High-dynamic ad hoc network routing method based on dynamic topology inference

Abstract

The invention discloses a high-dynamic ad hoc network routing method based on dynamic topology inference. The method comprises the following steps: step 1, providing traffic data of an unmanned aerial vehicle communication terminal for topology inference under the condition of low extra overhead by passively monitoring node service traffic characteristics of the unmanned aerial vehicle communication terminal; 2, implementing a connectivity analysis strategy and a connectivity judgment strategy, and judging the relevance between nodes of the communication terminal of the unmanned aerial vehicle; on the basis of the relevance of the flow data and according to deducing the physical topology of the whole network, topology information is obtained; step 3, based on the topological information, constructing a hybrid routing decision mechanism combining prior inference and on-demand discovery; on the premise of ensuring the route accessibility in a high-dynamic network environment, the control overhead of route discovery is reduced, channel congestion is relieved, and the network transmission efficiency under a limited bandwidth is improved. According to the method, the control packet overhead in a route discovery process is remarkably reduced while the route accessibility in a high dynamic environment is ensured, and the transmission efficiency of a network under limited bandwidth resources is improved.

Description

Technical Field

[0001] This invention belongs to the field of network communication technology, specifically relating to a highly dynamic ad hoc network routing method based on dynamic topology inference. Background Technology

[0002] In existing drone swarm ad hoc networks, the rapid movement of nodes leads to frequent changes in topology. Traditional routing protocols must rely on high-frequency periodic broadcasts or network-wide flooding of routing requests to maintain path validity. This blind probing mechanism has significant drawbacks: it generates a large number of redundant control messages for path discovery, causing routing control overhead to grow exponentially. This not only causes severe channel congestion but also directly squeezes out the limited bandwidth resources used for service data transmission, becoming a core bottleneck restricting the throughput of drone swarm networks.

[0003] Akash Kumar et al., in their paper "Queueing Network Topology Inference Using Passive and Active Measurements," proposed an innovative method for inferring the topology of queued networks. This method aims to infer a single-source routing tree through end-to-end measurements and addresses the issues of existing techniques relying on active probing and ignoring nodes with a degree of 2. However, this method assumes a static network state and may be limited in highly dynamic environments due to parameter estimation biases. Furthermore, its reliance on simplified queue models (such as M / M / 1) can lead to performance fluctuations in real-world packet-level simulations.

[0004] In their paper "Network Topology Inference With Partial Information," Brett Holbert et al. proposed iTop, a network topology inference method for non-cooperative routers, aiming to address the critical issue that the Traceroute tool can only obtain partial path information due to the presence of anonymous and congesting routers. However, its limitation lies in the high computational complexity of the merging phase, which may face scalability challenges in ultra-large-scale dynamic networks.

[0005] In their paper "Research on Topology Inference Method for Ad Hoc Network Simulation on Exata," Zhao Niu, Tao Ma, Xiao-feng Zhong, and others proposed a method for topology inference in Ad Hoc networks based on the AODV routing protocol, aiming to solve the problem of dynamic network topology inference and visualization. However, its limitation lies in its reliance on the integrity of RREP messages in the AODV protocol. In scenarios involving high-speed movement or frequent route updates, message loss may lead to incomplete topology reconstruction. Furthermore, this method only supports physical topology inference and does not address the real-time tracking of logical topology or dynamic topology changes. Summary of the Invention

[0006] To address the shortcomings of existing routing protocols for highly dynamic UAV swarm networks, which rely on periodic broadcasting or blind flooding for path discovery, resulting in high control packet overhead and low bandwidth utilization, this invention aims to provide a highly dynamic ad hoc network routing method based on dynamic topology inference. This method infers the entire network topology by passively sensing network traffic characteristics. While ensuring route reachability in highly dynamic environments, it significantly reduces control packet overhead during route discovery and improves network transmission efficiency under limited bandwidth resources.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A highly dynamic ad hoc network routing method based on dynamic topology inference includes the following steps; For communication in an ad hoc network of unmanned aerial vehicle (UAV) swarms, the UAV swarms are composed of communication terminals with high-speed random movement capabilities and whose relative positions between nodes change frequently over time. The network topology exhibits significant high dynamic time-varying characteristics. By passively monitoring the service traffic characteristics of UAV communication terminal nodes, UAV communication terminal traffic data is provided for topology inference with low additional overhead. Step 1: By passively monitoring the service traffic characteristics of UAV communication terminal nodes, UAV communication terminal traffic data is provided for topology inference with low additional overhead; Step 2: Apply connectivity analysis and connectivity decision strategies to the traffic data obtained in Step 1 to determine the correlation between UAV communication terminal nodes; based on the correlation of traffic data and by inferring the physical topology of the entire network, obtain topology information; Step 3: Based on topology information, construct a hybrid routing decision mechanism that combines prior inference and on-demand discovery; under the premise of ensuring route reachability in a highly dynamic network environment, effectively reduce the control overhead of route discovery, alleviate channel congestion, and improve network transmission efficiency under limited bandwidth.

[0008] Step 1 specifically involves: Step 1.1: Set the time interval for the drone communication terminal nodes in the network. , at time intervals The initial number of service data packets received for service transmission within the time period T is used as the statistical period. =0; Step 1.2: Whenever a service data packet is received within the statistical period, Increment by 1; Step 1.3: During each statistical period At the end, each UAV communication terminal node stores the first... The number of packages received within each time interval and the current time; a total of [number] packages need to be recorded. A time interval, if the currently recorded time is less than If the condition is met, return to step 1.1; otherwise, proceed to step 1.4. Step 1.4: Combine the recorded N packet reception statistics into a service traffic feature, send the service traffic feature as sampling data to the UAV communication terminal node with id 1, and clear the data recorded by this node.

[0009] In step 2, the connectivity analysis strategy specifically involves the following steps for processing the UAV communication terminal traffic data sampled in step 1: Step 2.1.1: Process the statistics sent by each UAV communication terminal node and combine the data from each node into a matrix. ; This represents the terminal traffic data sequence with node ID 1. There are p nodes in the network. This represents the traffic data sequence with id p; Step 2.1.2: Observational Statistics Matrix For each dimension of the data, missing data is filled using a fitting interpolation method. Step 2.1.3: Calculate the mean of the data in the current dimension. and variance ,mark The value is replaced by the arithmetic mean of adjacent points; It is a matrix The value in row i and column j; Step 2.1.4: Obtain the matrix based on the corrected sampled values ​​after interpolation completion in Step 2.1.2. precision matrix : in the formula The first in the precision matrix Line number The elements in the column represent drone communication terminal nodes. Communication terminal node with drone The joint correlation, and These are the corresponding UAV communication terminal nodes. and drone communication terminal nodes The conditional variance reciprocal, the precision matrix It describes the situation when other variables are given. and The correlation, when hour, and Mutually independent, that is and The two drone communication terminal nodes are not connected; A larger value indicates a stronger correlation. However, due to interference and other factors, the actual calculated value is mostly not 0. Simply using the definition for judgment will cause false positives (judging a link as connected when there is no link) and errors (judging a link as disconnected when there is a link). Therefore, three judgment methods are given below.

[0010] The connectivity decision strategy is specifically as follows: Method 1: Set point 0.5 as the threshold for connectivity judgment. When the precision matrix... Elements within At that time, the judgment and The two UAV communication terminal nodes are not connected; when the elements in the precision matrix At that time, the judgment and The two drone communication terminal nodes are connected; Method 2: Employ the maximum likelihood estimation method (ML) with regularization terms. The norm is used as a regularization term for estimation, and the accuracy matrix is ​​calculated. Some elements estimated in the model are penalized to 0 to ensure that the corresponding network topology satisfies sparsity. The model is as follows: in The sample covariance matrix Represents the trace of a matrix. Regularization parameters used to control sparsity; After regularization, when the elements in the precision matrix At that time, the judgment and The two UAV communication terminal nodes are not connected; when the elements in the precision matrix At that time, the judgment and The two drone communication terminal nodes are connected; Method 3: Perform multiple verifications: Step (1): Calculate the partial correlation coefficient matrix P, where each off-diagonal element represents a UAV communication terminal node. Communication terminal node with drone The direct correlation, excluding the influence of other UAV communication terminal nodes, is calculated using the following formula: In the formula, the molecule For the precision matrix The Middle Line number The off-diagonal elements of the column, in a physical sense, represent the UAV communication terminal node, assuming that the traffic characteristics of all other nodes in the control network remain unchanged. Communication terminal node with drone The strength of conditional dependence between them; in the denominator and These are the corresponding nodes in the precision matrix. and nodes The diagonal elements represent the conditional precision of the service traffic statistics for that node. Step (2): For each UAV communication terminal node: Here, i and j represent the number of rows and columns, meaning that the number of rows must be less than the number of columns, and only the lower left part of the matrix is ​​calculated. a. Calculate the partial correlation coefficient Conditional or partial correlations between variables; b. Calculate the t-statistic: ; Where p is the total number of variables. It is the number of samples in the time series. The partial correlation coefficient is calculated in step a; the independent samples t-test is used to compare whether there is a significant difference between the means of two independent samples. c. Calculate the two-sided p-value: , where T follows a t-distribution with (np) degrees of freedom; the t-test statistic needs to be compared with the t-distribution, and the shape of the t-distribution is determined by the degrees of freedom; Step (3): For all Multiple corrections (Benjamini-Hochberg method) were applied to control for FDR in each test. level; Step (4): The corrected p-value is less than... Edges that are connected are considered to be connected, and those that are not connected are considered to be disconnected.

[0011] Step 3 specifically involves: The operation of the hybrid routing decision mechanism is generally divided into three stages: topology data acquisition, topology structure initialization, and topology dynamic maintenance. During the topology data acquisition phase, data is collected based on the current network conditions, progressing in units of "topology sampling windows" (i.e., preset time periods, such as 5 seconds). In the topology initialization phase, a lightweight topology inference model is constructed based on sampled values ​​to replace the traditional flood detection network view and generate a dynamic topology view. During the topology dynamic maintenance phase, the UAV communication terminal node sends data packets by reading the inference results. At the same time, it parses the path feedback information (including node connectivity and link latency) of the RREQ (Route Request) and RREP (Route Response) control packets and updates the optimal path set in real time. After each "topology inference round", the path confidence is re-evaluated, retaining high-confidence paths and eliminating invalid paths, so that the UAV communication terminal node only initiates verification for the inferred valid paths, reducing redundant control packets from the source. This phase achieves a balance between topology dynamic adaptation and overhead control through "topology sampling window batch processing + real-time RREQ / RREP path fusion".

[0012] The topology data acquisition phase specifically includes: Step (1): Set the statistical period for each UAV communication terminal node as follows: ,initialization Number of data packets received within the time period When the value is 0, the system enters the statistical preparation state and waits for data packets to be received. Step (2): Whenever a data packet is in the time interval Once received, the drone communication terminal node will count the received packets locally. Increment by 1 to update the real-time statistics status; Step (3): When each time interval At the end, the node stores the number of packets received during the k-th time interval. (k is the current interval number); if the total number of recorded time intervals is less than the preset threshold N, return to step (2) to continue the next cycle of statistics; otherwise, proceed to step (4); Step (4): The node will record the number of N packets received ( to The data is sent to the target node with ID 1, and then all historical packet receiving records stored on this node are cleared, completing the data reporting for a single statistical cycle.

[0013] The specific steps for inferring the topology initialization phase are as follows: Step (1): Statistical reception and matrix construction: The UAV communication terminal central node (such as the node with id=1) receives statistics reported by all UAV communication terminal nodes in the network, that is, each UAV communication terminal node reports statistics at fixed intervals within a preset time window T. The collected sequence of received packets has a length of N; the N-dimensional statistics are arranged according to the node dimension and combined to form a statistical matrix A with a dimension of M×N, where M is the total number of nodes and each row corresponds to the sequence of received packets of one node. Step (2): Missing value completion and outlier correction: (2.1) If there are missing data in each dimension (column) of the observation statistics matrix A due to intermittent reporting by the UAV communication terminal node, then the estimated value of the missing point is calculated by fitting a Gaussian probability distribution to the adjacent valid data points before and after the missing point. (2.2) Calculate the mean of the data series for each dimension. and variance The marker deviates from the mean by more than Outliers (denoted as) The outlier was replaced with the arithmetic mean of the two valid data points immediately before and after the outlier, resulting in the corrected statistical matrix. ; Step (3): Precision matrix calculation and preliminary connectivity judgment: (3.1): Based on the corrected statistical matrix Calculate its covariance matrix (Dimensions M×M, elements) (The covariance of the sequence of packet counts received by node k and node l). (3.2): Calculate the covariance matrix The inverse matrix is ​​denoted as the precision matrix. ( (dimension M×M), where Characterize the conditional correlation between node k and node l under the condition of controlling other node variables; (3.3): Preliminary assessment of connectivity: If If so, then node k and node l are statistically independent in the current statistical period (i.e., there is no direct link connection). If so, mark it as a potential connected edge and proceed to the next step of detailed analysis; Step (4): Partial correlation analysis and significance test: (4.1): Calculate the partial correlation coefficient matrix P (dimension M×M), where the off-diagonal element P[k,l] represents the partial correlation coefficient between node k and node l under the control of other node variables. The calculation formula is as follows: , These are the diagonal elements of the precision matrix, i.e., the reciprocals of the conditional variances of nodes k and l; (4.2): For each off-diagonal element Perform a significance test: For each drone communication terminal node: a. Calculate the partial correlation coefficient Conditional or partial correlations between variables; b. Calculate the t-statistic: , where p is the total number of variables; the independent samples t-test is used to compare whether there is a significant difference between the means of two independent samples; c. Calculate the two-sided p-value: , where T follows a t-distribution with (np) degrees of freedom; the t-test statistic needs to be compared with the t-distribution, and the shape of the t-distribution is determined by the degrees of freedom; (4.3): For the test results of all M(M-1) / 2 off-diagonal elements, the Benjamini-Hochberg method is used to control the false detection rate (FDR) at a preset level q (e.g., q=0.05) to obtain the corrected p-value; Step (5): Connectivity determination and topology output: Edges with corrected p values ​​(i.e., nodes k and l) are considered to have valid connectivity, while those with negative p values ​​are considered to have no connectivity. Based on the determination results, the current network topology graph is generated (nodes are network nodes, and edges are valid links that are determined to be connected), thus completing the network topology inference. The dynamic maintenance phase of the topology structure specifically includes: Step (1) Data packet path establishment process: After the network obtains the initial topology inference results, the dynamic topology maintenance phase begins. i. Routing table establishment The steps for establishing a data packet path are as follows: Step (1): Each UAV communication terminal node reads the topology inference result, i.e., the known set of connectivity relationships. ,in Indicates the communication terminal node of the unmanned aerial vehicle. and There are bidirectional connected links between them. Construct a topology graph represented by an adjacency list. Where V is the set of nodes and E is the set of edges; the adjacency list data structure is defined as follows: each node maintains a neighbor list, storing all adjacent nodes directly connected to it; Step (2): Use the breadth-first search (BFS) algorithm to find the shortest path from the source node to the target node; The specific calculation process for the shortest path is as follows: Initialization phase: Create an access marker array visited[] to record whether each UAV communication terminal node has been visited; create a predecessor node array predecessor[] for path backtracking; initialize a FIFO queue Q, add the source node to the queue and mark it as visited; Traversal search phase Retrieve the currently pending node from queue Q. Then mark its status as "visited", and then traverse the nodes to be processed. All unvisited neighbor nodes in the adjacency list (i.e., satisfy) and (nodes); For each unvisited neighbor node Perform the following operations in sequence: Status flag: Unvisited neighbor nodes The access status visited[v] is updated to true to avoid duplicate processing; Precursor record: record predecessor[v]=u in the path tracing array predecessor[], which is used for subsequent path backtracking; Queue insertion: Inserting unvisited neighbor nodes Add to the end of the FIFO queue to ensure that traversal continues in hierarchical order; Repeat the above operation until the queue is empty, that is, all reachable nodes have been visited; this process ensures that the network topology is traversed in ascending order of hop count with the source node as the root, and finally a set of paths is generated. Path backtracking phase Starting from the target node, trace backwards along the predecessor node array until the source node to obtain the complete path; Step (3): Convert the discovered paths into routing table entries, in the format: {destination node, next-hop node, path weight, time-to-live}. The route weight is calculated based on the number of hops; fewer hops result in a higher weight. Set the route time-to-live. 300 seconds, automatically deleted after timeout; ii. The steps when a node needs to send packets are as follows: Step (1): The UAV communication terminal node checks whether the topology inference results obtained locally or in real time contain valid routing information of the destination node of the data packet; Step (2): If the topology inference result contains at least one path to the destination node, the node does not need to trigger the subsequent RREQ (request) routing process, but directly uses the path determined by the topology inference result as the transmission path of the data packet and performs the forwarding operation. Step (3): If there is no path to the destination node in the topology inference result, the node starts the RREQ pathfinding mechanism, generates an RREQ packet carrying the necessary pathfinding information such as the destination node identifier, and sends the RREQ packet to other nodes in the network in a network-wide flooding manner to trigger the path discovery process across the entire network. Step (II) Normal Topology Update i. Regular updates Because of the dynamic movement of UAV communication terminal nodes in self-organizing networks, the network topology will constantly change, therefore the topology inference results need to be updated periodically: Step (1): Each UAV communication terminal node continuously collects data. The number of packets received within a second is reported to the central node with ID=1 every few seconds; Step (2): After receiving the information reported by all nodes, the UAV communication terminal node with ID=1 executes the topology inference algorithm to generate a global topology view; Step (3): The UAV communication terminal node with ID=1 sends the inferred topology results to all nodes, and each node updates its local routing and forwarding strategy according to the new topology; ii. Updated irregularly When the UAV communication terminal node receives a control information packet of type Route Request (RREQ) or Route Response (RREP), it performs the following operations: Step (1): Parse the path sequence Extract the path information carried by the control message packet from specific fields. This information is an ordered sequence of node identifiers. ,in As the source node, For the current node or the destination node, the sequence describes the path of data packet transmission; Step (2): Extract path features From the path Obtain each hop connection And record the system time when this path information was received. If the new route information is better than the existing route in the routing table, then the new route will replace the existing route. Step (3): Topology correction and supplementation path Transform into a set of directed edges For each edge : If it is a new edge in the local topology, then add it to the topology graph. middle; If it already exists If the topology inference result of the path is incorrect, then the incorrectly inferred link is broadcast and flooded to the entire network, and each node corrects its local topology upon receiving the broadcast. Step (3) Topology Anomaly Update: When the UAV communication terminal node receives a control information packet of type Routing Error Report (RRER), it performs the following operations: Step (1): Parse the fault path sequence Extract the reported fault path and the node identifier sequence of the failed link from the specified fields of the RER message. And parse the fault type code encapsulated within it. (e.g., link interruption, node unreachable, etc.); Step (2): Deleting invalid entries in the routing table Traverse the local routing table, for any next hop or path containing The routing entry for the fault point mentioned above is set to invalid; Step (3): Topology Update Locally maintained topology graph In the middle, find with Corresponding edges Update the state of the edge to interrupted and remove the edge from the valid topology, thereby excluding this faulty link in subsequent route calculations.

[0014] The method is applied to highly dynamic UAV swarm ad hoc network communication scenarios, serving as the topology-aware foundation for the operation of routing protocols in highly dynamic UAV swarm networks. It replaces the path discovery methods that rely on high-frequency broadcasting or blind flooding in traditional routing protocols with a passive inference mechanism based on service traffic characteristics, thereby providing low-overhead link state information for routing decisions. It effectively overcomes the technical shortcomings of existing technologies, such as excessive routing overhead, channel resource congestion, and limited service bandwidth utilization caused by the surge in control messages in high-speed node movement environments.

[0015] The beneficial effects of this invention are: During network operation, nodes do not need to periodically send probe messages. Instead, they monitor and record their own service traffic statistics within a sliding time window in real time. These statistics include the number of data packets received per unit time and the message timestamp sequence. Nodes use this local observation data as the basic input for topology inference. This method passively samples existing service communication activities of the nodes, thus avoiding additional network overhead.

[0016] Based on the statistical correlation of traffic data between nodes, the topology inference strategy described in step two is executed to establish a network topology inference model. Specifically, this modeling process first constructs a precision matrix using the corrected service traffic feature sequence to quantify the strength of conditional dependencies between nodes; then, regularization sparsity constraints and a multi-verification mechanism for partial correlation coefficients are applied to statistically select node pairs with significant dependencies. The model uses this mechanism to mine the temporal and spatial correlation characteristics of the number of packets received by each node, calculates the confidence probability of physical links existing between nodes, and thus reconstructs a dynamic network topology view. This process realizes the transformation from discrete traffic statistics to logical network structure, providing information for subsequent accurate routing decisions.

[0017] A feedback loop is established to interact with the inference model and routing protocol, performing deep analysis of the topology information carried in received control messages (such as RREQ, RREP, and RERR). When a node receives a control message or data packet, it extracts deterministic information such as the source route list, cumulative hop count, and neighbor node relationships from the header, and uses the actual link states to perform real-time comparison and correction of the topology view generated based on traffic inference. This step constructs a dynamic feedback correction closed loop, utilizing the precise path information implicit in the control packets to eliminate failed links in the inference model and fill in missing connections, ensuring that the topology can adaptively follow the rapid movement of UAV nodes to achieve deterministic convergence and continuously approximate the real network state. Attached Figure Description

[0018] Figure 1 This is a flowchart of the data collection process based on passive perception of business traffic.

[0019] Figure 2 This is a flowchart of topology inference.

[0020] Figure 3 A flowchart is created for establishing a routing table based on topology information.

[0021] Figure 4 This is a flowchart of the routing decision process.

[0022] Figure 5 This is a performance comparison chart of topology inference methods. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings.

[0024] A highly dynamic ad hoc network routing method based on dynamic topology inference includes the following steps; Step 1: Passive data collection based on business traffic perception; This step aims to collect data for topology inference through passive traffic observation. Unlike traditional routing protocols that rely on periodic broadcast probe messages such as Hello packets for active probing, this invention directly utilizes the existing service communication characteristics of UAV communication terminal nodes as input, reducing the channel resource consumption of control packets at the data collection source. This design avoids channel congestion caused by frequent control packet transmission in highly dynamic environments and reserves valuable spectrum resources for service data, effectively improving network transmission efficiency under limited bandwidth resources.

[0025] Step 1 specifically involves: Step 1.1: Set the time interval for each node as follows. As the statistical period, the number of data packets received within the initial time period T is used. It is 0.

[0026] Step 1.2: Whenever a data packet occurs within a time interval Internally received, Increment by 1.

[0027] Step 1.3: When each time interval At the end, each node stores the first... The number of packets received within a time interval. If the currently recorded time is less than [a certain number of packets received]... Return to step 1; otherwise proceed to step 4.

[0028] Step 1.4: Send the recorded N data points to the node with id 1, and then clear the data recorded on this node.

[0029] Step 2: Topology inference strategy based on traffic correlation analysis; This step aims to infer the entire network topology based on discrete service traffic statistics through statistical correlation analysis. This strategy utilizes maximum likelihood estimation and multiple verification mechanisms to replace the traditional method of relying on network-wide flooding to obtain a topology view. This process provides high-confidence prior link states for subsequent routing decisions, thereby significantly reducing control packet overhead during route discovery and improving network transmission efficiency under limited bandwidth resources while ensuring route reachability in highly dynamic environments.

[0030] 2.1 Connectivity analysis strategy; The processing steps for the data sampled in step 1 are as follows: Step 2.1.1: Node ID processes the statistics sent by each node and combines the data from each node into a matrix. .

[0031] Step 2.1.2: Observe the data in each dimension and fill in the missing data using fitting interpolation.

[0032] Step 2.1.3: Calculate the mean of the data in the current dimension. and variance ,mark The value is replaced by the arithmetic mean of adjacent points.

[0033] Step 2.1.4: Obtain the matrix based on the sampled values ​​processed in Step 1. precision matrix : Precision matrix It describes the situation when other variables are given. and The correlation, when hour, and Mutually independent, that is and The two nodes are not connected.

[0034] 2.2 Connectivity Decision Strategy; 2.2.1. Threshold determination: Let 0.5 be the threshold for connectivity assessment. When the elements in the precision matrix... At that time, the judgment and The two nodes are not connected; when the elements in the precision matrix At that time, the judgment and The two nodes are connected; 2.2.2: Regularization: The maximum likelihood estimation method (ML) with regularization terms is typically used. The norm is used as a regularization term for estimation, and the accuracy matrix is ​​calculated. Some elements estimated in the model are penalized to 0 to ensure that the corresponding network topology satisfies sparsity. The model is as follows: After regularization, when the elements in the precision matrix At that time, the judgment and The two nodes are not connected; when the elements in the precision matrix At that time, the judgment and The two nodes are connected; 2.2.3. Multiple verification: Step (1). Calculate the partial correlation coefficient matrix P, where each off-diagonal element .

[0035] Step (2). For each: ; a. Calculate the partial correlation coefficient Conditional or partial correlations between variables; b. Calculate the t-statistic: , where p is the total number of variables; the independent samples t-test is used to compare whether there is a significant difference between the means of two independent samples.

[0036] c. Calculate the two-sided p-value: , where T follows a t-distribution with (np) degrees of freedom; the statistic of the t-test needs to be compared with the t-distribution, and the shape of the t-distribution is determined by the degrees of freedom.

[0037] Step (3): For all Multiple corrections (Benjamini-Hochberg method) were applied to control for FDR in each test. level.

[0038] Step (4). The corrected p-value is less than... Edges that are connected are considered to be connected, and those that are not connected are considered to be disconnected.

[0039] Step 3: Dynamic Routing Based on Topology Inference Results and Feedback Correction; This step, based on the topology information obtained in Step 2, constructs a hybrid routing decision mechanism combining prior inference and on-demand discovery. The UAV communication terminal node first constructs a local topology map based on the inferred adjacency relationships and obtains candidate paths from the source node to the destination node using the shortest path algorithm. During packet forwarding, the inferred valid paths are prioritized for routing; the active route discovery process is only initiated when the inferred path does not exist or becomes invalid. This mechanism, while ensuring route reachability in highly dynamic network environments, transforms the frequent network-wide flooding probing in traditional routing protocols into an inference process primarily based on local computation, effectively reducing the control overhead of route discovery, alleviating channel congestion, and improving network transmission efficiency under limited bandwidth.

[0040] Step 3 specifically involves: The routing protocol described in this invention generally operates in three stages: topology data acquisition, topology structure initialization, and topology dynamic maintenance. Its core is to achieve topology awareness and path updates through a low-overhead mechanism, replacing the blind probing of traditional routing protocols.

[0041] In the topology data acquisition phase, data is collected based on the current network conditions, progressing in units of "topology sampling windows" (i.e., preset time periods, such as 5 seconds). In the topology initialization phase, a lightweight topology inference model is built based on the sampled values ​​to replace the traditional flooding detection's full network view, generating a dynamic topology view. In the dynamic maintenance phase, nodes send data packets by reading the inference results, and simultaneously parse the path feedback information (including node connectivity and link latency) from RREQ (Route Request) and RREP (Route Response) control packets, updating the optimal path set in real time. After each "topology inference round," the path confidence is reassessed, retaining high-confidence paths and discarding invalid paths, ensuring that nodes only initiate verification for inferred valid paths, reducing redundant control packets from the source. This phase achieves a balance between dynamic topology adaptation and cost control through "topology sampling window batch processing + real-time RREQ / RREP path fusion."

[0042] Step 3.1 Topology Data Acquisition Phase like Figure 1 As shown, the steps for topology data acquisition are as follows: Step (1): Set the statistical period for each node as follows ,initialization Number of data packets received within the time period If the value is 0, the system enters the statistics preparation state and waits for data packets to be received.

[0043] Step (2): Whenever a data packet is in the time interval Once received, the node will count the received packets locally. Increment by 1 to update the real-time statistics status.

[0044] Step (3): When each time interval At the end, the node stores the number of packets received during the k-th time interval. (k is the current interval number); if the total number of recorded time intervals is less than the preset threshold N, return to step 2 to continue the statistics for the next cycle; otherwise, proceed to step 4.

[0045] Step (4) The node will record the number of N packets received ( to The data is sent to the target node with ID 1, and then all historical packet receiving records stored on this node are cleared, completing the data reporting for a single statistical cycle.

[0046] Step 3.2 Topology Initialization Phase like Figure 2 As shown, the steps for topology inference are as follows: Step (1): Statistical reception and matrix construction The central node (such as the node with id=1) receives statistics reported by all nodes in the network, that is, each node reports statistics at fixed intervals within a preset time window T. The collected packet count sequence has a length of N. The N-dimensional statistics are arranged according to the node dimension and combined to form a statistical matrix A of dimension M×N, where M is the total number of nodes, and each row corresponds to the packet count sequence of one node.

[0047] Step (2): Missing value completion and outlier correction (2.1) If there are missing data in each dimension (column) of the observation matrix A due to intermittent reporting by nodes, the estimated value of the missing point is calculated by fitting a Gaussian probability distribution to the valid data points before and after the missing point.

[0048] (2.2) Calculate the mean of the data series for each dimension. and variance The marker deviates from the mean by more than Outliers (denoted as) The outlier was replaced with the arithmetic mean of the two valid data points immediately before and after the outlier, resulting in the corrected statistical matrix. .

[0049] Step (3): Precision matrix calculation and preliminary connectivity judgment (3.1) Based on the correction matrix Calculate its covariance matrix (Dimensions M×M, elements) (The covariance of the sequence of packet counts received by node k and node l). (3.2) Obtain the covariance matrix The inverse matrix is ​​denoted as the precision matrix. ( (dimension M×M), where Characterize the conditional correlation between node k and node l under the condition of controlling other node variables; (3.3) Preliminary assessment of connectivity: If If so, then node k and node l are statistically independent in the current statistical period (i.e., there is no direct link connection). If so, it is marked as a potential connected edge, and we proceed to the next step of detailed analysis.

[0050] Step (4): Partial correlation analysis and significance test (4.1) Calculate the partial correlation coefficient matrix P (dimension M×M), where the off-diagonal element P[k,l] represents the partial correlation coefficient between node k and node l under the control of other node variables. The calculation formula is as follows: , These are the diagonal elements of the precision matrix, i.e., the reciprocals of the conditional variances of nodes k and l. (4.2) For each off-diagonal element Perform a significance test: For each: a. Calculate the partial correlation coefficient Conditional or partial correlations between variables; b. Calculate the t-statistic: , where p is the total number of variables; the independent samples t-test is used to compare whether there is a significant difference between the means of two independent samples; c. Calculate the two-sided p-value: , where T follows a t-distribution with (np) degrees of freedom; the statistic of the t-test needs to be compared with the t-distribution, and the shape of the t-distribution is determined by the degrees of freedom.

[0051] (4.3) For the test results of all M(M-1) / 2 off-diagonal elements, the Benjamini-Hochberg method is used to control the false detection rate (FDR) at a preset level q (e.g., q=0.05) to obtain the corrected p value.

[0052] Step 5: Connectivity determination and topology output Edges with corrected p values ​​(i.e., nodes k and l) are considered to have valid connectivity, while those with negative p values ​​are considered to have no connectivity. Based on the determination results, the current network topology graph is generated (nodes are network nodes, and edges are valid links that are determined to be connected), thus completing the network topology inference.

[0053] Step 3.2 Dynamic Maintenance Phase of Topology Step 3.2.1 Data Packet Path Determination Process After the network obtains the initial topology inference results, the dynamic topology maintenance phase begins.

[0054] i. Routing table establishment like Figure 3 As shown, the steps for establishing the data packet path are as follows: Step (1): Each node reads the results of topology inference, that is, the set of known connectivity relationships. ,in Represents a node and There are bidirectional connected links between them. Construct a topology graph represented by an adjacency list. Let V be the set of nodes and E be the set of edges. The adjacency list data structure is defined as follows: each node maintains a list of neighbors, storing all directly connected adjacent nodes.

[0055] Step (2): Use the breadth-first search (BFS) algorithm to find the shortest path from the source node to the target node.

[0056] The specific process is as follows: ① Initialization phase: Create an array of visit markers [visited] to record whether each node has been visited; create an array of predecessor nodes [predecessor] for path backtracking; initialize a FIFO queue Q, add the source node to the queue and mark it as visited.

[0057] ② Traversal and search phase Retrieve the currently pending node from queue Q. Then mark its status as "visited". Then, traverse the nodes. All unvisited neighbor nodes in the adjacency list (i.e., satisfy) and (nodes).

[0058] For each unvisited neighbor node Perform the following operations in sequence: Status marker: mark the node The access status visited[v] is updated to true to avoid duplicate processing; Precursor record: record predecessor[v]=u in the path tracing array predecessor[], which is used for subsequent path backtracking; Queue insertion: inserting a node Add to the end of the FIFO queue to ensure that traversal continues in hierarchical order; Repeat the above operation until the queue is empty, meaning all reachable nodes have been visited. This process ensures that the network topology is traversed in ascending order of hop count, with the source node as the root, ultimately generating a set of paths.

[0059] ③ Path backtracking stage Starting from the target node, trace backward along the predecessor node array until the source node to obtain the complete path.

[0060] Step 3: Convert the discovered paths into routing table entries, in the format: {Destination Node, Next-Hop Node, Path Weight, Time-to-Live}. The route weight is calculated based on the number of hops; fewer hops result in a higher weight. Set the route time-to-live. 300 seconds, will be automatically deleted after the timeout.

[0061] ii. Packet routing steps like Figure 4 As shown, the steps when a node needs to send packets are as follows: Step (1): The node checks whether the topology inference results obtained locally or in real time contain valid routing information of the destination node of the data packet.

[0062] Step (2): If the topology inference result contains at least one path to the destination node, the node does not need to trigger the subsequent REQ (request) routing process. Instead, it directly uses the path determined by the topology inference result as the transmission path of the data packet and performs the forwarding operation.

[0063] Step (3): If there is no path to the destination node in the topology inference result, the node starts the REQ pathfinding mechanism, generates a REQ packet carrying the necessary pathfinding information such as the destination node identifier, and sends the REQ packet to other nodes in the network in a network-wide flooding manner to trigger the path discovery process across the entire network.

[0064] 3.2.2 Normal Topology Update i. Regular updates Because of the dynamic movement of nodes in a self-organizing network, the network topology is constantly changing, therefore the topology inference results need to be updated periodically. Step (1): Each node continuously collects data. The number of packets received within a second is reported to the central node with ID=1 every few seconds.

[0065] Step (2): After receiving the information reported by all nodes, the node with ID=1 executes the topology inference algorithm to generate a global topology view.

[0066] Step (3): The node with ID=1 sends the inferred topology results to all nodes, and each node updates its local routing and forwarding policies according to the new topology.

[0067] ii. Updated irregularly When a node receives a control message packet of type Route Request (REQ) or Route Reply (REP), it performs the following operations: Step (1): Parse the path sequence Extract the path information carried by the control message packet from specific fields. This information is an ordered sequence of node identifiers. ,in As the source node, For the current node or the destination node, the sequence describes the path of data packet transmission.

[0068] Step (2): Extract path features From the path Obtain each hop connection And record the system time when this path information was received. If the new path information is better than the existing route in the routing table, then the new route will replace the existing route.

[0069] Step (3): Topology correction and supplementation path Transform into a set of directed edges For each edge : If it is a new edge in the local topology, then add it to the topology graph. middle,; If it already exists If the topology inference result for that path is incorrect, the incorrectly inferred link is broadcast and flooded to the entire network, and each node corrects its local topology upon receiving the broadcast.

[0070] 3.2.3 Topology Anomaly Update When a node receives a control packet of type Routing Error Report (RER), it performs the following operations: Step (1): Parse the fault path sequence Extract the reported fault path and the node identifier sequence of the failed link from the specified fields of the RER message. And parse the fault type code encapsulated within it. (such as link interruption, node unreachable, etc.).

[0071] Step (2): Deleting invalid entries in the routing table Traverse the local routing table, for any next hop or path containing The routing entry for the fault point mentioned above is set to invalid.

[0072] Step (3): Topology Update Locally maintained topology graph In the middle, find with Corresponding edges The state of the edge is updated to "discontinued" and the edge is removed from the valid topology, thus excluding this failed link in subsequent route calculations.

[0073] In the simulation experiment, the application platform was used for simulation based on MATLAB R2021a (or the corresponding version).

[0074] The simulation scenario is constructed as a 20km × 10km two-dimensional space, in which 6 mobile nodes are randomly deployed. By setting a uniform node transmission range (usually covering the entire scenario or a sufficiently large distance), it is ensured that each of the 6 nodes can communicate directly with each other, thus forming a number of C(6,2) = 15 undirected links.

[0075] To simulate the fluctuations in data transmission and reception in a real network, the number of successfully received data packets collected within a measurement period for each valid communication link is simulated using a Gaussian (normal) probability distribution. The mean of this distribution can be set according to specific network conditions to reflect random variations around a certain typical value.

[0076] Table 1 Simulation Parameters Simulation content and results In the experimental scenarios shown in Table 1, threshold determination, LASSO, and significance testing mechanisms were used for identification, with different sample sizes set. As variables, the recognition accuracy of each method under different sampling values ​​and the performance gap with the best method are compared. To avoid the randomness of the experiment, the data in the figure are the mean of 50 simulation results.

[0077] from Figure 5 It is evident that the significance test method achieves the best relevance identification effect across the entire sample range. While the threshold determination method performs reasonably well in the initial stage, its overall performance exhibits significant instability with varying sample size, lacking stability. Particularly on the right side, the gap with the optimal method consistently oscillates, indicating weak consistency in sample allocation. The LASSO method demonstrates a clear sample-dependent characteristic, showing poor performance with a small sample size (…). When the sample size is insufficient, the recognition accuracy is the lowest and the gap with the best method is the largest (around 0.18), indicating unsatisfactory performance. However, with a larger sample size... With the increase of the sample size, this method can introduce more effective information, and the recognition accuracy shows a significant upward trend, gradually narrowing the gap with the optimal solution. In contrast, the saliency test method always maintains the corresponding recognition accuracy level (around 0.6), and this mechanism is minimally affected by the uniform change of sample size. It can maintain flexible recognition performance under different data volumes, making it a suitable solution for the current scenario.

Claims

1. A highly dynamic ad hoc network routing method based on dynamic topology inference, used for ad hoc network communication of unmanned aerial vehicle (UAV) swarms, wherein the UAV swarm consists of communication terminals with high-speed random mobility and whose relative positions between nodes change frequently over time, and the network topology exhibits significant highly dynamic time-varying characteristics, characterized in that... Includes the following steps; Step 1: By passively monitoring the service traffic characteristics of UAV communication terminal nodes, UAV communication terminal traffic data is provided for topology inference with low additional overhead; Step 2: Apply connectivity analysis and connectivity decision strategies to the traffic data obtained in Step 1 to determine the correlation between UAV communication terminal nodes; based on the correlation and by inferring the physical topology of the entire network, obtain topology information; Step 3: Based on the topology information, construct a hybrid routing decision mechanism that combines prior inference and on-demand discovery; under the premise of ensuring route reachability in a highly dynamic network environment, reduce the control overhead of route discovery, alleviate channel congestion, and improve network transmission efficiency under limited bandwidth.

2. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 1, characterized in that, Step 1 specifically involves: Step 1.1: Set the time interval for the drone communication terminal nodes in the network. , at time intervals The initial number of service data packets received for service transmission within the time period T is used as the statistical period. =0; Step 1.2: Whenever a service data packet is received within the statistical period, Increment by 1; Step 1.3: At the end of each statistical period, each UAV communication terminal node stores the data for the first statistical period. The number of packages received within each time interval and the current time; a total of [number] packages need to be recorded. A time interval, if the currently recorded time is less than Return to step 1.1; Otherwise, proceed to step 1.4; Step 1.4: Combine the recorded N packet reception statistics into a service traffic feature, send the service traffic feature as sampling data to the UAV communication terminal node with id 1, and clear the data recorded by this node.

3. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 2, characterized in that, In step 2, the connectivity analysis strategy specifically involves the following steps for processing the UAV communication terminal traffic data sampled in step 1: Step 2.1.1: Process the statistics sent by each UAV communication terminal node and combine the data from each node into a matrix. ; This represents the terminal traffic data sequence with node ID 1. There are p nodes in the network. This represents the traffic data sequence with id p; Step 2.1.2: Observational Statistics Matrix For each dimension of the data, missing data is filled using a fitting interpolation method. Step 2.1.3: Calculate the mean of the data in the current dimension. and variance ,mark The value is replaced by the arithmetic mean of adjacent points; It is a matrix The value in row i and column j; Step 2.1.4: Obtain the matrix based on the corrected sampled values ​​after interpolation completion in Step 2.1.

2. precision matrix : in the formula The first in the precision matrix Line number The elements in the column represent drone communication terminal nodes. Communication terminal node with drone The joint correlation, and These are the corresponding UAV communication terminal nodes. and drone communication terminal nodes The conditional variance reciprocal, the precision matrix It describes the situation when other variables are given. and The correlation, when hour, and Mutually independent, that is and The two drone communication terminal nodes are not connected.

4. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 3, characterized in that, The connectivity decision strategy is specifically as follows: Method 1: Set point 0.5 as the threshold for connectivity judgment. When the precision matrix... Elements within At that time, the judgment and The two UAV communication terminal nodes are not connected; when the elements in the precision matrix At that time, the judgment and The two drone communication terminal nodes are connected; Method 2: Employ the maximum likelihood estimation method with regularization terms. The norm is used as a regularization term for estimation, and the accuracy matrix is ​​calculated. Some elements estimated in the model are penalized to 0 to ensure that the corresponding network topology satisfies sparsity. The model is as follows: in The sample covariance matrix Represents the trace of a matrix. Regularization parameters used to control sparsity; After regularization, when the elements in the precision matrix At that time, the judgment and The two UAV communication terminal nodes are not connected; when the elements in the precision matrix At that time, the judgment and The two drone communication terminal nodes are connected; Method 3: Perform multiple verification checks.

5. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 4, characterized in that, The specific steps of the multi-verification method are as follows: Step (1): Calculate the partial correlation coefficient matrix P, where each off-diagonal element represents a UAV communication terminal node. Communication terminal node with drone The direct correlation, excluding the influence of other UAV communication terminal nodes, is calculated using the following formula: In the formula, the molecule For the precision matrix The Middle Line number The off-diagonal elements of the column, in a physical sense, represent the UAV communication terminal node, assuming that the traffic characteristics of all other nodes in the control network remain unchanged. Communication terminal node with drone The strength of conditional dependence between them; in the denominator and These are the corresponding nodes in the precision matrix. and nodes The diagonal elements represent the conditional precision of the service traffic statistics for that node. Step (2): For each: Here, i and j represent the number of rows and columns, meaning that the number of rows must be less than the number of columns, and only the lower left part of the matrix is ​​calculated. a. Calculate the partial correlation coefficient Conditional or partial correlations between variables; b. Calculate the t-statistic: ; Where p is the total number of variables. It is the number of samples in the time series. The partial correlation coefficient is calculated in step a; the independent samples t-test is used to compare whether there is a significant difference between the means of two independent samples. c. Calculate the two-sided p-value: , where T follows a t-distribution with (np) degrees of freedom; the t-test statistic needs to be compared with the t-distribution; Step (3): For all Multiple correction control (FDR) is performed on each test. level; Step (4): The corrected p-value is less than... Edges that are connected are considered to be connected, and those that are not connected are considered to be disconnected.

6. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 5, characterized in that, Step 3 specifically involves: The operation of the hybrid routing decision mechanism is divided into three stages: topology data acquisition, topology structure initialization, and topology dynamic maintenance. The topology data acquisition phase collects data based on the current network conditions, progressing in units of "topology sampling windows"; The topology initialization phase constructs a lightweight topology inference model based on sampled values ​​and generates a dynamic topology view. During the topology dynamic maintenance phase, the UAV communication terminal node sends data packets by reading the inference results. At the same time, it parses the path feedback information of the route request (RREQ) and route response (RREP) control packets and updates the optimal path set in real time. After each "topology inference round", the path confidence is re-evaluated, high-confidence paths are retained and invalid paths are eliminated, so that the UAV communication terminal node only initiates verification for the inferred valid paths.

7. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 6, characterized in that, The topology data acquisition phase specifically includes: Step (1): Set the statistical period for each UAV communication terminal node as follows: ,initialization Number of data packets received within the time period When the value is 0, the system enters the statistical preparation state and waits for data packets to be received. Step (2): Whenever a data packet is in the time interval Once received, the drone communication terminal node will count the received packets locally. Increment by 1 to update the real-time statistics status; Step (3): When each time interval At the end, the node stores the number of packets received during the k-th time interval. k is the current interval number; if the total number of recorded time intervals is less than the preset threshold N, then return to step (2) to continue the statistics for the next cycle; Otherwise proceed to step (4); Step (4): The node will record the number of N packets received ( to The data is sent to the target node with ID 1, and then all historical packet receiving records stored on this node are cleared, completing the data reporting for a single statistical cycle.

8. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 7, characterized in that, The specific steps for inferring the topology initialization phase are as follows: Step (1): The UAV communication terminal central node receives the statistics reported by all UAV communication terminal nodes in the network, that is, each UAV communication terminal node reports statistics at a fixed period within a preset time window T. The collected sequence of received packets has a length of N; the N-dimensional statistics are arranged according to the node dimension and combined to form a statistical matrix A with a dimension of M×N, where M is the total number of nodes and each row corresponds to the sequence of received packets of one node. Step (2): Observe the data in each dimension (column) of the statistical matrix A. If there is data missing due to intermittent reporting by the UAV communication terminal node, fit a Gaussian probability distribution based on the valid data points before and after the missing point and calculate the estimated value of the missing point. Calculate the mean of the data sequence for each dimension. and variance The marker deviates from the mean by more than Outliers are denoted as The outlier was replaced with the arithmetic mean of the two adjacent valid data points before and after the outlier, resulting in the corrected statistical matrix. ; Step (3): Based on the corrected statistical matrix Calculate its covariance matrix Dimensions M×M, elements Let be the covariance of the sequence of packet counts received by node k and node l; Find the covariance matrix The inverse matrix is ​​denoted as the precision matrix. , , dimension M×M, where Characterize the conditional correlation between node k and node l under the condition of controlling other node variables; like If so, then node k and node l are statistically independent within the current statistical period, meaning they have no direct link connection. If so, mark it as a potential connected edge and proceed to the next step of detailed analysis; Step (4): Calculate the partial correlation coefficient matrix P, with dimension M×M, where the off-diagonal element P[k,l] represents the partial correlation coefficient between node k and node l under the control of other node variables. The calculation formula is: , These are the diagonal elements of the precision matrix, i.e., the reciprocals of the conditional variances of nodes k and l; For each off-diagonal element Perform a significance test: For the test results of all M(M-1) / 2 off-diagonal elements, the Benjamini-Hochberg method was used to control the false detection rate at a preset level q, and the corrected p-value was obtained. Edges with corrected p values ​​are considered to have valid connectivity, while those with lower p values ​​are considered to be disconnected. The current network topology is generated based on the judgment results.

9. The highly dynamic ad hoc network routing method based on dynamic topology inference according to claim 8, characterized in that, The dynamic maintenance phase of the topology structure specifically includes: Step (1): Packet path establishment process: After the network obtains the preliminary topology inference results, dynamic maintenance of the topology structure begins; Each UAV communication terminal node reads the results of topology inference, i.e., the known set of connectivity relationships. ,in Indicates the communication terminal node of the unmanned aerial vehicle. and There are bidirectional connected links between them. Construct a topology graph represented by an adjacency list. Where V is the set of nodes and E is the set of edges; the adjacency list data structure is defined as follows: each node maintains a neighbor list, storing all adjacent nodes directly connected to it; The breadth-first search (BFS) algorithm is used to find the shortest path from the source node to the target node. The discovered paths are converted into routing table entries, and the route weight is calculated based on the number of hops in the path; the fewer the hops, the higher the weight. A timeout is set for the route, and it is automatically deleted after the timeout. The steps when a node needs to send packets are as follows: The UAV communication terminal node checks whether the topology inference results, which are either stored locally or obtained in real time, contain valid routing information of the destination node of the data packet. If the topology inference result contains at least one path to the destination node, the node does not need to trigger the subsequent RREQ request routing process. Instead, it directly uses the path determined by the topology inference result as the transmission path for the data packet and performs the forwarding operation. If no path to the destination node is found in the topology inference results, the node initiates the RREQ pathfinding mechanism. It generates an RREQ packet carrying necessary pathfinding information such as the destination node identifier and sends the RREQ packet to other nodes in the network in a network-wide flooding manner to trigger a network-wide path discovery process. Step (2): Perform normal topology updates on the dynamically maintained topology structure; Each drone communication terminal node continuously collects... The number of packets received within a second is reported to the central node with ID=1 every few seconds; After receiving the information reported by all nodes, the UAV communication terminal node with ID=1 executes the topology inference algorithm to generate a global topology view; The drone communication terminal node with ID=1 will send the inferred topology results to all nodes, and each node will update its local routing and forwarding strategy according to the new topology. When the UAV communication terminal node receives a control information packet of type RREQ (Route Request RREQ) or RREP (Route Reply RREP), it performs the following operations: Extract the path information carried by the control message packet from specific fields. This information is an ordered sequence of node identifiers. ,in As the source node, For the current node or the destination node, the sequence describes the path of data packet transmission; From the path Obtain each hop connection And record the system time when this path information was received. If the new route information is better than the existing route in the routing table, then the new route will replace the existing route. path Transform into a set of directed edges For each edge : If it is a new edge in the local topology, then add it to the topology graph. middle; If it already exists If the topology inference result of the path is incorrect, then the incorrectly inferred link is broadcast and flooded to the entire network, and each node corrects its local topology upon receiving the broadcast. Step (3): Topology Anomaly Update: When the UAV communication terminal node receives a control information packet of type Route Error Report (RRER), perform the following operations: Extract the reported fault path and the node identifier sequence of the failed link from the specified fields of the RER message. And parse the fault type code encapsulated within it. ; Traverse the local routing table, for any next hop or path containing The routing entry for the fault point mentioned above is set to invalid; Locally maintained topology graph In the middle, find with Corresponding edges Update the state of the edge to interrupted and remove the edge from the valid topology, thereby excluding this faulty link in subsequent route calculations.

10. An application of the highly dynamic ad hoc network routing method based on dynamic topology inference as described in any one of claims 1-9, characterized in that, The method is applied to highly dynamic unmanned aerial vehicle (UAV) swarm self-organizing network communication scenarios.

Images