Physical layer fusion network construction method

Abstract

The invention discloses a physical layer fusion network construction method, which comprises the following steps of: acquiring physical layer feature data and preprocessing the physical layer feature data; constructing a node feature matrix and an edge feature matrix, and generating an initial graph structure; generating a position coding matrix by using the adjacent matrix; constructing a view pair, executing self-supervised graph pre-training, and generating a communication node initial representation; executing a local message passing operation, and obtaining a local context node representation of the communication node; performing global attention modeling; calculating a connection weight between the communication nodes, updating a graph connection relationship, and generating a dynamically evolved network graph structure; and performing optimal path planning and communication link selection to form a networking connection scheme. According to the method, physical layer multi-mode features and a graph neural network modeling method are fused, a self-adaptive high-stability topological structure is constructed, and the method has the advantages of being high in generalization, fast in response and accurate in perception.

Description

Technical Field

[0001] The present invention relates to the field of intelligent optimization of wireless communication, and particularly to a method for constructing a physical layer fusion network. Background Art

[0002] In a wireless communication system, the construction of a network topology is a key link to ensure communication quality and system stability. Traditional network construction methods mostly rely on static policy settings at the network layer or simply perform link determination based on a small number of physical layer metrics, making it difficult to accurately characterize the actual communication capabilities between nodes in a complex environment. With the development of diverse communication environments and the trend of device heterogeneity, using only a single physical layer metric is difficult to comprehensively reflect the link quality between nodes, resulting in the constructed topology showing poor robustness and adaptability in the face of dynamic factors such as channel fluctuations and node movements.

[0003] In recent years, graph neural networks have gradually been applied to the field of network structure modeling. Some studies have attempted to introduce graph structures into the modeling process of communication networks, leveraging the high-order relationships between node features and graph structures to achieve more effective network construction strategies. However, most existing methods focus on static graph modeling or explicit graph learning stages, have not fully integrated the rich physical layer feature information in wireless communication, often ignore the modeling capabilities of node position structural encoding and cross-frequency domain connection states, and lack a targeted self-supervised pre-training mechanism in the graph learning process, resulting in limited expressive ability of node representations and being unable to effectively adapt to the dynamic changes of the actual communication environment.

[0004] Existing physical layer network modeling technologies have limitations to varying degrees in aspects such as multi-modal feature fusion, self-supervised learning ability, dynamic structure adaptability, and path planning mechanisms. There is an urgent need to propose a new method that integrates physical layer features, multi-stage graph modeling, and adaptive optimization strategies to improve the network construction intelligence level and communication robustness of wireless networks in dynamic complex environments. Summary of the Invention

[0005] An object of the present invention is to propose a method for constructing a physical layer fusion network. The present invention integrates physical layer multi-modal features and graph neural network modeling methods, combines a self-supervised graph pre-training mechanism, and constructs a dynamically evolving network topology with adaptive ability and physical layer perception ability, which can achieve high-precision connection relationship modeling and optimal path planning in a complex and changeable wireless channel environment, and has the advantages of strong generalization ability, high network stability, fast network construction response speed, and high link selection accuracy.

[0006] A method for constructing a physical layer fusion network according to an embodiment of the present invention includes the following steps: Collect physical layer feature data of each communication node in the wireless communication system and perform preprocessing; Construct a node feature matrix and an edge feature matrix based on the preprocessed physical layer feature data, generate an adjacency matrix according to the physical connection status between communication nodes, and generate an initial graph structure at the same time; Generate a position encoding matrix using the adjacency matrix; Construct view pairs based on the initial graph structure, perform self-supervised graph pre-training, and generate initial representations of communication nodes through graph contrast learning between the main encoder and the target encoder; Perform local message passing operations based on the initial representations of communication nodes to obtain local context node representations of communication nodes; Concatenate the local context node representations with the corresponding position encoding matrix, perform global attention modeling, and generate updated node representations that fuse local and global information; Calculate the connection weights between communication nodes based on the updated node representations, construct a weighted adjacency matrix, and update the graph connection relationship according to the weighted adjacency matrix to generate a dynamically evolving network graph structure; Based on the network graph structure, perform optimal path planning and communication link selection to form a network construction connection plan at the current moment. When it is detected that the positions of communication nodes change, the physical layer feature data fluctuates abnormally, or the connection weights decrease, re-execute the above steps to achieve the adaptive construction and dynamic optimization of the network graph structure.

[0007] Optionally, the physical layer feature data includes channel state information values, received signal strength, signal-to-noise ratio, channel phase, frequency offset, and synchronization timestamps. The preprocessing includes outlier detection and removal, missing value filling, phase unwrapping and alignment, dimension unification, and normalization.

[0008] Optionally, the generation of the initial graph structure includes: Set the number of communication nodes in the wireless communication system to a fixed value. Each communication node has preprocessed physical layer feature data. Combine the feature data of all communication nodes in the order of node numbers to form a node feature matrix. Each row in the node feature matrix represents the complete physical layer feature entries corresponding to a communication node; Construct an adjacency matrix based on whether there is a physical connection relationship between communication nodes. The specific rules are: When the channel state information value between two communication nodes is greater than or equal to a preset connection threshold value, it is considered that there is a communication link between the two communication nodes, and the corresponding position in the adjacency matrix is assigned a value of 1; If the channel state information value is less than the preset connection threshold value, it is regarded that there is no stable communication path between the two nodes, and the corresponding position in the adjacency matrix is assigned a value of zero; The connection threshold value is set as a fixed constant according to the system environment; For all pairs of communication nodes marked as connected in the adjacency matrix, extract edge feature data, and combine the edge features of all connected node pairs in order to generate an edge feature matrix. Each row of the edge feature matrix corresponds to an established physical connection path; The edge feature data includes three types of indicators: channel state information difference, received signal strength difference, and channel phase difference, which describe the physical connection characteristics between communication nodes; Use the node feature matrix, adjacency matrix, and edge feature matrix as the input for constructing the graph structure, and define the initial graph structure in the form of a graph structure triple, which serves as the basic structure input for graph neural network modeling and graph position encoding calculation.

[0009] Optionally, the generation of the position encoding matrix includes: Construct a degree matrix based on the adjacency matrix to represent the connection strength of each node. Each element on the main diagonal of the degree matrix is equal to the number of connections of the corresponding node, and the remaining elements are zero; Construct the Laplacian matrix of the graph according to the element-by-element difference between the adjacency matrix and the degree matrix. By performing eigenvalue decomposition on the Laplacian matrix, obtain the eigenvector matrix and the eigenvalue diagonal matrix, and select the first several low-frequency eigenvectors as the Laplacian position encoding vectors of the communication nodes to capture the frequency domain structure information of the nodes; Further calculate the shortest path distance between each pair of communication nodes, construct a path distance matrix. Each element of the path distance matrix represents the number of hops required between two nodes, and perform dimensionality reduction processing on the path information of each node through a non-linear mapping function to form a path position encoding vector; Calculate the transition probability between communication nodes based on the adjacency matrix, construct a transition probability matrix, which represents the possibility of any node jumping to its adjacent node. Based on the transition probability matrix, statistically calculate the frequency information of each node being visited through multiple rounds of random walks to generate a random walk position encoding vector; Concatenate the Laplacian position encoding vector, path position encoding vector, and random walk position encoding vector in the column dimension to obtain a complete position encoding matrix. Each row in the position encoding matrix corresponds to a structural position vector of a communication node.

[0010] Optionally, the generation of the initial representation of the communication nodes includes: Based on the initial graph structure, use the node feature matrix and adjacency matrix of the communication nodes as inputs to construct view pairs. The view pairs include an original view and a perturbed view. The original view is the complete graph structure, and the perturbed view is a transformed graph structure with perturbations applied to the complete graph structure. The perturbation methods include masking some feature dimensions of the communication nodes, randomly discarding several edge connection relationships in the adjacency matrix, and performing random subgraph sampling operations on the graph structure; The original view and the perturbed view are respectively input into two graph encoders with the same parameter structure but different update strategies. The first encoder serves as the main encoder, and the parameters of the main encoder are updated through backpropagation gradients. The second encoder serves as the target encoder, and the parameters of the target encoder are updated through the exponential moving average of the parameters of the main encoder. The two encoders respectively generate structural representation vectors for each communication node; Both the main encoder and the target encoder adopt a graph attention network to achieve weighted aggregation of information between communication nodes through a multi-head attention mechanism; The parameter update rule of the target encoder is: ; Among them, represents the parameters of the target encoder, represents the parameters of the main encoder, represents the momentum coefficient, represents the assignment operation, represents the momentum compensation factor; Taking the cosine similarity between the representation vectors of the same communication node under the two views as the positive sample pair, and the representation vector pairs between the remaining different nodes as the negative sample pairs, introducing a temperature coefficient to adjust the gradient amplitude, constructing a self-supervised contrast loss function to minimize the distance between positive samples and maximize the representational discrimination between negative samples: ; Among them, represents the self-supervised contrast loss function, represents the natural exponential function, represents the cosine similarity between vectors, represents the representation vector of the main encoder for the th node, represents the representation vector of the target encoder for the th node, represents the temperature coefficient, represents the total number of communication nodes, represents the representation vector of the target encoder for the th node; After the training is completed, the output of the main encoder is used as the initial state of the structural representation of all communication nodes to generate the initial representation of communication nodes.

[0011] Optionally, the process of obtaining the local context node representation of the communication node is: ; Among them, represents the local context node representation of the communication node at the th layer, denotes the degree matrix, denotes the adjacency matrix, denotes the weight matrix of the denotes the non - linear activation function, denotes the local context node representation of the communication node at the layer.

[0012] Optionally, the generation of the updated node representation that fuses local and global information includes: For each communication node, obtain the local context node representation and extract the position encoding vector corresponding to the position from the position encoding matrix, which respectively represent the structural semantics and position information of the communication node; Concatenate the local context node representation and the corresponding position encoding vector along the feature dimension to generate a fused input vector; Construct a fused input matrix from the fused input vectors of all communication nodes in the order of node indices; Perform global modeling, and through the multi - head attention mechanism, perform information interaction and weight adjustment on the fused input matrix of each communication node to generate an updated node representation that fuses local and global information.

[0013] Optionally, the generation of the dynamically evolving network graph structure includes: Extract the embedding vector of each communication node from the updated node representation, set it as the updated node representation of the corresponding communication node, and determine the set of node pairs whose edge weights need to be updated according to the established connection relationships in the initial adjacency matrix; For each pair of nodes to be updated, respectively extract the channel state information, received signal strength, and signal - to - noise ratio in the physical layer feature data, and perform weighted summation according to the set weighting coefficients to obtain the physical layer perception factor corresponding to this pair of nodes, where the physical layer perception factor reflects the stability of the physical connection and the channel quality; Calculate the similarity score between the updated node representations of the two communication nodes in the node pair using cosine similarity, and jointly input the similarity score and the corresponding physical layer perception factor into a non - linear gating function, and dynamically output the adjustment value of the connection weight according to the physical connection quality and the similarity score, as the updated weight of this edge; If the wireless communication system supports channel acquisition of multiple frequency bands or sub - carriers, extract the channel state information of the node pair at multiple frequency points in the frequency domain, form a frequency - domain feature vector, and perform normalization and weighted fusion processing on the frequency - domain feature vector to obtain a frequency - domain fusion eigenvalue, which reflects the average effect of cross - frequency channel consistency; Jointly calculate the frequency - domain fusion eigenvalue and the edge weight output by the non - linear gating function to generate the final connection weight of the node pair, and construct a weighted adjacency matrix with the final connection weights of all node pairs; Based on the original node set and the updated weighted adjacency matrix, reconstruct the graph structure to generate an evolving graph structure with physical layer perception ability and frequency response ability.

[0014] Optionally, the formation of the network connection scheme at the current moment includes: Based on the dynamic evolving network graph structure, extract the connection weights between communication node pairs in the weighted adjacency matrix; For each source node and target node pair, set the total path cost, where the total path cost is equal to the sum of the connection weights of all edges on the path, to measure the communication cost of the path. The smaller the total path cost, the better the overall communication quality of the path; Perform an initial shortest path search operation on the dynamic evolving network graph structure using the Dijkstra method to obtain the initial shortest path between the source node and the target node; After obtaining the initial shortest path, perform a perturbation operation on each intermediate node in the initial shortest path to construct several perturbed paths. Specifically: Truncate the initial shortest path to form a path prefix from the source node to a certain intermediate node, and find an alternative path segment after the intermediate node to generate a new path without repeating loops; For each perturbed path, recalculate the total path cost, merge all the generated new paths with the original path, and sort them in ascending order according to the total path cost; Select a path with the smallest total path cost from the sorting result as the optimal path for the current source-target node pair, and record the node sequence, connection edge set, and total cost in this path as the network connection scheme. The network connection scheme represents the optimal network connection structure in the current topological state and guides the signal routing selection and communication link establishment process; When the network graph structure or connection weights change, repeat the above path planning process to ensure that the network connection scheme continuously maintains the global optimal state in a dynamic environment.

[0015] The beneficial effects of the present invention are: A method for constructing a physical layer fusion network proposed by the present invention fully integrates various physical layer feature data, including channel state information, received signal strength, signal-to-noise ratio, channel phase, frequency offset, and synchronization timestamp, etc. Through preprocessing methods such as normalization and phase alignment, high-quality node and edge feature representations are constructed. On the basis of maintaining the accuracy of physical perception, the stability and credibility of the modeling of the connection relationship between nodes are improved. Based on the constructed initial graph structure, combined with various position embedding methods such as Laplace features, shortest path coding, and random walk frequency, the relative positions and potential communication capabilities of communication nodes in the structure are systematically expressed, providing structural enhancement support for subsequent graph representation learning.

[0016] During the graph representation learning process, the present invention introduces a self-supervised graph pre-training mechanism, and conducts graph contrastive learning using a main encoder and a target encoder, effectively improving the discriminative ability and generalization ability of communication node representations. The local message passing operation is performed through a graph neural network to obtain context information, and then global attention modeling is carried out in combination with the position encoding vector, and finally a node update representation that fuses local perception and global structure is generated, enhancing the structural adaptability and information expression ability of the network in dynamic scenarios. At the same time, a dynamic edge weight update mechanism based on physical layer perception factors and node semantic similarity is innovatively introduced, and combined with cross-frequency domain feature fusion, an evolving graph structure with physical response ability and frequency consistency perception ability is formed.

[0017] In terms of path planning, the proposed method constructs a multi-path perturbation scheme on the dynamic evolving network structure, and realizes the optimal path selection through total cost evaluation, so as to still be able to respond quickly when the network structure changes or the connection weights fluctuate, and output a high-quality network construction connection scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings: Figure 1 is a flowchart of a method for constructing a physical layer fusion network proposed by the present invention; Figure 2 is a schematic diagram of generating an initial representation of a communication node by a self-supervised graph pre-training mechanism of a method for constructing a physical layer fusion network proposed by the present invention; Figure 3 is a flowchart of optimal path planning and selection of a network construction connection scheme of a method for constructing a physical layer fusion network proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0020] Refer to Figures 1 - 3 , a method for constructing a physical layer fusion network, includes the following steps: Collect the physical layer feature data of each communication node in the wireless communication system and perform preprocessing; Based on the preprocessed physical layer feature data, construct a node feature matrix and an edge feature matrix, generate an adjacency matrix according to the physical connection state between communication nodes, and at the same time generate an initial graph structure; Generate a position encoding matrix using the adjacency matrix; Construct view pairs based on the initial graph structure, perform self-supervised graph pre-training, and generate the initial representation of communication nodes through graph contrast learning between the main encoder and the target encoder; Perform local message passing operations based on the initial representation of communication nodes to obtain the local context node representation of communication nodes; Concatenate the local context node representation with the corresponding position encoding matrix, perform global attention modeling, and generate an updated node representation that fuses local and global information; Calculate the connection weights between communication nodes based on the updated node representation, construct a weighted adjacency matrix, and update the graph connection relationship according to the weighted adjacency matrix to generate a dynamically evolving network graph structure; Based on the network graph structure, perform optimal path planning and communication link selection to form a network construction connection plan at the current moment. When it is detected that the position of a communication node changes, the physical layer feature data has abnormal fluctuations, or the connection weight drops, re-execute the above steps to achieve the adaptive construction and dynamic optimization of the network graph structure.

[0021] The present invention breaks through the limitations of relying on a single indicator or static structure in traditional network construction methods by constructing a network construction method that integrates multi-dimensional feature information of the physical layer and a graph structure modeling mechanism, and realizes adaptive topology evolution for the actual channel environment of communication nodes. This method can perceive the multi-dimensional physical connection state between communication nodes, and through a dynamic graph construction and update mechanism, quickly respond to node state changes and channel fluctuations, effectively improving the stability, robustness, and transmission efficiency of network construction, and is applicable to high-dynamic communication networks in heterogeneous edge device environments.

[0022] In this embodiment, the physical layer feature data includes channel state information value, received signal strength, signal-to-noise ratio, channel phase, frequency offset, and synchronization timestamp, and the preprocessing includes outlier detection and removal, missing value filling, phase unwrapping and alignment, dimension unification, and normalization.

[0023] In the acquisition of physical layer feature data of the present invention, key communication parameters including channel state information value, received signal strength, signal-to-noise ratio, and channel phase are introduced, and time series information such as frequency offset and frame synchronization is supplemented, making the collected data more comprehensive in dimension and stronger in channel characterization ability. Through multi-step preprocessing operations such as anomaly detection, missing value filling, and dimension unification, the feature quality is improved, the interference caused by sensing errors is avoided, and the accuracy and consistency of subsequent feature expressions are ensured, laying a high-quality data foundation for node modeling and connection weight evaluation.

[0024] In this embodiment, the generation of the initial graph structure includes: Set the number of communication nodes in the wireless communication system to a fixed value. Each communication node has pre - processed physical layer feature data. Combine the feature data of all communication nodes in the order of node numbers to form a node feature matrix. Each row in the node feature matrix represents the complete physical layer feature entries corresponding to a communication node. Construct an adjacency matrix based on whether there is a physical connection relationship between communication nodes. The specific rules are as follows: When the channel state information value between two communication nodes is greater than or equal to a preset connection threshold value, it is considered that there is a communication link between the two communication nodes, and the corresponding position in the adjacency matrix is assigned a value of 1. If the channel state information value is less than the preset connection threshold value, it is regarded that there is no stable communication path between the two nodes, and the corresponding position in the adjacency matrix is assigned a value of zero. The connection threshold value is set as a fixed constant according to the system environment. For all communication node pairs marked as connected states in the adjacency matrix, extract edge feature data, and combine the edge features of all connected node pairs in order to generate an edge feature matrix. Each row of the edge feature matrix corresponds to an established physical connection path. The edge feature data includes three types of indicators: the difference in channel state information, the difference in received signal strength, and the difference in channel phase, which describe the physical connection characteristics between communication nodes. Take the node feature matrix, adjacency matrix, and edge feature matrix as the input for constructing the graph structure, and define the initial graph structure in the form of a graph structure triple, which is used as the basic structure input for graph neural network modeling and graph position encoding calculation.

[0025] In the process of constructing the initial graph structure of the present invention, a clear definition method of the node feature matrix and the edge feature matrix is introduced, and the connection threshold is set according to the channel state information to realize the structured expression of communication relationships. By performing three - dimensional difference modeling on the edge features, the ability to express the details of physical connection attributes is enhanced. Combined with the adjacency matrix generation mechanism, the physical perception accuracy of the initial structure of the network is ensured, providing accurate input for subsequent graph model reasoning and dynamic topology evolution, and improving the practical usability and robustness of the graph construction effect.

[0026] In this embodiment, the generated position encoding matrix includes: Construct a degree matrix based on the adjacency matrix, which is used to represent the connection strength of each node. Each element on the main diagonal of the degree matrix is equal to the connection number of the corresponding node, and the rest of the elements are zero. Construct the Laplacian matrix of the graph according to the element - by - element difference between the adjacency matrix and the degree matrix. Through eigenvalue decomposition of the Laplacian matrix, obtain the eigenvector matrix and the eigenvalue diagonal matrix, and select the first several low - frequency eigenvectors as the Laplacian position encoding vectors of the communication nodes to capture the frequency - domain structure information of the nodes. Further calculate the shortest path distance between each pair of communication nodes, construct a path distance matrix, where each element of the path distance matrix represents the number of hops required between two nodes, and perform dimensionality reduction processing on the path information of each node through a non-linear mapping function to form a path position encoding vector; Calculate the transition probability between communication nodes based on the adjacency matrix, construct a transition probability matrix, which represents the possibility of any node jumping to its adjacent node. Based on the transition probability matrix, statistically calculate the frequency information of each node being visited through multiple rounds of random walks to generate a random walk position encoding vector; Concatenate the three types of vectors, namely the Laplacian position encoding vector, the path position encoding vector, and the random walk position encoding vector, according to the column dimension to obtain a complete position encoding matrix, where each row in the position encoding matrix corresponds to the structural position vector of a communication node.

[0027] By integrating three types of position expressions, namely Laplacian position encoding, path distance encoding, and random walk position encoding, the present invention enables each communication node to possess three types of structural information: frequency domain structure, path reachability, and local traversal probability, thereby realizing a comprehensive modeling of the structural role of nodes in the network. This position encoding matrix enhances the global positioning ability of node representation, improves the discriminative ability of the subsequent fusion module to model the relationship between nodes, and provides strong structural support for updating node representation and path selection.

[0028] In this embodiment, the generation of the initial representation of communication nodes includes: Based on the initial graph structure, use the node feature matrix and adjacency matrix of communication nodes as inputs to construct view pairs, where the view pairs include an original view and a perturbed view. The original view is the complete graph structure, and the perturbed view is a transformed graph structure with perturbations applied to the complete graph structure. The perturbation methods include masking some feature dimensions of communication nodes, randomly discarding several edge connection relationships in the adjacency matrix, and performing random subgraph sampling operations on the graph structure; Input the original view and the perturbed view into two graph encoders with the same parameter structure but different update strategies respectively. The first encoder is used as the main encoder, and the parameters of the main encoder are updated through backpropagation gradients. The second encoder is used as the target encoder, and the parameters of the target encoder are updated through the exponential moving average of the parameters of the main encoder. The two encoders respectively generate structural representation vectors for each communication node; Both the main encoder and the target encoder adopt a graph attention network to achieve weighted aggregation of information between communication nodes through a multi-head attention mechanism; The parameter update rule of the target encoder is: ; Wherein, Represents the parameters of the target encoder, Represents the parameters of the main encoder, Represents the momentum coefficient, Represents the assignment operation, Represents the momentum compensation factor; Taking the cosine similarity between vectors represented by the same communication node under two views as the positive sample pair, and the vector pairs between the remaining different nodes as the negative sample pairs, introducing a temperature coefficient to adjust the gradient amplitude, constructing a self-supervised contrastive loss function to minimize the distance between positive samples and maximize the representational discrimination between negative samples: ; Among them, Represents the self-supervised contrastive loss function, Represents the natural exponential function, Represents the cosine similarity between vectors, Represents the representation vector of the main encoder for the th node, Represents the representation vector of the target encoder for the th node, Represents the temperature coefficient, Represents the total number of communication nodes, Represents the representation vector of the target encoder for the th node; After the training is completed, the output of the main encoder is used as the initial state of the structural representation of all communication nodes to generate the initial representation of the communication nodes.

[0029] The present invention introduces a self-supervised graph pre-training mechanism, constructs a structural perturbation view pair and generates the initial node representation through graph contrastive learning, effectively captures the graph structure semantics without a large amount of labeled data, the design of the collaborative update of the main encoder and the target encoder improves the stability and generalization ability of the node representation, and the multi-head attention mechanism further enhances the information aggregation effect. This method can more accurately depict the relationship structure between communication nodes and provide a high-quality starting state for the subsequent context modeling of graph neural networks.

[0030] In this embodiment, the process of obtaining the local context node representation of the communication node is as follows: ; Among them, Represents the local context node representation of the communication node at the th layer, Represents the degree matrix, Represents the adjacency matrix, Represents the weight matrix at the th layer, Represents the non-linear activation function, Indicates the local context node representation of the communication node at the layer.

[0031] The present invention uses a graph neural network for local message passing, utilizes the adjacency matrix and degree matrix to complete node - level feature aggregation, obtains node representations with local context structure semantics, and through the adjustment of non - linear activation functions and weight matrices, enables the model to adaptively strengthen or inhibit the information propagation path according to the adjacency relationship, retain effective neighbor features, while filtering redundant or irrelevant information, enhancing the expressive ability of communication node representations, and facilitating local relationship recognition under complex topological structures.

[0032] In this embodiment, the generation of updated node representations that fuse local and global information includes: For each communication node, obtain the local context node representation, and extract the position - encoding vector corresponding to the position from the position - encoding matrix, respectively representing the structural semantics and position information of the communication node; Concatenate the local context node representation and the corresponding position - encoding vector along the feature dimension to generate a fused input vector; Construct a fused input matrix from the fused input vectors of all communication nodes in the order of node indices; Perform global modeling, and through the multi - head attention mechanism, perform information interaction and weight adjustment on the fused input matrix of each communication node to generate updated node representations that fuse local and global information.

[0033] Based on fusing the local representation of nodes and structural position encoding, the present invention uses the multi - head attention mechanism for global modeling operations, fully explores the potential dependence relationships of communication nodes within the entire graph, enhances the input information density through feature concatenation, the fused input matrix can capture multi - scale structural semantics, improves the discriminability of updated node representations and the overall topological modeling ability. This strategy realizes local - global collaborative perception, and enhances the adaptability and robustness of graph structure expression and subsequent edge weight updates.

[0034] In this embodiment, the generation of a dynamically evolving network graph structure includes: Extract the embedding vector of each communication node from the updated node representation, set it as the updated node representation of the corresponding communication node, and determine the set of node pairs whose edge weights need to be updated according to the established connection relationships in the initial adjacency matrix; For each node pair to be updated, respectively extract the channel state information, received signal strength, and signal - to - noise ratio in the physical - layer feature data, and perform weighted summation according to the set weighting coefficients to obtain the physical - layer perception factor corresponding to this node pair, where the physical - layer perception factor reflects the stability of the physical connection and the channel quality; The cosine similarity is used to calculate the similarity score between the updated node representations of two communication nodes in a node pair, and the similarity score and the corresponding physical layer perception factor are jointly input into a non - linear gating function. According to the physical connection quality and the similarity score, the adjustment value of the connection weight is dynamically output as the updated weight of this edge; If the wireless communication system supports channel acquisition of multiple frequency bands or sub - carriers, the channel state information of the node pair at multiple frequency points is extracted in the frequency domain range to form a frequency - domain feature vector. The frequency - domain feature vector is normalized and weighted fusion processed to obtain a frequency - domain fusion eigenvalue, which reflects the average effect of cross - frequency channel consistency; The frequency - domain fusion eigenvalue and the edge weight output by the non - linear gating function are jointly calculated to generate the final connection weight of the node pair, and the weighted adjacency matrix is constructed using the final connection weights of all node pairs; Based on the original node set and the updated weighted adjacency matrix, the graph structure is reconstructed to generate an evolving graph structure with physical layer perception ability and frequency response ability.

[0035] The present invention designs a non - linear gating function to dynamically adjust the connection weight by comprehensively considering the embedding vector similarity of communication nodes and the physical layer perception factor, realizing fine - grained modeling of edge relationships. At the same time, a frequency - domain fusion mechanism is introduced to enhance the perception ability of the graph construction for cross - frequency channel response characteristics. Based on the updated weighted adjacency matrix, the evolving graph structure is constructed, improving the sensitivity of the graph structure to channel changes and the adaptability of the network construction strategy. This method enhances the modeling effect of the topological structure on heterogeneous communication environments.

[0036] In this embodiment, the formation of the network construction connection scheme at the current moment includes: Based on the dynamic evolving network graph structure, the connection weights between communication node pairs in the weighted adjacency matrix are extracted; For each source node and target node pair, a total path cost is set. The total path cost is equal to the sum of the connection weights of all edges on the path, measuring the communication cost of the path. The smaller the total path cost, the better the overall communication quality of the path; The Dijkstra method is used to perform an initial shortest path search operation on the dynamic evolving network graph structure to obtain the initial shortest path between the source node and the target node; After obtaining the initial shortest path, a perturbation operation is performed on each intermediate node in the initial shortest path to construct several perturbed paths. Specifically: The initial shortest path is truncated, a path prefix is formed from the source node to a certain intermediate node, and an alternative path segment is found after the intermediate node to generate a new path without repeated loops; For each perturbed path, the total path cost is recalculated, all generated new paths and the original path are merged, and they are sorted in ascending order according to the total path cost; Select a path with the minimum total path cost from the sorting results as the optimal path for the current source-destination node pair, and record the node sequence, connection edge set, and total cost in this path as the network construction connection scheme. The network construction connection scheme represents the optimal network connection structure under the current topological state and guides the signal routing selection and communication link establishment process; When the network graph structure or connection weight changes, repeat the above path planning process to ensure that the network construction connection scheme continuously maintains the global optimal state in a dynamic environment.

[0037] The present invention constructs an optimal network construction path planning scheme driven by connection weights, dynamically screens the communication quality optimal path through the total path cost function, and combines path perturbation operations to improve path diversity and fault tolerance, ensuring the global optimality and link stability of the connection scheme in the dynamic evolution of the network. This path planning method has high adaptability and real-time performance, and is particularly suitable for dynamic network construction tasks in environments where devices move frequently or channels change violently.

[0038] Embodiment 1: To verify the feasibility of the present invention in implementation, the present invention is applied to the 5G micro base station communication system in a certain coastal city. The communication system has a high coverage area density, complex communication node distribution, and is frequently interfered by environmental factors such as building reflections, spectrum congestion, and multipath effects. The traditional static topology network construction method has obvious defects in such an environment, with slow network topology response, large fluctuations in link quality, and frequent misconnections and link jitters, seriously affecting communication reliability and scheduling efficiency.

[0039] The present invention is deployed in the wireless communication cluster system in the southern part of the city. Using micro base stations as communication nodes, physical layer feature data including channel state information, received signal strength, channel phase, frequency offset, and frame synchronization time is extracted through the underlying driver, and the collected data is uploaded to the edge computing platform. In the data center, a node feature matrix and an edge feature matrix are constructed through normalization and standardization processing, and an adjacency matrix is generated based on the connection threshold, further forming a graph structure input. After the graph construction is completed, self-supervised graph pre-training is performed to fuse the structural differences between nodes and guide the stable convergence of the graph neural network parameters. Subsequently, the system performs local message passing operations to obtain local context representations, and then combines graph position encoding for global information modeling to generate fused node representations, and finally calculates the edge weights and dynamically updates the topological structure.

[0040] After being generated, this topology is used in the path optimization process. It combines the Dijkstra method to perform initial path planning, and conducts multi-path perturbation and global path optimization, enabling the finally generated network connection scheme to have higher adaptability and stability. During the one-week actual operation, the system automatically triggers the topology reconstruction mechanism according to the actual link status every day, enabling the network to maintain a high connectivity rate and stability under different communication periods and environmental perturbations.

[0041] To verify the performance of the present invention in implementation, a comprehensive performance comparison was conducted with the traditional physical layer fusion network method.

[0042] Table 1 Summary of the experimental comparison between the present invention and the traditional physical layer fusion network method

[0043] As can be seen from Table 1, the present invention shows advantages over the traditional method in multiple key performance indicators, fully demonstrating the effectiveness and advancement of the proposed method in the actual communication network construction task.

[0044] In terms of the topology adaptation delay, the method of the present invention shortens the time required for network structure adjustment to 168 ms, compared with 342 ms of the traditional method, indicating that it has a faster network self-adaptation ability, especially suitable for dynamic communication environments with frequent node changes. In terms of the path planning success rate, which directly reflects the effectiveness of network construction, the present invention reaches 90.6%, higher than 82.1% of the traditional method, indicating that the proposed self-supervised representation strategy has outstanding effects in enhancing the global path awareness.

[0045] In terms of the edge weight update accuracy, the method of the present invention reduces the error to 0.023, compared with 0.082 of the traditional scheme, and the error decreases significantly, further verifying its robustness in physical layer perception and connection weight reconstruction. At the same time, the packet loss rate per unit time in terms of communication link stability decreases from 0.071 of the traditional method to 0.030, indicating that the constructed network has stronger anti-interference ability and transmission reliability.

[0046] In the multi-frequency domain consistency scoring dimension, the average consistency score after the present invention fuses the frequency domain information is 0.890, higher than 0.630 of the traditional method, reflecting a significant improvement in channel stability in supporting multi-band environments. The adaptive network construction frequency also increases from the traditional 1.4 times per hour to 4.2 times per hour, indicating that it has stronger dynamic perception and real-time response capabilities.

[0047] In terms of the global path reconstruction rate, the present invention reaches 84.5%, which is higher than 56.7% of the traditional method, further demonstrating its global adaptation ability to the evolution of network topology. In terms of the average total path cost, it decreases from 0.720 of the traditional method to 0.350, indicating that the selected path has lower communication costs and energy consumption levels.

[0048] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for constructing a physical layer convergence network, characterized in that It includes the following steps: Collect the physical layer feature data of each communication node in the wireless communication system and perform preprocessing; Based on the preprocessed physical layer feature data, construct a node feature matrix and an edge feature matrix, generate an adjacency matrix according to the physical connection status between communication nodes, and generate an initial graph structure at the same time; Generate a position encoding matrix using the adjacency matrix; Based on the initial graph structure, construct view pairs, perform self-supervised graph pre-training, and generate initial representations of communication nodes through graph contrast learning between the main encoder and the target encoder; Perform local message passing operations based on the initial representations of communication nodes to obtain local context node representations of communication nodes; Concatenate the local context node representations with the corresponding position encoding matrix, perform global attention modeling, and generate updated node representations that fuse local and global information; Calculate the connection weights between communication nodes based on the updated node representations, construct a weighted adjacency matrix, and update the graph connection relationship according to the weighted adjacency matrix to generate a dynamically evolving network graph structure; Based on the network graph structure, perform optimal path planning and communication link selection to form a network construction connection plan at the current moment. When it is detected that the position of a communication node changes, the physical layer feature data has abnormal fluctuations, or the connection weight decreases, re-execute the above steps to achieve the adaptive construction and dynamic optimization of the network graph structure.

2. The method for constructing a physical layer convergence network according to claim 1, wherein The physical layer feature data includes channel state information values, received signal strength, signal-to-noise ratio, channel phase, frequency offset, and synchronization timestamps. The preprocessing includes outlier detection and removal, missing value filling, phase unwrapping and alignment, dimension unification, and normalization.

3. A method for constructing a physical layer fusion network according to claim 1, characterized in that, The generation of the initial graph structure includes: Set the number of communication nodes in the wireless communication system to a fixed value. Each communication node has preprocessed physical layer feature data. Combine the feature data of all communication nodes in the order of node numbers to form a node feature matrix. Each row in the node feature matrix represents a complete physical layer feature entry corresponding to a communication node; Construct an adjacency matrix based on whether there is a physical connection relationship between communication nodes. The specific rule is: When the channel state information value between two communication nodes is greater than or equal to a preset connection threshold value, it is considered that there is a communication link between the two communication nodes, and the corresponding position in the adjacency matrix is assigned a value of 1; If the channel state information value is less than the preset connection threshold value, it is regarded that there is no stable communication path between the two nodes, and the corresponding position in the adjacency matrix is assigned a value of zero; The connection threshold value is set as a fixed constant according to the system environment; For all communication node pairs marked as connected states in the adjacency matrix, extract edge feature data, and combine the edge features of all connected node pairs in order to generate an edge feature matrix. Each row in the edge feature matrix corresponds to an established physical connection path; The edge feature data includes three types of indicators: channel state information difference, received signal strength difference, and channel phase difference, which describe the physical connection characteristics between communication nodes; Use the node feature matrix, adjacency matrix, and edge feature matrix as the input for constructing the graph structure, and define the initial graph structure in the form of a graph structure triple as the basic structure input for graph neural network modeling and graph position encoding calculation.

4. A method for constructing a physical layer fusion network according to claim 1, characterized in that The generated position encoding matrix includes: Construct a degree matrix based on the adjacency matrix to represent the connection strength of each node. Each element on the main diagonal of the degree matrix is equal to the number of connections of the corresponding node, and the remaining elements are zero. Construct the Laplacian matrix of the graph according to the element-wise difference between the adjacency matrix and the degree matrix. By performing eigenvalue decomposition on the Laplacian matrix, obtain the eigenvector matrix and the eigenvalue diagonal matrix, and select the first several low-frequency eigenvectors as the Laplacian position encoding vectors of the communication nodes to capture the frequency domain structure information of the nodes. Further calculate the shortest path distance between each pair of communication nodes to construct a path distance matrix. Each element of the path distance matrix represents the number of hops required between two nodes, and perform dimensionality reduction processing on the path information of each node through a non-linear mapping function to form a path position encoding vector. Calculate the transition probability between communication nodes based on the adjacency matrix to construct a transition probability matrix, which represents the possibility of any node jumping to its adjacent node. Based on the transition probability matrix, statistically analyze the frequency information of each node being visited through multiple rounds of random walks to generate a random walk position encoding vector. Concatenate the three types of vectors, namely the Laplacian position encoding vector, the path position encoding vector, and the random walk position encoding vector, according to the column dimension to obtain a complete position encoding matrix. Each row in the position encoding matrix corresponds to the structural position vector of a communication node.

5. The method for constructing a physical layer convergence network according to claim 1, wherein The generation of the initial representation of the communication nodes includes: Based on the initial graph structure, use the node feature matrix and the adjacency matrix of the communication nodes as inputs to construct view pairs. The view pairs include an original view and a perturbed view. The original view is the complete graph structure, and the perturbed view is a transformed graph structure with perturbations applied to the complete graph structure. The perturbation methods include masking some feature dimensions of the communication nodes, randomly discarding several edge connection relationships in the adjacency matrix, and performing random subgraph sampling operations on the graph structure. Input the original view and the perturbed view into two graph encoders with the same parameter structure but different update strategies respectively. The first encoder is used as the main encoder, and the parameters of the main encoder are updated through backpropagation gradients. The second encoder is used as the target encoder, and the parameters of the target encoder are updated through the exponential moving average of the parameters of the main encoder. The two encoders generate structural representation vectors for each communication node respectively. Both the main encoder and the target encoder adopt a graph attention network to achieve weighted aggregation of information between communication nodes through a multi-head attention mechanism. The parameter update rule of the target encoder is: ； Among them, represents the parameters of the target encoder, represents the parameters of the main encoder, represents the momentum coefficient, represents an assignment operation, represents the momentum compensation factor; Use the cosine similarity between the representation vectors of the same communication node under the two views as a positive sample pair, and the representation vector pairs between the remaining different nodes as negative sample pairs. Introduce a temperature coefficient to adjust the gradient amplitude, and construct a self-supervised contrast loss function to minimize the distance between positive samples and maximize the representational discrimination between negative samples. After the training is completed, use the output of the main encoder as the initial state of the structural representation of all communication nodes to generate the initial representation of the communication nodes.

6. A method for constructing a physical layer convergence network according to claim 1, characterized in that The generation of the updated node representation that fuses local and global information includes: For each communication node, obtain the local context node representation, and extract the position encoding vector corresponding to the position from the position encoding matrix, which respectively represent the structural semantics and position information of the communication node; Concatenate the local context node representation and the corresponding position encoding vector according to the feature dimension to generate a fused input vector; Construct a fused input matrix by arranging the fused input vectors of all communication nodes in the order of node indices; Perform global modeling, and through the multi-head attention mechanism, perform information interaction and weight adjustment on the fused input matrix of each communication node to generate an updated node representation that fuses local and global information.

7. A method for constructing a physical layer fusion network according to claim 1, wherein The generation of the dynamically evolving network graph structure includes: Extract the embedding vector of each communication node from the updated node representation, set it as the updated node representation of the corresponding communication node, and determine the set of node pairs whose edge weights need to be updated according to the established connection relationships in the initial adjacency matrix; For each node pair to be updated, respectively extract the channel state information, received signal strength, and signal-to-noise ratio in the physical layer feature data, and perform weighted summation according to the set weighting coefficients to obtain the physical layer perception factor corresponding to this node pair, where the physical layer perception factor reflects the stability of the physical connection and the channel quality; Calculate the similarity score between the updated node representations of the two communication nodes in the node pair using cosine similarity, and jointly input the similarity score and the corresponding physical layer perception factor into a non-linear gating function, and dynamically output the adjustment value of the connection weight according to the physical connection quality and the similarity score as the updated weight of this edge; If the wireless communication system supports channel acquisition of multiple frequency bands or subcarriers, extract the channel state information of the node pair at multiple frequency points in the frequency domain range to form a frequency domain feature vector, and perform normalization and weighted fusion processing on the frequency domain feature vector to obtain a frequency domain fusion eigenvalue, which reflects the average effect of cross-frequency channel consistency; Jointly calculate the frequency domain fusion eigenvalue and the edge weight output by the non-linear gating function to generate the final connection weight of the node pair, and construct a weighted adjacency matrix using the final connection weights of all node pairs; Based on the original node set and the updated weighted adjacency matrix, reconstruct the graph structure to generate an evolving graph structure with physical layer perception ability and frequency response ability.

8. A method for constructing a physical layer fusion network according to claim 1, characterized in that, The formation of the network construction connection plan at the current moment includes: Based on the dynamically evolving network graph structure, extract the connection weights between communication node pairs in the weighted adjacency matrix; For each source node and target node pair, set the total path cost; Perform an initial shortest path search operation on the dynamically evolving network graph structure using the Dijkstra method to obtain the initial shortest path between the source node and the target node; After obtaining the initial shortest path, perform a perturbation operation on each intermediate node in the initial shortest path to construct several perturbed paths. Specifically: Truncate the initial shortest path to form a path prefix from the source node to a certain intermediate node, and find an alternative path segment after the intermediate node to generate a new path without repeating loops; For each perturbed path, recalculate the total path cost, merge all the generated new paths with the original path, and sort them in ascending order according to the total path cost; Select a path with the minimum total path cost from the sorting results as the optimal path for the current source-destination node pair, and record the node sequence, connection edge set, and total cost in this path as the network construction connection plan. The network construction connection plan represents the optimal network connection structure under the current topological state and guides the signal routing selection and communication link establishment process; When the network graph structure or connection weight changes, repeat the above path planning process to ensure that the network construction connection plan continuously maintains the global optimal state in a dynamic environment.

Images