random code number;

[0068] Radio positioning receiver node, the feature vector includes real-time position, velocity, received signal strength vector and carrier-to-noise ratio;

[0069] Interference source node, the feature vector contains the interference location, power, bandwidth and type code;

[0070] The imaginary multipath reflection node, the feature vector includes the reflection surface coordinates, incident angle, reflection coefficient and path delay;

[0071] Edge weights are calculated as follows:

[0072] Ground station-receiver side: based on the free space path loss formula and obstacle attenuation model;

[0073] Interference-receiver edge: based on the correlation between spectrum overlap integral and spatial distance;

[0074] Spatiotemporal trajectory edges: based on time decay factors or prediction confidence scores.

[0075] Furthermore, the dynamic update mechanism of step S2 includes:

[0076] The update frequency is adjustable from 50ms to 5000ms, with the default being 3000ms;

[0077] The reconfiguration mechanism is triggered when a new interference source is detected or the signal is lost. The reconfiguration conditions include:

[0078] The spectrum energy mutation exceeds the threshold of 3 times the background noise;

[0079] No signal is detected for N consecutive epochs, where N is dynamically adjustable and the default value is 5.

[0080] Furthermore, the construction of the Transformer-GNN joint model in step S3 further includes:

[0081] The GNN layer uses a graph convolutional network (GCN) or a graph attention network (GAT) to reduce computational complexity through neighbor sampling or TopK aggregation;

[0082] The Transformer layer uses a sparse attention mechanism to encode edge attributes into position embeddings;

[0083] Alternately stack GNN layers and Transformer layers 3 to 5 times, in the order of GNN→Transformer loop structure.

[0084] Furthermore, the residual connection and gating mechanism are implemented as follows:

[0085] The residual connection directly adds the GNN output features to the Transformer output features;

[0086] The gating mechanism dynamically controls the feature fusion weights based on GRU, and selects retained features through update gates and reset gates.

[0087] Furthermore, the dynamic weight adjustment (AdaLoss) in step S4 satisfies:

[0088] The joint loss function is ,in, is the positioning loss, To suppress the loss of interference, the weight λ is dynamically adjusted during the training process;

[0089] The training data is divided into training set, validation set and test set according to the ratio of 6:3:1, and GPU is used to accelerate parallel training.

[0090] Furthermore, the adaptive optimization in step S5 includes:

[0091] Adjust the number of GNN layers, number of Transformer heads, and hidden layer dimensions based on RMSE and MAE evaluation results;

[0092] Enhance the training set by rotating and scaling data to improve the model's adaptability to dynamic environments.

[0093] In order to solve the above technical problems, the present invention further proposes a computer-readable storage medium storing a computer program, which implements the above-mentioned method for eliminating the near-far effect when executed by a processor.

[0094] Example 2: A method for eliminating the near-far effect in a broadcast radio positioning system is based on the complementarity of the Transformer model and the GNN model in information processing mechanism and modeling capabilities, and based on global-local joint modeling and end-to-end learning capabilities. A Transformer-GNN near-far effect elimination model is constructed to improve the overall positioning performance of the system.

[0095] The Transformer-GNN model combines the Transformer model with a Graph Neural Network (GNN). This combination is not simply a stacking of modules, but rather a collaborative optimization based on their mechanisms. Addressing the near-far effect in radio positioning systems requires not only addressing the local issue of signals from nearby ground base stations potentially drowning out those from more distant ones, but also systemic considerations such as the radio positioning receiver's ability to capture and track signals from all ground base stations in the radio system, the effective geometric distribution of ground base stations, and the positioning accuracy of the radio positioning receiver. Research and analysis have shown that GNNs primarily target graph-structured data. By propagating and aggregating information between nodes and their neighbors, they can deeply explore the local topology of the graph and the local connectivity between nodes. The core advantage of the Transformer model is its self-attention mechanism, which enables it to capture global information, establish long-range dependencies in sequential data, and analyze global correlations between elements. Therefore, the Transformer-GNN model, formed by the synergistic combination of the two, combines a global perspective with detailed control, effectively addressing the near-far issue.

[0096] The Transformer-GNN model implements the following approach: GNNs are first used to perform preliminary processing on the graph-structured data of a broadcast radio positioning system, extracting local signal features. Through a message-passing mechanism, they accurately model local topological relationships (such as user-beacon distance attenuation and user-interference coupling strength). These processed features are then fed into a Transformer, which, through a self-attention mechanism, further analyzes and integrates the signal features from a global perspective, identifying cross-regional signal correlations (such as the co-localization patterns of distant beacons). This addresses the shortcomings of GNN modeling and enhances the model's global optimization capabilities. The GNN and Transformer share node embeddings, enabling joint training for end-to-end optimization. Their feature representations reinforce each other, forming a hierarchical information extraction process. Through model training and optimization, the weights for processing near- and far-field signals are adjusted, effectively eliminating the near-far effect and improving the positioning accuracy of a broadcast radio positioning system receiver for devices transmitting from ground stations at different distances.

[0097] The specific steps include:

[0098] Step S1, multi-source data collection and preprocessing:

[0099] The collected multi-source data includes: radio ground base station data: coordinates, transmission power, signal spectrum characteristics (pseudo-random code, carrier frequency, etc.); radio positioning receiver: real-time position, received signal spectrum (signal strength, carrier-to-noise ratio), motion status (speed, acceleration), etc.; interference source data: spectrum scanning results (interference center frequency, bandwidth, power), dynamic position, etc.; reflection source data: such as the position coordinates, incident angle, reflection coefficient and other data of the reflecting surface of buildings, ground, or water surface.

[0100] Data preprocessing includes radio ground base station data signal alignment: achieving multi-channel signal synchronization and signal timestamping; radio positioning receiver feature normalization processing: normalizing (Z-Score) features such as location information, signal power, and distance, and mapping them to the [0, 1] interval to improve model training stability; interference source data environment perception fusion: converting other collected auxiliary data or interference source data into obstacle masks for path loss correction; reflection source data missing and outlier detection and repair: identifying missing values ​​and outliers through statistical analysis, using interpolation to fill missing values ​​in the data, and repairing outliers through machine learning algorithms.

[0101] Step S2, construct dynamic graph data structure:

[0102] The construction of dynamic graph data structure mainly includes node definition, edge construction and dynamic graph update mechanism. ① Node definition takes each radio ground base station and radio positioning receiver as a node of the graph, and the node characteristics are the data collected and pre-processed in step 1; ② Nodes are connected to construct edges based on the signal propagation relationship, which includes three types of edge construction. The first type is the edge connecting the radio ground station and the receiver. When the radio positioning receiver receives a signal from a radio ground base station, an edge can be constructed between the two. The second type is the edge connecting the interference source and the radio receiver. When the radio positioning receiver receives a signal from a certain interference source, an edge can be constructed between the two. The third type is the spatiotemporal trajectory connection edge, which includes historical connection edges and connections formed by spatiotemporal trajectory deduction based on the motion trajectory of the radio positioning receiver. ③ Dynamic Update Mechanism: Mobile node positions and edge weights are updated using an adjustable spatiotemporal update mechanism, with an adjustable interval of 50ms to 5000ms, in steps of 10ms, and a default update frequency of 3000ms. The spatiotemporal association storage period ranges from 5 to 20 frames, in steps of 5 frames, with a default of 10 frames. When the system detects a new interference source or signal loss, the reconfiguration mechanism is initiated.

[0103] The detailed description of building dynamic graph data structure is as follows:

[0104] 1. Node parameter definition:

[0105] (1) Node type;

[0106] Radio ground base station node (GroundNode);

[0107] Physical meaning: Fixed or mobile radio ground base station transmission source.

[0108] Eigenvectors: ;

[0109] ① : 3D coordinates of the radio ground base station, unit: m;

[0110] ② : Transmit power, unit: dBm;

[0111] ③ : Carrier power, unit: MHz;

[0112] ④PRN: pseudo-random code number.

[0113] (2) Radio positioning receiver node (ReceiverNode);

[0114] Physical meaning: radio positioning receiver to be located.

[0115] Eigenvectors: ;

[0116] ① : Real-time position of radio positioning receiver, unit: m;

[0117] ② , : three-dimensional velocity, unit: m / s;

[0118] ③ : received signal strength vector, dimension = number of radio ground base stations;

[0119] ④ : Carrier-to-noise ratio (unit: dB-Hz).

[0120] (3) Interference Node;

[0121] Physical meaning: active or passive interference source.

[0122] Eigenvectors: ;

[0123] ① : interference source position, unit: m;

[0124] ② : Interference power, unit: dBm;

[0125] ③ : interference bandwidth, unit: MHz;

[0126] ④ : Interference source type code (e.g., 0: suppression interference, 1: deception interference; 2: new interference) (unit: dB-Hz).

[0127] (4) Fictitious multipath reflection node;

[0128] Physical meaning: The nodes that constitute the reflection surface that produces multipath effects.

[0129] Eigenvectors: ;

[0130] ① : coordinates of the center point of the reflecting surface, unit: m;

[0131] ② : incident angle of reflection source, unit: °;

[0132] ③ : reflection coefficient;

[0133] ④ : Delay of the reflected path relative to the direct path, unit: ns.

[0134] 2. Edge parameter definition;

[0135] (1) Ground-Receiver Link Edges

[0136] Weight calculation: Build a model based on path loss and obstacle attenuation. +

[0137] ① : Free space path loss value, calculated using the following formula.

[0138] , Unit: km, Unit: MHz.

[0139] ② : Obstacle attenuation, unit: dB, estimated through empirical model.

[0140] (2) Interference-Receiver Link Edges

[0141] Weight calculation: Build a model based on spectrum overlap and spatial correlation. ;

[0142] ① : Spectral overlap energy integral of the interference signal within the radio navigation signal band:

[0143] ;

[0144] : center frequency of radio navigation signal;

[0145] : radio navigation signal bandwidth;

[0146] : Power spectral density function of the interference signal;

[0147] ② : Weight coefficient, balancing the relationship between spectrum and spatial position, ∈[0,1];

[0148] ③ : The real-time distance between the interference source l and the radio receiver i, unit: m.

[0149] (2) TemporalandspatialdeductionEdges;

[0150] ① Receiver-RHistoryEdges;

[0151] Connection method: Connect the current node of the radio receiver with the node at the past (T) time points.

[0152] Weight calculation: Calculate weight based on time decay factor, , is the attenuation coefficient, ranging from 0.1 to 0.5, and t is the time.

[0153] ②DeductionEdges;

[0154] Generation method: Use algorithms such as LSMT to predict the estimated time position of the radio receiver and generate imaginary connection edges.

[0155] Weight calculation: Score the confidence of the prediction:

[0156] · ;

[0157] in, : The root mean square error between the predicted position and the actual position; Real-time adjustments are made based on real-time environmental changes and the maximum tolerable positioning error.

[0158] 3. Dynamic update mechanism:

[0159] (1) Adjustable spatiotemporal update mechanism;

[0160] Mobile node positions and edge weights are updated using an adjustable spatiotemporal update mechanism. The update frequency ranges from 50ms to 5000ms, with an adjustable step of 10ms and a default of 3000ms. The spatiotemporal association storage duration ranges from 5 to 20 frames, with an adjustable step of 5 frames and a default of 10 frames. When the system detects a new interference source or signal loss, the reconfiguration mechanism is initiated.

[0161] Update cycle: Typical value: 3000ms (suitable for scenarios with speeds no greater than Mach 3);

[0162] Calculation formula: .

[0163] (2) Reconstruction mechanism;

[0164] The reconstruction mechanism is activated in the following two situations: 1. New interference detection: detects whether there is a sudden change in spectrum energy, usually taking 3 times the background noise level as the detection threshold; 2. Detects whether the received radio ground base station signal is lost. If no signal is detected for N consecutive epochs, it is judged that the signal is lost. N can be dynamically adjusted, and the typical value is 5.

[0165] Step 3: Build the Transformer-GNN joint model:

[0166] The Transformer-GNN model is an optimized model that combines the Transformer and GNN. Therefore, the construction of the Transformer-GNN joint model includes the GNN layer design, the Transformer layer design, and the combined design of the Transformer and GNN models.

[0167] (1) GNN layer design:

[0168] A graph convolutional neural network (GCN) was chosen as the underlying architecture. Based on this, the number of layers and neurons per layer was determined, typically set to 2–4. The optimal number of neurons was determined through simulation experiments, with model complexity and performance as key metrics. The GNN layer extracts and updates local features from graph-structured data and propagates neighboring node information. During the information aggregation phase, node types are distinguished to accurately capture local physical constraints, such as the geometric distance between radio ground base stations and radio positioning receivers and the spectral impact range of interference sources. These local physical constraints are implemented through type-dependent message functions.

[0169] Message function definition: ;

[0170] in, : Feature vector of source node i in layer l;

[0171] : The feature vector of target node j in layer l, is a d-dimensional real vector space;

[0172] : edge features, is a real vector space of dimension e;

[0173] : The generated message vector is consistent with the node dimension.

[0174] (2) Transformer layer design:

[0175] Construct a Transformer model that includes a multi-head attention mechanism and a feedforward neural network. Set the number of attention heads (which can be set between 2 and 10) and the hidden layer dimension (which can be set between 64 and 1024). Determine the optimal hyperparameters through cross-validation. Use the output of the GNN layer as the input to the Transformer layer, which captures global information and adjusts the weights of signal features. Inject edge attributes (such as path loss) into the Transformer as positional encodings. Implement the edge feature encoding layer as follows:

[0176] Input: Original edge attributes , including path loss, interference source strength, spectrum overlap, etc.

[0177] Coding process:

[0178] : The weight matrix input to the hidden layer;

[0179] : hidden layer bias;

[0180] : Hidden layer attention space projection;

[0181] : Output bias.

[0182] (3) Transformer and GNN model combination design to eliminate the near-far effect:

[0183] The combined design of Transforme and GNN models eliminates the near-far effect. The combination of Transformer and GNN is not a simple stacking of modules, but is based on the complementarity of the two in information processing mechanisms and modeling capabilities, and achieves the purpose of eliminating the near-far effect through collaborative optimization. The combination of the two faces greater design difficulties and implementation challenges. It is necessary to solve system problems such as inconsistent feature representation distribution of GNN and Transformer, increased computational complexity, and difficulty in determining training stability weights. Careful design and engineering optimization are required to achieve the design goals. The specific steps for implementing the Transforme-GNN fusion construction in this patent are: ① Information input: At the beginning of the model, the node features are embedded; ② GNN layer design: Each layer of GNN processes the information of local neighbors, and GAT or GCN or other variants can be used, while considering the integration of edge features. Due to the high computational overhead of combining GNNs and Transformers, the multi-hop aggregation complexity of GNNs is O(K|E|). GNNs use neighbor sampling or TopK aggregation to reduce computational overhead. ③ Transformer layer design: Processing global information for all nodes requires custom position encoding to adapt to the graph structure. The Transformer's self-attention mechanism has a complexity of O(N²). Using sparse attention or local window attention on the Transformer reduces complexity. ④ Residual connections and fusion normalization: The feature representation distributions of GNNs and Transformers are inconsistent, and direct superposition can lead to performance degradation. This paper introduces residual connections and a gating mechanism to dynamically fuse features, and uses a joint loss function (localization loss + interference rejection loss) to achieve feature fusion normalization. The implementation of these methods requires extensive simulation and experimentation to determine the weight coefficient control during the dynamic fusion process and design appropriate residual connections, gating mechanisms, and joint loss functions, significantly improving the model's performance, stability, and flexibility. ⑤ Alternating stacking loop structure: GNN layers and Transformer layers are stacked alternately, typically 3-5 times, to fully integrate local and global information, forming a complete Transformer-GNN combined model. The alternation order is designed to be GNN layer → Transformer layer → GNN layer → Transformer layer, and this cycle repeats 3-5 times. ⑥ Information output: The features after alternating processing are aggregated to perform signal separation and interference suppression, eliminating the near-far effect.

[0184] Residual connection: In a residual connection, the input features are directly added to the output features through a skip connection to avoid losing key information during the feature transmission process. In the Transformer-GNN model, residual connections are used to fuse the features of GNN and Transformer.

[0185] GNN output features: ;

[0186] Transformer output features: Transformer );

[0187] Residual Connection: + .

[0188] Gating mechanism: This is primarily based on the GRU implementation, which uses dynamic weights to control feature fusion, selectively retaining or ignoring certain features. Key GRU parameters are the update gate, reset gate, candidate state, and output parameter. Dynamically controlling feature fusion through the update gate, reset gate, and candidate state allows for efficient feature alignment and fusion in the Transformer-GNN joint model.

[0189] Joint loss function: The joint loss function guides the feature alignment of GNN and Transformer through multi-task learning, while optimizing the localization task and interference suppression task.

[0190] The formula of the joint loss function is: ;

[0191] in: Positioning loss.

[0192] is the interference rejection loss.

[0193] λ is the weight of the positioning loss, and its value range is [0,1].

[0194] Step S4: Model training and optimization:

[0195] During Transformer-GNN model training, the optimization objectives of the GNN and the Transformer may conflict, leading to unstable training. This patent addresses this stability issue during model training through phased training and dynamic weight adjustment (AdaLoss). Model training is based on a parallel processing architecture and incorporates GPU acceleration to improve model training efficiency. ① Phased training: First, the GNN is trained independently to achieve good performance on graph-structured tasks. Then, based on the pre-trained GNN, the Transformer and GNN are jointly trained to adapt to global dependency tasks. ② Dynamic weight adjustment (AdaLoss): Weights are dynamically adjusted based on task losses to balance localization and interference suppression. The localization loss weights λ and the interference suppression loss weights 1−λ are dynamically adjusted during training. ③ The preprocessed graph data is partitioned into training, validation, and test sets in a 6:3:1 ratio. ④ The model is iteratively trained on the training set, calculating the loss at each iteration and updating the parameters through backpropagation. ⑤ Model performance is evaluated on the validation set, and hyperparameters are adjusted based on the results to prevent overfitting. ⑥ The results are validated on the test set.

[0196] Pre-training is performed in a laboratory calibration environment, and then adapted to specific scenarios through online learning.

[0197] Step S5: Model evaluation and adaptive optimization:

[0198] Evaluate whether the model trained in step 4 can eliminate the near-far effect. During the evaluation process, the positioning accuracy is evaluated by the radio positioning receiver terminal based on the radio ground base station signal with the near-far effect interference scenario.

[0199] Model evaluation includes two aspects: evaluation standard indicators and system positioning performance evaluation:

[0200] a. Evaluation criteria indicators:

[0201] The evaluation criteria include model accuracy (MSE), signal recognition rate (F1) and bit error rate (BER).

[0202] MSE= , is the true amplitude of the i-th signal, is the prediction result of the model for the i-th sample.

[0203] F1= ,F1-Score evaluates the model recognition accuracy and is an ,important indicator for measuring the model recognition accuracy.

[0204] in, is the accuracy, is the recall rate.

[0205] BER= ,BER is the core indicator for measuring the performance of digital communication ,systems, which represents the ratio of the number of error bits to the ,total number of bits during transmission.

[0206] in, It is the number of bits that are inconsistent with the original data at the sending end after demodulation at the receiving end. is the total number of bits transmitted by the sender.

[0207] To verify the accuracy (signal reconstruction accuracy) and recognition rate (target signal detection capability) of the Transformer-GNN model in signal separation and near-far effect elimination tasks, the following experiments were designed:

[0208] Signal modulation: QPSK, 16-QAM, OFDM (different modulation complexity).

[0209] Noise environment: Add Gaussian white noise (SNR range: 0~20 dB).

[0210] Simulation of the near-far effect: The power of a short-range signal is 10-30 dB higher than that of a long-range signal.

[0211] Number of signal sources: 10, including 1 close-range signal.

[0212] Number of experiments: 100 times.

[0213] Experimental method: Signal recognition and near-far effect elimination and suppression are performed through traditional power control method, Transformer method, GNN method and Transformer-GNN method.

[0214] After 100 tests, the following test results were obtained:

[0215] Serial number Model MSE F1 BER (1E-3) 1 Traditional power control 0.151 0.73 15.3 2 Transformer 0.049 0.84 8.2 3 GNN 0.062 0.83 9.2 4 Transformer-GNN 0.03 0.91 5.1

[0216] Accuracy advantage:

[0217] The MSE (0.03) of Transformer-GNN is about 35% lower than that of pure Transformer, proving that the joint model can reconstruct the signal more accurately.

[0218] Improved recognition rate:

[0219] At low SNR (0~20 dB), the F1-score still reaches 0.91, significantly better than traditional methods.

[0220] Robustness verification:

[0221] In the extreme case where the near-far power difference is as large as 30 dB, the BER remains below 5.1e-3.

[0222] b. System positioning performance evaluation:

[0223] The positioning error changes of radio positioning receivers at different distances before and after model processing are compared to evaluate the effectiveness of eliminating the near-far effect.

[0224] The model performance evaluation standard is based on the positioning accuracy of the radio positioning terminal, and the evaluation indicators are the system positioning root mean square error (RMSE) and mean absolute error (MAE).

[0225] ;

[0226] ;

[0227] in, For the real location, For the predicted location.

[0228] After simulation, the Transformer-GNN model was used to suppress the near-far effect, and the system root mean square error (RMSE) was reduced to within 5 meters and the mean absolute error (MAE) was reduced to within 3 meters.

[0229] Optimization strategy: Based on the above evaluation results, structural parameters such as the number of layers and neurons in the GNN and Transformer layers are adjusted. Through extensive learning and testing, the system can acquire dynamic adaptive adjustment capabilities. By rotating and scaling the training data, the data volume is expanded and the model's generalization ability is improved.

[0230] Key points of the present invention:

[0231] The idea of ​​implementing spatial anti-interference based on the Transformer-GNN deep learning method is the key technical point that the present invention must first protect;

[0232] Combining Transformer and GNN to realize a new Transformer-GNN model, which is applicable to solving the problem of eliminating the near-far effect in radio positioning systems and is the key technical point to be protected in this invention;

[0233] The construction, training and optimization methods of the Transformer-GNN model are the key technical points that need to be protected in this invention.

[0234] Advantages: Innovative Fusion Architecture: This patent pioneers the integration of two cutting-edge technologies, Transformer and GNN, breaking the limitations of traditional single technologies. Transformer excels at capturing long-range dependencies, while GNN efficiently processes graph-structured data. The fusion of the two enables the model to capture global signal correlations and finely analyze local signal propagation characteristics in radio positioning. This cross-domain fusion architecture is leading in the navigation field. However, the two are not simply superimposed. During the joint model construction and model training process, extensive simulation and experimentation are required to achieve the dynamic fusion goal of complementing the advantages of the two.

[0235] Leading near-far effect processing capabilities: Conventional methods for handling the near-far effect are often limited by differences in signal strength, making it difficult to balance near and far targets. Leveraging the Transformer self-attention mechanism, this approach treats near and far signal characteristics equally, independent of signal strength. Global modeling exploits the complex relationship between near and far signals, combined with GNN's in-depth understanding of local signal propagation characteristics, to precisely eliminate the near-far effect, far surpassing traditional solutions in balancing near and far signals.

[0236] Real-time, high-precision positioning: Leveraging its robust near-far effect mitigation capabilities, this patented technology significantly improves the accuracy of radio positioning systems, particularly for long-range receivers. Theoretically, it is expected to reduce the root mean square error (RMSE) to less than 5 meters and the mean absolute error (MAE) to less than 3 meters, far exceeding the performance of similar positioning technologies. This technology provides reliable support for applications requiring extremely high positioning accuracy, such as autonomous driving and precision drone flight. While the Transformer-GNN model is complex, through algorithm optimization and GPU acceleration, it meets the stringent real-time requirements of radio navigation systems, providing users with accurate and timely location information.

[0237] Strong adaptability to complex environments: In complex electromagnetic environments, signal interference and multipath effects can severely impact navigation performance, and radio signals are susceptible to fluctuations due to environmental interference. This technology utilizes normalization, missing value and outlier handling during data preprocessing, and data augmentation during model training to ensure robust adaptability to complex data. Furthermore, its unique global and local information processing capabilities effectively address interference and multipath signals, ensuring stable and accurate elimination of near-far effects and maintaining high-precision positioning in complex environments—a feat unattainable for traditional navigation technologies. Traditional navigation algorithms experience wide fluctuations in positioning errors in such environments. Signal fluctuations can instantly increase positioning errors by 20 to 30 meters or even more, and recovery from these errors can be slow. The Transformer-GNN model, by continuously learning and analyzing global and local signal information, significantly reduces positioning error fluctuations. Even in the face of severe signal fluctuations, positioning errors can be rapidly recovered, demonstrating enhanced stability.

[0238] Flexible Model Optimization: The structural parameters of the patented Transformer-GNN model, such as the number of layers and neurons in the GNN and Transformer layers, the number of Transformer heads, and the dimensions of the hidden layers, can be flexibly adjusted, and a dynamic adjustment mechanism can be established through training. Through cross-validation and experimental optimization, the optimal model configuration can be customized based on different application scenarios, hardware conditions, and data characteristics, enabling the technology to achieve optimal performance under different requirements, demonstrating a high degree of flexibility and optimizability.

[0239] Embodiment 3: A broadcast radio positioning system terminal device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the program according to the method for eliminating the near-far effect as described in embodiment 1.

[0240] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for eliminating the near-far effect of a broadcast radio positioning system, characterized in that: The following steps are involved: Step S1: Multi-source data acquisition and pre-processing: acquiring radio ground base station data, radio positioning receiver data, interference source data, and reflection source data; Perform signal alignment, feature standardization, environmental perception fusion, and outlier repair on the data; Step S2: Construct a dynamic graph data structure: define ground base stations, receivers, interference sources, and fictitious multipath reflection nodes as graph nodes, construct ground station-receiver connection edges, interference source-receiver connection edges, and spatiotemporal trajectory connection edges based on signal propagation relationships, and design a dynamic update mechanism to achieve real-time updates of node positions and edge weights; Step S3: Construct a Transformer-GNN joint model: Implement local-global joint modeling by alternating stacked GNN layers and Transformer layers, where: The GNN layer defines a message function based on the node type and extracts local topological features; The Transformer layer introduces edge attribute encoding and captures global signal correlation through sparse attention mechanism; Dynamically fuse GNN and Transformer features using residual connections and gating mechanisms; Step S4: Model training and optimization: A phased training strategy is adopted, first pre-training the GNN layer, and then jointly training the Transformer-GNN model; dynamic weight adjustment is used to balance the localization loss and interference suppression loss; Step S5: Model evaluation and adaptive optimization: Evaluate the positioning accuracy based on the root mean square error and mean absolute error, adjust the model structure parameters according to the evaluation results, and use data enhancement to improve the generalization ability.

2. The method for eliminating the near-far effect of a broadcast radio positioning system according to claim 1, characterized in that: In the step S2: Node types include: For radio ground base station nodes, the feature vector includes three-dimensional coordinates, transmission power, carrier frequency and pseudo-random code number; Radio positioning receiver node, the feature vector includes real-time position, velocity, received signal strength vector and carrier-to-noise ratio; Interference source node, the feature vector contains the interference location, power, bandwidth and type code; The imaginary multipath reflection node, the feature vector includes the reflection surface coordinates, incident angle, reflection coefficient and path delay; Edge weights are calculated as follows: Ground station-receiver side: based on the free space path loss formula and obstacle attenuation model; Interference-receiver edge: based on the correlation between spectrum overlap integral and spatial distance; Spatiotemporal trajectory edges: based on time decay factors or prediction confidence scores.

3. The method for eliminating the near-far effect of a broadcast radio positioning system according to claim 1, wherein: The dynamic update mechanism of step S2 includes: The update frequency is adjustable from 50ms to 5000ms; The reconfiguration mechanism is triggered when a new interference source is detected or the signal is lost. The reconfiguration conditions include: The spectrum energy mutation exceeds the threshold of 3 times the background noise; No signal is detected for N consecutive epochs, where N is dynamically adjustable and the default value is 5.

4. The method for eliminating the near-far effect of a broadcast radio positioning system according to claim 1, wherein: The construction of the Transformer-GNN joint model in step S3 further includes: The GNN layer uses graph convolutional networks or graph attention networks to reduce computational complexity through neighbor sampling or TopK aggregation; The Transformer layer uses a sparse attention mechanism to encode edge attributes into position embeddings; Alternately stack GNN layers and Transformer layers 3 to 5 times, in the order of GNN→Transformer loop structure.

5. The method for eliminating the near-far effect of a broadcast radio positioning system according to claim 4, characterized in that: The residual connection and gating mechanism are implemented as follows: The residual connection directly adds the GNN output features to the Transformer output features; The gating mechanism dynamically controls the feature fusion weights based on GRU, and selects retained features through update gates and reset gates.

6. The method for eliminating the near-far effect of a broadcast radio positioning system according to claim 1, characterized in that: The dynamic weight adjustment in step S4 satisfies: The joint loss function is ,in, is the positioning loss, To suppress the loss of interference, the weight λ is dynamically adjusted during the training process; The training data is divided into training set, validation set and test set according to the ratio of 6:3:1, and GPU is used to accelerate parallel training.

7. The method for eliminating the near-far effect of a broadcast radio positioning system according to claim 1, wherein: The adaptive optimization of step S5 includes: Adjust the number of GNN layers, number of Transformer heads, and hidden layer dimensions based on RMSE and MAE evaluation results; Enhance the training set by rotating and scaling data to improve the model's adaptability to dynamic environments.

8. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the method for eliminating the near-far effect described in any one of claims 1 to 7 is implemented.

9. A broadcast radio positioning system terminal device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the processor executes the program, the method for eliminating the near-far effect described in any one of claims 1 to 7 is implemented.

Images