Point-to-point communication method and system based on self-organizing direct connection

Abstract

The invention provides a point-to-point communication method and system based on self-organizing direct connection, and the method comprises the steps: obtaining network state information in a plurality of time periods, obtaining a historical communication data set sequence, carrying out the construction of a communication topological graph, and carrying out the time sequence feature modeling, and obtaining a dynamic communication time-space diagram sequence; performing space-time correlation feature analysis and time sequence communication state evolution on the dynamic communication space-time diagram sequence to obtain an expected communication state data set corresponding to each communication link in a future period; determining a communication device group needing point-to-point communication, and performing multi-constraint adaptive path decision on a local topology link of the communication device group according to the expected communication data to obtain a standard communication link; and performing signal transmission among the communication equipment groups by using a standard communication link to obtain a received signal, performing time-frequency transformation, image denoising and feature enhancement on the received signal to obtain an enhanced frequency domain feature map, and performing inverse transformation on the enhanced frequency domain feature map to obtain an enhanced received signal.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a point-to-point communication method and system based on self-organizing direct connections. Background Technology

[0002] In complex environments such as mining sites, underground spaces, emergency rescue areas, and large industrial parks, communication equipment typically communicates collaboratively in the form of self-organizing networks to achieve point-to-point information exchange and status synchronization between devices. This type of load scenario is characterized by discrete node distribution, frequent dynamic changes in network topology, unstable link quality, and limited communication resources, which places high demands on the real-time performance, reliability, and adaptability of the communication system.

[0003] However, existing self-organizing point-to-point communication methods still have significant technical limitations when dealing with the aforementioned dynamic and complex communication environments. First, most solutions rely on instantaneous link quality or static network topology for communication path selection, lacking the ability to predict the future state evolution of communication links. This results in a lag in path decision-making, making it difficult to avoid potential link degradation or communication interruption risks in advance. Second, existing methods typically design network layer path planning and physical layer signal processing independently, failing to establish an effective linkage mechanism between link decision results and signal processing strategies. This makes it difficult to specifically enhance the quality of received signals under conditions of high noise, high attenuation, or high link costs. Finally, facing topology changes caused by frequent node additions, removals, or movements, traditional communication models have limited ability to model the temporal correlation of nodes, making it difficult to achieve stable and accurate communication state prediction while ensuring computational feasibility. This, in turn, restricts the improvement of the overall communication performance of self-organizing networks. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a point-to-point communication method and system based on self-organizing direct connections. It has the advantages of dynamic spatiotemporal graph prediction-driven, multi-constraint adaptive decision-making, and cross-layer intelligent signal enhancement. It solves the problems of unstable communication links, poor end-to-end transmission quality, and low communication efficiency caused by the lack of forward-looking perception, cross-layer collaborative mechanisms, and adaptive processing capabilities in self-organizing network scenarios with dynamic topology, complex channels, and limited resources.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: This invention provides a point-to-point communication method based on self-organizing direct connections, comprising the following steps: The network status information of each communication device in the self-organizing network is obtained in multiple time periods to obtain a historical communication dataset sequence. Based on the historical communication dataset sequence, a communication topology graph is constructed and a temporal feature model is performed to obtain a dynamic communication spatiotemporal graph sequence. Spatiotemporal correlation feature analysis and time-series communication state evolution are performed on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period. Based on real-time communication requests, the communication device group that needs point-to-point communication is determined, and multi-constraint adaptive path decision is performed on the local topology links of the communication device group according to the expected communication data to obtain the standard communication link. Point-to-point signal forwarding is performed between each communication device in the communication equipment group according to the standard communication link. The destination device in the communication equipment group receives the signal to obtain the received signal. The received signal is then subjected to time-frequency transformation according to the link cost corresponding to the standard communication link to obtain the signal frequency domain feature map. Based on the link cost, the signal frequency domain feature map is subjected to image denoising and feature enhancement based on residual information to obtain an enhanced frequency domain feature map. The enhanced frequency domain feature map is then subjected to inverse transformation to obtain the enhanced received signal.

[0006] According to a preferred embodiment of the present invention, network status information of each communication device within an ad hoc network is obtained over multiple time periods to obtain a historical communication dataset sequence, including: In each past time period, the network layer state of the self-organizing network was extracted to obtain network structure data; The communication links in the self-organizing network are detected to obtain a set of communication links, and the link organization device group corresponding to the set of communication links is determined. The link status of each communication device in the link organization device group is collected to obtain a communication link status dataset. Device data is collected from each communication device within the self-organizing network to obtain a communication device dataset. Historical communication datasets are generated based on the network structure data, communication link status datasets, and communication device datasets, and historical communication datasets from multiple time periods are aggregated into a historical communication dataset sequence in chronological order.

[0007] According to another preferred embodiment of the present invention, a communication topology graph is constructed and temporal feature modeling is performed based on the historical communication dataset sequence to obtain a dynamic communication spatiotemporal graph sequence, including: The historical communication dataset sequence is unified in terms of fields and aligned with timestamps to obtain an aligned communication dataset sequence. The aligned communication dataset sequence is then cleaned to obtain a standard communication dataset sequence. The standard communication datasets in the sequence of standard communication datasets are selected one by one, and node identifier alignment and node feature mapping are performed to obtain the communication node set and the corresponding node edge set; The communication state feature mapping is performed on the node edge set to obtain the edge feature set corresponding to the node edge set; A communication network topology graph is constructed based on the communication node set, node edge set, and edge feature set, and the communication network topology graphs corresponding to all standard communication datasets are aggregated into a communication network topology graph sequence. The communication network topology graph sequence is dynamically embedded and aligned, and its spatiotemporal features are encoded to obtain a dynamic communication spatiotemporal graph sequence.

[0008] According to another preferred embodiment of the present invention, dynamic embedding alignment and spatiotemporal feature encoding are performed on the communication network topology graph sequence to obtain a dynamic communication spatiotemporal graph sequence, including: One by one, the communication network topology graphs in the communication network topology graph sequence are selected, dynamic node masks are initialized for the communication node set in the communication network topology graph, and low-dimensional feature mapping is performed on the communication node set in the communication network topology graph to obtain primary node embedding features. Based on the graph structure and edge feature set of the communication network topology, graph convolution and neighborhood aggregation are performed on the primary node embedding features to obtain node spatial embedding features, which are then aggregated into a node spatial embedding feature sequence according to time sequence, thereby obtaining the corresponding node spatiotemporal feature sequence. Based on the dynamic node mask, the spatiotemporal feature sequence of the node is aligned by cross-time step node embedding to obtain the aligned spatiotemporal feature sequence. The aligned spatiotemporal feature sequence is then used to embed features into the communication network topology graph sequence to obtain a dynamic communication spatiotemporal graph sequence.

[0009] According to another preferred embodiment of the present invention, cross-timestep node embedding alignment is performed on the spatiotemporal feature sequence of the nodes based on each dynamic node mask to obtain an aligned spatiotemporal feature sequence, including: The sampling anchor set is selected from each dynamic node mask, and positive sample sampling and negative sample sampling are performed on each dynamic node mask based on the sampling anchor set to obtain the positive sample node set and the negative sample node set. Construct a triplet set based on the sampling anchor set, the positive sample set, and the negative sample set; A dynamic temperature coefficient set is set for the sampling anchor set based on the triplet set, and a dynamic contrast loss value is calculated based on the dynamic temperature coefficient set and the triplet set. Based on the mask of each dynamic node, cross-time step smoothing constraints and temporal drift corrections are applied to the spatiotemporal feature sequence of the node to obtain smoothing loss values ​​and drift correction loss values. The node spatiotemporal feature sequence is aligned and optimized using a meta-learning algorithm based on the dynamic contrast loss value, smoothing loss value, and drift correction loss value to obtain an optimized spatiotemporal feature embedding sequence. The optimized spatiotemporal feature embedding sequence is subjected to alignment quality evaluation and adaptive realignment to obtain an aligned spatiotemporal feature sequence.

[0010] According to another preferred embodiment of the present invention, spatiotemporal correlation feature analysis and time-series communication state evolution are performed on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period, including: Based on a graph attention network, the dynamic communication spatiotemporal graph corresponding to each time step in the dynamic communication spatiotemporal graph sequence is spatially encoded to obtain an encoded communication graph feature sequence. The encoded communication graph feature sequence is weighted based on a multi-head self-attention mechanism and a temporal joint attention mechanism to obtain the corresponding enhanced communication graph feature sequence. The enhanced communication graph feature sequence is modeled temporally using a gated recurrent unit to obtain a communication graph temporal feature sequence, and then causal convolution is performed on the communication graph temporal feature sequence to obtain a communication graph evolution feature sequence. The communication graph evolution feature sequence is subjected to hierarchical pooling and multi-dimensional fusion to obtain a fused graph evolution feature sequence; The expected communication state dataset is obtained by performing context mapping encoding and multi-head prediction decoding on the fused graph evolution feature sequence.

[0011] According to another preferred embodiment of the present invention, a standard communication link is obtained by performing multi-constraint adaptive path decision-making on the local topology links of the communication device group based on the expected communication data, including: In the self-organizing network, multi-hop neighbor device extraction is performed on the communication device group to obtain a real-time neighbor device group, and the real-time neighbor device group and the communication device group are aggregated into a target communication device group; Based on the target communication device group, the dynamic communication spatiotemporal graph in the dynamic communication spatiotemporal graph sequence is pruned into a subgraph to obtain a local communication topology subgraph, and the reachable communication links of the local communication topology subgraph are extracted to obtain a local topology link set; Extract the link communication dataset corresponding to the local topology link set from the expected communication data, and perform basic weight allocation on the link communication dataset to obtain the link basic weight set; Obtain the target communication device data of the target communication device group, and extract the link resource constraints corresponding to the target communication device data; A multi-constraint link cost matrix is ​​generated based on the link basic weight set and the link resource constraints. A network dynamics score is extracted from the link communication dataset, a computing resource score is extracted from the target communication device data, and a path planning algorithm is matched based on the network dynamics score and the computing resource score. Based on the path planning algorithm and the multi-constraint link cost matrix, the link cost set corresponding to the local topology link set is calculated, and standard communication links are selected from the local topology link set according to the link cost set.

[0012] According to another preferred embodiment of the present invention, the received signal is subjected to time-frequency transformation based on the link cost corresponding to the standard communication link to obtain a signal frequency domain feature map, including: Determine whether the link cost corresponding to the standard communication link is greater than a preset cost threshold; If not, the received signal is subjected to basic filtering and enhancement to obtain an enhanced received signal; If so, then the transformation parameters are configured based on the link cost, and wavelet transform is performed on the received signal based on the transformation parameters to obtain a signal frequency domain feature map, wherein the transformation parameters include window length, window overlap ratio, and spectral resolution.

[0013] According to another preferred embodiment of the present invention, image denoising and feature enhancement based on residual information are performed on the signal frequency domain feature map according to the link cost to obtain an enhanced frequency domain feature map, including: Based on the link cost, the signal frequency domain feature map is normalized in amplitude and scale aligned to obtain an aligned frequency domain feature map. The aligned frequency domain feature map is then divided according to the time-frequency distribution characteristics to obtain a frequency domain feature sub-map set. The frequency domain feature sub-map set is low-pass filtered to obtain a reference feature sub-map set, and the primary residual frequency domain feature map is calculated based on the frequency domain feature sub-map set and the reference feature sub-map set. Based on the primary residual frequency domain feature map and the link cost, noise reduction parameters are matched to obtain the link noise reduction parameters; Based on the link denoising parameters, multi-band wavelet denoising is performed on the aligned frequency domain feature map to obtain a denoised frequency domain feature map, and the enhanced residual frequency domain feature map between the denoised frequency domain feature map and the aligned frequency domain feature map is calculated. The enhanced residual frequency domain feature map and the aligned frequency domain feature map are weighted and fused to obtain a fused frequency domain feature map. The fused frequency domain feature map is subjected to multi-scale convolution and global feature enhancement to obtain an enhanced frequency domain feature map.

[0014] To achieve at least one of the above-mentioned objectives, the present invention further provides a point-to-point communication system based on self-organizing direct connections, the system comprising a spatiotemporal modeling module, a communication evolution module, a path decision module, a time-frequency transformation module, and a signal enhancement module, wherein: The spatiotemporal modeling module is used to acquire network status information of each communication device in the self-organizing network over multiple time periods, obtain a historical communication dataset sequence, and construct a communication topology graph and perform temporal feature modeling based on the historical communication dataset sequence to obtain a dynamic communication spatiotemporal graph sequence. The communication evolution module is used to perform spatiotemporal correlation feature analysis and time-series communication state evolution on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period. The path decision module is used to determine the communication device group that needs point-to-point communication based on real-time communication requests, and to perform multi-constraint adaptive path decision on the local topology links of the communication device group according to the expected communication data to obtain a standard communication link. The time-frequency conversion module is used to perform point-to-point signal forwarding between each communication device in the communication equipment group according to the standard communication link, receive the signal using the destination device of the communication equipment group, obtain the received signal, and perform time-frequency conversion on the received signal according to the link cost corresponding to the standard communication link to obtain a signal frequency domain feature map. The signal enhancement module is used to perform image denoising and feature enhancement based on residual information on the signal frequency domain feature map according to the link cost, to obtain an enhanced frequency domain feature map, and to perform an inverse transformation on the enhanced frequency domain feature map to obtain an enhanced received signal.

[0015] (III) Beneficial Effects Compared with the prior art, the present invention provides a point-to-point communication method and system based on self-organizing direct connection, which has the following beneficial effects: This self-organizing direct-connection-based point-to-point communication method effectively models problems such as frequent node joining and leaving and drastic link quality fluctuations in mine self-organizing networks by constructing a dynamic communication spatiotemporal graph sequence under complex electromagnetic environments. By introducing dynamic node masks and cross-temporal embedding alignment mechanisms, the method maintains the consistency and comparability of communication characteristics in the time dimension even when nodes are discontinuous. By combining contrastive learning, cross-temporal smoothing constraints, and drift correction, the method effectively suppresses short-term noise fluctuations and long-term cumulative drift of communication quality characteristics, improves the stability and robustness of link state modeling, and provides highly consistent and reliable spatiotemporal characteristics for subsequent communication quality prediction and path decision-making.

[0016] This self-organizing direct-connection-based point-to-point communication method uses historical modeling and spatiotemporal prediction results to predict the quality of future communication links. It also constructs a multi-constraint link cost matrix by combining resource constraints such as node remaining power, communication load, and mobility status. This ensures that path selection not only focuses on current communication quality but also considers link stability and network sustainability. By matching path planning algorithms of different complexities with network dynamics scores and computational resource scores, it achieves rapid response in highly dynamic scenarios and fine optimization in stable scenarios, significantly improving the reliability, robustness, and overall communication efficiency of self-organizing networks in complex environments.

[0017] This self-organizing direct-connection-based point-to-point communication method enhances the noise resistance and stability of the received signal in complex, self-organizing network environments through link cost-driven residual denoising and deep feature enhancement. Simultaneously, it improves signal quality without significantly sacrificing detail by weighted fusion of the enhanced residual and the original image, forming a robust residual learning framework. This ensures effective recovery or enhancement of useful signal components after denoising. Multi-scale convolution and attention mechanisms are then used to further extract and calibrate deep spatiotemporal features from the fused image, suppressing noise and improving the signal-to-noise ratio. The final output is a highly clean and information-rich enhanced frequency domain feature map, improving signal transmission quality and thus increasing the efficiency of point-to-point communication. Attached Figure Description

[0018] Figure 1 The diagram shown is a flowchart of a point-to-point communication method based on self-organizing direct connection according to the present invention. Detailed Implementation

[0019] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0020] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0021] Example 1: Please combine Figure 1 This invention discloses a point-to-point communication method based on self-organizing direct connections, the method comprising the following steps: The network status information of each communication device in the self-organizing network is obtained in multiple time periods to obtain a historical communication dataset sequence. Based on the historical communication dataset sequence, a communication topology graph is constructed and a temporal feature model is performed to obtain a dynamic communication spatiotemporal graph sequence.

[0022] The self-organizing network is a network formed by multiple communication devices through self-organized connections in complex electromagnetic environments such as mines. Due to the complex communication environment, the self-organizing network exhibits characteristics such as frequent joining and leaving of network nodes, frequent changes in communication link quality, dust, metal obstruction, and severe multipath fading. This self-organizing network is formed based on LoRa communication mesh networking technology, combining the advantages of LoRa's low power consumption and long-distance wireless communication with the mesh topology, multi-hop relay, self-organization, and self-healing characteristics of mesh networks. It supports point-to-point multi-hop communication between devices without a central gateway, making it suitable for mining scenarios with large coverage dead zones and requiring decentralization. The communication devices are the communication nodes of the self-organizing network and can be intelligent terminals supporting LoRa or relay communication capabilities.

[0023] Specifically, network status information of each communication device within the ad hoc network is obtained over multiple time periods to generate a historical communication dataset sequence, including: In each past time period, the network layer state of the self-organizing network was extracted to obtain network structure data; The communication links in the self-organizing network are detected to obtain a set of communication links, and the link organization device group corresponding to the set of communication links is determined. The link status of each communication device in the link organization device group is collected to obtain a communication link status dataset. Device data is collected from each communication device within the self-organizing network to obtain a communication device dataset. Historical communication datasets are generated based on the network structure data, communication link status datasets, and communication device datasets, and historical communication datasets from multiple time periods are aggregated into a historical communication dataset sequence in chronological order.

[0024] The historical communication dataset sequence comprises network status information of all communication devices within an ad hoc network acquired within a specific time period. Each historical communication data point refers to the communication data of all detectable communication links within the ad hoc network during that time period, including communication link status data, network structure data, and communication device data. The communication link status data includes the Received Signal Strength Indicator (RSSI) and signal-to-noise ratio (SNR) of the received signals in the communication link. The network structure data includes: network ratio (SNR), number of successful / failed data packets, packet loss rate, communication latency, number of communication interruptions, and communication duration; the network structure data includes the neighbor node list of each communication device, link establishment and disconnection records, and routing table information; the communication device data includes the remaining battery power, communication load, and movement status of each communication device during communication; the signal strength and signal-to-noise ratio can be measured when receiving data packets; the number of successful / failed data packets, packet loss rate, communication latency, number of communication interruptions, and communication duration can be obtained by statistically analyzing the communication process; the neighbor node list can be obtained through broadcast detection in ad hoc networks; the link establishment and disconnection records and routing table information can be obtained through the maintenance process of routing protocols, reflecting the reachability of multi-hop paths and changes in path quality; the remaining battery power, communication load, and displacement during communication can be obtained by each communication device autonomously uploading data.

[0025] In detail, based on the historical communication dataset sequence, a communication topology graph is constructed and temporal feature modeling is performed to obtain a dynamic communication spatiotemporal graph sequence, including: The historical communication dataset sequence is unified in terms of fields and aligned with timestamps to obtain an aligned communication dataset sequence. The aligned communication dataset sequence is then cleaned to obtain a standard communication dataset sequence. The standard communication datasets in the sequence of standard communication datasets are selected one by one, and node identifier alignment and node feature mapping are performed to obtain the communication node set and the corresponding node edge set; The communication state feature mapping is performed on the node edge set to obtain the edge feature set corresponding to the node edge set; A communication network topology graph is constructed based on the communication node set, node edge set, and edge feature set, and the communication network topology graphs corresponding to all standard communication datasets are aggregated into a communication network topology graph sequence. The communication network topology graph sequence is dynamically embedded and aligned, and its spatiotemporal features are encoded to obtain a dynamic communication spatiotemporal graph sequence.

[0026] The term "field unification" refers to standardizing and unifying similar parameters from different formats and sources in various historical communication data into standardized and consistent field names, data types, and measurement units. This can be achieved through a data dictionary or pattern mapping table. "Timestamp alignment" refers to synchronizing data collected by different devices to the same timeline. This can be achieved through a network time synchronization protocol. "Data cleaning" includes missing value handling, outlier replacement, data deduplication, and noise smoothing. "Node identifier alignment" refers to mapping each communication device to a unique logical node ID. "Node feature mapping" refers to vectorizing the data of each communication device in the communication device dataset and mapping it to the corresponding logical node to obtain the communication node. "Generating node edges" refers to extracting communication node pairs corresponding to communication links from the network structure data and constructing corresponding node edges between these pairs. "Communication state feature mapping" refers to vectorizing the state data of each communication link to obtain edge features.

[0027] Specifically, the communication network topology graph sequence is dynamically embedded and aligned, and spatiotemporal feature encoded to obtain a dynamic communication spatiotemporal graph sequence, including: One by one, the communication network topology graphs in the communication network topology graph sequence are selected, dynamic node masks are initialized for the communication node set in the communication network topology graph, and low-dimensional feature mapping is performed on the communication node set in the communication network topology graph to obtain primary node embedding features. Based on the graph structure and edge feature set of the communication network topology, graph convolution and neighborhood aggregation are performed on the primary node embedding features to obtain node spatial embedding features, which are then aggregated into a node spatial embedding feature sequence according to time sequence, thereby obtaining the corresponding node spatiotemporal feature sequence. Based on the dynamic node mask, the spatiotemporal feature sequence of the node is aligned by cross-time step node embedding to obtain the aligned spatiotemporal feature sequence. The aligned spatiotemporal feature sequence is then used to embed features into the communication network topology graph sequence to obtain a dynamic communication spatiotemporal graph sequence.

[0028] The initialization of the dynamic node mask refers to setting a binary mask vector for each communication node in the communication node set. The mask vector represents the survival status of the communication node within a corresponding time period; for example, 0 represents inactivity and 1 represents activity. The low-dimensional feature mapping refers to projecting the original high-dimensional features of the nodes into a low-dimensional vector space through a shallow neural network, thereby removing feature redundancy and reducing noise. For example, low-dimensional feature mapping can be performed through a fully connected layer. The graph convolution refers to assigning different weights to different neighboring nodes of the communication network topology graph using a graph convolutional layer. The neighborhood aggregation refers to weighted summation of the features of the neighboring nodes of each communication node according to their weights, thereby obtaining the node spatial embedding features. The corresponding spatiotemporal feature sequence of the nodes is obtained. This refers to sequentially performing multi-scale temporal convolution, hierarchical graph pooling, and residual connections. The multi-scale temporal convolution refers to using multiple convolutional kernels with different time scales to perform temporal convolution on the node space embedded feature sequence, thereby capturing long-term, medium-term, and short-term temporal features. The acquired long-term, short-term, and medium-term temporal feature sequences are weighted and fused, and a gating mechanism is used to control the fused features. The hierarchical graph pooling refers to using pooling layers to gradually reduce the fused temporal feature sequence. The residual connection refers to upsampling the temporal feature sequence after hierarchical graph pooling to restore it to the same feature dimension as the node space embedded feature sequence, and then performing a residual connection between the upsampled temporal feature sequence and the node space embedded feature sequence.

[0029] In detail, based on the masks of each dynamic node, cross-timestep node embedding alignment is performed on the spatiotemporal feature sequence of the nodes to obtain an aligned spatiotemporal feature sequence, including: The sampling anchor set is selected from each dynamic node mask, and positive sample sampling and negative sample sampling are performed on each dynamic node mask based on the sampling anchor set to obtain the positive sample node set and the negative sample node set. Construct a triplet set based on the sampling anchor set, the positive sample set, and the negative sample set; A dynamic temperature coefficient set is set for the sampling anchor set based on the triplet set, and a dynamic contrast loss value is calculated based on the dynamic temperature coefficient set and the triplet set. Based on the mask of each dynamic node, cross-time step smoothing constraints and temporal drift corrections are applied to the spatiotemporal feature sequence of the node to obtain smoothing loss values ​​and drift correction loss values. The node spatiotemporal feature sequence is aligned and optimized using a meta-learning algorithm based on the dynamic contrast loss value, smoothing loss value, and drift correction loss value to obtain an optimized spatiotemporal feature embedding sequence. The optimized spatiotemporal feature embedding sequence is subjected to alignment quality evaluation and adaptive realignment to obtain an aligned spatiotemporal feature sequence.

[0030] The sampling anchor set refers to a subset of nodes selected from all active nodes at each time step as sampling anchors. Positive sample sampling refers to selecting the corresponding node from each sampling anchor and corresponding nodes from other time steps as positive samples. Negative sample sampling refers to selecting different nodes from other time steps as negative samples. Each triplet in the triplet set consists of the corresponding sampling anchor, positive samples, and negative samples. Setting a dynamic temperature coefficient set refers to setting a smaller temperature coefficient for highly similar sample pairs (positive and negative samples) to amplify the gradient, and a larger temperature coefficient for low-similarity sample pairs to reduce the gradient. For example, the formula for calculating the dynamic contrast loss value is as follows: in, This is the dynamic contrast loss value. It is the set of triples, It's a triplet. As the sampling anchor point, As a positive sample, For negative samples, For similarity function, This refers to the sampling anchor point. The corresponding dynamic temperature coefficient; The cross-time-step smoothing constraint refers to setting a loss value based on the changes of the same node in adjacent time steps. The temporal drift correction refers to detecting the loss value corresponding to the long-term feature centroid drift of a node, and setting a corresponding soft constraint when the drift exceeds the drift threshold to pull the feature back. For example, the formula for the cross-time-step smoothing constraint is as follows: in, It is the smoothing loss value. It is a time step. It is the total time step length. yes Time step downsampling anchor point Dynamic node mask, It is the square of the L2 norm. yes Time step downsampling anchor point The spatiotemporal characteristics of nodes; The formula for timing drift correction is as follows: in, It is the drift correction loss value. It is a sampling anchor point The long-term feature centroid of the spatiotemporal features of the nodes.

[0031] Specifically, the alignment optimization of the spatiotemporal feature sequence of the nodes using the meta-learning algorithm based on the dynamic contrast loss value, the smoothing loss value, and the drift correction loss value refers to using the weighted sum of the dynamic contrast loss value, the smoothing loss value, and the drift correction loss value as the inner loop loss of the meta-learning, updating the initial parameters of the meta-learning model by calculating the second derivative, and using the meta-learning model for alignment optimization. Model-Agnostic Meta-Learning (MAML) can be used for alignment optimization. The alignment quality evaluation refers to calculating an alignment score for each node. For example, the inverse of the variance corresponding to the node embedding is calculated as its consistency on the time axis. The higher the alignment score, the more consistent the representation of the node is at different time steps. Adaptive realignment refers to sorting according to the alignment score, setting an alignment threshold, re-executing the aforementioned alignment optimization steps for nodes with alignment scores lower than the alignment threshold and related affected nodes, while retaining nodes with alignment scores higher than the alignment threshold.

[0032] By constructing a dynamic communication spatiotemporal graph sequence under complex electromagnetic environments, we have achieved effective modeling of problems such as frequent joining and leaving of communication nodes and drastic fluctuations in link quality in mine self-organizing networks. By introducing dynamic node masks and cross-temporal embedding alignment mechanisms, we can maintain the consistency and comparability of communication characteristics in the time dimension even when nodes are discontinuous. By combining contrastive learning, cross-temporal smoothing constraints, and drift correction, we have effectively suppressed short-term noise fluctuations and long-term cumulative drift of communication quality characteristics, improved the stability and robustness of link state modeling, and provided highly consistent and reliable spatiotemporal characteristics for subsequent communication quality prediction and path decision-making.

[0033] Spatiotemporal correlation feature analysis and time-series communication state evolution are performed on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period.

[0034] The expected communication state dataset contains expected communication state data corresponding to each communication link of each self-organizing network in the future period. The expected communication state data includes the signal strength, signal-to-noise ratio, number of successful / failed data packets, packet loss rate, communication delay, number of communication interruptions, and communication duration of the received signals in the communication links in the future period.

[0035] In detail, spatiotemporal correlation feature analysis and time-series communication state evolution are performed on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period, including: Based on a graph attention network, the dynamic communication spatiotemporal graph corresponding to each time step in the dynamic communication spatiotemporal graph sequence is spatially encoded to obtain an encoded communication graph feature sequence. The coded communication graph feature sequence is spatially correlated and weighted based on a multi-head self-attention mechanism, and then dynamically weighted over time steps based on a temporal joint attention mechanism to obtain an enhanced communication graph feature sequence. The enhanced communication graph feature sequence is modeled temporally using a gated recurrent unit to obtain a communication graph temporal feature sequence, and then causal convolution is performed on the communication graph temporal feature sequence to obtain a communication graph evolution feature sequence. The communication graph evolution feature sequence is subjected to hierarchical pooling and multi-dimensional fusion to obtain a fused graph evolution feature sequence; The expected communication state dataset is obtained by performing context mapping encoding and multi-head prediction decoding on the fused graph evolution feature sequence.

[0036] The spatial encoding refers to calculating the similarity score between each node and all its neighboring nodes using a graph attention network in the dynamic communication spatiotemporal graph, and then normalizing it to attention weights using a softmax function. The updated feature of a node is the weighted sum of the features of all its neighboring nodes, and the weights are the learned attention values. The spatial association weighting refers to calculating the spatial association weights of nodes and edges in each encoded communication graph feature using a multi-head self-attention mechanism and then weighting them. That is, mapping each node and edge of the encoded communication graph feature to the query vector space, key vector space, and value vector space in the spatial dimension, and calculating the result after attention concatenation in parallel. The time-step dynamic weighting refers to assigning dynamic weights to the encoded communication graph features at each time step along the time dimension using Transformer-type attention, thereby achieving feature weighting enhancement. The temporal construction... The model refers to capturing the long-term temporal dependencies of the enhanced communication graph feature sequence using gated recurrent units. The causal convolution uses causal convolution to capture the local temporal features of the communication graph's temporal sequence, thereby achieving global and local temporal feature extraction. The hierarchical pooling refers to progressively compressing and pooling the communication graph evolution feature sequence from the node level, subgraph level, and global level, thereby reducing feature dimensionality and suppressing noisy nodes. This can be implemented using DiffPool or Top-KPooling. The context mapping encoding refers to using a Seq2Seq architecture encoder to map the fused graph evolution feature sequence into context features, and using a Transformer architecture multi-head decoder for decoding. In actual use, the above models and architectures are lightweight models obtained after pruning, model quantization, or knowledge distillation to meet the computing power requirements of communication equipment in mining environments.

[0037] By performing multi-level spatiotemporal correlation feature analysis and time-series communication state evolution modeling on dynamic communication spatiotemporal graph sequences, we can comprehensively characterize the evolutionary laws of communication links in self-organizing networks over time. Through the synergistic effect of graph attention networks and multi-head self-attention mechanisms, we can effectively mine the spatial correlation relationships between communication nodes and links. By combining temporal joint attention, gated recurrent units, and causal convolution, we can jointly model the long-term trends and short-term emergencies of communication states. Through hierarchical pooling and multi-dimensional fusion, we can suppress noise and enhance robustness. Finally, by utilizing context mapping coding and multi-head prediction decoding, we can accurately predict various communication performance indicators, thereby providing reliable data support for network scheduling, link optimization, and anomaly early warning, and improving the accuracy of communication link decisions and the quality of links.

[0038] Based on real-time communication requests, the communication device group that requires point-to-point communication is determined, and multi-constraint adaptive path decision is performed on the local topology links of the communication device group according to the expected communication data to obtain the standard communication link.

[0039] The real-time communication request refers to a communication request issued by a communication device in an ad hoc network that needs to perform point-to-point communication. The communication device group includes a source device and a destination device. The source device sends a signal to the destination device to complete the real-time communication.

[0040] Specifically, based on the expected communication data, multi-constraint adaptive path decision-making is performed on the local topology links of the communication device group to obtain a standard communication link, including: In the self-organizing network, multi-hop neighbor device extraction is performed on the communication device group to obtain a real-time neighbor device group, and the real-time neighbor device group and the communication device group are aggregated into a target communication device group; Based on the target communication device group, the dynamic communication spatiotemporal graph in the dynamic communication spatiotemporal graph sequence is pruned into a subgraph to obtain a local communication topology subgraph, and the reachable communication links of the local communication topology subgraph are extracted to obtain a local topology link set; Extract the link communication dataset corresponding to the local topology link set from the expected communication data, and perform basic weight allocation on the link communication dataset to obtain the link basic weight set; Obtain the target communication device data of the target communication device group, and extract the link resource constraints corresponding to the target communication device data; A multi-constraint link cost matrix is ​​generated based on the link basic weight set and the link resource constraints. A network dynamics score is extracted from the link communication dataset, a computing resource score is extracted from the target communication device data, and a path planning algorithm is matched based on the network dynamics score and the computing resource score. Based on the path planning algorithm and the multi-constraint link cost matrix, the link cost set corresponding to the local topology link set is calculated, and standard communication links are selected from the local topology link set according to the link cost set.

[0041] The multi-hop neighbor device extraction refers to extracting devices starting from the source device, exploring all direct neighbors of the source device with a 1-hop count, then exploring the neighbors' neighbors with a 2-hop count, and so on, until a preset hop count threshold is reached. All neighbor devices are then grouped into a neighbor device group. This extraction can be performed using a breadth-first search algorithm. The hop count threshold can be dynamically adjusted based on the remaining battery power of the target communication device, the communication load, and the device's movement status during communication. The subgraph pruning involves selecting the most recent dynamic communication spatiotemporal graph in the sequence and extracting only the segments containing the target communication device group using a graph analysis library or the subgraph function. The subgraph structure of points and their interconnections serves as a local communication topology subgraph. Reachable communication links can be extracted using a depth-first search algorithm, which enumerates all potential communication links from the source device to the destination device. The link communication data in the link communication dataset corresponds to communication parameters such as signal strength, signal-to-noise ratio, number of successful / failed data packets, packet loss rate, communication latency, number of communication interruptions, and communication duration for a local topology link. The basic weight allocation refers to the weighted summation of each communication parameter in the link communication data according to a preset weight function to obtain the corresponding basic link weight. The target communication device data is the real-time communication device data corresponding to the target communication device group.

[0042] In detail, the link resource constraint extraction refers to extracting resource data such as the remaining power, communication load, and movement status of each communication device in the target communication device data, and converting them into corresponding resource constraints. For example, the remaining power constraint can be defined as the minimum remaining power of all nodes on the path must be higher than a certain power threshold. Alternatively, the communication load can be converted into a penalty term; the higher the load, the greater the additional cost of the local topology link where the communication device node is located. The multi-constraint link cost matrix generation refers to fusing the link base weights and link resource constraints to generate a comprehensive, multi-constraint link cost matrix. Each element in the link cost matrix represents the final cost of connecting two nodes and is the direct basis for the path planning algorithm's decision-making. It can be set using a weighted linear combination method. For example, assuming the link base weight of a local topology link is... The corresponding link resource constraints are Then the final cost of this local topology link. for ,in To constrain the weighting coefficients, the path planning matching can obtain a network dynamics score that measures the degree of link fluctuation by weighting the communication parameters in the link communication dataset of each local topology link. It can also obtain a computing resource score that measures the available computing power of the current device or network by calculating the average of the weighted values ​​of the remaining power and communication load in the data of each target communication device. Matching a path planning algorithm refers to using a decision table to match the corresponding path planning algorithm based on various network dynamics scores and the computing resource scores. For example, when both the network dynamics score and the computing resource score are high, an algorithm with high computational cost but rapid adaptability to changes, such as the A* algorithm or a reinforcement learning-based algorithm, is matched. When both the computing resource score and the network dynamics score are low, a lightweight Dijkstra algorithm is used. Selecting a standard communication link from the local topology link set based on the link cost set means selecting the local topology link with the lowest link cost in the link cost set as the standard communication link.

[0043] By utilizing historical modeling and spatiotemporal prediction results, the quality of future communication links is predicted. Combined with resource constraints such as node remaining power, communication load, and mobility status, a multi-constraint link cost matrix is ​​constructed, enabling path selection to not only focus on current communication quality but also consider link stability and network sustainability. By matching path planning algorithms of different complexities with network dynamics scoring and computational resource scoring, rapid response in highly dynamic scenarios and fine optimization in stable scenarios are achieved, significantly improving the reliability, robustness, and overall communication efficiency of ad hoc networks in complex environments.

[0044] Point-to-point signal forwarding is performed between each communication device in the communication equipment group according to the standard communication link. The target device in the communication equipment group receives the signal to obtain the received signal. The received signal is then subjected to time-frequency transformation according to the link cost corresponding to the standard communication link to obtain the signal frequency domain feature map.

[0045] The signal transmission refers to the use of the source device in the communication device group to send a signal, which is transmitted along each relay node of the standard communication link and finally sent to the destination device. The source node sequentially extracts the next-hop communication device from the standard communication link, encapsulates the data to be sent into a signal with a multi-hop routing header, and the path information of the standard communication link is synchronized to each relay node along the way. Each relay node receives the signal and then sends it to the next-hop communication device in the standard communication link until it is sent to the destination device.

[0046] Specifically, the received signal is subjected to time-frequency transformation based on the link cost corresponding to the standard communication link to obtain a signal frequency domain feature map, including: Determine whether the link cost corresponding to the standard communication link is greater than a preset cost threshold; If not, the received signal is subjected to basic filtering and enhancement to obtain an enhanced received signal; If so, then the transformation parameters are configured based on the link cost, and wavelet transform is performed on the received signal based on the transformation parameters to obtain a signal frequency domain feature map, wherein the transformation parameters include window length, window overlap ratio, and spectral resolution.

[0047] The basic filter enhancement refers to enhancement based on a Finite Impulse Response (FIR) filter or a linear filter. The configuration transformation parameters refer to increasing the window length and increasing the window overlap ratio according to different amplitudes based on the ratio of link cost to cost threshold when the link cost is high, thereby improving the spectral resolution.

[0048] By comparing link cost with cost threshold, precise allocation of computing resources is achieved. When the link quality is good, basic filtering enhancement with low computational load is adopted to quickly output usable signals, which significantly reduces the computational burden and energy consumption of IoT devices. This is crucial for mining sensor devices with limited power, helps extend the overall network lifespan, improves robustness, real-time performance and resource utilization efficiency in complex network environments, and also improves communication efficiency.

[0049] Based on the link cost, the signal frequency domain feature map is subjected to image denoising and feature enhancement based on residual information to obtain an enhanced frequency domain feature map. The enhanced frequency domain feature map is then subjected to inverse transformation to obtain the enhanced received signal.

[0050] In detail, based on the link cost, the signal frequency domain feature map is subjected to image denoising and feature enhancement based on residual information to obtain an enhanced frequency domain feature map, including: Based on the link cost, the signal frequency domain feature map is normalized in amplitude and scale aligned to obtain an aligned frequency domain feature map. The aligned frequency domain feature map is then divided according to the time-frequency distribution characteristics to obtain a frequency domain feature sub-map set. The frequency domain feature sub-map set is low-pass filtered to obtain a reference feature sub-map set, and the primary residual frequency domain feature map is calculated based on the frequency domain feature sub-map set and the reference feature sub-map set. Based on the primary residual frequency domain feature map and the link cost, noise reduction parameters are matched to obtain the link noise reduction parameters; Based on the link denoising parameters, multi-band wavelet denoising is performed on the aligned frequency domain feature map to obtain a denoised frequency domain feature map, and the enhanced residual frequency domain feature map between the denoised frequency domain feature map and the aligned frequency domain feature map is calculated. The enhanced residual frequency domain feature map and the aligned frequency domain feature map are weighted and fused to obtain a fused frequency domain feature map. The fused frequency domain feature map is subjected to multi-scale convolution and global feature enhancement to obtain an enhanced frequency domain feature map.

[0051] Amplitude normalization can be achieved using a min-max normalization algorithm. This normalization scales the amplitude values ​​at each point in the frequency domain feature map to a uniform range, such as [0, 1], thereby eliminating deviations caused by differences in absolute signal strength. Scale alignment can be achieved through interpolation resampling, ensuring that frequency domain feature maps generated at different time steps or through different links have the same scale on the frequency axis. The time-frequency distribution characteristic division refers to dividing the entire aligned frequency domain feature map into multiple smaller frequency domain feature sub-maps based on the signal's distribution characteristics in the time and frequency dimensions, such as energy concentration areas in specific frequency bands. The diagram facilitates focusing on local features, enabling parallel processing and targeted analysis of information in specific frequency bands or time periods. The noise reduction parameter matching refers to matching the corresponding wavelet threshold size based on the value of the link cost. When the link cost is high, a more aggressive noise reduction parameter is matched, i.e., a lower wavelet threshold, to strongly suppress noise at the cost of sacrificing a small amount of detail. When the link cost is low, a more conservative noise reduction parameter is used to preserve signal details to the greatest extent. The multi-band wavelet noise reduction refers to using wavelet transform to decompose the signal into different scales and frequency bands, and different wavelet threshold strategies can be applied for noise reduction for different frequency bands.

[0052] Specifically, the fused frequency domain feature map is subjected to multi-scale convolution and global feature enhancement to obtain an enhanced frequency domain feature map, including: The fused frequency domain feature map is subjected to primary convolution to obtain a convolutional frequency domain feature map, and the convolutional frequency domain feature map is normalized to obtain a normalized frequency domain feature map. The normalized frequency domain feature map is activated to obtain an activated frequency domain feature map, and the activated frequency domain feature map is then subjected to adaptive convolution to obtain an adaptive frequency domain feature map. The adaptive frequency domain feature map is residually connected using the convolutional frequency domain feature map to obtain the convolutional residual frequency domain feature map; The convolutional residual frequency domain feature map is downsampled layer by layer at multiple scales to obtain a multi-scale frequency domain feature map set; The multi-scale frequency domain feature map set is sliced ​​and local attention encoded to obtain a local frequency domain feature map set; Global attention encoding is performed on the local frequency domain feature map set to obtain a global frequency domain feature map set; The global frequency domain feature map set is subjected to multi-level feedforward activation and residual connection to obtain an enhanced frequency domain feature map.

[0053] Specifically, feature activation can be performed using ReLU or sigmoid activation functions, and adaptive convolution can be performed using the Involution operator. The layer-by-layer multi-scale downsampling refers to reducing the scale of the frequency domain feature map of the convolution residual layer by layer, downsampling the frequency domain feature map of the convolution residual, and collecting the sampling results of each layer and the frequency domain feature map of the convolution residual into a multi-scale frequency domain feature map set. The feature slicing refers to dividing each multi-scale frequency domain feature map in the multi-scale frequency domain feature map set into frequency domain feature map groups, performing local attention encoding using a local attention mechanism, and performing global attention encoding using a global attention mechanism. The multi-level feedforward activation refers to activation using the structure of the Feed-Forward Network (FFN) and activation function.

[0054] By leveraging link cost-driven residual denoising and deep feature enhancement, the received signal exhibits stronger noise resistance and stability in complex, self-organizing network environments. Simultaneously, signal quality is improved without significantly sacrificing detail. The enhanced residual is weighted and fused with the original image, forming a powerful residual learning framework that ensures effective recovery or enhancement of useful signal components after denoising. Multi-scale convolution and attention mechanisms are then used to further extract and calibrate deep spatiotemporal features from the fused image, suppressing noise, improving the signal-to-noise ratio, and ultimately outputting a highly clean and information-rich enhanced frequency domain feature map. This improves signal transmission quality and, consequently, the efficiency of point-to-point communication.

[0055] Example 2: This invention discloses a point-to-point communication system based on self-organizing direct connections. The system includes a spatiotemporal modeling module, a communication evolution module, a path decision module, a time-frequency transformation module, and a signal enhancement module, wherein: The spatiotemporal modeling module is used to acquire network status information of each communication device in the self-organizing network over multiple time periods, obtain a historical communication dataset sequence, and construct a communication topology graph and perform temporal feature modeling based on the historical communication dataset sequence to obtain a dynamic communication spatiotemporal graph sequence. The communication evolution module is used to perform spatiotemporal correlation feature analysis and time-series communication state evolution on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period. The path decision module is used to determine the communication device group that needs point-to-point communication based on real-time communication requests, and to perform multi-constraint adaptive path decision on the local topology links of the communication device group according to the expected communication data to obtain a standard communication link. The time-frequency conversion module is used to perform point-to-point signal forwarding between each communication device in the communication equipment group according to the standard communication link, receive the signal using the destination device of the communication equipment group, obtain the received signal, and perform time-frequency conversion on the received signal according to the link cost corresponding to the standard communication link to obtain a signal frequency domain feature map. The signal enhancement module is used to perform image denoising and feature enhancement based on residual information on the signal frequency domain feature map according to the link cost, to obtain an enhanced frequency domain feature map, and to perform an inverse transformation on the enhanced frequency domain feature map to obtain an enhanced received signal.

[0056] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0057] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0058] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A point-to-point communication method based on self-organizing direct connections, characterized in that, The method includes: The network status information of each communication device in the self-organizing network is obtained in multiple time periods to obtain a historical communication dataset sequence. Based on the historical communication dataset sequence, a communication topology graph is constructed and a temporal feature model is performed to obtain a dynamic communication spatiotemporal graph sequence. Spatiotemporal correlation feature analysis and time-series communication state evolution are performed on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period. Based on real-time communication requests, the communication device group that needs point-to-point communication is determined, and multi-constraint adaptive path decision is performed on the local topology links of the communication device group according to the expected communication data to obtain the standard communication link. Point-to-point signal forwarding is performed between each communication device in the communication equipment group according to the standard communication link. The destination device in the communication equipment group receives the signal to obtain the received signal. The received signal is then subjected to time-frequency transformation according to the link cost corresponding to the standard communication link to obtain the signal frequency domain feature map. Based on the link cost, the signal frequency domain feature map is subjected to image denoising and feature enhancement based on residual information to obtain an enhanced frequency domain feature map. The enhanced frequency domain feature map is then subjected to inverse transformation to obtain the enhanced received signal.

2. The point-to-point communication method based on self-organizing direct connection according to claim 1, characterized in that, Obtain network status information of each communication device within the ad hoc network over multiple time periods to obtain a historical communication dataset sequence, including: In each past time period, the network layer state of the self-organizing network was extracted to obtain network structure data; The communication links in the self-organizing network are detected to obtain a set of communication links, and the link organization device group corresponding to the set of communication links is determined. The link status of each communication device in the link organization device group is collected to obtain a communication link status dataset. Device data is collected from each communication device within the self-organizing network to obtain a communication device dataset. Historical communication datasets are generated based on the network structure data, communication link status datasets, and communication device datasets, and historical communication datasets from multiple time periods are aggregated into a historical communication dataset sequence in chronological order.

3. The point-to-point communication method based on self-organizing direct connection according to claim 2, characterized in that, Based on the historical communication dataset sequence, a communication topology graph is constructed and temporal feature modeling is performed to obtain a dynamic communication spatiotemporal graph sequence, including: The historical communication dataset sequence is unified in terms of fields and aligned with timestamps to obtain an aligned communication dataset sequence. The aligned communication dataset sequence is then cleaned to obtain a standard communication dataset sequence. The standard communication datasets in the sequence of standard communication datasets are selected one by one, and node identifier alignment and node feature mapping are performed to obtain the communication node set and the corresponding node edge set; The communication state feature mapping is performed on the node edge set to obtain the edge feature set corresponding to the node edge set; A communication network topology graph is constructed based on the communication node set, node edge set, and edge feature set, and the communication network topology graphs corresponding to all standard communication datasets are aggregated into a communication network topology graph sequence. The communication network topology graph sequence is dynamically embedded and aligned, and its spatiotemporal features are encoded to obtain a dynamic communication spatiotemporal graph sequence.

4. A point-to-point communication method based on self-organizing direct connections according to claim 3, characterized in that, Dynamic embedding alignment and spatiotemporal feature encoding are performed on the communication network topology graph sequence to obtain a dynamic communication spatiotemporal graph sequence, including: One by one, the communication network topology graphs in the communication network topology graph sequence are selected, dynamic node masks are initialized for the communication node set in the communication network topology graph, and low-dimensional feature mapping is performed on the communication node set in the communication network topology graph to obtain primary node embedding features. Based on the graph structure and edge feature set of the communication network topology, graph convolution and neighborhood aggregation are performed on the primary node embedding features to obtain node spatial embedding features, which are then aggregated into a node spatial embedding feature sequence according to time sequence, thereby obtaining the corresponding node spatiotemporal feature sequence. Based on the dynamic node mask, the spatiotemporal feature sequence of the node is aligned by cross-time step node embedding to obtain the aligned spatiotemporal feature sequence. The aligned spatiotemporal feature sequence is then used to embed features into the communication network topology graph sequence to obtain a dynamic communication spatiotemporal graph sequence.

5. A point-to-point communication method based on self-organizing direct connections according to claim 4, characterized in that, Based on the dynamic node masks, the spatiotemporal feature sequences of the nodes are aligned across time steps to obtain aligned spatiotemporal feature sequences, including: The sampling anchor set is selected from each dynamic node mask, and positive sample sampling and negative sample sampling are performed on each dynamic node mask based on the sampling anchor set to obtain the positive sample node set and the negative sample node set. Construct a triplet set based on the sampling anchor set, the positive sample set, and the negative sample set; A dynamic temperature coefficient set is set for the sampling anchor set based on the triplet set, and a dynamic contrast loss value is calculated based on the dynamic temperature coefficient set and the triplet set. Based on the mask of each dynamic node, cross-time step smoothing constraints and temporal drift corrections are applied to the spatiotemporal feature sequence of the node to obtain smoothing loss values ​​and drift correction loss values. The node spatiotemporal feature sequence is aligned and optimized using a meta-learning algorithm based on the dynamic contrast loss value, smoothing loss value, and drift correction loss value to obtain an optimized spatiotemporal feature embedding sequence. The optimized spatiotemporal feature embedding sequence is subjected to alignment quality evaluation and adaptive realignment to obtain an aligned spatiotemporal feature sequence.

6. A point-to-point communication method based on self-organizing direct connections according to claim 1, characterized in that, Spatiotemporal correlation feature analysis and time-series communication state evolution are performed on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period, including: Based on a graph attention network, the dynamic communication spatiotemporal graph corresponding to each time step in the dynamic communication spatiotemporal graph sequence is spatially encoded to obtain an encoded communication graph feature sequence. The encoded communication graph feature sequence is weighted based on a multi-head self-attention mechanism and a temporal joint attention mechanism to obtain the corresponding enhanced communication graph feature sequence. The enhanced communication graph feature sequence is modeled temporally using a gated recurrent unit to obtain a communication graph temporal feature sequence, and then causal convolution is performed on the communication graph temporal feature sequence to obtain a communication graph evolution feature sequence. The communication graph evolution feature sequence is subjected to hierarchical pooling and multi-dimensional fusion to obtain a fused graph evolution feature sequence; The expected communication state dataset is obtained by performing context mapping encoding and multi-head prediction decoding on the fused graph evolution feature sequence.

7. A point-to-point communication method based on self-organizing direct connections according to claim 1, characterized in that, Based on the expected communication data, multi-constraint adaptive path decision-making is performed on the local topology links of the communication device group to obtain standard communication links, including: In the self-organizing network, multi-hop neighbor device extraction is performed on the communication device group to obtain a real-time neighbor device group, and the real-time neighbor device group and the communication device group are aggregated into a target communication device group; Based on the target communication device group, the dynamic communication spatiotemporal graph in the dynamic communication spatiotemporal graph sequence is pruned into a subgraph to obtain a local communication topology subgraph, and the reachable communication links of the local communication topology subgraph are extracted to obtain a local topology link set; Extract the link communication dataset corresponding to the local topology link set from the expected communication data, and perform basic weight allocation on the link communication dataset to obtain the link basic weight set; Obtain the target communication device data of the target communication device group, extract the link resource constraints corresponding to the target communication device data, and generate a multi-constraint link cost matrix based on the link basic weight set and the link resource constraints; A network dynamics score is extracted from the link communication dataset, a computing resource score is extracted from the target communication device data, and a path planning algorithm is matched based on the network dynamics score and the computing resource score. Based on the path planning algorithm and the multi-constraint link cost matrix, the link cost set corresponding to the local topology link set is calculated, and standard communication links are selected from the local topology link set according to the link cost set.

8. A point-to-point communication method based on self-organizing direct connections according to claim 1, characterized in that, The received signal is subjected to time-frequency transformation based on the link cost corresponding to the standard communication link to obtain a signal frequency domain feature map, including: Determine whether the link cost corresponding to the standard communication link is greater than a preset cost threshold; If not, the received signal is subjected to basic filtering and enhancement to obtain an enhanced received signal; If so, then the transformation parameters are configured based on the link cost, and wavelet transform is performed on the received signal based on the transformation parameters to obtain a signal frequency domain feature map, wherein the transformation parameters include window length, window overlap ratio, and spectral resolution.

9. A point-to-point communication method based on self-organizing direct connections according to claim 1, characterized in that, Based on the link cost, the signal frequency domain feature map is subjected to image denoising and feature enhancement based on residual information to obtain an enhanced frequency domain feature map, including: Based on the link cost, the signal frequency domain feature map is normalized in amplitude and aligned in scale to obtain an aligned frequency domain feature map. The aligned frequency domain feature map is then divided into time and frequency domain features to obtain a set of frequency domain feature maps. Low-pass filtering and residual calculation are performed on the frequency domain feature sub-map to obtain a primary residual frequency domain feature map. Based on the primary residual frequency domain feature map and the link cost, noise reduction parameters are matched to obtain the link noise reduction parameters. Based on the link denoising parameters, multi-band wavelet denoising is performed on the aligned frequency domain feature map to obtain a denoised frequency domain feature map, and the enhanced residual frequency domain feature map between the denoised frequency domain feature map and the aligned frequency domain feature map is calculated. The enhanced residual frequency domain feature map and the aligned frequency domain feature map are weighted and fused to obtain a fused frequency domain feature map. The fused frequency domain feature map is then subjected to multi-scale convolution and global feature enhancement to obtain an enhanced frequency domain feature map.

10. A point-to-point communication system based on self-organizing direct connections, characterized in that, The system includes a spatiotemporal modeling module, a communication evolution module, a path decision module, a time-frequency transformation module, and a signal enhancement module, wherein: The spatiotemporal modeling module is used to acquire network status information of each communication device in the self-organizing network over multiple time periods, obtain a historical communication dataset sequence, and construct a communication topology graph and perform temporal feature modeling based on the historical communication dataset sequence to obtain a dynamic communication spatiotemporal graph sequence. The communication evolution module is used to perform spatiotemporal correlation feature analysis and time-series communication state evolution on the dynamic communication spatiotemporal graph sequence to obtain the expected communication state dataset corresponding to each communication link in the future period. The path decision module is used to determine the communication device group that needs point-to-point communication based on real-time communication requests, and to perform multi-constraint adaptive path decision on the local topology links of the communication device group according to the expected communication data to obtain a standard communication link. The time-frequency conversion module is used to perform point-to-point signal forwarding between each communication device in the communication equipment group according to the standard communication link, receive the signal using the destination device of the communication equipment group, obtain the received signal, and perform time-frequency conversion on the received signal according to the link cost corresponding to the standard communication link to obtain a signal frequency domain feature map. The signal enhancement module is used to perform image denoising and feature enhancement based on residual information on the signal frequency domain feature map according to the link cost, to obtain an enhanced frequency domain feature map, and to perform an inverse transformation on the enhanced frequency domain feature map to obtain an enhanced received signal.

Images